The field experiment was conducted from 2019 to 2020 at the Facility Viticulture Test Base (36°43′34″N; 103°16′24″E; altitude: 2100 m) (Figure 1) in Yongdeng County, Lanzhou City, Gansu Province, China. The test base is located in a semi-arid region with a typical continental monsoon climate with an annual average rainfall, evapotranspiration, and temperature of 290 mm, 1,000 mm, and 5.9°C, respectively. The experimental field soil is classified as loam, with the following initial properties: volumetric soil moisture content (θ_f) 29.2%, density 1.42 g cm⁻³, and pH 8.15.

“Red globe,” a 5-year-old Eurasian grape variety, was used as the test material, which was planted in a plastic greenhouse of 8 m × 80 m with an earth wall straw curtain. The rows of grape were orientated perpendicular to the longitudinal direction of the greenhouse. Each row (each treatment) contained eight grapes with a row spacing 2.0 m and the plant spacing 0.8 m. The cultivation method of a single-arm Y-shaped low single-hedge frame was adopted.

The growing period of grapes were divided into five stages in the light of local protected-cultivation grape water consumption and irrigation experience: germination stage (May 15–May 24), shoot elongation stage (May 25–June 22), flowering stage (June 23–July 15), fruit enlargement stage (July 16–September 14), and coloring maturity stage (September 15–October 18). Three treatments of auxiliary experiment were set up to investigate the effects of different water regulation on soil physicochemical properties and enzyme activities in greenhouse grape during the growing season in 2019 (Table 1): W0 (CK) treatment (the control treatment) was adequate water supply (soil moisture content was (75% ~ 100%) θ_f); W1 treatment was mild water stress (soil moisture content was (65% ~ 90%) θ_f); W2 treatment was moderate water stress (soil moisture content was (55% ~ 80%) θ_f). In order to further study the effects of adding bacterial fertilizer under different water regulation on the rhizosphere soil environment of grape in the greenhouse, five treatments of main experiment were set up in 2020 (Table 1): W0F0 (CK) treatment (the control treatment) was adequate water supply and no fertilization (soil moisture content was (75% ~ 100%) θ_f); W1F1 treatment was mild water stress and small fertilizer (soil moisture content was (65% ~ 90%) θ_f and adding 60 kg hm⁻² of bacterial fertilizer); W1F2 treatment was mild water stress and maximize fertilizer (soil moisture content was (65% ~ 90%) θ_f and adding 120 kg hm⁻² of bacterial fertilizer); W2F1 treatment was moderate water stress and small fertilizer (soil moisture content was (55% ~ 80%) θ_f and adding 60 kg hm⁻² of bacterial fertilizer); W2F2 treatment was moderate water stress and maximize fertilizer (soil moisture content was (55% ~ 80%) θ_f and adding 120 kg hm⁻² of bacterial fertilizer). The experimental plots in both years were in a randomized block design with three replications for each treatment and plot size of 8 m × 2 m.

A drip irrigation system with a “one pipe per row” configuration was used in the greenhouse grape field trial, with emitters delivering water at a rate of 3 L h⁻¹. The irrigation amount and irrigation time were determined by the soil moisture content and measured using the water meter which was installed in each plot along with the valves. When the soil moisture in the field reached the lower limit of the experimental design, the irrigation was performed and stopped at the upper limit. The soil moisture ratio was maintained at 0.5, with a planned wetting depth of 80 cm. To monitor and maintain water stress levels in the greenhouse, soil water content was measured by the drying method, with soil samples taken every 7 d throughout the reproductive period and additional measurements taken before and after irrigation when soil water content approached the lower limit of irrigation control (Table 1), and a polyethylene geomembrane with a thickness of 2 mm was laid between the plots (to the soil depth of 1 m) to prevent the diffusion of soil water between different plots.

Pseudomonas fluorescens had the ability to antagonize many plants pathogenic fungi and was the dominant phosphorus bacterium in the soil. Pseudomonas fluorescent bacterial fertilizer was chosen in the experiment for its broad-spectrum bacteriostatic and growth promoting effects, which was purchased from Shanghai Ruichu Biotechnology Co., Ltd. ¹ with the brand REBIO, and the product number Y0024. The composition of the ingredients per kg of bacterial fertilizer was as follows: 55 g of fluorescent pseudomonas solution (1 g of fluorescent pseudomonas solution contains ≥10 billion live bacteria), 110 g of corn flour, 110 g of soybean cake powder, 73 g of bran powder, 73 g of K₂HPO₄, 29 g of KH₂PO₄, 3 g of MgSO₄, and 547 g of peat. To ensure uniformity of fertilizer application, Pseudomonas fluorescens was mixed with sand at a mass ratio of 1:50, and then applied in three separate applications approximately 20 cm around the roots of each grape plant in each plot at the beginning of germination period (May 15th), the flowering period (June 16th) and the fruit expansion period (August 16th), respectively. Pruning, fertilization, pest and disease control, weeding and other field management practices for each plot were uniformly carried out in accordance with local viticultural management experience, among which the fertilization measure was as follows: on February 24th, basal fertilizer (chicken manure; 5,000 kg hm⁻²) was applied along with 2 kg diammonium phosphate and 4 kg ammonium bicarbonate; On June 16th, each treatment received 1 kg diammonium phosphate, 0.8 kg calcium ammonium nitrate for agriculture, 0.8 kg of organic fertilizer, and 0.5 kg of potassium magnesium sulfate, respectively; On August 16th, 0.8 kg diammonium phosphate, 0.8 kg calcium ammonium nitrate for agriculture, 0.8 kg organic fertilizer, and 0.6 kg potassium and magnesium sulfate were applied.

Soil samples were collected during different grape growth periods in both 2019 (May 19th, June 18th, July 15th, August 20th, and October 20th) and 2020 (May 15th, June 15th, July 20th, August 15th, and October 15th). The surface litter and impurities were moved away, and then soil samples (depth of 20–40 cm) of 10 cm away from grape roots from three random regions of the plot were collected by using a core sampler (20 mm internal diameter). After the root system and other impurities were picked out, the soil samples of the same treatments were thoroughly mixed and sieved (2 mm mesh), and then divided into three parts, of which one part was stored at 4°C for analysis of enzyme activity, one part was frozen in refrigerator of −20°C for the extraction of soil macrogenomic DNA, and the third part were naturally air dried for the determination of physicochemical properties. The measurement of each index was completed within 1 month after sampling was completed.

Total nitrogen (TN), nitrate nitrogen (NO₃⁻-N), and ammonium nitrogen (NH₄⁺-N) contents were measured by the Kjeldahl method (Boell and Shen, 1954). Soil organic matter (SOM) was determined by the potassium dichromate external heating method (Wang et al., 2023). Total phosphorus (TP) and available phosphorus (AP) were analyzed by a UV2450 ultraviolet spectrophotometer (Shimadzu, Japan) (Li et al., 2022). The soil total organic carbon (TOC) content was measured using a carbon and nitrogen combined analyzer (Multi N/C 2100 s, Jena, Germany) after removing inorganic carbon with 0.5 mol L⁻¹ dilute hydrochloric acid. The dissolved organic carbon (DOC) content was determined with a carbon and nitrogen combined analyzer (Multi C/N 2100 s) after leaching with ultrapure water (water: soil = 5: 1). The microbial biomass nitrogen (MBN) and microbial biomass carbon (MBC) content were determined using a carbon and nitrogen combined analyzer after 0.5 mol L⁻¹ K₂SO₄ extraction (Multi C/N 2100 s) (Xu et al., 2016).

Soil urease activity was determined by the Kits of Art. No. G0301F (Suzhou Gerisi Biological Technology Co. LTD) through spectroscopic method. To determine the sucrase activity, 5 g of air-dried soils were incubated for 24 h at 37°C with 15 mL of 8% sucrose, 5 mL of phosphate buffer at pH 5.5, and 0.1 mL of toluene. The glucose released by sucrase reacted with 3-5-dinitrosalicylic acid and then was measured based on the absorbance at 508 nm (UV-2450, Shimadzu Corporation, Kyoto, Japan) (Wu et al., 2020). The results were expressed as mg glucose g⁻¹ h⁻¹. Soil catalase activity was determined by the KMnO₄ liquid titration method and expressed as the volume of 0.02 mol L⁻¹ KMnO₄ consumed of 2 g air-dried soil within 20 min (Li et al., 2014).

For the extraction of soil macrogenomic DNA was the soil samples at three time points (June 15th, August 15th, and October 15th, respectively) in 2020 corresponding to shoot elongation, fruit enlargement, and early coloring maturity stage, which were the three peaks of nutritional and reproductive growth of table grapes.

The total genomic DNA in each sample (2.5 g field-moist soil) was extracted using hexadecyltrimethy ammonium bromide (CTAB) method (Zhou et al., 1996). DNA concentration and purity was monitored on 1% agarose gels. According to the concentration, DNA was diluted to 1 ng μL⁻¹ using sterile water. The phusion^® High-Fidelity PCR Master Mix with GC Buffer (the New England Biological Laboratory Company) and the high-fidelity enzyme was used for the polymerase chain reaction (PCR). The V3–V4 region of the bacterial 16S rRNA gene was amplified using the PCR primers 515\u00B0F (5′-GTGCCAGCMGCCGCGGTAA-3′) and 806 R (50-GGACTACHVGGGTATCTAAT-30) (Guo et al., 2017), and the ITS2 region of fungi was performed using primers ITS3_KYO2F (50-GATGAAGAACGYAGYRAA-30) and ITS4-2409R (50-TCCTCCGCTTATTGATATGC-30) (Toju et al., 2012). The products from PCR amplification labeled with a barcode sequence were purified using GeneJET™ Gel Extraction Kit (Thermo Scientific) after electrophoresis in 2% agarose gel. Then, the 16S rRNA library was prepared by Ion Plus Fragment Library Kit 48 rxns (Thermo Scientific). Purified amplicons were sequenced on an Ion S5TM XL platform. Each sample data was separated from the off-line data according to the barcode sequence and PCR amplification primer sequence. After the barcode and primer sequence were cut off, the reads of each sample were spliced using FLASH (V1.2.7, http://ccb.jhu.edu/software/FLASH/) in order to obtain raw tags data (Magoč and Salzberg, 2011). Then, the high-quality tags data (clean tags) was obtained from strict filtering of raw tags using QIIME (V1.9.1, http://qiime.org/scripts/split_libraries_fastq.html) (Bokulich et al., 2013). After removing the chimeric sequence (http://www.drive5.com/usearch/manual/chimera_formation.html), compared with the database (Gold database, https://www.drive5.com/usearch/manual/chimera_formation.html) to detect the chimeric sequence, the final effective data (Effective Tags) was obtained. All effective tags of soil samples were clustered by Uparse software (Uparse v7.0.1001, http://drive5.com/uparse/). The sequences were clustered into OTUs (Operational Taxonomic Units) with 97% identity. At the same time, the sequence with the highest frequency in OTUs was selected as the COMMUNICATIONS IN SOIL SCIENCE AND PLANT ANALYSIS 5 representative sequence of OTUs, and the representative sequences of OTUs were annotated by using the Mothur method (http://www.arb-silva.de/) (Wang et al., 2007) and Silva’s SSUrRNA database (the threshold was set at 0.8 ~ 1) (Quast et al., 2013) to obtain taxonomic information of soil samples and make statistics on the community composition of each soil sample at each microbial community classification level.

The multi-sequence alignment was performed using MUSCLE software (Version 3.8.31, http://www.drive5.com/muscle/) to obtain the phylogenetic relationship of all OTUs representative sequences (Edgar, 2004). Finally, the data of each sample was homogenized according to the standard of the least amount of data in the sample for the subsequent analysis of microbial community diversity. The alpha diversity values of grape rhizosphere soil bacteria, such as the species richness index (ACE, and Chao), and the species diversity index (Simpson, and Shannon) were calculated using the Qiime software (Version 1.9.1).

Statistical analyses were performed using Microsoft Excel 2019 software (Microsoft Corporation). The SPSS Statistics 26.0 software (IBM, Armonk, New York, United States) with the factorial analysis of variance (ANOVA) was applied for statistical validation and plotting was performed using Origin 2024 software (Origin Lab Corporation). The Duncan’s multiple range tests were undertaken for comparing all treatment means for significant difference. A redundancy analysis (RDA) was used to analyze the relationship between microbial community distribution and environmental explanatory variables, enzyme activities based on OTU level. Correlations among soil physicochemical properties, enzyme activities, and microbial biomass were assessed using Pearson correlation analysis.

The W1F2 treatment exhibited the highest yield (Figure 2), which was 35.44% higher and significantly higher (p < 0.05) than the CK treatment. No other treatments demonstrated a significant difference (p < 0.05) from the CK treatment (Table 2). At the same water stress gradient, both water use coefficient and irrigation water use coefficient increased with increasing fertilizer application. The water use coefficients of W1F2 and W2F2 treatments were significantly higher (p < 0.05) than that of CK control, but the differences with other treatments were not significant (p < 0.05); the irrigation water use coefficients of all treatments were significantly higher (p < 0.05) than that of CK treatment. Compared to CK treatment, W1F1, W1F2 and W2F2 treatments increased economic benefit (Equation (3)) by 19.31, 44.04, 25.31% (Table 2), respectively, but W2F1 treatment decreased by 7.21%. Compared with CK treatment, better economic benefits were W1F2 and W2F2 treatments.

At the coloring maturity stage under water stress conditions in 2019 (Figures 3A,B), AP content decreased and then increased with increasing water stress gradient and was lower in both W1 and W2 treatments than in CK treatment, but the difference was not significant (p < 0.05) (Figure 3A), so there was a specific adaptation of AP content in the soil to water stress. While soil MBC, MBN and DOC contents were significantly lower (p < 0.05) in both W1 and W2 treatments compared with CK treatment (Figure 3B). It can be seen that water stress significantly suppressed soil MBC, MBN and DOC content, but had less effect on other physicochemical indexes.

Microbial fertilization additions significantly affected changes in rhizosphere soils AP, DOC, MBC and MBN (Figures 3C,D). Soil AP, DOC, MBC, and MBN contents were all highest under the W1F2 treatment, and all three indexes except AP were significantly higher than those of the CK treatment (p < 0.05). AP, MBC content among treatments showed W1F2 > W1F1 > CK > W2F2 > W2F1, indicating that the addition of bacterial fertilizer under mild stress could promote the accumulation of AP, MBC content, and the effect was more obvious with the increase of the amount of bacterial fertilizer. MBN, DOC contents showed W1F2 > CK > W2F2 > W1F1 > W2F1 among treatments, indicating that the increase in the amount of bacterial fertilizer promoted MBN, DOC contents more significantly under the same water stress gradient.

Urease, catalase and sucrase activities were significantly lower (p < 0.05) in W2 treatment compared with CK during coloring maturity stage (Oct-15) (Figures 4A–C). In addition, the activities of various enzymes were slightly higher under the W1 treatment than under the W2 treatment from the germination stage (May-19) to the coloring maturity stage (October-15).

The changes in urease, catalase and sucrase activities were more pronounced with the addition of microbial fertilizer (Figures 4D–F). At the coloring maturity stage (Oct-15), all three enzyme activities were significantly higher (p < 0.05) in the W1F2 treatment than in the CK treatment. The W1F2 treatment increased soil urease activity at all fertility stages compared to the CK treatment and the differences were significant except at flowering stage.

Based on 16S rRNA and ITS gene sequencing analysis, we obtained a total of 1,201,770 and 1,183,320 effective tags from all 15 samples after the quality control. After OUT clustering, 8,754 and 2,028 OTUs were remained for bacterial and fungal community, respectively. The rarefaction curve analysis of three replicates mean sequences of each treatment in grapevine coloring maturity stage showed that the data volume of sequenced reads was reasonable and that increasing the number of reads made only a small contribution to the total number of OTUs, because all rarefaction curves approached the saturation plateau (Figure 5). The number of observed OTUs of all treatments in the coloring mature period was relatively stable. The diversity index of bacteria was calculated in Table 3. There was no significant difference (p < 0.05) in both of fungi and bacteria Simpson, Chao1, and Shannon at each treatment. Compared with CK, all treatments increased fungal Chao1 index. And the Shannon and Simpson index increased except W1F2 treatment. For bacterial diversity, compared with CK, the Shannon, Simpson and Chao1 index in all treatments decreased, except for Chao1 index, which was increased under mild water stress (W1F1 and W1F2 treatments). And under the same water condition, with the increase of bacterial fertilizer amount, Chao1 index increases, but Shannon index decreases in all treatments.

Replicates from different treatments were largely clustered together indicating good sample reproducibility (Figure 6). The first axis of the PCoA analysis (Figures 6A,B) based on weighted unifrac distance for bacterial (a) and fungal (b) communities explained 23.09 and 33.11% of the variation, respectively. The bacterial communities were distinctly separated between the treatments with added bacterial fertilizer (as W1F1, W1F2 and W2F1) and CK, except for W2F2 (Figure 6A). The fungal communities however did not show a distinct separation among all treatments (Figure 6B). However, compared with CK treatment, soil bacterial and fungi communities were more heterogeneous with the addition of microbial fertilization, indicating variability in soil microbial communities. The NMDS (Figures 6C,D) based on bray-curtis distance revealed a clear separation among all treatments for both the 16S rRNA gene (c) and ITS gene (d) except W2F2 treatment, and showed that there were differences in the bacterial (R = 0.3333, p = 0.05) and fungal (R = 0.4763, p = 0.05) community structure between all treatments and CK according to Anosim test.

The results of UPGMA (Unweighted Pair-group Method with Arithmetic Mean) cluster analysis of bacteria (phylum (a) and genus (c)) and fungi (phylum (b) and genus (d)) based on weighted unifrac distance in grape rhizosphere soil (Figure 7) showed that there were significant differences in the structure of bacteria and fungi in grape rhizosphere soil at phyla and genus levels under water stress and bacterial fertilizer addition.

Bacterial gene sequences were affiliated with 68 phyla and 568 genera. The dominant phylum in all soil samples was Proteobacteria, followed by Bacteroidetes, Gemmatimonadetes, Actinobacteria, Acidobacteria, Planctomycetes, Chloroflexi, Thaumarchaeota, Firmicutes, Candidatus_Lloydbacteria, representing average of 42.2, 9.9, 8.9, 5.9, 4.7, 4.6, 4.5, 2.3, 0.9, 0.02% of all sequences, respectively (Figure 7A). There were not significant differences in the relative abundances of all bacterial genes among the five treatments. However, for the top 10 phyla, W1F1, W1F2 and W2F1 treatments increased the relative abundance of Actinobacteria and the decrease was observed in W2F2 treatments, while that of Thaumarchaeota was significantly decreased by W1F1, W2F1 and W2F2 treatments, and increased in W1F2 treatment. Additionally, all treatments increased the relative abundance of Proteobacteria and Firmicutes and decreased of Acidobacteria compared with CK. At the genus level (Figure 7C), the top 10 genera in all soil samples were Lysobacter, Pseudomonas, Nitrosospira, Candidatus_Nitrosoarchaeum, Chryseolinea, Sphingomonas, Terrimonas, Qipengyuania, Bacteroides, Faecalibacterium, together representing an average of 13.1% of all sequences. That is, bacterial gene species were very rich at the generic level (Others accounted for 86.9%). When compared with CK, all treatments increased the relative abundances of Lysobacter, Nitrosospira, Chryseolinea, Terrimonas, while decreased that of Pseudomonas and Faecalibacterium, and significant difference (p < 0.05) in the relative abundances of Pseudomonas was observed between W2F1 treatment and CK.

Fungal gene sequences were affiliated with 11 phyla and 263 genera. The dominant phyla in all soil samples were Mortierellomycota, Ascomycota, Blastocladiomycota, Basidiomycota, Rozellomycota, Aphelidiomycota, Chytridiomycota, Olpidiomycota, Mucoromycota, Glomeromycota, together representing an average of 56.6% of all sequences (Figure 7B). There were significant differences (p < 0.05) in the abundance of Glomeromycota between W2F1 treatment and CK. All treatments significantly decreased the abundance of Basidiomycota and Aphelidiomycota, while increased that of Ascomycota, Blastocladiomycota, Chytridiomycota, Olpidiomycota, Mucoromycota, Glomeromycota. At the genus level (Figure 7D), the top 10 dominant genera were Mortierella, Brachyphoris, Geomyces, Dactylonectr, unidentified_Mortierellomycota, Pseudogymnoascus, Beauveria, Exophiala, Metarhizium, Talaromyces, representing approximately 28.9% of all sequences. There were significant differences (p < 0.05) in the abundances of Mortierella between W2F1, W2F2 treatments and CK, extremely significant differences (p < 0.01) in the abundances of unidentified_Mortierellomycota between W2F2 treatment and CK, and all treatments decreased the relative abundances of Mortierella compared with CK.

Redundancy analysis (RDA) showed that SOM and TOC were key driving factors for the bacterial communities in the rhizosphere soils (Figure 8A), while TP, MBC, and MBN were the highly associated factors. For fungi, NO₃⁻-N and TN were key driving factors (Figure 8B). The correlations (Spearman) heatmap results were shown in Figure 9. In addition to the negative correlation between NO₃^-_N and TN, SOM, TOC, between TN and TP, AP, positive correlations were found between all of other soil physicochemical indicators. Among them, AP was significantly positively correlated with TP, MBC (p < 0.05); SOM was significantly positively correlated with TOC (p < 0.05); and MBN was positively correlated with DOC (p < 0.05). The correlation analysis between soil physicochemical indicators and soil enzyme activities showed that there were positive correlations between soil urease, catalase and physicochemical indicators to varying degrees, while there was a negative correlation between sucrase and soil physicochemical indicators. Among them, urease and catalase were significantly positively correlated with MBC, AP and TP (p < 0.05); sucrase was significantly negatively correlated with TP and AP (p < 0.05). Furthermore, the Acidobacteria abundance correlated negatively (p < 0.05) with NH₄⁺-N, the Chloroflexi abundance correlated negatively (p < 0.05) with TN, and the Planctomycetes abundance correlated negatively (p < 0.05) with MBC and TP, while the Firmicutes abundance correlated positively (p < 0.05) with TOC and SOM. For fungi, the Ascomycota abundance correlated positively (p < 0.05) with TN, the Mortierellomycota abundance correlated positively (p < 0.05) with TP, while the Rozellomycota abundance correlated negatively (p < 0.05) with MBC. There was a correlation between soil enzyme activities and the relative abundance of bacteria and fungi, with sucrase being positively (p < 0.05) correlated with the Planctomycetes abundance and negatively (p < 0.05) correlated with the Mortierella abundance. It is worth noting that urease and catalase had little effects on Acidobacteria.