Introduction

Forests are natural resources with important economic, cultural, and ecological values1. Many of the ecosystem processes that occur in forests are mediated by soil microbes, key among them is nutrient cycling, in which plant-derived organic matter (OM) is degraded and nutrients are released for consumption by plants and other organisms. Globally, forests are under pressure by changes in land use, global climate, and biomass harvesting2,3. Management practices that affect the function of the soil microbial community can potentially influence the growth, survival and economic value of trees in managed forest stands.

Established in the late 1980s, the Long-Term Soil Productivity Study (LTSP) focuses on the long-term effects of OM removal during harvesting on forest productivity4. Ten years after harvesting and replanting, varying levels of OM removal had only marginal effects on tree productivity5,6,7. On the other hand, harvesting has substantial short- and long-term effects on soil microbial communities. A meta-analysis of forest perturbation studies found that harvesting reduces microbial biomass by 19% on average8, with stronger effects on fungal than on bacterial populations (27% vs. 14% reductions, respectively). Molecular ecology studies of LTSP sites in multiple ecozones showed that harvesting caused long-term changes in the overall soil microbial community structure9,10, and in hemicellulolytic populations11. We recently showed that harvesting reduced biomass degradation potential for more than a decade after harvesting at one LTSP site in the interior Douglas fir ecozone of British Columbia12. However, it remains unclear whether this effect on biomass degradation potential is generalizable across different ecozones. We hypothesized that OM removal during harvesting would consistently decrease the genetic potential for organic carbon decomposition, reducing the abundance and diversity of genes encoding this process.

The carbon and nitrogen cycles are closely linked. In forest soils, organic decomposition is accompanied by mineralization of organic nitrogen, and growth of plant biomass creates high demand for available nitrogen. Nitrogen cycling is critical in temperate coniferous forests where primary productivity is typically N-limited13. Previous studies of American and European forests6,14,15 have shown that harvesting can increase rates of nitrogen mineralization and nitrification. We hypothesized that OM removal during harvesting would increase the relative importance of these catabolic processes in soil communities, resulting in an increase abundance and diversity of genes involved in nitrogen cycling.

We tested the above hypotheses using the LTSP experimental design on sites in five ecozones across North America. Metagenomics allowed us to identify and quantify the microbial community genes and characterize the diversity and abundance of those related to the functions of interest for this study. Our analyses focused on genes encoding carbohydrate active enzymes (CAZy) as well as nitrogen cycling enzymes. We compared the diversity and relative abundance profiles (genetic potential) of biomass decomposition and nitrogen cycle genes among ecozones, soil layers, and OM removal treatments. We related environmental gradients to the metagenomic responses to OM removal during harvesting. This is the most extensive metagenomic analysis of forest soil communities to date and among the first to use a field experiment replicated among different ecozones.

Methods

Study sites

Soil samples were collected from five sites that are part of the Long-term Soil Productivity (LTSP) Study in North America4,16. The sites, which varied in weather, vegetation, and soil chemistry, represent five distinct ecozones across North America named for important commercial tree species in each: BS, black spruce; IDF, interior Douglas-fir; JP, jack pine; LP, loblolly pine; and PP, ponderosa pine (Supplementary Table S1). At each site, we sampled harvested treatment plots with three levels of OM removal (and strictly minimized soil compaction): stem-only harvesting (OM1), whole-tree harvesting (OM2) and whole-tree harvesting plus forest floor removal (OM3). We additionally sampled unharvested reference plots (OM0).

Sampling

Sampling occurred 10 to 17 years after the OM removal treatments and replanting (Supplementary Table S1), as previously described9,12. Both organic and mineral soil layers were sampled; however, except for the LP ecozone, the organic layer had not yet re-developed in the OM3 treatments. The same samples were used for chemical analyses and DNA extraction, except for the IDF ecozone where samples for chemical analysis were taken the year prior to sampling for DNA extraction, as previously reported12.

Soil chemistry

Chemical analyses (Supplementary Table S2) were done using standard methods, as previously described9. Briefly, total C and N were determined with a combustion elemental analyzer. Water content was determined gravimetrically. pH was measured in a 2:1 slurry (5 mL water : 2.5 g dry soil). For the LP and PP ecozones, measurements of total C and N were only available per plot (not per sample) and only for the mineral layer.

DNA extraction and metagenome generation

Soil samples were processed for shotgun metagenome sequencing as previously reported12. That study of the IDF ecozone generated paired-end 75-bp metagenome reads, which were also used in the present study. For the remaining four ecozones, paired-end 150-bp metagenomics reads were generated at the Joint Genome Institute, Walnut Creek, CA, USA. Sequences were quality-filtered as previously reported12.

Carbohydrate degradation and nitrogen cycle gene databases

To identify genes potentially involved in carbon and nitrogen cycling (genetic potential), we compared our high-quality short reads against the carbohydrate active enzymes (CAZy) database17 and a custom database for nitrogen cycle enzymes. A database was created from the CAZy website (as of May 6, 2014) by downloading the corresponding proteins from Genbank using custom scripts. The CAZy database includes four conventional enzymatic classes: Glycoside hydrolase (GH), glycosyl transferase (GT), polysaccharide lyase (PL), and carbohydrate esterases (CE), as wells as the recently added auxiliary activities (AA) class. CAZy proteins are further classified into families, based on manual curation and phylogenetic analysis (CAZy similarity thresholds are not public).

The nitrogen cycle enzyme database included seven key enzymes of the nitrogen cycle (Fig. 1) representing nitrogen fixation (nifH), ammonia oxidation (bacterial and archaeal amoA), denitrification (nirS, nirK, norB, and nosZ), and dissimilatory nitrate reduction to ammonia (nrfA). We used a previously published database for the nifH gene18 and created custom databases for all the other nitrogen cycle genes. Protein sequences for amoA were retrieved from the FunGene database which uses Hidden Markov protein models19. On March 25, 2015, sequences were obtained and filtered according to their model fit. For filtering we used proteins with model coverage >75% and scores greater than 432.5 and 352 (for archaeal and bacterial versions, respectively). These minimum threshold scores were calculated using the score ratio between the lowest and highest sequences of the training set (i.e., the FunGene sequences) of archaeal amoA set. Custom databases for nirS, nirK, nosZ, norB, and nrfA were created by searching for proteins with related annotation in the NCBI’s Genbank and visually inspecting the alignments for the conservation of functional domains and motifs. The nitrogen cycle gene database was dereplicated by creating a group of sequences with a 95% identity threshold using cd-hit v4.620. We classified nitrogen cycle protein into enzyme families using a ≥95% identity threshold. Thus, families of CAZy and nitrogen cycle enzymes are differently defined and not completely comparable.

Figure 1
figure 1

Model of nitrogen cycle in forest soils. DNRA: dissimilatory nitrate reduction to ammonia. Nitrite (NO2−) appears twice since it is an intermediate in denitrification, nitrification, and DNRA. The four reactions in denitrification are shown in grey arrows. Significant differences in abundance for genes targeted in our database are shown in blue (Anova p < 0.05, controlling for ecozone differences). Abundances for the IDF ecozone were not included in the calculation due to differences in the sequencing method for that ecozone.

Full size image

Both forward and reverse reads as well as joined paired ends were compared against the databases using DIAMOND v0.5.342, highly correlated with bacterial abundances in Canadian agroforestry systems43, a driver of bacterial diversity under twelve North American forests44, and a driver of nitrogen cycle gene abundance over 107 sites across the Burgundy French region45.

Additional drivers of variation among ecozones in our study, which we did not measure, were likely the aboveground vegetation and soil nitrogen composition. Vegetation composition has been shown to be as important as soil chemistry in determining bacterial and fungal composition in forest soil43,46, because it influences the quantity and variety of carbon sources (as plant biomass) that enter the soil as carbon sources as well as the potential plant-microbe interactions. However, since the LTSP experiments in each ecozone had a different dominant tree species and associated vegetation, we cannot separate the ecozone and the vegetation effects. Different forms of nitrogen, such as nitrate and ammonia, as substrates for particular nitrogen-cycling populations, were found to be correlated with abundances of the corresponding genes47. These populations, in turn, modulate the fate of the nitrogen in the soil.

Forest soil is a highly stratified environment with particularly strong biochemical and biological differences between mineral and organic layers. The organic layer has higher carbon and nitrogen concentrations, higher microbial abundances, and higher denitrification and nitrogen fixation rates48. Accordingly, we observed large differences in soil chemistry and microbial biomass between layers (Fig. 4) as well as a strong influence of the soil layer on the distributions of decomposition and nitrogen cycle genes (Figs 2 and 3).

Despite our results that showed inconsistent effects of the OM removal treatments, we found a small set of gene families that were conserved in their distribution and seemed unaffected by OM removal. These core genes likely encode key functions in the degradation of plant biomass. These core genes were mainly bacterial; however, we did not sample the litter layer where fungal groups are reportedly more active35. The involvement of CAZymes in plant biomass decomposition is consistent with their skewed distribution toward the organic layer, and the predominance of plant cell wall degradation genes as soil layer predictors across ecozones. Additionally, CAZy genes showed more frequent and stronger correlations with C and the C/N ratio in the organic layer. These results are consistent with a meta-analysis of forest harvesting studies that showed that carbon storage in the organic layer was more sensitive to harvest impacts than in the mineral layer31. As the role of CAZy genes in decomposition would suggest, we found strong positive associations in the organic layer between CAZy GH and GT families and total carbon and the C/N ratio (Fig. 5, Supplementary Fig. S5). However, some AA families had negative associations with those two variables, which seems counterintuitive.

The distribution of nitrogen cycle genes suggests that they contribute proportionally more to catabolism in the mineral versus organic layer. However, this does not imply that more nitrogen cycling occurs in the mineral layer since it has lower biomass and lower total N than the organic layer. The C/N ratios imply that the organic layer is nitrogen-limited while the mineral layer is carbon-limited, likely making microbial catabolism of nitrogen compounds more favorable relative to heterotrophy in the mineral layer. This hypothesis is supported by the enrichment of nitrogen cycle genes in the mineral layer, both in terms of relative abundance and diversity (number of gene families) (Supplementary Fig. S1). The strongest case for specialization was from archaeal ammonia oxidizers, as the archaeal amoA gene was exclusively present or much more abundant in the mineral versus organic layer. Archaeal ammonia oxidizers are well adapted to low nitrogen concentrations, low pH, and low O249 and can outcompete their bacterial counterpart in deep soil layers. However, total bacterial amoA abundances were higher than archaeal ones in 94% of the cases.

In addition, we found that two bacterial amoA families were predictors of mineral soils across all ecozones. These families were affiliated with Nitrosospira (Betaproteobacteria) and Nitrosococcus (Gammaproteobacteria), which are frequently found in low-pH forest soils, where they may benefit from urease activity50. Layer specialization was also found for nitrogen fixers, as five nifH families, four from Alphaproteobacteria and one from Cyanobacteria were consistently associated with the organic layer and family 320, the most abundant nifH family from a Bradyrizobiaceae member was consistently associated with the mineral layer. Nitrogen fixation is a key link between the nitrogen and carbon cycles, as fixed N may prime degradation of high-C/N litters51. These results suggest that populations carrying nitrogen cycle genes depend not only on nutrient availability (which is higher at the organic layer) but also on other drivers such as pH and C/N ratio.

Harvesting had long-term effects on the capacity of the forest soil community for carbon and nitrogen cycling, which were consistent with our initial hypotheses. These impacts were distinct in the organic and mineral soil layers and were highly variable among different ecozones in a manner dependent on key environmental variables. These alterations of the soil community were not associated with major effects on tree productivity. However, these alterations might presage later effects on productivity as the trees grow and have greater nutrient demands or effects on the resilience of the forest system to future perturbations. Our results suggest a mechanism by which harvesting can exacerbate nitrogen losses at sites predisposed to such losses, potentially lowering plant productivity and increasing greenhouse gas emissions.