Soil and fungal samples were collected from the SPRUCE experimental site at the 8 ha S1 bog in the USDA Forest Service Marcell Experimental Forest in northern Minnesota, USA (47∘30.476′ N, 93∘27.162′ W). The bog is dominated by the trees Picea mariana (Mill.) Britton, Sterns and Poggenb. and Larix laricina (Du Roi) K. Koch; ericaceous shrubs (Ledum groenlandicum Oeder; Chamaedaphne calyculata (L.) Moench.); and Sphagnum mosses. Various forbs and sedges are also present. The bog topography can be separated into hummocks (protruding above the average water table) and depressed hollows, and divided into areas with trees (Picea or Larix) or without trees. Average annual air temperature from 1961 to 2005 was 3.3 ∘C with yearly mean extremes of -38 and 30 ∘C and average annual precipitation of 768 mm (Sebestyen et al., 2011). Average pH of the peat is 4.1 and average gravimetric water content is 7.40 g H2O g-1 dry peat (Iversen et al., 2014). The water table fluctuates about 30 cm annually and water table depth can reach 1.4 m over the course of a 50-year record (Sebestyen et al., 2011). The bog maintains a perched water table with a hydraulic gradient about 1 m above the surrounding aquifer, which is composed of outwash sand (Verry et al., 2011). Because of the high water table, root distribution is quite shallow compared to upland sites (Iversen et al., 2017). In total, 93 % of fine roots were in the 0–10 cm depth, 6 % at 10–20 cm depth, and only 1 % at 20–30 cm depth (A. Malhotra, personal communication, 2017). The boundary between the acrotelm and catotelm at SPRUCE is around 30–40 cm depth.

Peat cores for this analysis were collected in mid-August of 2012 from locations along three boardwalks extending out into the bog beyond the lagg region (Fig. 2). Surface peat (∼ 0–30 cm) was collected using a modified hole saw, while deeper samples down to mineral soil (∼ 30–250 cm) were collected using a Russian peat corer. Cores were taken in both hummocks and hollows, with 0 cm defined as the surface of hollows and hummock heights above that assigned positive depths. Cores were bulked and homogenized every 10 cm over the 0 to -100 cm depth, every 25 cm from -100 to -200 cm, and over the entire 50 cm increment from -200 to -250 cm (in some cases, -300 cm was reached before mineral soil was observed). Cores were sampled at 17 locations (Fig. 2; the locations of experimental plots distributed across the three boardwalks) and material from 16 of these locations was used for the δ13C, δ15N, and radiocarbon measurements reported here. At locations 4, 5, 6, and 7 along the southern boardwalk, separate cores were taken within 150 cm of Picea or Larix trees and in the open (no trees within 150 cm), and the distinction designated as “treed” or “untreed”.

Peat cores were analyzed for Δ14C, δ13C, δ15N, %C, %N, and C / N by depth increment, with the depth increment recorded as the average depth (for example, 0 to 10 cm in a hummock given as 5 cm). Peat cores were analyzed in hummocks to a depth of -10 cm and in hollows to the bottom of the core (between -200 and -300 cm). Live woody plant foliage and fine roots to -10 cm were collected in August 2012 and live Sphagnum in 2013.

To collect fungal hyphae, in-growth cores were constructed. Mesh (40 µm) in-growth bags (10 cm × 10 cm) were filled with sterile sand. Bags were incubated in the field in paired hummock and hollows at six locations in the bog. In hummocks, bags were inserted at +10 to 0 cm above the adjacent hollow and in both locations from 0 to -10 cm and -10 to -20 cm below the hollow surface. Bags were installed on 5 June 2013 and recovered on 20 September 2013. Sand from in-growth bags was combined with ultrapure water and mixed at 80 rpm for 20 min. Suspended hyphae were removed with tweezers and the process was repeated until all hyphae were collected. Hyphal biomass was dried in the oven at 60 ∘C for 48 h. Of 30 in-growth samples, 20 generated enough hyphal mass for analysis. All 20 samples were treated as independent replicates in statistical analyses.

Radiocarbon content of homogenized bulk peat was measured on the Van de Graaff FN accelerator mass spectrometer (AMS) at the Center for AMS at Lawrence Livermore National Laboratory. Peat samples were not chemically pretreated prior to 14C measurement. Samples were prepared by sealed-tube combustion to CO2 in the presence of CuO and Ag and then reduced onto Fe powder in the presence of H2 (Vogel et al., 1984). Radiocarbon isotopic values had an average AMS precision of 2.6 ‰ and were corrected for mass-dependent fractionation with δ13C values from analyses conducted at the Department of Geological Sciences Stable Isotope Laboratory at the University of California, Davis, using a GVI Optima stable isotope ratio mass spectrometer. Radiocarbon values are reported here in Δ14C notation corrected for 14C decay since 1950 (Stuiver and Polach, 1977). Calibrated ages were determined using Calib (http://calib.qub.ac.uk/calib/) or CaliBomb (Reimer et al., 2004) with INTCAL13 (Reimer et al., 2013) and Northern Hemisphere Zone 1 bomb curve extension (Hua et al., 2013) atmospheric 14C calibration curves. Years before present (BP) refer to years prior to 1950. For more recent samples, calendar years AD may also be used.

These same soil samples and additional samples of hyphae and foliage were analyzed for %C, %N, δ13C, and δ15N at the University of New Hampshire Stable Isotope Laboratory using a Costech 4010 elemental analyzer coupled to a Thermo Delta Plus XP IRMS. Standard deviations of laboratory standards (tuna, NIST 1515, and NIST 1575a) for δ15N and δ13C averaged less than 0.2 ‰. Fine roots of the woody vascular plants were analyzed for their stable isotopic composition at the Oak Ridge National Laboratory on an Integra CN mass spectrometer (SerCon, Crewe, UK), using standards traceable to NIST 8547-ammonium sulfate or 8542-sucrose (NIST, Gaithersburg, Maryland, USA).

The statistical program JMP (SAS Institute, Middleton, Massachusetts, USA) was used for statistical analyses. Reported values are ± SE, unless otherwise specified. Regression models for soil δ13C and δ15N were tested. Factors included in the regression model included nominal variables of vegetation type (treed or non-treed), topography (hollow or hummock), and plot number. Depths with only a single measurement were generally excluded, unless they were very similar in depth to another value. Continuous variables included %N, %C, and isotopic values.

To test whether plot location, proximity to trees, depth, topography, and elemental concentrations influenced the carbon and N isotope patterns in peat profiles, we used multiple regression analyses. Sample %C, %N, C / N, and either δ13C or δ15N were included as continuous variables. Because the effects of depth or plot location on N and C dynamics are unlikely to change continuously (for example, methanogenesis requires an anaerobic soil and plots at bog edges may have different hydrology and N dynamics than plots in the middle of the bog; Urban and Eisenreich, 1988), plot was treated as a nominal (categorical) variable in our regression models. To avoid over-parameterizing the model, depth was treated as a continuous variable with a cubic transformation (that is, the regression model included model depth, the square of model depth, and the cube of model depth as additional parameters). Stepwise regression was used and model selection based on the lowest Akaike information criterion with a correction for sample size (AICc). Within a given depth, values for radiocarbon were tested for correlations against δ13C and δ15N and the slope of the regression estimated.

Of the six vascular plant taxa tested, δ13C of foliage varied from -30 ‰ in Larix to -28 ‰ in Picea. The δ15N of plant foliage varied more widely than δ13C, from -8.5 ‰ for Picea to 2.5 ‰ for Eriophorum. Fine root δ13C averaged -27.4 ± 0.3 ‰ for Larix, -26.9 ± 0.1 ‰ for Picea, and -28.5 ± 0.2 ‰ for shrubs. Fine root δ15N averaged -4.7 ± 0.4 ‰ for Larix, -4.1 ± 0.3 ‰ for Picea, and -1.8 ± 0.1 ‰ for shrubs, whereas coarse roots of shrubs averaged -3.4 ± 0.3 ‰. If we assume plant productivity patterns are similar aboveground and belowground, then the productivity-weighted average in vascular plants for δ13C was -29.2 ‰ for foliage and -27.3 ‰ for roots (Table 2). Fungal hyphae from in-growth cores (n=20) averaged -26.0 ± 0.2 ‰ (se) for δ13C and -0.3 ± 0.2 ‰ for δ15N.

Carbon isotope (δ13C) values of peat in the profile increased from -29 ‰ in the top 10 cm of hummocks and hollows to -26 ‰ at -112 cm and then decreased slightly at greater depths. δ13C values changed most rapidly from 0 cm to -50 cm depth (Fig. 3a). Nitrogen isotope values in the peat profile increased from -3‰ in hummocks above the water level to around 1 ‰ at -50 cm. δ15N then decreased to 0 ‰ at -85 cm before increasing again to 1.5 ‰ at -200 cm. Similar to δ13C, δ15N changed most rapidly from 0 cm to -50 cm depth (Fig. 3b).

In a stepwise regression model for δ13C including δ15N, C / N, %N, %C, and depth as continuous variables and the vegetation type, topography, and plot sampling location as nominal variables, all factors were retained in our final model, including four terms for partitioning the 16 plots. The model explained 85 % of the total variance (n=238, adjusted r2) in peat δ13C. The three depth terms (depth, depth squared, and depth cubed) explained 45 %, %N explained 10 %, C / N 11 %, δ15N 15 %, %C explained 2 %, the four plot terms explained 7 %, and proximity to trees explained 3 % of the variance (Table 3). Topography explained 10 % of variance, with hollows lower than hummocks by 0.89 ‰ in δ13C (p<0.001). Results of the stepwise model testing δ13C and δ15N are given in Table S1.

The stepwise regression model for δ15N of peat included δ13C and the same variables as in the δ13C regression model. This model explained 66 % of the total variance (adjusted r2), with proximity to trees and topography dropping out of the regression model. Of the explained variance, the three depth terms accounted for 8 %, %N 22 %, %C 7 %, C / N 17 %, δ13C 34 %, and the three plot terms 11 % (Table 3).

The influence of depth and location in our regression models on δ13C and δ15N values is plotted in Figs. 4 and 5, respectively. The influence of depth was negative above 0 cm (corresponding to the surface of the hollows), increased regularly from -5 to -25 cm, and then varied little in δ13C in the deepest horizons while still increasing in δ15N. There was some spatial patterning of values across plots, with two plots (4 and 19) near to the western upland high in δ15N (Figs. 2 and 5).

Although overall patterns of radiocarbon with depth were clear, radiocarbon varied widely at any given depth, and correlated significantly with δ13C or δ15N at several depths (Table 4). Radiocarbon correlated positively with δ13C at -162, -5, 5, and 15 cm, and correlated negatively with δ13C at -35 and -65 cm. In contrast, radiocarbon correlated positively with δ15N at -65, -55, -45, -5, and 15 cm, and correlated negatively with δ15N at -162 and -25 cm. Overall patterns of Δ14C with δ13C or δ15N are shown in Fig. 6a and b, respectively.

Isotopic ratios within the profile can shift if elemental fluxes in or out of the system differ isotopically from profile material (Fig. 1). Loss of labile C via respiration, methanogenesis, or leaching (Kolka et al., 1999) could alter the δ13C of the residual material, as could inputs of 13C-enriched material such as roots or mycorrhizal hyphae. Similarly, changes in the δ13C of atmospheric CO2 can alter the δ13C of photosynthetically fixed C, whereas changes in moisture, temperature, or photosynthetic capacity can alter the 13C discrimination between atmospheric CO2 and fixed C. For N, loss of 15N-depleted material from the bulk peat via mycorrhizal transfer to fine roots, direct root uptake, denitrification, or leaching of organic or inorganic N could raise the δ15N of the remaining soil organic matter. Inputs of N via atmospheric deposition, N fixation, or transport from surrounding uplands could also influence δ15N if these inputs differ isotopically from peat profile values. These processes can be linked to past climate and vegetation with profile radiocarbon measurements that are calibrated to calendar years. Here, we used radiocarbon to indicate the potential timing of shifts in some of the primary drivers that influenced C and N stable isotope patterns within the peat profiles at SPRUCE, such as the Suess effect and the transition to a forested bog, but also examined processes that were not tied to a particular time period, such as differences in C and N stoichiometry or proximity to uplands.

%N, %C, and C / N contributed a combined 47 and 23 % of explained variance, respectively, to our regression models of δ15N and δ13C. These factors reflected the biochemical and isotopic composition of the original plant material, but also reflected how the chemical structure and isotopic composition of plant material has altered during its slow decomposition at S1.

Microbially driven C loss raises soil organic matter %N, lowers C / N, and enriches soil organic matter in 13C (resulting from loss of 13C-depleted CO2) (Ehleringer et al., 2000; Alewell et al., 2011). The positive correlation of %N with δ13C may therefore reflect an underlying correlation between the accumulation of 13C-enriched microbial necromass (Wallander et al., 2004) and the increased N content of the peatland organic matter. Fungal %N and δ13C are positively correlated (Hobbie et al., 2012) because of the high δ13C of microbially synthesized protein relative to other microbial components such as carbohydrates and lipids. In contrast, the positive correlation of C / N with δ13C and negative correlation with δ15N presumably reflect a legacy of buried wood, which, relative to other plant material, should be high in δ13C (Trudell et al., 2004) and high in C / N. This can be seen clearly in the few samples with C / N greater than 70, which is higher than any plant tissue measured in this study. Although some Sphagnum taxa under pristine conditions can be very low in %N (0.22 %) and δ15N (-3.6 ‰) and high in C / N (∼ 190) (Asada et al., 2005a), here, the presence of wood was noted seven times during laboratory examination of the 238 samples at depths from 15 to -35 cm, with those samples twice as high in C / N (average, 69) as other samples and also significantly higher in C / N in multiple regression analysis (see Table S2).

The negative correlation of %N with δ15N indicated that either added or removed N is low in δ15N. One possibility for removal is that 15N-depleted N has been transferred from mycorrhizal fungi to plants. Alternatively, N could be added via fixation with a δ15N value of -1 ‰, which would lower overall δ15N values deeper in the peat profile. However, we point out that only at -45 and -55 cm are %N and δ15N significantly and negatively correlated (Table S3). At these depths, %N is about 1.8 % and δ15N is about 1 ‰ (Fig. 3b, d). The value of the coefficient for %N in the δ15N regression, -2.96 ‰ %N-1, implied that the perturbing N has a δ15N value that is 1.8 % × 2.96 ‰ %N-1 less than that of 1 ‰, or -4.3 ‰, which is too low to be fixed N. The apparent 15N depletion of 5.3 ‰ against the source N is a plausible value for 15N discrimination between mycorrhizal fungi and host plants (Hobbie and Colpaert, 2003).

Correlations of carbon concentration with stable isotopes provide information about chemical composition or degree of processing. The negative correlation of %C with δ13C is expected based on the chemical composition of 13C-depleted compound classes of lignin, aromatics, and lipids, which are high in %C (Poorter et al., 1997; Hobbie et al., 2002). Initial decomposition of Sphagnum commonly decreases δ13C, as 13C-enriched soluble components are leached (Asada et al., 2005b). In contrast, aromatics and lipids do not generally contain N, so the positive correlation of %C with δ15N in bulk peat cannot be explained in the same manner. However, microbial processing generally enriches soils in 15N (Billings and Richter 2006; Templer et al., 2007) while increasing %C, which was also true at S1 (Tfaily et al., 2014).

Plot-specific coefficients for δ13C and δ15N may reflect site-specific differences in the dominance of conditions favoring 13C- or 15N-depleted losses during peatland development. The positive coefficients for δ15N are from two plots, 4 and 19, closest to the lagg region adjacent to the western upland. This suggests that the different hydrology in the lagg has enhanced 15N fractionation from N removal mechanisms such as denitrification, nitrification, or leaching of DON. Alternatively, dissolved N transported from the uplands during spring thaw and melt may have provided an additional 15N-enriched N source for plots 4 and 19 located near the bog edges (Fig. 5). Peatland DON appears enriched in 15N and 13C relative to bulk peat (Broder et al., 2012), and this is presumably true for upland sources as well. In the adjacent Marcell S2 kettle bog, large N fluxes from upland locations from both surface runoff and interflow led to much larger N losses in streamflow from the lagg region (∼ 32 kg ha-1 yr-1) than from the bog itself (2 kg ha-1 yr-1) (Urban and Eisenreich, 1988). The uplands here are dominated by ectomycorrhizal trees such as Populus, Quercus, and Pinus, which produce vertically stratified soil profiles with high δ15N values in lower organic and mineral horizons (Hobbie and Ouimette, 2009). We therefore expect DON produced in uplands to be high in δ15N, which will serve as a source of 15N-enriched N to lagg regions of peatlands.

Peatland succession and climate have been established previously at the nearby S2 bog and are summarized in Table 1. As the same climatological factors affected the S2 and S1 (SPRUCE) bogs, plant stratigraphy and isotopic patterns were probably similar, although accumulation rates are lower at S1 than at S2 (McFarlane, unpublished data).

In the following paragraphs, we link shifts in δ15N through the profile to radiocarbon ages and the corresponding patterns in vegetation and climate at the S2 bog. The high δ15N at -213 and -225 cm corresponded approximately to a warm period between 8000 and 9200 years BP, during a sedge-rich fen stage (Verry and Janssens, 2011), with mean annual temperatures of 4–5 ∘C. Given similar C / N ratios during this period to subsequent periods (∼ 20), nitrogen losses were probably more depleted in 15N than subsequent losses, that is, more losses via nitrate leaching or denitrification rather than via DON leaching. Nitrification and denitrification are higher in fens than in bogs and should change δ15N patterns along the core profile as the core reflects peatland succession and climatological changes (Regina et al., 1996; Bayley et al., 2005; Wray and Bayley, 2007).

The S2 bog shifted from a rich fen to a transitional fen by 5600 years BP, while mean annual temperature (MAT) decreased to perhaps 2 ∘C. At the S1 bog, the negative correlation between radiocarbon and δ15N at -162 cm (∼ 6000 years BP) suggested decreasing losses of 15N-depleted N during this transition. The depth increment from -85 to -112 cm corresponded to a transitional fen stage 3300–4800 years BP. The accompanying trough in δ15N at -400‰ Δ14C (4220 calibrated years BP, Fig. 6b) suggested decreased sequestration of 15N-enriched organic matter as nutrient availability declined during this transition.

The positive correlations between δ15N and Δ14C at -45, -55, and -65 cm (Table 4 and Fig. 6b) are linked to a parallel decline in %N over these time periods, so that younger samples are lower in %N and higher in δ15N than older samples, reflecting losses of 15N-depleted N from the younger samples. δ15N values peaked at -35 to -55 cm, and then declined above and below those depths. This pattern suggested that those depths were affected by a 15N enrichment process that did not affect deeper depths (that is, further in the past). The peak δ15N accordingly reflected 15N partitioning between surficial and deeper horizons by a new mechanism.

We suggest that ectomycorrhizal fungi are the most probable cause of this unusual peak in δ15N at intermediate depths in the peat bog. At the nearby S2 bog, poor fen transitioned to forested bog between AD 1610 and 1864, with a charcoal layer at S2 indicating that peat was consumed by fire during this period, precluding a more specific date for this transition (Verry and Janssens, 2011). This forested bog is dominated by ectomycorrhizal Picea and Larix. Transfer to plants of 15N-depleted N by ectomycorrhizal fungi leads to low plant δ15N and high 15N enrichment of the residual N not transferred (Hobbie and Hobbie, 2008; Hobbie and Högberg, 2012). Here, the apparent influence of ectomycorrhizal uptake can be traced to ∼ -55 cm.

The relatively low contribution of ectomycorrhizal and ericoid mycorrhizal fungi to total plant uptake here compared to forest sites dominated by ectomycorrhizal fungi accounted for the lower 15N enrichment between surficial layers and deeper layers here (∼ 3 ‰) than in forest soils, which averaged a 9.6 ‰ enrichment between surficial and deeper soils for ectomycorrhizal forests (reviewed in Hobbie and Ouimette, 2009). Ectomycorrhizal and ericoid mycorrhizal trees and shrubs contributed relatively little to total N uptake at the nearby Marcell S2 bog, where mosses contributed 75 % of total plant uptake, herbaceous plants 13 %, and mycorrhizal trees and shrubs 13 % (Urban and Eisenreich, 1988). A few studies in peatlands have reported δ15N values through profiles, but those values have peaked at -1 to 0 ‰ and have been in systems without ectomycorrhizal trees present (Krüger et al., 2015; Esmeijer-Liu et al., 2011).

Once shifts in δ13C caused by compositional shifts are removed, the remaining patterns can be explained by invoking climatic or vegetation shifts. The relatively low δ13C at -213 and -225 cm corresponded to a warm period between 8000 and 9200 years BP, during a sedge-dominated rich fen stage (Verry and Janssens, 2011), with mean annual temperatures of 4–5 ∘C. Higher temperatures in peatlands are associated with lower δ13C values (Skrzypek et al., 2005, 2008).

Radiocarbon and δ13C correlated positively at -162 cm, corresponding to a 1 ‰ rise in δ13C which accompanied a drop in MAT to perhaps 2 ∘C by 6000 years BP. This cooling trend was also accompanied by a slight rise in precipitation, so the decreased 13C discrimination could also be attributed to increased Sphagnum moisture (Rice and Giles, 1996), although it is difficult to distinguish between these two possible causes of 13C differences (Ménot and Burns, 2001). Sphagnum discrimination is less with increased moisture because CO2 diffusion is limited under wet conditions. The stratigraphy at the S2 bog indicated a vegetational shift from a rich fen to a transitional fen during this period.

The depth increment from -85 to -112 cm corresponded to a transitional fen stage 3300–4800 years ago. In our regression model for δ13C (Fig. 4), these two depths are about 0.5 ‰ higher in δ13C than at -162 cm. The peak in δ13C may reflect a phase during which sedges transported methane directly to the atmosphere, thereby minimizing the refixation in Sphagnum cells of 13C-depleted, methanotrophic-derived carbon dioxide (Raghoebarsing et al., 2005).

The strong dependence of δ13C on peat depth partially reflected the 1.7 ‰ decline in the δ13C of atmospheric CO2 since 1850, with the lowest δ13C values above the water table, where C is of recent origin. For example, the lowest estimated values of the δ13C depth coefficients (∼ -1 ‰) in hummocks at 15, 22, and 25 cm above the mean hollow surface reflect C from the last 50 years (Fig. 4), as confirmed by Δ14C averages of 59, 29, and 52 ‰ for these three depths, where only C influenced by 14C created during thermonuclear testing should have positive Δ14C values (Table 4). Although our sampling lacked sufficient vertical resolution to explicitly include corrections for the Suess effect (e.g., as done in Esmeijer-Liu et al., 2012), the ∼ 2 ‰ increase in the depth coefficient of our δ13C regression model from the hummocks to deeper in the profile correspond well to the long-term shift in δ13C of atmospheric CO2 from preindustrial times to the present. An additional factor contributing to the higher depth coefficient could be the 1–2‰ higher δ13C in roots than in foliage and the different input depths of foliage (surface only) and roots (distributed throughout the acrotelm). The steady increase in δ13C between -5 and -25 cm depth (Fig. 4) presumably reflected the increasing dominance of preindustrial C. Depths of -35 cm and below all had Δ14C values less than -100 ‰ (Table 4), indicating primarily pre-bomb and preindustrial C when the average δ13C of atmospheric CO2 was -6.5 ‰ (versus the current value of -8.2 ‰). In addition, modern production of organic matter averaged -29‰ in δ13C (Table 2), similar to values for surficial horizons, whereas deeper horizons were between -27 and -26 ‰. The Suess effect of ∼ 1.5 ‰ therefore accounted for at least half of this difference.