Temporal Shifts in Plant Diversity Effects on Carbon and Nitrogen Dynamics During Litter Decomposition in a Mediterranean Shrubland Exposed to Reduced Precipitation

Climate and plant diversity are major determinants of carbon (C) and nitrogen (N) dynamics in decomposing plant litter. However, the direction and extent to which these dynamics are affected by combined changes in climate and biodiversity are not well understood. We used a field experiment in a Mediterranean shrubland ranging from one to four shrub species with partial rain exclusion (− 12%) to test how lower precipitation interacts with shrub species diversity to influence C and N release during decomposition. We also distinguished between first-year (0–12 months) and second-year decomposition (12–24 months) to test the hypothesis of stronger diversity effects at the beginning of the decomposition process. Litter C and N release increased with litter species richness during the first year, but not during the second year of decomposition. However, these richness effects were weak and less consistent than litter composition effects, which persisted over time and became even stronger for C release after 2 years of decomposition. Partial rain exclusion reduced N release by 17% only during the first year and had no effect on C release in either year. Community-weighted mean (CWM) traits and functional dissimilarity (FD) of litter traits contributed both to explain litter species composition effects. These litter trait effects were not altered by partial rain exclusion, but were more important after 2 years than after 1 year of decomposition. Our findings suggest increasing trait legacy effects with ongoing decomposition. More generally, our data showed that changes in the diversity of dominant shrub species had stronger effects on C and N release during litter decomposition than a moderate reduction in precipitation.


9
underneath the gutters in both control and rain-excluded plots, and (iii) by determining the 176 gravimetric soil water content in soil samples from control and exclusion plots. These 177 measurements indicated that we only occasionally reached the target value and that our 178 exclusion system reduced the average annual precipitation by only 12 ± 2% compared to the 179 control plots. This seems to be mostly the consequence of wind turbulence during rainfall 180 rendering the gutters covering the relatively small plot area of 16 m 2 at a height of about 2 m 181 above the ground not as effective as we anticipated. An average of 12% less rainfall, however, 182 is close to the 15% mean annual decrease reported by the projections of 14 CMIP3 global 183 climate models for the Mediterranean Basin (Mariotti and others 2008) and to the 10% mean 184 annual decrease reported by the projections of 28 CMIP5 global climate models for southern 185 France (Polade and others 2014). Moreover, we stress that a given percentage of excluded 186 rainfall does not readily translate into a similarly reduced soil water content, which depends 187 also on the total amount of precipitation during a single rainfall event and how these events are 188 distributed. As a consequence, during certain rain events we measured between -13% and -24% 189 lower soil volumetric water content at 10 cm soil depth in plots with partial rain exclusion 190 compared to control plots ( Supplementary Fig. S1). where pi and pj are the relative abundance for shrub species i and j in the litter mixture, and dij 225 the Euclidian distance between species i and j for the trait considered. Because the measured 226 traits differ in their units, we used normalized values (using a z-scored standardization so as to 227 get a mean of zero and a standard deviation of one) to calculate functional dissimilarity. 228 The remaining leaf litter was collected from the plastic cylinders after one and two years 229 of field exposure. We considered the first year as an initial stage of decomposition that is 230 dominated by leaching losses, which can account for up to 30% of initial mass depending on 231 litter species (Berg and McClaugherty 2008). Mass loss in our study varied between 18.6% 232 (Ulex) and 36.5% (Cistus) in the single species treatments after one year. The advanced 233 decomposition during the second year, we defined here as a later stage of decomposition with 234 a range of mass loss between 27.0% (Ulex) and 53.0% (Quercus) in the single species 235 treatments after two years. We retrieved four replicates in December 2012 (368 mesocosms) 236 and three replicates in December 2013 (276 mesocosms) of plot-specific leaf litter. The 237 remaining leaf litter was put in plastic bags, and immediately transferred to the laboratory. Leaf 238 litter was separated into species, thoroughly brushed to remove adhering soil particles, freeze-239 dried (Lyovac GT2®) and weighed to obtain litter dry mass data of each species in each 240 mesocosm. After weighing the component litter species, all litter from an individual field 241 mesocosm was again put together and then ground using a ball mill to a fine powder before 242 chemical analyses. Carbon and N concentrations were measured from remaining litter material 243 using the same procedure as described for initial concentrations. Initial and final concentrations 244 of C and N after one year, and after two years of decomposition were multiplied with initial, 245 and final litter mass after one and two years, respectively, for the calculation of the amount of 246 C and N loss. The difference between initial amounts and those remaining after one year was 247 used to calculate total C and N release during the first year of decomposition. The difference 12 between the amounts remaining after one year and after two years was used to calculate total C 249 and N release during the second year of decomposition. 250 251

Statistical analyses 252
Statistical analyses were performed with the R software (R Core Team 2016) with 253 significance levels indicated as * for P < 0.05, ** for P < 0.01 and *** for P < 0.001. We used 254 a linear mixed-effects model approach ("nlme" package) to test the effect of diversity, partial 255 rain exclusion, and decomposition (initial and later) stage and their interactions on C and N 256 release. To take into account the fact that we had four and three replicate mesocosms per plot 257 for first year and second year, respectively, the random part of the model indicated that the 258 mesocosms were nested within plots with the following R syntax ("random = 259 ~1|plot/mesocosm"). Due to the large number of potentially important predictors, we carried 260 out three distinct statistical models to test for the litter diversity effect on C and N release. The Both C and N release from decomposing litter differed strongly between the first and 299 second year of decomposition, regardless whether the model fitted litter species richness (Table  300 1) or litter species composition ( Table 2). The C release was higher during the first year 301 compared to the second year of decomposition ( Figs. 1 and 2). In contrast, there was overall 302 more N released during the second year compared to the first year of decomposition (Figs. 1 303 and 2). 304 In the statistical models including the effect of species richness, we found that litter C 305 and N release were significantly affected by the interaction between year of decomposition and 306 species richness (Table 1). This interaction resulted because the proportion of C and N release 307 increased with species richness during the first year but not during the second year of 308 decomposition ( Fig. 1a and 1b). After one year of decomposition, the average C release 309 increased from 31% in monospecific litters to 35% in 4-species mixtures, whereas the N release 310 increased from 8% in monospecific litters to 17% in 4-species mixtures. The amount of 311 variation explained by species richness was comparatively small, especially for C release. 312 However, the positive relationship with species richness was robust, with similar or even higher 313 variation explained when the 4-species level was excluded from the analysis (R 2 = 0.14 and 314 0.13 for C and N loss, respectively), or when the few microcosms showing apparent N 315 immobilization (mostly at low species richness) were excluded from the analysis (R 2 = 0.17 for 316 N loss). 317 In the complementary statistical models testing the importance of species composition 318 (i.e. the 15 distinct litter treatments), we found that litter species composition strongly affected 319 C and N release (Table 2; Supplementary Table S2). As indicated by the significant litter 320 composition × year interaction, the differences between the 15 litter treatments depended on 321 the year of decomposition (Table 2; Supplementary Table S2). Most of the litter treatments lost 322 roughly twice to three times less C during the second year compared to the first year of 323 15 decomposition, but a few treatments lost much less (Ulex single species litter), or almost the 324 identical amount of C (Quercus alone or mixed with Cistus) during the second year compared 325 to the first year of decomposition. For N loss, the differences between one and two years of 326 decomposition varied even more (Supplementary Table S2). The differences between the litter 327 treatments increased during the second year compared to the first year of decomposition for C 328 release (CV = 16% and 36% for one and two years of decomposition, respectively), but not for 329 N release (CV = 53% and 45% for one and two years of decomposition, respectively) 330 (Supplementary Table S2). 331 As litter composition significantly affected the C and N release, we further assessed the 332 effects of the presence or absence of particular litter species (i.e. effects of litter species identity; 333 Table 3; Fig. 3). The statistical model incorporating species identity effects ( Table 3) showed 334 that the presence of Quercus consistently led to higher C and N release (Fig. 3). The presence 335 of Quercus litter enhanced C release to a larger extent during the second year than during the 336 first year of decomposition (+51% and +17%, respectively) but enhanced N release to a larger 337 extent during the second year (+240%) compared to the first year of decomposition (+69%) 338 (Table 3; Fig. 3). The presence of Cistus increased the release of C (+13%) but not that of N, 339 regardless of the year of decomposition (Fig. 3). The presence of Ulex generally decreased C 340 and N release, and these effects were more pronounced during the second year (-37% and -30% 341 for C and N release, respectively) compared to the first year of decomposition (-14% and -21% 342 for C and N release, respectively) ( Table 3

Consequences of reduced precipitation on C and N release during decomposition 350
The partial rain exclusion we applied to the experimental plots overall had weak effects 351 on C and N release during decomposition. In fact, partial rain exclusion did not have a 352 significant main effect in any of the statistical models we ran (Tables 1 to 4). The release of C, 353 but not that of N, however, was distinctly affected by reduced rainfall among the different litter 354 treatments (small but significant litter composition × partial rain exclusion effect, Table 2). 355 With less rainfall, the monospecific Rosmarinus litter released less C (25% vs. 37%) whereas 356 the monospecific Ulex litter released more C (23% vs. 20%). Overall, these differences were 357 small and most of the litter treatments did not show any difference between plots with partial 358 rain exclusion and control plots (data not shown). Partial rain exclusion interacted with the year 359 of decomposition to influence N release, but not that of C ( to control plots during the first year of decomposition, while N release was similar in both 362 treatments during the second year of decomposition (Fig. 2b). 363

CWM-versus FD-trait control over C and N release 365
Principal component analysis (PCA) of CWM traits showed that the first PCA axis 366 (CWM1) explaining 61.7% variation was determined by high scores of P concentration and 367 WHC, but low values of C concentration, and low ratios of C:P, N:P and lignin:P (Fig. 4a). 368 High scores of the second PCA axis (CWM2) explaining 28.2% variation were related to high 369 values of lignin concentration, and high ratios of C:N and lignin:N, while low scores were 370 associated with high concentrations of N and phenolics. When considering the functional trait 371 dissimilarity, the first PCA axis (FD1) explaining 43.3% variation separated litter mixtures 372 according to increasing dissimilarity of WHC values, the concentrations in C and P, and N:P 373 and C:P ratios (Fig. 4b). High scores along the second axis (FD2) explaining 28.1% variation 374 were largely determined by N-related traits with increasing dissimilarity in N concentration, 375 and C:N and lignin:N ratios, while low scores were related to increasing dissimilarity in the 376 concentrations of lignin and phenolics (Fig. 4b). 377 When evaluating conjointly the effects of CWM and FD traits in linear mixed-effects 378 models, we found that C release was mostly controlled by the CWM2 and the interactions 379 CWM1 × year and FD2 × year (Table 4) The N release was significantly affected by CWM2, FD1 as well as the interactions 389 CWM1 × year, FD2 × year and rainfall reduction × year (Table 4). Similar to what we observed 390 for C release, increasing CWM1 scores and decreasing CWM2 scores were related to higher N 391 release (Fig. 5), with a stronger impact of CWM1 during the second year of decomposition 392 (significant CWM1 × year interaction, Table 4; Fig. 5). In contrast to C release, functional trait 393 dissimilarity showed strong main effects on N release (Table 4)

CWM-versus FD-trait control over C and N release 509
The functional trait-based metrics CWM (community weighted mean) and FD 510 (functional dissimilarity) allow to distinguish between mass-ratio and niche differentiation as 511 two key mechanisms of diversity effects. Actually, the models incorporating CWM and FD 512 explained more variation than the models based on species richness alone and a similar amount 513 of variation as the models based on species composition and species identity (Tables 1 to 3). 514 The metrics CWM and FD both predicted C and N loss, with FD being more important overall 515 for N release compared to C release. Carbon release was most strongly related to the first 516 component of the CWM-trait PCA, indicating that increasing total litter P concentration and 517 water holding capacity (WHC) stimulate C release. Accordingly, when the two relatively P-518 rich litter species Cistus and Quercus, which also had comparatively high WHC, were present 519 in litter mixtures, C release was higher compared to when they were absent. Along with N, P is 520