Heterotrophic soil respiration and carbon cycling in geochemically distinct African tropical forest soils

Heterotrophic soil respiration is an important component of the global terrestrial carbon (C) cycle, driven by environmental 10 factors acting from local to continental scales. For tropical Africa, these factors and their interactions remain largely unknown. Here, using samples collected along topographic and geochemical gradients in the East African Rift Valley, we study how soil chemistry and fertility drive soil respiration of soils developed from different parent materials even after many millennia of weathering. To address the drivers of soil respiration, we incubated soils from three regions with contrasting geochemistry (mafic, felsic, and mixed sediment) sampled along slope gradients. For three soil depths we measured the potential maximum 15 heterotrophic respiration under stable environmental conditions as well as the radiocarbon content (ΔC) of the bulk soil and respired CO2. Our study shows that soil fertility conditions are the main determinant of C stability in tropical forest soils. We found that soil microorganisms were able to mineralize soil C from a variety of sources and C quality under laboratory conditions representative of tropical topsoil. However, in the presence of organic carbon sources of poor quality or the presence of strong mineral related C stabilization, microorganisms tend to discriminate against these energy sources in favor of more 20 accessible forms of soil organic matter, resulting in a slower rate of C cycling. Furthermore, despite similarities in terms of climate and vegetation, soil respiration showed distinct patterns with soil depth and parent material geochemistry. The topographic origin of our samples was not a main determinant of the observed respiration rates and ΔC. In situ, however, soil hydrological conditions likely influence soil C stability by inhibiting decomposition in valley subsoils. Our results demonstrate that even in deeply weathered tropical soils, parent material has a long-lasting effect on soil chemistry that can influence and 25 control microbial activity, the size of subsoil C stocks, and the turnover of C in soil. Soil parent material and its control on soil chemistry need to be taken into account to understand and predict C stabilization and rates of C cycling in tropical forest soils.


Introduction
Tropical forests and the soils therein are one of the most important and largest global terrestrial carbon (C) pools and serve as important climate regulators (Cleveland et al., 2011;Kearsley et al., 2013;Lewis et al., 2009;Sayer et al., 2011). They contain about one third (421 Pg C) of the global soil organic carbon stock (SOC) in the upper one meter of soil (Köchy et al., 2015) 35 and are characterized by high annual C turnover rates (Raich and Schlesinger, 1992). Generally, climatic parameters (temperature and precipitation) and vegetation input are regarded as the main factors controlling C dynamics in natural tropical systems (Davidson et al., 2000;Davidson and Janssens, 2006;Rey et al., 2005). Vegetation and climate can stimulate or hamper microbial activity and mineralization of C through quality and quantity of organic matter (OM) input to soil (Fontaine et al., 2007), and the availability of water and energy to drive microbial processes. However, recent studies show that SOC 40 dynamics are controlled by a much more complex interplay of geochemistry, topography, climate and biology (Doetterl et al., 2015b(Doetterl et al., , 2018Haaf et al., 2021;Luo et al., 2017Luo et al., , 2019 much like pedogenesis in general. For example, on average 72% of SOC in humid forest biomes is stabilized by interaction with the mineral phase in soil organo-mineral interactions and occlusion by aggregation (Kramer and Chadwick, 2018). Geology can control C dynamics as soils developed from felsic parent material (high SiO2, low Fe & Al, slow chemical weathering rate) provide less potential for C stabilization and a lower capacity 45 to release rock-derived nutrients than soils developed from mafic parent material (low SiO2, high Fe & Al, fast chemical weathering rate), limiting organic matter input. Additionally, topography through its control on water and soil fluxes may influence C dynamics by altering C respiration and input along slope gradients (Berhe et al., 2008). Hydrological features related to topography in tropical forests are likely to influence C cycling and explain spatial patterns of SOC distribution locally by limiting C decomposition in water saturated valleys (Kwon et al., 2013). Finally, some soils developed from sedimentary 50 parent material can contain a large fraction of fossil organic carbon (fFOC) of generally poorer quality than fresh organic matter inputs, which can be resistant to decomposition under in situ environmental conditions (Kalks et al., 2021). Hence, in order to explain SOC and its exchange between soil and the atmosphere, the interactions of geochemical, geomorphic and climatic drivers are central (Angst et al., 2018;Berhe et al., 2012;Doetterl et al., 2015b;von Fromm et al., 2021;Kramer and Chadwick, 2018;Luo et al., 2017) 55 To date, it is not clear if the relationships between soil geochemistry, topography and climate identified for temperate ecosystems also apply in the tropics. Especially for the African tropics, more work is required to understand how soil geochemical, physical, biological and topographic features interact to influence SOC dynamics. Established observatories in African tropical forests have focused mostly on biodiversity preservation and C storage in the phytosphere (Tyukavina et al., 2013;Xu et al., 2017), while soils have received much less attention and remain understudied. Generally, data on SOC 60 dynamics from tropical regions are rare compared to the temperate zone, originating mostly from the Amazon basin (Quesada et al., 2020;Schimel et al., 2015;Schimel and Braswell, 2005) and their application to the African tropics may be limited. For example, atmospheric nitrogen deposition is much higher in sub-Saharan Africa than in other tropical regions due to large amounts of recurring biomass burning originating from savanna and dry forests north and south of the humid tropics (Bauters et al., 2018). 65 Furthermore, long-term chemical weathering in tropical systems has led to the depletion of rock-derived nutrients in soils and has limited the capacity of microorganisms and plants to access these nutrients (Liu et al., 2015;Vitousek and Chadwick, 2013). It is likely that variation in soil weathering stage and nutrient availability in tropical forests affect soil C storage and the exchange of C between plants, soil and the atmosphere. For example, due to their tight coupling driven by the metabolic needs of plants and microorganisms, changes in nutrient availability such as nitrogen (N) and phosphorus (P) can greatly alter the 70 terrestrial C cycle, partly because CO2 uptake by terrestrial ecosystems strongly depends on N and P availability (Fernández-Martínez et al., 2014). Furthermore, low N and P availability limits microbial growth and activities and therefore affects the cycling of organic matter (Jing et al., 2020;Liu et al., 2015). Thus, nutrient limitations in highly weathered tropical soil likely force plant communities to alter belowground and aboveground C allocation (Doetterl et al., 2015a;Fisher et al., 2013;Wright et al., 2011), with more roots growing in organic rich topsoil, reducing C input to deeper soil layers (Addo-Danso et al., 2018), 75 thereby affecting SOC stocks.
Additionally, along soil age gradients, SOC stabilization by clay first increases and then decreases, with a reduction in reactive mineral surfaces as weathering advances (Doetterl et al., 2018;Kramer and Chadwick, 2018). As a consequence, clay in old tropical soils has a rather limited potential to protect C against microbial decomposers compared to younger temperate soils (Doetterl et al., 2018;Ngongo et al., 2009). In contrast, stable microaggregates rich in iron (Fe) and aluminum (Al) 80 oxyhydroxides found in abundance in tropical soils (Bruun et al., 2010;Torres-Sallan et al., 2017) seem to be of greater importance in stabilizing C in tropical soils, as concentrations of Al and Fe are commonly higher than in many temperate soils (Khomo et al., 2017). This is confirmed by studies conducted across a wide range of tropical ecoregions showing that SOC is mainly regulated by Fe or Al (hydr) oxides, more so than by clay content (Fang et al., 2019;von Fromm et al., 2021;Rasmussen et al., 2018). 85 Hence, understanding tropical soils C dynamics ultimately depends on our mechanistic understanding of these complex interactions and the ability to determine the primary environmental controls on SOC content and respiration. In our study, we aim to answer if C release through heterotrophic respiration from forest soils in the humid tropics follows predictable patterns related to geochemical soil properties and topography. We postulate that, in the absence of anthropogenic disturbance, soil geochemistry derived from its parent material has a lasting effect on soil C respiration due to its influence on stabilization 90 mechanisms and soil fertility, even in deeply weathered natural tropical soils. We selected soils in our study that developed from geochemically distinct parent material along slope gradients under comparable tropical climate and vegetation. We hypothesize that (1) specific soil respiration and the Δ 14 C signature of potential soil respiration in tropical soils are primarily controlled by geochemical properties related to soil fertility derived from and varying with soil parent material. These The presence or absence of C stabilization mechanisms, in soils, related to mineral geochemistry and soil formation, can increase SOC stocks and decrease heterotrophic C respiration rates by creating an energetic barrier for C decomposers, for example through complexation with organic molecules or by forming stable (micro) aggregates. (3) The topographic origin of a soil sample controls specific soil respiration and its Δ 14 C signature indirectly through the environmental conditions under which soil C decomposition took place in situ, modifying the quality and quantity of the available SOC stock prior to the 100 experiment.

Study sites
Our study sites are located in three forested national parks along the Albertine Rift System at the borders between Uganda, Rwanda and the Democratic Republic of the Congo (DRC), in the East African Rift Valley system . The 105 climate of the study region is classified as tropical humid with weak monsoonal dynamics (Köppen Af-Am). Mean annual temperature (MAT) is around 15.3-19.3 °C and mean annual precipitation (MAP) varies between 1697-1924 mm (Fick and Hijmans, 2017). Study sites are located between 1300-2200 m above sea level in sloping mountainous landscapes with small flat plateaus and ridges, followed by longer, steep slopes (up to 60% slope steepness) and small, v-shaped valleys. The dominant vegetation in all forests across the region is primary tropical mountain forest with smaller differences in biodiversity 110 and species composition (van Breugel et al., 2020;Doetterl et al., 2021;Verhegghen et al., 2012) Across the study region, we investigated soils developed from three geochemically distinct parent materials (mafic magmatic, felsic magmatic and mixed sedimentary rocks). Study sites in the DRC are located in Kahuzi-Biega National Park (-2.31439° S; 28.75246° E) where soils have developed from mafic magmatic rocks, a result of volcanism in the East African Rift System (Schlüter, 2006). Mafic magmatic rocks in the region are characterized by high Fe and Al and low Si content as well as a high 115 content of rock-derived nutrients such as base cations, and P (Table 1). Study sites in Uganda are located in Kibale National Park (0.46225° N; 30.37403° E) where soils have developed from felsic magmatic and metamorphic rocks. The felsic magmatic rocks in our study region are characterized by the gneissic-granulitic complex with low contents of Fe, and Al, and high Si content. Unlike mafic, felsic magmatic rocks in our study sites are characterized by low content of rock-derived nutrients (Table 1). Study sites in Rwanda are located in Nyungwe National Park (-2.463088° S; 29.103834° E) where soils 120 have developed from a mixture of sedimentary rocks of varying geochemistry. These sediments are mostly dominated by quartz-rich sandstones and schist layers spanning along the Congo-Nile divide in the western province of Rwanda (Schlüter, 2006). Similar to the felsic magmatic soils, mixed sediments in our study sites are characterized by low Fe and Al content but high Si content and low content of rock-derived nutrients. A specific feature of the mixed sedimentary rocks in our study a high C:N ratio (153.9 ± 68.5), and is depleted in N Reichenbach et al., 2021) Dominant soil types in the region are various forms of deeply weathered tropical soils. Following the World Reference Base for Soil Resources (IUSS Working Group WRB, 2015), soils in the mafic region are described as umbric, vetic and geric Ferralsol and ferralic vetic Nitisols. Soils in the mixed sediment region and the felsic region are described as geric and vetic

Soil sampling
As part of a larger project (Project TropSOC) , soil samples were collected following a catena approach with three 40 m x 40 m plots (field replicates) at four topographic positions (plateau, upper slope, middle slope and valley/foot 145 slope), resulting in 12 plots within each geochemical region (mafic, felsic, and mixed sediment). Each plot was subdivided into four subplots of 20 m x 20 m from which four 1-meter soil cores were taken using a cylindrical soil core sampler for undisturbed sampling. Cores were separated into 10 cm increments. The corresponding increments from four cores per plot were mixed into a depth-explicit composite sample. For the experiments conducted in this study, we selected 112 soil samples covering three depth categories including topsoil (0-10 cm), shallow subsoil (30-40 cm), and deep subsoil (60-70 cm). We 150 selected these three depth intervals as they cover a wide range of biogeochemical properties in soil and various levels of organic matter input to soil, both in terms of quantity (more C input near the surface, less at depth) and quality (leaf litter + root derived C in topsoil; root-derived C in subsoil). Note that as part of our own data analyses, published values for SOC stocks and bioavailable Phosphorus (bio-P) ) are reported in results section 3.1. For a more detailed description of the study design, soil sampling, sample treatment and analysis of all biogeochemical parameters used in this study, see the 155 complete database description in Doetterl et al. (2021).

Potential heterotrophic soil respiration
Heterotrophic respiration per gram SOC (specific potential respiration "SPR", µgCO2-C gSOC -1 h -1 ) was assessed in a lab-160 based incubation experiment and measured for the three sampling depths across geochemical and topographic gradients.
Briefly, 50 g of 12 mm sieved air-dried soil were weighed into a 100 ml beaker. Soil samples were sieved to 12 mm to homogenize the substrate while maintaining aggregate structure at a low level of disturbance. Soil moisture was adjusted to 60% water holding capacity, selected as the optimum water content level for microbial activity (Rey et al., 2005). Each beaker was placed inside an open 955.5±1.3 ml mason jar covered with parafilm, allowing for air exchange to avoid oversaturation 165 of CO2 within the jar that could inhibit microbial activity. Samples were then incubated at 20 °C, similar to the annual mean temperatures of the study sites. Except for keeping soil moisture steady by adding water when necessary, no further amendments were made to the incubated soils. Following a pre-incubation period of 4 days to allow equilibration, we incubated all samples for 120 days and sampled periodically every 1 to 14 days throughout the experiment with longer intervals towards the end of the experiment as respiration rates leveled off. The incubation experiment ended when additional CO2 production 170 was not detectable within measurement error. This was the case when the standard deviation of means of the respiration rate between three consecutive measurement time points was smaller than the standard deviation between three replicates of the same measurement time point. For CO2 accumulation prior to sampling, mason jars were sealed for several hours per measurement point. The accumulated CO2 was sampled using a syringe and transferred to pre-evacuated 20 ml vials. To avoid CO2 saturation effects during measurements, potentially influencing microbial decomposition processes, jars were flushed 175 with background air from the laboratory and checked for moisture content, before and after sealing to accumulate CO2.
Generally, CO2 samples were taken after accumulating between 1000-3000 ppm CO2. The CO2 concentration of the extracted gas was subsequently measured using a gas chromatograph (Trace TM 1300, Thermo Fisher Scientific, Massachusetts, United Stated) calibrated with five CO2 standards covering the range of measured concentrations (0, 500, 1000, 5000, 10000 ppm CO2). Further, the measured CO2 was corrected for the CO2 concentration of the ambient air that was used to flush the jars 180 before closing for CO2 accumulation. After each measurement was completed, each jar was opened and covered with parafilm to allow gas diffusion between CO2 accumulation periods. In this way, an average of 12 observations of CO2 production rate per incubated sample were conducted during the course of the experiment. Our aim was to compare average respiration between samples rather than the absolute values through the entire period of the experiment. Thus, we analysed data as the weighted average of SPR over the entire length of the experiment after respiration levelled off. We defined the weight by how 185 many days of the incubation experiment each observation represents. Additionally, we incubated 20% of all samples in triplicate to assess the average difference between samples for the experiment. The resulting average standard error of the mean between the three lab replicates was 9.6%.

Δ 14 C of bulk soil and respired CO2 190
We measured the soil radiocarbon ( 14 C) content of both bulk soil (SOC) and the corresponding respired CO2 from our incubation samples. Bulk soil Δ 14 C provides an indicator of the persistence of C in the soil and its age (Shi et al., 2020), while the Δ 14 C of the respired CO2 reflects more actively cycling C (Trumbore, 2009). The difference between these measures (Δ-Δ 14 C) can provide an indicator of how homogeneous or heterogeneous the system is , depending whether the Δ 14 C signature of the respired CO2 is similar to, or differs from the bulk soil Δ 14 C. Radiocarbon analyses were conducted 195 on composite samples of the bulk soil replicates used for incubation and correspondingly on composite samples of the respired CO2 during incubation. Bulk soil Δ 14 C was measured on soil samples before the incubation started. The Δ 14 C of respired CO2 was measured from CO2 that accumulated over the initial period following the pre-incubation period. The CO2 accumulation period varied depending on sample. For top and shallow subsoil with higher CO2 respiration rates it took on average 4-7 days while for deep soil with low CO2 it took 10-15 days to accumulate 1 mg C needed for Δ 14 C analysis. After accumulation, 120 200 ml of headspace gas from each field replicate incubation jar was sampled using a syringe. These replicate samples were transferred into a single 400 ml pre-evacuated Restek canister for composite analysis. Radiocarbon concentrations presented in our study are given as fractions of a modern oxalic acid standard following the conventions of Stuiver and Polach (1977).
All measurements were done with the MICADAS Mini Carbon Data System (IonPlus, Switzerland) at the AMS facility at Max Planck Institute for Biogeochemistry (Jena, Germany) (Steinhof et al., 2017). 205

Assessing fossil vs. biogenic organic carbon
Radiocarbon measurements were used to assess differences in the age of respired versus soil carbon, and to estimate the potential contribution of fossil organic C to the total soil organic C content and to CO2 respired during incubations. For the latter, we focused on the mixed sediment region, as this is the only geochemical region in our study where soil parent material 210 contains fossil organic C. We used a two-end member mixing model following (Schuur et al., 2016) to calculate the fraction of the C in the sample originating from biogenic vs. fossil organic C as follows: where FFOC, Fbio and Fsample represent the fraction modern radiocarbon (F), of fossil organic C, biogenic C and the measured sample (bulk soil organic C or respired CO2) respectively; and fFOC and fbio represent the proportion of fossil organic C and biogenic C contributing to a sample's total C content. For this estimate, we assumed that fossil organic C is free of 14 C due to the high age of parent material Schlüter, 2006). Further, we assumed that radiocarbon values of biogenic SOC (Fbio) in the mixed sediment region follow the same trend with soil depth as the mean radiocarbon content measured 220 depth-explicit from plateau soils of the mafic and felsic regions (regions without fossil organic C), and that these values represent biogenic SOC from active biological cycling in plant-soils systems (Cerri et al., 1985;Kalks et al., 2021). However, because rates of biogenic C cycling likely vary across sites, with potentially slower biogenic cycling in mixed sediment region (see discussion), this estimate is likely an upper bound on the fossil organic C contribution to these samples. Based on these assumptions, we reduced Eq. (1) and solved for the proportion of biogenic organic C (fbio) as follows: 225 The fraction of fossil organic C was then calculated as follows:

Assessing patterns of respiration and Δ 14 C
To examine differences in mean SPR in relation to the three main factors topographic position ̶ soil depth and geochemical region ̶ we conducted three-way analysis of variance (ANOVA). Before ANOVA, we conducted residual analysis to test for 235 the assumptions of ANOVA, Shapiro-Wilk's test of normality distribution and Levene's test for homogeneity of variances (Shapiro and Wilk, 1965). In most cases, the homogeneity and normality tests did not meet the requirement due to the natural variability of the samples and the factorial sampling design. Hence, we used square root and log transformation to approximately conform to normality and conducted the ANOVA tests on the transformed dataset. To compare the means of multiple groups, post-hoc pairwise comparison was applied using Bonferroni correction (Day and Quinn, 1989) or Tamhane 240 T2 in the case of unequal variances (Tamhane, 1979).

Predicting SPR and Δ 14 C
Multiple linear regression was used to assess the explanatory power of soil properties to predict SPR and the difference between the soil and respired CO2 Δ 14 C signature (Δ-Δ 14 C). Before running regression models, we extracted a wide range of physico-245 chemical soil properties and SOC quality indicator (C:N) for our investigated soils, where, in parallel studies, soil C stabilization mechanisms were assessed by Reichenbach et al. (2021) and microbial activity parameters (CNP enzymes and microbial biomass) assessed during our incubation experiment (Table B1) by Kidinda et al. (2020). Overall, our dataset consisted of 37 independent variables and 112 aggregated observations for each of our target variables (For CO2: Aggregated from 1350 individual observations of SPR over the course of the experiment). As multicollinearity and autocorrelation between 250 independent variables was to be expected due to this large number of independent variables and a relatively small number of aggregated observations, we conducted rotated principal component analysis (rPCA) for dimension reduction (Jolliffe, 1995), before regression analysis. We then named all retained rotated components (RCs) based on the loadings of the original variables and interpreted them for the likely underlying mechanisms that can affect C dynamics (Table B1). We used a threshold of r > 0.5 to decide whether an independent variable that is loaded into an RC is used for the mechanistic interpretation of the RC or 255 not. We used an eigenvalue >1 and explained proportion of variance >5% for each RC as criteria to include or exclude RCs into our regression models (James et al., 2013;Jolliffe, 1995). Furthermore, we used p-Values (p<0.1) and standardized coefficients to evaluate the contribution of explanatory power of individual RCs to the overall model while the F-statistic was used to evaluate the overall relationship between RCs and SPR or Δ14C for every model. Note that we found no statistical difference in SPR or Δ 14 C between the plateau and slope positions within each studied geochemical region (mafic, felsic, and 260 mixed sediment). Across geochemical regions and soil depths, SPR, and Δ 14 C differed only between valleys and non-valley positions. Hence, all further analyses were done after splitting the data into two subsets: (1) non-valley positions (plateau, upper slope, and middle slope) versus (2) valley positions (valleys and foot slope). When predicting our target variables, each regression analysis was done for three subsets of data: One model containing all data, one with only topsoil data and one with only subsoil data. Samples from valley positions were excluded from this part of the analysis due to the sample size of the 265 valley subset being too small (9 sites, 27 soil samples) for reliable regression analyses. Hence, our analysis to identify controls via regression and RCs to predict SPR and Δ 14 C is focused on non-valley positions (27 sites, 85 soil samples).

Assessing relative importance of explanatory variables
Lastly, we assessed the relative importance of each individual RC in predicting the target variables by interpreting the 270 standardized coefficients (-1 to 1) and p-values associated with each regression model. When predicting target variables with all data, soil depth was included as an additional explanatory variable beside the rotated components (RCs). We did this in order to avoid interpreting variables as being important for the model when they were instead just auto-correlated with soil depth. In a final step, we used partial correlation analysis following Doetterl et al. (2015b), to interpret the explanatory power of independent variables in our model, while controlling for soil depth. We contextualize our findings with respect to microbial 275 (extracellular enzyme activity) (Kidinda et al., 2020) , mineralogical (pedogenic oxides) (Reichenbach et al., 2021) and soil fertility parameters (available nutrients and exchangeable base cations) . For all statistical tests, due to the relatively small sample size and to avoid Type II statistical errors, a threshold of p<0.1 was used to indicate significant difference. R 2 and root mean squared error (RMSE) were used as evaluation metrics for model performance. All statistics were performed using R statistical software and the packages "psych", and "ppcor" (R Core Team, 2019). 280

Topography and soil depth
For all three geochemical regions in non-valley topographic positions, SPR and Δ 14 C decreased with soil depth (Fig. 1a-b and   2a-b, respectively). For SPR, differences with soil depth were smallest for sites in the mixed sediment and largest for sites in the mafic region. For Δ 14 C, relative changes with depth were similar for mafic and felsic geochemical regions in both soil and 290 respired CO2, but samples from the mixed sediment region were consistently more depleted in Δ 14 C than their counterparts from mafic and felsic regions (Fig. 2a-b).
In valley positions, SPR did not follow a clear trend with soil depth. In the mafic region, SPR decreased with depth, while in the felsic region it increased with depth (Fig. 1b). No statistically significant differences in SPR with depth were observed for the mixed sediment region (Fig. 1b). All regions show a strong trend of depletion of Δ 14 C with depth in valleys ( Fig. 2a-b). 295

Geochemistry and soil depth
Consistently, for non-valley profiles, SPR was higher in felsic and mafic regions than in the mixed sediment region (Fig.1a).
Additionally, while topsoil samples between mafic and felsic did not show differences in SPR, subsoil samples in the felsic region showed higher SPR than their mafic counterparts and SPR in the mixed sediment region was generally lowest. The 300 Δ 14 C values of both soil and respired CO2 in mafic and felsic regions were not significantly different from each other for either top-or subsoil. Bulk soil samples from the mixed sediment region were consistently depleted compared to their mafic and felsic counterparts.
At valley positions, SPR in topsoil was not significantly different for mafic and felsic samples. Mixed sediment samples were slightly lower in SPR in topsoil than their mafic and felsic counterparts, but not nearly as low as at non-valley positions (Fig.  305 1b). In subsoils, SPR was highest in the felsic and lowest in the mafic region, while the mixed sediment region was not significantly different from the mafic samples (Fig. 1b). As in non-valley counterparts, Δ 14 C activity in valley positions was lowest in samples from the mixed sediment region and differences between the mafic and felsic regions were generally small (Statistical test was not possible due to small sample size, Fig. 2a-b).

Patterns of SOC stock and available nutrient across geochemical region
SOC stocks in the topsoil were similar across all geochemical regions. However, SOC stock significantly decreased with depth for the felsic but not for the mafic or mixed sediment region (Figure 1c). Available phosphorus (Bio-P) was not significantly different between the mafic and felsic regions except in the topsoil where samples from the mafic region show the highest values ( Figure 1d). Bioavailable-P decreased with soil depth for the mafic and felsic regions, but not for the mixed sediment 315 region where it was consistently lower than in soils of the mafic and felsic regions across all sampled depths. top of bars indicate no significant difference following ANOVA tested for differences between geochemical regions and depth intervals. "x" indicates no significant difference between depth intervals within geochemical regions. ANOVA tests were performed separately for non-valley and valley positions. 325

Patterns and differences in Δ 14 C of bulk soils vs. Δ 14 C of respired CO2
Across all study regions we found a strong relationship (R 2 = 0.81, p<0.1) between Δ 14 C of the bulk soil and Δ 14 C of the respired CO2. In non-valley positions, soil C was consistently more depleted than its respired C counterparts, and depth trends in the Δ 14 C of respired CO2 were much less pronounced (Fig. 2a). Notably, the differences in Δ 14 C between soil and CO2 were 330 consistently smaller in the felsic and mafic regions than in the mixed sediment region. In valley positions, differences in Δ 14 C between soils and respired CO2 generally followed the same trends as for non-valley positions, with the exception of the Δ 14 C of respired CO2 in the mixed sediment region being similarly depleted to the soil (Fig. 2a-b). We found a significant contribution of fossil organic C to both SOC and respired CO2 in the mixed sediment region (Table 2).
There, the calculated contribution of fossil organic C to total SOC in bulk soil and respired CO2 increased with soil depth with similar trends for valley and non-valley positions. However, the calculated contribution of fossil organic C to respired CO2 was much higher in valley subsoil (19-39% fossil organic C in respired CO2) than in non-valley subsoil (7-9% fossil organic 345 C in respired CO2). Generally, the contribution of biogenic C to total C was consistently higher (61-97%) in respired CO2 than in SOC of the corresponding bulk soil (48-98%) in both valley and non-valley positions. Microbial respiration discriminated against fossil organic C in non-valley positions by a factor of 3-7 (fFOC bulk soil/respired CO2), but did not discriminate against fossil organic C in valley positions (0.7-1.5 fFOC bulk soil/respired CO2).
350 Table 2. Biogenic and fossil organic carbon contribution in the mixed sediment region to SOC and respired CO2 as % of total C and ratio bulk soil / respired C for both parameters. Values are displayed separately for non-valley and valley positions per soil depth (n = 1 per soil depth and position due to merging of replicates into composites prior to analysis). We note that these values are an upper bound on the contribution of fossil organic C, as these estimates may be affected by variable rates of

Explanatory variables and mechanistic interpretation
For the non-valley subset of our data, rotated principal component analyses yielded 5 significant rotated components (RCs) 360 that together explained 74.5% of the cumulative variance of the dataset (Table B1). From these components, RC1 and RC2 explained about 49% of the entire variance in the dataset, and were loaded with 13 (RC1) and 10 (RC2) independent but highly auto-correlated predictors within each RC. Predictors for RC1 related to soil organic matter characteristics and microbial activity. Predictors for RC2 related to the chemistry of the soil solution. RC3-RC5 explained about 5-11% of the variance aggregation (RC4) and C:N ratio + O horizon C stock (RC5).
Δ-Δ 14 C, in general, was predicted by a wider range of variables than SPR. Topsoil Δ-Δ 14 C was predicted by the RCs "soil 375 solution chemistry", "C:N ratio" and "soil texture". In subsoil, Δ-Δ 14 C was predicted by the RCs "SOM and microbial activity", "soil solution chemistry", "aggregation" as well as "C:N ratio and O horizon C stock". Note that RC4 "aggregation", related to the amount of C associated with microaggregates, played only a minor role as predictor in all data and subsoil predictions of Δ-Δ 14 C. Aggregation did not contribute to the predictive power of topsoil Δ-Δ 14 C and was not included in any model for predicting SPR. 380

Controlling for soil depth (partial correlations)
Partial correlation analysis revealed little to no statistically significant changes in correlation between most RCs and our target variables when comparing zero-order and depth-controlled correlations (Fig. 3). However, a marked and significant reduction in correlation was observed between SPR and RC1 (SOM and microbial activity) as well as between Δ-Δ 14 C and RC1 when 390 controlling for soil depth. A smaller but significant reduction in correlation after introducing soil depth as a control was observed for SPR and RC4 (aggregation). Thus, the reduction in correlation after controlling for soil depth indicates that the relationship of those RC1 and RC4 to target variables is, in part, dependent on soil depth and cannot be interpreted fully independent.

Discussion 405
Fertility and microbial activity Across soil depth, the chemistry of the soil solution (RC2 composed of pH, base saturation, CEC, exchangeable acidity etc.) (Table B1) played an important role in predicting SPR and Δ 14 C in our lab incubation experiment (Fig. 3, Table 3).
Additionally, available nutrients (dissolved N and bioavailable P) reported by Kidinda et al. (2020) for the same soils as investigated, were positively correlated to SPR and Δ 14 C of respired CO2 (Fig. A1) in the mafic and felsic regions. Note that 410 Δ 14 C signatures of the bulk soil and respired CO2 were nearly identical along depth intervals and between the two contrasting (mafic and felsic) parent materials (Fig. 2a). This suggests that the cycling of biogenic C, particularly in the topsoil, can act at a similar rate between soils developed from contrasting parent material (Fig. 3) if soil fertility constraints are satisfied. In contrast, in the mixed sediment region, poor soil fertility is likely one of the main causes of lower rates of C cycling in soil. Soils in this region had the lowest available nutrients, with substantially lower concentrations of bioavailable P (Fig.1d) and 415 NH4 + (data not presented) than soils in the mafic and felsic regions. This adds to existing literature suggesting that nutrient limitation, especially N and P, can significantly inhibit microbial growth and activity, hence lowering soil C turnover rates (Fang et al., 2014;Kunito et al., 2009). In addition, the depletion of N and high C:N values (153.9 ± 68.5) of fossil organic C, which encompasses a substantial part of total C in subsoils of the mixed sediment region (Table 2), was likely an additional factor reducing soil respiration rates (Whitaker et al., 2014). However, respiration rates in the topsoil of the mixed sediment 420 were also lower compared to the mafic or felsic region (Fig. 1), but fossil organic C content in the topsoil was low compared to the subsoil (Table 2). Thus, we conclude that for our investigated tropical forest systems, soil fertility constraints such as the composition of the soil solution (Table B2) are likely more important contributors to explain respiration rates than the presence of fossil organic C content or other C quality constraints.

The role of tropical weathering and mineral related C stabilization mechanisms in explaining soil respiration
In contrast to studies on soils in temperate climate zones (Franzluebbers and Arshad, 1997;Hassink, 1997;Schleuß et al., 2014), in our study aggregation and soil texture played only a secondary role in explaining variability in SPR and Δ 14 C, and their influence decreased with soil depth (Table 3). We explain this observation with the fact that clay minerals at advanced weathering stages such as kaolinite, dominating in tropical soils, generally show lower activity and reactive surfaces than clay 430 minerals dominating earlier weathering stages (e.g. Smectite, Vermiculite) (Doetterl et al., 2018). Lower reactivity of these clays and reduced ability to complex with organic compounds reduce the capacity of the clay fraction to stabilize C in tropical soils compared to temperate soils (Six et al., 2002). In contrast, high amounts of Fe and Al (hydro-) oxides as a result of long term soil weathering have been shown to have a greater influence on C stabilization mechanisms and soil C content (von Fromm et al., 2021;Khomo et al., 2017;Reichenbach et al., 2021). For example, amorphous, oxalate extractable Fe or Al-435 oxides improve the stability of aggregates and can ultimately limit microbial activity (Kirsten et al., 2021;Nagy et al., 2018).
Comparing our findings on SPR to the abundance of oxalate or DCB extractable Fe or Al amorphous and crystalline pedogenic oxides reported by Reichenbach et al. (2021), we found weak to no correlation (Fig. A3 b-c). We interpret this result as an indication that C stabilized by such minerals does not contribute to soil respiration in a significant way in our short-term respiration experiment. Its effects on the long-term SOC stability are more likely related to the formation of stable aggregates 440 (Kleber et al., 2005;Oades, 1988;Barthès et al., 2008;Rasmussen et al., 2018;Traoré et al., 2020;Quesada et al., 2020). Stable metal-organic complexes then represent energetic barriers in soil that are hard to overcome for microorganisms to access potential C resources (Bruun et al., 2010;Zech et al., 1997). The importance of these mechanisms is illustrated by the fact that although mafic soils were generally more fertile than soils in the felsic or mixed sediment region, SPR was lower and decreased more strongly with depth in mafic soils (75% decrease in deep subsoil compared to topsoil) than in felsic soils (33% decrease) 445 (Fig. 1a). We argue that SOC stocks in the mafic region are higher and SPR lower due to the presence of mineral related stabilization mechanisms that are lacking in other regions, consistent with the findings of Reichenbach et al. (2021).
Interestingly, our data suggests that C associated with pyrophosphate extractable oxides (organo-metallic complexes) is readily available to microbial decomposers and can contribute to respiration in a short-term experiment such as ours (Fig. A3a).
In summary, the contrasting relationship of pedogenic oxides of different origin and formation to SPR and Δ 14 C illustrates the 450 need to improve our understanding of metal-organic interactions and their role in C stabilization in tropical soils as our results seemingly confirm (role of metal oxides) and also contradict (role of clay) findings from younger soils in the temperate zone (Khomo et al., 2017). Our results, linked to those of Reichenbach et al. (2021), show that the presence or absence of mineral stabilization mechanisms is particularly important for long-term soil C stocks in tropical soils, varying largely with soil parent material, while short-term respiration relies on readily available C sources. However, given that annual plant C inputs are high 455 in tropical forest systems (Lewis et al., 2009;Sayer et al., 2011), exceeding what deeply weathered soils can stabilize, the soil and environmental conditions under which C can be decomposed or stabilized seem to be more important for short-term respiration.

Accessibility of old C sources to microbial decomposers and its contribution to SOC 460
The presence of fossil organic C in the mixed sediment region (up to 52% of SOC stock in deeper subsoil) (Table 2), had a marked effect on SOC stocks in subsoils that would otherwise be similarly low to those of the felsic region (Fig. 1c). Consistent with this finding, a recent study shows that fossil organic C can largely contribute to SOC in subsoils (Kalks et al., 2021).
While fossil organic C in our study region is of poor quality as indicated by depleted N and high C:N values (153.9 ± 68.5), our data shows that fossil organic C was microbially available (Fig. 2), leading to the respiration of CO2 with comparably old 465 14 C signatures. However, we were unable to quantitatively disentangle the slower biogenic C cycling from the contribution of fossil organic C using 14 CO2. Thus, whether the presence of FOC and/or other unfavorable chemical soil characteristics in the mixed sediment region contributed to a general slowing of C cycling remains unknown.
Nevertheless, under the ideal conditions for microbial activity evoked by our experimental setup, similar to in situ topsoil conditions, microbial organisms can decompose these older, less accessible C sources thereby decreasing the residence time 470 of the fossil organic C (Hemingway et al., 2018). The fact that Δ 14 C signatures in respired CO2 do not mirror the signature of their C sources in soil indicates that microorganisms do continue to discriminate against these older, poorer C sources if alternatives are available (Fig. 2)  .
Being a non-renewable source of organic matter, the fact that fossil organic C can still be found in topsoil is likely related on the one hand to the underlying erosion rates that continuously degrade the mountainous landscapes of the East African Rift 475 System, and on the other hand to the discrimination against fossil organic C by microbial decomposers in the presence of other, more available C sources. While erosion rates at annual or decadal timescales are negligible for the investigated tropical forests (Drake et al., 2019;Wilken et al., 2021), underlying geological erosion rates estimated for tropical mountain forests globally (Morgan, 2005) range between 0.03-0.2 t.ha -1 .y -1 . Assuming an average bulk density in our study area's topsoil of roughly 1.3 g.cm -3 (Doetterl et al., 2021), 6.8-45.3 k years are required to erode the top 10 cm of soil. Thus, slow erosion of soil at millennial 480 timescales may explain the residual content of fossil organic C in topsoil. The loss of soil material as a result of slow processes of landscape denudation do not directly affect the biological processes investigated in our study. However, erosion at geological timescales cannot be ignored as a mechanism for the long-term rejuvenation of soil surfaces (Flores et al., 2020;Montgomery, 2007) in pristine tropical catchments, and, in our study, leading to the exposition of fossil C sources to surface conditions.

Respiration in tropical forests unaffected by lateral fluxes and only by in situ forest hydrological conditions
While we did not observe differences in respiration and Δ 14 C along slope gradients within any of the geochemical regions, we observed significant differences in SPR between valley and non-valley positions ( Fig. 1a-b). The absence of differences along slopes is a strong indicator that lateral fluxes of matter and water do not significantly influence SOC dynamics at timescales 490 relevant to create topography dependent differences in C cycling. This finding is supported by work conducted at the global scale which found erosion in pristine tropical forests was negligible (Vågen and Winowiecki, 2019). Further, at the regional scale in our study region, Drake et al. (2019) found that riverine particulate matter draining from pristine tropical forest catchments are generally dominated by soluble and particulate organic matter fluxes, with little to no mineral sediment being transported. This result is a strong indicator for little to no erosion of mineral soil in pristine catchments, in agreement with 495 our own findings.
In our study, the effect of topography was limited to differences in hydrological conditions between valleys and non-valley positions. At valley positions decomposition of C in subsoil was generally reduced due to the nearly continuous water saturation, limiting the supply of oxygen (Linn and Doran, 1984;Skopp et al., 1990). These conditions are likely present at our study sites as supported by our findings of extensive gleyic features in all studied subsoils in valley positions. However, 500 under the ideal conditions for microbial decomposition of C during our laboratory experiment, decomposition of C from valley subsoil was often higher than at their non-valley counterparts (Fig. 1). We explain this observation with the presence of C sources that, although sometimes have old Δ 14 C signatures (Fig. 2), become readily available to decomposers once environmental constraints, such as water saturation, are removed (Fig. 1b).

Conclusions
Our study shows that geochemical differences in soils that are the result of soil formation from varying parent material (mafic, felsic and mixed sedimentary rocks) continue to influence the microbial activity, SOC stocks and C turnover in tropical soils even after many millennia of weathering and almost complete pedogenic alteration of the parent material. The chemistry of 510 the soil solution, namely soil fertility, and the availability of P and N for microbial decomposers were identified as the most important variables explaining patterns of heterotrophic respiration under idealized well-aerated topsoil conditions. C stabilization mechanisms including the presence or absence of pedogenic oxides between our geochemical regions were identified as indirect controls to explain variation in soil respiration through their effect on soil aggregation and as potential energetic barriers that decomposers are forced to overcome. Patterns of Δ 14 C with soil depth were largely driven by the 515 presence or absence of fossil organic carbon of low quality, inherited from parent material. Under idealized well-aerated topsoil conditions, these fossil C sources became available to microbial decomposers, especially in the absence of better alternative energy and nutrient sources.
Furthermore, our analyses revealed that soil respiration was driven in parallel by contrasting processes, limiting microbial activity and slowing down C cycling. C in soil of the studied mixed sediment region was low in quality, resulting in low 520 specific respiration, slower C cycling and high SOC stocks. C in soil of the mafic region was lower in accessibility due to its stabilization with minerals, also resulting in low specific respiration and high SOC stocks in subsoil. Thus, while geochemistry differed drastically between soils in those two systems, particularly in subsoils, both show low specific respiration for entirely different reasons. In contrast, soils in the felsic region showed high specific soil respiration, as no strong mineral driven stabilization mechanisms were present and as soil C of favorable quality was readily available for microorganisms to 525 decompose.
While the impact of geochemistry on C dynamics was clearly distinct between the studied soils, topography only played a secondary role in these densely vegetated tropical forest systems. Hydrological features such as water saturation in valleys partially inhibited microbial activity in situ, leaving labile C sources available for decomposition under the idealized laboratory conditions of our experiment. Erosional processes rejuvenating soils and landscapes at geological timescales did account for 530 significant differences in C cycling across geochemical regions due to the surfacing of fossil organic carbon, but did not act at timescales to create topography dependent differences in C cycling. We conclude from our findings that geochemistry, parent material and its lasting role on pedogenesis are key factors to consider to improve our understanding of C release from tropical forest soils. Improving the spatial representation of C dynamics at larger spatial scales using the variables and controls identified in this study could potentially be an important improvement for predicting and modeling future C turnover in tropical 535 forest soils.

Appendices 540
Appendix A- Figures   Figure A1. Pearson correlation between composite of corresponding replicates of Δ 14 C of respired CO2 and SPR to P (panels, 545 a-b), and N (panels, c-d) available nutrient data reported by Kidinda et al.,(2020) Table B2. Overview of soil properties and fertility indicators for the three geochemical regions and depth intervals.
Abbreviations: Baseexc= sum of exchangeable bases, CEC = potential cation exchange capacity, ECEC = effective cation 585 exchange capacity; pHKCl =, bio-P = bioavailable phosphorus (Bray P method). Values reported are averages plus standard deviations (n = 85). Data reported by Doetterl et al. (2021a,b Where sample material is left, soil samples can be made available upon request through the corresponding authors (S. Doetterl).

Acknowledgements 600
This study was financed within DFG Emmy Noether group through "Tropical soil organic carbon dynamics along erosional Science and Site Ecology at Technical University Dresden, Sustainable Agroecosystems Group and the Soil Resources Group ETH Zurich and the Faculty of Agriculture at the Catholic University of Bukavu, Institut Congolais pour la Conservation de la Nature (ICCN), Rwanda Development Board (RDB) and Uganda Wildlife Authority (UWA). The authors would also like to thank the whole TROPSOC team especially the student assistants for their important work in the laboratory and all field helpers who made the sampling campaign possible. A personal thanks to Sophie von Fromm for valuable feedback on the 610 manuscript. Special thanks also to the editors of SOIL and the two reviewers who have guided the review process of this manuscript and gave many valuable comments that improved the manuscript.

Competing interest
SD is a liaison editor of the special issue Tropical biogeochemistry of soils in the Congo Basin and the African Great Lakes 620 region and PF is a topical editor of the SOIL journal. However, none of them were involved in the review process of this manuscript. All other authors declare that they have no conflict of interest.

Special issue statement
This article is part of the SOIL Special Issue: Tropical biogeochemistry of soils in the Congo Basin and the African Great 625 Lakes region.