A selection of recently deposited sediments from the flanks of seven rivers was made from the sample collection of the Biogeoscience group at ETH Zürich (Figure 1; Table 1). The rivers were selected to cover a large range in latitudes, climate zones, lengths, or catchment area in order to enable linking potentially emerging patterns to environmental parameters, and include the east channel of the Mackenzie River delta (NT, Canada), Fraser River (BC, Canada), Connecticut River (CT, USA), Waimea River (HI, USA), Tempisque River (Costa Rica), upper Rhône River (Switzerland), and the Rakaia River (New Zealand). For each river, material from the site closest to the river mouth was chosen to yield a brGDGT signature that best represents the upstream catchment, as well as the signal that is discharged to the marine environment.

For the Mackenzie, Rhône, and Rakaia Rivers, surface soils (0–10 cm, without litter layer) from the catchment areas were also available. All materials were collected between 2005 and 2011 during different expeditions of the Biogeoscience group, and have been stored frozen after sampling. Details about each river catchment are listed in Table 1. Annual mean air temperatures for each site are from the Global Historical Climatology Network (GHCN; http://www.noaa.gov).

The river flank sediments and soils were freeze-dried, after which macroscopic root material and plant fragments were removed. The sediments and soils were subsampled three times (10–35 g) and separated into a silt/clay fraction (<63 μm), a medium/fine sand fraction (63–200 μm), and a coarse sand fraction (200–1,000 μm) by dry-sieving on an automated shaker. Due to the relatively coarse distribution of the Waimea River sediments, the silt/clay and fine sand fractions were combined into one fraction (<200 μm), and the coarse sand fraction was split into two separate fractions of 200–500 μm and 500–1,000 μm, respectively. The workup in triplicate is indented to average out potential variability or incomplete separation by dry-sieving. All reported values for the size fractions are the average of single measurements on each of the triplicate samples and the standard deviation of these three measurements.

All available grain size fractions were homogenized and decalcified by acidifying with concentrated HCl vapors at 60–65°C for 60–72 h. The total organic carbon (TOC) and δ¹³C value of each decalcified sample was determined using a ThermoFisher Delta V Plus isotope ration mass spectrometer (IRMS) coupled with a ThermoFisher Flash elemental analyzer 1112 at ETH Zürich.

One to five grams of each size fraction was extracted using a MARS Xpress microwave extraction system, using 9:1 (v/v) dichloromethane (DCM):MeOH at 100°C for 20 min after 10 min of ramping time. The samples were centrifuged for 5 min at 400 rpm in the extraction vessels, after which the extract was pipetted off and collected in a glass vial. The residues were rinsed with 9:1 (v/v) DCM:MeOH (2x), and all solvents were combined into a total lipid extract (TLE). Following addition of a C₄₆ GDGT standard the TLEs were dried down under N₂ before passing over a silica column (deactivated with Mili-Q water, 5% dry weight) to separate the extract into an apolar and a polar fraction, using 9:1 (v/v) hexane:DCM and 1:1 (v/v) DCM:MeOH, respectively. The polar fraction, containing the GDGTs, was dried under N₂, dissolved in 99:1 hexane:isopropanol, and filtered over a 0.45 μm PTFE filter prior to analysis with high performance liquid chromatography/atmospheric pressure chemical ionization-mass spectrometry (HPLC/APCI-MS) on an Agilent 1260 Infinity series LC/MS according to Schouten et al. (2007). Separation of the GDGTs was achieved with a Grace Prevail Cyano column (3 μm, 150 × 2.1 mm) after passing through a Prevail Cyano Guard column (5 μm, 7.5 × 2.1 mm) with hexane:isopropanol (99:1, v/v) as eluent. The flow rate was 0.2 ml/min. Each GDGT-fraction was eluted isocratically with 90% A and 10% B for 5 min, and then with a linear gradient to 18%B for 34 min, where A = hexane and B = hexane:isopropanol 9:1 (v/v). Selective ion monitoring of the [M+H]⁺ was used to detect and quantify the different GDGTs, according to Huguet et al. (2006), except for that a similar response factor was assumed for the GDGTs and the internal standard.

The TOC concentrations in river sediments and soils range between 0.1 wt% in the Waimea River (Hi, USA), and 6.0 wt% in the 63–200 μm fraction of the Connecticut River (CT, USA). Although the TOC concentrations show no clear trend with latitude (and thus temperature and/or precipitation) or basin size, highest concentrations of TOC are mostly found in the 63–200 μm fraction in the river sediments, followed by the <63 μm fraction. This latter fraction also contains most TOC in soils (Figure 2A), where OC tends to form aggregates or associations with mineral surfaces, and the smallest particle size fraction (i.e., clay) with the largest surface area exhibit the highest OC concentrations (e.g., Mayer, 1994; Kögel-Knabner et al., 2008; Keil and Mayer, 2014; Doetterl et al., 2016). Although chemical and physical protection of OC by minerals has been assessed in soils and during lateral transport (e.g., hill to valley), the stability of organo-mineral associations upon entering a river and subsequent transport remains poorly studied (Doetterl et al., 2016 and references therein). However, Keil et al. (1997) reported that the relation between OC content and mineral surface area of SPM and river sediments is relatively invariant for several rivers worldwide, but is altered upon entering the marine environment. Notably, in this dataset only the Fraser River sediments have the highest TOC concentration in the <63 μm fraction, whereas the other rivers carry more TOC associated with the 63–200 μm fraction. This may reflect the presence of fragments of undecomposed plant debris carried by the river that reside in these coarser size fractions.

The δ¹³C values of OC in the river sediments and soils vary between −24 and −30‰ (Figure 3), consistent with a primary origin from higher plant sources. Nevertheless, freshwater phytoplankton or fossil carbon carried by the river may also exhibit δ¹³C values within this range, so that their contribution to the OC pool cannot be excluded. Published δ¹³C values for carbonate bedrock (−5 to 0‰; Hitchon and Krouse, 1972) and OC in river SPM (−27.7 to −26.5‰; Vonk et al., 2016) in the Mackenzie River basin suggest that the enriched OC associated with the fine fraction of the Mackenzie River sediment (−12‰; Figure 3) could be caused by the incomplete removal of inorganic carbon.

The concentration of brGDGTs varies substantially among the river sediments and soils, as well as among the grain size fractions (See datasheet in supplementary information). There is no specific fraction that contains a systematically different concentration of brGDGTs among all rivers (Figure 2B). However, brGDGTs concentrations in the river sediments are significantly correlated with TOC content (r² = 0.79), suggesting that brGDGTs and OC associated with each fraction have a common preservation mechanism. Subsequent normalization of brGDGT concentrations to TOC reveals that the OC associated with the fine fraction contains greatest proportion of brGDGTs in soils and sediments (Figure 2C). This observation would seem to support the notion that organo-mineral associations formed in soils are maintained during river transport. Furthermore, the <63 μm fraction is most likely to reflect the suspended load of sediments, which is the material that is discharged and reaches the ocean. Thus, simply based on concentrations, this suggests that brGDGTs in river sediments associated with the clay/silt-sized river sediments fraction show potential as specific tracers for soil OC.

Information on brGDGT distributions in this study was obtained using a Prevail Cyano column for the separation of brGDGTs. A new method that has recently been proposed uses 2 silica columns in tandem, which improves the separation of brGDGTs and any occurring isomers with methyl branches attached on positions 6/6′ rather than 5/5′ (De Jonge et al., 2013; Hopmans et al., 2016). A subsequent re-investigation of the global surface soil calibration set indicated that 5-methyl brGDGTs relate to MAT, and 6-methyl brGDGTs to soil pH (De Jonge et al., 2014a). This explains the pH sensitivity of the MBT' index in the previous calibrations, where 5-methyl and 6-methyl brGDGTs were not fully chromatographically resolved (Weijers et al., 2007b; Peterse et al., 2012; De Jonge et al., 2014a). Furthermore, 6-methyl brGDGTs were among the dominant brGDGTs in SPM from the Yenisei River, suggesting an aquatic source for these isomers (De Jonge et al., 2014b). Even though the “classic” separation method has been used in this study, brGDGT index values vary only marginally among different size fractions from the same site (Figures 4A,B). This indicates that there are no marked differences in contributions of 5-methyl and 6-methyl brGDGTs as a function of grain size, as this would have influenced the total peak areas and thus the index values of the different fractions.

MBT' and CBT indices are calculated in order to study distributional differences of brGDGTs among grain size fractions in soils (cf. Weijers et al., 2007b; Peterse et al., 2012). Index values show that distributional differences are very minor (Figures 4A,B, and datasheet in supplementary information), varying on average by 0.03 for both the MBT' index (max. 0.10) and the CBT index (max. 0.16). This indicates that although brGDGTs are preferentially associated with the fine grain size fractions (as indicated by the concentration data), brGDGT distributions remain relatively invariant among size fractions from a given site. Hence, upon entering a river, all size fractions should initially carry the same brGDGT signals. However, small particles are generally transported fastest due to their hydrodynamic properties (e.g., McLaren and Bowles, 1985). This could result in an increasing deviation of brGDGT distributions among size fractions downstream, where brGDGTs associated with the fast traveling smallest particles would transport the initial soil signal, whereas brGDGTs associated with the more slowly moving coarser particles would be subject to more intense alteration, and have a greater opportunity for dilution by an aquatic contribution. However, MBT' and CBT index values for brGDGTs in particle size fractions of the river sediments are similar, with average variations of 0.01 (max. 0.04) and 0.06 (max. 0.16) for the MBT' and CBT indices, respectively. This minor variation in brGDGT parameters among river grain size fractions implies either that hydrodynamic sorting processes do not exert strong control on brGDGT distributions in any of the studied rivers, or that the organo-mineral associations change during river transport. Given the high TOC-normalized brGDGT concentrations in the <63 μm fraction of the river sediments, the uniformity in brGDGT distributions across grain sizes suggests that the organo-mineral associations must evolve during transport. Indeed, recent studies on the Yangtze and Danube Rivers showed that the brGDGT signature of river sediments resembles local soils rather than a basin wide signal, suggesting that upstream brGDGTs are continuously prone to degradation and subsequent replacement by more proximal sources during transport, even when associated with minerals (Li et al., 2015 and Freymond et al., 2017, respectively).

A consequence of this scenario of continuous replacement is that newly formed associations may also include brGDGTs with an aquatic source. Previous studies have identified the occurrence of in situ brGDGT production based on clear distributional offsets between river sediments and SPM and catchment soils, generally resulting in an underestimation of reconstructed MAT (e.g., Tierney and Russell, 2009; Zell et al., 2013, 2014a,b; De Jonge et al., 2014b). Discrepancies in brGDGT distributions between soils and river sediments are also evident in the present dataset, as manifested in differing MBT' and CBT index values for soils and sediments derived from the same location (Figures 4A,B). However, this observation is based on a single soil sample along the river, which is almost certainly unrepresentative of the spectrum of soil types within the respective catchments. Moreover, soil-derived brGDGTs will have entered the river upstream of the sample location, and can thus not be directly compared to the soils collected at the same site. They should carry an upstream (higher elevation) signal, and evidence for this can be found in derived temperatures from the MBT' and CBT index values which, with the exception of the Mackenzie River, are slightly lower than measured MAT at the sample location (Figure 4C). The higher MAT reflected by the Mackenzie River sediment may be explained by the South-to-North flow direction of the river. Additionally, the strong seasonality and related short growth period at this latitude likely causes a bias in brGDGT distributions toward summer temperatures in both soils and river sediments (Peterse et al., 2014).

Nevertheless, regardless of the actual source of the brGDGTs in river sediments, application of the MBT'-CBT proxy generates MATs that relate well with measured MATs at the sampling location (r² = 0.71), especially when the South-to-North flowing Mackenzie River sample is omitted (r² = 0.93). This observation, albeit on a very limited dataset, suggests that rivers worldwide transport a temperature-sensitive signal from land to sea, and that down-core changes in brGDGT distributions in near-river mouth sediments will reflect a record that is primarily forced by temperature. The uniform molecular distributions of brGDGTs among grain size fractions furthermore implies that their signature in river sediments will unlikely be altered by hydrological sorting, although these processes still could influence the overall flux of fluvially transported brGDGTs as a consequence of differential transport of particles with different brGDGT concentrations. Consequently, the BIT index may be most prone to changing hydrodynamic conditions, such as those associated with seasonal or shorter-term (e.g., storm events) variations in fluvial discharge (cf. Kim et al., 2007).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer HN and handling Editor declared their shared affiliation and the handling Editor states that the process nevertheless met the standards of a fair and objective review.