A deep-learning automated image recognition method for measuring pore patterns in closely related bolivinids and calibration for quantitative nitrate paleo-reconstructions

Vol.:(0123456789)

Scientic Reports | (2023) 13:19628

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports

A deep‑learning automated image

recognition method for measuring

pore patterns in closely related

bolivinids and calibration

for quantitative nitrate

paleo‑reconstructions

Anjaly Govindankutty Menon

, Catherine V. Davis

, Dirk Nürnberg

, Hidetaka Nomaki

Iines Salonen

4,5

, Gerhard Schmiedl

1,6

& Nicolaas Glock

Eutrophication is accelerating the recent expansion of oxygen‑depleted coastal marine environments.

Several bolivinid foraminifera are abundant in these oxygen‑depleted settings, and take up nitrate

through the pores in their shells for denitrication. This makes their pore density a possible nitrate

proxy. This study documents three aspects related to the porosity of bolivinids. 1. A new automated

image analysis technique to determine the number of pores in bolivinids is tested. 2. The pore patterns

of Bolivina spissa from ve dierent ocean settings are analysed. The relationship between porosity,

pore density and mean pore size signicantly diers between the studied locations. Their porosity is

mainly controlled by the size of the pores at the Gulf of Guayaquil (Peru), but by the number of pores

at other studied locations. This might be related to the presence of a dierent cryptic Bolivina species

in the Gulf of Guayaquil. 3. The pore densities of closely related bolivinids in core‑top samples are

calibrated as a bottom‑water nitrate proxy. Bolivina spissa and Bolivina subadvena showed the same

correlation between pore density and bottom‑water nitrate concentrations, while the pore density of

Bolivina argentea and Bolivina subadvena accumeata is much higher.

Oceanic oxygen concentrations are predicted to decrease globally aecting the stability of marine ecosystems

1–4

Global warming accelerates ongoing ocean deoxygenation

5,6

, and expansion of oxygen minimum zones

(OMZs)

1,2,7

. Increased ocean warming enhances upper-ocean stratication

, reduces ventilation, and has impli-

cations for biological productivity

as well as carbon, nitrogen

and phosphorus cycling

in the oceans. ese

processes are amplied by the large-scale use of chemical nitrogenous fertilizers to satisfy global demand for

food production which drastically disrupts the nitrogen cycle

11,12

. Oxygen is a major inuence on the marine

nitrogen cycle in the global oceans

as some microbial processes require oxygen while others are inhibited by it

When oxygen concentrations drop below ~ 4.5µmol/kg, nitrate becomes the major electron acceptor for respira-

tion replacing oxygen, a condition called suboxic

13–15

. e continued expansion of suboxia results in the loss of

xed nitrogen via denitrication

14,16

, a dissimilatory process in which nitrate (NO

) is ultimately converted into

dinitrogen gas

. erefore, denitrication reduces the supply of NO

in global oceans

14,16

. Nitrogen xation,

nitrication, and denitrication are major processes in the nitrogen cycle that are mainly facilitated by bacteria

while lower oxygen concentrations can either enhance or inhibit these processes

. erefore, the nitrogen cycling

OPEN

Department of Earth System Sciences, Institute for Geology, Universität Hamburg, Bundesstrasse 55,

20146 Hamburg, Germany.

Department of Marine, Earth, and Atmospheric Sciences, North Carolina State

University, 2800 Faucette Dr, Raleigh, NC 27695, USA.

GEOMAR Helmholtz Centre for Ocean Research Kiel,

Wischhofstr. 1-3, Geb. 8c, Raum 106, 24148 Kiel, Germany.

SUGAR, X-star, Japan Agency for Marine-Earth

Science and Technology (JAMSTEC), 2-15 Natsushima-cho, Yokosuka 237-0061, Japan.

Present address:

Tvärminne Zoological Station, Faculty of Biological and Environmental Sciences, University of Helsinki, Hanko,

Finland.

Center for Earth System Research and Sustainability, Institute for Geology, Universität Hamburg,

Bundesstrasse 55, 20146 Hamburg, Germany.

email: anjaly.govindankutty[email protected]

Vol:.(1234567890)

Scientic Reports

| (2023) 13:19628 |

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports/

in OMZs is dierent from the rest of the open ocean

. Approximately 30–50% of xed nitrogen loss in the world’s

oceans occurs in oxygen minimum and decient zones

. Quantitative paleo-reconstruction of nitrate levels could

provide a comprehensive understanding of how the dierent processes mentioned above interacted in the past.

is will help us to predict future changes in marine nutrient budgets and possible impacts of eutrophication.

Foraminifera are a group of amoeboid protists that are abundant in marine environments

, and account for

a major part of benthic denitrication in the OMZs

20–22

. Many calcareous foraminiferal tests (shells) are porous.

e pores in benthic foraminiferal tests play an important role in facilitating gas exchange and osmoregulation

between the foraminifera and the environment

. e pore density (number of pores per unit area), mean pore

size (average pore sizes of one individual), and shape of pores are important morphological features that vary

among dierent taxa

24–26

. e porosity (% of the area of the tests occupied by the pores), and pore density of

foraminifera are likely driven by environmental factors. Factors that have been suggested include latitude, water

density

27–30

, temperature, salinity

, oxygen, and nitrate concentrations

24,32–34

. Porosity might also be genetically

encoded

25,35

. Porosity is a species-specic trait that can be used to distinguish certain pseudocryptic species

such as Ammonia spp.

. Nevertheless, within a single species phenotypic plasticity exists. us, porosity can be

inuenced by environmental conditions, and hence used as a paleoproxy.

Porosity in benthic foraminifera plays an important role in adaptation strategies by facilitating gas exchange

through larger pore areas in low oxic conditions

24,32,36

. Cell organelles involved in respiration (i.e. mitochon-

dria) are more abundant around the inner pore surfaces of species living in oxygen-depleted conditions than in

well-oxygenated conditions

. In some foraminiferal species, increased gas exchange can be attained by either

increasing the number of pores or by increasing the surface area of the test (or shell)

. However, the function of

pores may vary among species because of their dierence in evolutionary history

e shallow oxygen minimum zones of the Eastern Pacic have large standing stocks of benthic foraminif-

eral species

. Several benthic foraminiferal species living in oxygen-depleted environments perform complete

denitrication, which is rare amongst eukaryotes

. Denitrication is the preferred respiration pathway in sev-

eral foraminiferal species from oxygen-depleted environments, making these eukaryotes an important part of

benthic nitrogen cycling in some environments

. Previously, it has been found that benthic foraminifera living

in oxygen- or nitrate-depleted environments have higher pore density and porosity than those living under well-

oxygenated conditions or high ambient nitrate concentrations

24,32,33

. erefore, pore parameters of fossil shells

are promising proxies for paleo oxygen and nitrate concentrations.

We determined pore parameters mean pore size, pore density, and porosity of the shallow infaunal species

Bolivina spissa (see Fig.1). Many bolivinids have an anity for low-oxygen environments

. Bolivina spissa is

well adapted to low oxygen conditions

32,43

, and has the ability to denitrify

, which makes it a promising species

that might facilitate quantitative NO

reconstructions.

We used foraminiferal specimens retrieved from ve oxygen-depleted locations around the Pacic: the Gulf

of Guayaquil (core M77/2-59-01), the Mexican Margin (core MAZ-1E-04), the Sea of Okhotsk (core MD01-

2415), and “core-top” (i.e. surface-sediment) samples from Sagami Bay, and the continental margin of Costa Rica

(Quepos Slide, core SO206-43-MUC) (Fig.2). Here, we present a non-destructive, fast and statistically robust

method for quantitatively describing the morphometrics in benthic foraminiferal tests. We applied an automated

image recognition technique on scanning electron microscope (SEM) images using a deep learning algorithm

to analyse the morphological features of B. spissa. Deep learning is a type of machine learning which is used to

identify objects in images and allows to process data in a way according to user’s interest

Figure1. (a) Scanning electron microscopic images of B. spissa collected from Mexican Margin (MAZ-1E-04),

(water depth: 1463m) and (b) their total area relative to rst (oldest) ~ ten chambers within 50,000–70,000 μm

measured using ZEN lite soware.

Vol.:(0123456789)

Scientic Reports

| (2023) 13:19628 |

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports/

We studied (1) the interdependence between the pore density, porosity and mean pore size of B. spissa to

demonstrate whether total porosity is mainly inuenced by the number or the size of pores and (2) whether

porosity or pore density can be used as a robust proxy for bottom-water nitrate [NO

–

]

reconstructions.

Finally, we compare the pore density between B. spissa, Bolivina subadvena, Bolivina subadvena accumeata, and

Bolivina argentea, and provide an extended nitrate vs. pore density calibration for B. spissa and B. subadvena

from dierent locations around the Pacic.

Results

Comparison between manual and automated pore density analyses

Pore density measurements showed a 0–20% dierence between manual and automated methods with an average

individual dierence at 4.2%. ere was no signicant dierence in the mean pore density of all 31 specimens

between the manual (0.0059 ± 0.0002 P µm

–2

; 1 SEM) and the automated (0.0059 ± 0.0002 P µm

–2

; 1 SEM) image

analyses (T-test, p = 0.99). In three out of 31 cases the dierence was 0% and the algorithm was counting exactly

the same number of pores that have been recognized manually (Supplementary TableST1). Only two specimens

of B. subadvena showed a relatively high oset (10% and 20%). e original training of the algorithm is based

on B. spissa. For future studies, which include a closer analysis of other species, we recommend an individual

training for each species.

Automated pore measurements with and without manual corrections

ere was no signicant dierence for porosity (t = 0.31, p = 0.75) and pore density (t = 0.58, p = 0.56) obtained

through automated image analysis with and without manual corrections, where artefacts of the automated image

analyses were manually removed (Supplementary TablesST2 and ST3).

Interdependence between pore parameters of B. spissa.

e overall porosity values of all locations varied between 2.66% and 16.03% with a mean (± SD) of 8.52%

(± 2.14%). e mean pore size varied between 5.98 µm

and 47.62 µm

with a mean (± SD) of 17.83 µm

(± 3.83

µm

). e overall pore density varied between 0.002 P/µm

–2

to 0.009 P/µm

–2

with a mean (± SD) of 0.004 P/

µm

–2

(± 0.001 P/µm

–2

Specimens of B. spissa from the Gulf of Guayaquil, (M77/2-59-01) had the lowest porosity (7.14% ± 1.62%)

and mean pore size (17.13 µm

± 4.37 µm

) of all analysed locations. e specimens from the Sea of Okhotsk,

(MD01-2415) had the highest porosity (10.83% ± 1.66%) and mean pore size (20.67 µm

± 3.54 µm

). e mean

pore density was not signicantly dierent for the core-top samples (Costa Rica and Sagami Bay) and the down

core samples from the Mexican Margin (MAZ-1E-04) and the Sea of Okhotsk. e pore density at the Gulf of

Guayaquil (0.0043 P/µm

± 0.0008 P/µm

) was lower than at the other locations (Supplementary TableST4).

In general, there was a signicant linear correlation between mean pore size and porosity (coecient of deter-

mination, R

= 0.27, p = 3.19E-93, Fig.3a; Supplementary TableST5) for all the analysed specimens. We observed

strong regional dierences in R

among the studied sites. e R

was highest for the specimens from the Gulf

of Guayaquil (R

= 0.45, p = 5.91E-89, Fig.3a), and lowest for the specimens from core-top samples (R

= 0.05,

p = 0.047, Fig.3a). We found a signicant linear correlation between porosity and pore density (R

= 0.42,

p = 1.36E-15, Fig.3b; Supplementary TableST6) among all the sampling locations with the highest R

of 0.45 at

the Mexican Margin, while the specimens from the Gulf of Guayaquil showed the weakest correlation (R

= 0.1,

p = 3.21E-17, Fig.3b) between porosity and pore density.

Figure2. Map showing site locations studied: Gulf of Guayaquil (M77/2-59-01, depth: 997m), Mexican

Margin (MAZ-1E-04, depth: 1463m), Sea of Okhotsk (MD01-2415, depth: 822m), core-top samples from

Costa Rica (Quepos Slide, SO206-43-MUC, depth: 568m), and Sagami Bay (Japan, depth: 1410m). e map

was produced using Ocean Data View (Schlitzer, R., Ocean Data View, odv.awi.de, 2017).

Vol:.(1234567890)

Scientic Reports

| (2023) 13:19628 |

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports/

All analysed specimens showed a signicant but weak negative linear correlation between pore density and

mean pore size (R

= 0.085, p = 1.34E-27, Fig.3c; Supplementary TableST7). We found a higher R

for the core-

top samples collected from Costa Rica and Sagami Bay (R

= 0.4, p = 7.18E-10, Fig.3c), and the weakest for the

samples from the Gulf of Guayaquil (R

= 0.20, p = 4.52E-35, Fig.3c).

e combined data from all studied locations clearly fall apart into two distinguishable clusters for both

porosity and pore density: “Cluster 1” (black dashed circle Fig.3a), grouped most of the specimens belonging

to the Gulf of Guayaquil (n = 669), and “Cluster 2” (red dashed circle, Fig.3a), consisted of specimens belong-

ing to the Mexican Margin (n = 445), the Sea of Okhotsk (144), and the core-top samples (n = 76). e porosity

was signicantly dierent between Cluster 1 and Cluster 2 (W = 50,716; p < 2.2e -16). is also accounts for the

pore density (W = 79,726, p < 2.2e-16) and the mean pore size (W = 170,008; p = 4.49e-15). All data have been

included in the Supplementary TableST8.

Figure3. Relationship between (a) porosity vs mean pore size (b) porosity vs pore density (c) pore density vs

mean pore size of B. spissa specimens from Gulf of Guayaquil (M77/2-59-01), Mexican Margin (MAZ-1E-04),

Sea of Okhotsk (MD01-2415), and the core-top samples (Sagami Bay and Costa Rica). Total number of

specimens utilized, n = 1344.

Vol.:(0123456789)

Scientic Reports

| (2023) 13:19628 |

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports/

Inter‑species comparison of pore parameters and pore density vs [NO

–

]

calibration in the

core‑top samples

While core-top specimens of B. spissa and B. subadvena from Costa Rica (Quepos Slide), and Sagami Bay (Japan)

had a very similar pore density, pore densities of B. subadvena accumeata and B. argentea were around 50–300%

higher (Fig.4; Supplementary TableST9). e new data for B. spissa and B. subadvena from Quepos Slide and

Sagami Bay t well into the pore density correlation with [NO

–

]

that has been found for B. spissa from the

Peruvian OMZ

(Fig.4). ere was a highly signicant linear correlation between the pore density of B. spissa

and B. subadvena from Peru, Costa Rica, and Sagami Bay (R

= 0.93, p < 0.0001, Fig.4b). e data of B. subadvena

accumeata and B. argentea were oset from this linear regression (Fig.4a). e relationships between the pore

density of B. spissa and B. subadvena from core-top samples (Costa Rica and Sagami Bay) and bottom-water

oxygen (R

= 0.43, p = 0.028; Supplementary Fig.SF1), temperature (R

= 0.50, p = 0.015; Supplementary Fig.SF2),

salinity (R

= 0.41, p = 0.035; Supplementary Fig.SF3), and water depth (R

= 0.48, p = 0.018; Supplementary

Fig.SF4) has been analysed to test, if nitrate is the main factor that controls the pore density. ese correlations

are signicant (R

varies between 0.41 and 0.50; P varies between 0.015 and 0.035) but clearly weaker than the

correlation of the pore density to nitrate (R

= 0.93, p = 1.4E-6; Fig.4b). e data for bottom-water oxygen,

temperature, salinity and water depth from core-top samples have been included in Supplementary TableST10.

Since pores were manually counted for the core-top pore density dataset o Peru from Glock etal.

, no data

was available for the porosity of these specimens. A comparison of the porosity in tests of core-top samples of

B. spissa from Costa Rica (9.5% ± 0.2%; 1SEM; N = 39) and Sagami Bay (9.1% ± 0.2%; 1SEM; N = 37) showed no

signicant dierence between these two locations (p = 0.25). e Costa Rica [NO

–

]

was lower and there was

Figure4. Correlation between the mean pore density of dierent closely related Bolivina species from core-top

samples and [NO

–

]

. If no species name is indicated, the analysed species was B. spissa. e specimens of B.

subadvena, B. subadvena accumeata and B. argentea are all from location SO206-43-MUC o Costa Rica, except

the one specimen of B. subadvena at ~ 42μmol/kg [NO

–

]

that was collected at Sagami Bay (Japan). e linear

t (all data) has been applied to all available data for B. spissa and B. subadvena, except B. subadvena accumeata

and B. argentea. (a) Pore density vs [NO

–

]

plot including all analysed Bolivina species. (b) Pore density vs

[NO

–

]

plot only including B. spissa and B. argentea. e linear t (Peru) alone was the published correlation

from Glock etal.

32,45

and only included B. spissa collected o Peru. Error bars are the standard error of the mean

(1SEM).

Vol:.(1234567890)

Scientic Reports

| (2023) 13:19628 |

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports/

a signicant dierence in the pore density between these two locations (p = 8.7E-5, Fig.4). is indicated that

the pore density of B. spissa might be more sensitive to changes in the [NO

–

]

than the porosity. In addition,

while the pore density of B. subadvena t very well into the pore density-[NO

–

]

correlation of B. spissa (see

Fig.4), the porosity of B. subadvena was signicantly higher than the porosity of B. spissa (10.9% ± 0.5% for B.

subadvena vs. 9.5% ± 0.2% for B. spissa from Costa Rica; p = 0.0002).

Discussion

Evaluation of the automatic image recognition technique

Our study tested the application of a newly developed automated image recognition method for the detection of

pore parameters of the benthic foraminiferal species B. spissa. is method can be used to accurately measure

pore parameters such as the mean pore size, porosity, and pore density of B. spissa. is allows a high and ecient

sample throughput (less than 1min for one specimen) compared to manual analysis (5–6min for one specimen)

of pores. is automated deep learning approach produces results statistically identical to manual analyses. No

signicant improvement is found, if the results from the deep learning image analyses are manually corrected

by removing artefacts from the images.

Both manual determination of pores using SEM images

32,46–48

and automated measurements

49,50

, have advan-

tages and disadvantages. For example, manual methods can be laborious and time-consuming. e fully auto-

mated method by Tetard etal.

is rapid, allows quick generation of data, and the image acquisition and processing

require no monitoring, however, it needs a very specic setup and is destructive, since the specimens are broken

to shards. e semi-automatic method by Petersen etal.

can produce reliable data in a short amount of time,

minimizes artefacts related to the curvature of the tests, and gives information on pore area, perimeter, and

circularity indexes but focuses only on a small part of the shell, which limits the amount of data per specimen.

By contrast, porosity measurements using deep-learning as applied in this study are non-destructive and

automatically determines various pore parameters on the fully visible test surface. Moreover, the fully automated

method is reproducible in comparison to manual methods where the analyses are performed by dierent opera-

tors. e application of a non-destructive method allows the use of the foraminifera for other analyses, thereby

providing the possibility to use a single sample population for a multiproxy paleo reconstruction.

Although this automated method generates large datasets, proper attention should be given to the processing

of curved specimens of B. spissa, because the curvature can create diculties in counting the exact number of

pores. erefore, we suggest utilizing specimens with at surfaces.

Variation of pore patterns in B. spissa from dierent environments

All specimens of B. spissa that have been analysed showed a positive but weak correlation between the poros-

ity and the mean pore size (R

= 0.27, p < 0.05; Fig.3a). Certain foraminifera species increase their porosity by

increasing the size of their pores to facilitate electron acceptor uptake from the environment

49,51

. e strongest

correlation between mean pore size and porosity at the Gulf of Guayaquil (M77/2-59-01) suggests that indi-

viduals at this location tend to increase the porosity by increasing their mean pore size rather by increasing its

pore density. Similar observations were documented on Ammonia spp. that typically dwells in shallow marine

environments such as tidal mudats

. ese species tend to increase their porosity by building fewer but larger

pores, which has been suggested to ensure optimal shell stability

34,49

. e notable weaker correlation between

porosity and mean pore size, for the other analysed sites (R

between 0.05 and 0.12, Fig.3a) implies that most

of the analysed B. spissa do not control their porosity by modifying the size of the pores. is weak correlation

between porosity and the mean pore size in B. spissa is an indicator that the size of the pores is only a secondary

control on overall porosity of B. spissa at most of the studied locations.

e strongest signicant linear correlation between porosity and pore density has been found at the Mexican

Margin (MAZ-1E-04) (Fig.3b), which suggests that B. spissa adjusts its porosity by adapting the number of pores

and not the pore-size. Specimens from the Gulf of Guayaquil are exceptional as they show only a weak correlation

between porosity and pore density (R

= 0.1, Fig.3b). Nevertheless, the negative correlation between pore density

and mean pore size among the studied sites (Fig.3c) are in good agreement with previous studies on Ammonia

spp.

34,49

. Mechanical constraints like shell stability could be a controlling factor leading to the inverse relationship

between pore density and mean pore size

. Our new data shows that, except in the Gulf of Guayaquil, B. spissa

mainly controls its porosity by the number of pores.

e dierent trends at dierent locations indicate that long-term environmental conditions or genetic fac-

tors likely play a pivotal role in contributing to the morphological dierences in benthic foraminifera since

the sediment cores cover periods of ~ 20 kyrs. Especially at the Gulf of Guayaquil, the pore parameters showed

signicant dierences to the other studied locations. We speculate that these dierences could be related either

to the mechanism of electron acceptor uptake or to genetic factors. Benthic foraminifera can actively migrate

within the sediment to their preferred microhabitat

52–54

which exposes them to an oxygen/nitrate concentration

gradient. e habitat preference of B. spissa in oxygen-decient zones necessitates the use of alternate electron

acceptors like nitrate for respiration

. In nitrate-depleted habitats, B. spissa optimizes its nitrate accumulation

by building more pores to eciently take up nitrate resulting in higher pore density

. Previous observations

found that the cell size of many denitrifying foraminifers is limited by nitrate availability instead of oxygen

Several denitrifying foraminiferal species, including B. spissa, have been shown to encode a NO

–

transporter

in their genome and transcriptome

55,56

. is means by using these NO

–

transporters they can actively pump

–

into their cells, since NO

–

is a charged ion. is NO

–

can be stored as intracellular nitrate (ICN) which

can be utilized as a source of energy for metabolic activities

21,40,57–59

via complete denitrication during oxygen-

depleted conditions.

Vol.:(0123456789)

Scientic Reports

| (2023) 13:19628 |

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports/

Biogeochemical controls on the pore patterns in the Gulf of Guayaquil

e site from where core M77/2-59-01 was retrieved (3.95° S, 81.23° W) is outside core oxygen minimum zone

o Peru. e modern oxygen concentration recorded closest to this site is 55µmol/kg, which is higher than at

the other studied locations (38–47µmol/kg)

. When oxygen concentration increases above a certain threshold,

there will be less overall denitrication

61,62

resulting in higher nitrate availability. We speculate that if there is

more nitrate in the Gulf of Guayaquil relative to the other studied locations in the modern ocean, this was likely

also the case in the past. is is supported by a sedimentary nitrogen isotope record on the same core M77/2-

59-01 by Mollier-Vogel etal.

and Mollier-Vogel etal.

, which indicated that pelagic denitrication was low at

this location over the entire last deglaciation. e regional dierences in the patterns at Gulf of Guayaquil could

be an adaptation to the continuously higher nitrate availability at this site.

Genetic controls on the pore patterns in the Gulf of Guayaquil

e B. spissa specimens from the Gulf of Guayaquil are, except for their pore characteristics, morphologically

similar to the B. spissa from the other locations but could be a dierent phylogenetic strain. Observations of

Ammonia specimens by Hayward etal.

suggested that genetically dierent species can also be morphologically

distinguished. Later studies found genetically well-separated species of the Ammonia genus, which have earlier

been considered as eco-phenotypes of Ammonia, can now be morphologically distinguished by their pore pat-

terns and other subtle morphological features

. Similarly, it is possible to have the existence of genetic variation

and cryptic species within a B. spissa morpho-group due to the wide geographical distances, and variability in

ecological conditions that separated oxygen-depleted regions in the Pacic. Nevertheless, the phylotypes of B.

spissa without a combined morphometric molecular analysis would be very dicult to discriminate as a separate

species.

An extended modern pore density vs. nitrate calibration

Since there are studies that use either the pore density or porosity to reconstruct past environmental

conditions

32,33,45,65,66

we intended to address whether pore density or porosity is a better proxy for quantitative

nitrate reconstructions. Although pore density in B. spissa shows a signicant correlation to nitrate (Fig.4b), the

correlation between porosity and nitrate availability has not been systematically tested, yet. In addition, an exten-

sion of the local nitrate vs. pore density calibration for the Peruvian OMZ

to other regions and foraminiferal

species would increase the applicability of this proxy.

Figure4 shows the relationship between pore density in other bolivinids and [NO

–

]

from core-top samples

at dierent locations of the Pacic. e linear correlation between the pore density of B. spissa and B. subadvena

and [NO

–

]

is highly signicant and much stronger than the correlation to oxygen, temperature, salinity or

water depth (Supplementary Figs.SF1 to SF4), making their pore density a promising proxy for present and past

[NO

–

]

. is also suggests a close phylogenetic relationship with similar metabolic adaptations of both species.

Indeed, B. spissa was originally classied as a variant of B. subadvena with the name B. subadvena var. spissa

and 7 out of 7 Bolivina species that have been tested for denitrication were able to denitrify and 11 out of 12

analysed species intracellularly stored nitrate [Ref.

and references therein]. Although B. subadvena accumeata

is still considered a subspecies of B. subadvena, the pore characteristics are distinct from either B. spissa or B.

subadvena. e pore density of B. argentea is elevated compared to the other species as it tends to build numer-

ous but very small pores (see Fig.5).

erefore, the pore density of B. spissa and B. subadvena both can be used to reconstruct past [NO

–

]

con-

ditions according to the following equation (Eq.1), However, B. subadvena accumeata and B. argentea should

be avoided, when the calibration shown in Eq.(1) is used. Future studies will show, if the latter two species also

show species-specic relationships that might be used for paleoceanographic reconstructions.

Figure5. Scanning electron microscopic images of bolivinids (a) B. spissa, (b) B. subadvena (c) B. subadvena

accumeata, and (d) B. argentea.

Vol:.(1234567890)

Scientic Reports

| (2023) 13:19628 |

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports/

where PD is the pore density.

While the pore characteristics of denitrifying foraminifera are promising paleoproxies for past [NO

–

]

32,45

pore characteristics of the epifaunal species Cibicidoides and Planulina spp. that likely rely on O

respiration seem

to be good indicators for past bottom-water oxygen concentration [O

]

33,65

. Intriguingly, while the new data

on B. subadvena, and B. spissa indicate that pore density is more sensitive to ambient [NO

–

] variations than the

total porosity, it appears that the opposite is the case for epifaunal species. In Cibicidoides and Planulina spp.

porosity is more sensitive to ambient [O

] uctuations than the pore density

33,65

Data from only two sites for the correlation between total porosity of bolivinids and [NO

–

]

are available.

Future studies should address this issue and include both the pore density and total porosity. e fact that porosity

of B. spissa from the Sagami Bay and Costa Rica core-tops are similar, but the pore density at Costa Rica is sig-

nicantly higher indicates that the Sagami Bay specimens build larger pores than the specimens from Costa Rica.

e dierent pore characteristics of denitrifying bolivinids and the aerobic epifaunal species might be related

to the mechanism of electron acceptor uptake. e uptake of O

is limited by passive diusion, since O

is not

charged and foraminifera have no respiratory organs that can actively take up O

. us, aerobic foraminifera

can only increase the O

uptake through the pores by increasing the area of pores on their test (i.e. total poros-

ity), which can be done by either creating more pores (increase in pore density) or larger pores (increase in

mean pore size). Some foraminifera species ensure better shell stability by increasing their porosity through

building less but larger pores

. us, the increase of total porosity of epifaunal Cibicidoides and Planulina

spp. might also be restricted by shell stability. ey tend to build larger pores to increase their porosity, which

might explain the weaker correlation between pore density and ambient [O

] compared to total porosity

33,45

Denitrifying bolivinids can actively pump NO

–

into their cells, since NO

–

is a charged ion and they genetically

encode nitrate transporters

55,56

. us, we hypothesize that the denitrifying bolivinids do not rely on the increase

of total porosity but rather on the number of pores to enhance electron acceptor uptake. For the moment, the

empiric correlation between the pore density of B. spissa and B. subadvena appears to be solid, since a deglacial

pore density record of B. spissa from the Peruvian margin reconstructed similar [NO

–

]

as other proxies and

various modeling studies

Conclusions

e application of automated image analysis through deep-learning provided a robust method for determining

the pore patterns in the shallow infaunal benthic foraminiferal species B. spissa. e dierences in pore patterns

of B. spissa found between dierent studied locations suggest caution in the interpretation of the results. Never-

theless, our new data shows that, except for the Gulf of Guayaquil, B. spissa mainly controls its porosity by the

number of pores. is gives additional validation that the pore density of B. spissa is a robust and reliable paleo-

proxy for nitrate concentrations in bottom-waters. Quantitative reconstructions of past bottom-water nitrate

concentrations could help us to predict the environmental and ecological impacts of future climate scenarios.

Moreover, understanding the factors controlling porosity in bolivinids provides insight into benthic denitrica-

tion, which is indispensable for future biogeochemical studies. Future studies concerning foraminiferal porosity

should consider both mean pore size and pore density, and a combined morphometric molecular approach for

the complete description of foraminiferal pore patterns. As the presence of cryptic species within a morpho-

group might complicate paleoceanographic interpretation of pore density or porosity in benthic foraminifera,

the phylogenetic analyses of Bolivina species is highly relevant for better proxy validations.

Methods

Sampling of sediment cores

e piston core M77/2-059-1 (03° 57.01′ S, 81° 19.23′ W, recovery 13.59m) was retrieved from the Gulf of

Guayaquil at 997m water depth during RV Meteor cruise M77/2 in 2008. e chronostratigraphy is based on

accelerator mass spectrometry radiocarbon dating (AMS

C) of planktonic foraminifers, supported by benthic

stable oxygen isotope (δ

O) stratigraphy from Uvigerina peregrina

69,70

. e CALYPSO giant piston core MD01-

2415 (53° 57.09′ N, 149° 57.52′ E, recovery 46.23m) was recovered from the northern slope of the Sea of Okhotsk

at 822m water depth during WEPAMA cruise MD122 of the R/V Marion Dufresne

71,72

. e chronostratigraphic

framework of core MD01-2415 is based on a combination of stable oxygen-isotope stratigraphy, AMS

C dat-

ing, and orbital tuning

. e piston core MAZ-1E-04, Mexican Margin was collected on board the RV El Puma

at a water depth of 1468m. e core, SO206-43-MUC was retrieved in 2009 from a sea mound slope (Quepos

Slide) o Costa Rica during RS Sonne cruise SO206 using a multicorer. Supernatant water of the multicorer

SO206-43-MUC was carefully removed. en, the core was gently pushed out of the multicorer tube. For the

foraminiferal analyses, the core was cut into 10mm thick slices (upto 20cm depth) and samples were transferred

to Whirl-Pack™ plastic bags and stored at a temperature of 4°C.

e sediment samples from central part of Sagami Bay were collected by a push core (inner diameter: 8.2cm,

tube length: 32.0cm) using the manipulator of human occupied vehicle Shinkai6500 in 2021 (Table1 shows

the details of all sampling locations). e surface 2cm of the sediment was subsampled by extruding from the

push core tube and then kept frozen prior to an isolation of foraminifera. Bottom-water temperature, salinity,

and dissolved oxygen concentrations were 2.3°C, 34.5, and 56.4µM, respectively, which were measured with

the CTDO sensor (Seabird SBE19).

(1)

[

−

]

=−3896

(

±350

)

PD + 61

(

±1

)

Vol.:(0123456789)

Scientic Reports

| (2023) 13:19628 |

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports/

Sample processing

All sediment samples from the Gulf of Guayaquil (M77/2-59-01), Mexican Margin (MAZ-1E-04), Sea of Okhotsk

(MD01-2415), Costa Rica (SO206-43-MUC), and the Sagami Bay were washed and wet-sieved through a 63µm

mesh sieve. e residues were dried in an oven at temperatures between 38 and 50°C. Aerwards the samples

were fractioned into the grain-size fractions of 63–125, 125–250, 250–315, 315–355, 355–400, and > 400μm.

Specimens of Bolivina spissa, Bolivina subadvena, Bolivina subadvena accumeata and Bolivina argentea were

picked from the 125–250μm fraction. Only megalospheric specimens of B. spissa, were used for the pore analysis.

Bottom‑water nitrate analyses at core‑top locations

Supernatant water was sampled for the analysis of bottom-water NO

–

concentrations in a core replicate from the

multicore deployment at Costa Rica, (SO206-43-MUC). For the bottom water sample, a total of 2ml was passed

through a cadmium (Cd) catalyst to reduce NO

–

to NO

–

(nitrite), which was then analysed on-board using

photometry. e resulting concentration is a mixture of NO

–

, and NO

–

. Since NO

–

is a transient intermediate

species in the benthic nitrogen cycle and is generally present at lower concentrations than NO

–

, the NO

–

con-

centration determined is assumed to approximately represent the concentration of NO

–

For Sagami Bay nitrate analyses, ~ 20mL of overlying water was gently collected using a tube. e overlying

water was ltered through a 0.45µm membrane lter and then stored at − 25°C before nutrient analyses back

in land-based laboratory. Nutrient concentrations were measured with a continuous-ow analyzer (BL-Tech

QUAATRO 2-HR system, Japan)

. e data for the Peruvian OMZ cores has been taken from

Bottom‑water salinity, temperature and oxygen at core‑top locations

Bottom-water conditions at the locations that have been used for the core-top calibrations are shown in Table2.

Salinity, oxygen and temperature for the Costa Rica core have been taken from the World Ocean Atlas location

24,671(B), 84.5° W, 8.5° N and 550m depth

. At the Sagami Bay location bottom-water temperature, salinity,

and dissolved oxygen concentrations were measured with the CTDO sensor (Seabird SBE19). Data for bottom-

water oxygen and temperature at the locations from the Peruvian OMZ were taken from

. Salinity data for the

Peruvian OMZ was taken from

, using the CTD-data at M77/1-501/CTD-RO-23.

Image acquisition

A total number of 23 sample depths from the Mexican Margin (MAZ-1E-04), 37 sample depths from the Gulf

of Guayaquil (M77/2-59-01), 12 sample depths from the Sea of Okhotsk (MD01-2415), and 2 core-top samples

from Sagami Bay (Japan) and Costa Rica, (SO206-43-MUC) were utilized. All specimens of B. spissa were

mounted onto carbon pads and photographed using Scanning Electron Microscope (version: Hitachi Tabletop

SEM TM4000 series). All images were captured at a magnication of 150x. Due to the more or less at surface

of B. spissa, pore openings were generally well-dened, and clearly distinguishable from the SEM images. e

total area on the tests of the specimens were determined using the Zeiss ZEN lite soware (version: ZEN 3.4 blue

edition; https:// www. zeiss. com/ micro scopy/ de/ produ kte/ sow are/ zeiss- zen- lite. html).

Table 1. Site location information and distribution of specimens (B. spissa) from dierent sampling locations

used in the study.

Locations Latitude Longitude Water depth (m) No of B. spissa specimens

Gulf of Guayaquil, (M77/2-59-01) 3.95° S 81.32° W 997 669

Mexican Margin, (MAZ-1E-04) 22.9° N 106.91° W 1468 455

Sea of Okhotsk, (MD01-2415) 53.95° N 149.96° E 822 144

Sagami Bay push core 35.09° N 135.38° E 1410 37

Costa Rica, (SO206-43-MUC) 8.87° N 84.23° W 568 39

Table 2. Bottom-water conditions at the sampling locations that have been used for the core-top calibration.

Sampling locations in italic letters have been taken from

Samplinglocations Nitrate (µM) Water depth (m) Salinity Oxygen (µmol/kg) Temperature (°C)

Costa Rica (SO206-43-MUC) 39.1 568 34.69 9.53 7.47

Sagami Bay push core (Japan) 42.2 1410 34.50 56.40 2.30

M77/1-455/MUC-21 (OMZ, Peru) 34.0 465 34.64 2.42 8.12

M77/1-565/MUC-60 (OMZ, Peru) 40.1 640 34.56 8.17 6.70

M77/1-445/MUC-15 (OMZ, Peru) 40.8 928 34.56 36.77 4.76

M77/1-487/MUC-39 (OMZ, Peru) 38.8 579 34.55 3.70 7.21

M77/1-459/MUC-25 (OMZ, Peru) 41.0 698 34.57 12.55 6.68

M77/1-604/MUC-74 (OMZ, Peru) 40.8 878 34.53 34.23 5.72

M77/1-516/MUC-40 (OMZ, Peru) 36.1 513 34.60 2.40 8.05

Vol:.(1234567890)

Scientic Reports

| (2023) 13:19628 |

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports/

Size normalization

To reduce ontogenetic eects, the total area equivalent to the rst (oldest) ~ ten chambers (covering 50,000–70,000

µm

) were measured for the quantication of pore parameters

. e pore density increases with each newly

built chamber (Fig.5), related to a decrease in the surface/volume ratio with the size of the specimens. If the

more recent 1–2 chambers would be analysed, only specimens within the same ontogenetic stage could be used.

i.e. the size of the specimen and the number of chambers should be the same in all the chosen specimens. It is

practically impossible to use only specimens having exactly the same number of chambers. By sticking to the

oldest chambers of the foraminifer the ontogenetic eects are minimized by size normalization. Considering

the short life span of foraminifera, the data from earlier ontogenetic stages still provide a proper representation

of the present situation.

Moreover, the larger area provides a statistically robust, and larger dataset for each analysed specimen

Automated image analysis

A total number of 1344 fossil specimens of B. spissa sampled from ve dierent sampling locations were ana-

lysed. Porosity measurements were made on 6–20 well-preserved specimens of B. spissa in each of the studied

locations. e pore density, mean pore size, and porosity were determined with an automated image analyzing

soware Amira (version: Amira

3D pro) using a previously trained deep-learning algorithm. e deep learning

algorithm that has been used for this study is included in the Amira soware package. We used a convolutional

neural network model (UNet) backboned with a resnet18 model for the deep learning training. e deep learn-

ing algorithm was trained with manually segmented pores on 52 images of B. spissa. In total 17,649 pores have

been segmented manually for the deep learning training.

Only those specimens that had a total area equivalent to at least 50,000 µm

were used for the automated

analysis. e main steps for porosity measurements in Amira were:

•

Import of multiple SEM images.

•

e deep learning algorithm to recognize the pores was applied on imported images.

•

Only the oldest chambers that t within the total area of 50,000 to 70,000 µm

were taken into acccount

(Fig.1b). All chambers beyond this threshold were manually removed, using the segmentation tools in the

Amira soware.

•

A table with all measured pore characteristics can be exported by the soware at the end of each set of analy-

sis.

Comparison of manual vs. automatic pore density determination

To assess the reliability of the deep learning algorithm pore density was determined manually for 31 specimens

belonging to the species B. spissa (27 specimens) B. subadvena (3 specimens) and B. subadvena accumeata (1

specimen). For four additional specimens of B. argentea pore density was determined manually, since the pores

in this species are very small and not recognized by the deep learning algorithm that was trained with images of

B. spissa. e detailed procedure for manual pore density determinations are published in Glock etal.

Automated pore measurements with and without manual corrections

To explore whether manual corrections (i.e. corrections done on the specimens that were automatically pore ana-

lysed) made a signicant dierence on automated data, a total number of 858 specimens were randomly selected

and analysed both with and without manual correction. To apply manual corrections, we removed all artefacts

(i.e. unwanted particles on the surface of B. spissa) on each specimen during the automated image analysis and

obtained porosity data. For the automated image analysis without manual corrections, we applied the method of

analyzing each specimen without manually removing the artefacts. Statistical analysis was carried out to decide

if the porosity data obtained through either of these methods were signicantly dierent or not. e data have

been included as supplementary information in Supplementary TablesST2 and ST3.

e preliminary statistical analysis was carried out in Excel and veried using R. To test the normality of the

samples, we used Shapiro–Wilk normality test whenever necessary. To determine the correlation between pore

parameters, a linear ordinary least-square regression was used. For normal distributions, we used the parametric

Student’s t test (t), and for non-normal distributions we used the non-parametric Wilcox test (W). All the data

generated or analysed during this study have been included in the supplementary information les.

Data availability

All data generated or analysed during this study are included in the tables of this published article (and its Sup-

plementary Information les).

Received: 16 June 2023; Accepted: 2 November 2023

References

1. Stramma, L., Johnson, G. C., Sprintall, J. & Mohrholz, V. Expanding oxygen-minimum zones in the tropical oceans. Science

320(5876), 655–658 (2008).

2. Schmidtko, S., Stramma, L. & Visbeck, M. Decline in global oceanic oxygen content during the past ve decades. Nature 542,

335–339 (2017).

3. Salvatteci, R. et al. Smaller sh species in a warm and oxygen-poor Humboldt Current system. Science 375, 101–104 (2022).

Vol.:(0123456789)

Scientic Reports

| (2023) 13:19628 |

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports/

4. Mott, S. E., Hill, T. M., Roopnarine, P. D. & Kennett, J. P. Response of seaoor ecosystems to abrupt global climate change. Proc.

Natl. Acad. Sci. U.S.A. 112, 4684–4689 (2015).

5. IPCC. Global Warming of 1.5°C. An IPCC Special Report on the Impacts of Gobal Warming of 1.5°C Above Pre-industrial Levels

and Related Global Greenhouse Gas Emission Pathways, in the Context of Strengthening the Global Response to the reat of Climate

Change, Sustainable Development, and Eorts to Eradicate Poverty (Eds. Masson-Delmotte, V. etal.) (2018).

6. Keeling, R. F., Kortzinger, A. & Gruber, N. Ocean deoxygenation in a warmingworld. Ann. Rev. Mar. Sci. 2, 199–229. https:// doi.

org/ 10. 1146/ annur ev. marine. 010908. 163855 (2010).

7. Stramma, L., Schmidtko, S., Levin, L. A. & Johnson, G. C. Ocean oxygen minima expansions and their biological impacts. Deep-

Sea Res. I Ocean. Res. Pap. 57(4), 587–595 (2010).

8. Voss, M. et al. e marine nitrogen cycle: Recent discoveries, uncertainties and the potential relevance of climate change. Philos.

Trans. R. Soc. B Biol. Sci. 368, 293–296 (2013).

9. Gruber, N. & Galloway, J. N. An earth-system perspective of the global nitrogen cycle. Nature 451, 293–296 (2008).

10. Wallmann, K. Feedbacks between oceanic redox states and marine productivity: A model perspective focused on benthic phos-

phorus cycling. Glob. Biogeochem. Cycles https:// doi. org/ 10. 1029/ 2002G B0019 68 (2003).

11. Caneld, D. E., Glazer, A. N. & Falkowski, P. G. e evolution and future of earth’s nitrogen cycle. Science 330, 192–196 (2010).

12. Sutton, M. et al. e nitrogen x: From nitrogen cycle pollution to nitrogen circular economy. In Frontiers 2018/2019 (eds Sutton,

M. et al.) 52–64 (United Nations Environment Programme, 2019).

13. Karstensen, J., Stramma, L. & Visbeck, M. Oxygen minimum zones in the eastern tropical Atlantic and Pacic oceans. Prog.

Oceanogr. 77, 331–350 (2008).

14. Codispoti, L. A. et al. e oceanic xed nitrogen and nitrous oxide budgets: Moving targets as we enter the anthropocene?. Sci.

Mar. 65, 85–105 (2001).

15. Lam, P. & Kuypers, M. M. M. Microbial nitrogen cycling processes in oxygen minimum zones. Ann. Rev. Mar. Sci. 3, 317–345

(2011).

16. Gruber, N. e dynamics of the marine nitrogen cycle and its inuence on atmospheric CO

variations. Ocean Carbon Cycle Clim.

https:// doi. org/ 10. 1007/ 978-1- 4020- 2087-2_4 (2004).

17. Korom, S. F. Natural denitrication in the saturated zone: A review. Water Resour. Res. 28, 1657–1668 (1992).

18. Klotz, M. G. & Stein, L. Y. Nitrier genomics and evolution of the nitrogen cycle. FEMS Microbiol. Lett. 278, 146–156 (2008).

19. Goldstein, S. T. Foraminifera: A biological overview. Mod. Foraminifer. https:// doi. org/ 10. 1007/0- 306- 48104-9_3 (1999).

20. Glock, N. et al. e role of benthic foraminifera in the benthic nitrogen cycle of the Peruvian oxygen minimum zone. Biogeosciences

10, 4767–4783 (2013).

21. Piña-Ochoa, E. et al. Widespread occurrence of nitrate storage and denitrication among Foraminifera and Gromiida. Proc. Natl.

Acad. Sci. U.S.A. 107, 1148–1153 (2010).

22. Dale, A. W., Sommer, S., Lomnitz, U., Bourbonnais, A. & Wallmann, K. Biological nitrate transport in sediments on the Peruvian

margin mitigates benthic sulde emissions and drives pelagic N loss during stagnation events. Deep Res. I Oceanogr. Res. Pap. 112,

123–136 (2016).

23. Leutenegger, S. & Hansen, H. J. Ultrastructural and radiotracer studies of pore function in foraminifera. Mar. Biol. 54, 11–16

(1979).

24. Kuhnt, T. et al. Relationship between pore density in benthic foraminifera and bottom-water oxygen content. Deep Res. I Oceanogr.

Res. Pap. 76, 85–95 (2013).

25. Richirt, J. et al. Morphological distinction of three Ammonia phylotypes occurring along European coasts. J. Foraminifer. Res. 49,

76–93 (2019).

26. Schönfeld, J., Beccari, V., Schmidt, S. & Spezzaferri, S. Biometry and taxonomy of Adriatic ammonia species from Bellaria-Igea

Marina (Italy). J. Micropalaeontol. 40, 195–223 (2021).

27. Bé, A. W. H. Shell porosity of recent planktonic foraminifera as a climatic index. Science 161(3844), 881 (1968).

28. Frerichs, W. E. & Bb, A. W. H. Latitudinal variations in planktonic foraminiferal test porosity part 1. Opt. Stud. 2, 6–13 (1972).

29. Bé, A. W. H., Harrison, S. M. & Lott, L. Orbulina universa d’Orbigny in the Indian Ocean. Micropaleontology 19(2), 150–192 (1973).

30. Frerichs, W. E. & Ely, R. L. Test porosity as a paleoenvironmental tool in the Late Cretaceous of the Western Interior. Rocky Mt.

Geol. 16(2), 89–94 (1978).

31. Bijma, J., Faber, W. W. & Hemleben, C. Temperature and salinity limits for growth and survival of some planktonic foraminifers

in laboratory cultures. J. foraminifer. Res. 20(2), 95–116 (1990).

32. Glock, N. et al. Environmental inuences on the pore density of Bolivina spissa (Cushman). J. Foraminifer. Res. 41, 22–32 (2011).

33. Rathburn, A. E., Willingham, J., Ziebis, W., Burkett, A. M. & Corliss, B. H. A new biological proxy for deep-sea paleo-oxygen:

Pores of epifaunal benthic foraminifera. Sci. Rep. 8, 1–8 (2018).

34. Richirt, J. et al. Scaling laws explain foraminiferal pore patterns. Sci. Rep. 9, 1–11 (2019).

35. Quillévéré, F. et al. Global scale same-specimen morpho-genetic analysis of Truncorotalia truncatulinoides: A perspective on the

morphological species concept in planktonic foraminifera. Palaeogeogr. Palaeoclimatol. Palaeoecol. 391, 2–12 (2013).

36. Pérez-Cruz, L. & Machain Castillo, M. L. Benthic foraminifera of the oxygen minimum zone, continental shelf of the Gulf of

Tehuantepec, Mexico. J. Foraminifer. Res. 20(4), 312–325 (1990).

37. Gary, A. C., Healy-Williams, N. & Ehrlich, R. Water-mass relationships and morphologic variability in the benthic foraminifer

Bolivina albatrossi Cushman, northern Gulf of Mexico. J. Foraminifer. Res. 19, 210–221 (1989).

38. Glock, N., Schoenfeld, J. & Mallon, J. e functionality of pores in benthic foraminifera in view of bottom water oxygenation; a

review. In Anoxia (eds Altenbach, A. V. et al.) 539–552 (Springer, 2012).

39. Phleger, F. B. & Soutar, A. Production of benthic foraminifera in three East Pacic Oxygen minima. Micropaleontology 19, 110

(1973).

40. Risgaard-Petersen, N. et al. Evidence for complete denitrication in a benthic foraminifer. Nature 443, 93–96 (2006).

41. Glock, N. et al. Metabolic preference of nitrate over oxygen as an electron acceptor in foraminifera from the Peruvian oxygen

minimum zone. Proc. Natl. Acad. Sci. U.S.A. 116, 2860–2865 (2019).

42. Harman, R. A. Distribution of foraminifera in the Santa Barbara Basin, California. Micropaleontology 10, 81 (1964).

43. Fontanier, C. et al. Living (stained) deep-sea foraminifera o hachinohe (NE Japan, western Pacic): Environmental interplay in

oxygen-depleted ecosystems. J. Foraminifer. Res. 44, 281–299 (2014).

44. Lecun, Y., Bengio, Y. & Hinton, G. Deep learning. Nature 521, 436–444 (2015).

45. Glock, N. et al. Coupling of oceanic carbon and nitrogen facilitates spatially resolved quantitative reconstruction of nitrate inven-

tories. Nat. Commun. 9, 1217 (2018).

46. Hayward, B. W., Holzmann, M., Grenfell, H. R., Pawlowski, J. & Triggs, C. M. Morphological distinction of molecular types in

ammonia–towards a taxonomic revision of the world’s most commonly misidentied foraminifera. Mar. Micropaleontol. 50,

237–271 (2004).

47. Constandache, M., Yerly, F. & Spezzaferri, S. Internal pore measurements on macroperforate planktonic Foraminifera as an alterna-

tive morphometric approach. Swiss J. Geosci. 106, 179–186 (2013).

48. Kuhnt, T. et al. Automated and manual analyses of the pore density-to-oxygen relationship in Globobulimina Turgida (Bailey). J.

Foraminifer. Res. 44, 5–16 (2014).

Vol:.(1234567890)

Scientic Reports

| (2023) 13:19628 |

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports/

49. Petersen, J. et al. Improved methodology for measuring pore patterns in the benthic foraminiferal genus Ammonia. Mar. Micro-

paleontol. 128, 1–13 (2016).

50. Tetard, M., Beaufort, L. & Licari, L. A new optical method for automated pore analysis on benthic foraminifera. Mar. Micropaleontol.

136, 30–36 (2017).

51. Moodley, L. & Hess, C. Tolerance of infaunal benthic foraminifera for low and high oxygen concentrations. Biol. Bull. 183, 94–98

(1992).

52. Geslin, E., Heinz, P., Jorissen, F. & Hemleben, C. Migratory responses of deep-sea benthic foraminifera to variable oxygen condi-

tions: Laboratory investigations. Mar. Micropaleontol. 53, 227–243 (2004).

53. Alve, E. & Bernhard, J. M. Vertical migratory response of benthic foraminifera to controlled oxygen concentrations in an experi-

mental mesocosm. Mar. Ecol. Prog. Ser. 116, 137–152 (1995).

54. Linke, P. & Lutze, G. F. Microhabitat preferences of benthic foraminifera-a static concept or a dynamic adaptation to optimize food

acquisition?. Mar. Micropaleontol. 20, 215–234 (1993).

55. Woehle, C. et al. A novel eukaryotic Denitrication pathway in foraminifera. Curr. Biol. 28, 2536-2543.e5 (2018).

56. Woehle, C. et al. Denitrication in foraminifera has an ancient origin and is complemented by associated bacteria. Proc. Natl.

Acad. Sci. U.S.A. https:// doi. org/ 10. 1073/ pnas. 22001 98119 (2022).

57. Høgslund, S., Revsbech, N. P., Cedhagen, T., Nielsen, L. P. & Gallardo, V. A. Denitrication, nitrate turnover, and aerobic respira-

tion by benthic foraminiferans in the oxygen minimum zone o Chile. J. Exp. Mar. Biol. Ecol. 359(2), 85–91 (2008).

58. Piña-Ochoa, E., Koho, K. A., Geslin, E. & Risgaard-Petersen, N. Survival and life strategy of the foraminiferan Globobulimina

turgida through nitrate storage and denitrication. Mar. Ecol. Prog. Ser. 417, 39–49 (2010).

59. Koho, K. A., Piña-Ochoa, E., Geslin, E. & Risgaard-Petersen, N. Vertical migration, nitrate uptake and denitrication: Survival

mechanisms of foraminifers (Globobulimina turgida) under low oxygen conditions. FEMS Microbiol. Ecol. 75, 273–283 (2011).

60. Garcia H.E., Boyer, T. P., Baranova, O. K., Locarnini, R.A., Mishonov, A.V., Grodsky, A., Paver, C.R., Weathers, K.W., Smolyar, I.V.,

Reagan, J.R., Seidov, M.M., Zweng, D., World Ocean Atlas 2018: Product documentation. A. Mishonov, Technical Editor (2019).

61. Parkin, T. B. & Tiedje, J. M. Application of a soil core method to investigate the eect of oxygen concentration on denitrication.

Soil Biol. Biochem. 16, 331–334 (1984).

62. Goering, J. J. Denitrication in the oxygen minimum layer of the eastern tropical Pacic Ocean. Deep. Res. Oceanogr. Abstr. 15,

157–164 (1968).

63. Mollier-Vogel, E. et al. Mid-holocene deepening of the Southeast Pacic oxycline. Glob. Planet. Change 172, 365–373 (2019).

64. Mollier-Vogel, E. et al. Nitrogen isotope gradients o Peru and Ecuador related to upwelling, productivity, nutrient uptake and

oxygen deciency. Deep Res. I Oceanogr. Res. Pap. 70, 14–25 (2012).

65. Glock, N., Erdem, Z. & Schönfeld, J. e Peruvian oxygen minimum zone was similar in extent but weaker during the Last Glacial

Maximum than Late Holocene. Commun. Earth Environ. 3, 1–14 (2022).

66. Lu, W., Wang, Y., Oppo, D. W., Nielsen, S. G. & Costa, K. M. Comparing paleo-oxygenation proxies (benthic foraminiferal surface

porosity, I/Ca, authigenic uranium) on modern sediments and the glacial Arabian Sea. Geochim. Cosmochim. Acta 331, 69–85

(2022).

67. Cushman, J. A. Foraminifera of the typical monterey of California. Contrib. Cushman Lab. Foraminifer. Res. 2(30), 53–69 (1926).

68. Glock, N. Benthic foraminifera and gromiids from oxygen-depleted environments–survival strategies, biogeochemistry and trophic

interactions. Biogeosciences 20, 3423–3447 (2023).

69. Mollier-Vogel, E., Leduc, G., Böschen, T., Martinez, P. & Schneider, R. R. Rainfall response to orbital and millennial forcing in

northern Peru over the last 18ka. Quat. Sci. Rev. 76, 29–38 (2013).

70. Nürnberg, D. et al. Sea surface and subsurface circulation dynamics o equatorial Peru during the last ∼17 kyr. Paleoceanography

30, 984–999 (2015).

71. Holbourn, A., Kiefer, T., Paumann, U., & Rothe, S. WEPAMA Cruise MD 122/IMAGES VII, Rapp. Campagnes Mer OCE/2002/01,

Inst. Polaire Fr. Paul Emile Victor (IPEV), Plouzane, France (2002).

72. Nürnberg, D. & Tiedemann, R. Environmental change in the Sea of Okhotsk during the last 1.1 million years. Paleoceanography

19, 1–23 (2004).

73. Nomaki, H. et al. Insitu experimental evidences for responses of abyssal benthic biota to shis in phytodetritus compositions

linked to global climate change. Glob. Change Biol. 27, 6139–6155 (2021).

74. Krahmann, G. Physical oceanography during METEOR cruise M77/1. IFM-GEOMAR Leibniz-Institute of Marine Sciences, Kiel

University, PANGAEA (2012). 10.1594/PANGAEA.777978.

Acknowledgements

We are grateful to the micropaleontology group at the Universität Hamburg. We greatly acknowledge the help

of Dr.Yvonne Milker with the SEM, Jutta Richarz, Kaya Oda for lab support, and student assistants Hanna

Firrincielli and Hannah Krüger. In addition, we thank Anke Bleyer for performing the nitrate analyses during

Sonne cruise So206. Funding was provided by the Deutsche Forschungsgemeinscha (DFG) through both N.G.’s

Heisenberg grant GL 999/3-1 and grant GL 999/4-1. Funding for the core MD01-2415 recovery was provided by

the German Science Foundation (DFG) within project Ti240/11-1. We thank Yvon Balut, Agnes Baltzer, and the

Shipboard Scientic Party of RV Marion Dufresne cruise WEPAMA 2001 for their kind support. e recovery

of core M77-59 recovery was a contribution of the German Science Foundation (DFG) Collaborative Research

Project “Climate–Biogeochemistry interactions in the Tropical Ocean” (SFB 754). e study is a contribution to

the Cluster of Excellence ‘CLICCS—Climate, Climatic Change, and Society’, and a contribution to the Center

for Earth System Research and Sustainability (CEN) of Universität Hamburg.

Author contributions

A.G.M. wrote the core manuscript, did the sample preparation, electron microscopy of the fossil foraminifera and

image and statistical analyses of all samples. N.G. planned the sampling strategy and study design, did onboard

sampling during So206 and did the electron microscopy and analyses of the core-top samples. G.S. hosted the

research group, and provided access to SEM, and lab facilities at the Universität Hamburg. D.N. provided sam-

pling material for cores MD01-2415 and M77/2-59-01. C.D. provided sampling material for core MAZ-1E-04

and H.N. and I.S. provided the core-top samples and environmental parameters from Sagami Bay. All authors

contributed to discussing the data and writing the manuscript.

Funding

Open Access funding enabled and organized by Projekt DEAL. We acknowledge nancial support from the Open

Access Publication Fund of Universität Hamburg.

Vol.:(0123456789)

Scientic Reports

| (2023) 13:19628 |

https://doi.org/10.1038/s41598-023-46605-y

www.nature.com/scientificreports/

Competing interests

e authors declare no competing interests.

Additional information

Supplementary Information e online version contains supplementary material available at https:// doi. org/

10. 1038/ s41598- 023- 46605-y.

Correspondence and requests for materials should be addressed to A.G.M.

Reprints and permissions information is available at www.nature.com/reprints.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and

institutional aliations.

Open Access is article is licensed under a Creative Commons Attribution 4.0 International

License, which permits use, sharing, adaptation, distribution and reproduction in any medium or

format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the

Creative Commons licence, and indicate if changes were made. e images or other third party material in this

article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the

material. If material is not included in the article’s Creative Commons licence and your intended use is not

permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from

the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.