Soil metabarcoding helps identify recalcitrant taxa from chaparral seed banks

Evaluating seed bank composition by germinating seeds from soil cores is a common technique used in ecological studies to identify the plant biodiversity reservoir of a site. However, failure to meet required germination cues or to correctly detect uncommon species are major hurdles to creating a comprehensive plant list from the soil seed bank. Identifying plant species from genetic material within the soil environment (eDNA or eRNA) via metabarcoding offers a potential solution that has not yet been widely utilized at least in part because interpretations of results are not always straightforward. To address this issue, we first assessed extraction and amplification protocols in a series of proof-of-concept experiments where we controlled the soil seed bank and soil environments. We found that barcodes from DNA were more consistently amplified than from RNA and adding a germination stimulant, such as water, did not significantly influence sequencing yield. We then compared our molecular methods to traditional methods of germinating seed banks using soil samples collected from a degraded chaparral site in southern California where germinating native plants ex situ is challenging. We found that the rbcL barcode identified the largest number of plant families while the ITS2 barcode identified the most plant genera. Species that are traditionally challenging to germinate, such as fire-followers and hemiparasitic plants, were among those identified by metabarcoding but not by traditional methods. Pairing molecular tools with ecological site familiarity will make the species identification process more efficient, complete, and especially conducive for identifying the recalcitrant species of soil seed banks.

Methods
MATERIALS and METHODS

Sequencing known seed bank (Proof-of-concept)

We tested the accuracy and efficiency of metabarcoding methods across a combination of different soil complexities, added plant species, and soil preparations, for a total of 12 treatment groups with five replicates each. We first tested how efficiently S. leucophylla seeds could be sequenced from the simplest soil matrix - sterilized sand - under three germination treatments prior to nucleotide extraction, which included no treatment, 24 hour incubation with deionized (DI) water, and 24 hour incubation with gibberellic acid (GA). We expected stimulation for germination with water or GA would increase RNA transcription and yield. S. leucophylla seeds were chosen as our control species since their seeds had been recently collected from the wild and were abundantly available for testing.

Next, we tested how accurately and efficiently S. leucophylla seeds could be sequenced with metabarcoding from sterilized field soils, which were collected from Piru, CA, a chaparral ecosystem undergoing restoration (site history described in next section). The site occurs on a mix of ~30% Lodo and ~25% Botella soils, plus similar soil series, and are associated with ~25% rock outcrops occurring on 30-60% slopes (Web Soil Survey). The soils were collected from a depth of 12 cm. The sand and field soil (<100 g) were both sterilized at 125 ℃ in the autoclave for 30 minutes.

We also tested whether metabarcoding would preferentially sequence S. leucophylla seeds over non-seed material by adding tissues of air-dried Bromus diandrus roots to some of the samples. Bromus diandrus, along with other non-native Bromus species, is a common non-native annual grass with a fibrous root system, generally found near the soil surface, that can outcompete shrub seedlings (Park, 2020) and invade chaparral stands (Haidinger and Keeley 1993). Bromus diandrus roots were freshly collected from senesced stands in July 2019 from Isla Vista, CA, and air-dried for three weeks on the lab bench. Large fresh quantities of B. diandrus were readily available and suitable to compare against S. leucophylla since they are taxonomically distant (Poaceae vs. Lamiaceae). Hence, we tested four combinations of added plant materials - S. leucophylla seeds only, B. diandrus roots only, S. leucophylla seeds with B. diandrus roots, and no plant material. These four treatment groups were tested with and without incubation in 24 hours of DI water. Finally, we sequenced the field soil without sterilizing (only air-dried and no incubation treatment) to assess potential sources of contamination. All 12 treatments were analyzed by both DNA and RNA and subsequently amplified for three barcode regions: rbcL (ribulose-bisphosphate carboxylase), ITS2 (second spacer of the internal transcribed region of the nuclear ribosome), and trnL (a non-coding chloroplast intron). We assessed accuracy based on identification of S. leucophylla reads, if the seeds were added, and efficiency based on number of replicates that were successfully amplified and sequenced.

Collecting chaparral soils

Soil samples were collected in June 2019 from two degraded chaparral stands as part of an ongoing restoration study at the Los Padres National Forest in Piru, California. One stand (Stand 1: Sites A and B) burned in a 2007 wildfire while the other (Stand 2: Sites C, D, and E) burned in 2003 and 2007. At both stands, native chaparral shrubs have declined in diversity and abundance and the canopy is co-dominated by a native, early-successional shrub, Malacothamnus fasciculatus, and various non-native grasses and forbs (e.g., Avena barbata, Bromus madritensis, Centaurea melitensis). Native shrubs Salvia leucophylla, Rhus ovata, Baccharis pilularis, Artemisia californica are also present at the northern Stand 1. Stands 1 and 2 were subdivided into a total of ten sites (approx. 350 m2 per site). From each site, 4.5 cm diameter soil cores were collected from 0-4 cm (top soil) or 4-12 cm (bottom soil) for a total of two soil samples per site. Deeper soil cores were predicted to include older more-native seed banks whereas the shallower soil cores were predicted to capture more non-native species due to recent history of introduction. Soil samples were transported to UC Santa Barbara where they were air dried for three weeks before being sieved through 4 mm and 2 mm sieves to remove rocks and to homogenize the soil sample, and then either prepared for the germination study or stored at 3℃ (35.6°F) prior to metabarcode sequencing.

Based on the results from the completed germination study (methods described below), five sites (Sites A, B, C, D, E) with the greatest and lowest plant species richness were chosen a posteriori for metabarcode sequencing in order to test correlations in species richness between the two methods. For each soil sample, two subsamples of 20 g of soil were placed in a 15 cm diameter Petri dish and incubated in a growth chamber at 35℃   for 18 hours to simulate a very hot day. Subsequently, one Petri dish of soil was incubated in DI water for 24 hours (+ DI), while the other received no treatment.

Nucleic Extraction, Amplification, and Sequencing

All soil samples were ground in liquid nitrogen with a mortar and pestle and stored at -80°C prior to isolation of DNA or RNA. The Qiagen RNeasy PowerSoil Total RNA Kit and the Qiagen RNeasy PowerSoil DNA Elution Kit (Qiagen, Maryland, United States) were used to simultaneously isolate DNA and RNA from the same replicate samples. The extracted nucleotides were quantified with dsDNA or RNA Qubit Assays (Qiagen, Maryland, United States).

All molecular work was performed with sterile techniques in a biosafety cabinet. Three ng of DNA and RNA were used for amplification of the target barcode regions. We tested the barcode regions rbcL, ITS2, and trnL for the proof-of-concept study, but we did not amplify the trnL region for the field study since trnL is considerably shorter, used less in other studies, and therefore unlikely to provide additional taxonomic resolution to the study. RNA samples were reverse-transcribed and amplified with the Qiagen OneStep RT-PCR Kit (Qiagen, Maryland, United States). Cleaned amplified products were standardized to 5 nM and pooled. The pooled libraries were sequenced with paired-end sequencing (2 × 250 bp) on an Illumina MiSeq sequencer at the California Nanosystems Institute at the University of California - Santa Barbara, CA, with a 15% spike-in of PhiX. 

Sequences were filtered and denoised into amplicon sequence variants (ASVs), which target variation within the finest level possible within a given barcode region (Callahan et al. 2017). ASVs were generated using a custom dada2 pipeline by combining identical reads, or nucleotide sequences (Callahan et al. 2016). Plant taxonomic identities were manually checked with BLAST (Altschul et al. 1990). The samples were sequenced across three separate MiSeq libraries with some of the same samples sequenced twice (Accession PRJNA1126262).