Agarose gel microcapsules enable easy-to-prepare, picolitre-scale single-cell genomics, yielding near-complete genome sequences

A novel type of agarose gel microcapsule (AGM), consisting of an alginate picolitre sol core and an agarose gel shell, was developed to obtain high-quality single-cell amplied genomic DNA of bacteria. The AGM is easy to prepare in a stable emulsion with oil of water-equivalent density which prevents AGM aggregation, with only standard laboratory equipment. Single cells from a pure culture of Escherichia coli, a mock community comprising 15 strains of human gut bacteria, and a termite gut bacterial community were encapsulated within AGMs, and their genomic DNAs were obtained with massively parallel amplications in a tube. The genome sequencing did not need second-round amplication, and showed an average genome completeness that was much higher than that obtained by the conventional amplication method in microlitre scale, regardless of the genomic guanine-cytosine contents. Our novel method using AGMs allows many researchers to perform single-cell genomics easily and effectively, and can enhance the genome analysis of yet-uncultured microorganisms.


Introduction
Microorganisms are ubiquitously distributed in diverse environments. They are often associated with other organisms, and play important roles in ecosystems. However, the majority of microbial species are di cult to culture with conventional methods and are called yet-uncultured microorganisms 1 . In the past two decades, they have been extensively studied using culture-independent methods such as amplicon sequencing analysis of small subunit ribosomal RNA (SSU rRNA) genes and shotgun sequencing analysis of metagenomes. Metagenomics is a powerful tool for investigating the ecological and physiological functions of microbiota based on their encoded genetic repertories. The recent development of the computational binning of metagenomic fragments into respective microbial taxonomic assemblies has reinforced the utility of metagenomics 2,3 . However, computational binning based on the sequence composition, homology to database sequences, and sequence read coverage of each fragment might often fail to discriminate genomes of closely related (sub)species and to correctly sort genomic regions exhibiting distinct features from others, such as rRNA genes, prophages, and plasmids 4,5 . Furthermore, a massive sequencing effort is required to obtain the genome sequences of minor species in the microbial community 3 .
Single-cell genomics, in which whole genomic DNA is ampli ed from physically isolated single cells 6 , has been used as an alternative or complementary method to reveal the metabolic capacity of yet-uncultured microorganisms 1 . Single-cell isolation can be achieved in many cases by using technologies such as uorescence-activated cell sorting (FACS) 1 , micromanipulators 7 , micro uidic devices 8,9 , and encapsulation in water-in-oil droplets 10 . Although various methods for whole genome ampli cation have been developed 11 , multiple displacement ampli cation (MDA) using phi29 DNA polymerase with random primers is the most widely used due to its relative simplicity, reliability, and applicability 6 . However, MDA inherently accompanies extreme ampli cation bias among genome regions 1 , which has been the major technical limitation in single-cell genomics. Additionally, the high guanine-cytosine (GC) content of genomic DNA tends to increase the ampli cation bias 1 . It has been reported that the MDA in a small reaction volume, for example, when using a microchannel chamber (60 nL) 9 , nano-litre microwells (12 nL) 12 , and digital droplets generated by a microchannel (9.53 pL) 13 , can suppress ampli cation bias.
However, these techniques are not available for many microbiologists, and the amount of the ampli cation product is often too small for the direct genome sequencing and a second-round MDA 14 may be required, which ultimately enhances the ampli cation bias.
To accelerate study using the single-cell genomics of yet-uncultured microorganisms, it is important to develop a simple method that prepares massive small reaction vessels using commercially available reagents, disposable plasticware, and equipment. For MDA reaction vessels, droplet microchannel massively generates single-cell encapsulating water-in-oil droplets 10 . However, MDA reagents are sequentially introduced into the droplets in the microchannel, which complexi es the microchannel structure. Single-cell encapsulating gel beads are easily prepared by gelating cell-containing solated gel droplets [14][15][16] . The MDA of the single-cell genome in the bead is done by exchanging reagents through gel. However, the amount of MDA product in the gel matrix is insu cient for next generation DNA sequencing. The hollow-core hydrogel microcapsule 17 , which consists of a hydrogel shell and a sol core, is expected to be an e cient vessel for single-cell MDA. Many hollow-core hydrogel microcapsules have been reported to be used for cell cultivations and cell transplantations: for example, a crosslinked polyethylene glycol (PEG) gel shell and an PEG sol core 18 , an alginate and poly-L-lysine complex shell and an alginate sol core 19 , and an agarose gel shell and an alginate sacri ce core 20 .
We here introduce a novel method for single cell genomics, using a newly developed agarose gel microcapsule (AGM) consisting of an agarose gel shell and alginate sol hollow-core, which enables single-cell isolation and picolitre-scale MDA (Fig. 1a). The fabrication process of the AGM was optimised here for easy preparation using inexpensive equipment and reagents, such as a vortex mixer, centrifuges, and commercially available reagents (Fig. 1b).
In this method, a single bacterial encapsulated cell in the picolitre-scale sol core of the AGM is subjected to MDA (designated here 'MDA-in-AGM'). Our study demonstrated that large quantities of AGMs containing single bacterial cells could be easily prepared at once, that massively parallel MDA-in-AGM in a single tube resulted in a large number of single-cell ampli ed genomes (SAGs) with high ampli cation e ciency, and that MDA-in-AGM greatly improves the genome completeness compared with conventional single-cell genomics.

Results
Preparation of AGMs containing single bacterial cells. Alginate gel cores containing bacterial cells were prepared using the emulsi cation/internal gelation method 21 (Fig. 1b). Brie y, a mixture of alginate sol solution, bacterial cells, and CaCO 3 nanoparticles was emulsi ed with isostearyl alcohol (ISA). The resulting microdroplets were gelated with calcium ions that were released from CaCO 3 nanoparticles by the addition of acetic acid to the water-in-oil emulsion. The concentration of bacterial cells was optimised using serial dilutions so that one alginate microdroplet contained one or no bacterial cells. The resulting alginate gel cores were washed sequentially with diethyl ether, 1-butanol, and Tris-HCl (pH 7.4) buffer.
Agarose solution (SeaPlaque, Lonza; nal 2%) was added to the alginate gel cores suspended in Tris-HCl (pH 7.4) at 35 °C, and the mixture was emulsi ed with 0.25% (v/v) Span 80 in polyglyceryl-6 octacaprylate (PGO) at 35 °C (Fig. 1b). The use of PGO with a density of 0.997 g/mL, which is equivalent to that of water and AGMs (~ 1.0 g/mL), enabled the formation of a stable water-in-oil emulsion during the gradual gelation of agarose by cooling at 4 °C. In addition, PGO improved the AGM preparation process by preventing the aggregation of AGMs and increasing their diameter and yield ( Supplementary   Fig. 1). After PGO was removed by mixing with diethyl ether and further with 1-butanol, the alginate gel cores were solated with a one-tenth volume of 0.5 M EDTA. The solation of alginate cores in AGMs was con rmed based on the core dissolution into the buffer after dissolving the agarose shells ( Supplementary Fig. 2). The alginate sol core is expected to provide an appropriate space for the MDA reaction, which was shown by the much greater yields of ampli ed DNA than those with an alginate gel core or agarose gel bead, as described below. The AGMs were washed with Tris-EDTA buffer (pH 7.4) to remove excess EDTA, and then suspended in a one-half volume of Tris-EDTA buffer and stored at 4 °C.
Escherichia coli DH5α was used as a model microorganism. E. coli cells were added at a concentration of 3 ⋅ 10 8 cells mL − 1 alginate-CaCO 3 solution and were encapsulated in 5.64 ± 1.72 ⋅ 10 5 AGMs (mean ± SD) in three independent experiments, which corresponded to 12.5% ± 5.4% of the total AGMs. Of the AGMs containing E. coli cells, 93.8% ± 8.8% harboured a single cell in the core (Supplementary Table 1 MDA of single-cell genomes within AGMs. AGMs (in 50 µL suspension) were washed with sterile water and collected by centrifugation. The lysis of bacterial cells and denaturation of double-stranded DNA within the AGMs were carried out at room temperature for 30 min with 50 µL of an alkaline solution, Buffer D2 of the REPLI-g UltraFast Mini Kit (Qiagen). Denaturation was stopped by neutralisation with the addition of the same volume of Stop solution from the kit, and the supernatant was then removed by centrifugation. Reaction Buffer (93.5 µL) containing 11 µL of REPLI-g UltraFast DNA Polymerase from the kit was added and the resulting mixture was incubated at 30 °C for 3 h. The reaction was stopped with one-tenth volume of 0.5 M EDTA, and the AGMs were washed three times with 200 µL of Tris-EDTA buffer.
All these reagents used in MDA and the AGM preparation were irradiated with ultraviolet light to degrade any contaminating DNA prior to use. Among 4.6 ± 2.9 ⋅ 10 5 AGMs prepared with E. coli cells, 8.9% ± 0.8% exhibited DNA ampli cation, which corresponded to 77.0% ± 27.2% of the AGMs containing E. coli cells (n = 3) (Supplementary Table 1). As shown in Fig. 2, ampli ed DNA lled the alginate sol core, whereas AGMs with alginate gel cores that were prepared without the solation step with EDTA or agarose gel beads containing no alginate core exhibited only limited or no DNA ampli cation by MDA.
Quality of single-cell genome sequences compared between MDA-in-AGM and a conventional method using FACS-MDA. The quality of SAGs was evaluated and compared with that obtained by a conventional method using a combination of FACS for single-cell isolation and MDA (FACS-MDA) on a microlitre scale.
AGMs containing ampli ed DNA that was stained with SYBR Green I were individually transferred to 0.2 mL PCR tubes with 29 µL of sterile water under an epi uorescence microscope equipped with a TransferMan NK2 micromanipulator (Eppendorf). Ampli ed DNA was released to water by heating the tubes at 60 °C for 5 min, and directly used for preparation of sequencing libraries using the QIAseq FX DNA Library Kit (Qiagen). No second round of MDA was necessary to obtain the libraries suitable for genome sequencing. Paired-end sequence reads were generated on the Illumina MiSeq platform with the Reagent Kit V3 (600 cycles), and a xed number of read pairs were randomly chosen and assembled de novo into contigs using SPAdes 3.13.0 22 after standard quality ltering. Contigs > 1 kb were used for the subsequent analyses. Several factors describing the genome sequence quality, including completeness and contamination rate, were evaluated using CheckM 23 and QUAST 24 , and compared between SAGs obtained using MDA-in-AGM and FACS-MDA, respectively.
The genome completeness and the total sequence length of E. coli SAGs obtained by MDA-in-AGM rapidly increased during the sequencing effort, and the average genome completeness reached 93.0% ± 4.7% (n = 10) for assemblies using 9 ⋅ 10 5 read pairs, while it was 33.6% ± 17.4% (n = 10) in FACS-MDA (Student's t-test, P < 0.01) (Fig. 3a). The genome coverage calculated by mapping unassembled reads to the reference E. coli DH5α genome sequence (BOCF01000000) was 98.0 % ± 1.7% in MDA-in-AGM (n = 10), which was much higher than 52.0% ± 13.8% in FACS-MDA (n = 10) for 9 ⋅ 10 5 read pairs (P < 0.01) (Supplementary Table 2). The number of contigs steeply decreased in MDA-in-AGM along the sequencing effort's depth probably owing to adequate assemblies of contigs, whereas it continued to increase in FACS-MDA (Fig. 3a). In addition, the ampli cation bias among genome regions, which is inherent to MDA, was much improved in MDA-in-AGM compared to FACS-MDA (Fig. 3b).
Single-cell genome analyses with a mock community and a natural sample. To evaluate the feasibility of MDA-in-AGM in a more realistic sample, we constructed a mock community comprising 15 cultured strains of human gut bacteria. The taxonomy and proportion of each bacterial strain are shown in Supplementary Table 3. SAGs from the mock community were obtained using either MDA-in-AGM or FACS-MDA in the same way as above (Supplementary Tables 4 and 5). The overall genome completeness of SAGs with < 5% contamination was signi cantly higher in MDA-in-AGM (68.0% ± 23.3%, n = 39) than in FACS-MDA (42.4% ± 20.5%, n = 36) for assemblies using 3 ⋅ 10 5 read pairs (P < 0.01) (Supplementary Tables 6 and 7). For example, SAGs of Bacteroides thetaiotaomicron and Parabacteroides distasonis exhibited clear differences between MDA-in-AGM and FACS-MDA in the genome completeness, total sequence length, number of contigs, and ampli cation bias, as seen in the E. coli SAGs (Fig. 3). The SAGs of most other strains also showed similar results ( Supplementary Fig. 3). Our results indicated that MDA-in-AGM is applicable to various bacterial species and improves genome completeness in both gram-positive and gram-negative bacteria, regardless of whether there is high or low GC content in the genomes (Fig. 4).
For SAGs with > 5% contamination, which would be caused either by contaminating extracellular DNA or by multiple cells being encapsulated in a single AGM, the contigs were binned into each bacterial species using metaWRAP 25 . As a result, 30 additional SAGs with < 5% contamination were obtained, and their average genome completeness and contamination rate were 74.1% ± 14.3% and 1.2% ± 1.1%, respectively (Supplementary Table 8).
Finally, we applied MDA-in-AGM to a natural environmental sample. We used the microbiota in the gut of the termite Reticulitermes speratus, which comprises several hundred bacterial species from diverse phyla 26 . The entire guts of ve worker termites were removed, the gut contents were suspended, and then bacterial cells were collected by centrifugation after digesting the extracellular DNA with DNase I and the cells were washed with sterile water as described previously 27 . The subsequent procedures were the same as those described above. We analysed 48 SAGs obtained by MDA-in-AGM and found high genome completeness (78.6% ± 18.2%) and a low contamination rate (1.0% ± 1.0%) ( Table 1). The SAGs were taxonomically classi ed using the Genome Taxonomy Database Tool Kit (GTDB-Tk) 28 and were found to be a liated with 13 bacterial classes, including known dominant groups in the termite gut, Spirochaetes, Bacteroidetes, Alphaproteobacteria, Betaproteobacteria, Deltaproteobacteria, Epsilonproteobacteria, and Clostridia 26 . In contrast, the genome completeness of SAGs obtained by FACS-MDA was only 37.7% ± 19.0% (n = 161). Although a different sequencing method, i.e., a combination of the Nextera XT DNA Library Prep Kit (Illumina) and the Illumina HiSeq 2500 platform (500 cycles), was used for the latter case, the number of reads (bases) used for assembly was much larger on average (551 Mb ± 64 Mb, n = 161) than in the analysis of SAGs obtained by MDA-in-AGM (256 Mb ± 49 Mb, n = 48) (Supplementary Table 9).

Discussion
MDA-in-AGM, the massively parallel MDAs in the picolitre-scale AGM cores, drastically improved the genome completeness of SAGs of E. coli, human gut bacteria in a mock community, and bacteria in the termite gut, compared to conventional microlitre scale FACS-MDA. Thus, the method is expected to contribute to revealing the metabolic capacity of yet-uncultured prokaryotes in environments.
It has been reported that the ampli cation bias in single-cell genomics can be suppressed by reducing the reaction volume of MDA using microfabrication techniques 9,12,13,15,16 . Although the uniformity of the reaction volumes of the AGMs developed in this study was lower than in previously developed methods using microchannels and nanowells 12,13,15,16 , MDA-in-AGM is much simpler and requires only inexpensive, commercially available reagents and standard molecular biology laboratory equipment, except for a micromanipulator. The AGM alginate sol core enhanced the yield of MDA compared to that obtained by MDA in gels (Fig. 2), which is probably attributable to the enhanced diffusivity of ampli ed DNA in the sol core. MDA-in-AGM therefore does not require a second round of MDA, which is often necessary in other methods using microfabrication techniques 15,16 . This direct use of the MDA product resulted in heightened genome completeness with suppressed ampli cation bias. Our MDA-in-AGM is applicable to both gram-positive and gram-negative bacteria and also to genomes with high GC content (Fig. 4). Additional advantages of the AGM are its physical stability and permeability for small molecules, which have made buffer exchange and physical handling much easier. To increase the uniformity and throughput of single-cell genomics using MDA-in-AGM, microchannels and molecular-barcoding 31 may be adapted, although these approaches decrease the simplicity of the procedure Recently, during our experiments were on going, MDA using a hollow-core hydrogel microcapsule consisting of a crosslinked PEG gel shell and a dextran sol core, a similar capsule structure to AGM, was reported 32 . Although ampli cation bias has not been evaluated, single-cell MDA in the core increases the amount of MDA product when compared with MDA in the gel bead. In addition, the shell of a hollow-core hydrogel microcapsule prevents genomic DNA leaking from the core after alkaline denaturation prior to MDA. However, near ultraviolet (UV) light (365 nm) used for photo chemical crosslink of the PEG shell damaged cells and DNA due to photooxidation 33 , a polymer shell of hydrogel microcapsule for single-cell MDA is required to gelate without the UV light. We selected agarose as the AGM shell because of thermally mild reversible gelation, the commercially availability, and stabilities in MDA reagents, especially alkaline, neutralisation buffer, or EDTA.
Any methodology relying on Poisson distribution to obtain AGMs or other reaction chambers containing single bacterial cells inevitably produces a proportion of chambers containing multiple bacterial cells 14 , and it is di cult to completely eliminate contaminating DNA in our experimental procedures, as in other previously developed methods 1 . Nevertheless, the application of a computer program for binning metagenomic fragments, such as metaWRAP 25 , enabled us to recover a considerable number of SAGs by eliminating contaminating sequences (Supplementary Table 8).
In recent years, numerous studies analysing metagenome-assembled genomes from environmental samples have been reported 34 . Although single-cell genomics has different advantages and potentially increases the quality of research in combination with metagenomics, for example, by revealing strainlevel heterogeneity 35 and by providing information on horizontally transferred genetic components 15 , it is di cult for many microbiologists to perform. MDA-in-AGM is a much easier and less expensive method, and in addition, only a small amount of sample is needed. Thus, this method is also suitable for tiny specimens, such as the intestinal tracts of small insects and protist cells with endo-and ectosymbiotic bacterial communities.
Additionally, AGM potentially enables cultivation of microorganisms in its solated hollow-core, using organic solvent-free washing buffer in the preparation. Actually, a new Dysgonomonas species has been successfully isolated from the termite gut using the AGM cultivation method and supplying nutrients from the outside of the shell (our unpublished data). AGM thus can be also an important tool to investigate cell-to-cell interactions of symbiotic bacteria using a co-culture of host-and-donor bacteria, and their genome analyses.
Reagents. Ultrapure DNase/RNase-Free Distilled Water (Thermo Fisher Scienti c) was used for all solutions prepared in-house. MDA was conducted in a biosafety cabinet while wearing sterile disposable gloves and sterile elbow covers. The surfaces of bottles and equipment were swiped with DNA Away (Thermo Fisher Scienti c). Solutions were autoclaved at 121°C for 15 min and further decontaminated by UV radiation at 7.2 mW/cm 2 overnight (12 h) in a laminar ow cabinet. CaCO 3 suspension and alginate solution. A suspension of uniform CaCO 3 nanoparticles (9.4% w/v, 100nm diameter, pH 10) was provided by Shiraishi Central Laboratories (Hyogo, Japan). The neness and uniformity of the particle size are important for the homogeneous gelation of alginate. To facilitate alginate gelation upon the addition of acetic acid, the conductivity and pH of the CaCO 3 suspension were lowered by washing the particles as follows: a suspension (10.6 mL) containing one gram of CaCO 3 particles was centrifuged at 2,000 ⋅ g for 3 min, and the pellet was re-suspended in 10 mL of water. After three washes, the resulting CaCO 3 suspension was autoclaved. The washing step decreased the conductivity and pH of the CaCO 3 suspension from 650 µS/cm and pH 10 to 82 µS/cm and pH 7, respectively. The CaCO 3 suspension was dispensed in screw cap tubes sealed in an aluminium bag at room temperature to prevent aggregation by freezing.
Alginate solution (5%, w/v) was prepared as follows: 2.5 g of sodium alginate (80-120 cP at 1% w/v, Wako, Osaka, Japan) was dissolved in 20 mL of water and brought to a volume of 40 mL with water in a 50-mL tube. The alginate suspension was mixed using a rotary shaker overnight until the alginate clods were completely dissolved. The alginate solution was then autoclaved and brought to a volume of 50 mL with water. Since autoclaving decreases the viscosity of alginate solution and affects the alginate core diameter, a consistent autoclave time (15 min) was used. The resulting alginate solution was stored at 4 °C. Since isostearyl alcohol (ISA; Kokyu Alcohol Kogyo Co., Ltd., Chiba, Japan) was easily removed from the emulsion with diethyl ether or 1-butanol, the alginate was emulsi ed with ISA containing 3% soy bean lectin (Wako) in the following experiments.
Observation of AGMs and agarose gel beads. To avoid the disruption of AGMs or agarose gel beads prepared in this study by a coverslip under microscopy, they were observed as follows. A silicone sheet with adhesive tape (20 ⋅ 500 ⋅ 0.5-mm thickness, Misumi, Tokyo, Japan) was cut into 20-mm squares, and the centres were punched with a square hole puncher (5 ⋅ 5 mm-tip, Kyoshin-elle, Tokyo, Japan). The silicone sheet was stuck on a glass slide (26 ⋅ 76 mm, Matsunami Glass Ind., Ltd., Osaka, Japan). To visualise Escherichia coli cells or DNA in AGMs, 50 mM Tris-HCl (pH 7.5) containing 0.5% SeaPlaque agarose, 0.1% p-phenylene diamine (Wako), and 10,000-fold diluted SYBR Green I (TaKaRa Bio Inc., Shiga, Japan) was prepared (designated as SYBR Green I solution). AGMs or agarose gel beads were suspended in 16 µL of SYBR Green I solution. The suspensions were individually dispensed into the holes in the silicone sheets. The holes were sealed with coverslips (18 ⋅ 18 mm, No. 1, Matsunami) and observed under an inverted uorescence microscope (IX71, Olympus, Tokyo, Japan) equipped with a CCD camera (BU-51LN, Bitran, Saitama, Japan).
Evaluation of PGO for agarose micro droplet gelation. SeaPlaque agarose (2%, 2 mL) was mixed with PGO or ISA, each containing 0.25% Span 80 (v/v) (20 mL, 35 °C), by vortexing for 1 min in a 50-mL conical tube, and the mixtures were cooled on ice for 1 h. PGO or ISA was removed by mixing with 10 mL of diethyl ether followed by centrifugation at 4 °C for 10 min using a swing rotor (3,000 ⋅ g, 5702R, Eppendorf, Hamburg, Germany). The addition of diethyl ether was required to decrease the density of the oil phase in order to separate the PGO from the agarose beads because PGO (0.997 g/mL) and agarose beads (approximately 1.0 g/mL) have very similar density. The agarose beads were further washed with 10 mL of TE containing 0.2% Tween 20 and 10 mL of 1-butanol, suspended in 5 mL TE, and ltrated with 300-µm cell strainers. The agarose aggregates on the strainers were collected into 2-mL dishes and observed. The ltrates were brought to 20 mL (10 volumes) of TE, and the supernatants were removed by centrifugation. The agarose beads were further washed twice with TE and collected into 2-mL tubes. After the volumes of the agarose beads were calculated from their weights, the agarose beads were suspended in a one-half volume of TE. When the agarose beads were observed as described above, their diameters were measured with ImageJ (https://imagej.nih.gov/ij/). The yields and the diameters of the agarose beads were statically analysed with R (https://www.r-project.org/) and RStudio (https://rstudio.com/). Preparation of Escherichia coli stock culture. Competent cells of E. coli strain DH5α were purchased from TaKaRa Bio Inc. and cultured on an LB plate (1% Bacto Tryptone, 1% NaCl, 0.5% Bacto Yeast extract, 1.5% Agar). A single colony was inoculated into LB medium (10 mL) in a 50-mL conical tube and cultured at 37°C in a rotary shaker (150 rpm) overnight. Two aliquots (2 mL) of the culture broth were inoculated into two LB (200 mL) media in 500-mL conical asks respectively and further cultivated at 37°C overnight in the rotary shaker. Cells of E. coli were harvested and washed twice in PBS (20 mL) by centrifugation. The cells were re-suspended in PBS (10 mL), and their density was determined with a cell-counting chamber.
Preparation of alginate cores containing E. coli cells. Cells of E. coli were encapsulated in alginate cores by the emulsi cation-internal gelation method as described previously 21 (Fig. 1b). E. coli cells from the stock culture (10 µL) were mixed with 990 µL of 2% sodium alginate solution containing 1% CaCO 3 and 50 mM acetate buffer (pH 7.0, 0.2-µm ltered, Sigma-Aldrich, St. Louis, MO). The mixture was emulsi ed with ISA containing 3% lecithin (9 mL) in a 50-mL tube by vortexing for 1 min. The emulsion was mixed with 2% acetic acid in ISA (0.1 mL each) by vortexing for 1 min with 10 repetitions. During this procedure, the pH of the mixture gradually decreased to 4.0 and the CaCO 3 nanoparticles were dissolved. The released calcium ions from CaCO 3 gelated alginate microdroplets in the emulsion. ISA was removed by mixing with diethyl ether and centrifugation, and the residual ISA was further removed from the alginate cores by mixing with 50 mM Tris-HCl buffer (pH 7.4) containing 0.2% (v/v) Tween 20 (Tris-Tween 20) (5 mL), followed by centrifugation, and then by repeating the procedure twice with 1-butanol (5 mL). The resulting alginate cores were suspended in one volume of 50 mM Tris-HCl (pH 7.4) and ltrated through a 100-µm cell strainer to remove large alginate aggregates. The ltrated alginate cores were further washed with Tris-HCl (pH 7.4) and collected in 2-mL microtubes. The alginate cores were weighed and stored at 4 °C after mixing with a one-half volume of Tris-HCl (pH 7.4). To adjust the number of E. coli cells in a single alginate core to less than two, 100-fold, 10-fold, and 1-fold dilutions of the E. coli stock were tested. The E. coli cells in these cores were counted, and a dilution that maximised the ratio of cores with a single E. coli cell was used in the following experiments.
Preparation of AGMs containing E. coli cells. The supernatant of the alginate core suspension (600 µL) was removed after centrifugation in a 50-mL tube. The residual alginate cores (400 µL) were incubated at 35 °C for 10 min and then mixed with the agarose solution as prepared above (2 mL, 35 °C). The mixture was emulsi ed with 0.25% (v/v) Span 80 in PGO (20 mL, 35 °C) by vortexing for 1 min (Fig. 1b). Then, the agarose was gelated by cooling. PGO was removed from the emulsion by washing as described above. The alginate gel core was solated by chelating calcium ions with the addition of a 1/10 volume of 0.5 M EDTA (pH 8.0, molecular biology grade, Thermo Fisher Scienti c). The AGMs were then suspended in one volume of TE buffer. Large debris in the suspension was removed through a 300 µm cell strainer, and the AGMs were further washed with TE and subsequently centrifuged. The AGMs (5 µL) were suspended in SYBR Green I solution prepared as above (100 µL), and the morphology, density, diameter, and number of encapsulated E. coli cells were observed under microscopy. The AGMs were stored at 4 °C. For long-term storage (more than a week), the AGMs were stored in 40% ethanol at −80 °C and washed three times in 10 volumes of water before use.
The alginate cores were mixed with 46 mM rhodamine 123 (54 µL), 0.5 M EDAC (50 µL), and 1 M NHS (50 µL), and 50 mM MES (pH 6.0) (646 µL). The mixture was left to react in the dark at room temperature with shaking for 16 h. The rhodamine 123-labelled alginate cores were washed three times with 10 volumes of 50 mM MES buffer and the cores were used for AGM preparation. AGMs containing rhodamine 123-labelled core were heated at 65 °C for 5 min in the presence of 50 mM EDTA or 50 mM CaCl 2 . Residual cores were observed as mentioned above.
Effect of alginate core gelation and solation on MDA. MDA was performed using the REPLI-g UltraFast Mini Kit (Qiagen, Hilden, Germany). Aliquots (50 µL) of the AGMs containing E. coli cells before solation of the alginate cores (i.e., alginate gel cores) were mixed with one volume of cell lysis solutions containing 400 mM KOH with or without 10 mM EDTA, then incubated at 4 °C for 1 h. Next, they were neutralised with 100 µL of Stop buffer from the REPLI-g UltraFast Mini Kit at 4 °C for 16 h. The AGMs were washed with 1 mL of water and were brought to a volume of 50 µL with water. Reaction buffer (32 µL) containing 2 µL of phi29 DNA polymerase from the kit was added to both sets of AGMs. The reaction mixtures were incubated at 30 °C for 6 h, and the reactions were stopped with 400 µL of 10 mM EDTA. After washing three times with 0.5 mL of water, each set of AGMs was brought to a volume of 100 µL with water. AGMs without E. coli cells were also subjected to the same procedure as a negative control. These samples were observed for DNA ampli cation under uorescence microscopy.
MDA-in-AGM. The genomic DNA of E. coli encapsulated within the AGMs was ampli ed by MDA (MDA-in-AGM). A 50 µL aliquot of AGMs was centrifuged, and the collected AGMs were washed with 200 µL water. The cell lysis and DNA denaturation were carried out with Buffer D2 (50 µL) of the REPLI-g UltraFast Mini Kit at room temperature for 30 min. After denaturation was stopped with the Stop buffer, the supernatant was removed by centrifugation. The Reaction buffer (93.5 µL) containing 11 µL phi29 DNA polymerase from the kit was added and incubated at 30 °C for 3 h. The reaction was stopped with 1/10 volume of 0.5 M EDTA, and the AGMs were washed with 200 µL of TE buffer three times. The washed AGMs were stored at 4 °C.
Single-cell MDA using a MoFlo XDP uorescence-activated cell sorter (Beckman Coulter Inc., Brea, CA) was performed as described in previous studies as a control 27 .
Mock community of human gut bacteria. A mock community of human gut bacteria was prepared from 15 isolates (Supplementary Table 3 Faecalibacterium prausnitzii JCM 31915, and Flavonifractor plautii JCM 32125. These were obtained from the Japan Collection of Microorganisms (JCM, https://jcm.brc.riken.jp/en/). The total genomic DNA of the mock bacterial community was extracted using the DNeasy Ultra Clean Microbial Kit (Qiagen). Amplicon sequencing of 16S rRNA genes in the mock community was performed by preparing a library with the Nextera XT Index Kit 96 indexes (Illumina, Inc, San Diego, CA) and the MiSeq platform with Reagent Kit V3 (600 cycles). The bacterial composition of the mock community was estimated on the basis of the results analysed in QIIME2 37 .
Termite gut bacteria. Specimens of the wood-feeding termite Reticulitermes speratus were collected at Tsukuba in Ibaraki Prefecture, Japan. The guts of ve worker termites were removed, and the gut contents were suspended in solution U, which consisted of 9.2 mM NaHCO 3 , 5.1 mM trisodium citrate dihydrate, 13 mM KH 2 PO 4 , 37 mM NaCl, 0.75 mM CaCl 2 , and 0.4 mM MgSO 4 . The extracellular DNA and protist DNA were digested using DNase I (Promega, Madison, WI), and bacterial cells were collected by centrifugation and washed with water. The detailed procedure has been described previously 27 .
Genome sequencing and bioinformatics. AGMs containing ampli ed DNA, detected with SYBR Green I, were transferred individually to PCR tubes, under an inverted uorescence microscope equipped with a micromanipulator (TransferMan NK2, Eppendorf, Hamburg, Germany). Water (29 µL) was added to each PCR tube with a single AGM and incubated at 60 °C for 5 min, to release DNA by solating the agarose shell. Sequencing libraries were prepared using the QIAseq FX DNA Library Kit (Qiagen). When the sequencing libraries were constructed using adaptors with the concentration described in the manufacturer's instructions, a large number of concatenated adaptor sequences were formed. To suppress the generation of the arti cial sequences, the concentration of adaptor sequences was adjusted to one-tenth of that in the manufacturer's instructions. Genome sequencing was performed on the MiSeq platform with Reagent Kit V3 (600 cycles). Sequence libraries for single cells isolated from the termite gut microbiota using FACS were prepared using the Nextera XT DNA Library Prep Kit (Illumina), and analysed on the Illumina HiSeq 2500 platform (500 cycles).