Roundup

All the projects included in this study are publicly available. A short
description of the chosen configurations and references are given in the
following. We used the default parameters unless mentioned otherwise.
RoundUp: RoundUp can be downloaded from https://rodeo.med.harvard.edu/tools/roundup/. It is available
with different parameter settings to tune for the desired sensitivity. In this
comparison we included the strictest parameter set (also default settings), i.e.
Blast E-value cutoff 10⁻²⁰ and divergence cutoff 0.2.
Inparanoid: Inparanoid is available from http://inparanoid.sbc.su.se. We used the release 6.0 from Aug
2007 including 35 species.
Ensembl Compara: The orthology predictions from Ensembl were
obtained from the Compara database version 47, which is available from http://oct2007.archive.ensembl.org/.
COG,KOG: Cluster of Orthologous Groups and its eukaryotic equivalent
are available from http://www.ncbi.nlm.nih.gov/COG/. We used the versions from Mar
2003 and Jul 2003 respectively.
OrthoMCL: We obtained the version from Sep 2006 of OrthoMCL from
http://orthomcl.cbil.upenn.edu/.
Homologene: Homologene is available from the NCBI webpage www.ncbi.nlm.nih.gov/HomoloGene/. For this comparison, we used
built 58 from Nov 2007.
EggNOG: EggNOG is available from http://eggnog.embl.de/. We
used the data from Oct 2007 including 373 species.
OMA: OMA is available in various formats on http://www.omabrowser.org. We used the the data from Nov 2007
including 550 species. OMA infer orthology at the level of pairs of sequences
(“OMA Pairwise”), from which it also computes groups of
orthologs (“OMA Group”). Both type of predictions are
included in the comparisons.
BBH: The typical Bidirectional Best Hit implementation uses BLAST
for aligning the protein sequences. We used the more accurate algorithm from
Smith and Waterman [35] (link) for the alignment with the same scoring
threshold as used by the OMA algorithm for the all-against-all step.
RSD: Reciprocal Smallest Distance orthology relations are computed
using ML distance estimates from pairwise alignments having significant
alignment scores (Dayhoff score >217, the cut-off used by OMA as
well)

According to interviews with smallholders, conducted by Euler et al. [unpublished data], the rubber and oil palm plantations in the clay Acrisol soil were planted after clearing and burning the previous forest or logged forest. In the loam Acrisol soil, oil palm plantations were established after clearing and burning the previous jungle rubber whereas the rubber plantations were established from previously logged forest. Based on our interviews, only the oil palm plantations were fertilized during our study year, 2013, while the rubber plantations were not. Oil palm plantations in the clay Acrisol soil were fertilized once in the rainy season (October to March), and in the loam Acrisol soil, these were fertilized once in the rainy season and once in the dry season (April to September). The most commonly used fertilizers were NPK complete fertilizer (i.e. Phonska, Mahkota), potassium chloride (KCl) and urea (CO(NH₂)₂). Fertilizer additions to the oil palm plantations ranged from 300 kg NPK-fertilizer ha^-1 year^-1 (for those plantations that were fertilized once) to 550 kg NPK-fertilizer ha^-1 year^-1 (for those plantations that were fertilized twice). In terms of unit nutrient element added, these rates were equivalent to 48–88 kg N ha^-1 year^-1, 21–38 kg P ha^-1 year^-1 and 40–73 kg K ha^-1 year^-1. Additionally, three of the smallholders applied 157 kg K-KCl ha^-1 year^-1 and 143 kg Cl-KCl ha^-1 year^-1 and two of the smallholders applied 138 kg urea-N ha^-1 year^-1. One of the smallholders also applied lime in 2013 at an average rate of 200 kg dolomite ha^-1 year^-1. Both manual and chemical weeding took place throughout the year at the rubber and oil palm plantations. The most commonly used herbicides were Gramoxone and Roundup; these were applied at an average rate of 2 to 5 L herbicide ha^-1 year^-1 [Euler et al. unpublished data].

The MultiParanoid algorithm [10] (link) is an extension of the
graph-based InParanoid clustering algorithm [11] (link), [42] (link) for
identifying orthologs and inparalogs across multiple species.
InParanoid uses bi-directional best BLAST [9] (link), [43] (link) to
identify putative orthologs and a clustering algorithm to identify their
inparalogs. To do so, InParanoid assumes that any sequences
from the same species that are more similar to the predicted ortholog than to
any sequence from other species are inparalogs [11] (link), [42] (link).
MultiParanoid generates multi-species orthogroups by
merging all pairwise InParanoid predictions, while minimizing
the number of internal conflicts. Furthermore, the algorithm uses a
‘cut-off’ parameter based on the distance of candidate inparalogs to
the predicted target ortholog to filter out weakly supported candidates.
MultiParanoid was obtained from http://multiparanoid.sbc.su.se and InParanoid(version 3beta) was obtained upon request from inparanoid@sbc.su.se.
The OrthoMCL algorithm also builds upon the InParanoidalgorithm [11] (link),
[42] (link) by
using the Markov Cluster (MCL) algorithm for predicting orthogroups across
multiple species based on their sequence similarity information [3] (link). The algorithm
uses an ‘inflation rate’ parameter, to regulate the
‘tightness’ of the predicted orthogroups. OrthoMCL (version
1.4) was obtained from http://orthomcl.org/common/downloads/software/v1.4/.
The Reciprocal Best Hit (RBH) algorithm [4] (link), [6] (link), [12] (link), [13] (link) relies on BLAST [9] (link), [43] (link) to
identify pairwise orthologs between two species. According to the RBH algorithm,
two proteins X and Y from species
x and y, respectively, are considered
orthologs if protein X is the best BLAST hit for protein
Y and protein Y is the best BLAST hit for
protein X. We integrated a ‘filtering’ parameter
r that enabled us to avoid constructing orthogroups that
contained distant homologs by considering the degree by which the two proteins
differed in sequence length or BLAST alignment [44] (link), [45] (link). Thus, putative
orthogroups are retained if:
From the above equation, it follows that r values close to 1 are
likely to filter out a larger number of putative orthologs, whereas
r values close to 0 are likely to include all putative
orthologs. The default mode of the algorithm does not use the filtering
parameter r.
The Reciprocal Smallest Distance (RSD) algorithm [14] (link) generates global sequence
alignments for a small number of top BLAST hits against a query gene
X from species x. RSD then calculates the
maximum likelihood evolutionary distance between X and its top
BLAST hits, identifying the gene with the smallest evolutionary distance from
X (e.g., gene Y from species
y). If the RSD search using gene Y from
species y as the query also identifies gene Xfrom species x as its closest relative, then proteins
X and Y are considered orthologs [14] (link), [15] (link). In RSD,
the user can modify the shape parameter a of the gamma
distribution, a key determinant of the estimated evolutionary distance between
genes. The RSD algorithm was obtained from http://roundup.hms.harvard.edu/site/.

Roundup (commercial grade, 41%, containing 360 g/L glyphosate) was diluted to prepare a 1000 mg/L glyphosate stock solution with a natural pH of 4.9–5.9. Calcium peroxide (CaO₂, 65%), ammonium molybdate (H₈MoN₂O₄, 99%), and ascorbic acid (C₆H₈O₆, 99%) was purchased from Alfa Aesar, Ward Hill, MA, USA. Calcium chloride (CaCl₂, ≥99.5%), hydrogen peroxide (H₂O₂, 30%), ammonia solution (NH₄OH, 25 wt%), ferrous sulfate (FeSO₄∙7H₂O, ≥98%), and ethyl alcohol (C₂H₅OH, 95%) were supplied from R&M Chemicals Sdn. Bhd. (Semenyih, Malaysia). Polyethylene glycol 200 [HO(C₂H₄O)nH, PEG 200], antimony potassium tartrate [K₂Sb₂(C₄H₂O₆)₂, 99%], sodium sulfite (Na₂SO₃, ≥98%, 2.0 M), sodium hydroxide (NaOH), and sulfuric acid (H₂SO₄, 95%) were procured from Sigma Aldrich (St Louis, MO, USA). Cerium (IV) sulfate tetrahydrate (CeO₈S₂∙4H₂O, ≥98%) was purchased from Acros Organics, Geel, Belgium. All the chemical reagents used in this study were of analytical reagent grade. Distilled water was used throughout the experiments to prepare the solutions and the stock solutions were freshly prepared every week. The pH of the solutions was adjusted by the addition of 0.1 M H₂SO₄ solution and 0.1 M NaOH solution.

The field trial for glyphosate tolerance of the transgenic maize CVC-1 followed a split block design (plots of twin rows spaced ~0.6 m apart) with three replications. Glyphosate isopropylamine salt (41%, Roundup, Monsanto, St. Louis, MO, USA) diluted at 1:100 (10 mL/L) was sprayed onto the CVC-1 and non-transgenic control Zheng 58 plants at the 4–5 leaf stage with the rate of 45 mL/m². The glyphosate tolerance was evaluated at two weeks after treatment.

The study was carried out in the Jena Experiment, a large-scale grassland diversity experiment in the floodplain of the Saale River near the city of Jena Thuringia, Germany; 50°57´N, 11°35´E^{39 (link),40 (link)}. In spring 2002, 82 experimental grassland plots of 20 × 20 m were established. Plots are arranged in four blocks to account for changes in soil characteristics with increasing distance from the river, typical for a flood plain. Specifically, soil texture in the upper soil ranges from sandy loam to silty clay with increasing distance from the river. Sand content declines from 40% near the river to 7% at the furthest plot, while silt content increases from 44 to 69%, respectively. The clay content ranges from 16 to 24%, but is not related to distance from the river. The experimental field site of this study was an arable field with mineral fertilizer input to grow wheat and vegetables for about 40 years before the establishment of the Jena grassland experiment. After the last harvest in autumn 2000 the field was ploughed and kept fallow throughout 2001. In preparation for the experiment, and in order to reduce the weed pressure the field was harrowed bimonthly (June, August, October) and treated with Glyphosate (N-(Phosphonomethyl)-glycin, Roundup) in July 2001. In spring 2002 the field was harrowed before the plots were established^{39 (link)}.
Plant communities of different plant species richness (PSR) were established building a gradient from monocultures to 60 plant species mixtures (diversity levels: 1, 2, 4, 8, 16 and 60 species). The plant diversity levels were replicated 16 times, except for the 16 (14 replicates) and 60 (four replicates) plant species mixtures (Fig. 1).Figure 1

The 10-ha site was used as arable land until 2000, and was converted to an experimental grassland in 2002. For this, the area was completely tilled and experimental plots of different plant diversity were sown, separated from each other by paths (left photo: field site in 2002; right photo: field site in 2016). The plant community composition of each plot was randomly selected out of a species pool of 60 grassland species. Each diversity level is replicated 16 times, except for 16 and 60-species mixtures, which have 14 and 4 replicates, respectively. After the plots became established, colonization by above- and belowground fauna began. Left photo by K. Kübler, copyright of right photo by the Jena Experiment.

The species were randomly chosen from a pool of 60 Arrhenatherion grassland plant species and from four functional groups, namely grasses, legumes, small herbs, and tall herbs (Table S1). Functional group classification was based on morphological, phenological, and physiological traits^{39 (link)}. For this study, the PSR levels 1, 4, 16 and 60 were used, resulting in fifty plots. Experimental plots are weeded manually two to three times a year to maintain the target plant community composition. Weeds were mainly species of the species pool that are not sown on the respective plot. The plots are mowed and the mowed plant material is removed twice a year in June and September, but not fertilized, which is typical for extensively managed hay meadows in Central Europe. In 2009 the original plot size was reduced to 105 m^{240 (link)}.