Canyon conditions impact carbon flows in food webs

savagecowcreekMechanics

Feb 22, 2014 (3 years and 6 months ago)

127 views


1

Canyon conditions impact carbon flows in food webs
1

of three sections of the Nazaré canyon

2



3

Dick van Oevelen
1,*
, Karline Soetaert
1
, Rosa García Novoa
2,3
, Henko de Stigter
4
,
4

Marina da Cunha
5
, Antonio Pusceddu
6
, Roberto Danovaro
6

5


6

1

Centre for Estuarine and Marine Ecology, Netherlands Institute of Ecology (NIOO
-
7

KNAW), POB 140, 4400 AC Yerseke, The Netherlands

8

2

Jacobs University Bremen, Campus Ring 1, 28759 Bremen, Germany

9

3

Department of Global Change Research, IMEDEA (CSIC
-
UIB) Ins
tituto
10

Mediterr
á
neo de Estudios Avanzados, Miquel Marqu
é
s 21, 07190 Esporles, Spain

11

4

Royal Netherlands Institute for Sea Research (
NIOZ), POB 59, 1790 AB Den Burg
-

12

Texel, The Netherlands


13

5

Centro de Estudos do Ambiente e do Mar (CESAM) &

Departamento de Biologia,
14

Universidade de Aveiro,
Campus de Santiago
, 3810
-
193 Aveiro, Portugal


15

6

Department of Marine Science, Polytechnic University of Marche, Via Brecce
16

Bianche, 60131 Ancona, Italy

17


18


19


20


21

*

corresponding author:
d.vanoevelen@nioo.knaw.nl


22


2

Abstract

1

Submarine canyons directly transport
large
amounts of sediment

and organic
2

matter (OM) from the continental shelf to the abyssal plain.

Three carbon
-
based food
3

web models were constructed for the

upper (
300


750

m

water depth), middle (2700
4



3500 m) and lower section (4000


5000 m) of the Nazaré canyon (eastern Atlantic
5

Ocean) using linear inverse modeling to examine how the f
ood web
is influenced by
6

the characteristics of the respective canyon

section. The models were based on an
7

empirical dataset consisting of biomass and carbon processing data, and general
8

physiological data constraints from the literature. Environmental conditions, most
9

notably organic matter (OM) input and hydrodynamic acti
vity, differed between the
10

canyon sections and strongly affected the benthic food web structure. Despite the
11

large difference in depth, the OM inputs into the food webs of the upper and middle
12

sections were of similar magnitude (7.98±0.84 and 9.30±0.71 mmo
l C m
-
2

d
-
1
,
13

respectively). OM input to the lower section was however almost 6
-
7 times lower
14

(1.26±0.03 mmol C m
-
2

d
-
1
). Canyon conditions greatly influenced OM processing
15

within the food web. Carbon processing in the upper section was dominated by
16

prokary
otes (70% of total respiration), though there was a significant meiofaunal
17

(21%) and smaller macrofaunal (9%) contribution. The high total faunal contribution
18

to carbon processing resembles that found in shallower continental shelves and upper
19

slopes, alth
ough the meiofaunal contribution is surprisingly high and suggest that high
20

current speeds and sediment resuspension in the upper canyon favor the role of the
21

meiofauna. The high OC input and conditions in the accreting sediments of the middle
22

canyon secti
on were more beneficial for megafauna (holothurians), than for the other
23

food web compartments. The high megafaunal biomass (516 mmol C m
-
2
), their large
24

contribution to respiration (56% of total respiration) and secondary production (0.08
25


3

mmol C m
-
2

d
-
1
)
shows that these accreting sediments in canyons are megafaunal
1

hotspots in the deep
-
sea. Conversely, carbon cycling in the lower canyon section was
2

strongly dominated by prokaryotes (86% of respiration) and the food web structure
3

therefore resembled that o
f lower slope and abyssal plain sediments.

This study shows
4

that elevated OM input in canyons may favor the faunal contribution to carbon
5

processing and create hotspots of faunal biomass and carbon processing along the
6

continental shelf.


7


4

Introduction

1

Submarine canyons are incisions of the continental margin and directly link
2

the continental shelf with deep
-
sea plains by transporting
large

amounts of sediment

3

(Canals

et al
.
, 2006; de Stigter

et al
.
, 2007)

and
OM

(Epping

e
t al
.
, 2002; Vetter and
4

Dayton, 1999)
.
T
he comparatively rapid transport in active canyons results in the
5

sedimentary OM being also of higher quality as compared to slope sediments at
6

similar water depth
(Garcia

et al
.
, 2007; Pusceddu

et al
.
, 2010; Vetter and Dayton,
7

1999)
. The high quantity and quality of the OM in canyon sediments results in carbon
8

oxidation rates
(Epping

et al.
, 2002; Rabouille

et al
.
, 2009)

and benthic standing
9

stocks of nematodes
(Ingels

et al
.
, 2009)

and deposit
feeding holothurians
(Amaro

et

10

al
.
, 2009; De Leo

et al
.
, 2010; Vetter and Dayton, 1999)

that are higher as compared
11

to adjacent open slopes and indicate extensive carbon cycling in the benthic food web.

12

These latter studies focus on individual components of the benthic food web
13

and suggest that different benthic components may benefit from the enhanced influx
14

of OM into canyons. These comparisons are, however, based
on single

biomass
-
to
-
15

biomass

or

process
-
by
-
process comparisons.
It is unclear how the structure of the
16

whole food web and carbon partitioning within the food web is affected by canyon
17

conditions. Moreover, it is unclear whether and how emerging properties at the whole
18

food web level are

impacted by canyon conditions. N
etwork analysis has been
19

developed to condense information contained in complex networks
, such as food
20

webs,

into interpretable indices
(Fath and Patten, 1999; Ulanowicz, 2004)
.

The index
21

total system throughput (
) sums carbon flows in the food web to obtain a measure
22

of total food web activity. The Finn cycling index summarizes

the fraction of total
23

carbon cycling that is generated
by recycling processes
(Allesina and Ulanowicz,
24

2004)
. Another index that is claimed to be related to food web maturity is average
25


5

mutual information (AMI), that ga
uges how orderly
and

coherently

flows

are

inter
-
1

co
n
nected

(Ulanowicz, 2004 and references therein)
. It is claimed that AMI is
2

indicative of the developmental status of an ecosystem and that while a food web
3

develo
ps specialization results in higher values of AMI.

4

The Nazaré canyon intersects the Portuguese continental shelf and
extends
5

from a water depth of 50 m near
the
coast
down
to 5000 m

at the abyssal plain and
6

presents an interesting case study because of the

varying conditions within the
7

canyon. The upper canyon section (
50


2700 m water depth) is characterized by a V
-
8

shaped valley that is deeply incised in the continental shelf. The middle canyon (
2
700

9



4000 m) is a broad meandering valley with terraced sl
opes that may experience high
10

rates of particle and organic matter
sedimentation
(Masson

et al
.
, this issue)
. The
11

upper and middle canyon sections capture suspended particulate matter from the
12

adjacent shelf and are affected by internal tide circulation of water with high bottom
13

current speeds, thereby imposing physical disturbance on the sedimentary
14

environment

(de Stigter

et al
.
, 2007)
. Finally, the lower canyon is a kilometers
-
wide
15

flat
-
floored valley that gently descends from
4
0
00
to
5000

m

depth
(de Stigter

et al.
,
16

2007; Masson

et al
.
, this is
sue)
.

17

The physical disturbance of s
ediments is especially strong in the narrow V
-
18

shaped valley of the upper canyon section and this may impose constraints on the
19

development of the food web. Especially large and longer
-
lived components of the
20

food web may be affected and carbon cycling may
be shifted towards microbes as
21

compared to sediments with similar OM input that are less frequently disturbed
(Aller
22

and Aller, 2004)
. Carbon recycling, quantified with the Finn cycling index, may
23

therefore be lower because fewer food web components give rise to more limited
24

recycling in the food web. A
lso
food web maturity
, as measured with the network
25


6

index AMI,

is expected to be lower as compared to the middle and lower canyon
1

sections.

2

The terraced slopes of the middle canyon section experience high rates of
3

sedimentation and associated organic matte
r input. Transport
of (semi)
-
labile OM to
4

these greater depths in the canyon may imply a deviation from the archetypical
5

relation between water depth and sediment oxygen consumption (SOC)
. The SOC and
6

the network index “total system throughput” is expected

to be comparatively elevated
7

in the middle section of the canyon due to the enhanced OM input as compared to
8

open slope sediments at similar water depth. The enhanced input OM ma
y not
be
9

partitioned equally among the food web compartments and may be influ
enced by the
10

environmental conditions in the respective canyon.
D
e Leo et al.
(2010)

for example,
11

reported extremely high biomass

levels of

particularly

deposit
-
feeding
holothurians
in
12

a low relief muddy sediment at 900


1100 m in the Kaikoura Canyon (New Zealand).
13

The conditions in the Kaikoura canyon are reported to be similar to the middle section
14

of the Nazaré canyon and indeed

high holothurian abundances are found there too
15

(Amaro

et al
.
, 2009)
. With
a whole food web approach as followed here it will be
16

possible to study quantitatively whether different food web compartments take
17

proportional advantage of the enhanced OM input in this section of the Nazaré
18

canyon.

19

The deeper canyon section is where the

canyon widens into a
kilometres
-
broad
20

channel
in the abyssal plain
(de Stigter

et al
.
, 20
07)
.

This deep canyon section, which
21

only intermittently receives material derived from up
-
canyon sections

via sediment
22

gravity flo
ws,
better resembles regular abyssal plain conditions with an associated
23

lower OM input. Under these lower OM inputs, lower faunal contributions to carbon
24


7

cycling are expected and the more steady conditions may imply a higher food web
1

maturity and higher r
ecycling within the food web.

2

Verifying
how specific conditions in the three canyon sections impose on the
3

benthic food web requires an analysis of the trophic structure of the complete benthic
4

food web. The quantification of complete food webs is however
a data
-
demanding
5

effort and canyon data sets are typically incomplete and limited in scope.
To
6

overcome
these
limitations and
maximize the amount of information gained from the
7

available data
, so
-
called linear inverse models (LIM) have been developed
. LIM
8

allow

quantifying biological interac
tions in a complex food web from an incomplete
9

and uncertain data set

such as encountered in the deep
-
sea
(Soetaert and Van Oevelen,
10

2009)
. For example,
Van Oevelen et al.
(2009)

using linear inverse modeling
to
11

quantify the interactions in the complex food web of a cold
-
water coral community at
12

Rockall Bank and

provid
ed

evidence that coral communities are hot
-
spots of biomass
13

and carbon cycling along continental margins
.

14

Here we develop
linear inverse models (
LIM)
to quantify carbon flows
in
the
15

complex food web
s characterizing upper, middle and lower sections of the Nazaré
16

canyon. The observed food web structures
and selected network indices are examined
17

as a function of the characteristics of the respective c
anyon section.

18

Methods

19

2.1 Nazaré canyon characteristics

20

The Nazaré canyon, one of the largest submarine canyons in Europe, intersects
21

the Portuguese continental shelf and has been intensively studied in the framework of
22

different European projects such as

OMEX
-
II, EUROSTRATAFORM and HERMES.
23

Expeditions carried out within these projects have resulted in comparatively high data
24


8

availability on different physical, chemical and biological aspects of the canyon
1

system.

De Stigter et al.
(2007)

proposed a division of the canyon into three sections
2

based on hydrographic and physical characteristics. The upper canyon is characterized
3

by a V
-
shaped valley that is deeply inc
ised in the continental shelf and starts at 50 m
4

water depth and runs down to a depth of 2700 m. The middle canyon (
2
700



4000 m)
5

is a broad meandering valley with terraced slopes and the lower canyon is a flat
6

floored valley that gently descends from
4
0
0
0
to
5000

m

depth. The water column
7

along the Western Iberian Margin is stratified, with relatively warm (14 to 18ºC) and
8

saline (35.4 to 35.8) water at the surface (North Atlantic Central Water) to cold (2ºC)
9

and less saline (34.8) water at 5000 m depth (
North Atlantic Deep Water). The upper
10

and middle canyon sections capture suspended particulate matter from the adjacent
11

shelf and are affected by internal tide circulation of water with high bottom current
12

speeds
(de Stigter

et al
.
, 2007)
.

13

The seabed of the Nazaré canyon is heterogeneous and consists of a highly
14

dynamic thalweg filled with coarse sandy and gravelly deposits, steep sloping canyon
15

walls with rocky outcrops, and terraces with

thick accumulations of soft muddy
16

sediments
(Tyler

et al
.
, 2009)
. The hard substrata in the thalweg and on steep walls
17

and outcrops are covered in places with a thin, centimeter
-
thick drape of soft mud,
18

where it is impossible to sa
mple with box
-

or multicorer to estimate biomass.
19

Moreover, to avoid large heterogeneity in the data set due to seabed differences, the
20

focus of this manuscript is on soft
-
sediments outside the thalweg, which were split
21

into the three sections as described

above. The depth range of the upper section was
22

here limited to 300


700 m.

23

Chemical and biological data were available on the concentration of total
24

c
arbohydrates
, l
ipids

and p
roteins

in the sediment
(Puscedd
u

et al
.
, 2010)
,
25


9

sedimentary chl
a
content
(Garcia and Thomsen, 2008)
, sediment diagenesis
(Epping

1

et al
.
, 2002)
, prokaryotic heterotrophic carbon production (Danovaro, unpub. data),
2

nematode trophic structure
(Danovaro

et al
.
, 2009)

and the macro
-

and megafaunal
3

community structure (Cunha et al., this issue and unpub. data).

Such data on biotic
4

and abiotic carbon stocks and transformat
ion rates are perfectly suited to quantify
5

food webs

of the three sections of the Nazaré canyon using linear inverse modeling.

6

2.2 Linear inverse models

7

The food web models developed for the Nazaré canyon are constructed using
8

linear inverse modeling
(Van Oevelen

et al
.
, 2010)
. In an inverse model, the food web
9

compartments and flows between them are
fixed
a priori

(see ‘Food

web structure’
10

below). The flow magnitudes are constrained within the boundaries that are defined
11

by the inclusion of empirical data on standing stocks, flux data and physiology into
12

the model. The food web topology and empirical data are included in a ma
trix
13

equation with equalities and in a matrix equation with inequalities
.

These matrix
14

equations are solved simultaneously to recover quantitative values for the flow values,
15

such that the flow values in a model solution are within the boundaries defined b
y the
16

matrix equations. The model was run 10,000 times and each time a different solution
17

is generated to allow estimating the mean and standard deviation of each unknown
18

flow. It is important to note that by running the model 10,000 times, the uncertainty

in
19

the empirical data (see ‘Data availability’ below) is propagated onto an uncertainty
20

estimate of the carbon flows as indicated by its standard deviation. Convergence of
21

the mean and standard deviation of the flows was used to verify whether the set of
22

10,000 model solutions was sufficiently large.

23

Several reviews on the technical and methodological aspects of linear inverse
24

modeling have been published and will therefore not be repeated here
(Soetaert and
25


10

Van Oevelen, 2009; Van Oevelen

et al
.
, 2010)
. These reviews contain simple models
1

to exemplify the setup and solution of linear inverse food web models using the
2

software
packages
LIM

(Soetaert and
Van Oevelen, 2008; Van Oevelen

et al
.
, 2010)

3

and
limSolve

(Soetaert

et al
.
, 2008)

that run in the R software
(R Development Core
4

Team, 2008)
.
The Nazaré food web models
are
made publically available in the
LIM

5

package
.

6

2.3 Food web structure

7

The compartments in the food web models w
ere chosen based on the classical
8

size distribution of prokaryotes (Pro), meiofauna (Mei), macrofauna (Mac) and
9

megafauna (Meg). The faunal compartments were further subdivided based on the
10

feeding classification for nematodes
(Wieser, 1953)

and feeding types for macro
-

and
11

megafauna were surface deposit
-
feeder (SDF),
deposit
-
feeder (DF), suspension feeder
12

(SF) and predator+scavenger (PS) (see below). The sedimentary organic matter was
13

divided into

dissolved organic carbon (DOC) and labile (lDet), semi
-
labile (sDet) and
14

refractory detritus (rDet).

15

Inputs to the food web

are deposition and/or suspension feeding of suspended
16

labile (lDet_w), semi
-
labile (sDet_w) and refractory detritus (rDet_w). Outputs from
17

the food web are respiration to dissolved inorganic carbon (DIC), burial of rDet, DOC
18

efflux to the water column and

export by the macro
-

and megafaunal compartments
19

(e.g. consumption by fish).

20

The detritus pools in the sediment can be hydrolyzed to DOC and the labile
21

and semi
-
labile detritus pools are grazed upon by meiofauna and MacSDF, MacDF,
22

MacPS, MegSDF and MegDF.

DOC is taken up by prokaryotes or fluxes out of the
23

sediment to the water column. Predatory feeding links are primarily defined based on
24

size class; prokaryotes are consumed by all meiofaunal and non
-
suspension feeding
25


11

macro
-

and megafaunal compartments,
meiofaunal compartments are consumed by
1

non
-
suspension feeding macro
-

and megafaunal compartments, the macrofaunal
2

compartments MacSDF, MacDF and MacSF are preyed upon by MacPS.

3

Part of the ingested matter by the faunal compartments is not assimilated but
4

instead expelled as feces, the non
-
assimilated labile (e.g. labile detritus, prokaryotes
5

and faunal compartments) and semi
-
labile (semi
-
labile detritus) carbon, flows into
6

semi
-
labile and refractory detritus, respectively. Respiration by faunal compartment
s
7

is defined as the sum of maintenance respiration (biomass
-
specific respiration) and
8

growth respiration (overhead on new biomass production). Prokaryotic mortality is
9

represented here as a flux to DOC and faunal mortality is defined as a flux to labile
10

de
tritus.

11

2.4 Data availability

12

The Nazaré canyon is one of the best studied canyons in Europe, with studies
13

on sediment transport and/or fate of organic matter
(e.g. de Stigter

et al
.
, 2007; Epping

14

et al
.
, 2002; García

et al
.
, 2008)
, concentration of total c
arbohydrates
, lipids and
15

p
roteins

in the sediment
(Pusceddu

et al
.
, 2010)

heterotrophic prokaryotic C
16

production (Danovaro unpub. data), nematode community structure
(Garcia

et al
.
,
17

2007; Danovaro

et al
.
, 2009; Ingels

et al.
, 2009)
, meiofaunal abundance
(Bianchelli

et
18

al
.
, 2010)
, macro
-

and megafaunal community structure
(Tyler

et al
.
, 2009, Cunha et
19

al., this issue and unpub. data)
. As stated above, empirical data were only included if
20

they were collected from the soft
-
sediments of the upper, m
iddle or lower section of
21

the canyon.

22

Detritus stocks were delineated as follows (Table 1): the stock of labile
23

detritus was defined as all carbon associated with chlorophyll
a
. Chlorophyll
a

24

concentrations were taken from the top 5 cm in sediments of the off
-
thalweg stations
25


12

(Garcia and Thomsen, 2008)
, which were converted to carbon units by assuming a
1

carbon to chl
a

ratio of 40. Semi
-
labile detritus was defined as t
he sum of the
2

carbohydrates, lipids and proteins (i.e. biopolymeric carbon) that were converted to
3

carbon equivalents
(Pusceddu

et al
.
, 2010)
. Biopolymeric carbon concentrations were
4

measured only in the top 1 cm and were linearly extrapolated to 5 cm depth under the
5

assumption that all semi
-
labile detritus is degraded in the top 5 cm. The latter
6

assumption is supported by Epping et al.
(2002)

who showed that carbon d
egradation
7

occurs primarily in the top 5 cm of the sediment. Refractory detritus was defined as
8

the degradable fraction of the particulate organic carbon in the top 5 cm of the
9

sediment
(derived from organic carbon content profiles in Epping

et al.
, 2002)
, minus
10

the labile and semi
-
labile detritus pools.

11

Biomass data were available for prokaryotes and all faunal compartments (i.e.,
12

mei
ofaunal, macrofauna and megafauna; Table 1).

Nematodes dominated the
13

metazoan meiofauna (on average 90% of total abundance) and the Wieser feeding
14

classification based on nematode mouth morphology was used to designate biomass
15

to selective feeding (Wieser
type 1A + 2A), non
-
selective feeding (Wieser type 1B)
16

and omnivore/predatory (Wieser type 1B). Polychaetes dominated the macrofaunal
17

compartments and these were grouped into surface
-
deposit, deposit, suspension and
18

predatory+scavenging feeding compartment
based on standard feeding type
19

classification from Fauchald and Jumars
(1979)
. Biomass
-
dominant polychaete
20

families in the upper section are
Onuphidae

(57%) and
Sigalionidae

(36%), in the
21

middle section Spionidae (61%),

Fauveliopsidae

(9%) and
Ampharetidae

(8%), and in
22

the lower section Spionidae (40%), Goniadidae (15%) and Siboglinidae (12%). Other
23

contributions to the macrofaunal biomass from Mollusca, Bivalvia and Crustacea are
24

low (< 3%) in the upper section, higher
in the middle section with 48%, 14% and
25


13

19%, and negligible in the lower section (<1%), respectively. Finally, the megafaunal
1

surface
-
deposit feeding community consists of
Ypsilothuria bitentaculata

2

(
Holothuroidea
) and deposit feeding community of
Molpadia

musculus

3

(
Holothuroidea
).

4

Since there were no data available on the temporal variability in benthic
5

biomass, these were neglected and it was assumed that the mass balances of all
6

compartments are in steady
-
state, i.e.,
. This assumption intro
duces only
7

limited bias in the model solution
(Vézina and Pahlow, 2003)
, primarily because net
8

biomass increases
(e.g. for the fauna and bacteria) are
small as compared to the other

9

flows
in the food web
.

10

In addition to the standing stock measurements, a variety of data on process
11

rates were available for the different sections of the Nazaré canyon (Table 2). These
12

data were implemented as inequalities by setting the minimum and maximum value
13

found in each
section as lower and upper bounds, respectively.

14

The determination of prokaryotic C production in

sediment samples was
15

carried out according to the procedure

described for marine sediments by

Danovaro et
16

al.
(2002)
. Sediment subsamples

from the top 1 cm were mixed with a

solution of
3
H
-
17

leucine (final concentration 0.2 mmol L
-
1
),

were incubated at
in situ

temperature for 1
18

h
our

in the dark.

After incubation, samples were supplemented with ethanol

(80%)
19

and processed
according to
V
an Duyl and Kop

(1994)

before scintillation counting.
20

Sediment blanks were

made adding ethanol immediately after
3
H
-
leucine addition.

21

The incorporated radioactiv
ity in all samples was

measured by a liquid scintillation
22

counter.

The following equation was used for calculating prokaryotic

C production:

23

PCP ~ LI


131
.
2



(%Leu)


1



(C: protein)


ID

24


14

where PCP is prokaryotic C production, LI is the leucine

incorporation rate
1

(mol m
l
-
1

h
-
1
), 131.2 is the molecular

weight of leucine, %Leu is the fraction of
2

leucine in protein

(0.073), C
:
protein is the ratio of cellular carbon to protein

(0.86),
3

and ID is the isotope dilution assuming a value of 2.

4

The prokaryo
tic C production was determined in the top 1 cm and this value was
5

taken as lower bound on prokaryotic production (Table 2).
Prokaryote production
6

typically decreases with depth in the sediment due to reduced availability of
7

degradable detritus and electro
n acceptors
(e.g. Nodder

et al.
, 2003; Glud and
8

Middelboe, 2004)
. The upper bound on prokaryotic C production for the top 5 cm was
9

set to five times the
prokaryotic C production
of
the top 1 cm
. As such, we impose
10

that the integrated prokaryotic C production does not increase within the top 5 cm of
11

t
he sediment, because the model solution is found between the lower bound
12

(production in top 1 cm layer) and the upper bound (5 times the production in the top
13

1 cm layer). Carbon burial rates, total respiration rates, total carbon deposition and
14

burial eff
iciencies for each section were taken from the diagenetic modeling work of
15

Epping et al.
(2002)

(Table 2). We imposed that total respiration and carbon
16

deposition in Epping et al.
(2002)

did not include the respiration and uptake by
17

megafauna, respectively, because the activity of these large burrowing
or surface
-
18

dwelling organisms is missed in a diagenetic modeling approach that is based on
19

small cores incubations and oxygen profiles in the sediment.

20

An additional number of general inequality constraints were taken from the
21

literature to constrain degr
adation rates of the labile, semi
-
labile and refractory
22

detritus pools, prokaryote growth efficiency, release of DOC from the sediment,
23

assimilation efficiency of all faunal compartments, net growth efficiency of all faunal
24

compartments, production and mor
tality rates of all faunal compartments (Table 2).
25


15

Since measurements of assimilation and growth efficiencies of deep
-
sea benthos are
1

very rare, we decided to use an extensive literature review
(Van Oevelen

et al
.
,
2

2006b)

of temperate benthos as basis for these constraints. Biomass
-
specific
3

maintenance respiration of all faunal compartments was defined as 0.01 d
-
1

at 20°C
4

(see references in Van Oevelen

et al
.
, 2006b)

and is corrected with
Q10 of 2, giving a
5

temperature
-
correction factor (Tlim) for each canyon section (Table 2).

6

Benthic organisms do not feed indiscriminately on the available food sources.
7

Both surface
-
deposit and deposit
-
feeding holothurians and echinoderms ingest
8

organic m
atter with higher than ambient
chl
orophyll
a

and total

hydrolysable amino
9

acid concentrations
(Ginger

et al
.
, 2001; Witbaard

et al.
, 2001; Amaro

et al
.
, 2010)
,

10

th
ough selectivity differs between feeding modes with surface
-
depo
sit feeders
11

typically exhibiting stronger selectivity than deposit feeders
(Wigham

et al
.
, 2003)
.
12

Se
lectivity

betw
een labile detritus and semi
-
labile detritus
for megafauna
was defined
13

as the ratio of
chl
orophyll

a

concentrations in
the
gut with respect to the ambient
14

surface sediment. The level of selectivity varies from 1
to

10 for deposit feeding
15

holothurians

to >
500 for the surface deposit feeding holothurians
Amperima ro
sea

16

(Porcupine Abyssal Plain, Wigham

et al
.
, 2003)
. Selectivity at the Antarctic
P
eninsula
17

was less
eviden
t (selectivity of 2
to

7), possibly
because of
the

existence of a food
18

bank, but there was a clear separation between deposit and surface deposit feede
rs
19

(Wigham

et al
.
, 2008)
. T
herefore, no to moderate selectivity of 1
to

10
for

deposit
20

feeders and strong selectivity (50
to

100) for surface
-
deposit feeders was assumed in
21

the model (Table 2). Since no comp
arable data are available for
macro
fauna
, similar
22

selectivity ranges were defined for these
compartments

(Table 2). Finally,
f
ew
23

organisms in benthic
food webs
can be considered as
sole

predators

(Fauchald and
24

Jumars, 1979)
, therefore
the predatory meio
-
, macro
-

and
mega
faunal

compartments
25


16

were assumed
rely between

75
% and 100
% through predatory feeding

to account for
1

this

(Table 2).

2

2.5 Network indices

3

The network indices
,

and

were directly cal
culated from the
4

output of the sampling algorithm in R using the newly developed R
-
package
5

NetIndices

(Kones

et al
.
, 2009)
. Details on the calculation of the indices can be found
6

in Ulanowicz
(2004)

and Kones et al.
(2009)
, but a summary of the nomenclature
7

(Table

3) and calculation algorithms (Table 4) are included in this manuscript.


8

Network indices were calculated for the complete set of food web solutions
9

(10,000 for each section). The network indices were compared between canyon
10

sections by calculating the fr
action of which the randomized set of indices of one
11

canyon section is larger than that of another section. For example, when this fraction
12

is 0.90, this implies that 90% of the values of section 1 are larger than the ones of
13

section 2 (and consequently 10
% of the values are lower). We define differences of
14

>90% and <10% as
significant

difference and >95% and <5% as
highly
significant
15

difference.

16

Results

17

3.1 Food web structure

18

The models of the upper and middle canyon could be solved with the default
19

equal
ity and inequality constraints. However, the first attempt to solve the model of
20

the lower section with the default set of constraints was unsuccessful, which indicates
21

that some of the data embedded in the linear inverse model are in conflict with each
22

ot
her. Subsequent analysis showed that the minimum degradation of semi
-
labile
23


17

detritus (4761


8.21
∙10
-
4

= 3.9 mmol C m
-
2

d
-
1
, Table 1 & 2) was higher than the
1

maximum rates of total carbon oxidation and carbon deposition (0.90 and 1.3 mmol C
2

m
-
2

d
-
1
, respec
tively). Since the latter two data are site
-
specific field data, it was
3

decided to modify the literature bound on the minimum rate of semi
-
labile
4

degradation through pre
-
multiplication with the temperature limitation factor (Tlim =
5

0.30
, Table 2). This all
owed the model to be solved and its implications will be
6

discussed below.

7

The mean flow values and standard deviations
for the three sections of the
8

Nazaré canyon
are reported in Web appendix 1.

9

The q
uality of the model solutions was evaluated with the Coe
fficient of
10

Variation (CoV), which is the standard deviation of a flow divided by the mean flow
11

value. As such, the CoV provides an indication for the residual uncertainty in the
12

solution, where flows with a relatively large residual uncertainty have a com
paratively
13

high CoV and flows with a relatively small residual uncertainty have a comparatively
14

low CoV. All flows in all three canyon sections had a CoV that was smaller than 1.
15

Maximum CoV were
0.86
,
0.90

and
0.86

for the
upper, middle

and lower canyon
16

section, respectively and were associated with transfer of one the nematode
17

compartments to the (surface) deposit
-
feeding macrobenthos. The CoV was smaller
18

than 0.75 for 81%, 73% and 82% of the flows of the upper, middle and lower canyon
19

section, respectively, and the CoV was smaller than 0.50 for 40%, 40% and 45% of
20

the flows.

21

Total carbon input (mmol C m
-
2

d
-
1
) to the different food webs was 7.98±0.84
22

(5% labile, 75% semi
-
labile and 20% refractory detritus), 9.30±0.71 (9% labile, 89%
23

se
mi
-
labile and 2% refractory detritus) and 1.26±0.03 (6% labile, 90% semi
-
labile and
24

4% refractory detritus) for the upper, middle and lower canyon section, respectively.
25


18

Total respiration was 4.52±0.28, 5.06±0.30 and 0.86±0.02 mmol C m
-
2

d
-
1

and organic
1

ca
rbon burial was 3.05±0.80, 3.85±0.35 and 0.34±0.04 mmol C m
-
2

d
-
1

for the upper,
2

middle and lower canyon section, respectively. Prokaryotes dominated carbon
3

respiration in the upper (70%) and lower (82%) section, but their contribution to total
4

respiration

is lower (38%) than the total megafaunal respiration in the middle section
5

(57%) (Table 5). Summed meiofaunal respiration contributes 21% tot total respiration
6

in the upper, 3% in the middle and 13% in the lower canyon section, whereas summed
7

macrofaunal
respiration contributes 8% in the upper, 1% in the middle and 5% in the
8

lower section. Summed export fluxes (i.e. secondary production not consumed within
9

the food web) differed between the sections with 0.18±0.08, 0.10±0.05 and
10

0.02±0.006

mmol C m
-
2

d
-
1

f
or the upper, middle and lower section, respectively.

11

The structural differences between the food webs become apparent when
12

flows are plotted as mean net values in a circular food web structure
(Fig. 1). The
13

main differences between the upper and lower section are the more important role of
14

the non
-
selective feeding meiofauna compartment (Fig. 1A vs. 1C) and MacPS
15

compartment (Fig. 1D vs 1F) in carbon cycling in the upper canyon section. Of
16

simil
ar importance, however, is the pathway of deposition of semi
-
labile, dissolution
17

to dissolved organic carbon, prokaryotic uptake of this DOC and prokaryotic
18

respiration in the upper and lower sections (Fig. 1A vs. 1C). Consistent with their
19

comparatively l
ow contribution to total respiration, the carbon flows related to the
20

macrofaunal compartments are small, except for the MacPS compartment in the upper
21

canyon section that show up mostly in the lower row of Fig.1. The food web structure
22

of the middle canyo
n section stands out primarily because of the dominant role of the
23

MegDF and, to a lesser extent, MegSDF compartments (Fig. 1B and 1H). Moreover,
24


19

carbon cycling by the macrobenthic compartments, especially MacPS, is less
1

important as compared to the upper
and lower canyon section.

2

There is a dominance of semi
-
labile detritus in the diets of most faunal
3

compartments in the upper section of the Nazaré canyon, with semi
-
labile detritus
4

supplying between 53% and 95% of carbon of the non
-
predatory compartments
and
5

11
-
12% of the predatory compartments MeiPS and MacPS, respectively (Fig. 2A).
6

Labile detritus (2


15%) and prokaryotes (2


22%) supply a comparable lower
7

fraction of carbon to the non
-
predatory compartments and 4


5% to the predatory
8

compartments. N
on
-
predatory meiofaunal compartments fuels the meiofaunal and
9

macrofaunal predatory compartments in similar amounts (21


50%). Faunal diets of
10

the non
-
predatory compartments in the middle section are comparable to the upper
11

section, with a dominance of se
mi
-
labile detritus (42


93%) and labile (2


21%)
12

detritus (Fig. 2B). The diet contribution of prokaryotes to non
-
predatory faunal
13

compartments varies between 2 and 21%. Dependence on selective and non
-
selective
14

feeding meiofaunal compartments is highest
for predatory meiofauna (80%), followed
15

by predatory macrofauna (48%) and <10% for the other macrofaunal and megafaunal
16

compartments. The diet of the predatory/scavenging macrofaunal compartment is
17

diverse, with no clear dominance of any resource (3


25%)
.

18

The diet compositions in the lower section of the Nazaré canyon resemble
19

overall those of the upper section (Fig. 2A vs. 2C). Again, semi
-
labile detritus is most
20

important (between 76


98%) in the diets of non
-
predatory faunal compartments.
21

Diet contrib
utions of labile detritus and prokaryotes are similar for

selective feeding
22

meiofauna (9
-
10%), non
-
selective meiofauna (each 1%), predatory/omnivore
23

meiofaunal (each 5%), surface
-
deposit feeding macrofauna (each 5%), deposit
-
24

feeding macrofauna (each 1%) an
d predatory/scavenging macrofauna (4
-
5%) (Fig.
25


20

2C). The meiofaunal compartments MeiSF + MeiNF are important resources for the
1

meiofaunal predators/omnivores (together 80% of the diet)

and predatory (69%)
2

macrofauna, but are of lesser importance for surface
-
deposit (10%), deposit feeding
3

(1%). The diet composition of predatory/scavenging macrofauna is diverse though
4

with a high importance of selective feeding meiofauna (54%) and lower contributions
5

ranging from 1
-

11% from other resources.

6

The diet of suspe
nsion
-
feeding macrofauna is similar among the canyon
7

sections and is partitioned among labile (32


36%) and semi
-
labile (64


68%)
8

detritus from the water column.

9

The dominant fate of prokaryotic production in all three sections is mortality
10

(52


88%) an
d grazing by meiofauna in the upper canyon section (31%) and by
11

megafauna in the middle section (36%) (Fig. 3A
-
C). The majority of the meiofaunal
12

secondary production is grazed by macrofauna in the upper (56%) and lower (47%)
13

canyon section, while megafaun
al grazing is important in the middle section (36%)
14

and grazing by meiofauna (MeiPO) is important with a consistent contribution of 18


15

23% in the three sections (Fig. 3D
-
F). The fate of macrofaunal production is
16

partitioned similarly in all three canyon

sections with maintenance representing 22


17

24%, mortality 29


34%, predation by macrofauna (MacPS) 2


20% and export 29


18

42% (Fig. 3G
-
I). The fate of megafauna is dominated by maintenance respiration
19

(91%) and with limited contributions of mortality
(5%) and export (4%) (Fig. 3J).

20

3.2 Network indices

21

The network indices total system throughput (
), Finn cycling index (
)
22

and average mutual information (
) were calculated for the three
section
s (Fig. 4)

23

and compared
(Table 6).
The

does not differ significantly
between the upper and
24


21

middle
sections

with median values of 41.
1 and

39.
7

mmol C m
-
2

d
-
1
,
respectively,
but
1


is
significantly

lower in the lower
section

with a median of
6.7
mmol C m
-
2

d
-
1

2

(Table 6). Differences in

are highly significant between
canyon sections

(Table
3

6)
and
median

values are
0.13
,
0.06

and
0.17

for the upper, middle and lower
section
,
4

respectively.


is
not significantly different

between

the
upper (
median of
2.21)
5

and middle (2.2
2
)
canyon section,
but
significantly
lower for the lower
section

(2.
12
).


6


22

Discussion

1

In this paper, we present the first quantitative analysis of carbon flows within
2

food webs of different sections of a submarine cany
on. This provides a unique
3

opportunity to study how different characteristics within a canyon influence food web
4

structure and attributes such as total system throughput, recycling within the food web
5

and food web maturity. The modeled food webs of the upp
er, mid and lower canyon
6

sections are based on a large variety of site
-
specific biological and biogeochemical
7

data and are combined with physiological constraints and empirical relations from the
8

literature. Despite the large amount of data that are implem
ented, this is insufficient to
9

uniquely quantify all carbon flows
(Van Oevelen

et al
.
, 2010)
. This implies that a
10

“solution space” exists, within which an infinite number of solutions are present that
11

are consistent with the data
(Soetaert and Van Oevelen, 2009)
. Conventional single
-
12

solution modeling approaches typically find a final solution at or clo
se to boundaries
13

of the solution space, making the final solution
sensitive to the

exact

boundaries of the
14

solution space
(

Vézina

et al
.
, 2004; Kones

et al
.
, 2006; Van Oevelen

et al
.
, 2010)
.
15

The

multi
-
solution approach followed here, samples the solution space
(Van den
16

Meersche

et al
.
, 2009)

such that the mean of this sampled set represents the best
17

central flow value that is less sensitive to the boundaries of the solution space
(Van
18

Oevelen

et al
.
, 2010)
. Moreover, the standard deviation on each carbon flow indicates
19

how the
uncertainty
in the data set
propagates
to an uncertainty on its value
(Van
20

Oevelen

et al
.
, 2010)
. The Coefficient of Variation (CoV) was smaller than 0.75 for
21

73


82% flows in the three sections (Web appendix), which ind
icates that the
22

residual uncertainty on the flows is comparatively low and that the food web is well
-
23

constrained. The lowest CoVs are associated with the respiration flows of the biotic
24

compartments, whereas highest CoVs are predominantly associated with c
arbon flows
25


23

that exist between biotic compartments. This directly relates to the data availability.
1

The carbon requirement of faunal compartments is constrained primarily by the
2

available biomass data. There are however few data that constrain the origin o
f this
3

carbon, such that the residual uncertainty on diet contributions and fates of secondary
4

production are comparatively high. Perhaps even more important than the residual
5

uncertainty on the flows, are the limitations and uncertainties with respect to
the
6

assumptions that were needed to setup the model. These sources of uncertainty mainly
7

concern substrate heterogeneity and combining different data sets and will be
8

discussed now.

9

The seafloor in the Nazaré canyon is heterogeneous and consists of rocks,
10

boulders, coarse gravel sediments, steep walls, a highly dynamic thalweg and terraces
11

consisting of soft
-
sediments. The hard substrata may be draped with a thin soft muddy
12

layer. Not surprisingly, also the associated fauna changes with substratum type and
13

condition. Rocky surfaces for example are dominated by suspension feeders such as
14

hard and soft corals, gorgonians, anemones, sea pens and crinoids
(Tyler

et al
.
, 2009)
.
15

In thalweg sediments, the biomass of nematodes
(Garcia

et al.
, 2007)

is about one
16

order of magnitude lower than in soft
-
sediment

terraces
(Ingels

et al
.
, 2009)
, which is
17

attributed to repeated sediment disturbance of thalweg sediments that prevents the
18

development of a mature nematode community
(Garcia

et al
.
, 2007)
. In addition,
19

megafauna and the
giant
epifaunal
protozoans

(xenophyophores) were not observed in
20

the thalweg
(Tyler

et al
.
, 2009)

but are f
ound outside the thalweg. Up to now, there
21

are no quantitative data available on the biomass and activity of the filter
-
feeding
22

community in the Nazaré canyon on rocky substrata. Moreover, quantitative data on
23

the faunal community in the thalweg is only sp
arsely available and its food web
24

structure is not representative for that of large sections of the canyon. Hence, in this
25


24

study we restricted our analysis to the soft
-
sediments of the terraces adjacent to the
1

thalweg and excluded other substrate types. Th
is implies for example that we may
2

miss the potentially high carbon processing activity associated with the canyon walls.
3

In terms of areal coverage however, these soft
-
sediments with net mud deposition
4

represent an appreciable ~70% of the total surface ar
ea of the canyon
(Masson

et al.
,
5

2010)
, such that a significantly large part of the Nazaré canyon is addressed here.

6

One compartment that is not included in the food web is
Foraminifera, which
7

are protozoans that are typically of meiofaunal size but can occur as giant epifauna
8

(xenophyophores). Meiofaunal foraminifera
(Koho

et al.
, 2008)

and epifaunal
9

xenophyophores
(Tyler

et al
.
, 2009)

have a high abundance in especial
ly the muddy
10

terraces with stable redox conditions and low disturbance. Foraminifera have been
11

shown to play an important role in the initial processing of fresh phytodetritus under
12

deep
-
sea conditions
(Moodley

et al
.
, 2002)

al
though their contribution

may also be
13

more limited
(Woulds

et al
.
, 2007)
. Moreover, their contribution to total respiration in
14

continental shelf sediments was recently found to be limited to <3%
(Geslin

et al
.
,
15

2010)
. Unfortunately, the available abundance data could not be converted

to biomass
16

with reasonable accuracy, and since biomass is essential to constrain their activity in
17

the food web we therefore decided to omit this compartment in this analysis.

18

The site
-
specific data that we include in this study were lumped into the three

19

canyon sections (Table 1 and 2). However, since deep
-
sea research is time
20

consuming, conducted over large spatial areas and depends on ship time availability
21

and meteorological/sea conditions, the data were not collected synoptically.
22

Inevitably, this dat
a ‘lumping’ into canyon sections will introduce errors in the food
23

web analysis linked to the spatial and temporal variability of the data collected.
24

Nevertheless, the Nazaré canyon is comparatively well
-
studied and one of the
25


25

strengths of linear inverse m
odeling is that datasets are merged and tested for internal
1

consistency
(Van Oevelen

et al
.
, 2010)
. Given the amount of da
ta in the models
2

(Table 1 and 2), the inverse model analysis at least showed that the different data sets
3

are consistent.
The only exception
was

that the minimum degradation rate of semi
-
4

labile detritus
in the lower canyon section
was higher than the maxim
um rates of
5

carbon oxidation and total carbon deposit
ion. The carbon oxidation and deposition
6

data are site
-
specific data and were therefore maintained. Instead, the minimum bound
7

on semi
-
labile degradation was reduced by multiplication with the temp
eratur
e
8

limitation factor, which allowed solving the food web model. Several explanations
9

may apply here. First, water temperature in the deep canyon section is about 2.5°C
10

and

lowest of the three sections. This low temperature may cause degradation to
11

proceed s
lower than in the higher
section
s of the canyon with comparatively higher
12

wat
er temperatures. Moreover, the quality of the semi
-
labile detritus may have
13

decreased during transport through the canyon and this may also lower the
14

degradation rates further. De
spite this minor adaptation that was needed, the results
15

from the present analysis serve as a significant first step in gaining insight in the food
16

web structure of submarine canyons.

17

4.1 Upper canyon section

18

The dynamic upper canyon receives about 8
±0.84

mmol C m
-
2

d
-
1
, which is
19

lower than the 15


23 mmol C m
-
2

d
-
1

that is predicted using an empirical relation for
20

continental shelf sediments
(i.e. summed burial and mineralization rates at 700 and
21

300 m, respectively, Middelburg

et al.
, 1997)
. However, carbon inputs at the open
22

slope sediments of the adjacent Iberian margin are substantially lower than pre
dicted
23

by the empirical relation by Middelburg et al.
(1997)

and are between 2.3 and 4.3
24

mmol C m
-
2

d
-
1

(Epping

et al
.
, 2002)
. Thus, carbon inputs to the upper canyon section
25


26

is higher those of adjacent slopes, but not ex
tremely high as compared to other slope
1

sediments. Burial rates in the upper and middle canyon are substantial flows in the
2

food web (Fig. 1A, B), but burial efficiencies are comparable to Iberian open slopes
3

and relate to sediment accumulations rates
(Epping

et al
.
, 2002)
. Hence, the efficiency

4

with which the food web processes organic carbon is similar to open slope sediments.

5

The model results allow detailed deciphering of the biotic compartments that
6

are responsible for carbon processing within the canyon. Woulds et al.
(2009)

used
7

the results
of isotope tracer experiments from different slope sediments to define
8

different categories of biological C
-
processing. In this categorization, the “active
-
9

faunal
-
uptake” category contains mostly shallow (<300 m) slope sediments and is
10

characterized by 10


25% metazoan uptake. This category matches best with the
11

upper canyon section that has a faunal contribution of ~40% and bacterial
12

contribution of 60% to total carbon assimilation.

13

The faunal contribution to total respiration and carbon processing typic
ally
14

decreases with increasing water depth and associated decrease in carbon input
(Heip

15

et al
.
, 2001; R
owe

et al
.
, 2008; Woulds

et al
.
, 2009)
. Henceforth, the high faunal
16

contribution in the upper canyon section is probably related to the h
igher OM content
17

and quality as compared to slope sediments at comparable water depth
(Garcia

et al
.
,
18

2007; Garcia and Thomsen, 2008; Pusceddu

et al
.
,
2010)
. One striking difference
19

however is that meiofauna dominated faunal processing and contributed around 33%
20

of the total carbon assimilation in the upper canyon section, which is muc
h higher
21

than in open slopes sediments included in the overview of Woulds et al.
(2009)
. This
22

high contribution also translates into a much higher meiofaunal respiration at 21% of
23

total respiration in the upper section of the Nazaré canyon as compared to oth
er open
24


27

slopes that vary from 4


8%
(Piepenburg

et al
.
, 1995; Heip

et al
.
, 2001; Soetaert

et
1

al
.
, 2009)
.

2

Rowe et al.
(2008)

and Bagulay et al.
(2008)

report even substantially higher
3

contributions ranging from ~20 up to 51% for the Northern Gulf of Mexico. Their
4

estimates are based on biomass
-
specific

respiration rates of 0.04 to 0.11 d
-
1

at a
5

temperature of 4


5°C. Moodley et al.
(2008)

used a novel micro
-
respiration system
6

and reported specific rates of 0.021 to 0.032 d
-
1

for intertidal (20°C) Nematoda,
7

Ostracoda and Foraminifera over a biomass range of 0.7 to 5.2 μC ind
-
1
. Nematodes
8

from the Gulf of Mexico are smaller
(~0.1μC ind
-
1, Baguley

et al
.
, 2008)
, but specific
9

respiration rates are still fairly high as compared to these intertidal meiofauna. The
10

high meiofaunal contribution to tota
l community respiration is therefore probably also
11

related to the comparatively high biomass
-
specific respiration rates that are estimated
12

for the Gulf of Mexico. Clearly more experimental work for especially small
13

nematodes at lower temperatures is needed

to better constrain these respiration rates.

14

The carbon sources that are consumed by meiofauna to fuel these respiration
15

rates are detritus and prokaryotes
(e.g., Rowe

et al
.
, 2008, this study)
. Stable isotope
16

tracer experiments allow direct quantification of labile food assimilation rates of
17

amongst others

meiofauna. Intriguingly, these results typically show low biomass
-
18

specific assimilation rates of <0.01 and mostly <0.001 d
-
1

(

Moens

et al.
, 2007; Franco

19

et al
.
, 2008; Ingels

et al.
, 2011;)
, a limited (<5%) contribution to
13
C uptake by
20

metazoan meiofauna on open slope
(Moodley

et al
.
, 20
02)

and abyssal plain
(Witte

et

21

al
.
, 2003)

sediments and negligible bacterivory by nematodes in a slope sediment
22

(Guilini

et al
.
, 2010)
. Irrespective of the labeled substrate or setting, meiofauna
23

consistently show an uptake of labile
13
C carbon that seems to be in imbalance
with
24

carbon requirements as estimated from biomass
-
specific respiration rat
es. This is not
25


28

in contrast with the meio
faunal

diet composition as inferred for

the Nazaré canyon
1

(Fig. 2), where semi
-
labile detritus (a carbon source not used in isotope tracer s
tudies)
2

is the dominant component. This dominance of semi
-
labile detritus in their diet would
3

explain the low labeling of metazoan meiofauna (dominated by nematodes) in isotope
4

tracer studies. It also agrees with Soetaert et al.
(1997)
, who found a strong positive
5

correlation between depth profiles of nematodes and or
ganic N content and suggested
6

that the concentration of lower quality food primarily determines nematode depth
7

distribution.

8

The elevated OM input in the upper canyon section combined with
9

hydrodynamic conditions with current speeds of up to 30


40 cm s
-
1
appear to
10

particularly favor meiofauna, whereas macro
-

and megafauna have a lower
11

contribution to carbon processing as compared to open slope sediments. As a result,
12

meiofaunal biomass in the upper canyon section rank among the highest reported in
13

marine

sediments
(Rex

et al.
, 2006)
, whereas macrofaunal bio
mass is comparatively
14

low.

15

Prokaryotes are responsible for the dominant part of carbon cycling and
16

respiration in the upper canyon section (Fig. 1 and Table 5).

An important pathway,
17

also seen in the middle and lower canyon section, is deposition of semi
-
labile detritus,
18

dissolution to dissolved organic carbon, to prokaryotic uptake of this DOC and
19

subsequent prokaryote respiration. A dominance of prokaryotes in carbon cycling and
20

respiration is commonly found in continental shelf sediments
(Canfield

et al
.
, 1993;
21

Piepenburg

et al
.
, 1995; Heip

et al
.
, 2001
; Rowe

et al
.
, 2008)
. Hence, it appears that
22

hydrodynamic conditions in the upper
canyon act predominantly on carbon
23

partitioning between faunal compartments rather than on the partitioning between pro
-

24

and eukaryotes.

25


29

4.2 Middle canyon section

1

Soft
-
sediment terraces in the middle section of the canyon experience high
2

sedimentation rates
(de Stigter

et al
.
, 2007; Tyler

et al.
, 2009; Masson

et al
.
, 2010)
,
3

which is accompanie
d by an input of organic matter of 9.30±0.71 mmol C m
-
2

d
-
1

that
4

is comparable to the upper canyon section. These high OM inputs clearly show that
5

the archetypical picture seen in open slope sediments that biomass, respiration and
6

carbon processing decreas
es with increasing water depth does not necessarily hold for
7

submarine canyons.

8

With respect to the carbon partitioning within the food web, the middle
9

canyon section seems to fall in the “metazoan
-
macrofaunal
-
uptake
-
dominated”
10

category, a category that is

typically found in shelf and upper slopes, with a
11

comparatively high macrofaunal biomass
(Woulds

et al
.
, 2009)
. An importanct
12

discrepancy with the categorization by Woulds et al. is that faunal carbon processing
13

in the middle canyon is not dominated by macrofauna, but by s
urface deposit
-
feeding
14

and deposit
-
feeding megafauna (i.e. the holothurians
Ypsilothuria bitentaculata

and
15

Molpadia musculus
, respectively
). The megafaunal importance is also apparent in
16

community respiration (57%) and export of secondary production from
the food web
17

(79%).

18

De Leo et al.
(2010)

reported recently for the Kaikoura Canyon (New
19

Zealand) an extremely high biomass of 89
±18 g C m
-
2

of megafauna (dominated by
20

M
.
musculus
) in low relief, muddy and accreting sediments at 900


1100 m of water
21

depth. Megafaunal biomass in the middle section of the Nazaré canyon is about an
22

order of magnitude lower (6.2 g C m
-
2
), but still 2


3 orders of magnitude higher than
23

found in open slopes at comparable depth
(Rex

et al
.
, 2006)
.

24


30

Amaro et al.
(2010)

conducted trophic studies on the holothurian
M
.

musculus

1

and estimated removal rat
es of 0.5 gC of semi
-
labile detritus m
-
2

d
-
1
. Our food web
2

analysis even suggests higher removal rates of 2.5

gC of semi
-
labile detritus m
-
2

d
-
1
,
3

showing that this
holothurian
can have an important impact on the sedimentary food
4

web
.
Amaro et al.
(2010)

also inferred that
prokaryotes delivered <0.1% of the
5

assimilated proteins and it was concluded that

holothurians do not appear to rely on
6

microbes for direct nutrition. This is also supported by our diet reconstruction of
7

deposit
-
feeding megafauna (i.e.,
M
.

musculus
), where prokaryotes play only a
8

marginal role (Fig. 2B).

9

Carbon partitioning with the f
ood web of the middle canyon section at
2
700



10

4000 m is comparable to much shallower shelf and upper
-
slope sediments, where also
11

an important faunal contribution is typically found. The large faunal contribution in
12

the middle canyon section is due to the
comparatively high input of OM, which is
13

quantitatively comparable to the upper canyon section. It is however unclear why
14

canyon
-
specific conditions in the middle section are particularly beneficial for
15

(surface) deposit
-
feeding holothurians as compared to

for example macrofaunal
16

polychaetes. The deposit
-
feeding megafauna consist predominantly of the holothurian
17

head
-
down feeder
M
.

musculus

and there was no evidence for a specialized
18

prokaryotic community in the guts of
M.

musculus

that may aid in the hydro
lyzation
19

of organic matter
(Amaro

et al
.
, 2009)
. Other possible explanations

for a strong
20

proliferation of
M
.

musculus

in soft accreting sediments within canyons may involve

a
21

better adaptation to high sediment rates, enhanced trapping of the depositing organic
22

matter in their feeding pits and negative feedbacks on macrofauna thro
ugh, for
23

example, predation or sediment disturbance.

24


31

4.3 Lower canyon section

1

The food web structure in the lower canyon section is markedly distinct from
2

the upper and middle sections (Fig. 1). Not only is total carbon input (1.26±0.03
3

mmol C m
-
2

d
-
1
) abo
ut an order of magnitude lower than in the upper and middle
4

sections, but also its partitioning within the food web differs considerably. OM input
5

in the lower section is lower, because
OM
delivery
from the upper and middle canyon
6

section
is
less frequent,

OM has been degraded during transport through the canyon
7

and the
lower canyon begins where the
V
-
shaped valley
widens into a kilometers
-
8

wide channel
thereby lowering the OM input per surface area.

9

Respiration in the lower canyon section is strongly dominated by protozoa
10

(82% of total respiration) whereas the faunal compartments each respire <10%. These
11

characteristics place the lower canyon section in the “respiration
-
dominated”
12

category, in which m
ost OM is respired by the prokaryotic community and the role of
13

benthic fauna in carbon cycling is low
(Woulds

et al
.
, 2009)
. Other sites that fall in
14

this category are lower slope sediments and abyssal plains
(Woulds

et al
.
, 2009)
,
15

suggesting that the benthic food of the lower canyon sectio
n resembles others sites at
16

similar depth . The lower canyon section seems to be less influenced by canyon
17

conditions as compared to the upper and middle section of the canyon.

18

4.4
C
omparison of canyon sections with network indices

19

The lower carbon process
ing in the lower canyon is also evident in the index
20

total system throughput (
), in which carbon flows are summed to obtain a measure
21

of total food web activity
(Ulanowicz, 2004)
. Total system throughput does not differ
22

significantly betwe
en the upper and middle sections (medians of
41.
1 and
39.
7 mmol
23

C m
-
2

d
-
1
, respectively), but is significantly lower in the lower canyon section (median
24


32

of 6
.
7 mmol C m
-
2

d
-
1
) (Table 6). Though community respiration and OM input is
1

higher for the middle ca
nyon section, total system throughput is slightly elevated (not
2

significantly) in the upper canyon section. This reversal in activity measures is
3

probably linked to the low recycling within the food web of the middle canyon as
4

quantified with the Finn cycl
ing index (Fig. 4B). This index summarizes the fraction
5

of total carbon cycling that is generated
by recycling processes
(Allesina and
6

Ulanowicz, 2004)
. Significant differences in recycling are found between the canyon
7

s
ections, with the most notable difference being low recycling in the middle canyon
8

section. One explanation relates to the viral shunt
(Danovaro

et al.
, 2008)
, in which
9

viral infection cause lysis of prokaryotes and the subsequent release of dissolved
10

organic matter that is again recycled by other heterotrophic prokaryotes
(e.g., Van
11

Oevelen

et al
.
, 2006a)
. Prokaryotes dominate carbon flows in the lower section, but
12

this dominance is reduced in the upper and particularly the middle canyon section. If
13

the viral
-
mediated shunt significantly
influences the FCI, this would explain the
14

decreasing FCI when going from the lower, upper to the middle canyon section. To
15

examine the impact of the viral shunt on the FCI, the viral shunt was eliminated from
16

the food web by only including the net fl
ow fr
om DOC to prokaryotes in the FCI
17

calculations. Though differences in FCI remain, the FCI of the upper and lower
18

sections

drops to medians of 0.07 and 0.04, respectively, whereas the middle
section

19

is
much
less affected with a drop to 0.03. This
exercise cl
early
shows that th
e viral
20

shunt increases carbon recycling in benthic food webs rendering recycling to be
21

higher in prokaryote
-
dominated food webs as compared to faunal
-
dominated food
22

webs.


23

The index average mutual information (AMI) gauges the developmen
tal status
24

of an ecosystem in the sense that while food webs develop, trophic specialization will
25


33

result in higher values for AMI
(Ulanowicz, 2004)
. The AMI

is that part of the flow
1

diversity
(i.e. the Shannon index applied to flow diversity, Ulanowicz, 2004)

that
2

quantifies how orderly
and

coherently

carbon
flows

are

inter
-
co
n
nected
. Since the
3

AMI is claimed to assess the developmental status of an ecosystems it is interesting to
4

as
sess whether differences in the food web structures are also reflected in the AMI
5

index. More specifically, we had expected the less
-
disturbed lower canyon section to
6

have highest AMI values with decreasing values going up
-
canyon. Differences in
7

AMI betwee
n the upper and middle canyon are non
-
significant (Table 6), though
8

large differences exist in environmental conditions and food web structure. The AMI
9

is significantly lower in the lower canyon section though this section is less impacted
10

by canyon condit
ions as compared to the other two sections.

Tobor
-
Kaplon et al.

11

(2007)

quanti
fied the AMI of soil food webs that were exposed to different stress
12

levels (i.e. pH and copper) and concluded that AMI appeared useful as an indicator of
13

environmental stress at the ecosystem level. For the benthic food webs analyzed here
14

however, there d
oes not seem to be a straightforward relation between AMI and
15

environmental stress. On the other hand, there is another important factor that
16

influences food web structure when going down
-
canyon, namely the reduced OM
17

input. To verify the usefulness of AMI

as a stress indicator it is therefore necessary to
18

compare the AMI of marine benthic food webs at similar levels of OM input, but
19

different levels of environmental stress.

20

In conclusion, benthic food web structures in the upper, middle and lower
21

sections

of the Nazaré canyon were shown to be influenced by the conditions in the
22

particular canyon section. The OM input in the upper and middle canyon sections is
23

elevated as compared to those of the surrounding open slope sediments and this
24

resulted in a highe
r contribution of fauna in carbon processing as compared to open
25


34

slope sites at similar water depth. The compartments that were responsible for the
1

faunal processing were strongly influenced by

conditions in the particular canyon
2

section. In the upper cany
on section, a dominance of meiofauna in faunal carbon
3

processing was evident, whereas a high faunal contribution to carbon processing in
4

open slope sediments is typically dominated by macrofauna. It is proposed that
5

hydrodynamic disturbance and resulting s
ediment resuspension in the upper canyon
6

shifts the balance towards the meiofauna. In contrast, the food web of the accreting
7

sediments in the middle canyon showed a completely different pattern where carbon
8

processing was dominated by the megafaunal holot
hurians. Our study confirms that
9

accreting sediments in canyons can be hotspots of megafaunal biomass and
10

production and megafauna can greatly influence carbon processing. The food web
11

structure of the lower canyon section resembled that of lower slope and

abyssal plain
12

sediment, where carbon processing is dominated by prokaryotes. The influence of the
13

canyon
-
specific processes seems to vanish in the deeper sections where the Nazaré
14

canyon widens and enters the abyssal plain. In all canyon sections, a domin
ance of
15

semi
-
labile detritus in the diet of (surface) deposit feeders is suggested. These results
16

are supported by stable isotope tracer (for meiofauna) and gut transformation
17

(holothurian
M.

musculus
) studies. This study shows that elevated OM input in
18

ca
nyons may favor the faunal contribution to carbon processing and creating hotspots
19

of faunal biomass and carbon processing along the continental shelf.

20

Acknowledgements

21

This research was supported by the HERMES project (contract

22

GOCE
-
CT
-
2005
-
511234), funde
d by the European Commission’s Sixth Framework
23

Programme under the priority “Sustainable Development, Global Change and
24


35

Ecosystems”, and HERMIONE project (
grant agreement n° 226354"
)
fund
ed by
the
1

European Community's Seventh Framework Programme (FP7/2007
-
2013)
. This is
2

publication
5018

of the Netherlands Institute of Ecology (NIOO
-
KNAW), Yerseke.


3


36

References

1

Aller, J.Y., Aller, R.C., 2004. Physical disturbance creates bacterial dominance of
2

benthic biological communities in tropical
deltaic environments of the Gulf of
3

Papua. Continental Shelf Research 24 (19), 2395
-
2416.

4

Allesina, S., Ulanowicz, R.E., 2004. Cycling in ecological networks: Finn's index
5

revisited. Computational Biology and Chemistry 28 (3), 227
-
233.

6

Amaro, T., Bianchell
i, S., Billett, D.S.M., Cunha, M.R., Pusceddu, A., Danovaro, R.,
7

2010. The trophic biology of the holothurian
Molpadia musculus
: implications
8

for organic matter cycling and ecosystem functioning in a deep submarine
9

canyon. Biogeosciences 7 (8), 2419
-
2432.

10

Amaro, T., Witte, H., Herndl, G.J., Cunha, M.R., Billett, D.S.M., 2009. Deep
-
sea
11

bacterial communities in sediments and guts of deposit
-
feeding holothurians in
12

Portuguese canyons (NE Atlantic). Deep
-
Sea Research Part I
-
Oceanographic
13

Research Papers 56 (10)
, 1834
-
1843.

14

Baguley, J.G., Montagna, P.A., Hyde, L.J., Rowe, G.T., 2008. Metazoan meiofauna
15

biomass, grazing, and weight
-
dependent respiration in the Northern Gulf of
16

Mexico deep sea. Deep
-
Sea Research Part II
-
Topical Studies in Oceanography 55
17

(24
-
26), 2
607
-
2616.

18

Bianchelli, S., Gambi, C., Zeppilli, D., Danovaro, R., 2010. Metazoan meiofauna in
19

deep
-
sea canyons and adjacent open slopes: A large
-
scale comparison with focus
20

on the rare taxa. Deep Sea Research Part I: Oceanographic Research Papers 57
21

(3), 42
0
-
433.

22

Burdige, D.J., Berelson, W.M., Coale, K.H., McManus, J., Johnson, K.S., 1999.
23

Fluxes of dissolved organic carbon from California continental margin sediments.
24

Geochimica Et Cosmochimica Acta 63 (10), 1507
-
1515.

25


37

Canals, M., Puig, P., de Madron, X.D.,

Heussner, S., Palanques, A., Fabres, J., 2006.
1

Flushing submarine canyons. Nature 444 (7117), 354
-
357.

2

Canfield, D.E., Jorgensen, B.B., Fossing, H., Glud, R., Gundersen, J., Ramsing, N.B.,
3

Thamdrup, B., Hansen, J.W., Nielsen, L.P., Hall, P.O.J., 1993. Pat
hways of
4

organic
-
carbon oxidation in three continental
-
margin sediments. Marine Geology
5

113 (1
-
2), 27
-
40.

6

Danovaro, R., Bianchelli, S., Gambi, C., Mea, M., Zeppilli, D., 2009. alpha
-
, beta
-
,
7

gamma
-
, delta
-

and epsilon
-
diversity of deep
-
sea nematodes in can
yons and open
8

slopes of Northeast Atlantic and Mediterranean margins. Marine Ecology
-
9

Progress Series 396, 197
-
209.

10

Danovaro, R., Dell'Anno, A., Corinaldesi, C., Magagnini, M., Noble, R., Tamburini,
11

C., Weinbauer, M., 2008. Major viral impact on the functio
ning of benthic deep