Volume 36, Issue 5 e2021PA004225
Research Article
Free Access

Diversification of Iron-Biomineralizing Organisms During the Paleocene-Eocene Thermal Maximum: Evidence From Quantitative Unmixing of Magnetic Signatures of Conventional and Giant Magnetofossils

Courtney L. Wagner

Corresponding Author

Courtney L. Wagner

Department of Geology and Geophysics, University of Utah, Salt Lake City, UT, USA

Correspondence to:

C. L. Wagner,

[email protected]

Search for more papers by this author
Ioan Lascu

Ioan Lascu

Department of Mineral Sciences, National Museum of Natural History, Smithsonian Institution, Washington, DC, USA

Search for more papers by this author
Peter C. Lippert

Peter C. Lippert

Department of Geology and Geophysics, University of Utah, Salt Lake City, UT, USA

Global Change and Sustainability Center, University of Utah, Salt Lake City, UT, USA

Search for more papers by this author
Ramon Egli

Ramon Egli

Division of Data, Methods and Models, Central Institute of Meteorology and Geodynamics, (ZAMG), Vienna, Austria

Search for more papers by this author
Kenneth J. T. Livi

Kenneth J. T. Livi

Materials Characterization and Processing Center, Department of Materials Sciences and Engineering, Johns Hopkins University, Baltimore, MD, USA

Search for more papers by this author
Helen B. Sears

Helen B. Sears

Department of Geology, Colby College, Waterville, ME, USA

Search for more papers by this author
First published: 23 April 2021
Citations: 7

This is a companion manuscript to Wagner et al. ( 2021), https://doi.org/10.1073/pnas.2018169118.


Conventional magnetofossils represent magnetic mineral remains of magnetotactic bacteria. Giant magnetofossils have no known modern analog. Both conventional and giant magnetofossil assemblages can record paleoenvironmental information through changes in magnetotactic bacteria diversity driven by nutrient supply and oxygenation. We use marine sediments that record a rapid global warming event, the Paleocene-Eocene Thermal Maximum (∼56 Ma), to assess how abundant well-preserved magnetofossils with high morphological disparity record paleoenvironmental information. We find that conventional magnetofossils can be distinguished from giant, needle-shaped magnetofossils using principal component analysis of first-order reversal curves (FORC-PCA); moreover, FORC-PCA may be able to distinguish between well-preserved magnetically soft and magnetically hard magnetofossils. FORC-PCA is a robust, nondestructive technique that can be applied to other marine archives to understand how these ecosystems respond to rapid environmental change. We hypothesize that the sudden appearance of giant magnetofossils represents a natural response to niche expansion within the water column (a thicker oxic-anoxic interface) and eutrophication (via iron supply) by protists that biomineralize giant magnetofossils. This application has potential as a proxy for changes in marine oxygen and iron concentrations stimulated by rapid planetary change.

Key Points

  • Giant magnetofossils are produced by microorganisms in response to vertical mixing and increased iron supply during rapid warming

  • High morphological disparity and abundance of magnetofossils are linked to seasonal nutrient cycling

  • Principal component analysis of first-order reversal curves can distinguish between groups of low and high coercivity magnetofossils

Plain Language Summary

Magnetofossils are fossilized magnetic particles produced by magnetotactic bacteria and other microorganisms. They are geologically significant because they record environmental and ecological information including bacterial diversity and marine oxygen levels. Here, we test if high-field magnetic measurements can discriminate between different sizes and shapes of magnetofossils within bulk samples of marine sediments that record rapid global warming 56 million years ago, the largest global warming event in the past ∼80 million years. These findings have important implications for tracking the timing and extent of deoxygenation and nutrient supply in the ocean stimulated by climate change.

1 Introduction

The Paleocene-Eocene Thermal Maximum (PETM) was a geologically rapid global warming event that occurred ∼56 million years ago. The Earth warmed 5°C–8°C over several thousand years during the PETM. The magnitude and pace of warming during the PETM led to increased ocean temperatures (3°C–4°C in surface waters and 2°C–6°C in bottom waters), ocean acidification and bottom water deoxygenation (and subsequent benthic foraminifera extinctions), rapid radiation of land mammals, expansion of tropical forests at high latitudes, the appearance of giant insects and reptiles, and stronger, more intense seasonal weather events like flash floods, hurricanes, and frequent wildfires (e.g., McInerney & Wing, 2011). The PETM is marked by a −3‰ carbon isotope excursion (CIE) in marine bulk carbonate, and it is an analog, albeit an imperfect one, for anthropogenic global warming and environmental change (Gutjahr et al., 2017; Kennett & Stott, 1991; McInerney & Wing, 2011; Zachos, Dickens, & Zeebe, 2008; Zachos, Pagani, et al., 2001; Zeebe et al., 2016).

Modern climate change affects both terrestrial and marine biota via habitat loss and mass extinction, but the full extent and timing of these effects throughout the modern Earth system remain unknown (Ceballos et al., 2015; Steffen et al., 2018; Zachos, Dickens, & Zeebe, 2008; Zeebe et al., 2016). One major question is the sensitivity of coastal ecosystems and the deep ocean to anthropogenic climate change (Calosi et al., 2019; Finnegan et al., 2015; Harnik et al., 2012; Hoegh-Guldberg & Bruno, 2010). This uncertainty in coastal ecosystem sensitivity is underscored by the fact that continental shelves host physical and biological processes that affect carbon and nutrient cycling between shallow and deep water, the generation of new biomass, and the fate of organic matter in general (Aller et al., 1996; Gooday et al., 2009; Hoarfrost et al., 2019; Humphreys et al., 2019; Middelburg & Levin, 2009). To better understand the interplay between coastal ecosystems and the deep ocean, and how they might respond to modern environmental change, we focus on continental shelf sediments deposited during the PETM.

PETM sediments from the New Jersey margin contain abundant single-domain magnetite that has been interpreted as the result of detrital inputs, impact melts, fires, magnetotactic bacteria (MTB), or a combination of these sources (Kent, Cramer, et al., 2003; Kent, Lanci, et al., 2017; Kopp, Raub, et al., 2007; Kopp, Schumann, et al., 2009; Lippert & Zachos, 2007; Schaller et al., 2016; Wang, Kent, & Jackson, 2013; Wang, Wang, et al., 2015). Electron microscopy and rock magnetic analyses of these and other PETM sections provide evidence for abundant magnetofossils (the fossilized remains of MTB) and reveal the occurrence of so-called giant magnetofossils (Chang, Harrison, et al., 2018; Chang, Roberts, Williams, et al., 2012; Kopp, Raub, et al., 2007; Kopp, Schumann, et al., 2009; Lippert & Zachos, 2007; Schumann et al., 2008). The unique morphology, crystallography, and magnetic properties of giant magnetofossils suggests that they have a biogenic origin (Schumann et al., 2008). Micromagnetic modeling of these and similar giant magnetofossils suggests that some giant magnetofossils may contribute to the single-domain magnetic signature typical of conventional magnetofossils (Chang, Harrison, et al., 2018; Chang, Roberts, Williams, et al., 2012; Wagner et al., 2021). Although most evidence is consistent with a biogenic origin for the single-domain magnetite, some researchers argue that impact ejecta or pyrogenic particles contribute to the single-domain enhancement of these sediments (Kent, Cramer, et al., 2003; Kent, Lanci, et al., 2017).

Magnetofossils have the potential to encode unique paleoenvironmental information, including changes in microbial ecosystem structure, in response to rapid planetary change. Therefore, it is important to determine the degree to which magnetofossils contribute to the single-domain magnetite on the New Jersey margin. For example, Chang, Harrison, et al. (2018) suggested that magnetofossil assemblage changes from a deep-marine sediment core from the South Atlantic may be linked to a gradual decrease in deep-sea oxygen levels during the PETM. Here, we test this interpretation in a setting sensitive to nutrient supply, sea-level change, and sediment loading (Kopp, Schumann, et al., 2009; Self-Trail, Robinson, et al., 2017; Sluijs & Brinkhuis, 2009; Stassen et al., 2012) by determining the response of magnetofossils, with different morphologies, to environmental change within this near-shore marine ecosystem. We also test the capability of magnetic unmixing techniques to resolve changes in magnetofossil morphology throughout the PETM at the Wilson Lake locality.

2 Geologic Setting

The Wilson Lake-A (WL-A) core from Wilson Lake, New Jersey, USA, contains a neritic record (∼70 m water depth) of the PETM (Figure 1) and is known for its exceptional preservation of marine microfossils and for geochemical proxy records within the Marlboro Clay (Olsson & Wise, 1987; Stassen et al., 2012). The fidelity and importance of the WL-A PETM record inspired drilling of a second Wilson Lake core (Miller et al., 2017). Several near- and off-shore records complement the WL-A core (Figure 1) (Miller, 19972017). This suite of cores, combined with deep ocean cores, provide an opportunity to correlate these sections and to determine the timing and spatial distribution of environmental changes during the PETM as a function of water depth (Chang, Harrison, et al., 2018; Kopp, Schumann, et al., 2009; Self-Trail, Robinson, et al., 2017).

Details are in the caption following the image

Paleogeographic map and cross-section of the New Jersey margin with the location of several New Jersey Coastal Plain Drilling Project and United States Geological Survey Eastern Mapping Team cores. (a) Paleogeographic map adapted from Kopp, Raub, et al. (2007) with the following cores highlighted: C, Clayton; WL, Wilson Lake (this study); An, Ancora; BR, Bass River. The red line labeled A and B represents an approximate depth-transect (dotted line separating northern New Jersey from Southern New Jersey). (b) Cross-section along line A and B, from the map, highlighting how the approximate water depth and thickness of the Marlboro Clay (brown) changes between cores. Approximate water depths are from Stassen et al. (2012). Cenozoic (Cz), Cretaceous (K), and pre-Cretaceous (pre-K) outcrop limits are indicated.

Marine Paleocene-Eocene boundary sediments from the WL-A core were deposited in approximately 70 m water depth (e.g., Gibson et al., 1993; Stassen et al., 2012) (Figure 1b). We collected WL-A sediments spanning the uppermost Paleocene and lower Eocene, with a focus on the earliest Eocene Marlboro Clay, from ∼109.85 to 96.94 m core depth. The Marlboro Clay is the regionally distinct lithofacies deposited during the PETM.

The PETM is generally characterized by three phases: (1) a pre-onset excursion (POE) in bulk carbon isotopes, (2) the main carbon isotope excursion (CIE), and (3) a recovery toward baseline, pre-PETM bulk carbon isotope values. The POE has only been identified in a few study sites (Bowen et al., 2014; Honegger et al., 2020; Luciani et al., 2007; Robinson & Spivey, 2019; Self-Trail, Powars, et al., 2012; Sluijs et al., 2007). The CIE is sub-divided into CIE onset and CIE core intervals. Following the age models and bulk carbon isotope trends summarized by McInerney and Wing (2011), the CIE onset is estimated to have lasted for ∼6–10 kyr and the CIE core for ∼100–200 kyr (Giusberti et al., 2007; Murphy et al., 2010; Röhl et al., 2007; Zachos, Röhl, et al., 2005). Similarly, the recovery interval is characterized by two subphases, Phase I and Phase II, lasting ∼33 and ∼50 kyr, respectively (McInerney & Wing, 2011; Murphy et al., 2010).

The WL-A core contains an expanded and nearly complete CIE onset (∼109.85–108.7 m) and core (∼108.7–96.94 m); it does not contain a recovery interval (Figure 2) (McInerney & Wing, 2011; Stassen et al., 2012). The upper PETM is punctuated by two unconformities, at ∼96.32 and ∼94.79 m depths (Gibson et al., 1993; Miller, 1997; Sluijs et al., 2007; Stassen et al., 2012). Sediments deposited during the CIE have an average accumulation rate of 19 cm/kyr based on the integrated stratigraphy and linear age model of Stassen et al. (2012). Preservation of calcareous and siliceous marine microfossils and nannofossils, as well as iron oxides, within the Marlboro Clay at WL-A is exceptional and well documented (e.g., Gibbs, Bown, et al., 2006; Gibbs, Bralower, et al., 2006; Schumann et al., 2008; Stassen et al., 2012; Zachos, Schouten, et al., 2006). This gives us the opportunity to examine an expanded, detailed, well-studied PETM CIE section, using electron microscopy and magnetic component unmixing to track magnetofossil assemblage changes during these key intervals of planetary change.

Details are in the caption following the image

Stratigraphy of the Wilson Lake-A (WL-A) core with bulk magnetic parameters. From left to right: stratigraphic intervals, δ13C from bulk carbonate and benthic foraminifera, magnetic parameters, including saturation magnetization (Ms), hysteresis squareness (Mrs/Ms), coercivity of remanence (Bcr), and, on the far right, our target intervals for high-field (HF) FORC measurements (black stars) and transmission electron microscopy (TEM) (red stars). Carbon isotope values are from Zachos, Schouten, et al. (2006) and magnetic parameters are from this study (solid lines) and Lippert and Zachos (2007) (single points). The CIE onset is highlighted in pink and the CIE core is highlighted in orange. The two wavy lines at ∼96.32 and ∼94.79 m denote unconformities at the top of the WL-A section.

3 Materials and Methods

3.1 Sample Preparation for Magnetic Measurements

We collected WL-A sediments spanning the uppermost Paleocene and lower Eocene, from ∼112.6 to 91.8 m depths, with a focus on the earliest Eocene Marlboro Clay, from ∼109.85 to 96.94 m depths. We prepared dry, bulk sediment chips (70–120 mg) every ∼24 cm throughout the Paleocene, every ∼8 cm throughout the early Eocene in the CIE onset interval, and every ∼23 cm throughout the CIE core interval (Figure 2). We collected 90 specimens in total, with 65 specimens from both the onset and core intervals at WL-A. This sample set has a stratigraphic resolution that is nearly triple that of previous studies, which allows us to verify finer details of the magnetic mineral stratigraphy during the CIE and to reveal finer-scale structure at the same resolution as geochemical and biostratigraphic studies. All specimens were used to measure magnetic hysteresis and direct current (DC) demagnetization curves. The 65 PETM-interval specimens were used for first-order reversal curve (FORC) measurements, and nine specimens from our target intervals (also from our PETM intervals) were used for high-field, high-resolution FORC measurements (Figure 2). Intact sediment chips were secured inside gelatin capsules and were then affixed to a vibrating sample magnetometer (VSM) probe for the measurements.

3.2 Magnetic Hysteresis and DC Demagnetization Curves

Hysteresis parameters, including saturation magnetization (Ms), saturation remanence (Mrs), and bulk coercivity (Bc) were determined from hysteresis loops measured using a Model 3900-4 Lakeshore Cryotronics/Princeton Measurements Corporation VSM at the Utah Paleomagnetic Center, University of Utah. Hysteresis measurements were made in 5 mT field increments to a maximum field of 1 T, and an averaging time of 0.3–0.5 s. DC demagnetization curves of the remanent magnetization acquired in a +1 T field were measured with 2 s averaging time using a sequence of ≥150 logarithmically spaced fields in the opposite direction, from −10 µT to −1 T. Coercivity of remanence (Bcr) was calculated from DC demagnetization curves. We uploaded raw data files from our hysteresis and DC demagnetization measurements to the Institute for Rock Magnetism database software (https://cse.umn.edu/irm/irm-software). This software package provides mass, noise, drift, and high-field slope correction from each hysteresis measurement (Jackson & Solheid, 2010). Magnetic hysteresis and DC demagnetization parameters provide an overview of the bulk magnetic stratigraphy of the section, which helps us identify intervals of magnetic interest (Figure 2) and enable rigorous correlation of our magnetic stratigraphy to published data sets (Kent, Cramer, et al., 2003; Kent, Lanci, et al., 2017; Kopp, Raub, et al., 2007; Kopp, Schumann, et al., 2009; Lippert & Zachos, 2007; Schumann et al., 2008)

3.3 First-Order Reversal Curves and Principal Component Analysis

First-order reversal curves (FORC) are a series of partial hysteresis curves obtained by decreasing the applied field from positive saturation to a reversal field Br, and then measuring the magnetization M while the field B is increased back to positive saturation (Pike et al., 1999). The succession of curves obtained from regularly spaced Br values “sweeps” the B-M space enclosed by the major hysteresis loop. The mixed second derivative of magnetization with respect to measurement field and reversal field (e.g., Pike et al., 1999; Roberts, Heslop, et al., 2014) highlights transitions between magnetization states of individual magnetic particles, or systems made of several particles, such as magnetofossils, that are not evident from simple bulk magnetic measurements. FORCs are sensitive to magnetic particle coercivity, domain state, and magnetic interactions, which makes them useful for distinguishing the mineralogy, size, shape, and spatial arrangement of magnetic particle assemblages within bulk materials (Pike et al., 1999; Roberts, Heslop, et al., 2014; Roberts, Pike, & Verosub, 2000; Roberts, Zhao, et al., 2018).

We performed FORC measurements for systematic characterizations and principal component analysis (FORC-PCA) using the same VSM as for hysteresis measurements, as well as a Lake Shore Model 8604 VSM at the Biogeomagnetism Laboratory, National Museum of Natural History, Smithsonian Institution. Each FORC data set was obtained using a 1 T saturation field, with Bc (the horizontal axis of the FORC diagram) ranging from 0 to 160 mT and Bu (the vertical axis of the FORC diagram) ranging from −40 to 40 mT, in 1.06 mT steps, with measurement averaging times of 0.3–0.5 s. Raw FORC data were processed using FORCinel 3.07 (Harrison & Feinberg, 2008) with the VARIFORC variable smoothing protocol (Egli, 2013). All FORC measurements were processed using the same parameters (supplemental text).

We also measured a set of high-field FORCs on nine specimens with the most extreme endmember proportions (asterisks in Figure 2), as identified with the FORC-PCA. We include these measurements to further constrain the characteristics and origin of these endmembers. These high-field measurements were performed at both the University of Utah and at the Smithsonian, using a saturation field of 1–2 T, Bu,min = −40 to −10 mT, Bu,max = 40–90 mT, Bc,max = 220–500 mT, an averaging time of 0.1–0.5 s, and a field increment of 0.8–1 mT. All high-field FORC data sets were processed using FORCinel 3.07 (Harrison & Feinberg, 2008) and the same VARIFORC variable smoothing parameters (supplemental text) (Egli, 2013).

FORC-PCA uses FORC data sets from a suite of samples. An important underlying assumption when applying this technique is that the selected samples contain a common set of magnetic components and that both reversible and irreversible contributions of these components are expressed as a linear combination of fixed endmembers. FORC-PCA can be a powerful tool for discriminating the contributions of different magnetic grain sizes (viscous particles close to the superparamagnetic/single-domain threshold size, single-domain, single vortex, multivortex, and multidomain) and mineralogical components (Channell et al., 2016; Harrison et al., 2018; Lascu et al., 2015; Pike et al., 2001; Roberts, Almeida, et al., 2017; Roberts, Zhao, et al., 2018). Feasibility metrics are incorporated into the unmixing analysis to guide endmember selection. These metrics help determine whether the endmembers selected are physically realistic (Harrison et al., 2018). We used all 65 original FORC data sets in our PCA. We extracted a grid over each data set from Bu = −90–150 mT, Bc = 0–150 mT with a 0.5 mT mesh size.

3.4 Magnetic Extracts and Transmission Electron Microscopy

We prepared magnetic extracts for three sampling intervals that were examined using FORC-PCA: WL35900 (109.4 m, onset), WL35800 (109.1 m, onset), and WL35245 (107.4 m, CIE Core). We modified the extraction protocol outlined by Strehlau et al. (2014), as described in the supplemental text. TEM analysis was performed using a TECNAI TF30 scanning/TEM at the Materials Characterization and Processing Facility at Johns Hopkins University using an accelerating voltage of 300 keV.

Approximately 20 TEM images of each extract were collected at 5900X magnification. We counted all identifiable magnetofossils (>950 magnetofossils per sample) in these images and separated them into nine categories based on their dimensions and morphologies: immature (<30 nm long), cuboctahedra, small bullet shapes (<85 nm long), medium-sized bullet shapes (90–180 nm long), elongated prisms, large bullet shapes (200–390 nm long), giant bullet shapes (>500 nm long), giant spindles, and giant needles (Figure 3). The first six categories are based on known and previously described morphologies (e.g., Posfai et al., 2013). The last three categories are giant magnetofossils similar in size and shape to giant magnetofossils identified in other studies (e.g., Chang, Harrison, et al., 2018; Chang, Roberts, Williams, et al, 2012; Kopp, Schumann, et al., 2009; Schumann et al., 2008). We also counted teardrop-shaped particles, but they did not fit our criteria for magnetofossils (more details below). We note that the categories “elongated prisms”, “rod”, and “needle” have been used interchangeably to refer to the same category of giant magnetofossils (e.g., Chang, Roberts, Williams, et al., 2012). Here, we refer to elongated prisms as their own category, within our conventional magnetofossil categories, based on known modern magnetofossil morphologies (e.g., Abreu et al., 2013; Posfai et al., 2013). We use the category needle in place of rods and recommend that needles must have aspect ratios >5:1 to be differentiated from prisms.

Details are in the caption following the image

TEM images of magnetofossils with highlighted examples for each of the nine magnetofossil categories. TEM images from (a) extract WL35900 with giant magnetofossils (giant bullets and spindles), (b) extract WL35800 with giant magnetofossil (needles) and nonmagnetofossil categories (teardrops), (c) extract WL35800 with conventional magnetofossils (cuboctahedra and elongated prisms), and (d) extract WL35800 with conventional magnetofossils (small, medium, and large bullets; and immature).

We collected energy-dispersive X-ray (EDX) spectra and selected area electron diffraction (SAED) patterns over multiple particles to ensure that each magnetofossil morphology, and most of the magnetic material observed in our extracts, is made of magnetite or its low-temperature oxidation product, maghemite (Figure 4). SAED patterns distinguish poorly between magnetite and maghemite. Our magnetic data sets suggest that the iron oxides are magnetite, but we cannot rule out the presence of maghemite without the use of Mössbauer or electron energy-loss spectroscopy. Here we refer to iron oxides with distinct, recurring morphologies as magnetofossils to distinguish them from irregular, nonrecurring morphologies that are more consistent with a detrital or authigenic origin. We note that some of the iron oxides in our magnetofossil counts may be authigenic or diagenetic, for example, produced by reductive alteration of nanogoethite (e.g., Till, Guyodo, Lagroix, Morin, Menguy, & Ona-Nguema, 2017) because some morphologies are not unique to biogenic magnetite. Additional crystallographic and higher resolution geochemical studies on these extracts may help identify the extent to which particles from these alternative sources are present.

Details are in the caption following the image

(a) EDX analysis of magnetic extracts containing magnetofossils mixed with clay minerals. (b) TEM image from WL35800 with a circle highlighting the area over which the diffraction pattern in (c) was collected. (c) Selected area electron diffraction (SAED) pattern from the circled area in (b). (d) Radially averaged profile of (c) along with calculated diffraction patterns for magnetite and basal reflections of montmorillonite. This indicates that the circled area in (b) contains a mixture of either magnetite or maghemite with a dioctahedral sheet silicate (clay).

We used the Gatan GMS 3 Digital Micrograph Software to measure the length and width of ∼30 conventional magnetofossils from each category (∼10 magnetofossils from each magnetic extract, per category). Giant magnetofossils were scarcer throughout our extracts and we typically measured 2–7 crystals, total, for each category. Using these measurements, we calculated average lengths, widths, width/length, volumes, and magnetizations for each category. We plotted lengths versus width/length of the magnetofossils used in these calculations to predict the average magnetic domain state of the nine categories (Figure 5).

Details are in the caption following the image

Dimensional analysis of magnetofossils identified in magnetic extracts from samples WL35900, WL35800, and WL35245 with predicted room-temperature low-field domain states: superparamagnetic (one unstable, homogeneously magnetized domain), single-domain (one stable, homogeneously magnetized domain), vortex (nonhomogeneous magnetization with low magnetic moment), and multidomain (several magnetic domains). The theoretical single-domain state thresholds for crystals occurring in isolated form or in chains are highlighted: (1) lower limit (solid lines) for a chain of six crystals separated by gaps of 0 (lower curve) and 0.6 particle lengths (Newell, 2009), (2) lower limit (dashed line) for isolated crystals (Butler & Banerjee, 1975), (3) upper limits (dashed lines) for isolated crystals with long axis parallel to the <100> and <111> crystallographic axes, respectively (Muxworthy & Williams, 2006), and (4) upper limit (solid line) for chains of three crystals (Muxworthy & Williams, 2006). Dimensions of conventional magnetofossils and modern biogenic magnetite of various shapes, from well-characterized MTB, are shown in purple, orange, and green (Hesse, 1994; Iida & Akai, 1996; Isambert et al., 2007; Lean & McCave, 1998; Lefèvre, Pósfai, et al., 2011; Lippert & Zachos, 2007; Mann, Sparks, & Blakemore, 1987; McNeill, 1990; Meldrum et al., 1993; Moisescu et al., 2008; Peck & King, 1996; Petersen, von Dobeneck, & Vali, 1986; Petersen, Weiss, & Vali, 1989; Simpson et al., 2005; Sparks et al., 1990; Vali et al., 1987). We also observed teardrop-shaded grains in our extracts and, although these are not considered magnetofossils, we include them for comparison. Figure includes data presented in Wagner et al. (2021).

To calculate approximate crystal volumes, we simplified the shapes of our categories as follows: a rectangular prism for elongated prisms and needles, two cones for spindles, and an ellipsoid for all other magnetofossil categories. The average volume V for each category was then multiplied by the spontaneous magnetization µs = 480 kAm−1 of magnetite at room temperature to find the average magnetic moment m = s per magnetofossil for each category. The resulting magnetic moment was then multiplied by the total number of magnetofossils counted in each category to find the total magnetic moment carried by each category. The relative magnetic contribution of each category to the extracted material was then obtained by dividing the corresponding calculated magnetic moment by the saturation moment of the magnetic extract. We performed a second calculation to estimate the relative contribution of each magnetofossil category to the total moment of all extracted magnetofossils. Although a decrease of the extraction efficiency is possible for smaller (<1 µm) sizes (Strehlau et al., 2014), each magnetofossil morphology is expected to be extracted with the same efficiency in all samples, so that downcore variations should reflect true changes in magnetofossil abundances. Immature magnetofossils, elongated prisms, giant bullets, and giant spindles have been excluded from single-domain moment calculations because they are more likely to exhibit superparamagnetic, vortex, or multidomain behavior unless they are preserved in chains (Figure 3) (Chang, Roberts, Williams, et al., 2012; Moskowitz, Frankel, & Bazylinski, 1993; Penninga et al., 1995).

4 Results

4.1 Hysteresis and DC Demagnetization Curve Parameters

Bulk magnetic hysteresis parameters are shown alongside the lithostratigraphy and stable isotope chemostratigraphy of the WL-A core in Figures 2 and S1. A large increase in Ms, Mrs, hysteresis squareness (Mrs/Ms), Bc, and Bcr coincides with the abrupt PETM onset. These values remain high throughout the PETM at WL-A and return abruptly to background levels above the first unconformity at ∼96.32 m (Figures 2 and S1). This bulk magnetic signature of the PETM is consistent with previous studies at nearby locations (Kent, Cramer, et al., 2003; Kent, Lanci, et al., 2017; Kopp, Raub, et al., 2007; Kopp, Schumann, et al., 2009; Lippert & Zachos, 2007; Schumann et al., 2008). Our high-resolution data set highlights additional trends in these parameters, including a large Bcr peak within the onset interval from ∼109.8 to 108.6 m and two smaller peaks in Bcr within the CIE core interval at ∼104.9 and ∼96.9 m (Figure 2); the same general trend is observed in Bc but the peak at ∼104.9 m is less obvious (Figure S1). We note a slight systematic difference between Bc and Bcr values reported here and by Lippert and Zachos (2007) (Figures 2 and S1), which can be attributed to measurement set-up, partial magnetite oxidation, and/or sample anisotropy. There are two broad Ms peaks over the CIE core interval that are separated by a small decrease at ∼101.5 m (Figure 2); the same general trend is observed in Mrs but this parameter is characterized by more random variations around a constant mean value (Figure S1). Mrs/Ms values of ∼0.4 suggest that single-domain particles dominate the magnetic remanence of the Marlboro Clay (Figure 2).

4.2 First-Order Reversal Curve Diagrams

FORC diagrams reflect additional changes throughout the PETM (Figures 6 and S2). All FORC diagrams from specimens within both the onset and CIE Core intervals contain a sharp horizontal ridge extending along the Bc axis from ∼0 to 160 mT, and concentrated along the Bu axis between −1.25 and 2 mT. The limited vertical extension of this so-called central ridge is controlled by the smoothing parameters used to process the FORC measurements (see Egli et al., 2010). FORC diagrams from the CIE interval are also distinguished by a region of negative amplitude in the lower left-hand corner, from Bc ≈ 0–30 mT and Bu ≈ −20 to −150 mT. The combination of a sharp central ridge and this negative region is typical of noninteracting uniaxial single-domain particles (Muxworthy, Heslop, & Williams, 2004; Newell, 2005) or linear chains of such particles (Egli et al., 2010).

Details are in the caption following the image

Nine high-resolution, high-field FORC diagrams with corresponding coercivity profiles along the central ridge (red curves). (a–f) CIE core interval. (g–i) CIE onset interval. For a low-noise FORC diagram for WL35950 see Wagner et al. (2021), where we stack and average multiple measurements from the same sample.

Horizontal profiles along the central ridges approximate the coercivity distribution of noninteracting uniaxial single-domain particles, or chains of such particles (Egli et al., 2010; Heslop et al., 2014). Central ridge profiles throughout the PETM interval are characterized by peak contributions in the 25–60 mT range. Average peak positions for onset specimens are toward the higher end of this range, whereas most CIE core specimens are near the lower end of this range. In most cases, horizontal profiles from the CIE interval feature a main peak at ∼30 mT and a distinct shoulder at 50–60 mT; these features indicate the presence of at least two coercivity components with median coercivities similar to those of the “biogenic soft” and “biogenic hard” magnetofossil components in Egli (2004a). Horizontal central ridge profiles for specimens from the CIE core interval have larger amplitudes than those of specimens from the CIE onset interval. All profiles taper toward coercivities >120 mT. The subset of extended FORCs measured at higher fields show that the central ridge is mostly comprised within Bc ≈ 200 mT (Figure 6), but continues to exist at higher fields. The minimum amplitude at ∼300–400 mT, close to the maximum theoretical coercivity of magnetite, marks the transition to small magnetic contributions from high-coercivity minerals, such as goethite (Wagner et al., 2021). Other features of the FORC diagrams in Figures 6 and S2 are much smaller than the central ridge in amplitude, but they contribute significantly to the bulk saturation remanence, due to their extension in both dimensions of the FORC space. They are discussed below.

4.3 Principal Component Analysis (PCA)

We performed two separate PCAs: one using a single principal component (PC1) which accounts for 65.8% of the variability in the data, and one using two principal components (PC1 and PC2), where PC2 accounts for an additional 17.4% of data variance. We attribute the residual data variability captured by a third principal component (6.5%) to small differences in instrument operating conditions between the two VSMs used, and to random measurement errors. The PETM interval of the WL-A core can therefore be modeled as either a binary (PC1) or a ternary (PC1 and PC2) mixture of magnetic endmembers. We defined our endmembers by moving along the trajectory of each PC, beyond the space defined by the data, but staying as close to the data as possible. The FORC-PCA software produces a synthetic FORC diagram for each endmember during this process so that the user can limit the endmember choice within plausibility limits based, for instance, on benchmark and micromagnetic modeling studies (Harrison et al., 2018). Below we describe our solution using three endmembers and refer readers to the supplemental text for a description of the two end-member model.

A FORC-PCA solution that includes PC1 and PC2 identified three endmembers: a magnetically hard EM1, a magnetically soft EM2, and a mid-range coercivity EM3 (Figure 7). The simulated FORC diagram for EM1 has a central ridge extending to >150 mT with a peak at ∼57 mT. The simulated FORC diagram for EM2 has a central ridge extending to ∼145 mT with a peak at ∼22 mT. The simulated FORC diagram for EM3, the mid-range endmember, has a central ridge extending to ∼150 mT with a peak at ∼32 mT. The central ridge of EM3 features a shoulder at ∼60 mT, which is also present, to a minor extent, in EM2. Feasibility metrics indicate that the chosen endmembers fall within the 97% significance interval (Figure 7e).

Details are in the caption following the image

PCA results from 65 FORC data sets using PC1 and PC2. (a) Simulated FORC diagram for the magnetically hard EM1. (b) Simulated FORC diagram for the mid-coercivity EM3. (c) Simulated FORC diagram for the magnetically soft EM2. (d) Horizontal profiles of the FORC diagrams for EM1, EM2, and EM3 along Bu = 0. (e) PCA score plot with samples highlighted according to depth, white to black (down-core to up-core), and feasibility metrics (blue confidence intervals) of Harrison et al. (2018). (f) Ternary diagram with relative proportions of EM1, EM2, and EM3 for each sample (also highlighted white to black, down-core to up-core).

Our three-endmember analysis indicates that EM1 is the dominant component within the onset interval and comprises ≥71% of the signal (Figure 8); there are four smaller peaks in EM1 within the CIE core interval at ∼104.9, 102.1, 99.7, and 97.6 m, respectively. There is a larger peak at the top of the section at ∼96.6 m. EM2 has a broad maximum within the lower part of the CIE core interval where it comprises ≥44% of the signal; there are two smaller EM2 peaks at ∼103.4 and 100.6 m. EM3 features a large spike at the start of the CIE core interval ∼108.6 m, where it comprises ∼74% of the signal, and a step-wise increase to ∼50% over the upper part of the section. Overall, EM1 dominates the onset interval, EM2 dominates the lower ∼4 m of the CIE core, and both EM1 and EM3 dominate the upper part of the CIE core interval.

Details are in the caption following the image

Stratigraphy of the WL-A core with unmixed magnetic components from FORC-PCA based on two PCs. From left to right: Stratigraphic column; relative contributions of EM1, EM2, and EM3 from FORC-PCA with EM peaks highlighted by black arrows; and, on the far right, our target intervals for high-field (HF) FORCs (black stars) and TEM (red stars) from intervals WL35900 (bottom star), WL35800 (middle star), and WL35245 (top star). Endmembers are expressed as percentages per specimen for all 65 PETM specimens. The symbols and color schemes follow those of Figure 2.

4.4 Transmission Electron Microscopy and Magnetofossil Calculations

TEM analysis of magnetic extracts from sampling intervals WL35900, WL35800, and WL35245 reveal a variety of iron oxide shapes and sizes (Figures 3, 4, and S5–S7). SAED patterns indicate a mixture of either magnetite or maghemite, with a dioctahedral sheet silicate, presumably kaolinite or illite-smectite (Figures 4c and 4d) (Lombardi, 2013).

Cuboctahedral crystals are the most abundant magnetofossil category in extracts from WL35900 and WL35800 (both from the CIE onset interval), in which they comprise 54% and 58% of the examined magnetofossils; they are the second most abundant category in WL35245 (from the CIE core interval), accounting for 38% of the magnetofossils. Immature magnetofossils are the most abundant category in WL35245, in which they comprise 42% of the magnetofossils; immature magnetofossils account for 29% and 12% of the total magnetofossils in WL35900 and WL35800. We identified all three giant magnetofossil categories in extracts from WL35900 and WL35800 but did not observe any giant magnetofossils in the WL35245 extract. Giant magnetofossils are the scarcest of all the categories (<1% of the total magnetofossils in each extract), consistent with findings from a different site reported by Wang, Wang, et al. (2015). Results from our magnetofossil counts are summarized in Tables S1–S3.

We calculated the average magnetic moment for each magnetofossil category (Table S4) and the percentage of each category that likely contributes to the saturation magnetization of each extract assuming they exist as isolated particles (Tables S1–S3); these calculations may be minimum estimates because we cannot guarantee that our magnetic extractions dislodged all magnetic particles from the bulk sediment. Results from these calculations indicate that two magnetofossil categories dominate the total magnetofossil moment of extract WL35900, during the CIE onset: giant bullets (43%) and cuboctahedra (34%). Cuboctahedral magnetofossils contribute the most to extracts WL35800 (68%) and WL35245 (71%). Separate calculations focused on the single-domain moment contribution; these indicate that cuboctahedral morphologies contribute the most to all extracts: 73% of WL35900, 74% of WL35800, and 77% of WL35245.

We also identified rounded, sometimes teardrop-shaped crystals in each extract (Figure 3). These crystals are scarce (<5 crystals observed per extract) and do not fit any of our criteria for magnetofossil identification, therefore, they are unlikely to be significant magnetic carriers (Li, Liu, Liu, et al., 2020). We note that similar particles were described as small spearhead magnetofossils in one study (Kopp, Schumann, et al., 2009) and larger (>63 µm), similar-looking particles were described as impact spherules in another study (Schaller et al., 2016). The tear-drop-shaped crystals from our extracts are smaller than the impact spherules described by Schaller et al. (2016), and they are neither pitted nor vesicular.

5 Discussion

5.1 A Magnetofossil Origin for the WL-A Sediment Magnetism

Our bulk magnetic hysteresis and Bcr data (Figures 2 and S1) have similar trends to previously reported values for the WL-A and WL-B cores (Kent, Lanci, et al., 2017; Lippert & Zachos, 2007). The high hysteresis squareness (Mrs/Ms) and bulk coercivity (Bcr) values are consistent with single-domain magnetite dominating the magnetic signal (Dunlop, 2002).

The central ridge and the negative region of the FORC diagrams are characteristic of uniaxial, noninteracting single-domain magnetite (Egli, 2013; Heslop et al., 2014; Ludwig et al., 2013; Newell, 2005; Roberts, Heslop, et al., 2014; Yamazaki, 20092012) or linear chains of magnetofossils that behave as uniaxial single-domain particles, as previously inferred from bulk hysteresis data from these PETM sediments (Lippert & Zachos, 2007) and in ferromagnetic resonance spectra from nearby sections (Kopp, Raub, et al., 2007; Kopp, Schumann, et al., 2009). These characteristics are also observed in the few FORC diagrams reported from the CIE onset interval from two nearby sections on the New Jersey margin (Kopp, Raub, et al., 2007; Wang, Kent, & Jackson, 2013). Our data set provides high-resolution FORC diagrams for the entire PETM interval and underscores the persistence of these features throughout the PETM interval at Wilson Lake.

The distinct FORC signatures of uniaxial single-domain magnetite in marine sediments are often interpreted to indicate abundant magnetofossils, particularly intact chains of conventional magnetofossils (Heslop et al., 2014; Roberts, Chang, et al., 2012). Similar signatures have also been attributed to isolated single-domain magnetite particles preserved in volcanic glass, igneous rocks, and detrital magnetite (Carvallo et al., 2006; Chang, Roberts, Heslop, et al., 2016; Kent, Lanci, et al., 2017; Lin, Wang, et al., 2013; Muxworthy, Evans, et al., 2013; Wang, Wang, et al., 2015). The sedimentology of WL-A sediments and TEM images of magnetic extracts (which visually document the abundance of magnetite or maghemite particles with magnetofossil morphologies) do not support these alternative sources.

Other studies suggest that abundant cuboctahedra particles are the result of impact ejecta and/or pyogenesis (Kent, Cramer, et al., 2003; Kent, Lanci, et al., 2017). The cuboctahedral morphologies observed in our extracts (1) are uniform in size and crystal structure and fall within the range for stable single-domain particles, (2) are well preserved (with no signs of dissolution), and (3) occur in large proportions, alongside other magnetofossils, throughout the PETM interval. We also note there are no environmental or stratigraphic changes that support an impact origin for the cuboctahedral particles. Such evidence would include extinctions, shocked quartz, iridium, or sudden changes in the lithology or preservation of the Marlboro Clay (e.g., the sudden occurrence of cuboctahedral particles at the onset followed by an exponential-like disappearance). Our results support the interpretation that our extracts and, by extension, the single-domain enhancement of the Marlboro Clay, are dominated by magnetofossils.

Nonzero central ridge contributions at Bc = 0 are compatible with minor contributions from isolated, nearly equidimensional single-domain particles; however, the lack of a specific endmember with an exponential-like coercivity distribution similar to that of known natural examples (Egli, 2004a; Geiss et al., 2008; Moskowitz, Frankel, & Bazylinski, 1993) means that the source of such particles covaries with EM1, EM2, and EM3, and was, therefore, subjected to the same factors that controlled magnetofossil production. The most affine endmember, EM2, has a concentration profile like that of impact products, such as microtektites; however, EM2 extends over >5 m in the CIE onset and CIE core, making it incompatible with the redistribution of an instantaneous input by sediment mixing.

Conventional magnetofossils fall within the single-domain stability range predicted for magnetite particle chains following Newell (2009) and Muxworthy & Williams (2006) (solid lines in Figure 5). Therefore, we expect all conventional magnetofossils to contribute to the noninteracting, uniaxial single-domain signature of the FORC diagrams, as long as the original chain configuration is preserved. The domain state of giant magnetofossils is more difficult to evaluate because it is not clear how they were arranged in the organisms that produced them. Giant needles and large bullet-shaped crystals fall within the stability range of isolated crystals calculated by Muxworthy & Williams (2006) (upper dashed line in Figure 5), so that they are also expected to contribute to the single-domain FORC signatures. Giant bullets, on the other hand, exceed the single-domain limit of isolated crystals but are well within the chain limit. Large and giant bullets could, therefore, have had the same magnetic navigation function as conventional magnetofossils, in which case the chain arrangement would make them also contribute to the single-domain signature. Elongated prisms are slightly less elongated than modern prismatic magnetosomal magnetite, but their dimensions are still within the single-domain limit if they are arranged in chains. Teardrops and spindles are the only crystals that clearly exceed the single-domain limit. Their predicted vortex state (Figure 5), with much lower magnetic moments, imply that these crystals do not contribute to the single-domain signature of our sediments and suggest that (bio)mineralization of these morphologies was not optimized for navigation purposes.

The FORC signatures described above, along with the coercivity distributions shown in central ridge profiles and our TEM images, bolster previous interpretations that the pronounced single-domain signature of the Marlboro Clay from the Salisbury Embayment is due to abundant and well-preserved magnetofossils rather than pyrogenic magnetite, impact spherules, or a mixture of these nonbiogenic sources (Kent, Lanci, et al., 2017; Kopp, Raub, et al., 2007; Kopp, Schumann, et al., 2009; Lippert & Zachos, 2007; Schaller et al., 2016; Schumann et al., 2008; Wang, Kent, & Jackson, 2013). Noncentral ridge contributions in our FORC diagrams are consistent with the magnetic response expected for collapsed conventional magnetofossil chains, single-domain particle clusters, detrital grains with single or multivortex domain configurations, as well as teardrops and spindle-shaped magnetofossils (Figures 6 and S2) (Chang, Harrison, et al., 2018; Egli, 2006; Harrison et al., 2018; Kobayashi et al., 2006; Li, Wu, et al., 2012; Till, Guyodo, Lagroix, Morin, Menguy, & Ona-Nguema, 2017; Till, Guyodo, Lagroix, Morin, & Ona-Nguema, 2015). Below we focus on our FORC-PCA-derived endmembers; for further discussion of magnetic signatures not associated with our endmembers, please see the supplemental text.

5.2 Unmixing the Magnetofossil Assemblage at Wilson Lake

Crystal preservation and structural integrity of fossil magnetosome chains are important prerequisites for interpreting our FORC-PCA unmixing results in terms of environmentally driven variations of magnetofossil morphologies. Reductive dissolution of conventional magnetofossils is morphology-dependent, with bullet-shaped and prismatic crystals the least and the most resistant shapes, respectively (Yamazaki, 2020). Good crystal preservation is confirmed for all morphologies by the lack of corrosion signatures (Vali & Kirschvink, 1989; Yamazaki, 2020) in our TEM images.

Preservation of intact chains cannot be inferred from our TEM analysis because the original crystal arrangement in the sediment matrix has been destroyed by magnetic extraction. Laboratory experiments with cultured MTB show that complete chain collapse removes the central ridge (Li, Wu, et al., 2012) and alters the coercivity distribution (Kobayashi et al., 2006). The presence of a central ridge contributing to ∼40% of the bulk Mrs (WL35950), which is a typical proportion for conventional magnetofossil-rich sediments (Ludwig et al., 2013), is a clear sign that such collapse does not occur. The tendency of magnetite nanoparticles to adhere onto clay platelets (Galindo-Gonzalez et al., 2009) might provide the necessary mechanical stabilization of chains once biological structures are dissolved (Shcherbakov et al., 1997), and this stabilization would prevent complete chain collapse during fossilization. We observe several clay particles in our magnetic extracts, which is not surprising given that clays are a distinguishing component of the CIE lithofacies (Gibson et al., 1993; John, Bohaty, et al., 2008).

The physical interpretation of FORC-PCA endmembers necessarily relies on the assumption that the properties of magnetic components within these endmembers remain relatively constant throughout the CIE event. For example, the coercivity of magnetosome chains depends on geometrical parameters such as particle spacing and the degree of chain bending (Berndt et al., 2020; Chang, Harrison, et al., 2018), which could vary with stratigraphic depth. Large variations of individual coercivity distribution components, however, are not expected: EM2 and EM3 have distributions and features that resemble coercivity components attributed to conventional magnetofossils in a variety of freshwater and marine sediments, which we explain below.

5.2.1 Conventional Magnetofossil Signatures

EM2 and EM3 contain two coercivity components similar to the biogenic soft and biogenic hard magnetofossil components found in magnetofossil-rich sediments (Abrajevitch & Kodama, 2009; Egli, 2004a2004b; Ludwig et al., 2013; Rodelli et al., 2019). Biogenic soft and biogenic hard components are well distinguished in lake sediments, where their coercivity distributions do not extend beyond ∼120 mT. They are less distinguishable in marine sediments, due to their broader coercivity distributions in those environments (Egli, 2004a; Ludwig et al., 2013). The biogenic soft component is usually more abundant in marine sediments (Roberts, Chang, et al., 2012) and is commonly attributed to equidimensional magnetosomes, such as those found in oligotrophic environments (Hassan et al., 2020; Jiang et al., 2020; Li, Pan, et al., 2009; Li, Wu, et al., 2012; Oda et al., 2018; Yuan et al., 2020). There is good agreement between the median field of the biogenic soft component and that of cultured MTB that produce cuboctahedral magnetite particles (Li, Pan, et al., 2009; Li, Wu, et al., 2012).

The biogenic hard magnetofossil component is attributed to elongated particles, such as those produced by MTB associated with eutrophication events (Egli, 2004a2004b). Few bacterial strains that produce elongated prisms have been isolated in culture, and no strains producing bullet-shaped crystals have been cultured successfully, so there are few direct comparisons for the biogenic hard component. Cultured MV-1 bacteria, which produce slightly elongated particles with a mean width/length value of ∼0.7 (Devouard et al., 1998), are characterized by a central ridge that peaks at ∼40 mT (Wang, Kent, & Jackson, 2013); this would contribute to the higher end of the coercivity range of the biogenic soft component or to the lower end of the coercivity range of the biogenic hard component of Egli (2004a). Bullet-shaped crystals contribute ∼8%–23% to the magnetization of our extracts. Elongated prisms contribute to <4% of the total magnetization, although their contribution to the biogenic hard component is questionable given their relatively low elongation compared to conventional prismatic magnetofossils.

The proportions of EM2 and EM3 are nearly equal at all depths within the CIE interval; the exception is between 106 and 108 m, within the CIE core, where the contribution of EM2 is greater. Given the dominance of the biogenic soft component in EM2, we conclude that the largest contribution of this coercivity component among the signature of conventional magnetosomes occurs within this interval. Furthermore, high-resolution FORC diagrams (Wagner et al., 2021) indicate that sediments from the onset interval have three clearly recognizable magnetofossil components: biogenic soft (∼10–50 mT), biogenic hard (∼50–120 mT), and biogenic needle (∼120–300 mT), as discussed in more detail below. Some of the FORC coercivity spectra from the CIE core interval also indicate that biogenic soft and hard components are distinguished in these intervals (e.g., Figures 6b and  6d–6f). Notably, FORC diagrams with greater peak signal densities (>800 Am2/T2) in the low-coercivity range are from intervals that have some of the highest percentages of EM2 and EM3 (Figures 8 and S4).

5.2.2 Giant Magnetofossil Signatures

The peak region of the coercivity distribution for EM1 overlaps with that of biogenic hard and, to an extent, some biogenic soft magnetofossils (Egli, 2004a; Wagner et al., 2021; Yamazaki & Shimono, 2013). However, the upper end of the coercivity distribution for EM1 extends well beyond that of conventional magnetofossils, signaling an additional magnetofossil contribution. Micromagnetic simulations of giant magnetofossils by Chang, Roberts, Williams, et al. (2012) suggest that giant, needle-like magnetofossils may have single-domain behavior. Micromagnetic simulations of the giant needles observed in our TEM images indicate that these particles are capable of exhibiting single-domain behavior and may have coercivities up to 300 mT (Wagner et al., 2021).

The occurrence of needles with a range of elongations, as indicated in Figure 5, together with additional contributions from the coercivity components of EM2 and EM3, can explain the wide coercivity distribution of EM1. FORC diagrams from our specimens with larger proportions of EM1 also have contributions that exceed the Bc = 140 mT absolute upper limit of conventional magnetofossil components (Egli, 2004a). This observation supports our interpretation that EM1 indicates the presence of giant needles (e.g., Figures 6 and S2). This interpretation is further supported by TEM observations: specimens with larger proportions of EM1 (WL35900 and WL35800) contain giant magnetofossils, including needles, whereas specimens with larger proportions of EM2 and/or EM3 (WL35245) contain no or few giant magnetofossils in our TEM micrographs (e.g., Figure 3). Our magnetization calculations suggest that extracted giant needles may account for ∼2% of the single-domain moment of some stratigraphic intervals (Table S1 and S2), compared to ∼6% central ridge contributions above Bc = 140 mT, relative to Mrs (WL35950). These calculations are based on just a few needles observed in our extracts; there are likely more needles (and other giant morphologies) than those observed in magnetic extracts, but they may not dislodge from clay particles as easily as smaller magnetofossils due to the larger contact area. We randomly selected TEM fields of view from each specimen to characterize the entire suite of iron oxide morphologies to minimize bias; that is, we did not search specifically for giant morphologies. Detailed magnetic analyses and micromagnetic simulations indicate that the coercivity distribution of giant needles peaks at ∼210 mT (Wagner et al., 2021). The unimodal nature of EM1 does not enable us to discriminate between the contributions of giant needles and other magnetic components over the coercivity range of conventional magnetofossils (i.e., up to 140 mT); however, a detailed analysis of the vertical offset of the central ridge, which is controlled by thermal relaxation processes that depend on the particle volume, suggests that the coercivity distribution of giant needles is mainly between ∼140 and ∼240 mT (Wagner et al., 2021).

5.3 Paleoecologic and Paleoenvironmental Change Recorded by Magnetofossils

The variety in both conventional and giant magnetofossil morphologies (i.e., disparity) at Wilson Lake highlights the coexistence of MTB and unidentified giant iron-biomineralizing organisms (GIBOs) during the PETM. Our FORC-PCA-derived endmembers allow us to look at changes in the abundance of these magnetofossil populations, providing information that might encode simultaneous responses by these organisms to local environmental changes during the PETM. Below, we provide a possible explanation for the emergence of GIBOs. We also use the observed high abundance and high morphological disparity of magnetofossils at Wilson Lake to explain possible changes to coastal marine water chemistry during the PETM. These associations, in combination with changes in our FORC-PCA-derived endmembers, allow us to predict how and why populations of environmentally sensitive organisms changed during the PETM.

5.3.1 Environmental Significance of Conventional Magnetofossils

TEM images of our magnetic extracts from WL-A showcase high morphological disparity of biogenic magnetite; several of these morphologies are similar to known taxa (Pósfai et al., 2013). Magnetofossils similar to magnetosomal magnetite produced by bacteria in the Proteobacteria, Nitrospirae, and candidate division OP3 phyla (potentially also from the Omnitrophica, Latescibacteria, and Planctomycetes phyla) suggest that microaerobic conditions were present for the duration of the CIE at the WL-A locality (Lefèvre, Viloria, et al., 2012; Lin, Zhang, et al., 2020; Mann, Frankel, & Blakemore, 1984; Pósfai et al., 2013).

A recent study by Li, Menguy, et al. (2020) suggests that some of the observed magnetofossils, which are the most dominant category in extract WL35245, might include immature bullets. The clear gap between the size and shape distribution of cuboctahedra and bullets that we measured from the magnetic extracts (Figure 5) does not support the presence of large proportions of immature bullet-shaped crystals. Rather, the uniaxial growth mechanism illustrated by Li, Menguy, et al. (2020) would populate the parameter space between the clusters formed by the two morphologies.

If we exclude immature crystals, then cuboctahedra are the most abundant morphology and contribute ∼37%–57% of the magnetofossils in each extract. In modern environments, these particle morphologies are most often associated with MTB living in oligotrophic environments, including ferromanganese crusts, the deep ocean, and oxygenated environments (Petermann & Bleil, 1993; Yamazaki & Shimono, 2013; Yuan et al., 2020). Importantly, however, abundant bullet morphologies in our extracts indicate that the environment at WL-A during the PETM was not oligotrophic: organisms that produce these particle morphologies are correlated with increased organic matter, eutrophication, and nutrient cycling (Egli, 2004a2004b; Moskowitz, Bazylinski, et al., 2008; Rivas-Lamelo et al., 2017; Yamazaki & Kawahata, 1998).

Environmental control of morphological disparity in MTB living in sediment and the water column might be fundamentally different and suggests another mechanism by which cuboctahedral forms flourished. For example, MTB living in the water column of a stratified estuary investigated by Chen et al. (2014) produce equidimensional magnetite and greigite particles that contribute exclusively to the biogenic soft coercivity component. Therefore, the magnetofossil assemblage described here may reflect sub-populations of iron-biomineralizing organisms living in distinct vertical positions of the coastal marine redox environment, or during distinct seasons, which were integrated during the taphonomic process. This is supported by the overall large magnetofossil abundance we observe throughout the PETM interval at Wilson Lake and, particularly, the large EM2 peak at the beginning of the CIE core, which is dominated by biogenic soft contributions (Figure 9). We interpret intervals that contain the largest EM2 proportions as intervals that experienced sustained water column stratification, specifically, a well-developed oxic-anoxic interface (OAI) just above the sediment-water interface. Benthic foraminiferal species associations described by Stassen et al. (2015) support our interpretation: the same interval that is dominated by EM2, from 106 to 108 m, corresponds to their association IIa, which contains stress-tolerant species that can withstand long-term water column stratification (Figure S8).

Details are in the caption following the image

Conceptual environmental model for the PETM based on our results: (a) bulk carbon isotopes, (b) Endmember 1 (EM1) interpreted as the biogenic needle (BN) component (giant magnetofossils), (c) inferred weathering events from (a and b), (d) Endmember 2 (EM2) interpreted as the biogenic soft (BS) component (conventional magnetofossils), and (e) Endmember 3 (EM3) interpreted as the biogenic hard (BH) component (conventional magnetofossils). Carbon isotopes, EM1, EM2, and EM3 are re-drawn schematically from Figure 8. Here we interpret δ13C enrichment and EM1 peaks as weathering events, associated with increased fluvial iron influx, that caused bottom water deoxygenation through upward shoaling and expansion of the oxic-anoxic interface (OAI). We interpret peaks in EM2 as intervals that experienced long-term stratification and had a thinner, yet stable OAI. Intervals with EM3 peaks are interpreted as corresponding to thinner, seasonally developed OAIs, likely caused by seasonal eutrophication and higher organic matter supply. Changes in relative stratification, and therefore oxygenation, are highlighted from light to dark gray.

The ∼19 cm/kyr sedimentation rate at WL-A implies that our samples are averages over multiple seasons, perhaps on the order of hundreds of years. Given the neritic depositional environment of these marine sediments, conditions at the Wilson Lake locality could have alternated between oligotrophic and eutrophic conditions during a single season. Seasonal algal blooms would have provided the organic matter necessary to raise the OAI into the water column (Gibbs, Bown, et al., 2006; Gibbs, Bralower, et al., 2006; Kopp, Schumann, et al., 2009; Sluijs & Brinkhuis, 2009), thus sustaining abundant MTB populations year-round. We interpret intervals in which the EM3 contribution, the endmember dominated by biogenic hard contributions, is largest to indicate times when stratification and nutrient supply were caused by sustained, but seasonal organic matter supply (by marine algae) over hundreds to thousands of years (Figure 9). Magnetobacterium bavaricum, an MTB that produces elongated magnetosomal particles, is known to perform redoxtaxis (shuttling across the OAI) and accumulates large quantities of elemental sulfur (Mao et al., 2014; Spring et al., 2000). This MTB can respond to sudden OAI shifts, while other MTB (e.g., cocci that produce more equant particles) cannot. This finding suggests that periodic stratification, or a more mobile OAI, favors MTB that produce elongated particles, which would contribute toward producing higher EM3 proportions. Our interpretation is further supported by benthic foraminifera association IIb of Stassen et al. (2015), which corresponds to periodic stratification over the same interval as EM3, that is, the upper half of the PETM (Figure S8). The lack of greigite might indicate that reducing conditions below the OAI never supported a massive, prolonged sulfate reduction regime, which would have led to the dissolution of single-domain magnetite particles, including magnetofossils (Leslie et al., 1990).

5.3.2 On the Origin and Environmental Significance of Giant Magnetofossils

Schumann et al. (2008) first reported giant magnetofossils in PETM sediments from Ancora, New Jersey, a section approximately 23 km east of WL-A (Figure 1). Chang, Roberts, Williams, et al. (2012) identified giant magnetofossils in deep marine sediments deposited before, during, and after the PETM from the Southern Ocean and equatorial Indian Ocean. Giant bullets were also found in Middle Eocene Climatic Optimum sediments (Chang, Roberts, Williams, et al., 2012). Chang, Harrison, et al. (2018) suggested that giant, needle-shaped magnetofossils in PETM sediments from Walvis Ridge, South Atlantic Ocean, are linked to changes in oxygenation and nutrient (iron) supply caused by rapid climate change. Chang, Roberts, Williams, et al. (2012) and Schumann et al. (2008) proposed that giant magnetofossil spindles, bullets, and spearheads may have been used for defense or were part of a feeding apparatus such as the mollusk radula. Another potential function of these magnetofossils could be their use as metabolic batteries. The ambiguous origin of giant magnetofossils and lack of modern analogs makes it challenging to use them as a proxy for environmental change. We use knowledge of extant organisms we think are most similar to GIBOs to understand their potential origin and function. We have detected all the giant magnetofossil morphologies identified by Schumann et al. (2008), except for spearheads.

Some protists contain magnetosomal crystals that they acquired either by ingesting MTB or biomineralizing them for magnetotaxis (Bazylinski, Lefèvre, & Frankel, 2012; Bazylinski, Schlezinger, et al., 2000; Leão et al., 2020; Martins et al., 2007; Monteil et al., 2019; Simmons & Edwards, 2007; Torres de Araujo et al., 1986). For example, Torres de Araujo et al. (1986) identified an alga in modern brackish, microaerobic mud that contains chains of bullet-shaped magnetite (80–180 nm long and 40–50 nm wide); these shapes and sizes are similar to the medium and large bullets in our samples (Figure 3 and Table S4). Other magnetite-containing protists were identified in Salt Pond, a seasonally stratified semi-anoxic eutrophic basin in Woods Hole, Massachusetts. These protists live at different depths and within different chemical gradients, including the OAI, but most are found in the anoxic zone (Bazylinski, Schlezinger, et al., 2000). Simmons and Edwards (2007) showed that these populations change seasonally in response to water column chemistry, specifically oxygen and iron concentrations (Bazylinski, Lefèvre, & Frankel, 2012; Moskowitz, Bazylinski, et al., 2008). These observations support the interpretation that giant magnetofossils, which are too large to be made by bacteria, were biomineralized by protists.

Protists are important in iron-cycling and may help generate bioavailable iron for other species and, therefore, may contribute to the high ferrous iron concentrations at the OAI (Barbeau et al., 1996; Bazylinski, Lefèvre, & Frankel, 2012; Martins et al., 2007; Pernthaler, 2005). If giant magnetofossils are a proxy for similar protists, then their existence during the PETM may be related to high iron levels, or perhaps development and expansion of a ferruginous zone. Such high levels of dissolved iron would require concomitant zones of low oxygen (Canfield & Thamdrup, 2009). Rates of physical and chemical weathering increased at the onset of the CIE and, therefore, bioavailable iron also may have increased (Carmichael et al., 2017; Kiehl et al., 2018; McInerney & Wing, 2011). For example, kaolinite-rich Cretaceous laterite deposits, which have high iron and aluminum concentrations, were eroded from the eastern North American continent and deposited onto the New Jersey shelf, including at WL-A (John, Banerjee, et al., 2012; Kopp, Schumann, et al., 2009). Given the distal location of WL-A to the paleoshoreline (Figure 1), the iron from these laterites was likely dissolved and bioavailable.

The largest proportions of EM1 to the total magnetic signal occur immediately after the PETM onset, where there is evidence of a marine transgression and increased riverine input (Sluijs & Brinkhuis, 2009; Stassen et al., 2015), and just before the first unconformity, which signals another erosional event at Wilson Lake. We interpret EM1 peaks (giant magnetofossils) to coincide with changes in iron supply and precipitation across the Salisbury Embayment. The most prominent EM1 peaks correspond to bulk δ13C peaks, which may signal productivity or nutrient supply changes, further supporting this interpretation (Figure 9). We argue that the inferred dependence of giant magnetofossils on increased bioavailable iron concentrations favors the interpretation that giant magnetofossils were produced by heterotrophic protists shuttling between pools of bioavailable iron and organic carbon. The shuttling hypothesis is further supported by a recent study that suggests that some MTB may have adapted magnetotaxis to move across the OAI for sulfur, carbon, nitrogen, iron, and other elemental cycling (Li, Liu, Wang, et al., 2020).

The rapid freshwater influx at the onset, and just before the first unconformity, could have created two potential scenarios. In the first, freshwater influx produced vertical water column mixing at Wilson Lake. This would have caused the OAI to de-stratify, resulting in a “patchy” OAI along the sediment-water interface, isolated oxygen-minimum pools within the water column, or an OAI that was just below the sediment-water interface. This scenario suggests that giant magnetofossils were produced by organisms that did not necessarily depend on a well-stratified OAI and were more dependent on large pools of nutrients (namely iron) coincident with localized oxygen minimum zones. Biomineralization of giant magnetofossils, therefore, would have been selected for horizontal or vertical navigation between changing nutrient pools. This hypothesis is best supported by abundant epibenthic foraminifera species that prefer more oxygenated environments during the same intervals where we observe EM1 peaks (Stassen et al., 2015; Figure S8).

The second scenario involves OAI shoaling and expansion, largely due to density differences within the water column. In this scenario, the OAI would become well-stratified and rise anywhere between a few centimeters to a few meters above the sediment-water interface. The result would be an overall seafloor deoxygenation. The absence of greigite magnetofossils and of a benthic foraminiferal extinction indicate that the water column remained at least microaerobic. In this scenario, giant magnetofossils would have been produced by organisms that were not only dependent on large pools of bioavailable iron, but also on a larger OAI diffusing out of the underlying sediment. A larger seafloor microaerobic zone would have selected for organisms that biomineralized giant magnetic particles to navigate within, or above, this expanded ecological niche. Abundant bioavailable nutrients would have decreased the pressure to biomineralize the most efficient magnetic particles for navigation, which provides an explanation for the variety of giant magnetofossil morphologies with different magnetic domain states (Figure 5). Conventional magnetofossils are also present within the same sediments dominated by EM1 proportions, which we interpret to favor a well-stratified OAI over a patchy OAI. The coexistence of epibenthic foraminifera (Figure S8) does not necessarily contradict this scenario: these foraminifera are mobile organisms that can adapt to life within the water column, and some are found within microaerobic environments (Stassen et al., 2015). We favor this second scenario which requires a combination of increased nutrient supply (namely iron) and an expanded niche (the OAI). This scenario supports the interpretation of Chang, Harrison, et al. (2018) that giant needles are related to deoxygenation.

In summary, our results suggest that magnetofossil disparity changes correspond to changes in water column oxygen concentrations at Wilson Lake. In our model, the Wilson Lake environment fluctuated between three general states: (1) a thicker OAI that moved upward within the water column, (2) a thinner but sustained, well-developed OAI, and (3) a seasonally controlled OAI (Figure 9). Each condition would have promoted biomineralization of distinct magnetofossil populations produced by organisms that prefer different environmental conditions. Enhanced surface water productivity would have supplied abundant bioavailable carbon and other nutrients for the PETM duration (e.g., Gibbs, Bown, et al., 2006; Gibbs, Bralower, et al., 2006). In summary, we interpret intervals in which the EM1 contribution—a proxy for giant magnetofossils—increases to correspond to times when OAI expansion and shoaling were driven by terrestrial weathering and erosional events that provided abundant dissolved iron to the continental shelf. We interpret intervals dominated by EM2—a proxy for biogenic soft magnetofossils—to correspond to times when the hydrologic cycle remained relatively constant, but still active, and the OAI was well-defined just above the sediment-water interface. Lastly, we interpret intervals dominated by EM3—a proxy for biogenic hard magnetofossils—as times when the formation of the OAI above the sediment-water interface was seasonally induced; this may have been linked to more of the baseline environmental changes associated with the PETM.

6 Conclusions

  1. We show that FORC-PCA can differentiate between distinct populations of conventional and giant needle-shaped magnetofossils in the WL-A core. This approach can be used in other magnetofossil-bearing marine and lacustrine records

  2. The characteristic single-domain magnetic signature at Wilson Lake is the result of abundant magnetofossils with a variety of crystal morphologies and sizes (i.e., high disparity). FORC diagrams of sediments from the WL-A core are characterized by the typical signature of noninteracting single-domain magnetite. This signature is due to three general magnetofossil assemblages: biogenic soft, biogenic hard, and biogenic needles

  3. We hypothesize that during the CIE interval there was an overall increase in seasonally produced organic matter, which in turn stimulated chemical water column stratification at Wilson Lake and other nearby Salisbury Embayment locations. This environment supported prolific magnetotactic bacteria communities

  4. We interpret giant magnetofossils to record a biological response to additional environmental pressures. We propose that giant magnetofossils are associated with iron-biomineralizing protists that exploited abundant bioavailable iron and niche expansion. An oxic-anoxic interface that expanded into the water column would provide distinct advantages for iron-biomineralizing protists: enhanced bioavailable iron and abundant organic matter would lower the selective pressure for the most efficient use of iron in magnetotaxis-supporting structures, enabling biomineralization of larger crystals with less-optimized magnetic properties. These environmental responses may have been spurred by elevated sea surface temperatures and average global temperatures, as well as increased precipitation and freshwater runoff that caused enhanced weathering of iron-rich laterites and clays and delivery of other nutrients from land to the continental margin. The distinct magnetic signature of giant needle-shaped magnetofossils in FORC diagrams can be used to identify giant magnetofossils and the abovementioned environmental pressures, both spatially and temporally, in response to rapid global change events

  5. We conclude that the New Jersey shelf experienced immediate deoxygenation and eutrophication via dissolved iron loading at the PETM onset, in contrast to deep marine records which indicate a delayed decrease in oxygen levels. The remainder of the PETM interval at Wilson Lake is characterized by either a long-term or seasonally stratified water column that was punctuated occasionally by iron loading events like that at the PETM onset


The authors thank Andrew Roberts and an anonymous reviewer for their helpful reviews of a previous version of this manuscript. This research used samples collected by the United States Geological Survey, in part collected by Peter Lippert and provided by Jim Zachos (UC Santa Cruz) and Ellen Thomas (Yale University). Financial support was provided by a Robert Hevey and Constance M. Filling Fellowship at the Smithsonian Institution, the N. Gary Lane Student Research Award through the Paleontological Society, an Evolving Earth Foundation Research Grant, a P.E.O. Scholar Award, and a Schlanger Ocean Drilling Fellowship, all to Courtney Wagner. Ioan Lascu is grateful for a Smithsonian Institution Edward and Helen Hintz Secretarial Scholarship and a National Museum of Natural History Research Grant.

    Conflicts of Interest

    The authors declare no conflicts of interest relevant to this study.

    Data Availability Statement

    The authors use data from Zachos, Schouten, et al. (2006) and Lippert and Zachos (2007) for Figures 2 and S1; we use data from Stassen et al. (2015) for Figure S8. Supporting materials are available in a separate document and Excel file uploaded with this submission. Data files are uploaded to figshare.com (https://doi.org/10.6084/m9.figshare.13182941.v1). Results from the companion manuscript are also available on figshare.com (https://doi.org/10.6084/m9.figshare.13182848.v3). Alternatively, the corresponding author may be contacted for files, directly.