More Efficient Nitrogen Recycling or Less Efficient Carbon Use to Decompose Nitrogen-Poor Residues?

2.1 Theory

2.1.1 Model rationale

We used macro-chemical (total C and total N) data from litter decomposition studies in combination with several variants of a minimal mathematical model to identify potential N limitation responses in litter decomposer communities. The model tracks C and N in litter substrates and microbial biomass (Figure 1) and includes parameters describing N versus C acquisition preference, CUE, and N recycling efficiency. Instead of using the model to determine the temporal trajectory of litter C and N during decomposition, we used it to find analytical expressions for the N import/release curve; that is, the relation between the fraction of initial N and the fraction of initial C that remain in the litterbags during decomposition. As shown mathematically in the following, the decomposition rates (and thus the effects of environmental conditions on these rates) do not affect the relations between litter N and C (Eq. (10)). Therefore, this approach allowed us to focus on microbial traits (at the community level) that affect litter stoichiometry, rather than on environmental and litter chemical conditions that control the speed of decomposition. Specifically, we estimated the degree of preferential N acquisition (denoted by α), microbial CUE (denoted by e), and efficiency of N recycling (denoted by η) within the community across litter types and climatic conditions. While the mathematical approach per se is not new (Ågren et al., 2013; Bosatta & Ågren, 1985; Manzoni, 2017; Manzoni et al., 2010), it was previously restricted to CUE estimation, neglecting the possible role of other traits (Table 1).

Figure 1

Table 1

Symbol definitions and units (see also Figure 1). State variables are expressed in terms of C or N mass per litterbag.

SYMBOL	EXPLANATION	UNITS
State variables and independent variables
cL	Fraction of initial litter C, cL = CL/CL,0	–
CB	Microbial biomass C	g C
CL	Total litter C, CL = CS + CB	g C
CS	Litter substrate C	g C
nL	Fraction of initial litter N, nL = NL/NL,0 (nL > 1 when N is imported)	–
NB	Microbial biomass N	g N
NL	Total litter N, NL = NS + NB	g N
NS	Litter substrate N	g N
t	Time	Y
Rates and other calculated quantities
D	Litter decomposition rate	g C y–1
G	Microbial growth rate, G = eD	g C y–1
M	Microbial mortality rate	g C y–1
φN	Net N import/release rate (net N import when φN < 0)	g N y–1
ē	Mean C use efficiency throughout the decomposition process	–
Initial conditions
CL,0	Initial total litter C	g C
C : NL,0	Initial litter C:N ratio, C : NL,0 = 1/rL,0	g C g N–1
L0	Initial lignin fraction	g lignin g litter–1
NL,0	Initial total litter N	g N
rL,0	Initial litter N:C ratio, rL,0 = 1/C : NL,0	g N g C–1
Parameters
a	Exponent in Eqs. (15)–(17)	–
b	Half saturation constant in Eqs. (15)–(17)	g N g C–1
b0	Half saturation constant at zero lignin content in Eqs. (15)–(17)	g N g C–1
b1	Sensitivity to lignin content of the half saturation constant in Eqs. (15)–(17)	g N g litter g C–1 g lignin–1
e	Microbial C-use efficiency (0 < e ≤ emax)	–
e0	Microbial C-use efficiency when rL,0 = 0 (0 < e0 ≤ emax)	–
emax	Maximum microbial C-use efficiency (emax = 0.6)	–
rB	N:C ratio of microbial biomass	g N g C–1
α	Preferential N acquisition coefficient (αmin ≤ α ≤ αmax)	–
α0	Preferential N acquisition coefficient when rL,0 = 0 (αmin ≤ α0 ≤ αmax)	–
αmin, αmax	Lower and upper bounds for the fitting of the preferential N acquisition coefficient	–
η	N recycling efficiency (0 ≤ η < 1)	–
η0	N recycling efficiency when rL,0 = 0 (0 ≤ η0 < 1 )	–

2.1.4 N import/release curves with microbial traits varying during decomposition

We also considered three more general model variants, in which α, e, and η are functions of the current litter C:N ratio. The rationale behind these variants is that N acquisition (model Adyn), CUE (model Cdyn), or N recycling (model Ndyn) could be regulated during litter decomposition so that N acquisition is prioritized, CUE is lower, or more N is resorbed at early stages of decomposition when the litter C:N is still high (Table 2). As in the model variants with time-invariant traits, traits are allowed to vary one at a time. In model Adyn, N acquisition is assumed to be higher (relative to the pool) than C acquisition at high litter C:N. This effect is captured by imposing that α decreases from a maximum hypothetical value characterizing a litter completely devoid of N (α₀) to α = 1 when the litter C:N reaches the microbial biomass C:N (solid green curve in Figure 2A). In model Cdyn, CUE is assumed to increase linearly with decreasing litter C:N, from a minimum hypothetical value for a litter without N (e₀) to a maximum value (e_max) when litter C:N reaches the microbial biomass C:N (solid orange curve in Figure 2B). In model Nflex, η is assumed to decrease linearly with decreasing litter C:N ratio, from a maximum hypothetical value for a litter without N (η₀) to zero when litter C:N reaches the microbial biomass C:N (Table 2, blue solid curve in Figure 2C).

Figure 2

In models Adyn and Cdyn, the relations between α or e and n_L/c_L introduce nonlinearities in Eq. (11) that prevent compact analytical solutions, so we solved Eq. (11) numerically. In contrast, model Ndyn can be formulated exactly as model Cflex (Eq. (13)), where e in Cflex is related to η₀ in Ndyn according to,

14

$e = \frac{\left(1–{\eta }_0\right)e_{max}}{1–{\eta }_0{e}_{max}}, \mathrm{\text{or}} {\eta }_0 = \frac{e_{max}–e}{e_{max}\left(1–e\right)}.$

Given the mathematical equivalence of models Cflex and Ndyn, η₀ can also be determined analytically.

2.2 Litter decomposition data

We started from the database of litter decomposition datasets used by Manzoni et al. (2017) and added datasets encompassing N-poor needles and wood (B. J. Wijas et al., 2024; Khanina et al., 2023; Smyth et al., 2016; Spohn & Berg, 2023) to better cover the higher end of the litter C:N range. Decomposition data from aquatic systems and wetlands were not included. Each dataset includes multiple measurements of remaining C and N in litterbags or wood blocks, as well as initial litter C:N ratios. The model does not account for leaching of soluble compounds that often occurs early during decomposition. Therefore, we followed the method described by Manzoni et al. (2010) to identify and remove the initial leaching phase. In practice, in datasets where leaching was present, we considered only data points after the initial leaching phase, resetting the initial litter C:N ratio as the C:N of the first point after the leaching phase. Moreover, we excluded datasets with fewer than six data points from the analysis (to ensure robust parameter estimation) as well as those datasets in which final mass loss was lower than 60% (to ensure the estimated parameters are representative of all or most of the decomposition process). The former requirement was relaxed for wood decomposition datasets, which often do not include more than five datapoints. Litterbags from plots fertilized during the field incubation were also disregarded. After screening, we retained ~500 N import/release curves from ~40 published articles (Ball, Bradford, & Hunter, 2009; Berg & McClaugherty, 1989; Blair et al., 1998; Busse, 1994; Chen et al., 2015; Chuyong, Newbery, & Songwe, 2002; Edmonds, 1987; Foster & Lang, 1982; Hirobe et al., 2004; Hobbie, 2008; Hobbie et al., 2012; Isaac & Nair, 2005; Jacob et al., 2009; Kamei, Barik, & Pandey, 2009; Khanina et al., 2023; Krankina, Harmon, & Griazkin, 1999; Lambert, Lang, & Reiners, 1980; Li, Han, & Zhang, 2007; Liu, Fox, & Xu, 2000; Long-term Intersite Decomposition Experiment Team, 1995; Means, Macmillan, & Cromack, 1992; Melillo et al., 1989; Moore et al., 2006; Osono & Takeda, 2004; Osono & Takeda 2005; Palm & Sanchez, 1990; Palviainen & Finer, 2015; Parsons & Congdon, 2008; Ricker et al., 2016; Rustad, 1994; Schroth, Zech, & Heimann, 1992; Smyth et al., 2016; Sollins et al., 1987; Spohn & Berg, 2023; Thompson & Vitousek, 1997; Tripathi & Singh, 1992; Trofymow & CIDET, 1998; Tu et al., 2014; B. J. Wijas et al., 2024; Xu, 2006; Xu & Hirata, 2005; Yavitt & Fahey, 1986; Zhu & Ehrenfeld, 1996) or online data repositories (Harmon, 2013; Hobbie, 2013; Hobbie 2015; B. Wijas et al., 2024).

For each litter decomposition dataset, we used the original sources to collect—when available—initial lignin contents and climatic data (mean annual precipitation, MAP, and mean annual temperature, MAT), which were used as covariates to predict the microbial traits (Tables S1 and S2). Wood lignin contents for the species in Khanina et al. (2023) were obtained from Kahl et al. (2017) and for the species in Wijas et al. (2024) from Lee et al. (2022).

2.3 Estimation of microbial traits

To estimate the microbial traits, we considered the three model variants with time-invariant microbial traits (Aflex, Cflex, and Nflex) and the three variants with traits that change through time during decomposition (Adyn, Cdyn, and Ndyn; Table 2). Parameters encoding preferential N acquisition (α or α₀), CUE (e or e₀), and N recycling efficiency (η or η₀) were estimated by fitting either the analytical N import/release curves [Eqs. (12) or (13)] or the numerical solution of Eq. (11) to litter remaining N and C data. All calculations were done in Matlab (MathWorks, 2024). When a numerical solution was necessary, we implemented the numerical integration of Eq. (11) (ode45 function) into a least square optimization algorithm (lsqcurvefit function). To constrain the parameter optimization algorithm, minimum and maximum values were set for α or α₀ (1 ≤ α or α₀ ≤ 10), e (0 ≤ e ≤ e_max = 0.6), e₀ (0 ≤ e₀ ≤ e_max = 0.6), and η or η₀ (0 ≤ η or η₀ ≤ 1). The maximum value of N recycling efficiency was set to one, but we expected that realistic values would be lower, as at least some N is likely to be left behind in necromass.

Model performance was evaluated by calculating the root mean square errors (RMSE) for each model variant. All model variants had the same degree of freedom (one fitting parameter), allowing a comparison only based on RMSE. To summarize and compare model performances, we finally calculated the percentage of datasets in which each model variant achieved the lowest RMSE, either among the models with time invariant traits (Aflex, Cflex, and Nflex) or among the models with dynamic traits (Adyn, Cdyn, and Ndyn). The estimated microbial traits and RMSE are reported in Table S3.

2.4 Analysis of C-use and N recycling efficiencies

We tested the combined effects of initial litter chemistry and climate on the microbial traits using linear statistical models (fitlm function) with initial litter C:N ratio (log transformed), initial lignin content, MAP, and MAT as predictors, including C:N-lignin and MAP–MAT interactions. All independent and dependent variables were normalized before the statistical analysis (subtracting the mean and normalizing by the standard deviation). Since climate in general did not affect the estimated microbial traits, but initial C:N and lignin content did, we focused on these chemical properties in more detailed analyses.

We assessed the relations between litter initial C:N ratios and the estimated e and η using sigmoidal functions,

where a (exponent) and b (half saturation constant) are fitting parameters representing the steepness of the sigmoidal curve at the inflection point, and the location of the inflection point, respectively; α_min and α_max are the lower and upper bounds of α . The choice of these functions is motivated by their flexibility despite a low number of fitting parameters. We also considered a variant to Eqs. (15)–(17) in which we could test if initial lignin content modulates the effect of initial litter C:N on the microbial traits. This analysis was restricted to the decomposition datasets in which initial lignin content was available (87% of the total number of datasets). Lignin effects were included in the parameter b, as b = b₀ + b₁L₀, where b₀ and b₁ are fitting parameters and L₀ is the initial lignin content expressed as a fraction of litter dry weight.

Nonlinear least square fitting was performed with the function lsqcurvefit. Positive values of a indicate that the sigmoidal curve is increasing with r_L,0, whereas negative values indicate a decreasing curve. Trends with increasing r_L,0 were considered significant if the 95% confidence interval of a did not bracket zero.

Once the microbial traits are estimated, it is possible to use the mass balance equations [Eqs. (7) and (8)] to calculate the mean CUE throughout the decomposition process—interpreted here as a measure of growth performance at the microbial community level. The mean CUE is defined as the total C converted into new biomass divided by the total C taken up,

It is convenient to calculate the integrals in Eq. (18) along the variable C_L (total remaining C) instead of time, recalling that at the beginning of decomposition, t = 0 and C_L = C_L,0 (initial litter C), while at the end of decomposition, t = ∞ and C_L = 0. Changing the variable of integration, we obtain,

We note that $\frac{d{C}_L}{dt}\hspace{0.17em}\hspace{0.17em}= \hspace{0.17em}–(1–e)D$ (Eq. (7)), and that we can change the variable of integration further from the total litter C to the fraction of initial litter C (c). As a result, the calculation of ē reduces to,

In models Aflex, Cflex, Nflex, Adyn, and Nflex, e is time-invariant, so ē = e in Cflex and ē = e_max in the other variants. In contrast, in model Cdyn, e varies during decomposition, so the integral cannot be further simplified and needs to be calculated numerically (trapz function). The relations between ē and initial litter C:N were assessed using Eq. (16).

3.1 Effect of preferential N acquisition, C-use, and N recycling efficiencies on the N import/release curves

N import/release curves illustrate how, the remaining litter N as a fraction of the initial litter N (n_L), changes as a function of the remaining litter C as a fraction of the initial litter C (c_L). These curves start from the initial litter C and N, at the point c_L = 1 and n_L = 1, and end when all litter has been decomposed, at the point c_L = 0 and n_L = 0. Their shape depends on both the initial litter C:N ratio and microbial traits—here α, CUE, and η. High values of initial litter C:N cause N to accumulate (n_L > 1) within decomposing litter, because microorganisms retain N in their biomass and import N from the environment to fulfill their stoichiometric requirements (Figure 3A). Net N release from these N-poor litter types starts relatively late during the decomposition process. In contrast, with N-rich litter (low initial C:N), N release can start from the beginning and no N accumulation occurs (n_L < 1). Preferential acquisition of N from the litter (i.e., α > 1) increases internal N availability and thus reduces external N requirements, ultimately lowering the peak in litter N (compare green dashed and solid gray curves in Figure 3A). Lower values of CUE decrease the microbial N demand linked to growth, and thus also lower N import (compare orange dashed and solid gray curves in Figure 3A). A similar effect is obtained by increasing microbial N recycling—stronger N recycling reduces N demand from external sources, thus lowering the N import/release curves (compare blue dashed and solid gray curves in Figure 3A). Despite a similar overall effect on the litter N peak, the three N limitation responses result in different shapes of the N import/release curves (compare dashed curves with the same thickness in Figure 3A). Most notable is the occurrence of a more pronounced, but earlier N peak when N is preferentially acquired (green curves).

Figure 3

Consistent with the differences shown in Figure 3A, an increase in preferential acquisition of N or N recycling causes a decrease in the predicted fraction of the initial litter N (green and blue curves are negative in Figure 3B), whereas an increase in CUE promotes N import (orange curve in positive in Figure 3B). A 10% relative change in α, e, and η causes a 5%–15% change in n_L, indicating that these three parameters have similarly strong effects on the predicted fraction of the initial litter N.

3.2 Model fitting and performance

Figure 4A shows an example of model fitting to data on remaining litter N and C. In this particular example, net N import occurred at the beginning of decomposition (when the fraction of remaining C was close to one), as apparent from values of n_L larger than one. This N import phase is often observed in litter decomposition studies (Moore et al., 2006; Parton et al., 2007; Spohn & Berg, 2023). N import continued until about 40% of the litter C had been lost. This specific example highlights contrasting modeled N import/release curves depending on the chosen model variant. When fitting the preferential N acquisition coefficient (Aflex and Adyn), the N peak occurred too early and N release in the later phase of decomposition was overestimated (green curves in Figure 4A). In models where only e was estimated and η = 0 (Cflex and Cdyn), e remained low (~0.2, see orange curves in Figure 4C). This means that low CUE allows microorganisms to reduce their N demand, resulting in relatively low N import. The same effect was obtained through high η and high e in models where η was estimated from the data (Nflex and Ndyn; blue curves in Figure 4A). Thus, high η and low e have complementary roles in lowering the microbial demand for external N, while flexible N acquisition alone does not explain the observed N import/release patterns well.

Figure 4

In models Adyn, Cdyn, and Ndyn we assumed that microbial traits could vary during decomposition. While we imposed linear relations between traits and litter N:C ratio, the (hypothetical) values for litter without N were estimated from the data, resulting in different relations between α, e, or η and the fraction of remaining litter C. In model Cdyn, e increased during decomposition (solid orange curve in Figure 4C), whereas in model Ndyn η decreased (solid blue curve in Figure 4D). In contrast, preferential N acquisition remained high throughout decomposition (green solid curve in Figure 4B), leading to a persistent (and unrealistic) high litter C:N (shallow slope of the green curves in Figure 4A indicates low N:C in the late decomposition phase).

Fitting of the N import/release curves to all the litter N and C data led to a range of model performances. Overall, prediction errors were larger at the beginning of decomposition, especially when N accumulated in the litter, and much lower when the fraction of remaining N approached zero (Figure S1). This error structure indicates heteroscedasticity, but it should be noted that the model is constrained at c_L = 1, when n_L(1) = 1, so it cannot capture rapid variations in remaining N induced by relatively large measurement errors near c_L ≈ 1. Generally, the RMSE were low (indicating a good fit) and around 0.1 for models where e or η were fitted, whereas they were notably higher (indicating a bad fit) when fitting α (Figure 5A). When comparing models with time-invariant parameters, model Cflex had the best performance in ~60% of the datasets, whereas among models with dynamic parameters Ndyn performed best in ~50% of the datasets (Figure 5B). These results demonstrate that patterns in the data are best captured by flexible CUE across litter types (model Cflex), or flexible N recycling efficiency across litter types when allowed to decrease as decomposition progresses (model Ndyn). It is important to recall that these two best-performing models are mathematically equivalent, suggesting that it is a single model structure (with contrasting interpretation of the meaning of its parameters) that captures most of the variation in the litter decomposition data.

Figure 5

3.3 Relations between litter C:N ratio and preferential N acquisition, C-use, and N recycling efficiencies

Climate was not a good a predictor of the microbial traits except for significant but minor positive effect of mean annual temperature on α and α₀, and marginally significant effects on e₀ (Figure 6). In contrast, litter chemical properties were good predictors of the N limitation responses. Specifically, initial lignin content and C:N were positively correlated with N acquisition and recycling traits, but negatively correlated with CUE (Figure 6). Lacking significant climatic correlations, in the following we focus on relationships between microbial traits and litter chemistry.

Figure 6

Proportionally more N was acquired in N-poor litter, as indicated by increasing α with wider initial C:N ratio (Figure 7A). The estimated C-use efficiency decreased with increasing initial litter C:N when N recycling was neglected (Figure 7C). If instead N recycling efficiency was estimated at fixed CUE, η increased with increasing initial litter C:N (Figure 7E). In the models with dynamic traits, the parameters representing preferential N acquisition, C use, and N recycling efficiencies at zero litter N (α₀, e₀, and η₀) followed similar patterns as the corresponding time invariant traits (Figure 7B, 7D, and 7F). The CUE at zero litter N in model Cdyn decreased with increasing initial litter C:N, but a large fraction of these CUE values were at their lower boundary, indicating that the model was not flexible enough to capture patterns in the data with physically meaningful parameter values (Figure 7D).

Figure 7

Initial lignin contents modulated the trait-litter C:N relations (compare gray and black curves in Figure 7). At any given initial litter C:N, higher initial lignin promoted the preferential acquisition of N (Figure 7A and 7B), decreased CUE (Figure 7C and 7D), and promoted N recycling (Figure 7E and 7F).

In models Adyn, Cdyn, and Ndyn, we imposed increasing trends for α and η, and a decreasing trend for e with increasing litter C:N, so the general shape of the relations of these traits with current litter C:N in Figure 8 are not surprising. In general, the temporal trajectories of these traits as litter decomposes and the C:N ratio is reduced followed the same pattern emerging across litter types with varying initial C:N. Notably, the variation of η₀ was minor compared to that of η, indicating that our assumed relation between η and current litter C:N can capture most of the trait variation both across litter types and through time.

Figure 8

3.4 Long-term mean C-use efficiency across model variants

The mean CUE in models where N acquisition or recycling were regulated is equal to the maximum CUE (e_max) by default, and thus always higher than in models Cflex and Cdyn. Between the latter two models, the mean CUE calculated for Cdyn was higher than for Cflex (Figure 9A). Moreover, the mean CUE calculated using model Cdyn decreased with increasing initial litter C:N (Figure 9B)—similar to e and e₀ (Figure 7C and 7D). Overall, the comparison of mean CUE among models implies that flexible N acquisition or recycling allows microorganisms to use a larger share of the residue C for growth compared to CUE regulation.

Figure 9

4.1 Contrasting theories on microbial responses to nitrogen limitation

Previous work on litter decomposition had concluded that CUE is lower in litter with high initial C:N ratio (Manzoni, 2017; Manzoni et al., 2010), but such work was based on a model that did not consider preferential N acquisition or N recycling (i.e., models Cflex and Cdyn), so that all variation in litter N was attributed to changes in CUE. To fill this gap, here we considered N acquisition and N recycling as alternative responses to N limitation. Regulation of N acquisition alone did not explain well the N import/release data (poor fitting, Figure 5), even though theory predicts that the most limiting element should be preferentially targeted by microorganisms (Averill, 2014; Wutzler et al., 2017). Given these theoretical expectations, the low predictive power of our models based on regulation of N acquisition was surprising. Empirical evidence of N acquisition regulation is not entirely consistent. The ratio of C-acquiring to N-acquiring enzyme activities decreased slightly with increasing soil C:N ratio (more so when labile C was added) in one study (Karhu et al., 2022), but in another it did not respond to N addition, suggesting that the expression of individual enzymes might not only be upregulated in response to a specific resource limitation (Mori et al., 2021).

Our model results show that regulation of N recycling can be as effective as CUE regulation to compensate for stoichiometric imbalances, as indicated by the comparable fitting performance of models based on these two mechanisms (Figure 5). This result might explain why CUE regulation is not always detected despite wide stoichiometric imbalances (Hasby et al., 2021), and suggests that we should interpret previous results focusing on CUE regulation alone with caution. However, there is also direct evidence of reduced CUE at high substrate C to nutrient ratio in litter (Lashermes et al., 2016; Voriskova et al., 2011), soil (Takriti et al., 2018; Z. M. Lee & Schmidt, 2014), and aquatic systems (del Giorgio & Cole, 1998; del Giorgio & Newell, 2012; Godwin & Cotner, 2015). How can we reconcile this perspective on N recycling as a major avenue of stoichiometric regulation, with the documented decrease in CUE as substrate C to nutrient ratio widens? The answer to this question might be linked to the type of decomposers and how they interact with their environment.

N recycling can occur within a fungal mycelium, in unicellular organisms, or within communities with microbial cells in close proximity. Within a mycelium, growth of hyphal tips at the mycelial front may be supported by reallocation of nutrients from older parts of the mycelium (Boberg et al., 2014; Falconer et al., 2005; Fricker et al., 2017; Tlalka et al., 2002; Wells, Harris, & Boddy, 1998), depleting the nutrient content of senescent mycelium (compartmental senescence, Camenzind et al., 2023). In Ghersheen et al. (2025), a mathematical model featured this mechanism by explicitly incorporating fungal physiological responses to N limitation during litter decomposition. Under low N availability, the model describes an increase in the proportion of vacuolised mycelium and accumulation of N in slowly decomposing necromass, thereby conserving internal N and buffering stoichiometric imbalances through vacuolization, without invoking CUE regulation. This approach functionally resembles the flexible N recycling response in our model variant Nflex, where N recycling efficiency increases in response to nutrient limitation, but it also provides a mechanistic basis by linking N retention and recycling directly to fungal growth traits and mycelial turnover.

Bacteria can recycle N within their cells by reusing amino acids that have already been used for protein synthesis to synthesize new proteins, instead of releasing old N and forming proteins from newly acquired N. This internal reuse of N does not alter the N content of the microbial necromass, and so cannot be captured by the parameter η in our model. In addition, in communities of unicellular microorganisms, N released at senescence can be rapidly reused by active cells in the community without being returned to the substrate pool. For example, nutrients contained in bacterial cell walls and intracellular enzymes of dying dells are efficiently recycled in the growing cells (Hartl et al., 2017; Mayer et al., 2019). Therefore, for a N recycling mechanism to promote N retention when necromass is formed, we need either multicellular organisms that can translocate N within their bodies (i.e., fungi), or tight interactions among bacterial cells within colonies or biofilms.

We assumed that different microbial responses to N limitation are mutually exclusive, but it is likely they coexist. For example, bacteria can both increase their C:P ratio and reduce their CUE as the C:P supply ratio is increased (Godwin & Cotner, 2015). From a statistical perspective, it is difficult to determine the relative contributions of combined responses because they give rise to similar decomposition patterns (Figure 3A), creating an equifinality problem. Ideally, one could constrain some of the model parameters using functional trait observations and fit the remaining parameters to infer which responses are expressed and to what extent. However, such detailed functional trait data is too scarce for this approach to work. Given these limitations, we can only discuss which response is more likely to be dominant based on the fitting performance of the different model variants.

4.2 More efficient nitrogen recycling under nitrogen limitation

In our model variants where N recycling was flexible (Nflex and Ndyn) η increased when N was scarce. In particular, η from model Nflex increased with increasing initial litter C:N (Figure 7E), allowing excellent model fit to the data without CUE regulation. In model Ndyn, η was assumed to decrease during decomposition, as litter C:N decreased. While this response within a litter cohort was hardwired in the model, we did not constrain the value of η at the beginning of decomposition (technically, the value of η in a hypothetical litter without N). Notably, we found that this initial N recycling was higher in litter with higher initial C:N (Figure 7F), suggesting that microbial communities colonizing N-poor litter have higher N recycling capacity.

Evidence from a field study in a boreal forest supports this conclusion. Fungal mycelium was collected using sand-filled ingrowth bags inserted into the organic topsoil (Högberg et al., 2021). The C:N of the mycelium decreased from almost 40 to 10 across a local gradient of increasing N availability, and since a large proportion of the collected mycelium may be expected to be dead (with higher C:N than the live biomass), a major contribution of recycling of N from senescent mycelium seems likely. When easily available N (methylurea) was added to the ingrowth bags, mycelial C:N remained low regardless of soil N availability, suggesting that in the presence of abundant external N, internal N recycling might be less necessary. This evidence is suggestive of putative mechanisms, but to our knowledge there is no direct evidence of regulation of N recycling in response to changes in substrate C:N ratio. Such evidence would validate our interpretation of the patterns in η that we found.

4.3 Effects of nitrogen limitation response on microbial growth

Microbial CUE integrates numerous metabolic processes and trade-offs in a single macro-trait, and it has been argued that evolutionary pressure acting on these processes should lead to CUE values that maximize microbial growth (Allison, 2014; Roller & Schmidt, 2015; Westerhoff, Hellingwerf, & Vandam, 1983). In this light, we could expect that CUE varies along gradients of nutrient availability to ensure continued growth despite imbalanced resources (Sterner & Elser, 2002). In fact, low nutrient content in the substrates requires CUE to be low, otherwise N demand for growth would be too large compared to the amount provided by the external environment (Figure 1). Indeed, CUE of microorganisms and animals in both terrestrial and aquatic systems decreases with increasing substrate C:N and C:P ratios and increases when mineral N and P are added to nutrient limited systems, as expected from a CUE optimization hypothesis (Manzoni et al., 2017). Our models in which CUE was flexible (Cflex and Cdyn) showed a clear declining trend of CUE as the initial litter C:N increased (Figure 7C and 7D), confirming previous findings (Manzoni et al., 2010; Manzoni et al. 2017).

In addition to CUE downregulation at high litter C:N, other traits could also vary and allow microorganisms to grow even more than by regulating CUE alone. Indeed, reducing CUE under N limitation leads to large losses of C that could sustain growth at a later stage when N becomes available, or meet the C demand from the growing mycelium (Boberg et al., 2010). N recycling under N limitation (as shown in Figure 8C) could sustain growth without wasteful expenditure of C. With flexible N recycling, microbial communities can achieve a much higher long-term mean CUE than without N recycling. Mean CUE is the amount of C used for growth throughout the decomposition process and can thus be interpreted as a measure of the overall microbial capacity to convert litter C into biomass [Eq. (18)]. The highest mean CUE was attained in the models in which N recycling was regulated, because in those models CUE was always equal to the maximum value e_max. Therefore, we propose that N recycling could be optimized along nutrient availability gradients, as this allows higher growth than CUE regulation alone.

CUE regulation has an obvious opportunity cost, as lower CUE implies lower growth, but also regulation of N recycling can have direct and opportunity costs. Direct costs could be related to N transport within the mycelium, but such costs are probably low. An indirect opportunity costs arises because recycling N within the mycelium is associated with vacuolization of the older sections of the mycelium, thus decreasing the proportion of mycelium that is actively engaged in resource acquisition. Therefore, it is reasonable to expect that these costs would lower CUE when N recycling occurs, due to the maintenance respiration of less active mycelium. We have not explored these trade-offs in a systematic way, but we tested an additional model variant in which CUE decreases with increasing N recycling efficiency. Such a model performed as well as those in which N recycling alone is regulated (results not shown), while also allowing reasonably high mean CUE. Based on these results, we can speculate that N recycling might still be advantageous to sustain growth despite its C cost, but a more detailed quantification of the C costs of N recycling would be needed to substantiate this argument.

4.4 Microbial traits are largely independent of climatic and soil conditions, but vary with litter lignin

The model estimates of microbial traits varied with initial litter C:N ratio but were largely independent of climatic conditions. Litterbags were incubated under a wide range of conditions: mean annual temperature between –10°C and 27°C and mean annual precipitation between 150 and more than 10,000 mm y^–1. With such a wide range, climatic effects on microbial traits should have been detectable, but the model only predicts minor increases in preferential N acquisition in warmer climates (Figure 6). In particular, estimated CUE was independent of climate, confirming previous results using the same method (Manzoni, 2017; Manzoni et al., 2010).

It could be argued that including leaching from the litterbags in the models could have altered this result. To understand this argument, we can assume that leaching occurs continuously, removing a fraction λ of the decomposed organic matter. Mathematically, this assumption leads to N import/release curves where CUE always multiplies 1–λ (Eq. (6) in Manzoni et al., 2010). This means that our estimates of CUE should be corrected by a factor $\frac{1}{1–\lambda }$ to be unbiased. For example, a short-term laboratory experiment showed that ≈ 20% of C loss during litter decomposition can be attributed to leaching (Hansson et al., 2010), so that λ ≈ 0.11 if CUE ≈ 0.5. In this example, our CUE estimates would need to be corrected by a factor 1.12. While this model including leaching would not allow independent estimation of CUE and λ because they always appear as a product (again an equifinality issue), we can speculate on how CUE would change if λ was affected by climatic conditions. We could expect leaching to be higher in wet conditions, so that the corrected CUE would also be higher. Therefore, when accounting for leaching effects, we would find a positive effect of precipitation on CUE, though the quantitative importance of such a trend is hard to assess without reliable estimates of λ across our datasets.

Independent empirical evidence supports the conclusion that CUE is not affected by experimental warming when measured at the same temperature as in the field (Zhang et al., 2024). A global CUE meta-analysis also showed that climatic effects on CUE are less important than those on microbial growth rate (Hu et al., 2025), indicating that temperature and water availability affect the rates of decomposition (e.g., Parton et al., 2007) and microbial metabolism (as expected), but not C allocation between growth and respiration (Spohn, 2016). In contrast, soil CUE calculated from modeled C fluxes at the global scale declines with mean annual temperature (Tao et al., 2023). While it is possible that climatic controls on soil and litter CUE differ, here we conclude that microbial traits modulating C and N relations during litter decomposition are largely independent of climate.

In addition to climatic conditions, also soil N availability could affect microbial responses to N-poor litter. High N availability could in fact allow higher rates of N import in the litter layer, reducing the need to regulate N acquisition, CUE or N recycling. For example, inorganic nutrients increase CUE in systems that were originally nutrient limited (Manzoni et al., 2017). Indeed, the litter C:N at which N release starts (threshold element ratio, TER) tends to increase with increasing soil C:N (Ågren et al., 2013). In turn, TER is linked to microbial traits, as TER = α/[(1–η)e_rB] (this relation can be obtained by setting d_nL/d_cL = 0 in Eq. (11)). Thus, higher TER could indicate higher N acquisition (α is the equation for TER), lower CUE (e), or higher N recycling (η). In contrast to this expectation, in the subset of our dataset including top soil C:N (data from Trofymow & CIDET, 1998), we did not find any significant effects of soil C:N on the microbial traits (results not shown).

Lignin content modulated the strong effect of initial litter C:N on microbial traits—litter with higher lignin content exhibited lower CUE and higher N recycling efficiency (Figure 6). The lower CUE can be explained by the C cost of ligninolysis, mostly due to continuous production of hydrogen peroxide to maintain the functionality of oxidative enzymes (Chakrawal, Lindahl, & Manzoni, 2024; Moorhead et al., 2013 and references therein). Lignin also increases N recycling efficiency, suggesting that it interferes with N acquisition. It is possible that by shielding N-rich compounds, lignin (or its decomposition products as well as other aromatic compounds included in the measured lignin pool) lowers the availability of organic N (Adamczyk et al., 2019), thus triggering more efficient N recycling (Sun et al., 2024). In other words, in lignin-rich litter, microorganisms perceive a higher litter C:N than the actual value and respond accordingly by regulating N recycling.