The rhizosphere microbiome of 51 potato cultivars with diverse plant growth characteristics

Abstract Rhizosphere microbial communities play a substantial role in plant productivity. We studied the rhizosphere bacteria and fungi of 51 distinct potato cultivars grown under similar greenhouse conditions using a metabarcoding approach. As expected, individual cultivars were the most important determining factor of the rhizosphere microbial composition; however, differences were also obtained when grouping cultivars according to their growth characteristics. We showed that plant growth characteristics were related to deterministic and stochastic assembly processes of bacterial and fungal communities, respectively. The bacterial genera Arthrobacter and Massilia (known to produce indole acetic acid and siderophores) exhibited greater relative abundance in high- and medium-performing cultivars. Bacterial co-occurrence networks were larger in the rhizosphere of these cultivars and were characterized by a distinctive combination of plant beneficial Proteobacteria and Actinobacteria along with a module of diazotrophs namely Azospira, Azoarcus, and Azohydromonas. Conversely, the network within low-performing cultivars revealed the lowest nodes, hub taxa, edges density, robustness, and the highest average path length resulting in reduced microbial associations, which may potentially limit their effectiveness in promoting plant growth. Our findings established a clear pattern between plant productivity and the rhizosphere microbiome composition and structure for the investigated potato cultivars, offering insights for future management practices.


Introduction
A projection of the world's growing population suggests that the demand for food will likely increase by 70% by 2050 (FAO 2019 ).Potatoes belong to the most important plants to ensure global food security and are in addition used as r ene wable r esources for many parts of industry (Priedniece et al. 2017, Wijesinha-Bettoni and Mouillé 2019, Bakhsh et al. 2020 ).In 2021, around 373 million tons of potatoes were harvested worldwide (FAOST A T 2021 ).Howe v er, potatoes ar e highl y susceptible to abiotic and biotic str essors (Levy and Veilleux 2007, Beddington 2010, Monne v eux et al. 2013, Bakhsh et al. 2020 ), and ther efor e it is assumed that in future yields will be strongly impacted mainly as a result of climate change (Hijmans 2003 ), which exacerbates prolonged drought periods, incr eased temper atur es, salinity, and emer ging pathogens (Handayani et al. 2019 , Raza andBebber 2022 ).Ther efor e, ther e is a strong need to develop alternative forms of management, as tr aditional a gricultur e pr actices will unlikel y meet the demand to maintain yields and quality (Singh et al. 2020, Sun et al. 2021 ).
Here, making use of the functional potential of the root-associated microbiome has been proposed as highly promising to ensure global food security while maintaining a healthy environment (Singh and Trivedi 2017 ).
The rhizosphere is the narrow zone of soil that is influenced by root secretions and drives complex plant-microbe interactions (Berendsen et al. 2012, Beattie 2018 ).The plant rhizosphere harbors diverse microbes, many of which extend plant capabilities including adaption to environmental stresses (Pascale et al. 2020 ), r esistance a gainst pathogen (Ber endsen et al. 2018 ), and impr ov ement of nutrient and water uptake (Richardson et al. 2009, Haney et al. 2015 ).While the composition of rhizosphere microbiota is str ongl y influenced by the soil microbiome (Inceo glu et al. 2012, Liu et al. 2019, Veach et al. 2019 ), in addition plants themselv es activ el y sha pe micr obial assembla ges within the rhizospher e thr ough the r elease of r oot exudates (Bulgar elli et al. 2013 ).For potato, those exudates can account for up to 20% of the total assimilated carbon (Gschwendtner et al. 2011 ), but composition and amount of root exudates can differ among potato cultivars and plant de v elopment sta ges (Gsc hwendtner et al. 2011 ).Indeed se v er al studies demonstrated an significant effect of potato cultiv ars on micr obial comm unities in their rhizospher es (Inceo glu et al. 2010, 2011, 2012, Gsc hwendtner et al. 2011 ).Copiotr ophic bacteria (known as r-str ategists), particularl y Pr oteobacteria ar e enriched in the rhizosphere, when compared to the surrounding soil, as demonstrated in a recent global study that integrated data fr om v arious ecosystems and soil bac kgr ounds (Ling et al. 2022 ).Consequentl y bacteria gener a like Sphingobium , Brad yrhizobium , Devosia , Microvirga , Rhizobium , Variovorax , or Burkholderia , which have been reported as potentially beneficial for plant growth, have been often described as important bacteria of the potato rhizosphere (Inceo glu et al. 2010, Barnett et al. 2015, Pfeiffer et al. 2017 ).Ho w e v er, besides Pr oteobacteria, other gr oups including Acidobacteria, Actinobacteria, Bacteroidetes , Firmicutes , as well as fungi including Ascom ycota, Basidiom ycota, or Zygom ycota are typical member of the microbiome of the potato rhizosphere.Thus assembly processes of the rhizosphere microbiome are essential to understand (Stegen et al. 2013 ).
Determinism and stochasticity are the two fundamental ecological processes that govern microbial assembly (Vellend 2010, Stegen et al. 2012 ).While deterministic processes reflect host selection, species filtering by environmental adaptation and interspecies inter actions, stoc hasticity involv es tr ait-and selectionindependent comm unity assembl y r egulated by stoc hastic e v ents such as random proliferation, death, and dispersal (Stegen et al. 2013 , Zhou andNing 2017 ).Despite the occurrence of both processes in most communities at local scales (Liu et al. 2023 ), previous studies reported that the rhizosphere microbiome composition is driven by deterministic processes (Zhang et al. 2018 ) and is selected according to the functional traits that benefit plant health (Mendes et al. 2014 ).Recently, another study involving eight potato cultivars grown under continuous cropping regime also demonstr ated a pr e v alent contribution of deterministic processes (64.19%-81.31%) in bacteria comm unity assembl y while the fungal comm unities wer e mainl y dominated by stoc hastic pr ocesses (78.28%-98.99%)(Gu et al. 2022 ).
Deplo ying microbial netw ork anal yses, especiall y cooccurrence networks based on DNA sequencing data sets offer great potential for unraveling the hidden patterns within large and complex microbial communities, and therefore have been lar gel y used to study community structures (Barberán et al. 2011, Banerjee et al. 2019, Niraula et al. 2022 ).These networks ar e inter pr eted as r eflecting inter-or intr akingdom inter actions between species that play diverse roles within the microbial ecosystems (Floc'h et al. 2020, Matchado et al. 2021 ).Pr e vious studies reported significant differences between microbial cooccurrence networks of bulk soil and those in the rhizosphere (Shi et al. 2016, Ling et al. 2022 ), mainly attributed to root exudates by which plants selectively recruit specific microbes (Blagodatskaya et al. 2014 ).Within association networks, nodes with extensive connections wer e r eferr ed to as hub taxa (Agler et al. 2016 ) and subsequently to k e ystone taxa, as their removal resulted in outsized impact on both the composition and functioning of the microbiome (Banerjee et al. 2018 ).Microbial co-occurrence networks can also involve the presence of modules (closely interconnected nodes), which can be interpreted as groups of taxa with ov erla pping ecological nic hes (Nir aula et al. 2022 ).A study sho w ed that v arious cultiv ars of Rabbiteye Blueberry had a substantial effect on the properties (modularity and robustness) of the rhizosphere bacteria co-occurrence network (Jiang et al. 2017 ).Furthermor e, the authors demonstr ated a positiv e and significant correlation between fruit yield and putative k e ystone taxa affiliated to Acidobacteria, Proteobacteria, and Actinobacteria (Jiang et al. 2017 ).In a recent study, members of modules significantl y positiv el y corr elated with yield wer e found to be more abundant in the rhizosphere of soybean with high productivity (Niraula et al. 2022 ).These findings suggest that microbial associations ma y ha v e implications for plant pr oductivity.Apart fr om these examples, v ery little is still known about the relation-ship between structure and complexity of microbial networks and the growth characteristics of plants, especially in potato.
Although studies have previously reported on the influence of genotype and environmental factors on the potato rhizosphere microbiome (Inceo glu et al. 2010, 2011, 2012, Gschwendtner et al. 2011, Weinert et al. 2011, Pfeiffer et al. 2017, Faist et al. 2023 ), they hav e onl y focused on a limited number of potato cultivars, thus pr oviding onl y a sna pshot of the lar ge genetic v ariability pr esent within this crop.We have expanded the scope by examining a collection of 51 potato cultivars and related them to the rhizosphere microbiome with the aim to: (i) gain a compr ehensiv e understanding of the selective effect potato cultivars have on the composition and structure of the rhizosphere microbiome; (ii) explore microbial co-occurrence patterns in response to plant growth characteristics .Furthermore , we used the β-nearest taxon index ( βNTI) a ppr oac h in a first attempt to understand whic h assembl y pr ocesses contribute to the structuring of microbial communities in the rhizosphere of the investigated potato cultivars.
To ac hie v e these goals, we gr e w the 51 cultiv ars under contr olled gr eenhouse conditions in soil fr om an a gricultur al field to allow for recruitment of a natural rhizosphere microbiome.We hypothesized that under these optimal growth conditions, the rhizospher e micr obiome w ould sho w differ entiall y higher r elativ e abundances of plant gr owth-pr omoting Pr oteobacteria, including Microvirga , Variovorax , Hyphomicrobium , Sphingobium , and Bradyrhizobium in high-performing potato cultivars compared to lowperforming cultiv ars (H1).Furthermor e, we expected that when the rhizosphere microbiome plays an important role in plant pr oductivity, the micr obial co-occurr ence network within highperforming cultivars would be larger and more complex than that of low-performing potato cultivars (H2).

Site description and experimental design
The soil for the experiment was collected in summer 2020 from the upper layer (0-20 cm) of an arable field at the experimental station Gut-Roggenstein (latitude 48.1824420, longitude 11.3126896, 508 m above sea level), Technical University of Munich in Southern Germany.The field had previously undergone a series of crop rotations: beans in 2015, wheat in 2016, rapeseed in 2017, wheat in 2018, maize in 2019, and summer barley in 2020.The soil, a loamy sand (55.9% sand, 27% silt, and 17% clay) classified as of luvisol was homogenized with a 2-mm diameter mesh sie v e, following the r emov al of stones and crop residues before the experiment.The soil was stored at 4 • C until further use.
The selection of 51 potato cultivars for this study includes varieties intended for food production and industrial purposes, name and c har acteristics, including country of origin, year of breeding, maturity, skin color, and shape of tubers are provided in Table 1 .Further information such as cultivar ID, resistance to abiotic and biotic stresses can be found on the Eur opean Cultiv ated Potato database available at www.europotato.org .These cultivars exhibited significant variability in the quality and quantity of root exudates (Joana Falcao Salles, personal communication), and were c hosen specificall y to inv estigate their inter action with the soil microbiome .T he cultivars were obtained from the Polish in vitro potato gene bank (Bonin, Poland).Plants were cultivated as tissue cultures for ∼56 days within the accredited gene bank laboratories at the Bonin Division of the Plant Breeding and Acclimatization Institute-National Research Institute (Bonin, Poland).Mur ashige and Sk oog nutrient medium, the a gar plugs attac hed to their roots were gently eliminated using tweezers and tap water.
Each plant was immediately transferred to 0.3 l (7 cm × 7 cm × 8 cm) pots filled with the homogenized soil and allo w ed to acclimate for 14 days (Fig. 1 ).Throughout the acclimatization period, the plants were water ed thr ee times a week to maintain a r oughl y 60% of the maximum water holding capacity .Subsequently , the acclimated plants wer e individuall y tr ansferr ed to 1.5 l (11 cm × 11 cm × 12 cm) pots and cultivated for 28 days in the greenhouse under r elativ e humidity of 65%, day/night temper atur es of 22 • C/18 • C, and a day/night photoperiod of 16/8 h.Each pot was r egularl y weighted, and soil moisture was maintained at 60% of the maximum soil water holding capacity throughout the experiment.Eac h pot r eceiv ed 50 ml of a low concentration nutrient solution ( Table S1 , Supporting Information ) corresponding to an addition of 1.1 mg N per pot ( ∼1 kg N ha −1 ) at 16 days after planting in order to not influence soil microbiome with plants.Each Figure 1.Experimental design of the greenhouse experiment using 51 in vitro propagated potato cultivars acclimated in homogenized soil for 14 da ys .Acclimated plants were transferred to 1.5 l pots containing the same soil and were cultivated for 28 days under continuous watering [60% maximum water holding capacity (mWHC)].At 42 days after planting, the experiment was terminated.Root, stem, and leaf samples were taken to assess growth traits and rhizosphere soil samples were collected for bacterial and fungal community analyses.Created with BioRender.com.cultiv ar was gr own in three replicates .T he experiment was terminated 42 days after planting, and various plant growth parameters above-and belo wground w ere recor ded (Fig. 1 ).Shoot fresh w eight w as measur ed immediatel y, and after drying for 2 days at 75 • C in the oven, dry weight was pr omptl y r ecorded.Fr esh weight and number of tubers for each individual plant wer e r eported as well as the av er a ge number of leaves.These plant growth data are available in Table S2 ( Supporting Information ).
For each individual plant, the thin layer of soil attached to the plant r oots, r eferr ed to as the rhizospher e, was collected, r esulting in 153 samples in total (51 cultivars × 3 replicates).These samples wer e immediatel y placed on dry ice and then stored at -80

Plant data analysis
To categorize the 51 potato cultivars according to their growth c har acteristics, w e emplo y ed k-means clustering.After calculating the mean values for all growth variables within each cultiv ar, our data wer e scaled.To ascertain the optimal number of clusters (k), w e emplo y ed the R function fviz_nbclust() from the Factoextr a pac ka ge v1.0.7.Visualization of the k-means clusters was accomplished through a principal component analysis using the fviz_cluster() function provided within the same R pac ka ge.For the r epr oducibility of the k-means cluster analysis, samples wer e hier arc hicall y cluster ed with the hclust() method using Euclidean distance and w ar d.D2 linka ge.Statistical differ ences betw een plant gro wth-based clusters of potato cultivars w ere as-sessed with analysis of similarity (ANOSIM) available in the packa ge v egan v2.6-4.The r esulting gr owth clusters (high, medium, and low productivity) were used to analyse the rhizosphere micr obial comm unity composition, co-occurr ence netw orks as w ell as community assembly processes.

Metabarcoding of bacterial and fungal communities
Total DN A w as extr acted fr om 250 mg of eac h individual rhizosphere soil sample using the DNeasy Po w erSoil Kit (QIAGEN GmbH, Hilden, Germany) following the manufacturer's instructions.Empty extraction tubes were used as negative controls to c hec k for contamination during the process .T he concentration of DN A extracts w as quantified in duplicate using SpectraMax Gemini EM Microplate Spectrofluorometer (Molecular Devices , C A, USA) and Quant-iT PicoGreen dsDNA Assa y Kit (T hermo Fischer Scientific, Waltham, MA, USA) according to the manufacturer's instructions.All samples were stored at −20 • C until further analysis.
The ITSmix3/ITSmix4 primer pair (Tedersoo et al. 2014 ), was used to amplify the ITS2 region of the fungal nuclear rRNA.PCR was performed with an initial denaturation phase at 95 • C for 15 min and 30 cycles of 30 s denaturation at 95 • C, 30 s annealing at 55 • C and 1 min extension at 72 • C, and a final extension of 10 min at 72 • C. PCR reaction mixtures contained 1 of 10 ng DNA templates, 0.5 μl of 10 pmol of each primer, 2.5 μl of 3% BSA, 12.5 μl of NEBNext High-Fidelity 2x PCR Master Mix (New England Biolabs, Fr ankfurt am Main, German y), and 8 μl of DEPC-treated water, resulting in a total volume of 25 μl.
For amplification of the V4 region of the bacterial 16S rRNA gene, we used universal primer pair 515F/806R (Apprill et al. 2015(Apprill et al. , P ar ada et al. 2016 ) ). PCR was performed under the following conditions: an initial denaturation phase at 98 • C for 1 min and 30 cycles of 10 s denaturation at 98 • C, 30 s annealing at 55 • C and 30 s extension at 72 • C, and a final extension for 5 min at 72 • C. PCR r eaction mixtur es contained 1 μl of 10 ng DNA templates, 0.5 μl of 10 pmol of each primer, 2.5 μl of 3% BSA, 12.5 μl of NEB-Next High-Fidelity 2x PCR Master Mix (New England Biolabs), and 8 μl of DEPC-treated water.
PCR pr oducts wer e v erified in 1% a gar ose gels, follo w ed b y Ma gSi NGSpr ep Plus bead purification (Steinbr enner, Wiesenbac h, Germany).The quality and quantity of purified amplicons and the presence of primer dimers wer e c hec ked with DNF-473 Standard Sensitivity NGS Fr a gment Kit (1-6000 bp) on a fr a gment analyser (Agilent Technology, Santa Clara, CA, USA).Both bacterial and fungal purified amplicons were indexed in an 8-cycle PCR, which contained 2.5 μl of each indexing primer (Nextera ® XT Index Kit v2; Illumina, San Diego, CA, USA), 12.5 μl NEBNext High-Fidelity 2x PCR Master Mix, 6.5 μl DEPC-treated water, and 10 ng of the purified amplicon.Subsequently, a second round of purification, follo w ed b y quality and quantity control w ere performed as described abo ve .Prior to sequencing, samples were diluted to 4 nM and equimolarly pooled into a single Eppendorf tube.P air ed-end sequencing was carried out using the MiSeq ® Reagent kit v3 (600 cycles) on the Miseq instrument ® (Illumina Inc., San Diego, CA, USA).
Pr epr ocessing of the bacterial and fungal raw sequencing data was conducted on the Galaxy web platform (Afgan et al. 2016 ) as pr e viousl y described by Kublik et al. ( 2022 ), with few modifications.Briefly, the following trimming and filtering parameters wer e consider ed for bacteria: 20 bp wer e r emov ed n-terminall y and reads were truncated at position 220 (forw ar d) and 150 (reverse) with expected error of 3 and 4, respectively.For fungi, forw ar d reads were trimmed to 20-220 bp, reverse reads to 20-160 bp with the same number of err ors.After mer ging r eads, the r esulting unique amplicon sequence variants (ASVs) wer e tr ained a gainst SILVA database v138.1 for bacteria and UNITE fungi database v9.0 released for QIIME, with 0.99 confidence threshold (Abarenkov et al. 2022 ).The R language and environment v4.2.1 were used for downstr eam anal ysis.Using Bioconductor decontam pac ka ge v1.13.0 (Davis et al. 2018 ), contaminant sequences were filtered based on pr e v alence acr oss negativ e contr ols, along with ASVs assigned to c hlor oplast and mitoc hondria.A phyloseq object was cr eated for eac h of bacterial and fungal datasets using the Phyloseq pac ka ge v1.42.0.Singletons (ASVs r epr esented by onl y one r ead acr oss all samples) wer e r emov ed.Furthermor e, onl y ASVs found in at least two out of three replicates per cultiv ar wer e kept for downstream analysis.We emplo y ed total-sum scaling (TSS) for data normalization.TSS involv es tr ansforming the abundance table into a r elativ e abundance table by scaling the data according to the library size of each sample.

Di v ersity and composition
Alpha diversity within bacterial and fungal communities was estimated through the observed species richness, Shannon index, and Simpson's dominance index in both, indi vidual culti vars and the growth clusters employing the Micr obiome pac ka ge v1.20.0.
A nonparametric Wilcoxon test was conducted to determine the influence of the individual cultivars and plant growth-based clusters on alpha div ersity.Differ ence between sample groups were considered significant when P < .05.
Beta diversity was evaluated through a principal coordinates analysis (PCoA) of weighted UniFr ac dissimilarity.A perm utational m ultiv ariate anal ysis of v ariance (PERMANOVA) with 999 permutations using weighted UniFrac distance (R package Vegan v2.6-4) was employed to estimate the r elativ e contributions of various factors including individual cultivars, plant growth-based clusters, br eeding pur pose, earliness and country of origin, on the structure of microbial communities.
Differ entiall y abundant taxa at different taxonomic levels within the plant growth-based clusters of potato cultivars were identified by pairwise Wilcoxon test ( α = 0.05) adjusted with FDR.
Bacterial and fungal ASVs found in 80% of samples with a relative abundance greater than 0.01% were considered for core micr obiome anal ysis using the amp_v enn function av ailable in the ampvis2 pac ka ge (v2.7.34).

Microbial assembly process based on entire-community null model
The βNTI and Raup-Crick-based Bra y-Curtis (RCbra y) were used to determine the contribution of deterministic and stochastic assembl y pr ocesses as pr e viousl y described (Stegen et al. 2013 ).Briefly, the βNTI measures the degree of deviation of the β-meannearest taxon distance from the null expectations based on 1000 random shuffles of ASVs across the phylogenetic tree.Values of | βNTI | > 2 indicate deterministic selection, which can be further partitioned into heterogeneous ( βNTI > 2) or homogenous ( βNTI < 2) selection.While heterogeneous selection implies that selectiv e pr essur es driv e comm unities to div er gent configur ations, in homogeneous selection, these selective pressures push communities to w ar d a common composition (Stegen et al. 2015 ).The r emaining comm unity pairs with | βNTI | < 2 indicate that the comm unity is mainl y assembled by stoc hastic pr ocesses.RCbr ay can be used to further classify this stochastic fraction.Values of RCbray < −0.95 indicate communities influenced by homogenizing dispersal (taxonomicall y mor e similar than expected; populations are capable of interactions, allowing members to freely exc hange), while RCbr ay > 0.95 suggest dispersal limitation; populations are unable to mix leading to de v elopment via ecological drift.Values of | RCbray | < 0.95 indicate an undominated processes, where no single assembly process is capable of explaining variation (Stegen et al. 2015 ).The βNTI, RCbray, and assembly processes based on entire-community null models (Stegen et al. 2013 ) were calculated using the qpen function from the iCAMP v1.5.12 R pac ka ge (Ning et al. 2020 ).

Co-occurrence networks
Microbial netw orks w ere constructed using the NetCoMi pac ka ge v1.1.0(Peschel et al. 2021 ).Initially, the dataset was subset according to the growth clusters of potato cultivars.After taxa were a gglomer ated at the genus le v el (pac ka ge Speedyseq v0.5.3.9018), the abundance table of each growth cluster underwent center log r atio tr ansformation (pac ka ge SpiecEasi v1.1.2) to addr ess compositionality bias.To compute associations between taxa, a Pearson correlation with a default threshold of 0.3 was performed.Additionally, sparsification and selection of connected nodes were conducted using a Student t -test with α = 0.0001 and 0.05 for bacterial and fungal networks, r espectiv el y.False positiv e was addressed using lFDR < 0.2.The resulting networks were structured into modules using the cluster_fast_greedy method.Nodes possessing the highest eigenv ector centr ality (abov e the 95% quantile the empirical distribution of centrality values) were designated as hub taxa.Eigenvector centrality measures the importance of a node within a network based on both its connections (degree) and the centrality of the nodes it is linked to.Lastly, the Fruchterman-Reingold layout algorithm from the package igraph v1.3.5 was used for network visualization.

Results
As expected, the potato cultivars were the strongest factor shaping the composition of both, the bacterial (PERMANOVA, R 2 = 0.4, P = .003)and fungal (PERMANOVA, R 2 = 0.4, P = .001)communities in the rhizosphere (Table 2 ).This was further confirmed with the PCoA of weighted UniFr ac distance, whic h showed a broad and uneven distribution of cultivars along the two axes ( Figure S1 , Supporting Information ).Ho w ever, the alpha diversity based on observed species richness and Shannon's index was similar amongst most of these cultivars ( Figure S2 , Supporting Information ).Within rhizosphere bacterial communities, most abundant taxa included Vicinamibacteria (Acidobacteriota), Gamma pr oteobacteria (Pr oteobacteria), and Actinobacteria (Actinobacteriota) ( Figure S3 , Supporting Information ).A similar pattern was found in the bare soil prior to planting ( Figure S4 , Supporting Information ).Rhizosphere fungal communities were mainl y r epr esented b y U. Sor dariales, Ramophialophora , Pseudeurotium , Podospora , Neoschizothecium , Fusarium , Lasiosphaeris , and Cercophora ( Figure S5 , Supporting Information ).In the next step, we wer e inter ested in how rhizospher e comm unities (bacteria and fungi) were associated with growth clusters.

The 51 cultivars can be grouped into plant growth-based performance clusters
The 51 potato cultivars grouped into three distinct clusters based on their growth characteristics, and each cluster contained a r elativ el y similar number of cultivars.Both approaches, the kmeans (Fig. 2 A) and hier arc hical clustering (Fig. 2 B) r esulted in the same number of potato growth clusters with lar gel y identical cultivars .T he first cluster comprised high-performing cultivars, the second cluster had cultivars with medium-performance, and the third cluster r epr esented low-performing cultivars .T he analysis of similarity between these growth clusters revealed a significant difference (ANOSIM, R = 0.6, P = .001)among the three groups.

Plant growth-based clusters influence the di v ersity and composition of the rhizosphere microbiome
Plant growth-based clusters of potato differed significantly in the bacterial ( Figure S3 , Supporting Information ) but not the fungal α-diversity ( Figure S6 , Supporting Information ).Specifically, the growth clusters comprising high-and low-performing cultiv ars consistentl y exhibited a higher observ ed species ric hness and Shannon's index compared to that of medium-performing culti vars (Wilco xon, P < .05)(Fig. 3 A and B).Ho w e v er, no significant difference in α-diversity was observed between clusters of high-and low-performing culti vars (Wilco xon, P > .05).Ho w e v er, the Simpson's index indicated lo w er species dominance in lowperforming cultivars compared to the other clusters (Wilcoxon, P < .05)(Fig. 3 C).
Although ov er all bacterial and fungal comm unity composition sho w ed no separation accor ding to the plant gro wth-based clusters of potato (Fig. 4 and Figure S7 , Supporting Information ), we found that these growth clusters as well as other factors including breeding year, purpose of breeding, and cultivar earliness had a significant effect on the β-diversity (PERMANOVA, P < .05)(Table 2 ).Additionally, the shape of the tubers only affected fungal communities.
A detailed analysis of microbial community composition of the plant growth-based clusters indicated Vicinamibacteria, Gamma pr oteobacteria, and Actinobacteria as the three most dominant bacterial classes for all growth-based clusters (Fig. 5 A).Pairwise comparisons of the different classes amongst the growth clusters r e v ealed that the r elativ e abundance of Actinobacteria and Verrucomicrobiae was higher in high-performing cultivars in comparison to low-performing cultivars (Wilcoxon, P < .05),whereas Vicinamibacteria (Acidobacteriota) and Polyangia (Myxococcota) exhibited the opposite pattern (Wilcoxon, P < .05)(Fig. 5 B).Due to the low taxonomic resolution, the analysis of Acidobacteriota could only be conducted at the order and famil y le v els.Ne v ertheless, the same observ ations wer e made at both le v els , i.e .Vicinamibacterales and Vicinamibacteraceae consistently exhibited higher relative abundance in the cluster of low-performing cultiv ars.Further anal ysis at the genus le v el confirmed the pr e vious r esults for Actinobacteriota.Specifically, the genus Arthrobacter , exhibited higher r elativ e abundance in high-performing cultivars (Wilcoxon, P < .05)(Fig. 5 C).Additionally, Arthrobacter was also influenced by the purpose of breeding as its proportion was found to be greater in cultivars bred for industry and starch production compared to generalpur pose cultiv ars (Wilcoxon, P < .05)(Fig. 5 D).Neither Alpha pr oteobacteria nor Gamma pr oteobacteria sho w ed differential abundances between the growth clusters (Wilcoxon, P > .05)( Figure S8 , Supporting Information ).Genus-based anal ysis r e v ealed that Massilia had a higher proportion in high-performing cultivars compared to low-performing cultivars (Wilcoxon, P < .05),MND1 and Ellin6067 sho w ed the opposite trends, and no clear pattern was found for Sphingomonas ( Figure S9 , Supporting Information ).Other genera including Bradyrhizobium , Hyphomicrobium , Lysobacter , Pseudomonas , Thermomonas , Ramibacter , Piscinabcter , Ellin6067 , Caenimonas , Acidibacter , and Arenimonas did not discriminate among the growth clusters (Wilcoxon, P > .05)( Figure S9 , Supporting Information ).
The fungal community composition in the rhizosphere of the investigated potato cultivars was dominated by U. Sordariales, Ramophialophora , Pseudeurotium , Podospora , Neoschizothecium , Fusarium , Lasiosphaeris , and Cercophora ( Figure S10 , Supporting Information ).Amongst these gener a, onl y Ramophialophora and Neosc hizothecium wer e found to be differ entiall y higher in the cluster of medium-performing cultivars compared to the other two clusters (Wilcoxon, P < .05)( Figure S11A , Supporting Information ).In addition, Pseudeurotium had a higher r elativ e abundance in cultivars used as renewable resources compared to the other cultiv ars, wher eas U. Sordariales showed a lower r elativ e abundance in starch and table potato as compared to the other cultivars (Wilcoxon, P < .05)( Figure S11B , Supporting Information ).
Cor e micr obiome anal ysis r e v ealed a stable bacterial cor e microbiome with 232 ASVs (50.5%) as part of the shared community (Fig. 6 A).The number of unique ASVs as well as their respectiv e r ead pr oportions within eac h gr o wth clusters w as quite small with 24 (2.1%), 3 (0.3%), and 30 (2.2%) unique ASVs in high-, medium-, and low-performing cultiv ars, r espectiv el y.Similar results were obtained for the fungal community where     the cor e micr obiome r epr esented 72 ASVs (75.7%) wher eby there was marginal or no influence of the growth clusters (Fig. 6 B).
The rhizosphere bacterial networks of high-and mediumperforming cultivars were larger compared to the cluster of lowperforming cultivars (Fig. 8 ).This difference was evident in the number of connected nodes within the r espectiv e networks: 86 nodes in high, 112 in medium, and 65 in low-performing cultiv ars.Additionall y, we found a positive relation between the potato growth clusters and bacterial network complexity, i.e. an increasingly higher network robustness (natural connectivity) and edge density from low-to high-performing cultivars (Table 3 ).Consequentl y, the cluster r epr esenting low-performing cultiv ars sho w ed a reduced bacterial co-occurrence netw ork ar chitecture, with the highest av er a ge path length observ ed.Ov er all, positiv e edges (Student test, P < .0001)r epr esented with gr een lines wer e pr e v alent, r anging fr om 87.06% to 94.16% (Table 3 ) in the different networks and seemed to be driven by U .Vicinamibacterales, U .Vicinamibacteraceae , U. Chthonomonadales , U. Rokubacteriales , Stenotrophobacter, U. Blastocatellaceae, and RB41 .These nodes identified as hub taxa wer e consistentl y observ ed in the different growth clusters with high co-occurrence within the same network modules (nodes of the same color).Mor eov er, we observ ed that hub modules were larger in high-and medium-compared to lo w-performing cultivars.Ho w ever, medium-and low-performing cultivars exhibited a topological structure with higher modularity and a greater number of modules compared to high-performing cultiv ars (Table 3 ).Inter estingl y, compar ed with high-performing culti vars, the n umber of modules containing less than four nodes was 3-fold higher in medium-and low-performing cultivars.Within the network of high-performing culti vars, n umerous Actinobacteriota lineages, including Lamia , Mycobacterium , Gaiella , Nocardioides , Agromyces , U. 67-14, U. MB-A2-108, Arthrobacter , U. Thermoleophilia, and Pseudonocardia were mainly distributed across thr ee modules.Mor eov er, these taxa co-occurred within their respective modules, but also exhibited positive interactions with some plant gr owth-pr omoting Pr oteobacteria, suc h as Massilia , Hyphomicrobium , and Microvir ga .T hese observ ations wer e ov erall consistent within the network of medium-performing cultivars .Con versely, for low-performing cultivars , the identified patterns were dispersed in the network and occurred within smallersized modules.Lastl y, the differ entiall y abundant gener a including Arthrobacter and Massilia were not observed in the network of low-performing potato cultivars.
Networks of fungal comm unities wer e also distinct between the thr ee gr owth clusters ( Figur e S13 , Supporting Information ).Notably, the number of nodes , modules , edge density, and the r obustness decr eased fr om high to low-performing cultiv ars ( Table S3 , Supporting Information ).Hub taxa were most common within high-and medium-pr eforming cultiv ars, and included Dichotomopilus , Dendryphion , Emericellopsis , Thielavia , Gibellulopsis , and U. Chaetomiaceae, whereas hub taxa of lowperforming cultivars were Myrmecridium , Penicillium , Pseudeurotium ,  Table 3. Topological features of bacterial co-occurrence network analysis in the rhizosphere according to plant growth-based clusters of potato cultivars.Fungal network properties in the rhizosphere of plant growth-based clusters of potato cultivars.Network property includes number of components, clustering coefficient, modularity, positive edge percentage, edge density, robustness , a verage path length, and number of modules.Pseudog ymnoascus , Tric hoderma , Scutellinia , and U. Pyr onemataceaeae ( Figure S13 and Table S3 , Supporting Information ).Interestingl y, the co-occurr ence patterns within the hub modules wer e stronger in high-and medium-than in low-performing cultivars ( Figure S13 , Supporting Information ).

Discussion
In this study, we assessed the comm unity structur e of rhizospher e bacteria and fungi of 51 distinct potato cultivars grown in the same soil under w ell-standar dized conditions in the greenhouse.Furthermore , we in vestigated microbial association networks and the ecological processes driving microbial community assembly in respect to the plant gr owth c har acteristics to understand how plant productivity governs the structure of these microbial communities.
Ov er all, the observ ed bacterial and fungal ASVs ric hness wer e mostly similar between the distinct potato cultivars used in this study.Mor eov er, the shar ed ASVs between all cultivars represented those with the highest read counts , i.e .around 50% and 75% in bacteria and fungi, r espectiv el y.This suggests that (i) a significant proportion of the potato microbiome regardless of genotypic differences is necessary for the plant when grown in the same soil, and (ii) this proportion may be v ariable fr om one community to another.Nevertheless, individual cultivars had the strongest influence on both bacterial and fungal community composition in the rhizosphere of potato, which is in line with pr e vious r eports wher e significant cultiv ar or genotype effects in the potato rhizospher e wer e shown (Gsc hwendtner et al. 2011, Inceo glu et al. 2012, Faist et al. 2023 ).
Compared with previous research involving a small number of distinct varieties, our approach with the 51 cultivars demonstrated that a significant part of the microbiome was lar gel y conserved in the potato rhizosphere regardless of the differences highlighted above between these genotypes.Using 51 cultivars, we were also able to define different classes of potato on the basis of plant gr owth par ameters, whic h so far has not been possible with a limited number of cultivars.To our knowledge, this is the first study linking the plant growth characteristics of various cultivars to micr obiome c har acteristics, in order to enhancing our understanding of the role of the rhizospher e micr obiome in potato productivity.

Microbial taxa driving plant performance in the potato rhizosphere
It is worth noting that the statement on plant productivity (low, medium, and high) made for the studied cultivars is specific to this study, and ther efor e is not indented to be generalized, as these cultivars may perform differ entl y when gr own in other soil and climatic conditions.Ne v ertheless, when they wer e gr ouped into plant growth-based clusters, we found a significant impact on rhizosphere microbial communities.Specifically, high-and medium-performing cultiv ars wer e dominated by certain bacterial ASVs whereas in the low-performing cultivars, ASVs were mor e e v enl y r epr esented amongst samples.Mainl y a higher r elative abundance of the genera Arthrobacter (Actinobacteriota) and Massilia (Pr oteobacteria) wer e observ ed in high-and mediumperforming cultivars compared to low-performing cultivars.Previousl y, pr oduction of sider ophor es and phytohormones such as indole acetic acid (IAA) were reported for Arthrobacter (Banerjee et al. 2010, Boukhatem et al. 2022 ), suggesting this genus encompasses a potential plant growth promoters.Another study evidenced an Arthrobacter strain to effectiv el y colonize rice roots thr ough biofilm formation, pr omoting later al r oot gr owth and carrying the potential of antimicrobial activities to w ar ds plant pathogens (Chhetri et al. 2022 ).The genus Massilia is a major rhizosphere and plant root-colonizing bacterium, capable to produce IAA, sider ophor es, and l ytic enzymes involv ed in the control of phytopathogens (Adrangi et al. 2010, Hrynkiewicz et al. 2010, Kuffner et al. 2010, Weinert et al. 2010, Ofek et al. 2012 ).Taken together, these data suggest that there may be a link betw een plant gro wth and specific genera inv olv ed in se v er al ways how to promote plant gr owth.Pr oteobacteria suggested as plant gr owth pr omoters including Brad yrhizobium , Microvirga , Variovorax , Hyphomicrobium , Sphingomonas , and Pseudomonas were not differentially abundant in the different growth clusters in our study.Ho w e v er, we found the genus Massilia , which has similar plant gr owth pr omoting tr aits as the afor ementioned gener a, in higher r elativ e abundance in the rhizosphere of high-performing cultivars suggesting support for H1 albeit for a different genus than expected.

Proteobacteria and Actinobacteriota synergy in the co-occurrence network of high-performing pota to cultiv ars
Additionall y, we demonstr ated that the rhizospher e bacterial comm unities wer e dominated by deterministic assembl y pr ocesses, mainly homogeneous selection (95%-98%) in the different growth clusters.Our findings are consistent with r esults fr om other studies where similar observations were reported for potato (Gu et al. 2022 ), but also for other plant species including soybean (Mendes et al. 2014, Zhang et al. 2018 ).This indicates that the consistent higher r elativ e abundance of certain beneficial bacteria taxa in the rhizosphere of high-and medium-performing cultivars and no dominance of these species in low-performing cultivars is the result of the homogeneous selection within each growth cluster.
Inter estingl y, some of the Proteobacteria members suggested as plant gr owth pr omoters ( Hyphomicrobium , Microvirga , and Bradyrhizobium ), the abovementioned genus Massilia as well as other Proteobacteria including Caenimonas , Skermanella , Janthinobacterium , and P edomicrobium co-occurr ed with man y Actinobacteriota lineages in high-and medium-performing cultivars.For e.g.Microvirga , pr e viousl y observ ed in potato rhizospher e (Pfeiffer et al. 2017 ) are known as nitrogen-fixing and plant gr owth-pr omoting bacteria (Ardley et al. 2012 ), while the genus Hyphomicrobium was associated with the suppression of Ralstonia solanacearum in tomato-cultivated soil microbiome (Wei et al. 2019 ).On the other hand, nodes belonging to Actinobacteriota present within these occurrence patterns included Arthrobacter , Agromyces , Pseudonocardia , Mycobacterium , Iamia , Nocardioides , Gaiella , and so on.The last tw o w er e found in the rhizospher e of Solanaceae, namel y tobacco, tomato, and potato cultivars (Hu et al. 2020, Chen et al. 2022, Mar-tins et al. 2023 ), and endosphere of maize and rice (Kämpfer et al. 2016, Wang et al. 2021 ), indicating their effective interactions with different plant species .T hese genera encompass species involved in N fixation and nitr ate r eduction (Tóth et al. 2008, Albuquerque et al. 2011, Wang et al. 2016, Nafis et al. 2019 ).These observations indicate that although most of beneficial Proteobacteria genera did not exhibit greater relative abundance in highperforming culti vars, the y potentially play a significant role in the recruitment of other beneficial bacteria, thus promoting the growth of these specific cultivars .T his was further illustrated by having Arthrobacter and Massilia , the two differ entiall y abundant genera interconnected, and this pattern was only observed within the network of high-and medium-performing cultivars.Further evidence for the significance of Proteobacteria is the exclusive presence of a module of diazotrophs (biological nitrogen-fixation micr oor ganisms) namel y Azohydromonas , Azospira , Azoarcus , and Dec hloromonas (Hur ek et al. 2002, Reinhold-Hur ek and Hur ek 2006, Rodrigues Coelho et al. 2008, Cheng et al. 2023, Guo et al. 2023 ) in the co-occurrence network of high-performing cultivars and to a lesser extent in medium-performing cultivars.In a pr e vious study, it was shown that reduced NPK inputs stimulated N-fixing activity and the number of diazotrophs in the potato rhizosphere (Volkogon et al. 2021 ).The presence of diazotrophic genera within the network of high-performance cultivars suggests they may start having a growing influence in the network due to a potential depletion of nitrogen by the plant.It is noteworthy to mention that plants r eceiv ed ∼1 kg ha −1 of N, whic h is far lo w er than the current field applications (180-252 kg N ha −1 ).Conversely, we found fewer and sparse co-occurrence patterns between these beneficial members affiliated to Proteobacteria and Actinobacteriota within low-performing cultivars .T his ma y indicate a less efficient network given that in a previous work, taxa positively correlated with yield were more abundant in the rhizosphere co-occurrence network of soybean with high productivity (Niraula et al. 2022 ).Additionally, the lo w est number of nodes, edge density, robustness, and particularly the highest av er a ge path length observed within this growth cluster suggest a network with limited effectiveness .T he modularity was ≥ 0.4 implying that all networks exhibited a modular topology (Newman 2006 ).Ho w ever, modules with less than four nodes in the network of low-performing cultivars was three times more important compared to those observed in high-performing cultivars.One may speculate that this could hav e r esulted in less efficient plant gr owth pr omotion.Bacterial communities being primarily governed by deterministic processes in low-performing cultivars, this implies that the reduced co-occurrence network arc hitectur e observ ed in the latter did not arise by chance, but rather was influenced by the cultivars themselves, since they exert the strongest influence on their rhizospher e micr obiome.Ov er all, we found support for H2, which pr edicted a lar ger co-occurr ence network with mor e intricate interactions within high-compared to low-performing cultivars of potato.
Further support for H2 is the largest number of hub taxa and their r espectiv e inter actions found in the network of highperforming cultivars.Ho w ever, despite the influence of growth clusters on these taxa, specific nodes in the hub modules primarily affiliated to Acidobacteriota (U .Vicinamibacterace, U .Blastocatellaceae, Stenotrophobacter , and RB41 ) emerged as core hub taxa, i.e. they were consistently present within the distinct networks .T his persistence suggests a fundamental and stable role of these Acidobacteriota-affiliated nodes in the structure of bacterial comm unities acr oss div erse contexts, particularl y because these taxa, whic h ar e highl y connected to others do not onl y fr equentl y act as gatek ee pers influencing ecosystem functions (Jiang et al. 2017 ), but also their r emov al r esulted in outsized impact on both the composition and functioning of the microbiome (Banerjee et al. 2018 ).Inter estingl y, pr e vious studies have reported on other Acidobacteria members as k e y players within co-occurrence networks in soil and plant-associated micr obial comm unities (Banerjee et al. 2016a,b , Jiang et al. 2017 ).In our study, Acidobacteriota was the most abundant phylum in the potato rhizosphere (U.Vicinamibacteraceae and RB41 sho w ed the largest node sizes, indicating their pr e v alence).These r esults ar e not aligned with pr e vious studies, which described the rhizosphere as an environment rich in carbon sources, thus favoring the pr olifer ation of copiotr ophs, particularl y Pr oteobacteria and Bacter oidetes (Ling et al. 2022 ).Similarl y, the anal ysis of the bar e soil prior to planting r e v ealed the same pattern observed in the rhizosphere , i.e .a prevalence of Acidobacteriota, suggesting that soil micr obial bac kgr ound is the primary contributor to the rhizosphere microbiome (Inceo glu et al. 2012, Liu et al. 2019, Veach et al. 2019 ).
Fungal communities were also driven by individual cultivars and the influence of growth clusters as observed in bacteria netw orks w ere consistent with fungal network anal ysis.Ov er all, we found that fungal comm unities wer e assembled by stochastic e v ents, mainl y undominated (75%-80%) in the potato rhizosphere.In a recent study (Gu et al. 2022 ), the analysis of fungal communities in (bulk) soil used for different potato cultivars demonstrated similar patterns (78.28%-98.99%),suggesting that stochastic e v ents ar e pr e v alent in fungal assembl y irr espectiv e of the compartment (bulk soil and rhizosphere).On the other hand, the difference between rhizosphere microbial communities has been related to the smaller body size of bacteria, their faster growth and higher dispersal rates than fungi, potentially resulting in relativ el y str onger deterministic pr ocesses for colonizing and establishing in new habitats (Po w ell et al. 2015, Aslani et al. 2022 ).In contrast, species with a low dispersal rate may be limited in their capacity to colonize different environmental niches, thus experiencing a lo w er influence of envir onmental selection on comm unity assembly (Leibold et al. 2004 ).T herefore , fungi likely underwent stochastic selection patterns (Nemergut et al. 2013, Zinger et al. 2019 ).
In conclusion, the gr owth c har acteristics of potato cultivars drive a deterministic and stochastic assembly of rhizosphere bacterial and fungal comm unities, r espectiv el y.Their influence was further exemplified in microbial co-occurrence patterns, which sho w ed larger netw orks with more intricate interactions within high-performing cultivars than in low-performing r epr esentatives .Within these networks , we found that high-performing cultiv ars featur ed unique positiv e inter actions between plant gr owthpr omoting Pr oteobacteria and Actinobacteriota, mainl y Massilia and Arthrobacter as well as a module of diazotrophs whereas in low-performing cultivars, none of these features were observ ed.This study demonstr ates that distinct cultiv ars with similar growth characteristics consistently recruit and structure their microbiome according to the patterns described abo ve .We also pinpointed microbial candidates that govern plant growth.Futur e r esearc h work aimed at ma pping r oot exudate patterns of these cultivars with the rhizosphere microbiome activity may provide a better understanding of how plant-microbe interactions can be used to impr ov e cr op pr oductivity.It is worth noting that our study involved a single timepoint sampling, thus providing a first insight into the assembly processes of microbial communities in a large number of potato varieties.Further studies including multiple sampling timepoints are required to comprehensiv el y addr ess forces that gov ern micr obial comm unity as-sembl y ov er a gr owing season and their implication for potato yield.

Figure 2 .
Figure 2. Cluster analysis using (A) k-means and (B) hierarchical clustering.51 different potato cultivars were grouped in three plant growth-based using shoot biomass , a v er a ge number of lea ves , tuber weight, and numbers upon harvest at 42 days after planting.k-means was performed using the lo w est within-sum of squares (means the sum of distances between the points and the corresponding centroids for each cluster) on the Euclidean distance.Hier arc hical was performed using ward.D2 linkage on the Euclidean distance.

Figure 3 .
Figure 3. Analysis of bacterial α-diversity.(A) Observed species richness, (B) Shannon's index, and (C) Simpson's dominance index of the rhizosphere bacterial community in plant growth-based clusters of potato.Boxplots display the medians, tops, and bottoms of the boxes r epr esent 75th and 25th quartiles, and whiskers outside this range; dots illustrate the individual observations in each cluster.Pairwise Wilcoxon test ( P < .05)was applied to calculate significant differences between potato clusters and numbers above the boxes indicate the corresponding P -values of the comparison.

Figure 5 .
Figure 5. Analysis of bacterial community composition.(A) Relative abundance of the top 10 classes in the rhizosphere of plant growth-based clusters of potato.(B) Pairwise comparison of bacterial classes and (C) bacterial genus Arthrobacter between the growth clusters and (D) breeding purpose.Boxplots display the medians, tops, and bottoms of the boxes r epr esent 75th and 25th quartiles, and whiskers outside this range; dots illustrate the individual observations in each cluster.Pairwise Wilcoxon test ( P < .05)was applied to calculate significant differences across sample groups and numbers above the boxes indicate the corresponding P -values of the comparison.

Figure 6 .
Figure6.Cor e micr obiome anal ysis.Venn dia gr am r epr esenting the shar ed and unique number of (A) bacterial and (B) fungal ASVs in the distinct growth clusters of potato cultivars (colored circles).brackets account for the respective read proportions associated with the ASVs.Core microbiome was defined as taxa consistently found in at least 80% of the samples with minimum relative abundance of 0.01%.

Figure 7 .
Figure 7. Comm unity assembl y pr ocesses dominating the rhizospher e bacterial comm unities in r esponse to the plant gr owth-based clusters of potato cultiv ars (A).Differ ences in βNTI, together with RCbr ay (B) serv ed to determine the comm unity assembl y pr ocess that dominated the gr owth clusters (C).The thresholds for the | βNTI | = 2 and | RCbray | are highlighted as horizontal red lines and the vertical dashed blue lines, r espectiv el y.Differ ences in βNTI are based on Wilcoxon rank-sum test and were considered significant when P -value < .05.

Figure 8 .
Figure 8. Bacterial co-occurrence network analysis in the rhizosphere according to plant growth-based clusters of potato cultivars.Nodes indicate the genera and edges represent robust ( P < .0001)Pearson correlation between a pair of nodes .T he green and red links represent positive and negative network inter actions, r espectiv el y.Modules r epr esent gr oup of interconnected nodes, eac h differ entl y color ed.Nodes with thic k blac k borders r epr esent hub taxa, whereas node sizes indicate their respective relative abundances.

Table 1 .
Name and c har acteristics of the 51 potato cultivars.

Table 2 .
PERMANOVA testing the r elativ e contribution of differ ent factors on the structur e of fungal and bacterial comm unities in the potato rhizosphere.