Shoot and root insect herbivory change the plant rhizosphere microbiome and affects cabbage–insect interactions through plant–soil feedback

Summary Plant–soil feedback (PSF) may influence plant–insect interactions. Although plant defense differs between shoot and root tissues, few studies have examined root‐feeding insect herbivores in a PSF context. We examined here how plant growth and resistance against root‐feeding Delia radicum larvae was influenced by PSF. We conditioned soil with cabbage plants that were infested with herbivores that affect D. radicum through plant‐mediated effects: leaf‐feeding Plutella xylostella caterpillars and Brevicoryne brassicae aphids, root‐feeding D. radicum larvae, and/or added rhizobacterium Pseudomonas simiae WCS417r. We analyzed the rhizosphere microbial community, and in a second set of conspecific plants exposed to conditioned soil, we assessed growth, expression of defense‐related genes, and D. radicum performance. The rhizosphere microbiome differed mainly between shoot and root herbivory treatments. Addition of Pseudomonas simiae did not influence rhizosphere microbiome composition. Plant shoot biomass, gene expression, and plant resistance against D. radicum larvae was affected by PSF in a treatment‐specific manner. Soil conditioning overall reduced plant shoot biomass, Pseudomonas simiae‐amended soil causing the largest growth reduction. In conclusion, shoot and root insect herbivores alter the rhizosphere microbiome differently, with consequences for growth and resistance of plants subsequently exposed to conditioned soil.


Introduction
Plants are members of complex communities, in which they interact with a plethora of other organisms such as insects and microbes (van der Heijden et al., 2008;Berendsen et al., 2012;Stam et al., 2014). Plant responses to the biotic or abiotic environment can affect many of these interactions and can shape the roots and their associated microbiome (Sasse et al., 2018;Stringlis et al., 2019;Wang et al., 2019;Delory et al., 2020;Kostenko & Bezemer, 2020). Shaping of the root-associated microbial community may impact future plants growing in the same soil. The net effect of all biotic and abiotic properties of soil conditioned by plants that previously grew in it on plants subsequently growing in the same soil is called plant-soil feedback (PSF) Kaplan et al., 2018;Bennett & Klironomos, 2019). PSF can affect the performance of plants positively (Kulmatiski et al., 2017) or negatively (Ma et al., 2017;Lekberg et al., 2018). Although an increasing number of studies focuses on the effects of PSF on plant growth, its effect on plant resistance is less explored, in particular plant defense against belowground insect herbivores (Hu et al., 2018).
Plants possess interconnected hormonal signaling pathways that respond to insect herbivory in both shoot and root tissue. Plant defenses to insect herbivores are mainly regulated by the phytohormones jasmonic acid (JA) and salicylic acid (SA), but also other plant hormones such as abscisic acid (ABA) and ethylene (ET) are involved (Erb et al., 2012b;Verma et al., 2016). Plants respond to herbivory by upregulating primarily JAor SA-associated signaling depending on the attacking insect species. Chewing insects generally induce JA production, whereas phloem-feeding insects induce SA biosynthesis (Erb et al., 2012b;Stam et al., 2014).
There are differences in plant defense and phytohormone regulation between plant shoot and root tissues . For instance, levels of the defensive glucosinolates in brassicaceous plants differ substantially between shoots and roots (Tsunoda et al., 2017). In terms of phytohormonal signaling, JA is thought to be less inducible in roots compared to shoots (Erb et al., 2012a;Tytgat et al., 2013), but increased levels do occur after herbivore attack (Erb et al., 2009;Lu et al., 2015;Karssemeijer et al., 2020), and SA may serve different functions in root and shoot tissues (Erb et al., 2012a;Lu et al., 2015).
Plant hormones do not only govern plant defense, they also influence root exudates and therefore consequently the microbiome around the plant root (Carvalhais et al., 2015;Eichmann et al., 2021). Therefore it is not surprising that feeding by shoot and root herbivores induces microbiome alterations, through altered plant root exudation (Dawson et al., 2004;Kostenko et al., 2012;Kim et al., 2016;Kong et al., 2016;Ourry et al., 2018;Friman et al., 2021b). Herbivores can also influence the soil microbiome directly, for instance through caterpillar frass or aphid honeydew that mixes with soil (Frost & Hunter, 2004). The resulting changes in microbiome and soil properties can affect the chemical composition of subsequently growing plants (Meiners et al., 2017) which in turn can affect herbivorous insects (Kostenko et al., 2012). In this manner, phytohormone-mediated signaling pathways and by extension plant defense relying on types and levels of secondary metabolites, can be modified by PSF (Ma et al., 2017;Hu et al., 2018;Zhu et al., 2018;Bennett & Klironomos, 2019). For instance, caterpillars of the cabbage moth Mamestra brassicae showed decreased performance when feeding on plants grown in soil conditioned by plants infested by root-feeding wireworms Agriotes lineatus, compared to caterpillars feeding on plants grown in soil conditioned by caterpillarinfested plants (Kostenko et al., 2012). Thus, herbivores can affect plant defense through PSF, and the identity of the herbivore species in the conditioning phase may be an important factor. Because plants respond differently to insect herbivores depending on their feeding guild and feeding site, it is plausible that different types of insects cause different changes to the plantassociated microbe community. Whether the underlying microbial community changes are comparable between insect feeding guilds and feeding location has received little attention so far.
Some root-associated bacteria are known to boost plant growth, and consequently have been coined plant-growthpromoting rhizobacteria (PGPR). A number of these PGPR can induce systemic resistance (ISR) in the plant, a mechanism that enhances resistance against a range of plant attackers (Pineda et al., 2010;Pieterse et al., 2014;Friman et al., 2021b). These ISR-inducing bacteria can mediate PSF. Arabidopsis thaliana recruited an assemblage of ISR-inducing microorganisms after infection with downy mildew, Hyaloperonospora arabidopsidis, which subsequently increased plant resistance of plants grown in the same soil against the same pathogen . Although plant-growth-promoting microbes are known to modulate plant resistance against insects (Pineda et al., 2010), it remains to be investigated how these rhizobacteria affect plant defense against insects in plant conspecifics growing in the same soil.
Here, we studied how shoot-and root-feeding insect herbivores and beneficial rhizobacteria affect the rhizosphere microbiome, and how these differences through PSF affect plant growth and defense against a root herbivore in plants subsequently growing in the same soil. We conditioned soil by growing Brassica oleracea plants induced by either root-chewing Delia radicum, leaf-chewing Plutella xylostella, phloem-feeding Brevicoryne brassicae, or by adding growth-promoting and ISRinducing PGPR Pseudomonas simiae WCS417r to the soil. These inducers have previously been tested for their influence on D. radicum performance through plant-mediated effects, where Plutella xylostella negatively influenced D. radicum performance, Brevicoryne brassicae had no effect (Karssemeijer et al., 2020), and Pseudomonas simiae positively affected the insect (Friman et al., 2021a). After removal of the conditioning plants and insects, we used a mixture of sterilized and conditioned soil to grow a consecutive set of Brassica oleracea plants, for which we assessed growth, defense-related gene expression, and resistance against the root herbivore D. radicum. We aimed to elucidate the effect of the inducers on the rhizosphere microbial community, and how these changes may moderate plant-mediated interactions between biotic inducers. We hypothesized that the induction by leaf-chewing, root-chewing, and phloem-feeding insect herbivores would have distinct effects on the rhizosphere microbiome due to their respective induction of different phytohormones, and that plants grown in these soils would differ in resistance against D. radicum. We expected that Pseudomonas simiae would increase plant growth in the feedback phase, and increase D. radicum performance.

Plant growth conditions
Our study system consisted of Brassica oleracea, a globally important cultivated crop plant. Brassica oleracea var. gemmifera cv. "Cyrus" seeds (Syngenta Seeds, Enkhuizen, The Netherlands) were germinated in a seeding tray with seedling soil in a glasshouse with 21 AE 3°C and 16 AE 3°C day and night temperatures respectively. Natural daylight was supplemented with 400 W metal halide lamps (200 µmol m À2 s À1 photosynthetically active radiation) when photosynthetic active radiation (PAR) dropped below 400 lmol m À2 s À1 , in a 16 h : 8 h, light : dark cycle. After 3 d, plants were transplanted to 1 l pots containing potting soil and grown in glasshouse conditions for 3 wk with identical settings as earlier at 60 AE 10% relative humidity (RH). Plants were watered three times per week from the bottom until the soil was moist. Plants were additionally fertilized twice per week with 50 ml of Hyponex solution (nitrogen, phosphorus and potassium (NPK) = 7 : 6 : 19, electrical conductivity = 1.6). As the staring soil can be important in PSF experiments (French et al., 2021), we used the same batch of soil throughout the experiment. Seedling and potting soil from the conditioning phase was bagged and stored at 4°C for use in the feedback phase (Fig. 1).

Insect rearing
Worldwide, the most important belowground feeding insect on Brassica oleracea is the specialist chewer cabbage root fly D. radicum L. (Diptera: Anthomyiidae). The female flies deposit a cluster of eggs in the soil near the plant stem base. After hatching, the larvae feed in the primary root. The larvae leave the root to pupate in the soil and emerge later as adult flies. Experimental D. radicum larvae were reared on rutabaga roots (Brassica napus var. napobrassica) at 22 AE 1°C, 70% RH and a 16 h : 8 h, light : dark cycle. The flies were caught in Zeewolde in the Netherlands in 2013 and reared in the laboratory since. Adult flies were fed honey and a 1 : 1 : 1 mix of milk powder, sugar and yeast flakes. Plutella xylostella L. (Lepidoptera: Plutellidae) were reared on Brassica oleracea var. gemmifera. Second instar larvae were used in this experiment. Brevicoryne brassicae L. (Hemiptera: Aphididae) were reared on Brassica oleracea var. gemmifera, and wingless adults were used as inducers in the experiment. These insects were reared at 22 AE 2°C, 70% RH and a 16 h : 8 h, light : dark cycle.

Conditioning phase: induction with insects and rhizobacteria
After 3 wk of growth, plants were infested with insects and/or exposed to Pseudomonas simiae inoculum. Each treatment had 24 replicates divided over four trays with six plants placed in individual pots on saucers, to prevent sharing water between plants. Treatments were D. radicum, D. radicum plus Pseudomonas simiae WCS417r, Plutella xylostella, Brevicoryne brassicae, Pseudomonas simiae WCS417r alone and control plants (Fig. 1). Fig. 1 Overview of the experimental design. Soil was conditioned by Brassica oleracea plants that, after 3 wk of growth, were induced by six treatments represented by colored boxes in the conditioning phase. The treatments were uninfested plants (no herbivores, green), Brevicoryne brassicae (yellow), Plutella xylostella (blue), Delia radicum (orange), Pseudomonas simiae WCS417r (purple), D. radicum and Pseudomonas simiae WCS417r (pink). Arrows in the leftmost panel indicate herbivore feeding locations (shoot or root). Additionally, soil was stored at 4°C to be used as nonconditioned treatment in the feedback phase (gray). After 2 wk of induction, plants and insects were removed and rhizosphere microbiome samples were taken. The remaining soil of each treatment was mixed with sterilized soil (40 : 60, v/v). These soil mixes were used to grow two new sets of Brassica oleracea plants, one set was used for gene expression assessment (24 h post-infestation) and the other set for plant and insect assessment (5 wk post-infestation). In the feedback phase, plants were exposed to D. radicum root herbivory, and the performance of the root herbivore was assessed, as well as plant performance and plant defense-related gene expression. Control plants were noninfested and noninoculated. For infestation with Plutella xylostella (L2) or Brevicoryne brassicae (apterous adults), 10 individuals were carefully transferred to the fourth leaf counted along the stem from the stem base to their respective treatment. To prevent insect contamination between the treatments, the petiole of the infested leaf was wrapped in cotton wool, bagged in a net and fixed with a piece of metal wire. The fourth leaves of the control plants were also wrapped in a similar manner. Delia radicum neonates were brushed on the carefully exposed stem base, just below soil level. For treatments that received Pseudomonas simiae WCS417r, bacterial suspension was applied next to the stem with a syringe. Each pot received 20 ml solution, which equals 2 9 10 10 CFU, and 8 9 10 7 CFU g À1 of soil. Control plants received 20 ml of sterilized 10 mM MgSO 4 , applied in a similar manner as treatment plants.
Conditioning phase: soil and microbiome collection Plants were exposed to insects and rhizobacterial inoculation for 2 wk. Aboveground plant parts and primary roots were then removed from the soil. For soil microbiome analysis, c. 3 g of secondary roots and root-attached soil were pooled from the six plants in each tray. Thus, the six plants in each tray were considered one biological replicate. Pooled roots were collected in 50 ml tubes containing 25 ml of sterilized buffer solution (6.33 g l À1 NaH 2 PO 4 and 10.96 g l À1 NaH 2 PO 4 9 2H 2 O). Tubes were vigorously shaken for 30 s, and centrifuged for 7 min at 3700 g. Supernatant was removed, as well as large chunks of root with sterilized tweezers. The soil slurry was transferred with a sterilized spoon into 1.5 ml tubes, and centrifuged for 5 min at 11 000 g. Supernatant was removed and samples were then stored at À80°C. After taking microbiome samples, soils of all plants from the same treatment were homogenized by mixing by hand, using clean gloves for each treatment. For soils conditioned with plants infested with D. radicum, special care was taken to remove any larvae from the soil.

Feedback phase: setup and measurements
Soil from the conditioning phase was mixed with c-irradiated soil (> 25 kGy; Steris, Ede, the Netherlands) in a ratio of 40% conditioned soil: 60% sterilized soil (v/v). The soil mixture was divided over 1 l pots, into 30 replicates per feedback treatment. We are aware of the discussion between mixed soil sampling strategy and independent soil sampling strategy in PSF experiments (Reinhart & Rinella, 2016;Cahill et al., 2017;Gundale et al., 2019). Since our experiment was performed in pots with similar starting soil, we believe the discussion is less applicable to our study. A soil treatment was added consisting of pots containing a 40 : 60 mix of sterilized soil together with the original potting soil that was used in the conditioning phase (stored for 6 wk at 4°C), to include a treatment consisting of soil with a microbiome similar to that of the soil used as starting material in the conditioning phase; this treatment is hereafter referred to as 'nonconditioned'. Brassica oleracea seeds were sown on seedling soil, that had been stored at 4°C from the start of the experiment, to expose the seeds to a similar microbiome as the first set of plants. After 3 d, the seedlings were transplanted to the feedback phase pots. Plants were grown for 25 d under the same glasshouse settings as during the conditioning phase. After 1 wk of plant growth the pots were provided with sticks to later support insect nets. Plants were divided into two sets, one for gene expression analysis after 24 h of exposure to D. radicum larvae and the other for assessing plant and D. radicum performance.
Feedback phase: plant and root herbivore performance After 4 wk of growth, plants were infested with 10 neonate D. radicum larvae. Half of the plants grown on nonconditioned soil were infested with larvae, to assess effects of D. radicum on plant performance. The larval infestation was performed as described earlier. For insect performance measurements, all plants were individually covered with a mesh bag 10 d after infestation. Plants were inspected daily for emerged D. radicum adults, which were then collected, frozen, and stored at À20°C. Delia radicum size was determined by measuring hind tibia length with a digital microscope (Dino-Lite Edge digital microscope, New Taipei City, Taiwan) as a proxy for fly body size (Soler et al., 2007;Karssemeijer et al., 2020). Developmental time was recorded as the time between larval infestation and adult emergence.
Plant performance in the feedback phase was assessed as leaf area of the second leaf after 3 wk of plant growth as a proxy for plant size. Since measuring the leaf area might damage the leaf, we measured only leaf width and length in experimental plants. We then calculated the leaf area from the leaf measurements using the following formula: length 9 width 9 leaf area coefficient = leaf area. The coefficient was calculated by measuring width, length, and leaf area of 10 Brassica oleracea nonexperimental plants' leaves of similar size using LeafByte (Getman-Pickering et al., 2020). Five weeks after infestation, the plant shoot was harvested and its biomass determined. Dry shoot biomass was recorded to the nearest 0.01 g (DK-6200-C-M; Allscales, Houston, TX, USA) after drying at 105°C for 24 h.

Plant defense-related gene expression analysis
After 4 wk of growth on conditioned soil, half of the plants were infested with 10 neonate D. radicum larvae (Fig. 1), to assess plant defense gene expression under PSF conditions. After 24 h of infestation, primary roots were harvested by uprooting the plants, cutting off secondary roots, and freezing the primary root directly in liquid nitrogen. One leaf disk from three leaves per plant was collected with a 1 cm diameter metal puncher. Samples were pooled for three plants, and immediately frozen in liquid nitrogen to form one replicate. Samples were stored at À80°C.
Frozen samples were ground in liquid nitrogen, with a mortar and pestle for roots, or with a small pestle directly in the collection tube for leaves. Plant RNA was extracted with Isolate II Plant RNA kit (Bioline, London, UK) following the manufacturer's instructions, and converted to complementary DNA (cDNA) (SensiFAST, Bioline). Quantitative polymerase chain reaction (qPCR) analysis was performed to test transcript levels

Research
New Phytologist of genes of interest (CFX96 TM Real-Time System; Bio-Rad, Hercules, CA, USA). The primer efficiency was calculated with qPCR by determining a standard curve with a dilution series. Reference genes SAR1a, Btub, Act-2, PER4, GADHP and EF1a were tested on 10 randomly selected samples from both roots and leaves to determine the optimal combination of reference genes using GeNorm (Vandesompele et al., 2002) in QBASE+ v.3.1 (Biogazelle, Zwijnaarde, Belgium). For roots, Act-2 and SAR1a were used as reference genes, while for leaves Btub and SAR1a were used. We analyzed transcript levels in roots for LOX6, MYB28, CYP81F1, MYB72 and PDR9, and in leaves for LOX2 and MYB28 (Supporting Information Table S1). For MYB72 and PDR9, two genes studied in Arabidopsis (At1g56160 and At3g53480, respectively), orthologous genes in Brassica oleracea were identified using the integrative orthology finder in PLAZA ( van Bel et al., 2018).

Delia radicum biomass assessment
One of the main challenges when working with D. radicum is the difficulty of assessing larval performance. The larvae are small and colorless, and during the first days of feeding they dig into the root, making it difficult to get them back. To overcome this obstacle, we developed species-specific primers (see Methods S1; Table S2; Fig. S1). These primers specifically target the 18S region of D. radicum, without amplifying nontargets such as those found in fungus gnats and nematodes which may occur in the experimental soil. We used these primers in the root samples collected for gene expression analysis (Fig. 1) as a proxy of larval performance and normalized the quantity relative to the plant reference genes Act-2 and SAR1a.

Soil microbiome analysis
Total genomic DNA (gDNA) from 0.25 AE 0.01 g of pooled rhizosphere soil was extracted using the DNeasy PowerSoil Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. The nucleic acid concentration and purity of samples were quantified with a spectrophotometer (DeNovix, Wilmington, DE, USA). For bacteria, the V4 region of the 16S gene was amplified using the 515F/806R primers (Caporaso et al., 2011) (Roche FastStart High Fi, 58°C, 26 cycles). For fungi, the ITS2 region was amplified using the fITS9/ITS4R primers (Ihrmark et al., 2012)  Raw fastq files were processed using CUTADAPT (Martin, 2011) and the DADA2 pipeline (Callahan et al., 2016). The code used for sample processing is available in the Notes S1. After processing, 62 735 to 97 854 bacterial reads and 47 339 to 98 457 fungal reads remained per sample. Taxonomy was assigned using the SILVA v.138 database (Quast et al., 2013) for bacteria and the UNITE v.8.2 database (Nilsson et al., 2018) for fungi. We filtered ASVs (amplicon sequence variants) with too few occurrences using the effective sample approach in metagenomeSeq (Paulson et al., 2013).

Statistical analysis
Statistical analysis was performed in R, v.4.0.0 (R Core Team, 2018), with RSTUDIO v.1.2.5042. For microbiome analysis, counts were normalized using METAGENOMESEQ (Paulson et al., 2013). Principle coordinate analysis (PCoA) was performed using Bray-Curtis dissimilarity in PHYLOSEQ (McMurdie & Holmes, 2013). Permutational multivariate analysis of variance (PERMANOVA) was done with 99 999 permutations using Bray-Curtis dissimilarity with the adonis function (Oksanen et al., 2007), post hoc analysis was performed using the RVAIDEMEMOIRE package (Herv e, 2020). We tested whether differences in variance could have caused significant differences using permutest, which were nonsignificant for both bacterial and fungal analyses, indicating that the PERMANOVA results are valid. Differential ASVs were calculated using DESEQ2 (Love et al., 2014), by comparing each treatment to the noninfested and noninoculated group with a false discovery rate of 0.05.
We used the packages TIDYVERSE, lME4, EMMEANS, LMTEST, lattice and FITDISTRPLUS for plant and insect data (Zeileis & Hothorn, 2002;Sarkar, 2008;Bates et al., 2015;Delignette-Muller & Dutang, 2015;Lenth et al., 2018;Wickham et al., 2019). The distribution of each dataset was explored with QQplots, histograms, Shapiro-Wilk test and the function descdist with 2000 bootstrapped values. Analysis of leaf length, plant shoot dry biomass and gene expression levels was performed with generalized linear models either using Gamma or Gaussian distributions. Development time, fly emergence and hind tibia length of D. radicum were analyzed by using generalized linear mixed models with Poisson, binomial and gamma distributions, respectively. Plant ID was used as a random factor to avoid pseudoreplication. Models were compared and chosen based on Akaike Information Criterion (AIC) values. In the case of multiple fixed factors, the best model that included both factors ('soil treatment' and 'sex' or 'time') was chosen. Significance of fixed factors was assessed using the lrtest function.

Results
Insect herbivore-induced alterations in the plant rhizosphere microbiome Rhizospheres from plants in the conditioning phase were extracted and analyzed for bacterial and fungal communities. We found 1311 bacterial and 187 fungal ASVs, the majority of which belong to the phyla Proteobacteria and Ascomycota, respectively (Fig. S2).
To identify specific changes caused by our treatments, we analyzed differentially abundant ASVs (Fig. 3). Based on visual representation of the Euclidean distance hierarchical tree, for both bacteria and fungi, rhizospheres of plants treated with root herbivory were separated from the shoot herbivory and no herbivory groups. Rhizospheres of plants treated with shoot herbivores also clustered in terms of bacteria, but not for fungal ASVs. For bacteria, most ASVs were differentially abundant between rhizospheres of plants treated with Brevicoryne brassicae and Plutella xylostella and control plants. For fungi, the largest numbers of ASVs were found for plants infested by D. radicum and Pseudomonas simiae + D. radicum.
A cluster of five bacterial ASVs is present in rhizospheres of plants treated with root herbivory, while being absent in the control treatment; these include two members of the family Enterobacteriaceae, a Klebsiella, a Pseudomonas, and Verruccomicrobiom spinosum. Among the fungal ASVs, Candida tropicalis has the most striking difference between treatments, and was strongly associated with rhizospheres of plants treated with D. radicum. Several differentially abundant ASVs, both bacteria and fungi, were negatively affected by infestation of the plants by D. radicum (without Pseudomonas simiae); these ASVs are members of the bacterial families Nocardiaceae and Chitinophagaceae, genera Bryobacter, Chryseobacterium and Roseiarcus, and fungal   order Helotiales, class Microbotryomycetes, and species Candida palmioleophila and Coniochaeta fasciculata. Further, a group of highly abundant bacterial ASVs were quantitatively affected in the rhizospheres of Plutella xylostella and Brevicoryne brassicae-treated plants compared to control plants. For instance, a member of the genus Rhodanobacter was the most abundant ASV in the overall bacterial community, and it was reduced from an average of 3700 normalized counts (4.8% relative abundance) in the rhizosphere of control plants, to 2600 (3.9% relative abundance) and 2500 (3.8% relative abundance) in rhizospheres of Plutella xylostella and Brevicoryne brassicaetreated plants, respectively. Interestingly, several bacterial ASVs were depleted specifically in rhizospheres of Brevicoryne brassicaetreated plants compared to rhizosphere of control plants, including members of the genera Flavobacterium, Azospirillum, Hyphomicrobium, Alkanibacter, Cytophaga, and the species Parafilimonas terrae.
Rhizospheres of plants inoculated with Pseudomonas simiae only differed from those of noninfested/noninoculated plants in four bacterial ASVs, while eight fungal ASVs were affected. Of those four bacterial ASVs in rhizospheres of Pseudomonas simiaeinoculated plants, one is a Pseudomonas fully matching Pseudomonas simiae WCS417r through a BLAST search. However, the sequenced 16S fragments are identical to many strains in the related group Pseudomonads. Therefore we cannot verify that these fragments are explicitly from the strain used in the experiment; without specific bacterial testing, we cannot be certain of the origin of our recovered ASV. Two fungal ASVs, Hawksworthiomyces lignovirorous and Trichoderma hamatum, are specifically depleted in rhizospheres of plants inoculated with Pseudomonas simiae.

Plant-soil feedback effects on plant performance
To assess whether rhizosphere microbiome alterations affected consecutively growing plants and their resistance to insect herbivores, Brassica oleracea plants were grown in the same soil previously conditioned by conspecific plants exposed to different treatments. The surface area of the second leaf was affected by soil conditioning (Fig. 4a; Table 2): plants grown on conditioned soil had smaller leaves. Plant shoot dry mass was also affected by soil conditioning (Fig. 4b; Table 2), where dry shoot biomass of plants grown on conditioned soil was lower compared to plants grown on nonconditioned soils. Plants grown on soil conditioned by plants inoculated with Pseudomonas simiae were smaller compared to plants grown on soil conditioned by noninfested/noninoculated plants. Plants grown on soil conditioned by plants treated with Plutella xylostella were larger, both in terms of leaf size and biomass.

Plant-soil feedback effects on D. radicum performance
To examine belowground plant resistance in a PSF context, we infested Brassica oleracea plants grown in conditioned soils with D. radicum larvae. Overall, D. radicum adult emergence was low in the experiment, on average 11.4% (N total = 1970) of larvae developed into adults. In addition to these performance measurements, in the plants used for gene expression analysis, we examined larval performance through analysis of D. radicum 18S ribosomal RNA.
Emergence of D. radicum was affected by soil conditioning in a treatment-specific way ( Fig. 5a; Table 2). Fewer flies emerged from plants grown on soil conditioned by plants infested by D. radicum compared to plants grown on soils conditioned by plants treated with Brevicoryne brassicae, Pseudomonas simiae or D. radicum together with Pseudomonas simiae. Tibia length of adult flies was affected by soil conditioning (Fig. 5b; Table 2). Flies with smaller tibia length emerged from plants grown on soil conditioned by plants infested with Plutella xylostella compared to flies that emerged from plants grown on nonconditioned soil. Fly development time was similar for all treatments (data not shown).
In the set of plants used for gene expression analysis 24 h postinfestation, we quantified D. radicum 18S ribosomal RNA relative to plant reference genes as a proxy of D. radicum performance ( Fig. 5c; Table 2). Relative quantities of D. radicum 18S were affected by soil-conditioning treatments. This analysis supports the observation that D. radicum performance was reduced in plants grown on soil conditioned by D. radicum compared to plants grown on nonconditioned soil or soil conditioned by control plants. Taken together, the results show that D. radicum was negatively affected when feeding on plants that had been growing in soil conditioned by plants also exposed to feeding by conspecific larvae.

Gene expression in response to D. radicum infestation and plant-soil feedback treatments
We assessed primary root defense responses to herbivory by D. radicum in plants grown on conditioned and nonconditioned soil, measured after 24 h of D. radicum infestation of the primary root. Expression in the roots of LOX6, a gene involved in JA biosynthesis, was induced by D. radicum regardless of soil conditioning ( Fig. 6a; Table 2). Root transcript levels of MYB28, involved in the biosynthesis of aliphatic glucosinolates, were downregulated by D. radicum infestation ( Fig. 6b; Table 2). The soil conditioning treatments did not affect root MYB28 expression, but there was a significant interaction effect between D. radicum infestation and soil conditioning. When infested with D. radicum, transcript levels of MYB28 were lower in plants grown on conditioned soils compared to nonconditioned. In contrast to MYB28 downregulation by D. radicum infestation, messenger RNA (mRNA) levels of CYP81F4, encoding an enzyme involved in indole glucosinolates biosynthesis, were strongly upregulated by infestation. Type of soil conditioning did not influence CYP81F4 transcript levels, but there was an interaction between D. radicum and soil conditioning ( Fig. 6c; Table 2).
Expression of root MYB72, a transcription factor involved in induced systemic resistance and iron acquisition (van der Ent et al., 2008;Palmer et al., 2013), was affected by soil conditioning in a treatment-specific way, but not by D. radicum infestation New Phytologist (2021) Fig. 6d; Table 2). Transcript levels of PDR9, a gene encoding a transporter involved in root exudation of coumarins, were affected by both soil treatment and D. radicum infestation, and there was an interaction between soil treatment and D. radicum infestation ( Fig. 6e; Table 2). When no D. radicum was present, expression of PDR9 was upregulated in primary roots of plants subjected to all soil conditioning treatments compared to plants grown on nonconditioned soil, especially when soil was conditioned by plants infested with D. radicum. This effect was attenuated upon D. radicum infestation, in which case transcript levels of PDR9 did not differ between soil conditioning treatments.
Leaf transcript levels of LOX2, a marker gene for JA biosynthesis expressed in the shoot, were increased by root herbivory but not by soil conditioning; there was a significant interaction effect between soil conditioning and root herbivory ( Fig. S3a; Table 2). MYB28 transcript levels in leaves were affected by soil conditioning treatments ( Fig. S3b; Table 2), but not by D. radicum infestation.

Discussion
Our study shows that the plant root microbiome is affected by insect attack to the plant and that plant growth and insect resistance are influenced via PSF mechanisms. Our results demonstrate that the bacterial rhizosphere community is differentially affected by shoot and root herbivory, whereas the fungal rhizosphere community is mostly affected by root herbivory. Although previous research shows that plant defense against shoot-feeding insects can be altered through PSF (Kostenko et al., 2012;Bezemer et al., 2013;Kos et al., 2015a,b;Hu et al., 2018;Pineda et al., 2020), we here show novel evidence that the root-feeding insect D. radicum is negatively affected by conspecific feeding through PSF. While our data do not allow an unambiguous link to be established between the rhizosphere microbiome in the conditioning phase and the results in the feedback phase, it is most plausible that microbial changes underlie the reported PSF effects on plant growth and insect resistance.

Rhizosphere microbiome composition is differentially affected by shoot and root herbivory
We observed that herbivores feeding on the root or the shoot influenced the rhizosphere microbial community. Multivariate analysis revealed that bacterial rhizosphere communities were separated into three groups: (1) plants exposed to shoot herbivory, (2) plants exposed to root herbivory and (3) noninfested plants. We further observed that the fungal rhizosphere community was similar between plants fed on by shoot-feeding insects and noninfested plants, but was different from the fungal community of plants with root-feeding D. radicum. Thus, our results show that root herbivory has more impact on the plant rhizosphere community than the addition of Pseudomonas simiae WCS417r. A previous study showed that D. radicum herbivory led to only minor changes in the fungal community, but caused major changes in both endosphere and rhizosphere bacterial communities of oilseed rape, Brassica napus (Ourry et al., 2018). Interestingly, our results show that D. radicum herbivory strongly increased the abundance of the soil yeast Candida tropicalis, a species containing known plant growth promoting strains (Amprayn et al., 2012). None of the fungal ASVs that were different between the treatments are known to have entomopathogenic properties, although this was not directly studied for most of these species. Rhizospheres of D. radicum-infested plants showed an accumulation of several bacterial taxa (Enterobacteriaceae, Klebsiella, and Pseudomonas) that were previously found to be associated with the D. radicum gut microbiome (Lukwinski et al., 2006;van den Bosch & Welte, 2020). The gut microbiome of another much-studied root herbivore, western corn rootworm, is thought to consist mostly of microbes selected from the surrounding soil (Dematheis et al., 2012;Ludwick et al., 2019). Our findings hint at the interesting possibility of direct interactions between the microbiomes of the plant rhizosphere and the root herbivore gut. Perhaps, by selecting specific microbes from the soil and excreting them, root herbivores can influence the rhizosphere microbiome. Herbivory by shoot-feeding insects was previously shown to affect the rhizosphere community, in line with our results (Yang et al., 2011;Lee et al., 2012;Bezemer et al., 2013;Kong et al., 2016;Malacrin o et al., 2020;Zytynska et al., 2020). However, some studies report similar rhizosphere microbiomes between shoot-herbivore-infested and noninfested plants (O'Brien et al., 2018;Malacrin o et al., 2020). The variation seen in the literature regarding rhizosphere microbiome responses to shoot herbivory could be explained by factors such as plant-and insect-specific responses, or different bulk soil bacterial communities in the starting soil.

Plant-soil feedback by differently treated conspecifics has adverse effects on plant growth
In the feedback phase of our experiment, we observed treatmentdependent responses in plant growth when grown on conditioned soils. Regardless of the treatment, plant growth was inhibited on conditioned soil compared to nonconditioned soil. Generally, such unfavorable legacy from plant conspecifics is termed negative PSF. In our experiment, shoot herbivory by Plutella xylostella on plants during the conditioning phase led to increased growth of plants in the feedback phase, compared to plants grown in soil conditioned by plants without herbivores. Hence, herbivory can affect not only the attacked plant, but also the growth of future plants growing in the same soil, via soilmediated effects.

New Phytologist
It is challenging to directly link changes in the rhizosphere microbiome of plants in the conditioning phase of our experiment with findings in the feedback phase. One potential discrepancy is that we sampled rhizosphere soil for microbiome analysis but transferred all soil in the pot to the feedback phase. The soil in the pots was completely colonized by roots at the end of the conditioning phase, therefore we believe that the overall bacterial community we transferred is representative of the rhizosphere community. Several PSF mechanisms other than the transfer of microbes could have contributed to our results. Fresh litter, such as fine roots, can stimulate the microbial activity (Fontaine et al. 2003), but can also negatively affect plant growth through the release of phytotoxic (allelopathic) and autotoxic compounds when decomposing (Bonanomi et al., 2006). Extracellular self-DNA (eDNA) is also released from decomposing tissue, and can exert plant growth inhibition on grasses, forbs and Arabidopsis thaliana in vitro (Mazzoleni et al., 2015). These PSF mechanisms are likely to have contributed to our results to some extent, as root fragments were present in the soil we transferred.
Surprisingly, the performance of Brassica oleracea was drastically decreased when grown in soil on which previously growing plants had been inoculated with Pseudomonas simiae, compared to the other soil conditioning treatments. Root herbivory by D. radicum together with Pseudomonas simiae inoculation of the plants during the conditioning phase restored plant biomass to a certain degree in the feedback phase. Although this PGPR strain is usually considered a beneficial rhizobacterium when applied to plants, including Brassica oleracea (Friman et al., 2021a), our results suggest that this beneficial effect may not persist through PSF. Notably, there are reports of rhizobacteria causing effects varying from plant-growth promotion to inhibition, depending on e.g. phosphate availability or rhizobacterial population density   (Ciccillo et al., 2002;Morcillo et al., 2020). Although plant growth may have been boosted in 2 wk of the conditioning phase, we regard this period as too short to leave lasting nutrient deficiencies in the soil, and therefore unlikely to have influenced our results. Further, we assume that the nutrient availability was sufficient for the experimental plants due to regular fertilization in our experiments and hypothesize that changes in the microbiome underlie the reduction in growth.
In contrast to our hypothesis, we found that inoculation with the rhizobacterium Pseudomonas simiae did not affect overall microbial communities in the rhizosphere. Although there are studies that find an altered root community after addition of individual rhizobacterial species, others report no such effects (Herschkovitz et al., 2005;Gadhave et al., 2018;Wang et al., 2018;Zytynska et al., 2020). Even though the microbial community composition was not affected by the addition of Pseudomonas simiae WCS417, the abundance of several distinct species was changed. It has been demonstrated that only a set of three bacterial soil species are sufficient to increase resistance in Arabidopsis thaliana against a foliar fungal pathogen . For example, Trichoderma hamatum was absent in rhizospheres of Pseudomonas simiae-induced plants while it was

Research
New Phytologist present in the other treatments. This species is a known growthpromoting fungal species in e.g. pepper (Mao et al., 2020). In this way, the addition of Pseudomonas simiae may have suppressed other beneficial microbes in the rhizosphere, leading to a net negative effect on plant growth in our study.

Root herbivores can be affected via plant-soil feedback
Root herbivory by D. radicum during the conditioning phase led to lower performance of D. radicum in the feedback phase, in line with previous studies that recorded an alteration of plant resistance against insects through PSF (Kostenko et al., 2012;Bezemer et al., 2013;Kos et al., 2015a,b;Hu et al., 2018;Pineda et al., 2020). Overall D. radicum adult emergence in our experiment was low compared to other studies using similar methods (Soler et al., 2007;van Geem et al., 2015;Karssemeijer et al., 2020). As a root miner, the insect is difficult to quantify in the early stages of its lifecycle. Therefore, we developed primers to supplement the emergence data with the quantification of D. radicum 18S ribosomal RNA after 24 h of feeding. This is a novel method to quantify root fly larval performance in planta; yet, similar methods are used to quantify plant parasitic nematode abundance in roots (Zijlstra & Van Hoof, 2006;Braun-Kiewnick et al., 2016). The D. radicum 18S ribosomal RNA method confirmed a lower performance of D. radicum on plants in the feedback phase growing in soil conditioned with D. radicum-infested plants. Notably, this technique can be further fine-tuned, for instance by dilution or selecting the optimal time-point for harvesting, and the results here should be interpreted in conjunction with the emergence data. Differences between the emergence data and 18S measurements may be due to different life stages targeted, as one measures performance of neonates while the other measures survival to adulthood.
The performance of D. radicum may have been affected by a change in plant defense, or by a direct influence of the soil microbiome. Lachaise et al. (2017) reported that differences in the soil microbiome affected D. radicum performance. Delia radicum infestation was previously shown to increase the abundance of Bacillus and Paenibacillus in the rhizosphere, which could have entomopathogenic properties (Ourry et al., 2018). These bacterial species were not differentially affected in our study, perhaps due to different plant growth substrates. Without isolating specific rhizosphere microbes and testing their effects on the plant and the root herbivore larvae, we can only speculate about the underlying mechanisms.
In roots, most defense markers we studied were not affected by soil conditioning treatments, and thus they do not explain the difference in insect performance. However, we cannot rule out that soil microbes may have primed defense against D. radicum, leading to a faster defensive response. Indeed, two genes involved in ISR, MYB72 and PDR9, were affected by soil conditioning treatments. The role of these genes in ISR has been especially studied in Arabidopsis thaliana. Here, we found that soil conditioning changed the expression of their orthologues in Brassica oleracea. The transcription factor MYB72 has been identified as a key regulation node in Arabidopsis thaliana roots in iron uptake and communication with the beneficial rhizobacterium Pseudomonas simiae WCS417r (Verhagen et al., 2004) and was later verified to play a central role in rhizobacterial ISR (van der Ent et al., 2008). This transcription factor regulates the expression of genes involved in the shikimate, phenylpropanoid and nicotianamine biosynthesis pathways, including genes leading to the production and exudation of coumarins . These coumarins, in particular scopoletin, are secreted by the roots by the transporter PDR9, where they play a dual role in both the plant response to iron deficiency and influencing the rhizosphere microbiome (Stringlis et al., 2018(Stringlis et al., , 2019. This could be an indication that ISR plays a role in PSF. Interestingly, transcript levels of LOX2 and MYB28 in leaves were affected by soil conditioning treatments, a result which is in line with previous studies that found a link between shoot defense and PSF in maize plants (Hu et al., 2018). Our gene expression results underline that defense signaling in shoot and root is fundamentally different .

Conclusion
In conclusion, our study demonstrates that shoot and root herbivory lead to distinct plant rhizosphere microbial communities, whereas inoculation of Pseudomonas simiae to the soil has limited effects on the rhizosphere microbial community. Through PSF, plant performance and defense is altered in a treatmentdependent way for Brassica oleracea plants growing in soil conditioned by conspecific plants. The results presented here suggest that changes in the abundance of specific microbes, rather than the overall microbiome, may be more important for plant performance and defense.  Relative abundance of bacterial and fungal phyla in rhizospheres of Brassica oleracea plants exposed to herbivory, rhizobacterial inoculation or a combination.

Fig. S3
Relative gene expression of LOX2 and MYB28 in leaves of Brassica oleracea plants grown in soil conditioned by conspecific plants exposed to herbivory, rhizobacterial inoculation or a combination.
Notes S1 Code used for processing and analyzing microbiome samples.
Table S1 Primers for target and reference genes in Brassica oleracea.