Ensemble learning algorithms to elucidate the core microbiome's impact on carbon content and degradation properties at the soil aggregate level.

Journal: The Science of the total environment
Published Date:

Abstract

Soil aggregates are crucial for soil organic carbon (OC) accumulation. This study, utilizing a 32-year fertilization experiment, investigates whether the core microbiome can elucidate variations in carbon content and decomposition across different aggregate sizes more effectively than broader bacterial and fungal community analyses. Employing ensemble learning algorithms that integrate machine learning with network inference, we found that the core microbiome accounts for an average increase of 26 % and 20 % in the explained variance of PCoA and Adonis analyses, respectively, in response to fertilization. Compared to the control, inorganic and organic fertilizers decreased the decomposition index (DDI) by 31 % and 38 %, respectively. The fungal core microbiome predominantly influenced OC content and DDI in larger macroaggregates (>2000 μm), explaining over 35 % of the variance, while the bacterial core microbiome had a lesser impact, explaining <30 %. Conversely, in smaller aggregates (<2000 μm), the bacterial core microbiome significantly influenced DDI (R > 0.2), and the fungal core microbiome more strongly affected OC content (R > 0.3). Mantel tests showed that pH is the most significant environmental factor affecting core microbiome composition across all aggregate sizes (Mantel's r > 0.8, P < 0.01). Linear correlation analysis further confirmed that the core microbiome's community structure could accurately predict OC content and DDI in aggregates (R > 0.8, P < 0.05). Overall, our findings suggested that the core microbiome provides deeper insights into the variability of aggregate organic carbon content and decomposition, with the bacterial core microbiome playing a particularly pivotal role within the soil aggregates.

Authors

  • Fengwu Zhou
    Jiangsu Provincial Key Laboratory of Materials Cycling and Pollution Control, School of Geography, Nanjing Normal University, Nanjing 210023, China.
  • Yunbin Jiang
    Jiangsu Provincial Key Laboratory of Materials Cycling and Pollution Control, School of Geography, Nanjing Normal University, Nanjing 210023, China; Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application, Nanjing 210023, China.
  • Cheng Han
    State Key Laboratory of Polymer Materials Engineering, Polymer Research Institute, Sichuan University, Chengdu, 610065, China.
  • Huan Deng
    School of Medical Informatics and Engineering, Southwest Medical University, Luzhou, China.
  • Zongren Dai
    Department of Environmental Science and Engineering, Fudan University, Shanghai 200433, China.
  • Zimeng Wang
    Department of Environmental Science and Engineering, Fudan University, Shanghai 200433, China. Electronic address: zimengw@fudan.edu.cn.
  • Wenhui Zhong
    Jiangsu Provincial Key Laboratory of Materials Cycling and Pollution Control, School of Geography, Nanjing Normal University, Nanjing 210023, China; Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application, Nanjing 210023, China. Electronic address: zhongwenhui@njnu.edu.cn.