Global microbial necromass contribution to soil organic matter


 Soil organic matter (SOM) plays an important role in mitigating climate change and sustaining soil health and food production 1,2. Mounting evidence suggests that microbial necromass is the main contributor to SOM 3; however, we lack quantification of microbial necromass at a global scale, especially in subsoils. Here, we generate, for the first time, global distribution maps of microbial necromass carbon (C) and nitrogen (N) and contributions to SOM in topsoil and subsoil. Globally, necromass concentrations varied widely across ecosystems and by latitude, contributing 19-60% to SOC and 41-92% to soil N stocks, with particularly large accumulations in boreal and tropical ecosystems. On average, fungal necromass contributions to SOM are 3x greater than bacterial, although this varied across ecosystems. Microbial necromass contributions to SOC are strongly associated with soil C:N ratios and pH; necromass contributions are greater in soils with narrow C:N ratios and higher pH. Microbial necromass is on average 23 and 77 times greater than living microbial biomass in topsoil and subsoil, respectively. These data highlight the importance of necromass contributions to SOM, especially soil N, and the need for spatially resolved necromass data sets that can be used in biogeochemical models to estimate SOM dynamics more accurately.

As the keystone of soil health and the dominant nutrient pool in terrestrial ecosystems, SOM contains 25 more C than the atmosphere and terrestrial vegetation combined 4 . Therefore, a better understanding of the 26 mechanisms underlying the formation and stabilization of SOM is critical for many climate change mitigation 27 strategies and for providing the foundations for food security 1,2,5 . Traditional views suggest that selective 28 preservation and molecular recalcitrance determine the long-term persistence of SOM 4 , while emergent views 29 suggest that SOM persistence is largely due to complex interactions between SOM, microorganisms and soil 30 minerals 6 . Chemically, it appears that much of the C stabilized onto mineral fractions -40-80% of SOC 3 -31 originates from microbial necromass, with a longer lifespan than plant derived SOM 7 . Indeed, incorporation 32 of microbial necromass into models significantly improves model performance 8 ; however, the model 33 simulated contributions of microbial necromass to SOC (10-27%) that were much lower than previously 34 measured 3 . Lack of explicit quantification of microbial necromass contributions to SOM at the global scale 3 , 35 especially in subsoils 9 , greatly hampers our ability to constrain and incorporate soil microbial processes into 36 biogeochemical models, which is needed to precisely predict soil responses to management, agricultural 37 intensification and potential feedbacks to climate change 8 . 38 Amino sugars are major components of microbial cell walls and have widely been used as biomarkers 39 to track microbial residues 3 . Muramic acid exclusively originates from bacteria, whereas glucosamine occurs 40 in either fungal or bacteria cell walls 10 . Fungal and bacterial necromass C and N can be estimated by 41 measuring amino sugar biomarkers in the soil and then multiplying by conversion factors based on previously 42 determined molecular stoichiometry of the biomarkers 3 . Using this approach, and a relatively limited data set 43 (n=122), Liang et al. (2019) estimated that microbial necromass comprises ~30-60% to total SOC in temperate 44 topsoil 3 . 45 Using a comprehensive global dataset (n=902) of amino sugar concentrations in both topsoil (0-30 cm) 46 and subsoil (30-100 cm) from peer-reviewed literature and our own data (Extended Data Fig. 1, Extended 47 Data Fig. 2 and Supplementary Data), we demonstrate for the first time a strong and universal correlation 48 between total soil N (TN) and amino sugars (Extended Data Fig. 3). This is important because measuring TN 49 is relatively inexpensive and standardized, while measuring amino sugars is relatively rare, expensive, and 50 time consuming. Therefore, researchers and modelers can now use this robust relationship to estimate 51 microbial necromass contributions to SOM across different ecosystems at a large scale without having to 52 measure amino sugars in every soil. While the relationship between TN and amino sugars has been reported 53 11 , the limited data sets in previous studies were not large enough to establish the strong universal and by 54 ecosystem type relationships we report here. We demonstrate the power of using this relationship by 55 extrapolating soil amino sugar (Extended Data Fig. 4) and then microbial necromass distributions at a global 56 scale using existing global soil TN data at a spatial resolution of 0.083° (~10 km). Global amino sugar 57 concentrations were converted to global microbial necromass C and N using previously estimated 58 stoichiometric conversion factors 11, 3 .   SOC (19%,Fig. 1b,e), also reflected in its low ratio of necromass C to living microbial biomass C (MBC, 69 Table 1) compared to other biomes. This is likely a result of slow growth and turnover of microbial biomass 70 at low temperatures 12 . Apart from tundra, boreal and tropical forests also had relatively lower microbial 71 necromass contributions to SOC compared with other ecosystems (Table 1). Relatively higher lignin versus 72 protein in forest ecosystem organic matter inputs 13 or a low rhizosphere-to-bulk soil ratio in forests compared 73 with other ecosystems could decrease SOC accumulation due to relatively low C transformation rates and 74 efficiency 14 . 75 We present the first ever estimates of microbial necromass N as well as its global distribution. Similar 76 to microbial necromass C, we found a high accumulation of microbial necromass N in boreal and tropical 77 ecosystems in the whole soil profile (Fig. 1g, j, Extended Data Fig. 5d, e). Globally, microbial necromass N was estimated to contribute 49 and 63 Pg N to topsoil and subsoil N stocks, respectively ( contribute a greater proportion of C than bacteria due to wider biomass C:N ratios and generally higher 102 substrate use efficiencies, leading to greater fungal biomass and then necromass C produced relative to C lost 103 as CO2 19 . Also, formation of hyphal condensed tannins and melanin complexes 20 , and higher sorption rates 104 of melanized hyphae to soil minerals 21 , significantly slow fungal necromass decomposition and increase 105 incorporation into stable SOM. 106 The ratio of fungal to bacterial necromass contributions to SOM tended to increase with increasing 107 latitude ( Fig. 1). On average, the fungal to bacterial necromass ratio was lowest in tropical and subtropical 108 forests (1.8) and was highest in temperate forests (4.7, Table 1, Extended Data Table 1)

117
At the global scale, random forest analysis showed that microbial necromass C contributions to SOC 118 were strongly associated with soil C:N ratio and soil pH ( Fig. 2a-d). The microbial necromass C contributions 119 to SOC increase as soil C:N ratios narrow in both topsoil and subsoil (Fig. 2e, g). While the relationship is 120 correlative, a causative relationship makes sense because a narrowing of soil C:N ratios would indicate 121 alleviation of microbial N limitation, which would lead to increased microbial biomass production and 122 substrate or C use efficiency (CUE) 25 , and ultimately, increased microbial necromass accumulation 14 and 123 contributions to SOC. 124 Soil pH was also a strong predictor of microbial necromass contributions to SOC, as pH increased 125 microbial necromass contributions to SOC increased (Fig. 2f, h). Increases in soil pH have been shown to 126 increase overall microbial CUE 26 . This effect is apparently more important than a concomitant shift in fungi 127 and bacteria relative abundance. Fungi are favored by acidic conditions, and are generally considered to have 128 a greater CUE than bacteria 27 . Therefore, we expected to see a decrease in microbial necromass C 129 contributions to SOC accompanied by decreasing fungal biomass with increasing pH. While a bit surprising, 130 this result supports the idea that pH is a driver of microbial community structure and CUE beyond just controlling fungal to bacterial ratios 28 . 132 133 Links between necromass and living biomass 134 Quantitative connections between living microbial biomass and microbial necromass are a rarity 15 and 135 because living microbial biomass is estimated to represent less than 4% of SOM it is often ignored in terms 136 of SOM accrual 3 . Using a recently generated global living microbial biomass dataset 29 , we estimated the 137 ratios of microbial necromass to living biomass C and N at a global scale. The living microbial biomass does 138 indeed comprise only a very small proportion of the total microbial residue (Fig. 3). The ratio of microbial 139 necromass to living biomass is quite stable across biomes in topsoil (Table 1); however, in subsoil this ratio 140 was almost two times higher in boreal forest, montane grassland, and wetland compared to other ecosystems 141 (Extended Data Table 1). In topsoil, microbial necromass, on average, is 23 times greater than living biomass, 142 while in subsoils microbial necromass was more than 70 times greater (Table 1, Extended Data Table 1). This There are very few studies reporting amino sugars in important SOM storage ecosystems such as tundra, 151 wetlands, permafrost. Our data suggest cold temperatures may limit microbial necromass C accrual in SOM, 152 which may mean that boreal and tundra ecosystems could potentially accrue more microbial necromass C 153 under climate warming, with resulting increases in microbial growth and biomass turnover, and accelerated 154 formation microbial necromass-mineral associations 30 or warming may decrease microbially derived SOM 155 31 . More research is needed to examine how these vulnerable ecosystems will respond to climate change. Our 156 results show the importance of microbial necromass in subsoil and a general lack of data, highlighting the 157 need for more studies that include subsoil or are focused on subsoil with regard to microbial necromass and 158 SOM. 159 With the development of isotopic and spectroscopy techniques, there is a wide recognition that microbial 160 residues are a critical component of stabilized SOM, however, microbial necromass has not been adequately 161 represented in most microbial explicit or general SOM models. Only recently researchers attempted to 162 incorporate microbially derived C into models to show they did indeed perform better than existing models 8 . 163 However, due to limited observation data and no quantitative constraints at global scale, their estimation of 164 microbial necromass C within total SOC were much lower than previous 3 and our current estimates. The data 165 set presented here can serve as a benchmark for biogeochemical and earth systems models to parameterize 166 and more accurately estimate and constrain microbial necromass contributions to SOM, which is a 167 fundamental step for the improvement of nutrient management in agroecosystems and for mitigating soil 168 degradation and climate change.

Fig. 2 Climate factors and soil characteristics that control microbial necromass contributions to SOC in topsoil 250
and subsoil at a global scale. a-d, Random forest analyses of the relative importance (IncNodePurity) of different 251 variables in relation to microbial necromass C contributions to SOC with and without inclusion of soil C:N ratios in 252 topsoil (a-b) and subsoil (c-d), respectively. Red asterisks indicate significant influence at P < 0.05. e-h, Relationships 253 between soil C:N ratios or soil pH and microbial necromass C contributions to SOC in topsoil (e-f) and subsoil (g-h). 254 Relationships are based on data from 10,000 randomly sampled pixels across the globe and all data were log-transformed. 255 MAT: mean annual temperature, MAP: mean annual precipitation, pH: soil pH measured by potassium chloride, BS: 256 base saturation, CEC: cation exchange capacity, MBC: microbial biomass carbon, MBN: microbial biomass nitrogen. 257 28 °S to 70 °N latitude, range in annual mean precipitation from 19 to 2942 mm, and in annual mean 292 temperature from -11 to 32°C (Extended Data Fig. 1).

294
Missing amino sugar data filling 295 There were around 120 measurements reporting total amino sugar while not reporting glucosamine or 296 muramic acid (Supplement Data1). We found a strong relationship between different amino sugar compounds 297 (glucosamines, galactosamine, muramic acid, mannosamine) and total amino sugar (with R 2 range from 0.83 298 to 0.98, Extended Data Fig. 2a-d). In addition, our results showed that glucosamine (GluN) and galactosamine 299 (GlaN) are the major components of total amino sugars, while muramic acid (MurA) and mannosamine (ManN) 300 only comprise a small proportion of the total amino sugar (Extended Data Fig. 2e), consistent with previous 301 work 10 . In order increase the available data, we first filled in the data sets missing glucosamines and muramic 302 acid cases based on the strong relationships with total amino sugars (Extended Data Fig. 2).

305
Based on the complied dataset, we first conducted a correlation analysis between amino sugar with 306 climate (MAT and MAP), and edaphic properties (SOC,TN,C:N ratio,pH,sand,silt,clay,MBC,and MBN) 307 using corrplot package in R (http://cran.r-project.org/). Then we conducted a random forest analysis to identify 308 the main predictors for the amino sugar. The importance of each predictor variable is determined by evaluating 309 the decrease in prediction accuracy (that is, increase in the mean square error between observations and 310 predictions) when the data for that predictor is randomly permuted 32 . This accuracy importance measure was 311 computed for each tree and averaged over the forest (5,000 trees). Random forest analyses were conducted 312 using the randomForest package in R. In addition, the significance of the importance of each predictor on 313 amino sugar was assessed by using the rfPermute package in R. Random forest analysis results explained 74.3% 314 to 87.6% of the variance for different amino sugars. Besides, random forest analysis suggested that TN and 315 SOC were the two most significant predictors for different amino sugars (Extended Data Fig. 3b). Then we 316 built a linear mixing model using nlme package in R to compare different model performance that include TN, 317 to predict the variation of amino sugar, models that adding more variables did not significantly improve 319 prediction of the amino sugar variation (Extended Data Fig. 3c). Therefore, in the following analysis, we only 320 used TN as the predictor for amino sugar to simplify the prediction model. R 2 of the prediction model with 321 TN in different ecosystems varied from 0.74 to 0.94 for total amino sugar (TAS), 0.74 to 0.94 for GluN, and 322 0.74 to 0.95 for MurN, respectively (Extended Data Fig. 3d). Overall, TN explained 91% of the variation for 323 TAS, 90% variation for GluN, and 79% variation for MurA, respectively (Extended Data Fig. 3e). is determined by evaluating the decrease in prediction accuracy (that is, the mean node impurity between 384 observations and predictions) when the data for that predictor is randomly permuted for 500 times. Random 385 forest analyses were conducted using the randomForest package in R, and the significance of the importance of each predictor on amino sugar was assessed by using the rfPermute package in R.  Fig. 1 Soil amino sugar data distribution across biomes and climates. We collected a total of 902 434 soil amino sugar measurements made in either topsoil (0-30 cm) or subsoil (30-100 cm) spanning all major land cover 435 types (a) and climate zones (b). These data were used to build the relationship(s) with soil TN, that could then be used 436 to extrapolate global amino sugar stocks. 437 438 Extended Data Fig. 2 Relationship between different amino sugar components and total amino sugar. a-d, 440 Strong relationships between total amino sugar (TAS) and different amino sugar components, glucosamine (GluN, a), 441 muramic acid (MurA, b), galactosamine (GlaN, c) and mannosamine (ManN, d) were used to fill missing amino sugar 442 data where only total amino sugar data were reported (n = 782