The research area is situated in the permafrost zone of the Beiluhe Basin on the central Tibetan Plateau. According to meteorological data, the Beiluhe Basin has an arid environment with an average yearly temperature in the range of −5.4°C to −3.6°C and mean annual precipitation of 369.8 mm, which is much less than the region’s mean annual evaporation of 1317 mm (Peddle and Franklin, 1993; Ni et al., 2021). Stipa purpurea and Carex rigescens dominate the alpine steppe site (Zhang et al., 2013). There is no flora growing on the alpine desert’s surface. The area has continuous permafrost, with a maximum depth of 3 m and an average yearly ground temperature of −0.9°C (Niu et al., 2005; Yu et al., 2015). During the thawing season, the active layer arises. Human intervention in this region is minimal, and the region is mostly unaltered.

At the Beiluhe Long-term Permafrost Research Station (34°50′57.42″N, 92°54′23.96″E, 4,694 m), nine locations with two types of ecosystems were 300–400 m apart and the collection of 27 soil samples occurred in mid-September 2021. The soil from each sampling site was collected in three layers: 0–5 cm, 20–30 cm, and 50–60 cm (Chen et al., 2021). Five sampling sites were chosen from the alpine desert and four were selected from the alpine steppe (Figure 1). Three soil replicates from each sampling site were merged into a single replicating sample using a soil auger with a diameter of 5 cm. Each collected soil sample was sieved (with a 2 mm mesh) to remove big particles and surface vegetation before being split into two subsamples. Fresh samples were used to determine the soil water content (SWC). The remainder of the subsample was air-dried to determine the soil’s physical and chemical characteristics, namely pH, SOC content, soil texture, and mineral composition. The first subsample was kept at −20°C for PLFA analysis (Shao et al., 2022).

Samples were dried for 24 h at 105°C in an oven before being given time to absorb as much moisture as they could. The SWC of each soil sample was calculated by gravimetric analysis. A 1:2.5 (w/v) mixture of soil and deionized water was used to measure the pH of the soil using an electrode from a pH meter (PB-10, Sartorius, Germany) (Liu et al., 2020). In a 1:5 (w/v) mixture of soil and deionized water, the electrical conductivity (EC) of the soil was measured using a DDS-307A conductivity meter (Precision and Scientific Corp., Shanghai, China). A CS-902G analyzer was used to find the SOC content after the samples were acidified with 7% HCl to get rid of carbonates and rinsed with deionized water till neutrality (Zhang et al., 2022). The total carbon content (TC) in the soil was directly measured using the CS-902G analyzer, and SIC was calculated using the following formula: SIC = TC – SOC. The total nitrogen (TN) content in the soil samples were determined using a Kjeldahl nitrogen meter (SKD-200, Shanghai Peiou Analytical Instruments Corp., China), and C/N was calculated. Moreover, the chemical index of alteration (CIA) of all the samples were calculated using the formula: CIA = [Al₂O_3/(Al₂O₃ + CaO + K₂O + Na₂O)] × 100 (Parker, 1970). The supernatant of soil samples after pressurized acid digestion and the deionized water treatment (v/v = 1:5), were measured separately by inductively coupled plasma-optical emission spectroscopy (ICP-OES, Optima800, PerkinElmer, USA) to determine major element concentrations, i.e., Ca, Mg, Fe, Al, K, Na, Ti, and P (Yang et al., 2016).The Cl^–, NO^– 3, and SO2^– 4 contents were determined by ion chromatography (ECO IC, Metrohm, Swiss), and the CO2^– 3 and HCO^– 3 contents were determined according to a Chinese industrial standard (DZ/T 0064. 49-93).

After removing organic matter and carbonates with hydrogen peroxide and HCl, respectively, the soil texture (clay/silt/fine sand/coarse sand%) was categorized in accordance with the international system using a particle size analyzer (Malvern Masterizer 2000, UK) in a measuring range of 0.02–2,000 μm. Mineral composition was evaluated using CuK radiation at 40 kV and 40 mA while operating at divergence, scattering, and reception slits of 1° and 0.15 mm, respectively. The scanning angle spanned from 2° to 52°, the step interval was 0.02°, and the pace was 4°(2)/min. MDI Jade 5 software was used to calculate the mineral contents and percentages of the samples. The Chinese industrial standard SY/T 5163-2010 was used to calculate the mineral content (Sun et al., 2020).

The δ¹³C analysis of carbonates was performed using a gas stable isotope mass spectrometer (Thermo Scientific Delta V – GasBench II). Before measurement, plant residues and microbiological shells were picked out under a magnifying glass. After that, each sample was sieved less than 200 mesh. The treated sediment samples were loaded into glass vials and placed in an aluminum heating block of GasBench II. Then the samples (∼0.1 g) were purged with helium for 7 min followed by a treatment with supersaturated phosphoric acid (98%) for 2 h; the reaction temperature was 72°C. Ultimately the CO₂ generated by the reaction was carried into the Delta V for detection using helium as the carrier gas. NBS-18 was used as the international reference material, while GBW04405 was used as the national reference material. The δ values are reported in per mil relative to the Vienna Pee Dee Belemnite, with an error less than 0.1‰.

We estimated microbial abundances and community composition by PLFAs. PLFAs were isolated from lyophilized soil biomass using the single-phase Bligh and Dyer method (Qin et al., 2021). The combined soil sample (5 g) was extracted twice with a single-phase mixture of chloroform, methanol, and phosphate buffer (1:2:0.8, v/v/v; pH = 7.4), and the extracted sample was vigorously shaken each time for at least 2 h. Water and chloroform in equal parts were added to the solution to divide it into two phases. By sequentially eluting the organic phase with chloroform, acetone, and methanol, the neutral lipids, glycolipids, and phospholipids were extracted from the organic phase on a silica acid column. The phospholipids were moderately alkaline methanolyzed to create fatty acid methyl esters, and the PLFA samples were then kept at −20°C until analysis. The amount of n-tetracosane-D50 (C₂₄D₅₀), which was added as an internal standard, was the same in each test sample. The isolated PLFAs were then identified and quantified using the MIDI Sherlock Microbial Identification System (Agilent 7890A-5977N) (MIDI Inc., Newark, DE, USA). Bacteria (i14:0, a15:0, i16:0, 16:17c, a17:0, cy17:0, i17:0, 18:17, and cy19:0) and fungi (18:26 and 18:29c) were quantified using particular biomarkers, and the microbial community composition was analyzed using the ratio B/F (Qin et al., 2021). G⁺ PLFAs included i14:0, i15:0, a15:0, i16:0, i17:0 and a17:0; and G^– PLFAs included 16:1ω7c, cy17:0 and cy19:0 (Bai et al., 2017). Moreover, the ratio of monounsaturated to saturated fatty acids (MUFA/SFA) was used as a stress indicator to indicate microbial response to environmental stresses (Pivnickova et al., 2010; Sun et al., 2018).

The Kolmogorov–Smirnov test and Levene’s test were used to check the normality and homogeneity of the variance of data before conducting statistical analyses. All components were log-transformed because they did not follow a normal distribution, with the exception of pH, SWC, EC, TN, C/N, CIA, MUFA/SFA, coarse sand, and fine sand. To consider multicollinearity among variables before multiple linear regression, the variance inflation factor (VIF) was calculated; acceptable collinearity is defined as a VIF of less than 2 (Craney and Surles, 2002; Qin et al., 2019). SPSS 26.0.0.0 was used for the statistical studies mentioned above at a significance threshold of 0.05.

The following three steps were involved in the statistical analysis after that. First, we looked at significant variations in soil physicochemical parameters and PLFA values between the three depths and two ecosystems using mixed-effects models (R package: nlme). Second, the Pearson correlation was used to examine the relationship between soil environmental properties. Third, linear regression models were used to determine the relationships between calcite content and SIC and variables with significant differences in soil depth.

Also, multiple linear regression models were employed to examine soil calcite content and SIC with all the variables by using maximum likelihood estimation, which included the variables significantly affecting calcite content and SIC in pairwise analyses. The absolute values of the standardized regression coefficients of the explanatory variables accounting for the percentage of the sum of all standardized regression coefficients were used to calculate the relative effect of all parameter estimates to assess the relative significance of the drivers in predicting calcite content and SIC. The contribution of various types of predictor factors to a dependent variable was comparable to the simultaneous testing of each predictor variable in the model. Then, to quantify their relative contribution to calcite content and SIC in different soil depths across all sites, 21 predictor variables were categorized into four groups, namely edaphic factors (pH, SWC, EC, SOC, TN, C/N, and CIA), soil texture (coarse sand, fine sand, silt, and clay), soil minerals (the contents of quartz, k-feldspar, plagioclase, and clay mineral), and microbial factors (B/F, G⁺/G^–, MUFA/SFA, microbial biomass, bacterial biomass, and fungal biomass).

Based on the results of the mixed-effects models, SIC and soil calcite content did not exhibit significant differences between the two vegetations (Supplementary Table 1). Soil pH, quartz content, and coarse sand content in the alpine desert were significantly higher than those in the alpine steppe (all p < 0.05, Supplementary Table 1), whereas SWC, EC, SOC content, plagioclase content, fine sand content, microbial biomass, bacterial biomass, and fungal biomass showed the opposite results (all p < 0.05, Supplementary Tables 1, 2).

Four kinds of soil texture were identified: clay (< 2 μm), silt (2–20 μm), fine sand (20–200 μm), and coarse sand (200–2,000 μm). In all samples, clay made up 0.1–11.8%, silt made up 2.3-17.7%, fine sand made up 44.3-87.7%, and coarse sand made up 1.5–43.4% (Supplementary Table 2). Analysis of the soil’s mineralogy revealed that all samples had similar mineralogical makeup (Supplementary Figure 1). All samples contained quartz, which was predominant (55.5–91.3%), along with calcite (3.5–20.4%), plagioclase (2.4–15.2%), K-feldspar (0.0–10.5%), and clay minerals (0.0–12.6%) (Supplementary Table 1). The results of the soil particle size analysis were consistent with the findings that the clay mineral concentrations of 22 samples were below the instrument’s detection limit (Supplementary Table 2).

The SWC of the alpine desert ranged from 3.5 to 23.4%, while that of the alpine steppe ranged from 6.7 to 29.4%. All 27 soil samples were alkaline in nature, with pH values of 8.3–9.0. The SOC content varied from 0.4 to 8.3‰ in the alpine desert and from 1.4 to 6.8‰ in the alpine steppe (Table 1 and Supplementary Table 1). However, there were no significant differences in soil properties (including calcite content and SIC) among the three layers, except for the MUFA/SFA value in soil (p = 0.035) (Figure 2). The mean MUFA/SFA values for soil layers 0–5 cm, 20–30 cm, and 50–60 cm were 0.50, 0.33, and 0.20, respectively (Table 1 and Supplementary Table 1). This indicated that the available carbon resources in soil decreased significantly with the increasing soil depth. Moreover, the mean B/F value in the 50–60 cm soil layer (mean = 9.98, Table 1) was significantly higher than that in the 0–5 cm soil layer (mean = 4.64, Table 1; p = 0.045). With increasing in soil depth, the microbial community structure changed significantly, and the bacterial biomass increased significantly compared with that of fungi (Kieft et al., 1997).

We established a relationship between carbonate content and soil properties in all samples. The calcite content in soil was significantly associated with the quartz, k-feldspar, and silt contents (all p < 0.05, Figure 3). Particularly, soil calcite content was positively correlated with SIC and the B/F values (p = 0.011). Moreover, there was also a significant correlation between SIC and quartz content in soil (p = 0.007, Figure 3).

To better understand the relationship between carbonate content and microbial community structure, linear regression models were used to characterize the correlation between the two for each soil layer. In the 0–5 cm and 50–60 cm soil layers, calcite content and SIC had no significant correlation with B/F and MUFA/SFA. In the 20–30 cm soil layer, calcite content and SIC had a significant positive correlation with B/F (Figures 5A, B) and a significant negative correlation with MUFA/SFA (Figures 4C, D).

Furthermore, we tested the carbon isotopes of soil carbonate minerals to confirm the influence of microorganisms on soil carbonate. The results showed that the δ¹³C values in soil carbonate were significantly associated with plagioclase content and bacterial biomass (Figures 5A, B). Lower δ¹³C values of soil carbonate were reported, the increasing plagioclase mineral content and bacterial biomass in the soil.

The dominant factors affecting the soil carbonate and SIC contents were different among the three soil layers. Specifically, for the 20–30 cm soil layer, the linear model revealed that microbial and mineral factors largely contributed to the variations in the calcite content, followed by edaphic factors, and soil texture contributed the least (Figure 6D). Among these factors, B/F, SOC, and quartz content exhibited larger effects on the calcite content compared with other predictor factors (Figure 6B). For the 0–5 cm soil layer, the linear model showed that edaphic factors largely explained the variations in the content of calcite, followed by soil texture and microbial factors and lastly soil mineral (Figure 6D). Among these factors, SWC and the coarse sand content had larger contributions to the variation in calcite content than others (Figure 6A). For the 50–60 cm soil layer, the linear model showed that soil texture largely explained the variations in the calcite content, followed by edaphic factors, and lastly microbial factors (Figure 6D). Among these factors, silt content had larger contributions to the variation of calcite content than others (Figure 6C).

Similarly, linear models showed that edaphic factors and soil texture largely explained the variations in SIC for the 0–5 cm and 50–60 cm soil layers, respectively (Figure 6E). The difference, however, was that pH exhibited larger effects on SIC compared with other predictor factors in the 0–5 cm soil layer (Figure 6A). Moreover, edaphic factors largely explained the variations in SIC in the 20–30 cm soil layer, followed by soil mineral and soil texture and lastly microbial factors (Figure 6D). With increasing soil depth, the influence of edaphic factors on SIC decreased while the influence of microorganisms on SIC increased.

Due to considerably stable characteristics and the long turnover time, SIC stock is traditionally considered to be dominated by abiotic factors including soil moisture, soil pH, CO₂ partial pressure, and Ca²⁺ concentrations (Yang et al., 2012; Rey, 2015; Yang and Yang, 2020). However, our results revealed that the influence of microbial factors on the variation in SIC gradually increased with soil depth, and B/F was the strongest predictor for the change in calcite content among microbial factors in the 20–30 cm soil layer (Figure 6B). The significant positive correlation between calcite content and B/F suggested that the higher the bacterial biomass in soil was beneficial to the formation of calcite minerals compared with the fungal biomass (Figures 3, 4A, B). Furthermore, the significant negative correlation between bacterial biomass and carbonate carbon isotopes not only proved that carbonate minerals were mainly secondary in the 20–30 cm soil layer of the study area, but also proved the influence of bacteria on the carbonate mineral formation (Figure 5B).

Combined with the significant negative correlation between calcite content and MUFA/SFA (Figures 4C, D), we suggested that bacteria-mediated calcite formation is related to resource competition against fungi. In arid and semi-arid regions, where SOC is very low, competition for resources affects the fungi and bacteria biomass (Figure 3; Bahram et al., 2018). Also, high free Ca²⁺ concentrations in alkaline soil contribute to SOC stabilization through the form of cation bridging (Rowley et al., 2018). This is consistent with our result that SOC has a significant positive correlation with the Ca²⁺ concentration (Figure 7A). This form of cation bridging by Ca²⁺ (Ca-OC) has been highlighted as an important component of SOC stabilization by many authors and is well-documented (Oades, 1988; Clough and Skjemstad, 2000; Edwards and Bremner, 2010). To survive, bacteria have to decompose Ca-OC because of the broader activities of fungi than bacteria in using complex carbon substrates (Bahram et al., 2018). Thus, bacteria would mediate the calcite mineral formation to obtain OC bound to Ca²⁺ in two types: biologically controlled and induced mineralization (Dong et al., 2022). Biologically controlled mineralization is the active absorption of calcium ions by microbial cells and the formation of carbonate precipitation by regulating the saturation conditions of cells (Bazylinski and Frankel, 2004; Benzerara et al., 2011). Biological induced mineralization is a passive process in which microorganisms alter the pH through metabolic activities and bind large amounts of calcium ions to their cell surface and extracellular polymers, making their microenvironment conducive to carbonate mineral formation (Muynck et al., 2010; Phillips et al., 2013).

It was found that as the proportion of available organic carbon decreases, bacteria can take advantage of the more stable organic matter that binds to calcium ions through mediating CaCO₃ formation in alkaline soil environments. However, this process can be heavily influenced by external factors, including edaphic factors and soil texture, which results in the discovery only in the 20–30 cm soil layer of this study area. Moreover, SOC can also be stabilized through either inner- or outer-sphere interaction with minerals (Sposito, 1989; Sutton et al., 2010). Thus, plagioclase can provide a site for microbial colonization, leading to a significant positive correlation between plagioclase content and microbial biomass in soil with little clay mineral content (Figure 3). At the same time, the negative correlation between plagioclase content and carbonate carbon isotope in the 20–30 cm soil layer indicated that plagioclase provided the Ca²⁺ for bacteria-mediated calcite formation (Figure 5A).

Although much attention has been paid to the effect of microorganisms on soil carbonate content in recent years, the effect of soil texture on SIC pools has not been studied. However, our results showed that the silt content in the soil had a significant relationship with calcite content and played a dominant role in regulating the calcite content and SIC stock in the 50–60 cm soil layer (Figures 3, 7C). Soil texture is fundamentally affected by soil mineral composition and content: the higher the quartz content, the lower the feldspar content, and the more difficult it is to be weathered (Franke, 2009). The quartz content in the soil of the study area was much higher than those of other minerals, resulting in the soil texture being sandy soil and significantly affecting the silt content (Figure 3 and Supplementary Table 2). The silt content was positively correlated with the water-soluble ion content and EC, especially the Ca²⁺ content (Figure 7B and Supplementary Table 2). Thus, the silt content of soil can not only directly affect the calcite content by controlling soil the Ca²⁺ content of soil but also indirectly affect the calcite content through EC affecting microbial activity (Yuan et al., 2007; Prindle et al., 2015; Yan et al., 2015).

Taken together, our results demonstrate that in arid and semi-arid areas, bacteria acquire Ca-OC in the soil environment by mediating calcite mineral formation for survival, leading to the conversion of organic carbon to inorganic carbon. Moreover, the effect of soil texture on soil biogeochemical processes was further amplified with increasing in soil depth.