Search is not available for this dataset
text
stringlengths 4.99k
445k
|
|---|
# Introduction
Since the early days of MRI, a constant drive to higher main magnetic field
strengths (B<sub>0</sub>) has existed. While the increase in B<sub>0</sub> leads
to a superlinear increase in SNR at higher magnetic fields, the wavelengths of
the electromagnetic fields at the corresponding Larmor frequencies decrease. At
ultra-high field (≥ 7T), the wavelength in tissue is shorter than the diameter
of a human torso. Consequently, substantial wave effects can give rise to
inhomogenous spin excitations, which can lead to contrast artifacts and complete
signal dropouts in the field of view. Not only do these inhomogeneities affect
the SNR, they also lead to varying contrast throughout the field of view,
severely reducing the clinical utility of the resulting images.
To cope with the challenges introduced by the decreased wavelength at 7T and
above, many different techniques have been proposed. Examples of these
techniques are adiabatic pulses, RF shimming, kT-points, 2D spokes, 3D tailored
radiofrequency pulses, Transmit SENSE, and TIAMO. Although there are many
differences between these techniques, most of them require a multi-channel
transmit system. Moreover, the obtainable transmit power efficiency, specific
absorption rate (SAR) efficiency, and pTx speed-up factor improve as the number
of independent transmit channels increases.
Using 8- and 16-channel transmit systems, an increasing number of clinical
feasibility studies at 7T have aimed at regions where imaging is hampered by
severe transmit inhomogeneities, for example liver, kidneys, prostate,
female pelvis, the hip joints, shoulder, the heart, the spinal cord, and
the breasts, or they cover extended regions such as the complete lower
extremities.
Most of these studies use local transmit arrays to enhance transmit efficiency,
and many of these studies report a need for higher transmit power than what was
available (generally 8 kW) despite the use of a local array. Such arrays that
are placed either directly on or very close to the patient take up significant
space in the patient bore of the magnet, reducing the maximum patient size. To
get to a more clinical workflow, where a transmit body coil is placed further
away from the patient, several stand-off arrays and arrays integrated into the
bore have recently been proposed.
In this paper we present a 32-channel transmit system add-on for 7T, including
modulator system, power amplifiers, and SAR supervision system. The 32-channel
body array is integrated into the MRI system by placing it in the gap between
the gradient coil and liner of the patient tunnel, similar to body coils at
lower field strengths and thereby enabling a workflow very close to conventional
clinical MRI.
# Methods
## Ethics approval
For in-vivo measurements, informed consent was obtained from both participants
included in the study. In vivo studies were approved by the institutional review
board of the University Clinic Essen. Approval nr. 16 7214BO.
## Original MR system
The vendor-provided MR System in this study is a Magnetom 7T with an AS095
gradient coil (Gradient strength 38 mT/m, slew rate 200 mT/m/ms; Siemens
Healthineers, Erlangen, Germany). The nominal field of view is 50 cm in all
directions, and the system has 32 receivers in total. It is equipped with the
“Step 2” 8-channel transmit system, which was not used in this work.
## Overview of system add-on
shows a schematic overview of the add-on system. The system only uses 4 input
signals for synchronization with the original system, three digital signals and
one analog signal. The digital signals are 1 bit for transmit or receive state,
1 bit for the unblank signal for the transmit amplifiers, and 1 bit as a trigger
from the sequence to set the modulators to the next pre-programmed state. The
analog signal is the RF signal coming from the exciter, which is the signal that
would be sent to the power amplifier in the single-channel mode of the original
system.
The signal from the exciter is split up and fed to the modulators. In the
original single-channel system, this signal is directly fed into the vendor
power amplifiers that are combined into a single high-power output. The output
of the modulators is connected to the power amplifiers in the magnet room. The
high-power signals from the amplifiers are fed to the RF coil via directional
couplers (DiCos) and transmit/receive (T/R) switches.
To be able to switch between reception with local coils or the body array, the
receive chain of the system is modified to include RF switches. This is the only
change to the vendors transmit or receive path.
The RF signals for forward and reflected power are taken from the DiCos and fed
into a multiplexing power supervision unit (MUX). This unit feeds a portion of
the forward power into the receive chain and sends a logarithmic baseband
voltage signal to the operating PC system for amplitude supervision.
The operating PC is used to pre-program the modulators with a set of states
before a sequence is run. Using the trigger from the sequence, the modulators
are successively switched to the next state.
The logic controller receives the information about whether the MR system is in
transmit or receive mode as well as the unblank signal and distributes the
signals across the add-on system.
## Modulators
The custom-built modulator units use AD8345 (Analog Devices, Norwood, MA, USA)
IQ modulators to modify the exciter signal. The exciter signal is fed into the
local oscillator input at a very low power (\<-30 dBm), to ensure that changes
in the input are transferred linearly to the output, since in normal operation
the input power for the AD8345 is far beyond the 1 dB compression point. The
analog input voltages for I and Q are provided by MAX5100 8-bit digital-to-
analog converters (DACs) (Maxim Integrated Products, Sunnyvale, CA, USA). These
DACs have an input register that allows programming them to a new value before
switching them to that new value by a trigger signal. Due to the rise time of
the output operational amplifiers of the MAX5100, output settling time is up to
3 μs, depending on the degree of change of the outputs. The modulator system is
divided into two half systems with 16 modulators each. Each half system is
connected to a 16-bit parallel bus provided by a digital I/O card. On a trigger
from the pulse sequence of the MRI system, a new bit pattern is loaded into the
input register of the DACs. After all input registers are filled with the new
pattern, a trigger is issued to simultaneously move the data from the input
register to the DACs, changing the modulator state. To account for delays, the
first trigger from the sequence is sent 10 μs before the start of the RF pulse.
Adjustment of the data rate of the bus allows to fine tune the delay between the
trigger from the MRI pulse sequence and the point in time when the DACs change
their output.
To account for small offsets in the I and/or Q channel and other systematic
errors after calibration, the modulation strategy for pTx is performed as
follows. Instead of playing out a rectangular pulse on the exciter and
performing the modulation completely with the modulators (“Strategy 1”), the
exciter plays out a signal in which each sample is the maximum amplitude across
all channels. In this way, a variable dynamic range is achieved for the 8-bit
modulators. Errors due to offset in the I and/or Q channel are further decreased
by shifting the exciter phase by 180° for each sample, which is accounted for in
the phases the modulators apply (“Strategy 2”). Changing the phase each sample
by 180° leads to alternatingly adding and subtracting the offset. shows a
picture of a modulator unit containing two modulator circuits.
The modulators are run at a 100 kHz sampling rate, in accordance with the
sampling rate of the system’s gradients. Signals with higher baseband frequency
can be generated by the system’s exciter, and these are fed through the
modulators, providing extra pulse form control in addition to the modulators.
## High-power amplifiers
The custom-built high-power amplifiers are designed to be used close to the
magnet in order to minimize cable losses. Each amplifier consists of three
amplifier stages with an overall gain of slightly more than 60 dB. The first
stage is a MGA-31189-BLKG (Broadcom Inc., San José, USA), the second stage is a
MRF6V2010NR1 (NXP Semiconductors N.V., Eindhoven, The Netherlands), and the
final amplification stage of each amplifier is a BLF188XR power LDMOS transistor
(Ampleon BV, Nijmegen, The Netherlands). With the final stage driven in AB mode,
the amplifier has a peak output of 1 kW. Without an unblank signal, the bias
voltages of the last two amplifier stages are set to 0 V, and the 50V power
supply for the final stage is cut off via a high power MOSFET. This ensures that
no power can be transmitted without an unblank signal present. Each amplifier
contains a 61 mF capacitor array and a high-power voltage regulator to supply
peak power during transmission. DC power supplies for the whole system can
continuously deliver 6 kW in total. In case of an emergency shutdown, the
capacitor arrays are discharged through load resistors inside the power
supplies.
shows a picture of an opened power amplifier. shows the position of the two
power amplifier racks at the back of the magnet; the second rack is just visible
at the right border.
To ensure stable behavior independent of reflected power, circulators
(JCC0296T0298N20-GIG, JQL Electronics Inc., Rolling Meadows, USA) are connected
to the outputs of the amplifiers. These circulators have an insertion loss of
0.15 dB. Since circulators are highly magnetic, the circulators are placed just
outside the magnet’s passive shielding, resulting in an extra cable length of 7
m in total (Aircell 7, SSB-Elektronik GmbH, Lippstadt, Germany), which is
shorter than the minimum cable length necessary to place the amplifiers outside
the magnet room at our site (14 m). The overall attenuation of these cables is
0.78 dB. The air to cool the amplifiers is taken from the room’s air-
conditioning input duct by two radial fans in series and delivered via a tubular
system over a distance of 15 m.
Due to the strong currents on the power supply cables of each amplifier rack,
the ground of the DC power supplies is only connected to a common ground
directly at the amplifier housing, ensuring that the current of up to 60 A on
the feed line is identical in both directions and resulting in no net force on
the cable as it crosses the magnetic field.
## Directional couplers
The custom-built directional couplers are fabricated in strip-line technology
with 1.5-mm-thick FR4 substrate material separating the strips from the top and
bottom copper layers. The distance between the conductors is 3.4 mm (center to
center) and the width of 0.94 mm was calculated numerically using CST Microwave
Studio (CST AG, Darmstadt, Germany). The coupled length is 69 mm, resulting in a
coupling of -29 dB. The output is further reduced by 20 dB attenuators for both
the forward and reflected output signals. To maximize directivity, the coupled
lines are terminated by a variable resistor and a trimmer capacitor, which were
fine-tuned after production. The insertion loss is 0.2 dB. The couplers are
placed in two boxes with 16 DiCos each on the tables next to the amplifier
racks.
## Power supervision
The power supervision units get the forward and reflected signals from the
directional couplers. A power splitter in the forward path feeds half of the
power directly to the receive chain while the other half as well as the
reflected signal are each fed into a logarithmic amplifier AD8307 (Analog
Devices, Norwood, MA, USA). With the use of the logarithmic amplifiers, the
forward and reflected signals are translated into voltage signals proportional
to the logarithm of the input power. Since the envelope of the RF power only
changes slowly, the result is a low-frequency signal that is multiplexed and
sent to two NI5751 digitizer cards (National Instruments, Austin, TX, USA) with
16 analog inputs each that are connected to the operating PC I/O system. Here,
the forward power in each channel is sampled every microsecond for amplitude
supervision. The software is implemented in a LabView environment (National
Instruments, Austin, TX, USA). If the maximum allowed input power is exceeded,
the supervision prevents the unblank signal from being conveyed to the
amplifiers.
## Receive chain
To enable automated switching between local and body coil reception as well as
to feed power supervision signals into the system’s receivers, several changes
were made to the receive chain. shows the original receive chain of the system
as provided by the vendor. It contains a second-stage receive amplifier as well
as a switch to allow input of power supervision signals from the vendor-provided
8-channel transmit system. shows the changes made to the receive chain. A
custom-built second-stage receive amplifier and power supply for the on-coil
pre-amps is used for reception with the body coil. The second-stage amplifier
uses an MGA-62563-BLKG (Broadcom Inc., San José, USA), has a gain of 22 dB and a
noise figure of 1 dB. The input 1 dB compression point is -3.5 dBm. The power
supply for the on-coil pre-amps and the second-stage receive amplifier can be
switched off during transmit mode. This helps to protect the pre-amplifiers and
reduces spurious signal in the receive path that might affect the power
supervision, reducing the demand on the off state-isolation of the following
switches. A custom-built 32-channel 2:1 switch (SRS) is used to select between
local and body coil reception and another switch (SSS) is used to integrate the
signals from the directional couplers for transmit supervision. These switches
are identical. Each channel contains three MASWSS0155 GaAs SPDT switches
(M/A-COM Technology Solutions Inc., Lowell, MA, USA) to achieve an off-state
isolation of better than -55 dB and an insertion loss of better than -0.6 dB.
## Logic controller
The logic controller receives the unblank and Tx/Rx signals from the original
system. Furthermore, it accepts signals from a digital I/O card from the
operating PC. These signals go through a logic circuit and are then distributed
to the power amplifiers (unblank signal), to the receive chain (Rx amplifiers
on/off, SRS and SSS switching states), and to the T/R switch controller
circuitry (transmit/receive state, tune/detune). All logic is implemented in
hardware using standard logic gate ICs to reduce the probability of failure as
compared to microcontrollers or embedded systems.
## Body coil
Since the 7T MR system is equipped with the AS095 gradient coil, there is a
space of 34 mm between the bore liner (615 mm diameter) and the gradient coil
(683 mm diameter). This space is used to accommodate the integrated body array.
The body array is mounted on a polycarbonate frame that consists of two halves
forming a cylinder with 22 outer faces. The length of the cylinder is 60 cm and
the inside of the polycarbonate frame rests directly on the bore liner. The
longitudinal bars stand off from the bore liner so that the coil elements have a
distance of about 8 mm from the bore liner.
The coil elements are micro strip-line elements with meanders, which were chosen
because of their good intrinsic decoupling when lying side by side even in low
load conditions. The elements have a length of 25 cm and are arranged in 3
rings: two outer rings of 10 elements and one inner ring of 12 elements. The
distance between the ground plane and the strips is 20 mm as was used in
previous studies. The ground plane completely encloses the coil. It is made of
0.127-mm-thick RO3010, both sides clad with copper. The copper of both sides is
slotted in the longitudinal direction in such a way that the copper strips of
the inner and outer layers overlap. In this way, RF currents can flow in the
circumferential direction, while eddy currents from the gradient fields are
effectively blocked. For mechanical stability the ground plane is glued with
epoxy to a 1-mm FR4 sheet on the inner surface.
While micro strip-line elements with meanders are well decoupled when lying next
to each other, additional decoupling is needed for diagonally adjacent elements
and elements placed head-on. shows an excerpt of the schematic of the body
array. Decoupling is done using 90° delay lines connected through a very low
impedance to ground. These lines provide another path for coupling that reduces
overall coupling due to a different phase compared to the parasitic coupling via
air and tissue. The two halves of the array are not electrically connected other
than the connection of the ground plane.
The elements are connected to custom-built, compact T/R switches with a design
similar to; the only difference to the published design is that all components
were placed on one side of a 81-mm by 42-mm PCB rather than using a folded
layout to achieve a flatter profile. An appropriate length of low-loss cable
(Aircell 5, SSB-Electronic GmbH, Lippstadt, Germany) transforms the impedance of
the pre-amp to ensure pre-amp decoupling. The T/R switches introduce an
attenuation of 1.1 dB. The Aircell 5 cables on the bore liner, connecting the
T/R switches with the coil elements and the directional couplers, have an
overall attenuation of 0.58 dB.
Detuning of the elements is performed with dedicated detuning boards containing
PIN diodes that produce an RF short circuit of the transmit cable to ground when
a forward current is applied. These detuning boards are placed in the transmit
chain between the power amplifiers and the T/R switches with an appropriate
length of cable to transform the short circuit to an impedance that detunes the
elements.
## EM simulations
A detailed simulation model of the prototype coil was constructed for validation
purposes and safety assessments. The model includes coil elements, the
supporting frame, and the MR environment, i.e. RF shield, cryostat, bore liner,
and patient table with a length of 200 cm to account for wave propagation.
Time-domain simulations were performed in CST Studio Suite 2016 (CST AG,
Darmstadt, Germany) using the finite integration technique (FIT). The simulation
domain was discretized with a hexahedral grid and 105 million mesh cells. A
finer resolution was used inside the coil volume to account for coil details and
a coarser mesh was used outside the coil volume. Tuning and decoupling of the
coil elements were performed in a circuit co-simulation. All cables between the
coil elements and the directional couplers were considered.
For coil validation, a homogeneous body-size phantom filled with tissue
simulating liquid (*ɛ*<sub>*r*</sub>*’* = 45, *σ* = 0.55 S m<sup>-1</sup>) was
simulated in the center of the coil. The phantom is a roughly oval-shaped
cylinder with a length of 50 cm, a width of 34 cm, and a height of 20.7 cm; the
filling volume is 32 liters. *B*<sub>*1*</sub><sup>*+*</sup> maps for excitation
of individual channels and CP+ and CP2+ mode were exported for measurement
validation. The simulated *B*<sub>*1*</sub><sup>*+*</sup> maps were compared to
maps acquired with the B1TIAMO technique.
For safety assessments, a male and female anatomical body model with a
resolution of 2x2x2 mm³ were placed in head-first supine position on the table
in the coil center. The normalized electric field distributions were exported
and 10g-averaged Q-matrices were calculated for every voxel in the body model.
Subsequently, the Q-matrices of both body models were compressed together using
the ‘virtual observation point’ (VOP) algorithm with a maximum overestimation of
10% to 420 VOPs. No differentiation was made between the extremities and the
rest of the body, so the head and trunk local SAR limit, which is more
restrictive, was used for the extremities as well.
## Quality assurance
A routinely applied quality assurance (QA) protocol was used to investigate the
influence of the add-on system on the performance of the original system. A
1-channel Tx / 32-channel Rx head coil (Nova Medical, Inc., Wilmington, USA) was
loaded with a tissue simulating phantom (*ɛ*<sub>*r*</sub>*’* = 55, *σ* = 0.6
S/m). The QA protocol includes tests to check for the performance of the local
RF coil as well as gradient and system stability. The proper function of the RF
head coil was verified by *B*<sub>*1*</sub>, SNR, and coupling measurements
(i.e. noise correlation). Unwanted noise sources and RF spikes were searched for
with noise measurements. System stability was measured with high-duty-cycle EPI
imaging (TR 1000 ms, TE 30 ms, echo spacing 0.54 ms, BW 2112 Hz/pixel, 64 phase-
encoding steps, 16 slices, FOV 220 mm, 3.4x3.4x2 mm<sup>3</sup>, 3x 250
measurements), and stability parameters such as SNR, Signal-to-Fluctuation-Noise
Ratio (SFNR), signal drift, fluctuation, and ghosting were calculated according
to the recommendations of the FBRIN consortium.
## Imaging experiments
In all imaging experiments the 32-channel body array was used for transmission
as well as reception.
For phantom measurements a torso phantom filled with tissue simulating
polyvinylpyrrolidon solution (*ɛ*<sub>*r*</sub>*’* = 45, *σ* = 0.55 S
m<sup>-1</sup>) was used that has the same dimensions as the one used in the
simulations.
For in-vivo measurements, informed consent was obtained from both participants
included in the study. All imaging experiments with the volunteers were
performed with an input power not exceeding the maximum allowed input power for
the worst-case shim set used within an experiment, with an extra safety margin
of a factor of 2. This was done to ensure that even if an erroneous trigger
signal would alter the shim setting, the SAR would remain within the allowed
limits.
*B*<sub>*1*</sub> mapping was performed using the B1TIAMO method. Absolute
*B*<sub>*1*</sub> maps were acquired in a central transversal slice in the torso
phantom and compared with simulation data. Furthermore, *B*<sub>*1*</sub>
mapping was performed in a human volunteer (1.86 m, 80 kg) in an axial slice
through the kidneys and a coronal slice through the center of the torso with the
CP+ and CP2+ modes. In this case we define the phase for each element via its
respective geometrical angle in the xy-plane.
Structural imaging was performed in a healthy volunteer of above-average size
(1.85 m, 95 kg). A 2D gradient echo sequence with 1.1-mm in-plane resolution and
5-mm slice thickness, a TR of 50 ms, and a TE of 6.1 ms was used to acquire
slices with a 50-cm field of view. Data sets from the head to the thighs were
acquired in three stations with approximately 10 cm overlap. TIAMO with the CP+
and CP2+ modes was used for transmit homogenization. For the slices in the
abdominal region, the volunteer was asked to place the arms above the head. Each
image was acquired during a breath hold. The vendor’s gradient distortion
correction was used.
To demonstrate the selective excitation capability of the system, four different
patterns were excited in the torso phantom: the letters “ELH”, the logo of the
Erwin L. Hahn Institute for MRI, a checker board, and the logo of the German
Cancer Research Center (“DKFZ”). The pulse length was 5.36 ms (536 samples) with
a variable density spiral for a 64 x 64 voxel target in 2D spatially selective
excitation, implemented in a 3D gradient echo sequence with 128 x 128 in-plane
voxels. The sampling rate for the transmit pulse was 100 kHz.
# Results
Implementation of the add-on system did not have detectable influence on the QA
parameters of the system apart from a 1.2 dB reduction in received signal due to
the switches integrated in the receive chain. Since these switches are behind
the second receive amplifier stage, no change in SNR was detectable. Neither an
immediate impact after integration nor a longer-term drift during the following
6 months was found; system users did not report any impairments in image quality
that could be correlated with the integration of the add-on system.
The amplifiers were successfully tested to play out 10 ms pulses at full peak
power. Duty cycle was limited to 10% or smaller to prevent overheating of the
amplifiers.
shows a comparison of the simulated and measured *B*<sub>*1*</sub><sup>*+*</sup>
maps in the torso phantom for the CP+ mode (A, C) and the first 16 individual
elements of the body array (B, D, E, F). The phase maps are normalized to the
phase of the first channel. The maps show good agreement qualitatively and
quantitatively in the absolute values (A, B, C, D) as well as in the depicted
phase (E, F). The single-channel maps show good agreement even in areas far away
from the transmitting elements. The normalized root-mean-square error for the
CP+ mode is 16.6%. Further values are shown in. With the shim shown in, a
*B*<sub>*1*</sub><sup>*+*</sup> amplitude of 13 μT can be reached in the center
of the phantom at full peak power. shows an axial cut through the body model at
the position of the highest SAR in the CP+ mode, with the SAR distribution as an
overlay. The highest SAR in the case of the CP+ mode occurs in the left arm. The
values are normalized to 1W total accepted power.
shows in-vivo flip angle distributions for the CP+ and CP2+ modes of the body
array. shows a coronal structural image including the position of the transverse
slice for orientation. show the CP+ mode in transverse and coronal orientation;
show the CP2+ mode in the respective orientations. The flip angle is fairly
constant in z-direction over a 50 cm field of view.
Structural body images are presented in. Due to the capability of both the
gradient and RF coil to cover a 50-cm field of view, the head, torso, and thighs
of a large volunteer could be imaged in three stations. The images show good
uniformity in contrast, and no complete signal dropouts are visible.
Both volunteers had undergone many 7T examinations in the past and reported that
the examination was very comfortable compared to examinations with local
transmit coils.
2D selective excitation examples are shown in. The images show the central slice
of the 3D data set for which the pulse calculation was performed. The patterns
show sharp edges and only low residual background excitation.
# Discussion
The results presented here show the feasibility of the presented system.
Implementation is straightforward since only very few interfaces are necessary.
Only one analog and three digital signals have to be taken from the original MRI
system to make the add-on fully functional. In comparison to the development of
complete consoles, which might pose more powerful tools, the big advantage is
the use of the standard vendor software that clinicians and researchers are
accustomed to. For simple RF shimming with the add-on system, the vendor-
provided sequences can remain completely unchanged. When selective excitation is
used, the sequences only need minimal reprogramming to allow loading of user-
provided pulses, greatly simplifying the workflow.
Originally the RF amplifiers were placed in the magnet room to reduce the cable
length to the coil for three reasons, namely to reduce the cable losses, to use
the impedance of the amplifiers for coil decoupling, and to enable a feedback
loop for coil decoupling. With the feedback loop included, the amplifiers had an
overall gain of about 90 dB. Together with the high in-room RF due to waves
traveling out of the bore, this led to unwanted feedback, so the loop had to be
disabled. Without the feedback loop, utilizing the amplifiers’ impedance for
decoupling led to non-linear behavior depending on the reflected power, which in
turn depends on the shim. Therefore, the circulators were included, also
increasing the cable length by 7 m. Putting the amplifiers in the same space as
the circulators was not possible because of space constraints and cooling
constraints.
The overall losses between the amplifiers and the inputs of the elements of the
body array amount to 2.79 dB. This includes 0.78 dB for the cables to and from
the circulators, 0.15 dB for the circulators, 0.20 dB for the directional
couplers, 0.58 dB for the cables on the bore liner (Aircell 5) and 1.1 dB for
the T/R switches. This leads to approximately 520 W per element at the coil.
An improved version of the system that is being installed in Heidelberg now has
the amplifiers outside the magnet room, with twice the amplifier peak power and
cables with much lower attenuation to account for the increased cable length,
leading to approximately 1.5 dB attenuation. This will result in roughly 900 W
peak power per coil element. Moving the amplifiers outside the magnet room also
facilitates cooling, since active cooling with fans directly at the amplifiers
is possible and the duty cycle can therefore be increased.
The low transmit efficiency of integrated body arrays is partly offset by the 32
kW total peak power provided by the amplifiers. With the shim shown in, a
*B*<sub>*1*</sub><sup>*+*</sup> of 13 μT was achieved in the center of the
phantom. A dedicated shim for the center of the phantom using a local 8-channel
transmit array comprised of the same transmit elements and 8 kW total peak RF
power resulted in a maximum achievable *B*<sub>*1*</sub><sup>*+*</sup> of 12.4
μT, but covering a smaller field of view in z-direction. A mean peak
*B*<sub>*1*</sub><sup>*+*</sup> of 20 μT was reported in the prostate over 6
healthy volunteers for a close-fitting 16-channel loop-dipole array as well as
13.2 μT for a 10-element array of fractionated dipoles. Higher amplifier power
for the integrated body array presented here would clearly be beneficial to
improve transmission. Furthermore, a more efficient design for the body array
could lead to further improvements, since the transmit elements in this project
were not chosen for their efficiency, but for their good decoupling; dipoles
have been shown to be more efficient in integrated designs. Using a more
efficient coil design as proposed in and increasing the amplifier peak output
power to 2 kW per channel could increase the peak B1+ amplitude to 30 μT in the
center of the body. Possible further improvements in addition to switching to
dipoles as transmit elements include using T/R switches with reduced attenuation
(or even excluding them and only using dedicated receive coils) and using cables
with reduced attenuation on the bore liner.
Although the hardware of the system is conceived to apply low-power samples of
the forward waveforms to the receive chain during transmit, this capability was
not used in the current experiments. Instead, more conservative SAR limits based
on amplitude-only supervision were utilized. Future work will target real-time
monitoring of both the amplitude and phase of the forward signal on all 32
channels, which represents a significant computational challenge but would
further relax SAR restrictions.
Integrating a body array into the MRI system leaves more space in the bore,
allowing for larger patients to be imaged, potentially with slim receive-only
arrays, while local transmit arrays tend to take up quite some space in the
bore. The large field of view allows imaging of large areas, which enables
whole-body imaging with only a few stations. Among other applications, this new
coil configuration might be beneficial for contrast agent free time-of-flight
imaging of the lower extremities where local coils are known to limit the
patient diameter. First structural images show promising homogeneity and image
contrast. In this study, the arms were placed outside the field of view because
their high signal caused image artifacts due to their proximity to the coil
elements. Since reception in this work was done with the body coil, a large SNR
boost can be expected from local receive-only coils, also reducing the signal
intensity differences between arms and torso. Ultimately, this would allow a
workflow at UHF that mimics common practice at lower field strengths like 1.5
and 3 T.
Even though the DACs used in the presented systems are comparably slow and need
up to 3 μs until the output is settled, the selective excitation examples show
that 100 kHz sampling is possible. Imprecisions of the IQ modulators were
successfully countered by using the system’s exciter in combination with the IQ
modulators to achieve a coupled modulation strategy. Together with the large
field of view, the capability for 32-channel selective excitation could
potentially be used for reduced field of view imaging of, for example, large
parts of the spinal column at UHF, which has already been shown for 16 transmit
channels.
Since the add-on system does not affect the use of the system in standard
configuration, all imaging protocols relying on the single-channel transmit
system and the affiliated local Tx/Rx coils can still be used without any
change, leading to an overall very versatile system. Future work will have to
look more deeply into the capabilities, advantages and drawbacks of a 32-channel
transmit system with an integrated body coil compared to standard systems with
local transmit coils.
# Conclusion
We present a 32-channel transmit system add-on that is capable of 100 kHz
sampling. Structural imaging with an integrated 32-channel body array is
possible and selective excitation is shown. The system does not affect the
operation of the original MRI system in single-channel mode. The presented add-
on system provides a tool for large field of view imaging and a potentially more
clinic-like workflow.
The research leading to these results has received funding from the European
Research Council under the European Union’s Seventh Framework Programme
(FP/2007-2013) / ERC Grant Agreement n. 291903 MRexcite.
The authors thank Kevin Kolpatzeck, Dominik Beyer, Tristan Mathiebe, Jonathan
Weine, and Sarah Handtke for help with hardware assembly.
[^1]: The authors have declared that no competing interests exist. |
# Introduction
Inflammatory bowel diseases (IBD), which include ulcerative colitis (UC), are
characterised by uncontrolled intestinal inflammation. Colorectal cancer (CRC)
is a life-threatening complication of UC, and UC patients are estimated to be at
a 2.4 fold increased risk of CRC relative to the general population. The
risk of CRC in UC patients increases considerably with duration and extent of
active inflammation, diagnosis at a younger age, concomitant primary sclerosing
cholangitis (PSC) and family history of CRC. Relative to sporadic CRC, UC
associated CRC (UC-CRC) is linked to an earlier age of onset and higher rates of
mortality; in one study, 59% of UC-CRC patients had died at 5-years of follow-
up.
Given the increased risk of UC-CRC, it is necessary to screen UC patients for
the development of precancerous lesions (dysplasia), with periodic
chromoendoscopies starting 8–10 years after the first appearance of colitis-
associated symptoms. During surveillance procedures, targeted biopsies are
collected from suspect lesions and assessed by pathologists. The presence of
dysplasia is then used to guide and inform the subsequent clinical management of
the patient: for low-grade dysplasia (LGD) on flat mucosa it is advisable to
reduce the interval between surveillance colonoscopies, in the case of high-
grade dysplasia (HGD) colectomy is indicated.
UC-associated post-inflammatory polyps (UC-IPs) (pseudopolyps) are raised areas
of inflamed mucosa or granulation tissue seen after intestinal mucosal recovery
from inflammatory damage. A history of UC-IP may be an indication of the
severity of prior inflammation. Current endoscopic surveillance programmes do
not consider UC-IPs *per se* important clinically and their recording may well
be incomplete. Yet, a correlation between the presence of colonic UC-IPs and
increased risk of CRC has been proposed.\[–\] However, it remains to be
determined whether the link between UC-IPs and CRC is indirect or whether UC-IPs
show any features or changes in gene expression that might be associated with an
increased risk of malignancy.
The molecular mechanisms that lead to cancer from LGD in UC are poorly
understood, and rates of UC-CRC occurrence in patients with diagnosed LGD vary
widely between studies 0–54%;\[–\] current estimates are that 19% of patients
with a diagnosis of LGD develop HGD or cancer. This variation in outcomes
probably reflects the fact that although LGD is treated as a single entity it
encompasses a range of underlying molecular changes which cannot be
distinguished macroscopically or histologically. Hence there is a significant
clinical need for molecular biomarkers to identify those dysplastic lesions that
are at high risk of neoplastic progression and malignancy.
MicroRNAs (miRNAs) are small non-coding RNAs that inhibit protein translation by
binding to complementary sequences in the 3’UTR of target mRNAs. Each miRNA is
predicted to bind to multiple potential mRNAs targets and thereby alter
signalling pathways within cells. MiRNAs have important roles in development and
disease and have attracted particular attention for their role in cancer and
IBD, and for their potential use as biomarkers and therapeutic targets. A
role for miRNAs in sporadic CRC has been established. However, far less is
known about their function in UC-CRC and dysplasia with only two studies
reported assessing patients with both UC and CD. The techniques used and
the clinical characteristics of the specimens analysed also differ between
studies. Moreover, because of the low frequency of IBD-dysplasia the cohorts
analysed are relatively small and the result requires independent
validation.
In order to address the role of miRNAs in UC-dysplasia, we have determined miRNA
expression in a cohort of UC patients with and without dysplasia, using samples
collected from IBD surveillance centres across the UK. We have also compared
miRNAs identified in our study of UC with findings in the published literature
on IBD-dysplasia. In addition, we provide the first comprehensive analysis of
miRNAs in UC-IPs and compare changes in miRNA profiles between these polyps and
UC controls. As miRNAs are also exported in the wider circulation, the potential
to utilise miRNAs as non-invasive biomarkers of UC-Dysplasia (UC-DYS) was also
explored in a proof-of-principle study.
# Materials and methods
## Ethics statement and patient selection
Ethics were obtained to recruit patients from multiple hospitals in UK (East
London and The City Research Ethics Committee 1, REC 09/H0703/116; UKCRN ID
8099). Inclusion criteria in the study included a diagnosis of UC and enrolment
in a surveillance programme. Patients gave written informed consent on forms
approved by the Ethics committee and were recruited between 2010 and 2014.
The clinical and endoscopic disease activity was recorded for each patient,
using the Mayo score. UC phenotype was classified according to Montreal
criteria. Further clinical characteristics were also recorded, including
age, ethnicity, disease duration, smoking habit, medications, adherence to
therapy, family history of colon cancer, concomitant diseases and previous
history of dysplasia.
## Collection and selection of samples for microRNA analysis
Multiple biopsies were collected from 5 sites throughout the bowel during
routine endoscopy for histopathology and miRNA analysis. Biopsies for miRNA
analysis were stored in RNAlater at -80°. The local pathologist assessed
biopsies taken for histopathology. The resulting pathology reports were used to
classify the biopsies taken from the same site and stored in RNAlater into those
with and without dysplasia, and with and without evidence of histological
inflammation, (neutrophil infiltration of the epithelium). A peripheral blood
sample was also collected and centrifuged to obtain serum, and stored at -80°C
until further experiments.
To reduce possible confounding factors, samples from patients with coexisting
primary sclerosing cholangitis, previous history of dysplasia or CRC and colitis
limited to the left-sided colon or to the rectum were excluded from this study.
Those samples used here were from patients with pancolitis (E3 Montreal).
## Tissue miRNA arrays
Biopsies stored in RNA later were homogenised in 700μL Qiazol, and RNA
extractions were performed using the miRNAesy Kit (Qiagen, UK) as per the
manufacturer’s instructions. The resultant RNA was sent to Exiqon (Denmark) for
miRNA profiling using the miRCURY LNA™ microRNA Array (7th Gen). The quantified
signals were background corrected and normalized using the quantile
normalization method. The normalised data was then log2 transformed prior to
statistical analysis.
## Serum miRNA arrays
Serum samples were sent to Exiqon (Denmark) and RNA extracted and analysed by
qPCR using miRCURY LNATM Universal RT microRNA PCR Human panel I+II. Samples
were run in two batches, and each batch contained UC controls and UC-dysplasia
samples (batch one: n = 5 and n = 3, respectively; batch two: n = 5 and n = 5,
respectively). The data normalisation was performed based on the average of the
assays detected in all samples in each batch respectively, and the normalised
data log2 transformed. The data for each batch was then median-centred and
combined prior to statistical analysis to account for ‘batch’ effects. Finally,
miRNAs were filtered to include only miRNAs that were detected above the
background threshold in all samples tested. The aim was to ensure that only
robustly expressed miRNAs were considered.
## MiRNA qPCR validation
A select number of miRNAs were evaluated by qPCR using the miScript primer
assays. For qPCR analysis 1000 ng of RNA was reversed transcribed using the
miScript II RT Kit (Qiagen) and the miScript HighSpec buffer. The resultant cDNA
was then diluted 10-fold prior to qPCR analysis using the miScript miRNA
quantification system (Qiagen). Each reaction was performed in duplicate and
melt curves performed to ensure a single PCR product was synthesised. The
resultant cycle threshold values (Ct’s) were exported into Excel, duplicates
were averaged, and data normalised to the geometric mean of RNU6 and SNORD42b,
two stably expressed small non-coding RNAs. The normalised expression values
were then log2 transformed to equalise variance prior to statistical analysis.
## microRNA in situ hybridization
The levels of miR-200b-3p and miR-21-5p were investigated using miRNA *in situ*
hybridization (ISH) technology (Exiqon, Denmark) in formalin fixed paraffin
embedded (FFPE) tissue taken from the colon of UC patients. All blocks were
obtained from the Royal London Pathology archives, and 4 μm sections were cut
under RNAse-free conditions. Prior to staining sections were baked at 65°C for
10 mins before deparaffinising in xylene for 10 mins and hydrating through
graded alcohols to RNAse-free H<sub>2</sub>O. Sections were digested with
10ug/ml proteinase K in PBS at 37°C for 30 minutes, washed in RNAse-free
H<sub>2</sub>O and dehydrated. Sections were then incubated with Double-DIG
(digoxigenin) labelled Locked Nucleic Acid (LNA) ISH probes (60 nM, Exiqon) for
miR-200b-3p (619853–360), miR-21-5p (619870–360) or a scrambled negative control
(699004–360) at 56°C for 2 hours before a series of stringency washes and
staining performed as given previously. A clinical pathologist \[RF\] at the
Royal London Hospital confirmed positive staining and reviewed all slides.
## Statistical analyses
Unsupervised principal component analysis (PCA) of miRNA levels in UC controls,
UC-DYS and UC-IPs was performed to explore natural grouping in the data using
Solo software (Eigenvector, USA). The software’s proprietary algorithms were
used to exclude miRNAs with excessive missing values and replace any further
missing values in the remaining data. The PCA was carried out on the covariance
matrix using the log2 transformed array data. Three components were fitted to
the model and these were sufficient to separate UC controls from UC dysplasia.
To identify individual differentially expressed miRNAs, the array data (both
tissue and serum) were analysed using the ‘Significance Analysis of Microarrays’
package in R. Missing values were compensated for using the K-Nearest Neighbours
Imputer algorithm, and the number of neighbours set to 6. The log2 transformed
data were median-centred, and a two-class unpaired analysis was performed with
1000 permutations. A significance cut-off was set using a delta value of 0.5 and
a 2-fold minimum change. Under these stringent cut-offs the estimated false
discovery rate was 0.00%.
For the qPCR validation analysis a two-way analysis of variance (ANOVA) test was
used to explore the relationship between miRNAs in dysplasia and inflammation,
each modelled as independent discrete factors. *Post-hoc* analyses were then
performed to identify difference between groups with alpha adjusted for multiple
testing using the Bonferroni criterion.
## In silico identification of microRNA targets
Experimentally validated target genes for candidate miRNAs were obtained using
miRTarBase, which contains targets that have been validated by reporter
assays, western blot, qPCR, microarray and next-generation sequencing
experiments. miRNA-target interactions were also investigated based on the
collective information of functional studies of miRNAs in the database. To
bioinformatically predict target genes based on the miRNA seed sequence, two
predictive algorithms were used, TargetScan as well as miRDB.
To investigate the influence of candidate miRNA transcript levels in UC
dysplasia, we interrogated the gene signatures obtained by cDNA microarrays in 4
UC patients without dysplasia and 11 UC patients harbouring remote
neoplasia. To obtain a list of all differentially expressed genes across the
two groups, the microarray datasets were obtained from the GEO repository
(accession number GSE37283) and analysed with GEO2R, a web-based application
that performs comparisons between groups on the original submitter-supplied
microarray datasets. GEO2R uses the GEOquery and limma R packages from the
Bioconductor project. For consistency, to identify differentially expressed
genes, we applied the following thresholds: adjusted p-value \< 0.05 (based on
the Benjamini & Hochberg false discovery rate method) and \|log2(fold-change)\|
\> 1.
# Results
## Study design and biopsy characteristic
To interrogate miRNA profiles in UC-DYS lesions, RNA extracted from biopsies
collected during routine surveillance endoscopies with and without evidence of
dysplasia was analysed (n = 10 and n = 7, UC controls and UC-DYS, respectively);
the clinical characteristics of patients from whom a UC dysplasia sample was
taken is given in. For the array, the cohort was predominantly White (6/7), the
median age was 73 years, and all patients had a history of pancolitis (E3
disease). Biopsies were taken from sites of dysplasia throughout the colon: 2
from the ascending colon, 2 from the transverse colon, 2 from sigmoid colon and
1 from the rectum. Of these, 6/7 (86%) was classified as LGD and 1/7 had HGD;
3/7 biopsies also had evidence of histological inflammation. RNA isolated from
UC-IPs (n = 7) was also analysed.
To identify miRNAs associated with UC-DYS that were independent of intestinal
location and disease activity, control biopsies were pooled from a number of UC
patients with respect to presence or absence of histological inflammation
(active and inactive) and the five-biopsy sites (ascending colon, transverse
colon, descending colon, caecum, or rectum). This gave a final 10 control
samples. RNA isolated from UC-IPs (n = 7) was also analysed to determine whether
these possessed similar changes in gene expression to dysplasia specimens.
## Identification of miRNAs associated with dysplasia in ulcerative colitis
The expression levels of 1899 miRNAs were assessed by array, of which 1240 (65%)
were not detected above the background threshold levels in any of the 24 samples
analysed and therefore excluded. One control sample (an active biopsy from the
caecum) was further excluded on the basis of a lower call rate than the
remaining samples. To explore natural grouping within the data unsupervised
principal component analysis (PCA) was performed. For PCA, miRNAs with excessive
missing values were excluded (a total of 258), and only the remaining 401 miRNAs
were considered. The resultant PCA model demonstrated some separation of UC-DYS
from control samples, suggesting that UC-DYS is associated with a shift in miRNA
expression profiles. Control samples were separated completely from dysplasia
samples within the first three principal components with the sole exception of
one sample, an active rectal biopsy.
To identify differentially expressed individual miRNAs the array data was
analysed using ‘Significance Analysis of Microarrays’ software with an estimated
false discovery rate of 0% and a fold change \>2 set as stringent cut-off values
for significance. This analysis identified 22 miRNAs that were upregulated in
dysplasia and 8 miRNAs that were present at lower levels in UC-DYS lesions. A
number of the miRNAs identified had been previously linked to IBD-DYS, including
miR-200b-3p, miR-200a-3p, miR-192-5p, and miR-155-5p13. Previously reported
associations between miR-21-5p12 were also confirmed; reported difference in
miR-31 in IBD-dysplasia could not be fully assessed here as miR-31-5p
expression was only detected in 2 of the 24 samples analysed by array. However,
it is noteworthy that both biopsies where miR-31-5p levels were detected above
the background threshold were dysplastic.
## Changes in miRNA profiles in post inflammatory polyps
To determine whether the miRNA profiles of UC-IPs resembled UC-DYS specimens or
not, the UC-IP miRNA expression changes were fitted to the PCA model.
Interestingly, UC-IPs appeared to form an intermediate group, with some samples
clustering with controls and other samples with the UC-DYS specimens. However,
relative to controls, in terms of absolute differences in expression, there were
only two miRNAs with a fold change\>2 at a FDR of 0.00% in UC-IPs using the SAM
software: miR-144-3p was increased in UC-IPs (2.017 fold, q\<0.001) and
miR-4795-3p was reduced (0.456 fold, q\<0.001). There was no overlap between
miRNAs that marked UC-IPs and those that were associated with UC-DYS.
## qPCR validation of miRNA changes in dysplasia
To confirm the findings of the array, qPCR validation was performed for 7 miRNAs
upregulated and 1 miRNAs down-regulated in UC-DYS in the array. The validation
cohort was expanded in number, but included samples previously analysed by array
and therefore does not represent a wholly independent cohort (UC-DYS n = 10 and
UC controls n = 16); where multiple biopsies were taken from individual control
UC patients they were averaged with respect to disease activity to avoid pseudo-
replication of the data. There were no significant differences between UC-DYS
and UC controls in our initial analysis (t-test).
One potential confounding factor in this analysis is a change in the proportions
of UC-DYS samples with active disease analysed in the validation cohort compared
to the array cohort: 3/4 in the array cohort vs. 7/11 in the validation cohort.
Hence a two-way ANOVA was performed to investigate the relationship between
dysplasia, disease activity and miRNAs levels, followed by *post-hoc*
comparisons of groups. This demonstrated that for 3 of the miRNAs tested
(miR-200b-3p, miR-451a, and miR-27b), there was a significant interaction term
for UC-DYS and disease activity. A subsequent *post-hoc* test demonstrated that
miR-200b-3p was significantly up-regulated (fold change \>2) in UC-DYS specimens
in the absence of histological inflammation relative to inactive controls in
line with the results of the array (p = 0.033). However, as disease activity
tended to supress miR-200b-3p levels in the UC-DYS group (p = 0.010),
dysplasia–associated differences in expression were not seen on a background of
histological inflammation. Conversely, and in contrast to the array, a
significant reduction in miR-27b was observed in UC-DYS relative to controls on
an inactive background (p = 0.016); in the UC-DYS group histological
inflammation was associated with higher miR-27b levels (p = 0.015). No
significant differences in miR-451a were identified in the *post-hoc*
comparisons of the groups once alpha was adjusted for multiple testing. In
contrast to previous studies miR-21-5p showed a stronger association with
disease activity than with dysplasia.
## miRNA in situ hybridization
Associations between miR-200b-3p and UC-DYS detected in the array and validation
cohort were further investigated using miRNA ISH techniques on archival FFPE
blocks. To determine whether miR-200b-3p was linked to CRC progression in UC
patients *in situ* analysis was performed on a series of FFPE specimens from UC
controls and from UC patients with confirmed UC-DYS and UC-adenocarcinomas;
miR-21 levels were analysed in serial sections of a subset of blocks as a
positive control for staining. Representative images are shown in. Discrete
miR-200b-3p staining was identified in the epithelial cells of controls, UC-DYS
and UC-adenocarcinomas. In general, the strength of staining increased from UC
controls through to UC-adenocarcinomas. By comparison, miR-21 was more robustly
expressed and localised to the stroma of UC adenocarcinomas. No positive
staining was observed in sections incubated with a scrambled negative control
probe.
## Identification of miR-200b-3p targets in UC-dysplasia
By interrogating the gene signatures from cDNA microarray analysis of UC
patients with neoplasia (UCN) compared to UC patients, we found 129 mRNAs
that were downregulated, corresponding to 76 genes. Since an increase in miRNA
levels would result in mRNA downregulation, we investigated whether the reduced
expression of these genes is a result of the miRNA action. Among the 76 down-
regulated genes, one gene, *INO80D*, was among the targets of miR-200b-3p
obtained by miRTarBase , a database of experimentally confirmed miRNA targets.
*INO80D* is a regulatory component of the chromatin remodelling INO80 complex,
which is involved in transcriptional regulation, DNA replication and probably
DNA repair. We expanded our search to target genes that are predicted by
bioinformatics algorithms based on the miRNA seed sequence. Five downregulated
genes (*INO80D*, *SHROOM1*, *HMBOX1*, *SLC4A4* and *PLEKHA6*) were among the
targets predicted bioinformatically by TargetScan, while six genes (*PPARA*,
*SHROOM1*, *SMIM5*, *HMBOX1*, *ANK3*, *KMT2C*) were predicted by miRDB. Among
these genes, *KMT2C* is a lysine methyltransferase found to be associated with
colorectal cancer as well as acute myeloid leukemia. KMT2C has also been
implicated in gastric cancer and could be useful as a marker of prognosis.
summarises these results.
## Analysis of serum miRNA profiles in dysplasia and controls
MiRNAs are also present in the serum, where there is the potential to exploit
them as non-invasive biomarkers of dysplasia. Hence, we sought to determine
serum miRNA levels in UC-DYS patients (n = 8) and UC controls (n = 10). Levels
of miRNA in the serum were assessed by qPCR array (Exiqon) and the normalised
data were analysed using SAM software as per the tissue arrays (with a cut-off
of a fold change\>2 set). However, no significantly altered miRNAs in the serum
of UC-DYS patients relative to UC controls were identified; with an adjusted
cut-off of a fold change\>1.5 there were 4 miRNAs were identified: miR-423-3p
and miR-28-5p were present at higher levels, whereas serum levels of miR-150-5p,
miR-32-5p were reduced in UC-DYS. However, with a false discovery rate of 1.12 a
cut-off of a fold change\> 1.5 is not stringent enough, and these miRNAs are
highly likely to represent false positives.
# Discussion
In this study miRNA profiles associated with dysplastic lesions in UC patients
were investigated. Multivariate analysis demonstrated good separation of UC-DYS
from UC controls, indicating a shift in miRNA expression profiles in tissue from
UC-DYS. However, there was a larger degree of variance between the UC-DYS
samples with respect to their miRNA profiles, highlighting the inherent
heterogeneity of this group. For future studies it will be important to
understand the significance of the profile heterogeneity for UC-DYS progression,
as not all LGD will progress to HGD and/or CRC. Whether this heterogeneity in
miRNA profiles could be exploited to identify the dysplastic lesion most likely
to progress remains to be seen. Encouragingly, from a molecular biomarker
perspective, the observed changes in miRNA expression in this study appeared
specific to UC-DYS because similar changes were not observed in UC-IPs. In fact,
the miRNA profiles of UC-IPs were largely indistinguishable from UC controls,
suggesting that these lesions might not have pre-malignant potential that could
be linked to miRNAs expression changes.
A number of differentially expressed miRNAs in UC-DYS that were identified in
this study have been previously linked to IBD-dysplasia, providing
independent validation of our results. For example, Olaru and colleagues
demonstrated broad increases in the miR-200 family in IBD-dysplasia (both UC and
CD patients) relative to chronically inflamed controls, a finding that is in
keeping with the up-regulation of miR-200b-3p reported here. While both studies
show that increased miR-200b-3p expression is a feature of UC-DYS, it is
important to note that in our study, miR-200b-3p was only upregulated in
dysplastic lesions of patients in clinical and histological remission, i.e. no
inflammatory infiltrates. Inflammation supressed miR-200b-3p and so in patients
with ‘active’ disease there was no difference in miR-200b-3p in our validation
cohort. Surveillance colonoscopies are already indicated for patients in
clinical remission, our data would now indicate that the presence of
inflammatory infiltrates in the mucosa should also be considered when assessing
the utility as a biomarker. The observation that miR-200b-3p is supressed in
patients with evidence of active disease is consistent with reports that show a
reduction in miR-200b in inflamed UC colon and reports that the reduced
expression of miR-200b in IBD patients is linked to epithelial mesenchymal
transition (EMT). The effect of inflammation may be to down-regulate miR-200b-3p
in other cell types within the mucosa.
As this study focuses predominantly on the analysis of low-grade UC-DYS, it
seems likely that the observed up-regulation of the miR-200 family is a very
early event in the UC-CRC development pathway. The analysis of miR-200b-3p using
miRNA-ISH techniques further supports a functional role for this miRNA in
development of UC-CRC as miR-200-3p is localised to intestinal epithelial cells
in UC patients consistent with its known role in determining epithelial cell
fate. Notably, the strength of miR-200b-3p staining increased at each stage
of malignancy, UC-DYS and UC-CRC. In line with the findings of this study,
increased expression of miR-200b-3p in colon cancer cells *in vitro* has
previously been reported, and high expression of miR-200b-3p observed in
sporadic colorectal tumours compared to adjacent mucosa.
Previous studies have identified altered mRNA signatures associated with
dysplasia in UC that were progressively altered during neoplastic
progression. In this study we have utilised this available data set and
interrogated it for down-regulated genes with a miR-200b-3p binding site. This
list included genes associated with chromatin remodelling and DNA transcription
(e.g. INO80D and KMT2C). The functional significance of the suppression of these
genes on the progression of UC-DYS is currently unknown. Little is known about
the role of the INO80D subunit. However, loss of function mutations in KMT2C, a
histone lysine methyltransferase, have been identified in a number of
cancers, and colorectal cancer cell lines, pointing towards a tumour
suppressor role. This is supported by the fact that KMT2C knockout mouse model
that developed ureter epithelial tumors. Correlating miRNA and mRNA
expression data sets can give useful insights into the potential action of a
miRNA *in vivo* and provide direction for future experimentation that can
confirm whether, for example, the suppression of 200b-3p targets in UC-DYS is a
direct consequence of the up-regulation of miR-200b-3p.
While there was good concordance between our study and the previous studies for
IBD-dysplasia and miR-200b-3p, we were unable to fully validate the increase
in miR-21-5p in UC-DYS that was observed in the array and has previously been
reported by Ludwig and colleges; instead in the validation analysis this
miRNA was much more strongly associated with active inflammation as opposed to
the presence of dysplasia, which is known to be an independent risk factor for
the development of UC DYS and UC CRC. The link between high miR-21-5p may
therefore be highly dependent on the patients inflammatory status. Nevertheless,
high levels of miR-21 were observed using miRNA ISH techniques in UC-CRC; in
these tumours miR-21 was localised to the stroma and not epithelial cells, a
finding which contrasts with Ludwig et al, but accords with Hedbäck et
al. The conflicting results between studies for miR-21 localisation is
likely a results of using different *in situ* hybridization probes, this will
need to resolved before a definitive answer on miR-21 localisation can be given.
Unfortunately, reported differences in miR-31 in IBD-dysplasia, could not be
fully interrogated because of its low expression levels in the samples analysed
in this study. However, the only two samples from the array where miR-31 was
detected above the background threshold levels were dysplastic, which is
consistent with the earlier reported increase in dysplasia in the literature.
Our data suggest that up-regulation of miR-31 is not sufficiently consistent
across UC-DYS specimens for this miRNA to be a good biomarker candidate of LGD,
as the absence of its expression could lead to false negative diagnoses of
dysplasia.
Attempts to identify a non-invasive serum biomarker of UC-DYS were also made,
but no significant differences in serum miRNA expression were observed. This was
a novel analysis, and the lack of statistical differences likely reflects the
fact that any serum dysplasia signal will be very weak given only a small
proportion of cells in the gut are affected, and the serum miRNA pool is derived
from all cells in the body. Also, the current study may have been underpowered
to detect small differences in miRNA levels in the serum of UC-DYS patients.
However, it should be acknowledged that the rarity of the UC-DYS makes large
population biomarker studies extremely challenging. Nevertheless, we believe
that the data from this study can be used to inform future studies.
There are some limitations to our study. The estimated prevalence of LGD in UC
is 15 cases per 1000 patients per year in the UK. Therefore significant numbers
of patients need to be screened to identify dysplasia specimens for analysis. To
address this we recruited from ten NHS trusts with major inflammatory bowel
disease surveillance centres across the UK. Despite this, a clear limitation of
this study is the sample size, which reflects the low frequency of UC-DYS, and
the difficulty in obtaining sufficient specimens from well-phenotyped patients.
Independent validation is therefore an important step that requires adequate
numbers. It is therefore encouraging that several miRNAs identified here had
been previously linked to IBD-dysplasia, e.g. miR-200b-3p. For future
studies a standardized miRNA analysis pipeline should be implemented to
facilitate direct cross-comparisons between studies, enabling pooling of data
and large population studies, as we have recommended in a recent review;
each of the studies performed to date have used different miRNA profiling
platforms and methods. The data from this study also highlight the need to
control for heterogeneity within the UC-DYS group, as potential confounding
variables include the size of the field of dysplasia, the extent of inflammation
within a lesion, length of disease duration and patient’s medications. Given the
relatively advanced age of the UC-dysplasia cohort (median 73-years), there is
an increased risk of sporadic CRC and neoplasias, which may not be directly
attributed to the underlying presence of UC. These confounders are inherently
difficult to control for in small populations, although this is essential if
more accurate biomarkers of progression are to be developed, as dysplasia is not
a single homogenous group.
In summary, CRC is a life-threatening complication of UC, and surveillance
programmes use regular colonoscopy to identify early pre-cancerous lesions
(dysplasia) and guide patient care. However, we know very little about the
molecular changes that occur in these dysplastic lesions and rates of CRC
occurrence in patients with confirmed LGD vary widely. A better understanding of
the underlying mechanisms of dysplasia may allow for more accurate
identification of the lesions that harbour the highest risk of malignancy. In
this study we have shown that dysplasia is associated with alterations in the
miRNA expression profiles and have demonstrated localisation of this miRNA to
epithelial cells in dysplastic lesions and in UC associated CRCs. This study
highlights increased expression of miR-200b-3p as a critical component of UC-DYS
and a relevant target for therapy.
# Supporting information
We are indebted to the patients who participated in this study and thank the
gastroenterologists (Ailsa Hart, Donagh O’Riordan, Konrad Koss, Matthew Brookes,
Stephen Lewis, Tariq Iqbal,) and research nurses and practitioners (Carol Gray,
Fiona Stead, Helen Fairlamb, J Rankin, Janet Morse, Judy Sercombe, Keely Lane,
Marie Green, Natalie Wheatley, Nicola Lunt, Sue Inniss, Zacharias Tsiamoulos)
who contributed samples. Jake Bundy (Imperial College London) and Robert Lowe
(Queen Mary’s University London) are thanked for discussion of statistical
analysis.
ANOVA analysis of variance
CRC colorectal rectal cancer
CD Crohn’s disease
FDR false discovery rate
FFPE formalin fixed paraffin embedded
HGD high grade dysplasia
IBD inflammatory bowel diseases
UC-IP post inflammatory polyps
ISH *in situ* hybridization
LNA locked nucleic acid
LGD low grade dysplasia
miRNAs microRNAs
PCA principal component analysis
UC Ulcerative colitis
UC-CRC UC associated CRC
UC-DYS UC-Dysplasia
[^1]: The authors have declared that no completing interests exist.
[^2]: **Conceptualization:** AS JOL. **Data curation:** AL CL. **Formal
analysis:** AL EG AS. **Funding acquisition:** AS JOL NJ. **Investigation:**
AL CF TK KS RJ RF AA NJ. **Methodology:** AL EG JOL AS. **Project
administration:** AL CL AS JOL. **Supervision:** AS AA JOL. **Validation:**
AL CF AS. **Visualization:** AL EG RF. **Writing – original draft:** AL AS.
**Writing – review & editing:** AL CF TK CL KS RJ RF EG AA NJ JOL AS. |
# Introduction
Chikungunya virus (CHIKV) is an enveloped positive-strand RNA virus belonging to
genus *Alphavirus* of the family *Togaviridae*. It is an epidemic viral disease
responsible for significant global public health problem mainly in Asian and
African continents. Efforts are underway in developing prevention strategy.
Recently a VLP based chikungunya vaccine was developed in USA that was found to
be protective in primates. Chikungunya infection is generally characterized by
fever and joint pains with additional symptoms including chills, vomiting,
nausea, headache and rashes.
Historical data report the first detection of chikungunya in 1952 in the Makonde
Plateau in Africa – where the virus is known to be maintained in the sylvatic
cycle of wild primates and mosquitoes such as *Aedes taylori*. Later in 1958 it
was detected in urban Asia such as Thailand mainly transmitted by *Aedes
aegypti*. In India, where both *Aedes aegypti* and *Aedes albopictus* are known
to exist and are widely prevalent during the post monsoon season, CHIKV was
first detected in 1963 in West Bengal. It was followed by several epidemics in
Chennai, Pondicherry, Vellore, Visakhapatnam, Rajmundry, Kakinada, Nagpur and
Barsi between 1964 and 1973. Recently CHIKV resurfaced in India affecting
several South Indian states. The outbreak started in 2005 from the coastal
regions of Andhra Pradesh and Karnataka. With more than 1.3 million people
estimated to be affected CHIKV prevailed across 150 districts of 8 states in
India. Despite the number estimated, the actual disease burden was thought to be
much higher due to potential underestimation from lack of accurate reporting.
CHIKV being an RNA virus is susceptible to high mutation rates which may help
the virus to evade the immune response and thus adapt efficiently. However,
phylogenetic analysis of E1 gene of CHIKV indicates only three lineages with
distinct genotypic and antigenic characteristics i.e. the “Central/East African
genotype”, the “Asian genotype” and the “West African genotype”. CHIKV strains
with an Asian genotype of E1 gene were reportedly detected during the 1963–73
outbreaks in India, while the more recent outbreaks since 2005 have been caused
by the “Central/East African genotype”. Additionally, a mutation at 226 amino-
acid (Valine–Alanine) of E1 gene was observed during the recent outbreaks and
has been associated with the more efficient replication of CHIKV in *Aedes
albopictus*.
The high morbidity and loss in daily activity associated with CHIKV infection
results in considerable economic loss among the affected nations, specifically
India. This emphasizes the need to have a detailed understanding of epidemiology
and strain diversity for planning a prevention strategy. Towards this end the
present study aimed to evaluate the disease prevalence at various geographical
regions and assess the genomic diversity among CHIKV strains infecting the
Indian population.
# Methods
## Study sites
Patients were recruited at three distantly located regions of India (North, West
and South) from 1<sup>st</sup> June, 2008 through 31<sup>st</sup> May, 2009.
Patient enrolment was carried out year round at Karnataka Institute of Medical
Sciences (KIMS), Hubli, Karnataka (south); Sawai Man Singh Medical College (SMS)
Jaipur, Rajasthan (West), and All India Institute of Medical Sciences (AIIMS)
New Delhi (north) with AIIMS as the coordinating and testing centre.
## Patients and clinical specimens
Patient enrolment involved those with fever ≤7 days duration at the time of
their visit to the Out/Inpatient departments at AIIMS (n = 178), KIMS (n = 233)
and SMS (n = 129) hospitals and those who were willing to participate in the
study (n = 540). The study was approved by each Institution's Ethics Committee
(Institutional Ethics Committee, AIIMS; Ethics Committee SMS Medical College,
Ethics Committee Karnataka Institute of medical Sciences, Hubli). All patients
(or for children parents or guardians) gave written informed consents.
Demographic details, clinical data including symptoms and onset and duration of
fever were recorded in the specified proforma. Approximately 3 ml of blood was
collected from each patient enrolled at the time of admission. Convalescent
phase (15–90 days of fever onset) sample could be collected from 30 patients
only at KIMS center. Sera were separated from the blood by centrifugation and
stored at −70°C in aliquots till further use. Specimens from KIMS and SMS
centers were subsequently transferred to AIIMS center within two to three months
of collection. Both RTPCR assay and serology (IgM) were employed to test CHIKV
in all study specimens at the AIIMS centre.
## Diagnostic RT-PCR
Genomic viral RNA from CHIKV infected culture-lysate/patient sera were extracted
using QIA Amp Viral RNA minikit (QIAGEN, Germany) according to manufacturer's
protocol. The 294b E1 gene fragment of CHIKV was amplified with Qiagen one Step
RT-PCR kit as described previously. Briefly, the RNA fragment was reverse
transcribed and amplified in a single step and the 294 bp amplified product was
analyzed on 2% agarose gel as reported earlier.
## IgM-ELISA
All serum specimens were screened for CHIKV specific IgM antibodies by ELISA
using a commercial kit (Standard Diagnostics, Inc., Korea) according to
manufacturer's recommended procedure.
## Nucleotide sequencing
Sequences were determined from the 913 bp E1-amplicon (nt 10246–11158) that had
been generated following RT-PCR (n = 15) with forward (5′-TACCCATTCATGTGGGGC-3′)
and reverse (5′-TGACTATGTGGTCCTTCGGGGG-3′) primers. Amplification product was
gel purified by the Gene-Clean purification method (Q-biogene, Cambridge, UK)
and sequenced using a previously described method. Lineage identification and
assessment of evolutionary relationships were performed using Bioedit and Mega2
programs as described previously.
## Statistical analysis
Statistical analysis was performed using Stata statistical software, version 7.0
(StataCorp LP, College Station, TX, USA).
# Results
## Clinical features among suspected chikungunya patients
lists patient demographics from a total of 540 cases with fever up to 7days
duration regardless of chikungunya virus infection. Among the clinical features
recorded, joint pain (62.8%) and headache (63.3%) were most frequently observed
while other features included abdominal pain (48.1%), vomiting (43.9%), rash
(36.1%), joint swelling (27.8%), cough (27.2%) and restlessness (21.3%).
Symptoms such as diarrhea, rhinitis, conjunctival congestion, hepatomegaly,
splenomegaly and lymphadenopathy were rarely observed. Symptoms commonly caused
by CHIKV i.e. joint pain, joint swelling, abdominal pain and conjuctival
congestions were more common among patients at KIMS as compared to SMS and
AIIMS.
## Laboratory based diagnostics of Chikungunya viruses
CHIKV infection was tested both for the presence of anti-CHIKV IgM antibodies by
ELISA and CHIKV RNA by RT-PCR. A total of 137 (25.4%) subjects tested positive
for CHIKV through either of the two tests. A solitary case was detected at AIIMS
while relatively higher percentages were observed at SMS (16.3%) and KIMS
(49.4%). Rate of chikungunya detection was nearly similar between ELISA and RT-
PCR. Year round detection of chikungunya was observed at KIMS while at SMS it
was rarely detected during summer. Moreover, majority of the CHIKV cases were
observed during monsoon and post monsoon seasons both at SMS and KIMS.
## Time kinetics of CHIKV RNA detection by RT-PCR and anti-CHIKV IgM antibodies by ELISA
Chikungunya detection was performed by both RT-PCR and IgM-ELISA with detection
rates ranging from 4.3%–90% for RT-PCR and 10%–95.6% for ELISA depending on the
days of fever onset. This difference in detection rates between RT-PCR and ELISA
with respect to days of fever onset was found to be statistically significant
through chi square analysis (p\<0.05). RT-PCR assay was more sensitive till
first 4 days of fever onset declining after day 4 at which point serology (IgM)
was observed to be more effective.
Additionally 30 cases from KIMS were assessed for persistence of IgM during
convalescence (15–90 days of fever onset). While majority of the chikungunya
positive cases (9 of 11) were detected anti-CHIK IgM between 15–90 days post
illness, none of the chikungunya negative patients (n = 19) were detected the
same (data not shown).
## Association of clinical features with CHIKV laboratory diagnosis
Comparison of clinical features between chikungunya positive and negative
patients demonstrated rashes, headache, joint pains, joint swelling, abdominal
pain, cough and vomiting to be significantly associated with CHIKV confirmed
cases (p\<0.05) whereas differences for other symptoms were not found
significant.
## Age wise distribution of CHIKV infection among the study population
We compared the rate of CHIKV prevalence based on patient age groups 0–5, \>5–18
and \>18 years of age and observed significantly higher detection (p\<0.05)
among adult populations (aged \>18 years). An apparent increase in frequency of
CHIKV detection with age, from about 16% cases among children to 50% among
adults was observed.
In addition Chi square analysis revealed significant differences for presence of
rashes, headache, joint pain and joint swelling (p\<0.05), between children
(\<18 years) and adults (\>18 years). Interestingly, while most of the rashes
among adult cases (\>18 years) were of erythrematous type, that of children were
maculaopapular type (\<18 years) (data not shown).
## Phylogenetic tree based on E1 gene of CHIKV demonstrates high homology with strains belonging to Central/East African genotype
Comparative sequence analysis of chikungunya E1 gene from 15 CHIKV infected
patients from KIMS revealed 98.3%–99.9% homology among them, clustering within
the Central/East African genotype. Furthermore, all recently detected CHIKV
strains worldwide including India clustered in this lineage with 93.4% to 99.9%
homology. However, recently detected CHIKV strains from India appeared different
from those of previous 1963 and 1973 outbreaks. were similar when the analysis
was performed with amino acid sequences deduced from the E1 gene sequences (data
not shown). Additionally we detected an amino acid change from lysine to
glutamine at residue 132 in E1 gene among 5 of 8 CHIKV infected children, while
none among the adults (data not shown). Of note, all these 15 cases were
enrolled in the same place and during same period.
# Discussion
The present multi-centre study confirms and extends the findings of recent
reports from India and other parts of world indicating a re-emergence of severe
chikungunya disease which is becoming a major public threat in India. This is
the first multi-centre study with sites from three distinct geographical
locations except the eastern part of the country. The study was carried out with
year round enrolment of subjects in comparison to previous CHIKV epidemiological
data that was derived mainly from outbreaks,. For the first time both genomic
(RT-PCR) and serology (IgM ELISA) based assays were employed throughout the
study for laboratory confirmation of chikungunya in a large sample population.
However our data on IgM antibody persistence emphasizes the need to analyze
paired sample for IgM or IgG test for confirmation of chikungunya infection,
particularly for those sampled after day-3 post illness when viremia is low.
Amongst the three centres, AIIMS and SMS had a relatively higher male∶female
ratio. However this could be attributed to the social bias generally observed
against females in the north and west India rather than to disease
susceptibility based on gender. Various other studies conducted worldwide have
shown inconsistencies on gender bias to disease susceptibility with most studies
reporting each gender to be equally susceptible.
Discrepancies in the symptoms of Chikungunya among suspected cases across
different studies warrant the requirement of a clear and well defined clinical
profile of the disease. For example our study detected rashes more frequently
among CHIKV cases as compared to other. One reason for this discrepancy could be
due to the study design. While earlier studies were retrospective focussing
largely on outbreaks, our study was conducted throughout the year and across
different sites.
Our study also demonstrated a link between chikungunya detection rate and
geographical location of study centre. We observed highest percentage of cases
at KIMS which is located along the coast in the southern part of the country
compared to the other two centres that are landlocked regions with extended
periods of dry summer and extreme winter. This is in agreement with findings of
other studies those reported the prevalence of chikungunya mostly in the coastal
regions including South India and South East Asia.
One interesting observation made in the present study was the near absence of
chikungunya case at AIIMS site despite reports of dengue prevalence. As both
dengue and chikungunya are transmitted by the same vectors, it is not
unreasonable to expect similar disease prevalence. However reports of the Delhi
State Government showed large number of dengue cases in Delhi during 2008, while
only few clinically suspected chikungunya cases were reported. Similarly a study
from Delhi detected CHIKV virus in serum samples of patients hospitalized during
a dengue outbreak in the year 2006. Since then no confirmed cases of CHIKV
infections have been reported from Delhi. One can speculate the reason for the
absence of chikungunya to be simply a time factor and that in time CHIKV may
make its way from the coastal regions to more inland areas such as Delhi. Hence,
this warrants year round surveillance of CHIKV in regions that already have high
prevalence of dengue. Beside the geographical location another reason for the
lower detection of chikungunya could be due to silent infections that may have
gone unnoticed in the past and thus may have primed the immune system against
this pathogen. A study from Chennai, nearly a decade prior to the 1964 outbreak,
observed high seropositivity among older subjects and they speculated the waning
antibodies to CHIKV as a possible reason for the subsequent outbreak. More
studies need to be undertaken to prove this hypothesis and to further address
the susceptibility of the Delhi populace to chikungunya infection. This is
supported by reports of periodic outbreaks of dengue fever in Delhi every 3–4
years.
Chikungunya fever has been observed to occur mostly during the monsoon and post
monsoon seasons during which time there is high breeding of both *Aedes
albopictus* and *Aedes aegypti.* In India the first CHIKV outbreak in 1963 was
observed during July to December, coinciding with the monsoon and post monsoon
seasons. However, in the present study, CHIKV was detected throughout the year
at KIMS centre although with higher rate during the monsoon season, a situation
similar to dengue.
Our study has confirmed the findings of various other studies that certain
symptoms like rashes, headache, joint pain/swelling and abdominal pain are
significantly associated with chikungunya infections.
Our finding on higher detection rate of chikungunya among adults as compared to
children confirms most of those reported worldwide. In addition we found adults
more frequently exhibiting symptoms such as rashes, headache, joint pain and
joint swelling as compared to children. Interestingly among rashes,
erythrematous rash was frequently observed in adults while maculopapular rash
was common in children confirming earlier reports in infants. Another
interesting observation made in our study was the higher incidence of
hepatomegaly among children aged below 5 years. Support for this result can be
found in a recent pathogenesis study in mice that demonstrated hepatic
involvement during chikungunya infection and showed CHIKV to replicate first in
the liver before targeting the muscle and joints. In addition, a recent review
reported many atypical symptoms including raised levels of hepatic enzymes
associated with chikungunya infection.
In addition to assessing clinical and epidemiological implications of
chikungunya we tried to identify the molecular composition of the infecting
strains. High nucleotide and amino acid homology was observed among the
infecting strains which clustered within the Central/East African genotype
confirming earlier reports that all recent CHIKV outbreaks in India have been a
result of the Central/East African genotype. This is markedly different from the
CHIKV strains that caused the outbreak in India during 1963–73 that resulted
from the Asian genotype. Our study also confirms earlier reports on the presence
of A226V mutation in the E1 gene. It has been suggested that a change in the
amino acid at this particular position may provide a selective advantage for
CHIKV transmission by mosquitoes by reducing their cholesterol dependence. Our
study also observed a lysine-glutamine change at aa 132 in the E1 gene of CHIKV
infecting children, an implication that warrants further investigation.
In conclusion, to our knowledge this is the first multi-centre study undertaken
to determine the epidemiology of chikungunya fever across a wide age group with
year round enrolment of large number of patients simultaneously in three
different geographical regions of India. Both genomic and serology based assays
were employed throughout the study for confirmation of CHIKV. Our study clearly
demonstrates high incidence of CHIKV in India particularly in the southern
region with moderate prevalence in the western and rare in the northern region.
This warrants continuous CHIKV surveillance in India to determine the disease
burden for improved healthcare.
The authors would like to thank the doctors and nursing staff of AIIMS, KIMS and
SMS hospitals for sample collection and the patients and parents of the children
for their participation. The authors thank Prof. M. K. Bhan for helpful
suggestions and Prof. V.K. Paul, the department chair for general support.
[^1]: Conceived and designed the experiments: PR VHR SKK BSS NW. Performed
the experiments: PR. Analyzed the data: PR RL MK. Contributed
reagents/materials/analysis tools: PR SKK VHR BSS NW. Wrote the paper: PR
SS.
[^2]: The authors have declared that no competing interests exist. |
# Introduction
In 1990, the Commission for Health Research and Development (COHRED) reported
that less than 10% of global health research funding was spent on health
conditions that account for 90% of the global disease burden. This led to an
outcry that if it remained the status quo, global inequities in health will
persist. COHRED also highlighted the important role health research could play
in reducing global health inequities. This has since led to an increase in the
number international health research projects in Africa. International health
research is research conducted in low and middle income countries (LMICs) with
funding from institutions and organisations in high income countries (HICs).
Clinical trials have dominated the international health research landscape,
however there is currently a growing number of genomics research and biobanking
projects in Africa which are funded by institutions and organizations in HICs.
Examples include: the African Genome Variation Project; Human Heredity and
Health in Africa (H3Africa) Consortium; and the Bridging Biobanking and
Biomedical Research across Europe and Africa (B3Africa) initiative. The
explanation for the growing interest in genomics research and biobanking in
Africa is that it will facilitate cutting-edge health research on African
populations; prevent a genomics divide between HICs and Africa; reduce global
health inequities or–more skeptically–generate evidence that will be of benefit
to genetically homogeneous populations.
Despite the potential of international health research to reduce global
inequities in health and research, it also has the potential to exploit research
participants and African researchers. In the late 1990s, for example, there was
a spark of debates on the ethics of international health research in Africa, a
number of which highlighted the impact social and economic differences between
HICs and LMICs could impact on international health research. The debates also
suggested that the asymmetrical nature of these partnerships tend favor research
collaborators in HICs while their African counterparts ended up as sample
collectors.
Contemporary genomics research and biobanking initiatives in Africa are
promising to transform the way international health research has typically been
structured in Africa. They hope to minimize the possibility of exploitation of
African researchers. Their plan is to go beyond the traditional practice of
collecting samples in Africa and conducting the scientific analyses outside of
the continent, to one that fosters equitable research collaborations. While in
theory, the intention to foster equitable research partnerships is laudable,
these promises remain largely unchecked. There is a need to go beyond the
documented promises and explore whether and how genomics research and biobanking
initiatives in Africa are going about realizing these goals, and how successful
their efforts are perceived to be by their LMIC partners. In this paper, we
document the perceptions and expectations of genomics researchers in Africa, on
the benefits of externally funded genomics research and biobanking initiatives
in Africa. This study identifies challenges pertaining to equity in research
collaborations which HIC partners should consider when they engage in
international health research in Africa.
# Methods
We adopted a qualitative research methodology. Between September 2014 and June
2015, 17 face-to-face semi-structured interviews were conducted with genomics
researchers in Africa. Purposive sampling was used to select research
participants. Research participants were genomics researchers based in an
African research institution and who were directly involved in an international
genomics research and biobanking project. Principal investigators and co-
principal investigators within the H3Africa Consortium were first invited to be
part of the study. We selected this group of participants because, we are part
of the consortium and therefore were certain that participants were actively
involved in international health research projects. After interviews with these
group of researchers, we invited non-H3Africa researchers, involved in other
collaborative genomics research or biobanking initiatives in Africa, to be part
of the study. The reason was to attain data saturation. H3Africa is an African-
led genomics research and biobanking initiative with funding from the National
Institute of Health, USA and The Wellcome Trust, United Kingdom.
Research ethics clearance was obtained from the Faculty of Health Sciences
Health Research Ethics Committee, University of Cape Town (HREC-Ref-618/2014).
Written informed consent was obtained from all research participants before the
start of the study.
Of the 17 interviewees, 14 were part of the H3Africa consortium while 3 were not
involved in H3Africa. Interviewees were based in 8 African countries and
included biomedical and social scientists. At the time of the interviews, 13 of
the 17 researchers were principal investigators or co-principal investigators
for a genomics research and/or biobanking project in Africa, while four were
research scientists.
The researchers were asked questions about their experiences of being part of an
international health research consortium; what they perceived to be the benefits
and risks; and their recommendations on how international health research should
be done in Africa. All interviews were conducted in English and lasted
approximately one hour.
Audio recorded interviews were transcribed verbatim. Iterative data analysis
started immediately after the first few interviews and continued throughout the
data collection phase. This allowed for the identification of themes and
patterns that were emerging from the data, which were further probed in
subsequent interviews. Data collection continued until we reached saturation
point where no new patterns or categories were emerging from the interviews. To
facilitate analysis and management, we usedNVivo 10, a text-based data analysis
software. We performed inductive thematic analysis of the interview transcripts
with the objective of identifying, examining and recording patterns of meaning
across the dataset. These patterns were identified through a process of data
familiarization, data coding and the development and revision of themes and
models. NSM did the initial coding for the first few interview transcripts.
Following the first round of coding, codes were discussed with JDV. NSM and JDV
then separately coded several transcripts to establish validity of the coding
scheme and its application. Differences in coding were discussed and the coding
scheme was adapted to clarify ambiguities. NSM continued to code the entire
dataset and interpreted the coded data. Challenges and evolving data
interpretations were discussed regularly with JDV.
# Results
In the interviews, as genomic researchers shared their perceptions, expectations
and recommendations for international health research partnerships in Africa, a
number of significant insights emerged.
All interviewees were of the view that externally funded genomics research and
biobanking initiatives in Africa provided a platform for researchers in Africa
to to collaborate, on a large scale, with researchers from different
disciplinary backgrounds and from different parts of the world. This, they
hoped, will play a critical role in narrowing the existing gap in genomics
research capacity between HICs and LMICs and in preventing a possible genomic
divide between Africa and HICs.
> This opportunity will allow us to bring genomics in Africa to a level > that
is comparable to other parts of the world and it could > potentially lead to new
therapies and treatment strategies that are > relevant, needed in Africa for
which we couldn’t do by using results > from genomics research from other parts
of the world either because of > the differences in the inheritability of the
different conditions in > Africa compared to other parts of the world or because
of the > gene-environment interactions in different parts of Africa and that is
> different from what you see in other parts of the world. (R-01)
A view that has also been articulated by other researchers in Africa. In
one case, An African researcher said that without international health research,
researchers in Africa, would be sitting all day in their offices, ‘reading
newspapers’ instead of doing research. Although all our interviewees cited the
opportunity to collaborate as a benefit, some concurrently stressed that these
collaborations carry mutual benefits for all partners, including collaborators
in HICs.
> This part of the world has not got the necessary equipment to carry > out
certain experiments. The way to go is to collaborate. By > collaborating, each
individual group brings in particular expertise. > So if we are working on
African genomics then the first thing that the > African scientists can bring in
is the African genome and then the > other people can come in from other sides
and bring in their expertise > and all that. (R-17)
Recognition of this mutual benefit is important for international health
research partnerships in Africa. International health research has often been
portrayed as a form of development aid or altruistic philanthropy, where the
receiver is expected to show some gratitude to the giver. This has led to
situations whereby HIC collaborators had said to their African collaborators
that they were only hired to do the research and not to be involved in key
decision making activities. This is problematic in that it highlights
international health research as a patronizing and neocolonial activity.
Nevertheless, African researchers are increasingly judging these collaborations
as being of mutual benefit to all partners, and as a result want their HIC
collaborators not to treat them as employees but as partners. For example, some
African researchers are now willing to ask their collaborators, upfront, without
fear of losing a funding opportunity, “*Do you want me to work with you or to
work for you?*”. These are indeed good signs for international health research
in Africa because when research collaborations are seen as mutually beneficial
to all partners, they are arguably closer to being equitable. It is therefore
imperative that HIC partners involved in these collaborations recognize the
implications this may have in achieving equitable collaborations when they setup
international health research partnerships in Africa.
## Fears of exploitation of African researchers
The recognition by African researchers that international health research
partnerships are mutually beneficial has however not allayed fears of
exploitation of African researchers by their partners in HICs. Despite being
generally supportive of African genomics research and biobanking initiatives in
Africa, all our interviewees expressed concerns that they may end up being
exploited in these collaborations. These fears were primarily shaped by past
experiences of exploitation of African scientists.
> It could be that African researchers, as it has happened before, are > just
being used for collecting materials and that is a very big > potential \[for
exploitation\] because if you don’t have capacity to > analyse, make sense out
of it, you just collect the material and the > data and send it to people who
can analyse and that could be a benefit > not only for publication of an article
but longer term benefit of > patents and discoveries… Those are some of the
potential risks. (R-07)
The fear of exploitation is made worse by uncertainty about sustainability of
genomics research and biobanking projects in Africa. One skeptical view
explaining the growing interest in African samples is that access to Africa’s
rich human genetic diversity makes the continent a choice destination for
population genomics studies –though not always to the benefit of African
populations, patients or researchers. Interviewees expressed concerns that they
were unsure of what will happen to their research projects once the current
funding period is over and they no longer have funds to enable them use the
samples and data collected in the initial stages of these genomics research and
biobanking projects.
For example, about two thirds of our interviewees mentioned that whilst their
collaborators in HICs have the capacity and resources to continue research on
stored samples and the data that were jointly obtained during the funded
collaboration, they would not have access to similar research funds. The risk,
therefore, is that though initial studies may seek to establish more equitable
forms of collaboration, African researchers may still be marginalized in
subsequent research projects.
> Our capacity just to handle the data first, to analyse them, to handle > the
samples, to analyse them or even to have research means in terms > of funding,
to take advantage of those samples or data is by far > limited as compared to
our international partners that are in the same > project. So what will happen
if we design our own research questions > to take advantage of the biobank? Will
we still be able to have the > amount of funding as we are having now in
\[consortium name\] to > respond to our own question? Otherwise you will see
that after > \[consortium name\] the biobanks will benefit more the
international > scientists because they will have more resources and we might
end up > working in collaboration again with international scientists on their >
agenda and not our agenda (R-05)
This researcher highlights not only the importance of equity to access to
samples, data and funding to support their continued work but also that
researchers in Africa want to be able to drive their own research agendas. A
theme is further discussed below. The inequitable nature of international health
research collaborations have led to calls for funders of international health
research to build capacity for health research in Africa as well as for HICs
partners to engage with their African collaborators in ways that build mutual
trust and respect \[, –\]. In the interviews, we asked the researchers how these
fears of exploitation may be assuaged. Their responses can be grouped into two
broad categories: Equitable research partnerships and research capacity
building.
## Establishing equitable research partnerships
The nature of research collaborations between LMIC researchers and HIC
researchers is often uneven in terms of access to research funding, research
resources and involvement in decision making. These power imbalances have made
African research partners to oftentimes remain silent about worrying
inequalities. As we described above, a two-thirds majority of our interviewees
expressed the desire to be in equitable research partnerships. They also had
suggestions for how equitable collaborations may be achieved. This included:
setting the “rules” of engagement, involvement of African researchers in
decision making and African leadership of international health research
projects.
### Setting the “rules” of engagement
The first suggestion advanced by our interviewees as one of the major ways of
achieving transparent and equitable research partnerships is having “rules of
engagement” between collaborators.
> I think importantly around developing the rules of engagement, so, > making
sure that Africa benefits from the research, not just > \[consortium name\]
research but that this kind of work done in > Africa, I mean, setting the pace,
that this is how genomics research > should be done in Africa so that Africa
benefits. (R-12)
Slightly over half of our interviewees who expressed the desire for equitable
research partnerships specified that the rules of engagement need to be defined
before the start of the collaborations. For these interviewees, it is critical
that at an early stage of the collaboration, all partners have an idea of what
is expected of each collaborator (both LMIC and HIC partner). Recently, some
authors have recorded how collaborators in HICs have dictated what needs to be
done in research collaborations and in some cases have said their African
collaborators were hired to do research and should implement as requested. The
resistance of our interviewees to sign up to this way of collaborating may
therefore be evidence that they are fighting what Okeke has termed the ‘little
brother effect’ in African biosciences, that is, north-south partnerships that
are a mirror of paternalism and colonial hegemony even though the HIC partners
may have initially displayed good intentions at the beginning.
### Involvement of African researchers in decision making
A second way in which more than half of the interviewees felt equitable
partnerships may be achieved was for researchers in Africa to be actively
involved in decision making processes in the consortia. As one of the
interviewees stated, these collaborations can only be considered fair if they
are treated as equal by their collaborators through being involved in decision
making activities.
> There should also be regulatory procedures that will make sure that > the
African scientists are involved and are central to any decision > involving the
use of the samples and are actually involved in the > publications and the
intellectual property that emanates from such > processes. African scientists
should be at the centre of all of this. > So it is not disadvantaging anyone
(R-06)
Again, these are indications that African researchers do not want to be
considered as hired workers but as equal research partners. From the interviews,
it is clear that African researchers want to be involved in making decisions on
the use of samples collected as part of genomics research and biobanking in
Africa. Genomics research and biobanking are generally characterized by the
sharing of samples and data and most times the cross-border movement of samples
for the purposes of analysis.
### African leadership of genomics research and biobanking in Africa
The perceived fears of exploitation has also led to an expressed yearning for an
African leadership of international health research consortia in Africa. More
than half of the interviewees recommended that African researchers should be at
the forefront in the design and conduct of genomics research and biobanking
projects in Africa. As one interviewee said, the story of Africans needs to be
told by Africans. Interviewees who articulated a preference for African
leadership of research projects in Africa, were clear that they wanted to be
involved in all phases of the studies including research design and analysis.
> First of all, African scientists should be involved in doing this > research.
It should not be research that comes from outside Africa, > being implemented by
non-Africans researchers in Africa. (R-14) > > Well I think that it \[African
ownership\] is a good idea in the sense > that we will not have a stranger
telling us our story that is the > positive part of it. (R-16)
This perception is similar to what Okwaro and Geissler reported in a study on
scientific collaborations in HIV research whereby researchers in an African
research institution expressed interest in leading projects in Africa and
criticized the use of the term “local PI” as a way of demonstrating local
leadership of projects whilst in actual facts ‘local PIs’ are often only used to
implement projects designed by northern collaborators. Our interviewees objected
to such tokenistic involvement in decision-making and were rather expressing a
desire for substantive involvement.
The desire for equitable research partnerships does not necessarily mean that
African researchers thought all collaborators needed to be equal. In fact all
the researchers acknowledged that there were disparities in resources and
capacity for genomics research between HIC and LMIC investigators, but indicated
that this should not stop them from active and equal participation in network
activities as their collaborators in HICs.
> We must have a critical mass of people who can talk the language of >
genomics, who can, you know, when there is collaboration their > participation
is equal. Maybe not in terms of funding but certainly in > terms of intellectual
contribution. (R-03)
The acknowledgment of these inequalities and how it may impact on the success of
international health research in Africa is a step towards achieving equitable
international health research partnerships. It is also contrary to what may have
been expected a few years back, and shows the willingness and eagerness of
African researchers to take up roles that go beyond sample collection.
## Research capacity building
Besides building equitable research partnerships as a way of minimizing the
possibility of exploitation of African researchers, all our interviewees also
said research capacity building will play a big role in minimising exploitation
of African researchers and in achieving equitable research partnerships. Some
ongoing genomics research and biobanking initiatives in Africa are promising to
build research capacity in Africa to ensure that genomics research is done in
Africa by Africans. But whilst building health research capacity in LMICs is key
to promoting justice in global health research, the central question is which
type of capacity building efforts can foster equitable genomics research and
biobanking collaborations in Africa. When tour interviewees talked about
capacity building for genomics research and biobanking in Africa, it was in
three major areas: building of human capacity, infrastructural capacity building
and the sustainable access to funding.
### Research capacity building: Training and skill development
Training of researchers in LMICs is critical for the success and sustainability
of international health research in Africa. Our interviewees confirmed that this
was one of the key features of the genomics research and biobanking initiatives
they were involved in. Also, researchers who were part of H3Africa or had
knowledge of H3Africa activities, identified the training of the next generation
of African scientists as one of the main benefits of the collaboration.
> In terms of building capacity, training and giving opportunities for > many
African students. So that will accelerate capacity within Africa. > (R-05)
Training of researchers within these consortia are taking place in institutions
in both LMICs and HICs One of our interviewees mentioned that in their project
some of the students are being trained out of their home countries and this
gives them the opportunity to expand their skills and research networks.
> We currently have two trainees from Country Y \[LMIC\] at Country Z > \[HIC\]
at the Z college of Medicine, doing their rotation of > laboratories towards
their PhDs in genomic sciences and genomic > medicine. We also have three from
Country X \[LMIC\] and we are > expecting two more from Country Y \[LMIC\] next
year. (R-02)
Another form of human capacity building that has been part of the activities of
H3Africa and which more than three quarters of the interviewees identified as a
major benefit, is postdoctoral skills development, including a focus on
transferable skills. These included training in laboratory techniques,
statistics, data analysis and grant writing.
> Capacity building in our research to help young African researchers, > develop
themselves in many fields related to research: statistic, >
epidemiology……Through this project, I have developed some skills in > writing
grant proposals (R-09)
Genomics research and biobanking are relatively new in Africa. Many researchers
who now work in this field did not receive formal training in genetics, genomics
or associated fields such as bioinformatics. In fact, many of our interviewees,
and of the H3Africa researchers more generally, are medical doctors with a
strong research record in investigating particular diseases. Whilst being world-
experts on these diseases, they do not necessarily have expertise in genomics.
Training in key genomic research skills such as for instance those relating to
bioinformatics is essential for establishing successful and equitable
partnerships in genomics research and biobanking agenda in Africa. The
sustainability of such training is a key concern, however. One researcher
described that his involvement in H3Africa has provided an opportunity for their
research team, together with their collaborators in HICs, to develop degree
programs in genomics at African universities as a way of ensuring
sustainability.
> We are working with the Faculty of Health Sciences in both Country X >
\[LMIC\] and Country Y \[LMIC\] to create masters programs, PhD > programs in
genomics and bioinformatics. We hope that what will be a > lasting and
sustainable benefit to those countries, is by having a > local program in
genetics and genomics (R-02)
Although just one of our interviewees mentioned this as a sustainable capacity
building effort within H3Africa, it is one that will be worth emulating and
should be extended to other African countries. Research capacity building
strategies aimed at developing undergraduate and postgraduate research programs
in African universities and research institutions are vital in sustaining health
research in Africa.
The third and last form of training and skills development articulated by almost
a third of our interviewees was capacity building in research ethics at both the
professional and institutional level. This benefit was only mentioned by
H3Africa researchers and they were of the opinion that there has been some
improvement in research oversight for genomics research and biobanking in Africa
as a result of H3Africa activities.
At the professional level, it has made African biomedical researchers to think
beyond the science of their projects to the ethical issues raised by their
research projects. Typically, African researchers have demonstrated a knowledge
gap in research ethics and so engaging them in research ethics debates may help
improve research oversight in Africa
> I can almost be sure that if you visited any study site at the > beginning of
H3Africa and visit them now, the way the will be > consenting and engaging with
the community will be completely > different, I mean in ethics, they have been
challenged to think far > beyond just the compensation for travel, for time and
everything … > even our ethics committees have been touched to think far more
(R-13)
Weak research oversight systems in Africa challenge the execution of health
research projects in Africa and make African researchers and research
participants more vulnerable to exploitation. Genomics research and biobanking
in particular raise unique ethical challenges that may be quite different from
other forms of international health research such as clinical trials that ethics
committees review more habitually. Furthermore, methods of genomics research and
biobanking are relatively new to REC members in Africa and they may not be fully
equipped to review such studies. It is therefore important that for
international health research projects that will involve sharing of samples and
data and that raise complex ethical issues, ethics committee members be
appropriately engaged in discussions on genomics research and biobanking. One
interviewee mentioned that as a result of meetings organised by H3Africa,
research ethics committee members in Africa are becoming familiar with concepts
in genomics research and biobanking. In his/her opinion, this had implications
in research oversight for genomics research and biobanking in Africa.
> Another thing which is already happening that is good is involving the >
ethical review committees in giving them additional training in > genomics
research because that is a new area for a lot of the IRBs. In > Country X we
found that the IRBs had not reviewed this type of > research before and this
caused a lot of delay in terms of getting > approvals and feedback from them but
things improved after one of the > IRB members attended the last consortium
meeting for the special > session for ethical review committees. (R-02)
Capacity building in research ethics has impact on regulation of health research
in Africa as well as the protection of research participants. Africa has
historically been characterised by a history of “parachute” or “postal” research
whereby researchers from HICs have come to Africa just for the samples and then
“disappear” once samples have been collected. This has led to research ethics
committees being overly cautious when reviewing projects and in some cases it
has led to tight national regulation for the export of samples. It will appear
that genomics and biobanking initiatives in Africa are trying to engage research
ethics committees in Africa and to build their capacity to be able to identify
ELSIs pertinent to genomics research and biobanking. Infrastructural research
capacity building.
A second major form of capacity building articulated by all our interviewees is
infrastructural capacity building, mainly the setting up of biobanks and the
acquisition of laboratory equipment. For these researchers, having a biobank in
Africa will be a great resource for biomedical research in Africa.
> It \[consortium name\] has really emphasized on building > infrastructure in
Africa, giving Africans the tools to solve their > problems.....One big
advantage of having a biobank in Africa is that > it really puts all of our
resources together. I think that is very > huge for Africa, to begin with. And
it is a way of putting our > resources together and giving us the infrastructure
to produce high > quality research (R-13)
Biobanks are repositories where organized collections of human biological
materials, and associated data from large numbers of individuals, are collected,
stored and distributed for the purpose of scientific investigations or public
health use. Biobanks are therefore an important resource for biomedical
research. Also, well curated biobanks in Africa could foster international
health research collaborations, south-south collaborations and promote
biomedical research in Africa and globally. However, for researchers in Africa
to make optimal use of the samples and data stored in the biobanks, they will
have to secure the necessary laboratory equipment. One of our interviewees
mentioned that their genomics research and biobanking project has given them an
opportunity to source for recent and novel laboratory equipment.
> In terms of infrastructure, we are building capacity at Q University > and
university of Country X \[LMIC\] to do genome sequencing and to > get their
sequencers, their Illumina sequencer and so all of those > tools will help in
the building of infrastructure in the different > laboratories. (R-02)
Besides the modest infrastructural capacity for genomics research and Biobanking
in Africa, limited access to technology remains a serious impediment to
biomedical research in Africa. It is also a major reason for the export of
samples from Africa to HIC, especially when high-throughput technology is
required. To minimize export of samples, researchers in Africa will need to
acquire up-to date laboratory equipment. As one interviewee explained, African
countries must emulate LMICs such as India and China and build their
infrastructural capacity for genomics research and biobanking.
> We should not just be thinking that anything we want to do, we have to >
collect samples and send to Europe, samples to China, samples to > America. We
need to build the capacity that we can do work in Africa. > If that is the case
then we can produce things that will be valuable > and useable in Africa. And I
think that is just what has been done in > several other countries like India.
It is very difficult now to take > anything out of India for analysis. (R-14)
Export of samples from Africa is a major contributor to exploitation of African
researchers. The acquisition of laboratory equipment that will permit research
work to be done in Africa will minimize this risk–but only if equipment is
maintained. Unfortunately, in many African research institutions, laboratory
equipment which are primarily obtained through funds from international health
research tend to lie loose with little use once the projects they were obtained
for are over. In the unfortunate case of a breakdown, equipment worth thousands
of dollars are usually abandoned to the dust for sheer of lack of resources to
maintain them. The virtual absence of an infrastructural maintenance culture by
research institutions in Africa (both in providing resources and tools required
for maintenance) and the impact it has on health research cannot be missed even
by a passing observer.
### Access to funding
The last and major form of research capacity building is sustainable access to
funding for health research. This was mentioned by almost half of the
interviewees. For these interviewees, being part of a funded genomics research
and biobanking consortium has been a unique opportunity to access and administer
funds for large-scale research projects.
> For researchers, what it is really doing is, it’s building their > capacity to
do research, to do large scale research that they haven’t > done before because
it costs a lot of money to do this type of > research. (R-15)
Research funding is a major driver of change for biomedical research. A dearth
of funding for health research coupled with the lack of research infrastructure
(laboratories, biorepositories, databases has held back African scientists from
carrying out rigorous health research. The importance of funding health research
in Africa has been presented to African governments and while there was a
general buy-in by African governments followed by a commitment to allocate a
small proportion of their annual national growth domestic product (GDP) to
research, there has been little compliance by a vast majority of African
governments. Where there has been government commitment, such as in South
Africa, tremendous progress has been made in advancing health research. African
governments will have to take responsibility and devise research funding schemes
that can sustain and foster health research in Africa. One African researcher
and a pioneer in research capacity building in Africa has suggested that a Pan-
African research funding agency may solve the problem of limited local funding
for health research. Organizations like the Alliance for Accelerating Excellence
in Science in Africa (<http://aesa.ac.ke/>), a new partnership between the
African Academy of Sciences and the New Partnership for Africa’s Development
(NEPAD) may play the role of such a Pan-African funding agency especially as it
is acting as an agenda setting and funding platform to address Africa’s health
and development challenges. However, it is still funded by the Wellcome Trust
and may therefore face the crisis of non-sustainability should there be no
financial commitment from African governments and the private sector in Africa
to support its activities.
## Challenges in building capacity for genomics research and biobanking in Africa
While research capacity building could be a major way of achieving fairness in
international health research partnerships in Africa, it is hard to sustain
research capacity building efforts and to retain human research capital in
Africa. Brain drain, for example, is a major impediment to research capacity
building efforts in Africa. About a third of our interviewees expressed concerns
that although young African scientists are being trained in genomics research
and biobanking, the lack of an enabling research environment in Africa could see
them migrating to research institutions in HICs. Some interviewees suggested
that assurances of a research career in African institutions could curb brain
drain. One interviewee explained how their project is working towards overcoming
the problem of brain drain
> We want these trainees to come back and do research that benefits > their own
countries. And so there is an agreement with the host > institutions that there
will be faculty positions available for them > when they complete their degrees,
so they will be able to go straight > to those faculty positions and to be able
to use the infrastructure > that we have been building in the meantime to
continue their genomic > research, to apply for additional grants and to nurture
their own > students one day. (R-02)
Generating a pool of qualified African researchers must be accompanied by a
parallel interest in maintaining them in research institutions in Africa.
Considering that most research institutions in Africa are government
establishments, African governments will have to play a key role in retaining
emerging researchers in Africa. And whilst African governments may find it hard
to direct funds for research activities, they can at least provide research jobs
and research support for emerging African researchers.
Equally, whilst the establishment of well-curated biobanks in Africa is one of
the greatest benefits of these genomics research and biobanking consortia,
maintaining them will be a serious challenge for host institutions. To solve the
problem of sustainability a few interviewees again suggested that African
governments and the private sector in Africa invest in health research
> What I know will be essential will be for African governments to > invest more
in biomedical research and development at all levels and > for African
governments and the private sector to improve the health > research systems that
are currently in place in many African > countries. Those efforts will benefit
not just genomic research but > all aspects of health research. (R-01)
The quote by the researcher above further highlights the importance of African
governments to support health research in Africa as this will enable African
researchers to work on local health needs rather than having to rely heavily on
external support which could come with the specific research interest and
priorities of the funders or collaborators in HICs.
# Discussion and conclusion
International health research has great potential in fostering health research
in Africa. It is also a platform for both LMIC and HIC researchers to share
expertise and resources for the purpose of advancing scientific discovery and to
access funding for health research. However, power imbalances between
collaborators in Africa and HICs may hinder successful research partnerships and
has been known to lead to the exploitation of African researchers and research
participants. In this study, we document the results of a qualitative study that
aimed at exploring African researchers’ perceptions and expectations of the
risks and benefits of international health research collaborations, with a
particular focus on genomics research and biobanking in Africa. All the
interviewees acknowledged that the opportunity to collaborate and access funding
is a benefit for African researchers. However, they expressed fears that may be
exploited within these collaborations. In their opinion, fears of exploitation
may be minimized through setting up equitable research collaborations and
building capacity for genomics research and biobanking in Africa. In this paper,
we document that despite persisting fears of exploitation of African scientists,
all interviewees were of the opinion that the international genomics research
and Biobanking consortia in Africa have provided African researchers with a
platform to build their capacity to conduct cutting edge genomics research.
There is a growing acceptance of the importance of research capacity building
and African leadership and ownership of health research in Africa as a means of
achieving equitable international health research partnerships. The assumption
is that equitable research partnerships would ultimately stand a better chance
of building trust between research partners and in fostering health research on
African health problems, thereby reducing global health inequities.
Extrapolating this trend to genomics research and biobanking in Africa, it is
equally important to ensure that international collaborations in African
genomics research are fair and equitable–and the rhetoric surrounding these
initiatives reflects this trend. Some genomics research and biobanking
initiatives in Africa such as the H3Africa Consortium have gone some way in
defining some of the aspects of fair and equitable international collaboration
in African genomics research and to speculate that H3Africa may be setting a
gold standard for how collaborative international health research in Africa
should be done to benefit African populations. In this study, we go beyond this
documented promise to explore African researchers’ perceptions and expectations
of benefits and risks of Africa’s participation in collaborative genomics
research and biobanking in Africa. For our interviewees, a key hallmark of
equitable collaboration is equip African researchers to make equivalent
intellectual contributions to the design and conduct of genomics research
through a rigorous process of research capacity building. This may take several
forms including training of Masters and PhD students, skills development for
postdoctoral scientists, mentorship, peer support through continental networks
and infrastructural support.
Different forms of research capacity building have been described in the
literature and one key component that emerged from the interviews, that goes
beyond individual training, is the creation of postgraduate degree programs in
genomics in African universities. This approach could foster sustainability of
genomics initiatives in Africa. Such degree programs could be designed to meet
current needs of African genomics research and need to include training in
bioinformatics, genomics medicine, genetic counselling, Bioethics and the social
sciences. Equally important in the sustainability of capacity building efforts
is the creation of networks that could foster interdisciplinary research in
Africa. The H3Africa bioinformatics network (H3ABionet) is an example of a
network that provides peer support for bioinformatics research in Africa with
nodes in more than 30 different African countries. Such an approach will have to
be adopted especially in emerging disciplinary areas that provide support for
genomics research and for which there is limited capacity. Examples include
genetic counselling, bioethics, sociology and anthropology. Also, because
research is a complex activity and process that requires the interplay of
individuals, organizations, national and international research systems.
Genomics research and biobanking initiatives in Africa will also need to expand
their capacity building efforts to include the establishment of centers of
excellence for genomics research in Africa, supporting research administration
in Africa, strengthening capacity for research-to-policy and promoting public
education in genomics in Africa. This will be particularly crucial if these
initiatives are to achieve their aim of being an exemplary model for
collaborative health research in Africa.
A challenge to the H3Africa experience is that it is based on a set of shared,
yet often unarticulated, values and principles that seek to promote equity and
fairness. Whilst these principles are incorporated into the design of projects
and policies of the H3Africa Consortium, they are not necessarily visible to or
shared by other initiatives that support genomics research in Africa. A question
is how the H3Africa infrastructure could be effective in influencing the design
of such initiatives, to ensure that the principles of fairness and equity are
also woven into such future endeavors. It is possible that this could be
achieved through for instance the African Academy for Sciences (AAS) and its
affiliate organization the Alliance for the Acceleration of Excellence in
Science in Africa, but only if there is active advocacy on behalf of the African
genomics community, including H3Africa researchers, for this to be the case.
# Supporting information
We are grateful to all researchers who participated as interviewees in this
study. The empirical work for this project was supported by an RHDGen MSc
Studentship to NSM. RHDGen is an H3Africa collaborative center led by Professor
Bongani M Mayosi and funded by the Wellcome Trust (WT099313MA). We thank the two
reviewers of the article for their comments and suggestions which helped
improved the first version of the manuscript.
[^1]: The authors have declared that no competing interests exist. |
# Introduction
In the past two decades, *Staphylococcus aureus* has emerged one of the most
important pathogens causing infections with indwelling medical devices, such as
prosthetic joints, prosthetic heart valves, intravascular catheters, and
cerebrospinal fluid shunts, which creates an increasing health care problem. For
example, prosthetic joint infections occur at a frequency of 1.5–2.5% in primary
total hip or total knee arthroplasty with a mortality rate of up to 2.5%. By far
the most frequently isolated species from these infections are *Staphylococcus*
species, i.e. *S. aureus* (22–39%) and coagulase-negative staphylococci
(15–37.5%).
The pathogenesis of device-associated infections with staphylococci is mainly
characterized by the pathogens ability to colonize the surfaces of the implanted
medical device by the formation of a three-dimensional structure of
microorganisms embedded in a thick extracellular matrix composed of
polysaccharides, proteins, extracellular DNA, and host factors, known as
biofilm. Microorganisms within a biofilm are protected against antimicrobial
chemotherapy as well as against the immune system of the host.
Biofilm formation occurs in a two-step process. The first step involves the
adherence of the bacteria to artificial surfaces that can occur either directly
or via host factors acting as bridging molecules, such as the extracellular
matrix and plasma proteins fibrinogen (Fg) and fibronectin (Fn) or platelets. In
the second step, the bacteria proliferate and accumulate into a biofilm
requiring intercellular adhesion. Direct *S. aureus* adherence to the unmodified
artificial surface may be mediated by the major autolysin Atl, which is highly
homologous to the *S. epidermidis* autolysin/adhesin AtlE shown to be involved
in the attachment to polymer surfaces. *S. aureus* host-factor binding proteins
that typically belong to the family of <u>m</u>icrobial <u>s</u>urface
<u>c</u>omponents <u>r</u>ecognizing <u>a</u>dhesive <u>m</u>atrix
<u>m</u>olecules (MSCRAMM) are involved in binding to host factor-coated foreign
material, among them Fn-binding proteins (FnBpA, FnBpB, Ebh), Fg-binding
proteins (ClfA, ClfB), collagen-binding protein (Cna), bone-sialoprotein-binding
protein (Bbp), and von Willebrand factor (vWf)-binding protein A (Spa).
Staphylococcal biofilm accumulation is mediated by polysaccharide as well as
protein factors. The intercellular polysaccharide adhesin (PIA), a
β-1,6-N-acetylglucosaminoglycan, is produced by the gene products encoded by the
*icaADBC* operon that was first identified in *S. epidermidis* and is also
present in *S. aureus*. Surface proteins conferring biofilm accumulation include
the accumulation-associated protein (Aap) from *S. epidermidis*, and the
homologous *S. aureus* surface protein G (SasG). In *S. aureus*, another
protein, the biofilm-associated protein (Bap), is involved in biofilm
accumulation. However, so far the *bap* gene has not been found in any *S.
aureus* isolate of human origin, but has only been identified within bovine
mastitis isolates.
A recent study demonstrated that all 18 *S. aureus* isolates from prosthetic
joint infections carry the *icaADBC* operon, produce PIA, and are biofilm-
positive. Surprisingly, the biofilms of all 18 *S. aureus* isolates could be
almost completely eradicated by the treatment with dispersin B (DspB), an enzyme
with specific β-1,6-hexosaminidase activity as well as by the treatment with
trypsin suggesting that both, proteinaceous adhesins and PIA contribute to
biofilm formation in these *S. aureus* isolates. This was in contrast to *S.
epidermidis* isolates from prosthetic joint infections. Only 62% of the 52 *S.
epidermidis* isolates carry the *icaADBC* operon and *S. epidermidis* biofilms
produced by *icaADBC*-positive strains were disintegrated by DspB, but not by
proteases. Furthermore, biofilms produced by *icaADBC*-negative strains were
disintegrated by proteases, but not by DspB. Thus, different mechanisms seem to
be involved in biofilm formation in clinical *S. aureus-* and *S.
epidermidis-*associated prosthetic joint infection isolates. More specifically,
in *S. aureus*-associated prosthetic joint infections, polysaccharide and
protein factors seem to act synergistically in biofilm formation. Only 33% of
the analyzed *S. aureus* strains carry the *sasG* gene and none of them carry
the *bap* gene, indicating the existence of further, not yet identified surface
proteins involved in biofilm accumulation of *S. aureus*.
Upon a search for LPXTG-motif containing surface-anchored proteins encoded by
the *S. aureus* Col genome (<http://www.tigr.org>), we chose to study the so far
uncharacterized SasC. We heterologously expressed *sasC* from *S. aureus* Col
and from the clinical *S. aureus* isolate 4074 in *Staphylococcus carnosus*
under the control of a xylose-inducible promoter. *S. carnosus* expressing
*sasC* formed huge cell aggregates indicative of intercellular adhesion, which
were disintegrated by protease treatment. Upon plasmid-encoded expression of
*sasC*, *S. carnosus* as well as *S. aureus* not only formed huge cell
aggregates, but also formed a much more pronounced biofilm in microtiter plates
as well as in glass tubes than the respective wild-type strains. The domain
conferring cell aggregation and biofilm formation was localized to the
N-terminal domain of SasC. In conclusion, we identified SasC as a novel *S.
aureus* factor involved in intercellular adhesion and biofilm accumulation.
# Results
## Identification and cloning of the *sasC* gene of *S. aureus*
As a candidate gene conferring biofilm formation in *S. aureus*, we amplified a
DNA fragment containing the *sasC* gene including the ribosome binding site by
polymerase chain reaction (PCR) from *S. aureus* 4074 and *S. aureus* Col
genomic DNA using the primers CHsasCfor and CHsasCrev yielding 6577 bp DNA
fragments. The DNA fragments were cloned into the *Bam*HI and *Sma*I sites of
the vector pCX19 in *S. carnosus*, creating plasmids pSasC4074 or pSasCCol.
## Nucleotide sequence of *sasC* and amino acid sequence analysis of the deduced protein
The nucleotide sequence of the cloned *sasC* gene from *S. aureus* 4074 was
determined on both strands. *sasC* consists of 6558 nucleotides and encodes a
deduced protein of 2186 amino acids (aa) with a predicted molecular mass of
237.9 kDa. The ATG start codon is preceded by a putative ribosome binding site
at a distance of 8 bp. Putative −10 (TATATT, nucleotides −61 to −56) and −35
(TAAACA, nucleotides −80 to −75) promotor sequences were deduced from homologous
DNA sequences from strain *S. aureus* Col. A putative ρ-independent terminator
consisting of two stem-loops is located downstream of the TAA stop codon. The
deduced SasC sequence contains a putative signal peptide in the first 37 aa that
contains an YSIRK motif, which seems to play a role in signal peptide
processing. The predicted *sasC* gene product is composed of 25.1% hydrophobic,
12.1% basic, and 13.2% acidic aa. The theoretical pI value of SasC is 5.08. The
deduced aa sequence of SasC of strain 4074 shares 97% identical aa with
homologous proteins from strains *S. aureus* MW2 and MSSA476, 96% identical aa
with strains USA300, COL, Newman, NCTC8325 (accession number: Q2FXH4), and N315
and 89% identical aa with strain MRSA252. Besides, SasC shares 31% identical aa
with Mrp and FmtB proteins from strain Col that have been implicated in
methicillin-resistance.
Sequence comparison of the deduced SasC sequence with known protein sequences in
databases revealed a domain structure of SasC. The central portion of SasC
contains a domain, which is similar to the motif <u>f</u>ound <u>i</u>n
<u>v</u>arious <u>ar</u>chitectures (FIVAR; 54 aa, starting at N-590 and ending
at D-643). The FIVAR domain is followed by 17 direct repeated sequences of 72 aa
each separated by a stretch of 5 aa, which are homologous to a <u>d</u>omain of
<u>u</u>nknown <u>f</u>unction (DUF1542) also found in other cell surface
proteins. The first DUF1542 repeat starts at Q-671 and the last ends at I-1974.
The DUF1542 repeats share between 18 and 47% identical aa. The FIVAR motif and
the DUF1542 domain are also present within Mrp and FmtB (see above) as well as
within the cell surface protein Ebh from *S. aureus* and the homologous Embp
from *S. epidermidis*.
## SasC mediates strong cell aggregation in *S. carnosus*
After overnight growth in tryptic soy (TS) broth supplemented with 1% xylose,
*S. carnosus* expressing *sasC* formed huge cell clusters that were visible
macroscopically and microscopically. In contrast, the strains did not form cell
clusters without induction by xylose. The cell clusters were dissolved upon
treatment with trypsin or proteinase K. Disruption of cell clusters by trypsin
was concentration-dependent.
## Characterization of *S. carnosus* and *S. aureus* expressing *sasC* or *sasC*-subclones
In order to dissect the functional domains within SasC, we constructed subclones
of *sasC* expressing either the N-terminal domain including the FIVAR motif
(subclone 1) or 8 of the 17 DUF1542 repeats (subclone 2) in *S. carnosus*
yielding *S. carnosus* (pSasCsub1) or *S. carnosus* (pSasCsub2), respectively.
An inverse PCR with the plasmid pSasC4074 as a template was carried out for the
construction of subclone 1 and subclone 2 by using the primers
CHsasCSub1rev/CHsasCSub1for and CHsasCSub2rev/CHsasCSub2for, respectively.
The resulting PCR fragments were ligated and subsequently, *S. carnosus* was
transformed with the ligation mixtures. Correct clones were verified by PCR and
DNA sequencing. However, DNA sequence analysis of subclone 1 revealed a 33-bp
deletion in the C-terminal region so that aa T-654 is fused to N-1987 rather
than to D-1976. To functionally characterize *sasC* also in the *S. aureus*
background, we transformed *S. aureus* 4074 and *S. aureus* SH1000, which is a
8325-4 derivative reconstituted for its *rsbU* mutation, with the plasmids
pSasC4074, pSasCsub1, or pSasCsub2 leading to overexpression of *sasC* and its
subfragments in these strains.
To verify the production of the whole SasC or the truncated SasC proteins, cell
lysates of the strains were prepared and analyzed by SDS-PAGE. *S. carnosus* and
the *S. aureus* strains expressing *sasC* revealed an additional protein band
corresponding to the size of SasC of 238 kDa. The strains expressing
subfragments 1 or 2 revealed additional protein bands at 93 or 96 kDa,
respectively. The strains expressing *sasC* or the subfragments 1 or 2 were
further characterized.
## (i) Cell aggregation
In contrast to the wild-type strains, *S. carnosus* (pSasC4074), *S. aureus*
4074 (pSasC4074) and *S. aureus* SH1000 (pSasC4074) as well as *S. carnosus*
(pSasCsub1) and *S. aureus* SH1000 (pSasCsub1) formed huge cell aggregates that
were visible macroscopically in cultures grown overnight on TS agar supplemented
with 1% xylose and resuspended in PBS on glass slides. The cell aggregates
formed by *S. aureus* 4074 (pSasCsub1) were somewhat smaller in size (comparable
to those mediated by the *icaADBC* operon encoding the production of the
polysaccharide intercellular adhesin, PIA –13) corresponding to a lower
expression level of the fusion protein. In contrast, the parent strains or
strains expressing the subfragment 2 did not form visible cell aggregates. Thus,
*sasC* and its subfragment 1 confer cell aggregation.
## (ii) Biofilm formation on polystyrene and glass
Biofilm formation on polystyrene was determined in the quantitative biofilm
assay. Whilst *S. carnosus* (pCX19) revealed a biofilm-negative phenotype
(*A*490 value: 0.11), *S. carnosus* (pSasC4074) and *S. carnosus* (pSasCSub1)
showed an enhanced biofilm formation corresponding to an *A*490 value of 0.43
and 0.4. In contrast, biofilm formation of strain *S. carnosus* (pSasCsub2) is
comparable to that of the wild type (*A*490 value: 0.05).
In *S. aureus*, the expression of *sasC* or the subfragment 1 also led to a
markedly increased biofilm formation as shown for strain SH1000 \[*A*490 value
of 1.5 or 1.4 for *S. aureus* SH1000 (pSasC4074) or *S. aureus* SH1000
(pSasCsub1), respectively, versus *A*490 value of 0.7 for *S. aureus* SH1000\].
As observed with *S. carnosus*, the biofilm formation of *S. aureus* SH1000
(pSasCsub2) is comparable with that of its wild type (*A*490 value: 0.8). The
results for *S. aureus* 4074 are similar (not shown).
The stronger capacity for biofilm formation mediated by *sasC* or the
subfragment 1 also could be observed on a glass surface with *S. carnosus* as
well as with *S. aureus* SH1000 and *S. aureus* 4074 (not shown) and was
comparable with that of *S. carnosus* (pCN27) producing PIA. Thus, *sasC* and
its subfragment 1 mediate biofilm formation on polystyrene and on glass.
## (iii) Initial attachment to polystyrene and to glass
Bacterial biofilm formation results from initial attachment of bacterial cells
to a surface and subsequent accumulation into multilayered cell clusters, which
requires intercellular adhesion visible as cell aggregation. To determine,
whether *sasC* not only mediates cell aggregation, but also initial attachment,
we analyzed the capacity of the respective strains for attachment to a
polystyrene or a glass surface. Initial attachment of *S. carnosus* (pCX19) to
polystyrene was very low indicated by a low number of attached bacteria. *S.
carnosus* strains expressing *sasC* or its subfragments showed higher initial
attachment with *S. carnosus* (pSasCsub1) yielding the highest numbers of
attached bacteria. Essentially the same was observed with *S. aureus* SH1000
strains albeit at a higher level of attachment and with *S. aureus* SH1000
(pSasCsub2) showing numbers of attached bacteria that were comparable with the
wild type.
Initial attachment of the strains to a glass surface generally was higher and
similar with *S. carnosus* and *S. aureus* SH1000 wild-type strains and the same
strains expressing *sasC* or the subfragment 2. In contrast, both strains
expressing subfragment 1 showed a slightly lower number of attached bacterial
cells. Thus, expression of *sasC* slightly increased attachment to a polystyrene
surface, but did not increase attachment to glass.
## (iv) Binding to Fg, vWF, thrombospondin-1 (TSP-1), and platelets
The ability of *S. aureus* to bind to extracellular matrix and plasma proteins
and to host cells determines its capacity for tissue colonization. The potential
of *sasC* to mediate binding to the extracellular matrix and plasma proteins Fg,
TSP-1, and vWf as well as to platelets was analyzed by flow cytometry. While *S.
aureus* 4074 bound to Fg, TSP-1, and vWf as well as to activated platelets in a
dose-dependent fashion, *S. carnosus, S. carnosus* (pSasC4074), and *S.
carnosus* (pSasCsub2) did not. Thus, *sasC* does not mediate binding to these
extracellular matrix proteins or to activated platelets.
## Biofilm formation of a *sasC* transposon (Tn)*917* insertion mutant (SMH2035)
To further support the role for SasC in biofilm formation, we analyzed the
capacity of a *sasC* Tn*917* insertion mutant (SMH2035) for biofilm formation in
comparison to its wild-type strain *S. aureus* SH1000. Under all conditions
tested, the *sasC* mutant strain showed reduced biofilm formation in microtiter
plates, i.e. the addition of 1% xylose \[*A*490 value of 0,4 for SMH2035 versus
*A*490 value of 0,7 for SH1000\], 0.25% glucose \[*A*490 value of 1,6 for
SMH2035 versus *A*490 value of 1,8 for SH1000\] or no additional carbohydrate
source in the biofilm assay \[*A*490 value of 0,5 for SMH2035 versus *A*490
value of 1,0 for SH1000\]. The biofilm-forming capacity of both strains was most
pronounced upon addition of 0.25% glucose, which is known to induce the
production of PIA in *S. epidermidis* and probably leads to enhanced PIA
production also in *S. aureus*, which might partially obscure the function of
SasC. The presence of 1% xylose seems to slightly reduce biofilm formation. So
far, the reason for this is unknown.
## Expression and purification of the 6 x Histidine (His)-DUF1542 fusion protein in *Escherichia coli*
For expression of the *sasC* portion encoding 8 of the 17 DUF1542-repeats in *E.
coli*, the PCR-amplified fragments were cloned into the expression vector
pQE30Xa. One representative clone expressing the DUF1542 repeats (6 x His-
DUF1542) contained the plasmid pHis-DUF1542. Subsequently, the 6 x His-DUF1542
fusion protein was purified from *E. coli* (pHis-DUF1542) via its His-tag using
Ni-NTA affinity chromatography under native conditions. SDS-PAGE of the
affinity-purified fusion proteins revealed an approximately 70-kDa protein for
the 6 x His-DUF1542 fusion protein. A protein with the same size was present in
cell lysates of an induced culture of *E. coli* (pHis-DUF1542) and absent from
cell lysates of an induced culture of *E. coli* (pQE30) (not shown) and of a
non-induced culture of *E. coli* (pHis-DUF1542).
## Surface location of SasC
To detect the surface location of SasC, the antiserum against the 6 x His-
DUF1542 fusion protein that was raised in rabbits was used in immunofluorescence
microscopy. The anti-His-DUF1542 antiserum strongly reacted with cells of *S.
carnosus* (pSasC4074) and *S. aureus* 4074 (pSasC4074) (not shown) indicating
the cell surface location of SasC. With *S. carnosus* (pSasC4074), no
immunofluorescence was detected with the preimmune serum (not shown) and with
*S. carnosus* (pCX19), no immunofluorescence was detected with the anti-His-
DUF1542 antiserum. However, there was some immunofluorescence detectable with
the preimmune serum and strain *S. aureus* 4074 (not shown), which may be due to
the IgG-binding and surface-associated proteins A and Eap.
## Characterization of *sasC* expression in *S. aureus* strains
To analyze the production of SasC in *S. aureus*, we performed Western blot
analysis using the anti-His-DUF1542 antiserum. In cultures of *S. aureus* 4074,
Col, and SH1000, a faint band corresponding to SasC was detected in lysostaphin
lysates of cultures after 10, 16, 24, and 48 h of growth (not shown). A strong
production of SasC was observed in control strains *S. aureus* 4074 (pSasC4074)
and *S. carnosus* (pSasC4074), while no protein isolated from *S. carnosus*
(pCX19) reacted with the anti-His-DUF1542 antiserum (not shown). SasC did not
react with the preimmune serum (not shown).
## Prevalence of *sasC*
To determine the prevalence of the *sasC* gene among clinical *S. aureus*
isolates, we performed PCR analysis using primers CHsasC1for and CHsasC1rev. An
approximate 500 bp fragment encoding a portion of the N-terminal SasC domain was
found to be present in 97% (66/68) of the clinical *S. aureus* strains that were
tested (not shown). This indicates a very high prevalence of the *sasC* gene
among clinical *S. aureus* isolates.
# Discussion
The most frequently isolated bacteria from implant-associated infections are *S.
aureus* and coagulase-negative staphylococci causing significant morbidity and
mortality. The pathogenicity of these infections is characterized by the
pathogens pronounced ability to form biofilms. To date, several *S. aureus*
genes have been implicated in biofilm formation, among them *atl*, *dltA*, and
the *icaADBC* gene cluster. *S. aureus* surface proteins reported to be involved
in biofilm formation include SasG, the Fn- and Fg-binding proteins FnBPA and
FnBPB , and the biofilm-associated protein Bap. SasG is homologous to the
accumulation-associated protein Aap, which mediates biofilm accumulation in *S.
epidermidis*, and the plasmin-sensitive surface protein Pls, which so far has
not been implicated in biofilm formation. Furthermore, extracellular genomic DNA
(eDNA) has been established as another important component of *S. aureus*
biofilms. eDNA is released from the bacteria by cell lysis and may implicate an
additional role for the major *S. aureus* autolysin Atl in biofilm development
besides its function in initial attachment.
However, a recent study indicated the existence of further, yet unidentified
surface proteins contributing to *S. aureus* biofilm formation. As a potential
candidate, we identified SasC in the *S. aureus* genome and expressed its gene
heterologously in *S. carnosus* as well as in the *S. aureus* strains SH1000 and
4074 under a xylose-inducible promotor. All strains expressing *sasC* showed a
high-level production of a protein with a molecular size of approximately 240
kDa corresponding to SasC. Expression of *sasC* led to strong cell cluster
formation, intercellular adhesion, and biofilm formation. Furthermore, a *sasC*
Tn*917* insertion mutant (SMH2035) showed a reduced capability for biofilm
formation in comparison to its wild type.
SasG and Aap promote biofilm formation via their B-repeats. Each B-repeat also
known as G5 domain consists of 128 aa and is present 7 and 5 times in SasG and
Aap, respectively. Recently, the G5 domains were found to be zinc-dependent
adhesion modules and a “zinc zipper” mechanism was suggested for G5 domain-based
intercellular adhesion in SasG- or Aap-mediated biofilm accumulation. SasC also
contains a repeat region with 17 repeats of 72 aa being similar to the DUF1542
domain (see). The SasC repeats do not share sequence similarities with the
B-repeats of SasG and Aap and are not involved in biofilm formation. Instead in
SasC, the domain conferring cell aggregation and biofilm formation could be
localized to the N-terminal domain by subcloning experiments.
Furthermore, SasG and Aap must undergo proteolytic cleavage to become active,
while SasC-mediated biofilm formation does not depend on proteolytic cleavage:
the formation of SasC-mediated biofilms was unchanged in the presence of the
protease inhibitor α2-macroglobulin (data not shown) that blocked SasG and Aap-
mediated biofilm formation,. Moreover, *S. carnosus* expressing *aap* only
formed a detectable biofilm after treatment with 2 µg/ml of trypsin leading to
the production of a truncated version of Aap or when an N-terminally truncated
version of *aap* was cloned in *S. carnosus*. In contrast, no truncation of SasC
was involved in SasC-mediated biofilm formation. Thus, the mechanism of SasC-
mediated biofilm accumulation clearly differs from that mediated by SasG/Aap.
Because of the lack of sequence similarities, the SasC-mediated mechanism
probably is also distinct from that mediated by Bap.
Recently, a role for FnBPA and FnBPB in biofilm accumulation, which is
independent of their Fn- and Fg-binding activities was reported. Like with SasC,
the domain conferring biofilm accumulation is located within the N-terminal
domain (A domain), which also includes the domain for Fg-binding. However,
biofilm formation seems to be independent of Fg-binding, because a single aa
exchange within that region abolished Fg-binding, but did not influence the
capability for biofilm formation. Because the FnBPs and SasC do not share
significant sequence similarities within their N-terminal domains, the mechanism
of biofilm accumulation between the two also seems to differ.
The N-terminal SasC domain shows significant homology to the N-terminal domains
of Mrp and FmtB, both of which have been reported to be involved in methicillin
resistance. However, the mechanism of Mrp and FmtB conferring methicillin
resistance is not completely clear yet and may be indirect. A function of these
proteins in biofilm formation has not been proposed so far.
The mechanism of SasC-mediated biofilm accumulation is not known yet and may
involve either protein-protein interactions conferred by the N-terminal SasC
domain or protein-carbohydrate interactions conferred by the FIVAR motif. The
FIVAR motif is located at the C-terminus of the N-terminal domain and has been
proposed to have a sugar-binding function. Further analyses are necessary to
elucidate these possibilities.
SasC not only mediates cell cluster formation and intercellular adhesion, but
also slightly increases the attachment of the cells to polystyrene. Initial
attachment to a polystyrene surface depends on cell surface hydrophobicity. With
25.1% hydrophobic aa, the percentage of hydrophobic aa of SasC is comparable to
that of AtlE (26.4%) previously found to mediate initial attachment to
polystyrene in *S. epidermidis*. Remarkably, the higher level of initial
attachment to polystyrene observed with *S. carnosus* and *S. aureus* producing
subclone 1 correlated with a higher percentage of hydrophobic aa (28.1%) and a
lower percentage of charged aa (23.7%: 12.3% basic and 11.4% acidic aa) of
subclone 1 in comparison to subclone 2, which contains 26.6% hydrophobic aa and
30.1% charged aa (14.7% basic and 15.4% acidic aa).
Although the *S. aureus* wild-type strains Col, SH1000, and 4074 all harbour the
*sasC* gene, we were not able to detect SasC in lysostaphin lysates by SDS-PAGE,
but only by Western immunoblot analysis indicating very low *sasC* expression
*in vitro*. However, the identification of SasC as an *in vivo*-expressed
antigen during infection delineates its potential clinical significance. Thus,
*sasC* expression may be induced *in vivo*. Several genes are involved in the
regulation of *S. aureus* biofilm formation, such as *agr*, *sarA*, *sigB*,
*rbf*, *tcaR* , *arlRS*, and *alsSD*. Further experiments are necessary to
characterize the expression of *sasC*.
Homologous *sasC* genes were found in the eight sequenced *S. aureus* genomes
analyzed and the sequence similarities of the SasC gene products ranged from 89%
to 97%. Consistently, we found a very high prevalence of the *sasC* gene among
clinical *S. aureus* strains delineating the potential importance of SasC in
colonization and infection. In conclusion, we identified and characterized a
novel *S. aureus* surface protein, SasC, involved in cell aggregation and
biofilm formation, which may play an important role in colonization during
infection with this important pathogen.
# Materials and Methods
## Ethics Statement
This study was conducted according to the principles expressed in the
Declaration of Helsinki. The study was approved by the local ethics committee
(Ethikkommission der Aertzekammer Westfalen-Lippe und der Medizinische Fakultaet
der WWU Muenster) (reference number: Sitzung 19.05.1999). All volunteers
provided written informed consent for the collection of samples and subsequent
analysis.
## Bacterial strains, plasmids, and media
The *sasC* gene was cloned from the clinical strain *S. aureus* 4074 and from
*S. aureus* Col. *S. carnosus* TM300, *S. aureus* 4074, and *S. aureus* SH1000
were used as cloning hosts. The *sasC* Tn*917* insertion mutant SMH2035 was
isolated by sequencing Tn insertion junctions from a Tn*917* mutant library of
*S. aureus* SH1000, using direct sequencing from isolated genomic DNA. One of
the mutants gave a junction sequence, which mapped to an insertion point at
1812698 in the *S. aureus* N315 genome. This corresponded to inactivation of
SasC at aa 934. *E. coli* TG1 was used to construct the plasmid for the
production of the 6 x His-DUF1542 fusion protein and its purification.
For the expression of *sasC* in staphylococci, the vector pCX19, a derivative of
the xylose-inducible expression vector pCX15 and for the production and
purification of the 6 x His-DUF1542 fusion protein, the Qia*express* vector
pQE30Xa (Qiagen) was used.
To determine the prevalence of *sasC*, a 489 bp internal *sasC* fragment was
amplified from genomic DNA of 68 clinical *S. aureus* isolates obtained from
patients at the University Hospital of Münster (Münster, Germany).
The staphylococci were routinely cultivated in TS broth (TSB). *E. coli* strains
were grown in Luria-Bertani (LB) medium. TS and LB agar contained 1.4% agar. TSB
and TS agar were supplemented with 1% xylose to induce *sasC* expression.
Selection for resistance to antibiotics in *E. coli* was performed with 100
µg/ml ampicillin and in staphylococci with 10 µg/ml chloramphenicol or 5 µg/ml
erythromycin, when appropriate. Wild-type strains not harboring a plasmid were
grown in the presence of 0.07% ethanol in the respective assays, when compared
to the strains harboring a plasmid to rule out an effect of the ethanol. Because
the *sasC* Tn*917* insertion mutant SMH2035 is stable, no antibiotic or ethanol
was included in the assays comparing the SH1000 strain and its *sasC* mutant.
Bovine serum albumin (BSA) and α-thrombin (bovine) were purchased from Sigma.
Human Fg was purchased from Enzyme Research Labs. The monoclonal antibody
against human CD42a (GP IX) (conjugated with phycoerythrin \[PE\]) was delivered
by Exalpha via NatuTec. Syto 13 for labeling of staphylococcal cells was
purchased from Molecular Probes via Mobitec.
## DNA manipulations, transformation, PCR, DNA sequencing, websites, and Pfam accession numbers
DNA manipulations and transformation of *E. coli* were performed according to
standard procedures. *S. carnosus* and *S. aureus* strains were transformed with
plasmid DNA by protoplast transformation. Plasmid DNA was isolated using the
Qiagen Plasmid Kit and chromosomal DNA was isolated using the QIAamp DNA Blood
Mini Kit according to the instructions of the manufacturer (Qiagen). PCR was
carried out with the PCR Supermix High Fidelity (Invitrogen) or with the Expand
Long Template PCR System (Roche, Mannheim, Germany) in accordance with the
protocol of the supplier. The primers were synthesized by MWG-Biotech
(Ebersberg, Germany).
The DNA sequence of both strands of the *sasC* gene of the clinical isolate *S.
aureus* 4074 was determined by MWG-Biotech using a LI-COR DNA sequencer.
The DNA and deduced protein sequences were analyzed using the program “JustBio”
at <http://www.justbio.com>. The protein sequences were compared with those of
known proteins using the programs BLASTP and FASTA. The alignments were done
using the program ClustalW at the European Bioinformatics Institute (EBI,
Cambridge, UK). The Pfam accession numbers are: PF04650 for the YSIRK_signal;
PF07554 for the FIVAR motif; PF07564 for the DUF1542 domain; PF00746 for the
Gram_pos_anchor (LPXTG_anchor) available at: <http://pfam.sanger.ac.uk/>. The
signal peptide of SasC was predicted by using the program “SignalP” at
<http://www.cbs.dtu.dk/services/SignalP/>).
## Biofilm formation assay
For quantification of the biofilm-forming capacity, a test for biofilm
production was performed essentially as described previously. Briefly, strains
were grown in TSB supplemented with 1% xylose for 24 h at 37°C in 96-well
polystyrene microtiter plates (cell star; Greiner, Frickenhausen, Germany), the
wells were washed with phosphate-buffered saline (PBS) and adherent biofilms
were stained with 0.1% safranin (Serva). In some experiments, instead of 1%
xylose, 0.25% glucose or no additional carbohydrate source was added. Absorbance
was measured with a Micro-ELISA-Autoreader at 492 nm. Strains were tested at
least in triplicate. Determination of biofilm formation on a glass surface was
carried out essentially in the same way, except that glass tubes were used
instead of microtiter plates and 5 ml of TSB were inoculated instead of 200 µl.
## Initial adherence to a polystyrene or a glass surface
Initial cell attachment was tested as described previously. Briefly, diluted
cell suspensions of bacteria in PBS were incubated for 30 min in polystyrene
Petri dishes (Sarstedt) at 37°C or on glass slides and after a washing
procedure, attached bacteria were evaluated by phase-contrast microscopy.
## Construction and purification of the 6 x His-DUF1542 fusion protein and anti-His-DUF1542 antiserum
For the construction of the His-tagged fusion protein, the primers CHsasCDUFfor
and CHsasCDUFrev were used to amplify 8 of the DUF1542-repeats of *sasC* from
genomic DNA of *S. aureus* Col, introducing a *Bam*HI-site at the 3′ end. The
PCR-amplified fragment was cloned into the vector pQE30Xa, so that the gene
fragment is in frame with the His-codons. One representative clone expressing
the DUF1542-repeats was designated *E. coli* (pHis-DUF1542). The 6 x His-DUF1542
fusion protein was purified under native conditions via its His-tag using Ni-NTA
affinity chromatography (Ni-NTA Spin Kit; Qiagen) according to the protocol of
the suppliers. The yield of the 6 x His-DUF1542 fusion protein was in the range
of 150 µg per 10 ml culture volume as determined by the Coomassie Plus™ Protein
Assay (PIERCE, Rockford, IL, USA distributed by Perbio Science, Bonn, Germany).
The purified 6 x His-DUF1542 fusion protein was used to immunize rabbits by
Eurogentec (Belgium) according to their standard immunization program.
## Protein isolation, SDS-PAGE, and Western blot analysis
Staphylococcal surface proteins covalently linked to the peptidoglycan were
prepared from cultures that were grown overnight in TSB broth at 37°C by
lysostaphin treatment. For this, the staphylococcal cells were harvested by
centrifugation, washed and then, the cell pellet was resuspended in 20 ml Tris-
buffered saline (TBS) pH 7,4 per g bacteria. After adding 500 µg lysostaphin
(Ambi Products LLC, Lawrence, NY, USA; distributed by WAK Chemie, Steinbach,
Germany) and 100 µg DNAse (Sigma) per g bacteria, the suspension was incubated
at 37°C with shaking. Afterwards, the cell debris was removed by centrifugation
(45 min, 13.000 rpm, 4°C). Lysostaphin lysates were stored at −20°C. Crude cell
lysates of *E. coli* (pHis-DUF1542) were prepared from non-induced and induced
(addition of 1 mM Isopropyl-β-D-thiogalactoside \[IPTG\] and continued growth of
4 h) cultures by harvesting the cells, resuspending the cell pellet in sample
buffer, and heating the suspension for 5 min at 95°C. Additionally, as a
negative control, a crude cell lysate was prepared from an induced culture of
*E. coli* (pQE30). After centrifugation, 7 µl of the cell lysates from *E.
coli*, 9 µl of the staphylococcal cell lysates, or 2 µl of the purified protein
(containing 1.5 µg) were subjected to SDS-PAGE (10% separation gel and 4.5%
stacking gel). Proteins were stained with Coomassie brilliant blue R250 (0.1%).
For Western immunoblot analysis, staphylococcal surface proteins, crude cell
lysates, or purified proteins were prepared and separated by SDS-PAGE as
described above and transferred to a nitrocellulose membrane (Schleicher and
Schuell, Dassel, Germany). The membranes were then blocked in Tris-buffered
saline (TBS)/3% BSA (overnight) and washed three times with TBS/0.5% Tween 20
(TBST). Afterwards, the nitrocellulose membranes were incubated for 2 h with the
anti-His-DUF1542 antiserum diluted 1∶2000 in TBST/3% BSA. As a negative control,
incubation was performed in TBST/3% BSA with (1∶2000) or without preimmune
serum. The reaction of proteins with specific antibodies was detected by
incubation (1 h) with anti-rabbit immunoglobulin G (IgG)/alkaline phosphatase
(AP) conjugate (Dako GmbH, Hamburg, Germany) diluted 1∶5000 in TBST/3% BSA and
subsequent color reaction.
## Purification and fluorescence-labelling of Fg, thrombospondin-1 (TSP-1), or von Willebrand factor (vWF)
Labelling of Fg was performed as described previously. TSP-1 in an adhesive
conformation was purified from freshly isolated human platelets as described
before. For labelling TSP-1 with fluorescein-isothiocyanate (FITC) (Calbiochem;
La Jolla, CA, USA), FITC was added to TBS/2 mM CaCl<sub>2</sub> containing TSP-1
in a molar ratio of 600∶1 and incubated for 24 h at 4°C. Unbound label was
removed using a Sephadex G-25 PD-10 column equilibrated with TBS. The
concentration of FITC-labeled TSP-1 was determined by using the Pierce BCA
protein assay (Pierce Europe B.V., BA oud-Beijerland, The Netherlands) according
to the manufacturers instructions. vWf was purified and labeled according to
Hartleib *et al*.
## Flow cytometric analysis of Fg-FITC, TSP-1-FITC, or vWF-FITC binding to staphylococci
Measurement of the binding of Fg-FITC, TSP-1-FITC, or vWf-FITC to staphylococcal
cells was analyzed as described before for the binding of vWf-FITC to *S.
aureus*. Briefly, bacteria from an overnight culture (120,000 cells/µl) were
incubated with Fg-FITC, TSP-1-FITC, or vWf-FITC (final concentrations 0, 50, or
100 µg/ml) in TBS/2 mM CaCl<sub>2</sub> for 10 min at room temperature. After
washing and sonication, bacteria (5,000 cells/determination) were analyzed in a
flow cytometer (Becton Dickinson, FACSCalibur flow cytometer, Heidelberg,
Germany) using an excitation wave length of 488 nm at the FACSCalibur standard
configuration with a 530 nm bandpass filter. Data were obtained from
fluorescence channels in a logarithmic mode.
## Preparation of platelets
Blood was taken from healthy adult volunteers who had not taken any medication
affecting platelet function for at least 2 weeks before the study. Platelet-rich
plasma (PRP) was prepared from anticoagulated blood by centrifugation and the
platelets were gel-filtered on a Sephadex Cl-2B column. To inhibit fibrin
polymerisation, experiments were performed in the presence of the peptide GPRP
(1.25 mM) as described previously. The platelets were labeled by incubation with
a monoclonal anti-CD42a (GP IX) antibody conjugated with PE at saturated
concentrations for 30 min.
## Preparation of bacteria and flow cytometric measurement of *Staphylococcus*-platelet associate formation
*Staphylococcus*-platelet associate formation was measured essentially as
described before. Briefly, bacteria grown over night were washed in TBS, briefly
sonicated, and diluted with TBS/2 mM CaCl<sub>2</sub> to 250,000 bacteria/µl.
Bacteria were labeled with the fluorescent dye Syto 13 (emission similar to
FITC) at a concentration of 2 µM for 10 min, washed in TBS and briefly sonicated
again.
Platelets were activated with α-thrombin at the given concentrations for 4 min
and subsequently, labeled bacteria were added. Bacteria and platelets (10∶1)
were coincubated for 15 min at room temperature and conjugate formation was
measured immediately thereafter in a flow cytometer. Associates were identified
by double labelling with Syto 13 (FL-1, “FITC like” signal) and anti CD42a-PE
(FL-2, PE signal), and given as the rate of bacteria-positive platelets. Given
are the mean values of three independent experiments.
## Immunofluorescence microscopy
The detection of SasC by immunofluorescence microscopy was performed essentially
as described before. Briefly, cultures were grown aerobically in TSB for 16 h at
37°C. After washing, aliquots (30 µl) were applied to glass slides. The slides
were air-dried and the bacteria were fixed by heat. The fixed cells were
incubated with anti-His-DUF1542 antiserum or preimmune serum diluted 1∶500 in
PBS for 2 h at 37°C in a humid chamber, washed 4 times with PBS, and then
incubated with FITC-conjugated anti-rabbit F(ab′)<sub>2</sub> fragment diluted
1∶500 for 1 h at 37°C in a humid chamber. The slides were washed twice with PBS
and twice with double-distilled water. Then, the slides were dried, covered with
fluorescent mounting medium (Dako, Hamburg, Germany), and viewed with a
fluorescence microscope (Zeiss, Oberkochen, Germany).
## Nucleotide sequence accession number
The EMBL/GenBank/DDBJ accession number of the *sasC* DNA sequence of strain 4074
is FM202067.
We thank S. Weber and M. Rauth for excellent technical assistance. Sequence data
for identification and cloning the *sasC* gene was obtained from the Institute
for Genomic Research website at <http://www.tigr.org>.
[^1]: Conceived and designed the experiments: SF BK GP CH. Performed the
experiments: KS MJ SMH NH AB AS CH. Analyzed the data: KS MJ SMH AB AS BK
CH. Contributed reagents/materials/analysis tools: CN. Wrote the paper: CH.
[^2]: The authors have declared that no competing interests exist. |
# Introduction
The discovery of the process of transformation was key to the development of the
field of molecular genetics. The first evidence that genetic information could
be introduced into a cell came in 1928 when Frederick Griffith discovered that a
“transforming factor” could make a harmless strain of *Streptococcus pneumoniae*
virulent after being exposed to a heat-killed virulent strain, giving rise to
the term transformation. It was not until 1944 that Avery and colleagues used
transformation to prove that this factor was DNA. The era of eukaryotic
molecular genetics began over thirty years later when Hinnen and colleagues
employed transformation in brewer’s yeast to integrate a plasmid into the
*Saccharomyces cerevisiae* genome. Beggs subsequently demonstrated that *S*.
*cerevisiae* could maintain a plasmid carrying the 2μ origin of replication
without the need for integration. These discoveries established *S*.
*cerevisiae* as the premier eukaryotic model for molecular genetics.
Transformation protocols were subsequently developed for *Neurospora crassa* and
*Aspergillus nidulans*, and over the following decades, the development of
transformation protocols made many previously intractable species easier to
study.
*Cryptococcus neoformans* is one such species. Found worldwide in association
with bird guano, *C*. *neoformans* primarily causes disease in immunocompromised
individuals, disseminating *via* the lungs to cause life-threatening
meningoencephalitis; it is classified as an AIDS-defining illness. In developed
countries, the mortality rate is as high as 20%, but in developing countries
where there is limited availability of treatment, infection can result in close
to 100% mortality. While transformation of *C*. *neoformans via* electroporation
was achieved over 25 years ago, the technique was not widely adopted due to its
extremely low homologous integration efficiency and the instability of
transformants. It was not until the development of a biolistic protocol in 1993
that molecular genetic manipulation in this organism became routine. Although
biolistic technology is now widely employed, creating gene deletions in *C*.
*neoformans* can still be difficult due to the poor reproducibility of the
biolistic technique and low levels of integration *via* homologous
recombination; the majority of transformants are either ectopic integrants or
unstable.
Upon introduction of genetic material into a cell *via* transformation there
are, broadly, four possible fates. First, the exogenous DNA may be maintained
extrachromosomally in the form of a plasmid or minichromosome, provided this is
possible in the host species and the DNA sequence is appropriate. Second, the
foreign DNA may simply be degraded by the host machinery. Third, the exogenous
DNA may integrate into the genome in a targeted manner *via* homologous
recombination, and lastly, the exogenous DNA may integrate at a random site in
the genome. These two mechanisms of integration into the genome occur by very
different mechanisms. Homologous recombination occurs through crossing over
where DNA sequences are exchanged between two similar molecules of DNA; this
method is the basis for creating targeted gene deletions. While creating gene
deletions *via* homologous recombination occurs readily in species such as *S*.
*cerevisiae*, it occurs at a lower frequency in many organisms, including *C*.
*neoformans*. These species instead tend to predominantly employ non-homologous
end joining (NHEJ) and integrate the transformed DNA into a random location in
the genome.
In eukaryotes, NHEJ begins with the DNA-dependent protein kinase heterodimeric
regulatory factor Ku70-Ku80 forming a link between two broken DNA ends and
acting to structurally support, align and protect them from further degradation.
In humans, the Ku70-Ku80 heterodimer recruits the catalytic subunit DNA-
PK<sub>cs</sub>, which is thought to induce conformational changes that allow
end-processing enzymes to access the DNA ends; in *S*. *cerevisiae*, a complex
consisting of Mre11, Rad50 and Xrs2 is thought to perform this function instead.
In both species the process of NHEJ is completed when the DNA ligase IV-Xrcc4
complex ligates the broken ends of the DNA back together \[, –\].
Homologous recombination and NHEJ compete directly against one another to
incorporate foreign DNA into the genome. Unfortunately for molecular
geneticists, in most species NHEJ usually triumphs over homologous
recombination, frustrating researcher’s attempts to make precise genetic
modifications. One solution to this problem is to enhance homologous
recombination by creating mutants defective in NHEJ. Deletion of the *ku* genes
in *N*. *crassa*, *Kluyveromyces lactis*, and *A*. *nidulans* have all resulted
in increased gene deletion success, with targeted integration rates exceeding
90%.
Ku deletion mutants have also been generated in *C*. *neoformans*, with mutants
exhibiting almost 100% homologous recombination following biolistic
transformation. Lin *et al*. have also shown that the use of a *ku80Δ* mutant
strain increases the rate of homologous integration when using electroporation
up to 75%, making this previously superseded technique a viable alternative to
biolistic transformation provided the recipient strain is a *ku* mutant.
Unfortunately, using Ku deletion mutants to ensure targeted integration
subsequently requires sexual crosses (both time consuming and technically
difficult) with a wild-type partner to restore NHEJ because loss of the Ku
heterodimer alters virulence. Expression of *ku80* is increased during infection
in a human host, and a *ku80Δ* mutant is less successful in a competition model
of murine infection. Consequently, while useful, Ku deletion strains have not
been adopted for widespread use by the *C*. *neoformans* community.
Here we describe an alternate strategy to enhance homologous integration in *C*.
*neoformans* by transient inhibition of NHEJ with the aid of chemical
inhibitors. Using a range of candidate drugs that have been shown to inhibit
NHEJ in mammalian cell lines, we have successfully identified compounds that
enable rates of homologous integration consistently greater than 50%. The
efficacy of these compounds in *C*. *neoformans* provides an attractive tool for
the community as they provide the ability to increase homologous integration in
a strain-independent fashion, making molecular genetic manipulations easier to
achieve.
# Methods
## Strains and growth conditions
*C*. *neoformans* var. *grubii* type strain H99 was stored in 15% glycerol at
-80°C until use, at which point it was maintained on YPD at 4°C for a maximum of
two weeks; this strain was used for all transformations. *C*. *neoformans* was
cultured in liquid (2% bacto-peptone, 1% yeast extract, and 2% glucose (Sigma,
USA)) or solid (2% agar added) YPD media at 30°C. Biolistic transformations were
performed on solid YPD media supplemented with 1 M sorbitol (Sigma, USA), and
transformants were selected on solid YPD medium containing 100 μg/mL G418
(Sigma, USA). Adenine auxotrophy was detected on YNB media (0.45% yeast nitrogen
base w/o amino acids and ammonium sulfate, 2% glucose, 0.5% ammonium sulfate, 2%
agar (Sigma, USA)), and melanin production on L-DOPA media (Sigma, USA).
## Checkerboard assays
The inhibitors NU7026, NU7441, AG14361 (Selleckchem, USA), mirin (Life
Chemicals, Canada), SCR7 (DSK Biopharma Products, USA), W7 hydrochloride (TCI UK
Fine Chemicals, UK), vanillin, chlorpromazine, as well as the DNA damaging
agents phleomycin, hydroxyurea and 6-mercaptopurine riboside (Sigma, USA) were
stored at -20°C until use. Stock solutions and serial two-fold dilutions of each
drug were prepared immediately prior to MIC testing according to the
recommendations of CLSI (CLSI M27-A2) modified for *C*. *neoformans*. For the
Fractional Inhibitory Concentration (FIC) assays, inhibitors were serially
diluted along the abscissa of a 96 well plate, while the DNA damaging agents
were diluted along the ordinate. Plates were incubated at 35°C and visually
scored at 24 and 48 hr. All MIC and FIC assays were performed in duplicate.
MIC was defined by the lowest concentration of a specific drug that inhibits all
visual growth, and was used to determine a ΣFIC value for each combination set.
The ΣFIC was calculated as follows: ΣFIC = FIC A + FIC B, where FIC A equals the
MIC of the DNA damaging agent in the combination divided by the MIC of the DNA
damaging agent alone, and FIC B equals the MIC of the inhibitor in the
combination divided by the MIC of the inhibitor alone. The combination was
considered strongly synergistic when the ΣFIC value is \<0.5, weakly synergistic
when ΣFIC is 0.6 to 1.0, additive when ΣFIC is 1.0 to 2.0 and antagonistic when
the ΣFIC is \>2.
## Gene deletion construct generation
Primers used in this study are listed in ; all PCR was performed using Phusion
DNA Polymerase (New England Biolabs, USA). A deletion construct for the *ADE2*
gene was generated as per Arras *et al*.. Briefly, the construct was prepared
using overlap PCR, employing primers UQ1439 and UQ1442 to join the *ADE2* 5’
region (primers UQ1439 and UQ1440), the G418 resistance marker *NEO* (UQ234 and
UQ235) and the *ADE2* 3’ region (UQ1441 and UQ1442). For *LAC1*, primers UQ3686
and UQ3691 were employed to join the *LAC1* 5’ region (primers UQ3686 and
UQ3687), the G418 resistance marker (UQ3688 and UQ3689) and the *LAC1* 3’ region
(UQ3690 and UQ3691); H99 genomic DNA was used as the template for *ADE2* and
*LAC1*, and the plasmid pJAF1 for *NEO*.
## Transformations
Transformations were performed using a procedure modified from Tofaletti *et
al*. and Davidson *et al*. On the day of transformation, fresh YPD agar plates
supplemented with 1 M sorbitol and the tested inhibitor were prepared and dried.
A 50 mL *C*. *neoformans* culture grown at 30°C for 16 hr was pelleted by
centrifugation, washed in dH<sub>2</sub>O, centrifuged and the pellet
resuspended in 500 μL dH<sub>2</sub>O. Approximately 10<sup>7</sup> cells were
plated onto the inhibitor media, dried in a biosafety cabinet, then incubated
for 4 hr at 30°C.
For transformation, 0.25 g of 0.6 μm Gold Microcarriers (Bio-Rad No. 165–2262)
were resuspended in 2 mL 100% ethanol. Transforming DNA samples were prepared by
mixing 10 μL Gold Microcarriers (Bio-Rad, USA), 10 μL of construct DNA (1
μg/μL), 10 μL 2.5 M CaCl<sub>2</sub> (Sigma, USA) and 2 μl fresh 1 M spermidine
(Sigma, USA), then incubated for 5 min at 20°C. Following centrifugation and
washing in 500 μL 100% ethanol, the sample was pelleted and resuspended in 5 μL
100% ethanol.
For each transformation a Macrocarrier disk (No. 165–2335, Bio-Rad, USA) and a
Stopping Screen (No. 165–2336, Bio-Rad, USA) were sterilized in 100% ethanol and
dried in a Petri dish. Transforming DNA was pipetted onto the center of the
disk, dried then inserted into the Macrocarrier Holder of a Bio-Rad Biolistic
PDS-1000/He Particle Delivery System (No. 165–2257, Bio-Rad, USA). A 1,350 psi
Rupture Disk (No. 165–2330, Bio-Rad, USA) sterilized in 70% isopropanol was
inserted into the Rupture Disk Retaining Cap. The sterilized Stopping Screen was
inserted into the Microcarrier Launch Assembly, the Macrocarrier Holder attached
with the Macrocarrier Cover Lid, and placed in position one in the PDS-1000/He
chamber. Transformation was performed with the *C*. *neoformans* plate in
position three, bombarded at 1,350 psi under a vacuum of 27 in.Hg.
After recovery at 30°C for 4 hr, 2.5 mL dH<sub>2</sub>O was added to each plate,
the cells suspended with a 1 mL pipette tip, and spread over five YPD plates
containing 100 μg/mL G418. Transformants appeared after 3–5 days incubated at
30°C.
## Mutant verification
Correct *ade2Δ* integrants were initially identified *via* their pink coloration
on YPD media and adenine auxotrophy on YNB media. Representative pink
transformants were selected for Southern blot analyses to validate correct
integration of the *ADE2* deletion cassette using the UQ1439 and UQ1442 PCR
product as a probe. Correct *lac1Δ* integrants were identified by their lack of
melanization on L-DOPA media, with random transformants verified by Southern
blot using the UQ3686 and UQ3691 PCR product as a probe. All transformations
were performed in triplicate and only transformations that resulted in 10 or
more colonies were included in subsequent analyses. To determine significance of
correct integration rates between conditions, two-tailed *t*-tests were
completed in GraphPad Prism Version 6.0c (GraphPad Software, USA). *P*-values of
\<0.05 were considered statistically significant.
# Results
## Identification of potential inhibitors of *C*. *neoformans* NHEJ
Given the success in increasing homologous integration in many fungal species by
disrupting NHEJ *via* deletion of genes encoding the Ku proteins, we
hypothesized that it might be possible to achieve similar results by instead
transiently inhibiting key proteins involved in NHEJ repair. Through literature
searches we were able to identify seven candidate chemical compounds that are
known to inhibit different aspects of mammalian NHEJ, are cost effective, and
are readily available on the market.
Therapeutic inhibition of the mammalian DNA-dependent protein kinase
heterodimeric regulatory factor (the Ku heterodimer) can be achieved through
depletion of its cofactor inositol hexakisphosphate (InsP<sub>6</sub>) *via* the
calmodulin antagonists N-(6-aminohexyl)-5-chloro-1-naphthalenesulfonamide (W7)
and 2-chloro-10-(3-dimethylaminopropyl)phenothiazine hydrochloride
(chlorpromazine). Likewise, inhibition of the catalytic subunit of the mammalian
DNA-dependent protein kinase catalytic subunit (DNA-PK<sub>cs</sub>) is also
possible; the *Vanilla planifolia*-derived phenolic aldehyde vanillin, the
benzochromenone 2-(morpholin-4-yl)-benzo\[h\]chomen-4-one (Nu7026) and the aryl-
substituted chromenone 8-dibenzothiophen-4-yl-2-morpholin-4-yl-chromen-4-one
(Nu7441) all inhibit mammalian DNA-PK, and have been investigated as anticancer
agents. Z-5-(4-hydroxybenzylidene)-2-imino-1,3-thiazolidin-4-one (mirin) is a
known inhibitor of Mre11, and
5,6-bis(((E)-benzylidene)amino)-2-thioxo-2,3-dihydropyrimidin-4(1H)-one (SCR7)
targets DNA ligase IV.
In addition to these known NHEJ inhibitors, we also chose to study a compound
not directly associated with NHEJ—the poly(ADP-ribose) polymerase-1 (PARP-1)
inhibitor 1-(4-((dimethylamino)methyl)phenyl)-8,9-dihydro-2,7,9a-
triazabenzo\[cd\]azulen-6(7H)-one (AG14361), a tricyclic benzimidazole that
inhibits single strand DNA repair. As mammalian cells lacking PARP-1 have been
shown to have an increased frequency of homologous recombination, we reasoned
that treatment with an inhibitor of this enzyme may also lead to an increase in
homologous recombination in *C*. *neoformans*.
To our knowledge, none of these eight drugs have ever been tested against *C*.
*neoformans*.
## Growth inhibition of *C*. *neoformans* by inhibitors of mammalian NHEJ
It has previously been demonstrated that the eight chosen inhibitors abolish
growth in mammalian cell lines at high concentrations, however it is unknown if
they exhibit toxicity against *C*. *neoformans*.
To determine the minimum inhibitory concentration (MIC) for each compound in
*C*. *neoformans*, we based our initial concentration ranges on MIC values
identified in previous studies against mammalian cell lines. Initial tests
revealed that growth of *C*. *neoformans* var. *grubii* type strain H99, the
strain in which the vast majority of molecular genetics is performed in this
species, is unaffected at the concentrations toxic to mammalian cell culture.
Following testing with increasing concentrations, growth inhibition was
eventually observed at concentrations substantially higher than that tolerated
by mammalian cells.
## Synergy between inhibitors and DNA damaging agents
While our MIC assays showed that the candidate NHEJ inhibitors were toxic to
*C*. *neoformans* in sufficient concentrations, it did not prove that the
observed effects were mediated *via* DNA repair-mediated processes. To provide
this evidence we investigated their synergy with known DNA damaging agents; if
the observed toxicity was due to inhibition of DNA repair, treatment with the
inhibitors should increase sensitivity to DNA damaging agents.
The growth of strain H99 was tested in a checkerboard assay, with increasing
concentrations of each DNA damaging agent tested in combination with increasing
concentrations of each inhibitor to determine the factorial inhibition
concentration (FIC). The FIC index is based on the Loewe additive zero-
interaction theory, which states that a drug cannot interact with itself and
therefore a drug combination will always be additive with an index of 1. If the
index fluctuates lower indicating synergy it is because more drug would be
required in order to produce the same affect of the drugs alone, and the reverse
also. As the DNA damaging agents are designed to induce double strand DNA breaks
and the NHEJ proteins should be involved in the repair of these breaks, we
hypothesized that if a synergistic effect was observed the tested compounds were
potentially inhibiting *C*. *neoformans* NHEJ.
As DNA damaging agents vary in their effectiveness and method of action, we
selected three for our FIC assays. Phleomycin attracts hydrogen atoms from
deoxyribose, producing apurinic and apyrimidinic sites that enable single strand
breaks, which can in turn become double strand breaks. Hydroxyurea depletes
cells of dNTPs *via* inhibition of ribonucleotide reductase, resulting in
stalled replication forks that can collapse into double strand breaks.
6-mercaptopurine is converted into thioinosinic acid, which perturbs purine
metabolism and results in double strand breaks. As a starting point for FICs,
MIC values were determined for the DNA damaging agents: 10 μg/mL for phleomycin,
1500 μg/mL for mercaptopurine, and 1500 μg/mL for hydroxyurea.
An FIC is considered synergistic when the ΣFIC is \<0.5 and weakly synergistic
when ΣFIC is \>0.5 to \<1.0. All of the inhibitors tested showed a weak
synergistic effect against at least one of the DNA damaging agents tested.
Nu7026, SCR7 and chlorpromazine exhibited synergy with all three DNA damaging
agents, and AG14361 with two. Vanillin, Nu7441, mirin and W7 all showed a weak
synergistic effect with only one DNA damaging agent. These results were
promising, given the *C*. *neoformans ku80Δ* mutant also exhibits slight
sensitivity to phleomycin.
## Treatment with W7, chlorpromazine, NU7026, mirin and AG14361 enhance frequency of deletion of *ADE2*
Given that all eight tested inhibitors exhibited synergy with DNA damaging
agents, all were trialed in biolistic transformation by including the inhibitor
into the YPD + sorbitol plates used at the beginning of the protocol; the strain
to be transformed is incubated on this media for four hours prior to
transformation and recovers for four hours after. The fungus is therefore only
exposed to the inhibitor for eight hours. Adding inhibitor in varying
concentrations at this point would ensure disruption of NHEJ at the time of
transformed DNA entry and integration. After recovery, the *C*. *neoformans* was
transferred to G418 selective media lacking inhibitor, enabling reestablishment
of NHEJ. Four concentrations per inhibitor were trialed, all less than the MIC.
In some instances, much lower concentrations were needed; for example, after no
colonies would grow on high concentrations of NU7026, concentrations were
subsequently dropped to 5, 2.5 and 1.125 μg/mL.
The first gene ever deleted *via* biolistic transformation in *C*. *neoformans*
was the phosphoribosylaminoimidazole carboxylase-encoding *ADE2* gene, loss of
function mutants of which are easily identified due to their pink colonies and
adenine auxotrophy. We employed the same locus in our initial studies as we had
data indicating it was unusually easily deleted, with \~35% of deletion
construct transformants of the *ade2Δ* genotype.
Inclusion of either W7 or chlorpromazine, inhibitors of mammalian calmodulin
proposed to inhibit NHEJ through depletion of Ku cofactor InsP<sub>6</sub>,
resulted in a significant increase in the deletion of *ADE2*. Treatment with W7
at either 5 or 12.5 μg/mL was particularly effective, resulting in deletion
frequencies equivalent to those reported for a *ku80Δ* mutant. A statistically
supported increase in *ADE2* deletion was also observed, but to a lesser extent,
for one of the three inhibitors of mammalian DNA-PKcs (Nu7026), the Mre11
inhibitor mirin and the PARP-1 inhibitor AG14361. Importantly, none of the
strains obtained exhibited changes in growth rate or morphology. Furthermore,
while we observed an increase in the proportion of correct integrants, there was
no gross change in the total number of colonies obtained from each
transformation event.
## Four of the inhibitors also showed an increase when using *LAC1* as a target
Our success in enhancing deletion of the *ADE2* gene was tempered by one
significant fact; it is known that the rate of homologous recombination differs
from locus to locus, and that *ADE2* is particularly easy to delete. To address
concerns of locus specificity, we focused on the five most promising
inhibitors—W7, chlorpromazine, NU7026, mirin and AG14361 –in biolistic
transformations to delete the laccase-encoding *LAC1* gene Each of the
inhibitors were this time tested only at the optimal concentration observed in
the previous experiment.
Without addition of any inhibitor, our control frequency for *LAC1* deletion was
only \~10%; much lower than for *ADE2* and more in line with typical deletion
frequencies observed in the field. Once again W7 and chlorpromazine performed as
the best inhibitors, along with Nu7026. In contrast to the *ADE2* experiment,
inhibitor treatment during *LAC1* deletion resulted in approximately half of all
transformants exhibiting the deletion genotype, a 5-fold increase in frequency
when compared to the no inhibitor control. The failure of mirin to yield a
statistically significant result in the *LAC1* experiment could be due to the
typically large fluctuations in the number of transformants observed during
biolistic transformation.
# Discussion
Despite the clinical importance of *C*. *neoformans*, there have been few
improvements over the last decade in the molecular tools available to
characterize this species. The current methodologies have been invaluable in
elucidating the function of many genes, however these techniques have their
limitations and new advances are required to facilitate everyday experiments,
particularly for large-scale genomic studies. A good example of this is the
recent publication of the sequence of *C*. *neoformans* var. *grubii* type
strain H99, which represents one of the most complete eukaryotic genomes to
date, due in no small part to the extensive transcriptomics performed to enable
its annotation. Of the 6967 predicted protein coding genes, 45.5% (3170) are
listed as hypothetical proteins. The functions of the 1197 miscRNAs also
identified during transcriptome sequencing are likewise unknown. To enable a
deeper understanding of these genes, as well as those with an annotation based
purely on bioinformatic homology, a more robust procedure for performing
molecular genetic modifications is essential.
Efforts to enhance the biolistic transformation procedure through disruption of
NHEJ have had some success. The use of a *ku80Δ* strain can increase rates of
recombination to nearly 100%, but recent publications have revealed that the Ku
proteins may be playing a role during infection of the mammalian host. Liu and
colleagues demonstrated that a *ku80Δ* mutant has diminished virulence in murine
infection competition experiments, and Chen *et al*. discovered transcription of
*KU80* is higher in cells isolated from the cerebrospinal fluid of an infected
human than from cells grown on YPD. Given these observations, it would be unwise
to employ strains carrying the *ku80Δ* mutation during *in vivo* studies since
it may influence the outcome of infection. Consequently, the *ku80Δ* mutant has
seen little use by the community since its development a decade ago.
However the concept underlying the use of *kuΔ* mutants still has merit. By
employing discoveries made in the field of mammalian NHEJ, we have been able to
emulate the higher homologous recombination rates of the *ku80Δ* mutant without
the experimental consequences of permanently compromising NHEJ at the genetic
level. Of the eight inhibitors of mammalian NHEJ that we tested, four influenced
the rate of homologous recombination for multiple loci. Notably, some of the
best results observed correlated with employing the inhibitor at a concentration
much lower than the MIC, and much lower than the concentration that confers
increased sensitivity to DNA damaging agents. For example, W7 enhanced the rate
of generation of a gene deletion at only 5 μg/mL, substantially lower than the
concentration at which it displayed weak synergy with the DNA damaging activity
of hydroxyurea.
Importantly, the use of transient drug treatment to enhance homologous
integration efficiency is also highly cost effective. The best performing
inhibitor, W7, is inexpensive and used at such low concentrations that it only
costs around \$USD 0.26 per plate; a negligible expense that significantly
increases the chance of gaining a correct gene deletion on the first attempt,
saving both time and reagents on repeated efforts to perform a simple genetic
modification. We now employ W7 in all transformations in our laboratory,
routinely achieving correct integration rates exceeding 50%.
In this work we demonstrated a new, easily applied methodology that will
expedite acquisition of precise genetic alterations in *C*. *neoformans* that
has distinct advantages over current protocols. We have successfully employed
multiple inhibitors to reproducibly enhance the deletion rate at multiple loci.
Based on this success, we anticipate that the use of these inhibitors will not
only become widespread in the *Cryptococcus* community, but may also find use in
other fungal species as well. More importantly, they will greatly facilitate the
ongoing efforts designed to elucidate the function of all genes in the *C*.
*neoformans* genome, enabling a much deeper understanding of the process of
pathogenesis in this clinically important species.
# Supporting Information
[^1]: The authors have declared that no competing interests exist.
[^2]: **Conceptualization:** SDMA JAF. **Formal analysis:** SDMA. **Funding
acquisition:** JAF. **Investigation:** SDMA. **Methodology:** SDMA.
**Supervision:** JAF. **Writing – original draft:** SDMA JAF. **Writing –
review & editing:** SDMA JAF. |
# Introduction
*Coxiella burnetii* is the Gram negative bacteria that causes the multifaceted
human disease Q fever, an ever-present global public health threat. Endemic and
hyperendemic regions experience regular cases of Q fever. Sporadic outbreaks
also transpire, the largest being in The Netherlands between 2007 and 2010,
which emphasized both the serious health and economic impact of Q fever. In
particular, the more insidious chronic Q fever can cause significant morbidity
and potentially death. Current therapeutics can be problematic due to contra-
indications or long courses so there is considerable need for novel drugs to
treat Q fever effectively.
Recent stable isotope labelling studies demonstrated <sup>13</sup>C-label
incorporation into lactate by *C*. *burnetii*, despite the apparent lack of a
known lactate synthesis pathway within the *C*. *burnetii* genome. Lactate
production can be beneficial to bacteria by several means. It can be the end
product of efficient metabolism pathways, play a role in redox and energy
homeostasis, be metabolized an alternative carbon source, or modify host immune
response. The ability to synthesize lactate has been linked to virulence in
disparate species such as *Bacillus cereus*, *Staphylococcus aureus*, *Neisseria
gonorrhoeae*, and lactic acid bacteria *Enterococcus faecalis*, *Streptococcus
pyogenes*, *Streptococcus mutans* and *Streptococcus pneumoniae*. Furthermore,
inhibitors of a unique lactate synthesis enzyme of *Cryptosporidium parvum*
reduced replication and pathogenicity. As yet, the method and the essentiality
of lactate synthesis to *C*. *burnetii* replication has not been investigated.
Should unusual enzyme(s) responsible for lactate production in *C*. *burnetii*
be determined and found to play an important role in the organism’s metabolism
and pathogenicity, they may represent useful future anti-*Coxiella* targets.
Lactate dehydrogenases (LDHs) are widely distributed within prokaryotic and
eukaryotic organisms, but no LDH is annotated on the *C*. *burnetii* genome.
LDHs produce lactate from pyruvate and regenerate NADH from NAD<sup>+</sup> in
the process. An LDH is also absent from genome annotations of the close *C*.
*burnetii* relative *Legionella pneumophila* and some strains of fellow
acidophile *Helicobacter pylori*. LDHs share a common ancestor with malate
dehydrogenase enzymes (MDHs) within the same superfamily of dehydrogenases.
Their highly similar structures, containing many conserved residues, can result
in substrate flexibility. Some can use multiple substrates in their native
state, others can be manipulated to switch substrate preferences with very few
residue substitutions. This feature is more widespread in LDHs than MDHs. The
*C*. *burnetii* gene *cbu1241*, annotated as a putative malate dehydrogenase,
has previously been shortlisted as a potential virulence factor that is also
expected to play a vital metabolic role and thus provides a promising
anti-*Coxiella* drug target.
Similarly, malic enzymes (MEs) and malolactic enzymes (MLEs) are closely related
enzymes capable of catalyzing multiple reactions. MEs act to decarboxylate
malate to pyruvate, whereas MLEs decarboxylate malate to lactate without
reduction of an essential NAD<sup>+</sup> cofactor. MEs are broadly distributed
throughout eukaryotes and prokaryotes, with the exception of most lactic acid
bacteria, whereas malolactic enzymes (MLEs) are yet to be found outside lactic
acid bacteria. Substrate flexibility has been demonstrated for numerous members
of both MEs and MLEs, with the MLE of *Oenococcus oeni* capable of particularly
diverse catalytic activity, possessing MDH, LDH and ME functions in addition to
MLE function.
In this study, we analyzed bioinformatic information regarding two *C*.
*burnetii* genes *cbu1241*, annotated as a putative malate dehydrogenase, and
*cbu0823*, annotated as a putative NAD<sup>+</sup>-dependent malic enzyme, in
relation to likely substrate preferences. We then characterized the biosynthetic
capabilities of recombinant CBU1241 and CBU0823 *in vitro* using both
spectrophotometric absorbance measurements and gas chromatography-mass
spectrometry (GC-MS) analysis, with a particular focus on the ability of these
enzymes to synthesize lactate.
Furthermore, we investigated the involvement of *cbu0823* in lactate
biosynthesis *in vivo* utilising stable isotope labelling techniques and a
previously constructed *cbu0823* transposon mutant. This approach also provided
additional information as to the broader effect of a non-functional malic enzyme
on the central carbon metabolism of *C*. *burnetii*. Finally, we examined the
necessity of *cbu0823* for efficient replication of *C*. *burnetii* both
axenically and intracellularly.
# Materials and methods
## Cell strains and culture conditions
*C*. *burnetii* Nine Mile RSA439 (phase II, clone 4) referred to as NMII RSA439
and strains derived from this parent were axenically cultured in liquid
acidified citrate cysteine medium 2 (ACCM-2) at 37°C in 2.5% O<sub>2</sub> and
5% CO<sub>2</sub>. Antibiotic selection for *C*. *burnetii* transposon mutants
was accomplished using chloramphenicol (3μg/ml) with additional kanamycin
(350μg/ml) for complemented mutant strains. All plasmid construction was carried
out in *Escherichia coli* DH5α and recombinant protein expression in *E*. *coli*
JM109. *E*. *coli* strains were cultured in LB broth at 37°C with agitation
unless otherwise specified, adding ampicillin (100 μg/ml), chloramphenicol
(25μg/ml), or kanamycin (50 μg/ml) as required for plasmid selection. THP-1
human monocytic cells (ATCC TIB-202), were propagated in RPMI + GlutaMAX medium
(Gibco, California, USA) supplemented with 5–10% fetal calf serum at 37°C in 5%
CO<sub>2</sub>.
## Bioinformatics
Protein sequences were retrieved from the NCBI Protein Database within the
Geneious Prime software. Sequence alignments were generated using Clustal Omega
1.2.2 within Geneious Prime on default settings, with the alignment order set on
“Group sequences by similarity”. Residues were compared to known key residues
and motifs.
## Protein expression
Full-length *cbu0823* was amplified using PCR with gene-specific
oligonucleotides from purified *C*. *burnetii* genomic DNA. The gene was cloned
into pQE-30 to produce recombinant CBU0823 with an N-terminus 6x-His tag.
*cbu1241* was amplified with gene-specific oligonucleotides and cloned into
pGEX-4-T1 to produce an N-terminus glutathione S-transferase (GST)-fusion
protein. A commercial wine making strain of *Oenococcus oeni* was purchased for
PCR template as a freeze-dried preparation (Vitilac-F, Martin Vialatte, Magenta,
France). The full-length malolactic enzyme was amplified using gene-specific
oligonucleotides then cloned into pQE-30. Cloned genes and affinity tags were
confirmed with Sanger sequencing for all plasmids.
Proteins were expressed by culturing *E*. *coli* containing the relevant protein
expression plasmid construct, the empty pQE-30 or pGEX-4-T1 plasmids (negative
control) in 500 ml LB broth. At OD<sub>600</sub> 0.6–0.8, protein expression was
induced using 0.5 mM Isopropyl β-d-1-thiogalactopyranoside (IPTG) for 16 hours
with agitation at 19°C. Whole cell lysate was then analyzed for protein
expression using SDS-PAGE in TGX Stain-Free protein gels (BioRad, California,
USA) under denaturing conditions and immunoblotting using primary mouse anti-His
IgG (Invitrogen, California, USA) and secondary sheep anti-mouse IgG (GE
Healthcare, Illinois, USA) for His-tagged proteins or with anti-GST-HRP
conjugate (GE Lifesciences, Illinois, USA) for GST-tagged proteins.
The purification procedure used was dependent on affinity tag and protein
characteristics. Pellets containing 6xHis-CBU0823 and empty pQE-30 were
resuspended in 10 ml 50 mM NaH<sub>2</sub>PO<sub>4</sub>/300 mM NaCl/10 mM
imidazole pH 8.0 lysis buffer containing 1 mg/ml lysozyme, then incubated on ice
for 30 minutes. 1 mM phenylmethylsulfonyl fluoride (PMSF) was added prior to
overnight incubation at 4°C. Cells were sonicated and 1% Triton X-100 added
before centrifugation. The resultant supernatant was dialyzed against the same
lysis buffer overnight at 4°C, before being filtered and loaded onto 1 ml NTA-Ni
agarose beads (QIAGEN, Hilden, Germany). Column washing with 50 mM
NaH<sub>2</sub>PO<sub>4</sub>/300 mM NaCl/20 mM imidazole pH 8.0 and elution
with 50 mM NaH<sub>2</sub>PO<sub>4</sub>/300 mM NaCl/250 mM imidazole pH 8.0 was
performed as per manufacturer’s recommendations, except elution volume was
increased to 12 ml. All elutions were pooled and buffer exchanged into
phosphate-buffered saline (PBS), then concentrated to a final volume of 5 ml
using Amicon Ultra-15 UltraCel-30K MWCO Centrifugal Filter (Millipore,
Massachusetts, USA) before storage at 4°C.
6xHis-*O*. *oeni* MLE (denoted 6xHis-oMLE) purification was carried out as above
except buffers were altered to accommodate preferences of a previously described
6xHis-oMLE. oMLE lysis buffer contained 100 mM HEPES/100 mM KCl/10 mM pH 6.5,
wash buffer 100 mM HEPES/100 mM KCl/20 mM imidazole pH 6.5, and elution buffer
100 mM HEPES/100 mM KCl/250 mM imidazole pH 5.9. Elutions were buffer exchanged
into oMLE storage buffer containing 100 mM HEPES/0.1 m MnCl<sub>2</sub> pH 6.0.
Bacterial pellets containing GST-1241 and GST alone were resuspended in a PBS
buffer pH 7.4 and cell lysis performed as above. The resultant supernatants were
syringe filtered prior to loading onto Glutathione Sepharose 4B beads (GE
Healthcare, Illinois, USA). PBS pH 7.4 was used as column wash before eluting
with 50 mM Tris/10 mM reduced glutathione (Sigma, Massachusetts, USA) pH 8.0.
GST-1241 and GST were buffer exchanged into PBS pH 7.4 as above, with the
exception of an Amicon Ultra-15 UltraCel-10K MWCO Centrifugal Filter (Millipore,
Massachusetts, USA) for the smaller GST protein.
Purified protein was visualized on TGX Stain-Free gels and immunoblotted for
His-tagged or GST-tagged proteins as above to confirm the presence of the
correct protein tags on purified proteins. Purified protein was quantified using
Qubit<sup>™</sup> Protein Assay Kit with the Qubit<sup>™</sup> 3.0 Fluorometer
(Invitrogen, California, USA) as per manufacturer’s instructions.
## Spectrophotometric enzyme activity assays
NADP<sup>+</sup> and NADPH were sourced from Roche (Basel, Switzerland) and all
other reagents were purchased from Sigma (Massachusetts, USA). Reactions were
prepared in triplicate in a single tube and the respective protein added
immediately prior to dispensing single reactions of 200 μl into 96-well plates.
This set of reactions was repeated three times in total for each measurement
point, with the mean of each set used to represent the experiment in data
analysis. Assays were repeated a minimum of 3 times. Immediately after
dispensing, measurement of light absorbance at 340 nm, with software pathlength
correction on, was performed on either Synergy H1 Hybrid multi-mode reader
(BioTek, Vermont, USA) or FLUOstar microplate reader (BMG Labtech, Ortenberg,
Germany). Measurements were taken every minute for 10 measurements.
Within this paper, the standard MDH forward reaction was 2 mM oxaloacetate (OAA)
and 0.5 mM NADH in PBS pH 7.4. The standard LDH forward reaction was considered
to be 2 mM pyruvate and 0.5 mM NADH in PBS pH 7.4. This can also detect ME
reverse activity. The standard ME forward reaction for this paper contained 3 mM
malate, 2.5 mM NAD<sup>+</sup> and 1 mM MnCl<sub>2</sub> in PBS pH 7.4. This can
also detect MDH enzyme reverse activity. To assess the necessity of the metal
cation cofactor for 6xHis-CBU0823 activity, selected reactions were repeated
with or without 1 mM MnCl<sub>2</sub>. Suitability of other 2+ metal ions for
6xHis-CBU0823 cofactor function was tested in the standard ME reaction by
replacing MnCl<sub>2</sub> with 1 mM of CaCl<sub>2</sub>, CuCl<sub>2</sub>,
MgCl<sub>2</sub>, ZnSO<sub>4</sub>, FeSO<sub>4</sub>, or NiSO<sub>4</sub>. To
examine cofactor preference, NAD<sup>+</sup> was replaced with NADP<sup>+</sup>
in the standard ME assay and NADH with NADPH in the standard MDH assay.
Standard reactions contained 2 μg GST-CBU1241, 2 μg GST (negative control), 20
μg 6xHis-CBU0823, 28.8 μg 6xHis-oMLE (equivalent to amount used in), or
equivalent volume as 6xHis-oMLE of pQE-30 negative control. Commercially
prepared enzymes were used as positive controls, namely 0.013 units MDH (from
pig heart mitochondria, Roche, Basel, Switzerland) and 0.1 units LDH (from beef
heart, Sigma, Massachusetts, USA). To verify 6xHis-oMLE protein activity, oMLE
assays were performed at 45°C with oMLE storage buffer replacing PBS and altered
substrate concentrations to reproduce the previous characterisation of this
enzyme.
The effect of pH on enzyme activity was determined relative to pH 7.4 by
adjusting the pH of the PBS buffer before mixing of assay reactions through the
range pH 4 to 11 for 1 μg GST-CBU1241 and pH 5 to 11 for 2 μg 6xHis-CBU0823 per
well.
Enzyme kinetic properties for GST-CBU1241 were assessed using 0.1 μg GST-CBU1241
per well. Kinetics for OAA were assessed by increasing OAA concentration from
0.1 mM to 5 mM while maintaining NADH at 0.5 mM. Kinetics for NADH were assessed
by increasing NADH concentration from 0.05 mM to 1.5 mM while maintaining OAA at
0.6 mM.
Enzyme kinetic properties for 6xHis-CBU0823 were assessed using 2 μg 6xHis-
CBU0823 per well. Kinetics for malate were assessed by increasing malate
concentration from 0.5 mM to 10 mM in reactions while maintaining
NAD<sup>+</sup> at 2.5 mM and MnCl<sub>2</sub> at 1 mM. Kinetics for
NAD<sup>+</sup> were assessed by increasing NAD<sup>+</sup> concentration from
0.1 mM to 5 mM while maintaining malate at 3 mM and MnCl<sub>2</sub> at 1 mM.
Kinetics for MnCl<sub>2</sub> were assessed by increasing MnCl<sub>2</sub>
concentration from 0.01 mM to 10 mM while maintaining malate at 3 mM and
NAD<sup>+</sup> at 2.5 mM.
Michaelis constant (K<sub>m</sub>) and maximal reaction velocity within the
system (V<sub>max</sub>) were calculated for GST-CBU1241 for OAA and 6xHis-
CBU0823 for malate, NAD<sup>+</sup> and MnCl<sub>2</sub> using substrate
concentrations below inhibitory levels on a Lineweaver-Burke plot using
nonlinear regression (straight line) within GraphPad Prism 9.0 on default
settings. This equates to above 1.5 mM<sup>-1</sup> OAA for GST-CBU1241, and
above 0.3 mM<sup>-1</sup> malate, above 1.25 mM<sup>-1</sup> NADH and above 1
mM<sup>-1</sup> MnCl<sub>2</sub> for 6xHis-CBU0823. Rate constant
(k<sub>cat</sub>) was calculated from V<sub>max</sub>. Substrate inhibitor
constant (K<sub>Si</sub>) was calculated using equations from. Briefly, using
the calculated V<sub>max</sub>, K<sub>Si</sub> was represented by the slope when
v/(V<sub>max</sub>−v) was plotted against 1/\[S\] (reciprocal of the quotient
velocity plot). For GST-CBU1241, this slope was calculated for 1/\[OAA\] between
0.5 and 4 mM<sup>-1</sup>. For 6xHis-CBU0823, the slope was calculated for
1/\[malate\] between 0 and 0.4 mM<sup>-1</sup>, for 1/\[NADH\] between 0 and
1.25 mM<sup>-1</sup>, and for 1/\[MnCl<sub>2</sub>\] between 0 and 1
mM<sup>-1</sup>.
## Enzyme activity assay with GC-MS product detection
200 μl reactions containing 3 mM malate, 5 mM NAD<sup>+</sup> and 1 mM
MnCl<sub>2</sub> and either 54 μg 6xHis-CBU0823, 28.8 μg 6xHis-oMLE, 3 μl of
negative control purification (equivalent volume as 6xHis-CBU0823) or 3 μl PBS
of blank control were prepared, with six replicates of each. After 20 minutes,
the reaction was stopped by adding 100 μl of the reaction to 300 μl 100%
methanol, then centrifugation used to remove the protein. Internal standards of
1 nmol <sup>13</sup>C<sub>6</sub>-sorbitol and 10 nmol
<sup>13</sup>C<sub>5</sub>,<sup>15</sup>N-labelled valine were added to a 40 μl
aliquot of the previous supernatant before storage at -80°C. For GC-MS analysis,
5 μl of this aliquot was dried completely in glass vial inserts within a
rotational vacuum concentrator (Concentrator *Plus*, Eppendorf, Hamburg,
Germany), including a final drying step of 30 μl 100% methanol.
The samples were derivatized with methoxyamine (Sigma, Massachusetts, USA) and
N-bistrimethylsilyltrifluoroacetamide (BSTFA) containing 1%
trimethylchlorosilane (TMCS; Thermo Scientific, Massachusetts, USA) before
metabolites were analyzed on the Agilent 6545 series quadrupole mass
spectrometer (Agilent Technologies, California, USA), using a protocol described
previously. Metabolite identification was performed by comparing molecular mass
and retention time to authentic standards using MassHunter software (Agilent
Technologies, California, USA). Results for each replicate were normalized to
the valine internal standard value.
## *C*. *burnetii* transposon mutant complementation
*C*. *burnetii cbu0823* transposon (Tn) mutant from a previously generated
transposon mutant library was clonally isolated on semi-solid ACCM-2 agarose
containing chloramphenicol. PCR screening using transposon- and gene-specific
oligonucleotide pairings confirmed the transposon insertion within the
*0823*::Tn mutant 832 basepairs downstream of the start codon, previously
documented in.
To create the plasmid for genetic complementation of the *0823*::Tn mutant, full
length *cbu0823* was amplified using gene-specific oligonucleotides, then cloned
into the *C*. *burnetii* complementation vector pJB-kan:3xFLAG-MCS. Sequence and
FLAG tag was confirmed by Sanger sequencing.
The complemented *0823*::Tn mutant strain was generated by introducing the pJB-
kan:3xFLAG-*cbu0823* plasmid into the *0823*::Tn mutant as described previously.
Recovered bacteria were plated onto ACCM-2 agarose with chloramphenicol and
kanamycin. After 7 days of incubation, colonies were picked into 1 ml ACCM-2
with chloramphenicol and kanamycin selection in 24-well plates. Wells containing
visible growth were harvested and whole cell lysate screened for 3xFLAG-CBU0823
expression by immunoblotting with primary anti-FLAG antibody (Sigma,
Massachusetts, USA) and secondary anti-mouse IgG (GE Healthcare, Illinois, USA).
## Stable isotope labelling analysis
The method used in Hauslein et al. was chosen to investigate lactate
biosynthesis in *C*. *burnetii* as it produced the highest stable isotope label
incorporation into lactate of published works. Five replicates of 20 ml ACCM-2
containing 5 mM <sup>13</sup>C-U-glucose (Sigma, Massachusetts, USA), each
inoculated with 2 x 10<sup>6</sup> GE/ml of the relevant *C*. *burnetii* strain,
were incubated for 7 days as previously described,then processed for metabolite
extraction in methanol:water:chloroform 3:1:1 v/v as previously described.
Preparation of glass inserts, derivatization and GC-MS analysis was performed as
for the enzyme activity GC-MS assay, except the whole aqueous phase sample was
dried down. DExSI software was used for metabolite identification by comparison
with an in-house Metabolomics Australia library of authentic standards for
molecular masses and retention times. The peak integrations for all relevant
mass isotopologues were combined for every detected metabolite and corrected to
natural background isotopic abundance to give fractional labelling. The
fractional labelling was then normalized to the initial <sup>13</sup>C-glucose
level within each replicate. Isotopologues were graphed using DExSI applying the
inbuilt natural isotopic abundance correction for 0% unlabelled biomass.
Detected metabolites were mapped to known *C*. *burnetii* metabolic pathways.
## Intracellular replication assays in THP-1 cells
THP-1 cells were seeded at 5 x 10<sup>5</sup> cells per well into 24-well
plates, with or without sterile glass coverslips, differentiated into
macrophage-like cells by treating with 10 nM phorbol 12-myristate 13-acetate
(PMA) and incubated for 3 days. 7-day cultures of *C*. *burnetii* strains were
pelleted and resuspended in PBS. Genome equivalents were quantified using a
quantitative qPCR that targets *ompA* as described previously, using
oligonucleotides listed in. Each well was infected with a multiplicity of
infection (MOI) of 25 with *C*. *burnetii* resuspended in 500 μl RPMI + 5% FCS
for 4 hours at 37°C 5% CO<sub>2</sub>. Each strain was prepared in triplicate
per replicate and six independent biological replicates were performed.
After infection, all wells were washed with PBS, then, except for day 0, 500 μl
RPMI + 5% FCS was added per well for incubation at 37°C 5% CO<sub>2</sub>.
Bacteria were harvested on days 0, 1, 3, 5, and 7. With the exception of day 0,
the media from each well was collected and pelleted for 15 minutes at 13 200 x
g. Attached cells were lyzed with nuclease-free water for 20 minutes. The lyzed
cells were scraped from the well and added to the pellet from the media, before
repeat centrifugation. The resulting pellet was resuspended in 100 μl nuclease-
free water and *C*. *burnetii* genome equivalents (GE) determined by the *ompA*
qPCR. Fold change was calculated for each time point relative to day 0.
Immunofluorescence microscopy slides were prepared on day 3 post-infection. The
media was removed from wells containing coverslips and cells fixed in 4%
paraformaldehyde before being permeabilized with 0.05% saponin + 2% BSA in PBS
for 1 hour, then washed three times with PBS. Primary antibodies against *C*.
*burnetii* (Roy Laboratory, Yale University, Connecticut, USA) and LAMP-1
(Developmental Studies Hybridoma Bank, Iowa, USA) at 1:10,000 and 1:500
respectively were applied for one hour, before further washing with PBS.
Secondary antibodies anti-mouse AlexaFluor 488 and anti-rabbit AlexaFluor 568
(Thermo Fisher Scientific, Massachusetts, USA) were, both at 1:3,000, were
applied for one hour and the first PBS wash that followed contained 1:10,000
4′,6-diamidino-2-phenylindole (DAPI) (Invitrogen, California, USA). Dako
Fluorescent Mounting Media (Agilent Technologies, California, USA) was used to
mount coverslips to glass slides. Images were obtained using Leica 780 and 700
confocal microscopes (Biological Optical Microscopy Platform, University of
Melbourne, Melbourne, Australia) and processed in ImageJ.
Vacuole size quantification was completed in ImageJ, by measuring vacuole area
in μm<sup>2</sup> as seen on the anti-LAMP channel of at least 50 vacuoles per
replicate. Four of the six biological replicates had sufficient quality
immunofluorescence images to allow for vacuole quantification.
## Statistical analysis
All graphs were prepared using Prism 9.0 (GraphPad, California, USA). Unpaired
two-tailed *t*-tests were performed in Prism 9.0 to compare the means of each
enzyme activity in the GC-MS detection activity assay, for each metabolite in
the stable isotope labelling experiment, as well as average vacuole area.
Ordinary one-way ANOVA was performed in conjunction with Tukey’s multiple
comparisons test in Prism 9.0 to examine differences in cofactor preference of
6xHis-CBU0823. Statistical differences in the intracellular replication assays
between strains at each time point were calculated in Excel using unpaired two-
tailed Student’s *t*-tests. A threshold significance of *p \<* 0.05 was used in
all analyses.
# Results
## The CBU1241 amino acid sequence is consistent with other malate dehydrogenases and CBU1241 possesses malate dehydrogenase function *in vitro*
Clustal Omega alignment of select MDH and LDH protein sequences revealed that
CBU1241 and the putative MDH from *Legionella pneumophila* str. Philadelphia 1
shared the greatest identity at 63%. CBU1241 was more similar to the putative
*Thermus thermophilus* MDH (56% identity) and the pig cytoplasmic MDH isozyme
(46% identity) than to *E*. *coli* MDH (24% identity). This subgrouping of *T*.
*thermophilus* with eukaryotic cytoplasm MDHs, as well as *Mycobacterium*
species, has been noted previously and *C*. *burnetii* and *L*. *pneumophila*
MDHs appear to lie within the same subgroup.
The presence of a Glu at position 42 and an Arg at position 92 predicts that
CBU1241 should have MDH function. Other residues typical of the LDH and MDH
superfamily involved in substrate binding (Arg98, Asn131 and Arg162) and proton
transfer during catalytic action (Asp159 and His187) are conserved in CBU1241.
No residues uncharacteristic of an MDH were identified in CBU1241.
To assess the *in vitro* catalytic activities of CBU1241 in enzyme activity
assays, recombinant GST-CBU1241 was expressed and purified. When GST-CBU1241 was
used in the MDH assay containing OAA and NADH, oxidation of NADH was observed as
decreasing absorbance at 340 nm. GST alone did not produce a measurable
absorbance change, suggesting that CBU1241 exhibited its predicted MDH function
*in vitro*. The positive control, a commercially produced MDH, produced robust
absorbance change in the same assay conditions.
To evaluate lactate production *in vitro*, GST-CBU1241 was used in the LDH assay
containing pyruvate and NADH. No change in absorbance was detected with either
GST-CBU1241 or GST. By contrast, the commercially produced LDH showed activity
in standard assay conditions. GST-CBU1241 MDH activity was maximal at pH 9.0,
showing a preference for alkaline conditions.
Enzyme kinetics showed that GST-CBU1241 activity was maximal between 0.3 and 0.6
mM OAA and substrate inhibition was observed above this concentration. For OAA,
K<sub>m</sub> was calculated as 0.11 mM (95% CI 0.07–0.17), V<sub>max</sub> as
63.61 μmol/min per mg (95% CI 54.61–76.22), k<sub>cat</sub> as 65.20
s<sup>-1</sup> (95% CI 55.98–78.13) and K<sub>i</sub> was calculated as 1.25 mM
(95% CI -0.85–3.39). Lineweaver-Burke and reciprocal of the quotient velocity
plots are included in the supplementary information. Enzyme activity more
closely resembled classical Michaelis-Menten kinetics with increasing NADH
concentrations. The calculated K<sub>m</sub> for NADH was 0.12 mM (95% CI
0.06–0.22), V<sub>max</sub> 49.13 μmol/min per mg (95% CI 40.68–59.45), and
k<sub>cat</sub> 50.36 s<sup>-1</sup> (95% CI 41.70–60.94). These values fall
amongst those reported for other bacterial MDH proteins.
Together, this *in vitro* assessment demonstrated that GST-CBU1241 acts as an
MDH, having no LDH activity, with a preference for alkaline conditions.
## CBU0823 shares many key residues with other bacterial malic enzymes and functioned as a malic enzyme and malate dehydrogenase *in vitro*
Clustal Omega alignment of ME and MLE protein sequences showed CBU0823 was most
similar to one of the putative NAD<sup>+</sup>-dependent MEs of *L*.
*pneumophila* Q5ZRB1 with 55% identity, which was expected given their close
phylogeny. CBU0823 also showed high identity with MLEs, 42% with the
multifunctional *O*. *oeni* MLE and 41% with the MLE from *Lactobacillus casei*,
permitting that CBU0823 could possess MLE function.
CBU0823 contained some key residue motifs not predicted by its
NAD<sup>+</sup>-dependent ME annotation. The presence of Gly162 in CBU0823 is
more consistent with MLEs, as bacterial MEs usually contain an Arg, however,
yeast MEs also contain a Gly at the equivalent residue. Moreover, Val117 in the
malate binding site of CBU0823 differed from both MLEs (Ile) and bacterial MEs
(Cys), although all residues are hydrophobic.
In CBU0823, the GXGXXG motif beginning at Gly307 is consistent with other
prokaryotic ME sequences and corresponds to the observed NAD<sup>+</sup> binding
site in resolved ME crystal structures. Additional evidence for an
NAD<sup>+</sup>-preference by CBU0823 can be found where the strictly conserved
Asp341 and Gly344 is not surrounded by the motif of NADP<sup>+</sup>-preferring
enzymes (Ser342, Lys343 and Arg351).
Residues involved in catalytic activity (Tyr109, Lys180 and Asp274) and residues
known to bind the metal cation cofactor (Glu251, Asp252, Asp275) were strictly
conserved in CBU0823, reflecting their crucial functions.
To assess the *in vitro* catalytic activities of CBU0823, recombinant 6xHis-
CBU0823 was expressed and purified. When the 6xHis-CBU0823 recombinant protein
was added to the ME assay containing malate, NAD<sup>+</sup> and
MnCl<sub>2</sub>, NAD<sup>+</sup> reduction was observed as increasing
absorbance at 340 nm, indicating 6xHis-CBU0823 exhibited the annotated ME
function *in vitro*. The pQE-30 negative control produced no absorbance change
in this ME assay.
Akin to GST-CBU1241, there was no change in absorbance when 6xHis-CBU0823 was
used in the LDH standard assay, suggesting no lactate production. Interestingly,
6xHis-CBU0823 was capable of oxidizing NADH in the presence of OAA in the
standard MDH reaction, indicating it possessed both ME and MDH activity.
6xHis-CBU0823 preferred increasingly alkaline conditions and activity was not
detected in acidic conditions. The ME activity of recombinant 6xHis-CBU0823
exhibited a strong preference for NAD<sup>+</sup>, as activity with
NADP<sup>+</sup> cofactor was significantly reduced to 8.23±0.24% to that with
NAD<sup>+</sup>. This is consistent with other enzymes from the same enzyme
class. Removal of the Mn<sup>2+</sup> cofactor also significantly reduced ME
activity of 6xHis-CBU0823 to 20.14±13.38%. Fe<sup>2+</sup> effectively replaced
Mn<sup>2+</sup> as cofactor. All other tested metals provided significantly
reduced activity, with Zn<sup>2+</sup> the next most proficient cofactor
replacement with only 29.98±12.68% activity.
Michaelis-Menten plots revealed substrate inhibition also occurred with 6xHis-
CBU0823. Maximum activity was between 2 and 3 mM malate, at 0.8 mM NADH, and at
1 mM MnCl<sub>2</sub>. For malate, K<sub>m</sub> was calculated as 0.40 mM (95%
CI 0.01–1.25), V<sub>max</sub> as 1.06 μmol/min per mg (95% CI 0.78–1.67),
k<sub>cat</sub> as 1.12 s<sup>-1</sup> (95% CI 0.82–1.77) and K<sub>i</sub> was
calculated as 10.86 mM (95% CI -7.31–29.05). These values fall amongst those
reported for other bacterial MDH proteins. Lineweaver-Burk and reciprocal of the
quotient velocity plots are included in the supplementary information.
For NADH, K<sub>m</sub> was calculated as 0.31 mM (95% CI 0.03-undefined),
V<sub>max</sub> as 2.06 μmol/min per mg (95% CI 0.78-undefined), k<sub>cat</sub>
as 2.18 s<sup>-1</sup> (95% CI 0.83-undefined) and K<sub>i</sub> was calculated
as 6.30 mM (95% CI 0.27–12.33). Lineweaver-Burk and reciprocal of the quotient
velocity plots are included in the supplementary information.
For MnCl<sub>2</sub>, K<sub>m</sub> was calculated as 0.06 mM (95% CI
0.04–0.09), V<sub>max</sub> as 1.34 μmol/min per mg (95% CI 1.07–1.90),
k<sub>cat</sub> as 1.45 s<sup>-1</sup> (95% CI 1.13–2.01) and K<sub>i</sub> was
calculated as 6.82 mM (95% CI -5.15–18.79). Lineweaver-Burk and reciprocal of
the quotient velocity plots are included in the supplementary information.
6xHis-oMLE did not perform well in the standard assay conditions. For example,
in the ME standard reaction, 6xHis-oMLE did not have detectable activity. To
confirm the purified protein could behave similarly to the previously published
study, reactions were repeated with substrate concentrations as previously
published at 45°C using oMLE storage buffer as diluent. 6xHis-oMLE activity was
improved in these conditions. Notably, the observed reduction of OAA indicating
MDH activity was not detected in the previous characterisation and represents
yet more substrate flexibility of this versatile enzyme.
These *in vitro* data demonstrated that CBU0823 acts preferentially as an ME,
converting malate to pyruvate. It is also capable of MDH activity but not of MLE
or LDH activity. It has a strong preference for alkaline conditions and
NAD<sup>+</sup>/NADH and Mn<sup>2+</sup> as cofactors, though it will tolerate
some substitutions.
## CBU0823 does not exhibit malolactic enzyme activity *in vitro*
The GC-MS analysis of *in vitro* activity indicated that 6xHis-oMLE could
generate amounts of lactate and pyruvate significantly above baseline,
corroborating the previously published work showing oMLE has ME and MLE
activity. 6xHis-CBU0823 similarly produced pyruvate significantly above baseline
supporting that malic enzyme activity was demonstrated in, but did not produce
lactate above baseline levels , suggesting it cannot act as an MLE.
Comparatively 6xHis-CBU0823 generated significantly less pyruvate than 6xHis-
oMLE, particularly as double the amount of 6xHis-CBU0823 enzyme was used. This
suggests that despite the ME activity of CBU0823 being more efficient than its
MDH activity, it is still less active than the malic enzyme activity of the
multipurpose enzyme oMLE in these assay conditions. The instability of OAA
precludes its detection by GC-MS, therefore we were unable to confirm whether
CBU0823 could produce OAA by means of a reverse malate dehydrogenase activity.
## Loss of *cbu0823* causes no detectable change in <sup>13</sup>C-incorporation into lactate
In order to evaluate the *in vivo* capacity of CBU0823 to synthesize lactate in
*C*. *burnetii*, we wanted to compare stable isotope label enrichment in lactate
in the *cbu0823* transposon insertion mutant to that of the wildtype strain.
Following confirmation of the transposon insertion in the *0823*::Tn mutant, it
was genetically complemented with pJB-kan:3xFLAG-*cbu0823* plasmid to provide
constitutively expressed 3xFLAG-CBU0823. Expression of a correct sized product
was confirmed on a western blot probed for FLAG tag.
The *C*. *burnetii* strains were cultured for 7 days in ACCM-2 containing
\[<sup>13</sup>C\]glucose before harvest and analysis, conditions in which *C*.
*burnetii* can incorporate labelling into lactate. In wildtype, label enrichment
into lactate was 14±4.6%, compared to 9.3±4.0% in the *0823*::Tn mutant and
11.2±3.8% in the *0823*::Tn pFLAG-CBU0823 complemented mutant. These
<sup>13</sup>C-incorporation levels were not statistically significantly
different (*p* \> 0.05), suggesting CBU0823 is not involved in lactate
biosynthesis. The wildtype label enrichment in lactate detected in this study is
similar to the previously published study, despite differing label enrichment
calculation methodologies. In the small amount of labelled lactate detected, all
strains showed an isotopologue pattern of mostly M+2, with a small amount of
M+1, once natural isotopic abundance correction was applied, suggesting the
precursor might be a TCA cycle intermediate. In contrast, Hauslein et al.
observed a predominantly M+3, more suggestive of a pyruvate or glycolysis
intermediate precursor.
## Loss of *cbu0823* alters central carbon metabolism of *C*. *burnetii*
The stable isotope labelling studies examining lactate biosynthesis
advantageously provided information about the wider consequences of disabling
*cbu0823* to *C*. *burnetii* central carbon metabolism. After 7 days incubation
with \[<sup>13</sup>C\]glucose, the fraction labelled was significantly reduced
in lower glycolysis intermediates within *0823*::Tn mutant compared to wildtype.
For example, glycerate 3-phosphate within the *0823*::Tn mutant contained only
60.7±2.1% <sup>13</sup>C-incorporation compared to 95.0±8.0% in wildtype and
87.6±13.5% in the *0823*::Tn pFLAG-CBU0823 complemented mutant. This suggests
that overall the glycolytic pathway is less active in the *0823*::Tn mutant
compared to wildtype, at least in its utilization of glucose.
The significant reduction in <sup>13</sup>C-incorporation from
\[<sup>13</sup>C\]glucose was also observed in TCA cycle intermediates in the
absence of CBU0823. All metabolites contained significantly lower label
enrichment in the *0823*::Tn mutant than in wildtype, except for succinate. For
instance, label incorporation into malate, the substrate for CBU0823 and the
product of CBU1241 in the *in vitro* assays, was 34.2±3.2% in wildtype yet only
15.4±1.3% in the *0823*::Tn mutant and 20.6±3.3% in the *0823*::Tn pFLAG-CBU0823
complemented mutant strain. Additionally, TCA cycle derivates aspartate and
glutamate had lower label inclusion levels, suggesting the decreased fraction
labelling continued into these derivative biosynthesis pathways as well. The M+2
isotopologue was most prevalent through detected TCA cycle intermediates,
supporting previous findings of carbon entering the TCA cycle via a fully
labelled acetyl-CoA intermediate.
It is worth noting that the inconsistent pattern on label enrichment before and
after citrate/isocitrate is likely due to citrate/isocitrate enrichment dilution
by unlabelled citrate, present in large amounts in *C*. *burnetii* axenic media
and remaining despite multiple washes, and is not related to any unusual TCA
cycle direction within *C*. *burnetii*. The negative label enrichment in serine
detected is physiologically impossible and probably detection of noise rather
than true <sup>13</sup>C-incorporation levels. The differences observed between
wildtype and the *0823*::Tn pFLAG-CBU0823 complemented mutant strain are likely
explained by the non-physiological, constitutive expression of pFLAG-CBU0823.
Taken as a whole, these data suggest the loss of CBU0823, a malic enzyme, within
*C*. *burnetii* led to significant reductions in carbon flux from glucose
through glycolysis and within the TCA cycle as well. This is potentially due to
an inability to maintain adequate metabolic intermediates in the absence of the
malic enzyme, particularly malate and acetyl coenzyme A (acetyl CoA), thus
preventing metabolic processes from continuing as normal.
## *cbu0823* is not required for efficient *C*. *burnetii* replication
Finally, we investigated the impact what impact the metabolic alterations seen
in the *0823*::Tn mutant might have on *C*. *burnetii* replication inside host
cells. The fold change in genome equivalents of *C*. *burnetii* strains was
measured within THP-1 macrophage-like cells infected with an MOI of 25 over 7
days by qPCR. No significant differences were observed in replication between
strains over 7 days (*p* \> 0.05). The strains were visibly indistinguishable on
representative immunofluorescent images taken on day 3 post-infection and no
significant differences in vacuole size were observed.
# Discussion
It is yet to be determined how *C*. *burnetii* accomplishes lactate synthesis,
though it can reproducibly incorporate stable isotope label into lactate in
multiple studies. This study has demonstrated that CBU1241 and CBU0823 are
unlikely to be responsible for the lactate production observed. Neither CBU0823
nor CBU1241 exhibited any LDH activity in *in vitro* enzyme activity assays and
CBU0823 did not produce lactate from an MLE reaction. *In vivo* assessment
indicated the *0823*::Tn mutant was as able to incorporate <sup>13</sup>C label
into lactate in an equivalent manner to wildtype. Taken together, these results
indicate that the metabolic pathway by which *C*. *burnetii* synthesizes lactate
is still to be established.
The characterization in this study of CBU1241 revealed robust OAA reduction. *in
vitro* CBU1241 is clearly affected by substrate inhibition, with activity
diminishing above 0.6 mM OAA. 0.6 mM is a lower concentration than for many
other published bacterial MDHs, though it is comparable to pig heart
mitochondrial MDH and higher than some, such as *T*. *flavus* MDH. Due to its
instability, accurate intracellular OAA values are currently not available.
Together with the lack of knowledge regarding the MDH malate reduction reaction,
the level of substrate inhibition CBU1241 experiences *in vivo* remains unknown.
The variation of two key residues within CBU0823 from other ME suggested the
enzyme could possess the substrate flexibility characteristic of malic enzymes
(MEs). In this study, the *in vitro* experiments demonstrated recombinant
CBU0823 lacked any detectable LDH or MLE function but did possess both ME and
MDH activity. Nonetheless, the simple *in vitro* assessment of CBU0823, and
CBU1241, function may not fully represent enzyme capabilities *in vivo*.
The reduced incorporation of label into metabolites across central carbon
metabolism in the *0823*::Tn mutant cultures appears similar to the rerouting of
carbon away from glycolysis and the TCA cycle noted in ME mutant strains of
*Sinorhizobium meliloti*, a soil dwelling symbiont bacteria associated with
alfalfa roots. These metabolism modifications may be attributable to the loss of
the cataplerotic reaction catalyzed by MEs, removing malate from the TCA cycle,
and the resultant anaplerotic reaction of pyruvate conversion to acetyl CoA for
entry into the TCA cycle. Instead, acetyl CoA replenishment from TCA cycle
intermediates would require the anaplerotic reaction of OAA to
phosphoenolpyruvate (PEP) by PEP carboxykinase (PckA). In other words, TCA cycle
function in the ME mutants likely would rely on the slowest reaction in the
cycle, the MDH enzyme, to produce sufficient OAA to both remain in the TCA cycle
and to exit the TCA cycle for replenishment of acetyl CoA. From this data, we
cannot determine how the loss of the secondary MDH function of CBU0823 in the
*0823*::Tn mutant contributed to TCA cycle dysfunction.
The replication of the *0823*::Tn mutant was not lower than wildtype
intracellularly. The *S*. *meliloti* ME mutants with comparable metabolome
changes did exhibit slowed replication, in contrast to *E*. *coli* ME mutants
where simultaneous removal of the alternate pyruvate producing PckA pathway was
required to affect replication. The MDH-PckA pathway may provide sufficient OAA
and PEP to maintain typical *C*. *burnetii* replication, especially as the
normal replication rate of *C*. *burnetii* is comparatively slow. Furthermore,
reduction in TCA cycle activity reduces oxidative stress, an issue for *C*.
*burnetii* within its replicative niche, and the reduced TCA activity may be of
benefit to the *0823*::Tn mutant. Moreover, this study concentrates on glucose
utilization whereas *C*. *burnetii* has a metabolism capable of utilizing amino
acids, thus the additional energy required by the *0823*::Tn mutant for
replication could be provided by carbon sources other than glucose.
This study has demonstrated the *C*. *burnetii* gene *cbu1241* encodes an enzyme
with *in vitro* MDH function and *cbu0823* encodes an enzyme with both ME and
MDH function *in vitro*, albeit a less efficient MDH than *cbu1241*. Neither
enzyme demonstrated the capacity to produce lactate *in vitro*, and *in vitro*
labelling studies suggested that CBU1241 was not responsible for lactate
production *in vivo*. Although not required for efficient replication in axenic
culture or within host cells, *cbu0823*, most likely through its malic enzyme
function, is required for normal glycolysis and TCA cycle function in *C*.
*burnetii*. Future work examining lactate production by *C*. *burnetii* may
include large scale screening studies of mutant libraries and/or sophisticated
bioinformatic analysis, in order to identify new potential candidate enzymes.
# Supporting information
Technical support was provided by Chen Ai Khoo, Newton Laboratory, Peter Doherty
Institute, University of Melbourne. Confocal imaging was performed at the
Biological Optical Microscopy Platform, University of Melbourne
([www.microscopy.unimelb.edu.au](https://www.microscopy.unimelb.edu.au)).
[^1]: The authors have declared that no competing interests exist. |
# Introduction
Persons with severe mental illness (SMI) including schizophrenia, bipolar
disorder and major depression are approximately twice as likely to have a
stroke. Increased stroke risk among those with SMI has been attributed to
unhealthy lifestyle choices like tobacco use, substance abuse, low physical
exercise, poor nutrition, and morbid obesity. Use of atypical antipsychotics,
such as olanzapine and clozapine, also account for significant weight gain and
may contribute to risk for stroke. Additionally, functional impairments
associated with SMI place them among the most vulnerable of social groups where
they encounter poverty, chronic unemployment, substandard housing, poorer access
to quality health care, disrupted social relations, and delayed preventative
care, which exacerbate overall risk for poor health outcomes. Consequently,
individuals with SMI experience about 25 years of reduced life expectancy.
Little is known about stroke survivors with SMI regarding their patterns of
post-stroke hospitalization. Unplanned hospitalizations among stroke survivors
are costly, may reflect suboptimal patient outcomes, and are often related to
other co-morbid conditions. However, there is conflicting information on whether
persons with SMI are at increased risk for non-psychiatric hospitalization after
an initial stroke. For instance, a recent systematic review of hospitalization
among stroke survivors does not mention SMI as a contributing factor in any
reviewed studies. Yet, previous research suggests that stroke survivors with SMI
may be particularly vulnerable to poor stroke outcomes requiring hospitalization
because they have high readmission rates across a spectrum of other health
conditions, even after adjusting for other comorbidities and lifestyle factors.
Factors that have been related to post-stroke hospitalizations among stroke
survivors include age, race, education, insurance coverage, history of physical
comorbidity prior to stroke, hospitalizations prior to stroke, hypertension,
coronary artery disease, diabetes, and residential proximity to hospital.
The current study compares the risk of non-psychiatric hospitalizations for
veteran stroke survivors with and without SMI and examines patterns of
hospitalization in Veterans Health Administration (VHA) hospitals over a five-
year period. Veteran stroke survivors, whether or not they have severe mental
illnesses, are vulnerable and have relatively poor health status. For example,
the health status of veterans is consistently found to be poorer in comparison
to the rest of the United States with increased rates of obesity, diabetes,
homelessness, and substance abuse. Individuals with SMI in the general
population who have never had a stroke have higher non-psychiatric
hospitalizations even after adjusting for other co-morbidities and lifestyle
factors. Yet, there is a paucity of research on the influence of SMI on non-
psychiatric hospitalizations among veteran stroke survivors receiving care in
the VHA system. It is notable that in a relatively recent systematic review of
predictors of hospital admissions after stroke that not a single reviewed study
considered SMI as either a primary independent variable or as a covariate
related to post-stroke hospitalizations. The aim of this study is to begin to
fill the gap in the literature on the role of SMI in post-stroke non-psychiatric
hospitalizations.
# Materials and methods
## Hypothesis
We conducted a five-year retrospective cohort study of veterans identified
through the U.S. Department of Veterans Integrated Service Network Five as
having an index stroke in 2003 and receiving their care in the VHA system.
Our hypothesis was that stroke survivors with SMI would have increased risk of
non-psychiatric hospitalizations over five-years compared to stroke survivors
without SMI after adjustment for inpatient stroke treatment variation, age,
race, marital status, education, income, health insurance coverage above VHA
benefits, proximity of patient’s home to a VHA hospital, physical co-morbid
conditions, prior pre-stork hospitalizations, mental co-morbid conditions and
history of stroke risk factors of hypertension, diabetes, and peripheral
vascular disease. The rationale for this hypothesis was that individuals with
SMI have myriad lifestyle factors, health care access difficulties, and
medication side effects, which increase their risk of physical diseases
resulting in greater non-psychiatric hospitalization rates.
## Defining stroke, non-psychiatric hospitalizations and SMI
The International Classification of Diseases, Ninth Revision (ICD-9) was used to
identify ischemic and hemorrhagic stroke participants. ICD-9 diagnosis codes
have been extensively used for stroke outcomes research and have been validated
with high sensitivity (86%), specificity (95%), and positive predictive value
(86–92%). The following ICD-9 CM diagnostic codes were used to identify stroke
patients from either inpatient or outpatient encounters: 433.x1, 434 (excluding
434.x0), or 436. Outpatient records were important to consider because many VHA
medical centers contract with their affiliated medical school hospital to
provide specialty care services like acute stroke care. Thus, a majority of
acute stroke care is not provided in the VHA facility. Once a patient has been
stabilized they may be transferred to the VHA medical center or to some other
level of rehab or nursing care. Thus, an outpatient claim may be the first
instance that the stroke event gets captured within the claims system.
Outpatient claims were used to capture those patients whose index stroke
hospitalization occurred in a non-VA facility. Additionally, ICD-9 code 436 is
not routinely used to identify stroke, however, after consulting with stroke
specialists at the VHA, it was recommended for inclusion because it was a code
commonly used to capture strokes in patients for whom they personally did not
provide care. Moreover, based on previous work in identifying neurologic cohorts
from VHA data, the stroke diagnostic codes must occur as the first or second
diagnosis indicating a primary diagnosis. The VHA captures up to ten diagnosis
codes per encounter with codes listed in descending order of their impact on
that episode of care. This method identified a total of 2,299 veterans. Nearly
80% of those identified based on an inpatient encounter had diagnosis codes of
431–434, whereas 89% of those identified based on an outpatient encounter had
only a ICD-9 436 diagnostic code. Thus, ICD-9 code 436 was used when there was
unambiguous evidence of stroke but that index event (acute care inpatient
encounter) was not managed in the VHA facility. Finally, we then looked back
three years for the presence of a stroke diagnosis to identify only veterans
with an index stroke hospitalization in 2003 yielding 523 such cases who
comprise the study group.
Non-psychiatric hospitalizations were defined as inpatient admissions for any
cause except treatment for psychiatric or substance abuse disorders from the VHA
Inpatient data file. Consistent with previous research, non-psychiatric
hospitalizations were considered within the following time intervals from index
stroke discharge: within 30 days, 31 to 90 days, 91 to 364 days, and within
year-long intervals thereafter up to five years. All admissions within 30-days
post-stroke discharge were due to stroke-related complications. Causes of
hospitalization were determined by examining the principle ICD-9 diagnosis code
for each hospitalization. ICD-9 codes were further categorized into eleven
discrete groupings, including: neurologic, cardiovascular, orthopedic,
infections, kidney disorders, endocrine disorders, digestive disorders, cancer,
blood disorders, respiratory disorders, or miscellaneous. SMI was defined in our
sample by ICD-9 codes for schizophrenia (295.00–295.05; 295.10–295.15;
295.20–295.25; 295.30–295.35; 295.40–295.45; 295.50–295.55; 295.60–295.65;
295.70–295.75; 295.80–295.85; and 295.90–295.95), bipolar disorder
(296.00–296.06; 296.10–296.16; 296.40–296.46; 296.50–296.56; 296.60–296.66;
296.70; 296.80–296.82; 296.89–296.90; and 296.81), and major depression
(296.20–296.26 and 296.30–296.36). ICD-9 codes 297, 298, and 300 were also
initially included in our determination of SMI, however, none of these codes
were present in the administrative data over the two-year look-back period.
Likewise, A patient was categorized as having SMI if at least one instance of an
ICD-9 code occurred in the two-year look-back period in either inpatient or
outpatient diagnostic fields. Randomly selected medical charts were reviewed (n
= 20) to validate agreement with this classification algorithm. In this review,
every patient categorized as having SMI was found to have had more than three
previous encounters with the VHA related to SMI.
## Study covariates
defines the covariates, their central tendency, and coding scheme. Demographic
covariates included age (18–44, 45–64, 65–74, or \>75 years), marital status
(married or not married), annual income (in 2002 dollars), health insurance
coverage (dual beneficiary or only VHA benefits), and proximity of patient’s
home to a VA hospital (in miles). Race was coded dichotomously (Caucasian or
non-Caucasian) because 97.8% of non-whites were African American. Sex was
excluded as a covariate because only 3% of the sample were women. History of
physical comorbidity was measured as a composite score using the Elixhauser
Index, which is a method of categorizing 30 comorbidities of patients based on
the ICD-9 diagnosis codes found in administrative data. History of mental health
comorbidity was measured using ICD-9 codes for post-traumatic stress disorder
(309.81), acute depression (311; 308; 309.0; 309.1; and 313.1) and substance
abuse (291; 292, 303; 304; 305; 535.3; 571.1; 648.3; and 790.3). History of
stroke risk factors was measured by presence of ICD-9 codes for hypertension
(401.x; 402.x; 403.x; and 405.x), diabetes (250.0x–250.4x; 250.7x; and 250.9x),
and peripheral vascular disease (440.x; 441.x; 443.1; 443.9; 447.1; 557.1;
557.9; and V43.4) calculated for the year prior to stroke. Counts of 1-year pre-
stroke non-psychiatric hospitalizations may be predictive of future
hospitalizations and were considered as a covariate. Length of hospital stay at
index stroke was measured in days. Quality of stroke treatment was measured
using three of ten indicators identified by the Joint Commission as evidence for
appropriate stroke care. These indicators included: appropriate assessment for
rehabilitation (yes or no), appropriately discharged on a statin (yes or no),
and appropriately discharged on antithrombotic medication (yes or no). SAS
algorithms (Appendix 1) were written to identify ICD-9 diagnosis, procedure, and
inpatient pharmacy administration codes indicative of the Joint Commission
stroke processes of care.
## Statistical methods
Unadjusted bivariate analyses were used to describe the differences between
stroke survivors with SMI compared to those without SMI for all covariates and
hospitalizations over time. Bonferroni corrected significance was used to
minimize the chances of making a Type I error across multiple comparisons with a
p\< 0.0030 needed for statistical significance. Hospitalizations in the VHA
dataset were skewed to the right with substantial zero admissions at each time
interval and had distributions with long right tails. Deviance and Kolmogorov-
Smirnov tests for goodness-of-fit were calculated for negative binomial, zero
inflated negative binomnial, ordered probit and logit, and Poisson distribution
models for each time period examined. The data best fit a Poisson distribution
model for every time period except for between 4 and 5 years post-stroke, where
a negative binomial model had preferable deviance and Kolmogorov-Smirnov scores.
However, the Poisson model for between 4 and 5 years post-stroke had a nearly
equal mean (1.48) and variance (1.63). Therefore, for consistency Poisson
regression was used at every time period with log-linked functions and
corrections for overdispersion to test the hypothesis that SMI is associated
with non-psychiatric hospitalizationsAdditionally, negative binomial regression
model for the time period between 4 and 5 years post-stroke was calculated
yielding equivalent results as the Poisson regression. Confounding between the
variable SMI and other independent variables was assessed by adding and then
removing each independent variable to a Poisson model. Independent variables
that altered the parameter estimate for SMI by greater than ±15% were considered
to be confounding and were removed from final models. All statistical
calculations were performed using SAS version 9.3. The University of Maryland,
Baltimore institutional review board approved the study.
# Results
The study population consisted of 523 patients with an index stroke in fiscal
year 2003, 100 with SMI comorbidity and 423 with no SMI comorbidity. summarizes
the demographic, comorbid, and treatment differences between those with and
without SMI. Patients with SMI comorbidity were significantly more likely to be
younger and without a marital partner, only insured by the VHA, have more
comorbid diagnoses, have more pre-stroke non-psychiatric hospitalizations and
peripheral vascular disease. Patients with SMI did not significantly differ from
those without SMI by race, proximity to hospital, income, history of
hypertension or diabetes, hospital length of stay, or by any of the quality of
stroke care indicators.
Causes of non-psychiatric hospitalizations were examined in the year preceding
stroke and after being discharged from inpatient care for index stroke. In the
year prior to stroke, patients with SMI had significantly (p\<0.05) more
admissions related to cardiovascular (SMI 46.3% vs. No SMI 31.6%) and infectious
(SMI 17.0% vs. No SMI 8.7%) causes, and fewer admissions related to orthopedic
(SMI 1.4% vs. No SMI 10.1%) causes. Patients with SMI had significantly
(p\<0.05) more hospitalizations related to cardiovascular (SMI 42.5% vs. No SMI
31.2%) causes and significantly fewer hospitalizations related to recurrent
stroke (SMI 1.5% vs. No SMI 6.7%) during the first year post-stroke. Patients
with SMI, who were re-hospitalized with a stroke complication within 30 days
after being discharged for index stroke had significantly more cardiovascular
complications (SMI 42.8% vs. No SMI 32.4%), but fewer hospitalizations related
to neurological causes (SMI 18.4% vs. 32.4%). During the period 31 to 90 days
post-stroke SMI patients had significantly more hospitalizations related to
cardiovascular (SMI 40.0% vs. No SMI 25.5%) and infectious (SMI 20.0% vs. No SMI
9.9%) causes. Patients with SMI had significantly more cardiovascular causes
(SMI 37.0% vs. No SMI 26.5%) of hospitalization during the period of 1 to \<2
years post-stroke. There were no significant differences between patients with
and without SMI regarding causes of hospitalization in any time period greater
than two years.
summarizes the differences in hospitalizations for patients with and without SMI
at various time intervals from index stroke. Mean pre-stroke non-psychiatric
hospitalizations were higher (p = 0.0004) among patients with SMI (1.47 ± 0.51)
compared to those without SMI (1.00 ± 1.33) in our sample. Mean non-psychiatric
hospitalizations continued to be significantly higher among patients with SMI
compared to patients without SMI at all the examined time intervals during the
first year after stroke, and for the entire first year considered cumulatively.
In time intervals 1 to \<2 years, 2 to \<3 years, and 3 to \<4 years post-stroke
there was no statistical difference between mean non-psychiatric
hospitalizations among patients with and without SMI. However, by the fourth
year after stroke, patients with SMI have significantly fewer non-psychiatric
hospitalizations compared to those without SMI.
Tables and summarizes the results of the multiple Poisson regression by
transforming beta coefficients from the regression into relative risk
calculations, which allow for interpretation and testing of the hypothesis that
SMI is associated with increased non-psychiatric hospitalizations, after
adjustment by covariates. PTSD exhibited a confounding relationship with SMI
during the time intervals of 91 to 364 days and 3 to \< 4 years post-stroke.
History of depression was also confounding with SMI during the time intervals of
≤ 30 days, 91 to 364 days, and 1 to \< 2 years post-stroke. No other independent
variables exhibited a confounding relationship with SMI. After adjusting for
covariates, there are no longer significant differences in the risk for non-
psychiatric hospitalizations between patients with and without SMI at any time
period after stroke.
While history of SMI was not associated with post-stroke non-psychiatric
hospitalizations after adjustment, a number of other independent variables were
signficantly associated with non-psychiatric hospitalizations. Patients
discharged on antithrombotic medications had a lower risk for re-admission (RR
=.62; 95% CI:.50–.74) for a stroke-related complication within 30-days of index
stroke discharge compared to patients not discharged on antithrombotics. During
the time period 91 to 364 days post-stroke discharge minority patients (RR =
0.66; 95% CI:.46–.91) had signficantly lower risk for non-psychiatric
hospitalizations compared to white patients, while patients with higher
Elixhauser co-morbidity index scores (RR = 1.21; 95% CI: 1.12–1.33) were at
greater risk for non-psychiatric hospitalizations. Additionally, during the time
period 91 to 364 days the longer the initial stroke hospital length stay the
greater risk for non-psychiatric hospitalizations (RR = 1.19; 95% CI:
1.07–1.30). During the time period 1 to \< 2 years post-stroke patients with
higher Elixhauser co-morbidity index scores (RR = 1.13; 95% CI: 1.02–1.24) and
patients with more hospitalizations in the year preceding index stoke (RR =
1.06; 95% CI: 1.01–1.11) were at signficantly greater risk for non-psychiatric
hospitalizations, while patients with a history of substance abuse (RR = 0.50;
95% CI: 0.30–0.81) had lower risk for non-psychiatric hsopitalization.
Additionally, patients discharged on statin medications (RR =.63; 95%
CI:.45–.88\] also had lower risk of non-psychiatric hospitalizaiton. During the
time period 2 to \< 3 years post-stroke only patients with higer Elixhauser co-
morbidity index (RR = 1.11; 95% CI: 1.03–1.26) were at signficantly increased
risk for non-psychiatric hospitalization. During the time period 3 to \< 4 years
post-stroke patients with higher Elixhauser co-morbidity index scores (RR =
1.18; 95% CI: 1.04–1.35) were at significantly greater risk for non-psychiatric
hospitalizations, while patients with at history of depression (RR =.43; 95%
CI:.20–.92) and a history of substance abuse (RR =.49; 95% CI:.25–.98) were at
lower risk for non-psychiatric hospitalizations. Lastly, during the time period
of 4 to \< 5 years post-stroke, patients with higher Elixhauser co-morbidity
index scores (RR = 1.16; 95% CI: 1.02–1.36) and patients with more
hospitalizations in the year preceding index stroke (RR = 1.09; 95% CI:
1.01–1.17) were at signficantly greater risk for non-psychiatric
hospitalizations.
Tables and present results using an approach of fitting a full model and using
data reduction for collinear variables to test the hypothesis that SMI is
associated with non-psychiatric hospitalizations. As a final measure to ensure
accurate interpretation of results, variables except for SMI were removed from
the model if α \>.5 to achieve greater parsimony. For instance, in the time
period of \<30 days only the variables history of SMI, race, history of PTSD,
marital status, income, distance from VA to home, antithrombotic at discharge,
and assessed for rehab were retained in the model, while ten other variables
were excluded because they had α \>.5. This procedure was repeated for each of
the time periods yielding no meaningfully divergent results from a full model
presented in Tables and. The same variables retained statistical significance at
each time period with similar relative risks for non-psychiatric
hospitalization. None of the parsimonious models demonstrated history of SMI was
significantly associated with increased risk of non-psychiatric hospitalization
after controlling for covariates.
# Discussion
The aim of this study was to determine the relationship between SMI and non-SMI
and the risk for non-psychiatric hospitalization in the five years after index
stroke. In unadjusted descriptive analyses, patients with a history of SMI had
significantly higher mean non-psychiatric hospitalizations in the year prior to
index stroke and over the course of the entire first year post-stroke compared
to patients with no SMI comorbidity. In Poisson regression analysis after
adjusting for a broad set of patient characteristics, differences in non-
psychiatric hospitalizations between the two groups were no longer significant.
Patient characteristics included in the regression analysis were age, marital
status, income, insurance status, proximity to hospital, physical co-morbidity,
acute mental health history, hospital length of stay at index stroke, and
quality of inpatient stroke treatment. This analysis highlights that patients
with SMI are medically complex with multiple risk factors for hospitalization,
including significantly higher levels of physical co-morbidity. Patients with
this type of profile, notwithstanding mental illness, are likely to have
increased non-psychiatric hospitalizations. Because individuals with SMI have
extensive co-morbidity, tend to delay care, engage in risky health behaviors,
and do not effectively adhere to prescribed treatment, their physical health and
hospitalization rates are influenced by their mental health status.
Our findings differ with several studies examining the association between
hospital admissions and mental illness in the general population. Saravay and
colleagues found that those with post-stroke depression averaged twice as many
readmissions and spent twice as many days re-hospitalized over a 4-year period.
Similarly, Borckardt and colleagues found that psychiatrically involved
outpatients had higher average readmissions (mean = 1.6) compared to non-
psychiatric outpatients (mean = 1.34) over one year. However, both studies only
report unadjusted readmissions and do not utilize regression techniques to
account for the effects of other patient characteristics on hospitalizations.
In our sample, 43.3% of veterans had health insurance coverage with Medicare or
Medicaid or both. Patients with SMI were significantly more likely (70%) to have
only VHA benefits compared to those without SMI (54%). Those veterans with
Medicare or Medicaid are likely to have received a portion of their care outside
of the VHA system. Since we did not have access to data describing non-VHA care
we are unable to account for the full spectrum of care delivered to stroke
patients in our sample. However, it is more likely that stroke survivors with
SMI received more of their care in the VHA system compared to those without SMI.
Nevertheless, insurance status was not a significant predictor of
hospitalizations in our sample.
It is possible that the results obtained by the regression analysis reflect the
success of longstanding effort to integrate physical and mental health care in
the VHA. Hankin and colleagues indicate that the VHA health care system has been
intentionally designed to effectively manage patients with mental illness. In
the mid-1990s, the VHA health care system launched an extensive reengineering
process with the intent of creating greater integration of primary care with
mental health services. The VHA’s design process recognized and attempted to
correct system-level factors that contribute to poor health outcomes such as
poor access to care, and ubiquitous segregation of medical and mental healthcare
leading to disjointed, uncoordinated, and poor quality care. Several subsequent
studies evaluating the effectiveness of the VHAs care integration have emerged
over the last decade, all of which demonstrate substantial improvement in mental
health quality of care indicators. In the general population (non-VHA),
suboptimal management of comorbidities is widely reported among patients with
SMI, as are unfavorable disparities in morbidity and mortality. In VHA systems
of care, individuals with SMI die 13.8 years earlier than those without SMI,
contrasted with non-VHA systems of care where individuals with SMI die 25 years
earlier. In a comparison of the quality of medical care in VHA and non-VHA
settings between 1990 and 2009, the care delivered in VHA showed greater
adherence to accepted processes of care, greater rates of evidence-based
pharmaceutical therapy, greater mental and primary care provider co-location,
greater collaboration on diagnostic and treatment plans, improved quality
monitoring and better health outcomes. These findings suggest that the VHA’s
model for integrated medical and mental health care reduces disparities in years
of lost life and could partially explain the findings of our study. Future
research should examine the impact of the integrated nature of VHA health care
among patients with SMI to determine the extent to which coordinated care
confers benefits to this population, and whether the VHA model could provide
lessons to improve care for the SMI in health care reform.
It is notable that in the time period of within 30-days post-stroke, both SMI
and non-SMI patients who were discharged on antithrombotic medications had
significantly lower risk for stroke-related readmission. Blood clots complicate
recovery from stroke and antithrombotic medications reduce the formation of such
clots. Our findings validate the use of antithrombotic agents to reduce
unnecessary future morbidity, which may have relevance and implications in the
current landscape of health care reform. Additionally, patients with higher
Elixhauser comorbidity index scores were at greater risk of non-psychiatric
hospitalization after 91 days post-stroke, which is consistent with the results
of several studies. Our finding validates previous work indicating that physical
comorbidity increases risk for hospitalizations after surviving a stroke.
There are several limitations of our study that may have implications on the
interpretation and generalizability of results. First, caution should be used
when generalizing results beyond the VHA because of the unique sociodemographic
and health profiles of the veteran population and the integrated nature of
mental health care within the VHA. Second, the study utilized administrative
data that, as we have noted, has limitations. Third, as mentioned previously,
our data set does not contain health care utilization outside of the VHA system.
Since veterans without SMI were more likely to have Medicare, Medicaid or other
insurances, they would be more likely than the SMI to have utilized health
services that remain unmeasured in this study. Further, since the VHA does not
provide emergency department services for acute stroke, patients being
transported to the hospital in response to a 911 call will be taken to a non-VHA
hospital. We assume that veterans without health insurance, and particularly
those with SMI, who require hospitalization will be transferred as soon as
possible to the VHA. Consequently, examining non-VHA care might yield different
results regarding the relationship between SMI and non-psychiatric
hospitalizations. Fourth, several additional relevant variables were not
available for our analysis. For instance, there were no direct measures of
stroke severity based on clinical measures, rather we relied on index stroke
hospital length of stay as a proxy for severity. Data on lifestyle factors such
as obesity, alcohol use, smoking, and physical activity were also not available.
Additionally, antipsychotic medication usage was not considered in regression
models and may have provided valuable insight into the results. These factors
should be assessed in future studies.
Strengths of the study are 1) the VHA system is one of the largest integrated
and standardized health systems in the United States and regular audits
conducted by clinical specialists are in place to cross-check diagnosis codes
with more robust patient chart review. Moreover, research from as early as 1998
has demonstrated adequate reliability for demographic variables (kappa = 0.92),
diagnosis variables (kappa = 0.39–1.0) and cohort identification in
administrative data in the VHA system. 2) This longitudinal VHA database
provides a good opportunity to examine post stroke utilization for the SMI, in
part because a large percentage of this population has no other source of health
coverage.
# Conclusion
The severely mentally ill present a challenging and costly patient population
for health systems. Although previous studies have examined hospitalizations
among stroke survivors, limited data are available on the burden of SMI on post-
stroke non-psychiatric hospitalizations. Unadjusted results highlight
differences between the SMI and non-SMI stroke survivors, with the SMI
experiencing more non-psychiatric hospitalizations both prior to and up to one
year after their initial stroke. After adjusting for numerous patient
characteristics, we found that patients with SMI, who were receiving services in
the VHA system, were not at higher risk for non-psychiatric hospitalization
during any time period after their index stroke for up to five years, when
compared to patients without SMI. These findings suggest that the integrated
nature of VHA healthcare for patients with SMI may confer benefits to this high
risk population through coordination between mental and physical health care.
Because integrated services for the SMI are considered one of the strengths of
the VHA, our findings are provocative, suggesting that the VHA model may be
useful to other health systems as the country is challenged by an aging
population living with SMI, stroke, and other chronic conditions. Additionally,
the adjusted difference demonstrating that individuals discharged on
antithrombotic medications were at lower risk of re-admission within 30 days
highlight the importance of this quality indicator for effective stroke
management, irrespective of mental illness.
# Supporting information
[^1]: The authors have declared that no competing interests exist.
[^2]: ‡ These authors also contributed equally to this work. |
# Introduction
Facial expressions are critical elements in nonverbal communication, which allow
observers to infer the internal state of others and - in case of negative states
- to be alarmed of impeding danger or be prepared for empathic behavior. The
question is how specific this warning signal can be. Is the observer informed
about an impeding threat in general by suggesting that somebody is experiencing
a negative affective state or can facial expressions point to the specific type
of threat? Clearly, facial expressions can only be specifically perceived if
they are sufficiently distinct.
Pain and disgust are suitable models to address this question because of their
minimal expressive difference, which allow for very critical testing. Both
states are elicited by harmful or potentially harmful stimuli (that can be
threatening to our physical integrity) and are characterized by strong feelings
of unpleasantness and both states result in defensive behavior. However, despite
of these similarities, the respective threat to the body and the resulting
subjective experiences are fundamentally different. Are facial expressions
during the experience of pain and disgust specific enough to capture the
distinctness of these two negative affective states? When looking at previous
findings, there is on the one hand evidence that would suggest high distinctness
of facial expressions of pain and disgust whereas on the other hand some
findings favor the assumption of expressive overlap between the two facial
expressions.
Evidence for the latter can be found in research that focused on the facial
muscle movements elicited during pain and disgust. Using the Facial Action
Coding System (FACS) – which is considered to be the gold standard in facial
expression research – it has been shown that the experience of disgust elicits
contraction of the eyebrows (Action Unit (AU) 4), nose wrinkling (AU9), upper
lip raise (AU10), jaw drop (AU25/26/27), raising of the chin (AU17) and
narrowing the eyes (AU6/7). Amongst these facial movements, the activity of the
musculus levator labii superior (which leads to the upper lip raise and nose
wrinkling) seems to be the most central one whereas for the other movements
there is no complete consensus between studies. Interestingly, pain similarly
elicits contraction of the eyebrows (AU4), nose wrinkling (AU9), upper lip raise
(AU10) and narrowing the eyes (AU6/7). In addition, closing the eyes for more
than 0.5 seconds (AU43) has also been shown to occur in the context of pain.
Thus, there seems to be a great overlap between the facial movements that are
elicited during pain and disgust, which might challenge the idea of distinctness
between the two facial expressions. In line with this, observers often confuse
facial expressions of pain with those of disgust which might not be surprising
given the similarity in facial ovements involved in the two expressions.
However, a certain degree of expressive overlap or confusion of both states by
observers would not necessarily preclude sufficient distinctness of facial
expressions to act as specific warning signals. Especially since observers can
distinguish facial expressions of pain and disgust clearly above chance level,
there is some evidence for sufficient distinctness of these two facial
expressions. Furthermore, the overlap might be in fact less than it seems to be
due to the characteristics of the material used for disgust induction. Disgust
has often been induced using pictures and films that also contain pain-related
content, i.e. individuals with injuries or pictures of ripped off limbs,
mutilations, etc.. Hence, it is difficult to decide whether facial expressions
elicited by these confounded stimuli are specific for disgust or are also
representative for pain.
So far, the overlap in facial expressions between pain and disgust has not yet
been systematically studied. The study would be worthwhile just because of the
proximity of the two facial expressions and the risk of overlap. If the facial
expressions are distinct in these two cases they are in other cases too.
No study has yet investigated these two facial expressions in one sample; which
is the only way to test whether these two facial expressions are – despite
similarities – still distinct in the same person (intra-individual comparison).
Thus, the aim of the present study was to investigate whether individuals
facially encode the specific emotional quality of pain and disgust (leading to
distinct facial expressions) or whether they simply encode the similar negative
valence and arousal of both states (leading to similar facial expressions). We
assessed facial and subjective responses to pain (induced by heat stimuli) and
to disgust (induced by pictures) in one sample and used disgust stimuli with and
without pain-related content to be able to compare facial expressions elicited
by “pure” disgust and by disgust due to pain-confounded contents. Based on
previous findings, we hypothesized that pain and disgust elicit a very similar
set of single facial muscle movements. However, given that observers can
differentiate between pain and disgust expressions above chance level, we
expected that the distinctness between the two facial expressions might rely on
different combinations of these single facial movements.
# Methods
## Participants
Sixty healthy volunteers (30 males and 30 females, mean age 22.9 ± 4.3), were
recruited via advertisements posted in the university buildings of the
University of Bamberg. Exclusion criteria were current experience of acute or
chronic pain, psychological illnesses (especially any kind of anxiety disorders)
and physical illnesses. None of the participants had taken analgesics,
psychotropic medication or alcohol the day before testing.
### Ethics statement
All participants provided written informed consent and received monetary
compensation (25 €) or course credits for their participation. The study
(including the consent procedure) was approved by the ethics committee of the
University of Bamberg.
## Materials and procedure
### General protocol
The study was composed of two experimental blocks: a pain induction block and a
disgust induction block. The order of blocks was balanced across participants,
with half of the participants starting with the pain induction block, whereas
the other half started with the disgust induction block.
In the pain block participants received heat stimulation of non-painful and
painful heat intensities. In the disgust block participants viewed pictures of
different emotional content: neutral (as fillers), positive (to counteract an
unspecific lowering of mood), disgust due to pain-related content (e.g.
mutilation) and “pure” disgust pictures (e.g. bodily excrements). In both the
pain and the disgust blocks participants were seated in front of a 19-inch
computer screen positioned 50 cm in front of them. Each target stimulus (heat or
picture stimuli) was preceded by a fixation cross (duration 1s - 3s) to orient
eye gaze to the center of the screen and to minimize eye and head movements for
later off-line analyses of facial display. The fixation cross was followed by 5
s of heat stimulation (pain block) or by 5 s of picture presentation (disgust
block), respectively. After stimulus offset participants had to rate their
experiences on separate scales. To familiarize participants with the rating
procedures, two familiarization trials were conducted at the beginning of each
block. During the whole session, which lasted almost 2 hours, participants sat
upright in a comfortable chair.
### Pain block
Following a previous protocol that has been shown to successfully elicit facial
expressions of pain, pain was induced by use of a Peltier-based, computerized
thermal stimulator (Medoc TSA-2001; Medoc Ltd, Ramat Yishai, Israel) with a 3 ×
3 cm<sup>2</sup> contact probe attached to the outer part of the left lower leg
(midpoint between ankle and knee). To ensure that painful stimuli were indeed
perceived as being painful without exceeding individuals pain tolerance, we
adjusted stimulation temperature to the individual pain threshold. Thus, heat
pain thresholds were determined first, using the method of adjustment.
Participants were asked to adjust pain threshold starting from 38 °C, using
heating and cooling buttons (rate of change: 0.5 °C/s), until they obtained a
level which was felt as barely painful. This procedure was repeated in 4 trials,
the first trial was a familiarization trial. The threshold estimate was the
average of the last 3 trials. Thereafter, the “pain block” started. Here, 10
painful stimuli (pain threshold +3°C) and 10 non-painful heat stimuli (pain
threshold -3°C) were applied in randomized order. The non-painful heat stimuli
were applied to have a neutral reference for facial expression analyses. The
painful intensity of +3°C above threshold was chosen since this intensity
elicits a painful sensation of mild to moderate intensity and thus, the arousal
and valence of this painful intensity should be comparable to the disgust
pictures we selected. The temperature increased (rate of rise: 4°C/s) from
baseline (38°C) to these pre-set temperatures, was kept constant for 5 s
(plateau phase) and returned to baseline. Facial as well as subjective responses
(ratings) to each stimulus were assessed.
### Disgust block
Color pictures of emotional content (800x600 format) were presented for 5
seconds on the computer screen in a randomized order. Pictures were mostly
selected from the International Affective Picture System. (Those IAPS pictures,
listed by their identification numbers, are as follows: neutral – 219, 1731,
2745.1, 5551, 5720, 5800, 5900,7009, 7041, 7052; pain-disgust – 3101, 3150,
3261, 3400, 7361, 8230, 9405; disgust – 9301, 9320; happy – 1340, 1440, 1441,
1710, 1750, 2091, 2165, 2501, 4625, 7325). Since the IAPS does not include a
sufficient number of disgust pictures, 12 pictures were selected from the
internet: 3 pain-disgust pictures (content: decubitus, open fracture, suppurated
sore) and 8 “pure” disgust pictures (content: moldy toast, rotten teeth,
excrements, a woman vomiting, a man vomiting, snot, vomit in a toilet, infected
toenail, spitting person). We made sure that the pictures taken from the
internet matched the IAPS pictures with regard to arousal and valence and thus,
asked 40 individuals in a pilot study to rate the IAPS and non-IAPS pictures. No
difference was found between non-IAPS and IAPS pictures (non-IAPS pictures:
valence: 3.4 (SD 1.2), arousal: 5.7 (1.3); IAPS pictures: valence: 3.2 (SD 1.1),
arousal: 5.9 (SD 1.2); all p-values\>0.05). We also included neutral pictures
from the IAPS to have a neutral reference for facial expression analyses (as we
did with the non-painful heat intensities in the pain block). The picture set
also included pictures with happy content, which were only presented to avoid a
lowering of mood by only showing negative pictures, but were excluded from
further analysis.
## Assessment of facial responses
Faces of the participants were continuously videotaped throughout both testing
blocks using a camcorder (JVC GZ-MG30) mounted on top of the computer screen.
Subjects were instructed not to talk during the experimental blocks. A LED
behind the subject, visible to the camera, but not to the participant, was
lighted concurrently with the (thermal stimuli or pictures) to mark the on- and
offset of the stimulation.
Facial expressions were coded from the video recordings using the Facial Action
Coding System, which is based on anatomical analysis of facial movements and
distinguishes 44 different Action Units (AUs) produced by single muscles or
combinations of muscles. Two FACS coders identified frequencies and intensities
(5-point scale) of all Action Units that occurred during the stimulation (inter-
rater reliability =.90 tested in a sub-set (15 %) of the video segments). To
segment videos and to enter the FACS codes, we used the Observer Video-Pro
(Noldus Information Technology). Time segments of 5 s, beginning just after
stimulus onset (in case of the heat stimuli, stimulus onset was defined as start
of the 5 second plateau phase), were selected for scoring (the onset of a facial
action had to lie in this time window in order to be scored). In total, 20 video
segments for thermal stimulation (10 painful and 10 non-painful) and 30 segments
of picture presentation (10 neutral, 10 pain-disgust and 10 “pure” disgust) were
analyzed per subject. As has been done in preceding studies (especially on
facial responses to pain) we combined those AUs that represent similar facial
movements (AU1/2, AU6/7, AU9/10 and AU25/26/27).
## Self-Report
After stimulus offset, participants were asked to rate their sensations on four
separate scales. Two of these scales were Visual Analogue Scales (VAS; which
appeared simultaneously on the screen) which assessed pain and disgust
intensity, respectively. Participants were asked to rate both intensities by
moving a cursor on the 100 mm VAS scales with the endpoints “no pain” and
“extremely strong pain” or “no disgust” and “extremely strong disgust”,
respectively. The cursor appeared in random positions to avoid response
tendencies due pre-selection of scale ranges. Participants had 16 s to provide
both VAS ratings. Following the VAS ratings, participants were asked to rate the
valence and arousal of the pain and disgust stimuli using Self Assessment
Manikins (SAM) scales that appeared on the computer screen (rating was done by
mouse click on the manikins or spaces in-between, resulting into 9 categories).
Both SAM scales appeared simultaneously on the screen and participants had 12
seconds to provide their ratings. To familiarize subjects, two practice trials
were conducted in each block.
## Statistical Analyses
### Ratings of pain and disgust (VAS, SAM)
To investigate whether we succeeded in inducing painful experiences in the pain
block and disgust experiences in the disgust block, respectively, we compared
VAS ratings between the different types of stimuli (painful heat vs. pain-
disgust pictures vs. “pure” disgust pictures) using a multivariate analysis of
variance (VAS pain and VAS disgust) with repeated measurement. To investigate
whether the different types of stimuli elicited comparable levels of valence and
arousal ratings, we again used a multivariate analysis of variance (SAM valence
and SAM arousal) with repeated measurement. If a MANOVA revealed significant
effects, post-hoc T-Tests (bonferroni-corrected) were conducted for single
comparisons.
## Facial responses
To investigate whether facial responses elicited during pain and disgust
induction differ from each other we used a two-step approach. In a <u>first
step</u>, we analyzed which individual facial actions are elicited during pain
and disgust (step 1a) and whether the strength with which these individual
facial actions are displayed (frequency and intensity values) differs between
the affective states (step 1b). However, given that facial expressions are not
only characterized by single facial actions but more importantly by the specific
combination of facial actions, we compared in a <u>second step</u> the
occurrence of facial action combination during pain and disgust induction.
### Step 1a: Which individual facial actions are displayed during pain and disgust experiences, respectively?
First, we wanted to analyze which single Action Units (AUs) are relevant for
pain, pain-disgust and “pure disgust” expressions, respectively. For that
purpose, we calculated which AUs occurred with a frequency of at least 5% in the
segments recorded (this was done separately for pain, pain-disgust and “pure”
disgust stimuli; see for details). This critical level of 5 % was derived from
earlier studies. Furthermore, in order for an AU to be classified as being
relevant for the respective state, it also had to occur more often during the
respective state compared to neutral conditions (neutral pictures or non-painful
intensities, respectively). To determine this, effect sizes (Cohen’s *d*) for
frequency differences in AUs between pain and non-painful heat, between “pure”
disgust and neutral pictures as well as between pain-disgust and neutral
pictures were computed. This procedure has been proven to be successful in
former studies \[20,24.25\] to identify AUs that are relevant for specific types
of affective states. AUs which showed an effect size d ≥.05 (medium effect) were
selected as pain- or disgust-relevant facial responses (these AUs are displayed
bold).
### Step 1b: Are these individual facial actions that encode pain and disgust displayed with similar strength during pain and disgust induction?
In a next step, we analyzed whether those single facial actions that are used to
encode pain as well as pain-disgust and “pure” disgust are displayed with
similar strength during pain and disgust induction. The strength of each single
AU was computed by forming product terms (multiplying the frequency and
intensity value of each individual AU). These product terms were then entered
into a multivariate analysis of variance with repeated measurement. If the
MANOVA revealed significant effects, post-hoc T-Tests (bonferroni-corrected)
were conducted for single comparisons.
### Step 2: Which combinations of single facial actions occur during pain and disgust induction?
In order to compare combinations of facial actions between pain, pain-disgust
and “pure disgust” expression, we focused on those AUs that proved to be
relevant for all three affective states (see results of step 1a). These were the
AUs 4, 6/7 and 9/10, allowing for combinations of 3 elements at maximum. We
assessed in which combinations these selected AUs were displayed during pain and
during disgust induction and then used chi-square analyses to compare the
frequency of facial action combinations between pain, pain-disgust and “pure”
disgust expressions.
Statistical analyses were run by means of the statistic software SPSS 21.0.
Findings were considered to be statistically significant at p \< 0.05.
# Results
## Ratings of pain and disgust (VAS, SAM)
VAS: The MANOVA revealed that VAS intensity ratings differed significantly
between the different types of stimuli (painful heat vs. pain-disgust pictures
vs. “pure” disgust pictures) (F(4,236)=234.00; p\<0.001). As the univariate
results showed, these differences were significant for the VAS pain ratings
(F(2,118)=331,86; p\<0.001) as well as for the VAS disgust ratings
(F(2,118)=146.47; p\<0.001). As expected, pain stimuli were rated to be more
painful compared to the disgust pictures (both pain-disgust and “pure” disgust;
p-values of the post-hoc comparisons: p\<0.001; see also). Moreover, the pain-
disgust pictures were also rated to be more painful compare to the “pure”
disgust pictures (p-value of the post-hoc analysis: p\<0.001). With regard to
the VAS disgust ratings, participants rated the disgust pictures (both pain-
disgust and “pure” disgust) to be more disgusting compared to the pain stimuli
(p-values of the post-hoc comparisons: p\<0.001; see also). VAS disgust ratings
of the two picture categories (pain-disgust and “pure” disgust) did not differ
significantly (p-value of the post-hoc analysis: p=0.151).
SAM: The MANOVA revealed that the different types of stimuli (painful heat vs.
pain-disgust pictures vs. “pure” disgust pictures) elicited significant
differences in the SAM ratings (F(4,236)=7.33; p\<0.001). As univarate results
revealed, both valence (F(2,118)=10.30, p\<0.001) as well as arousal ratings
(F(2,118)=12.40; p\<0.001) changed significantly depending on the types of
stimuli. However, as post-hoc comparisons revealed, these differences were only
due to pain-disgust pictures being rated to be more arousing and more negative
in valence than the pain and “pure” disgust stimuli (all p-values \<0.05);
whereas pain stimuli and “pure” disgust pictures elicited comparable levels of
arousal (p=0.562) and valence ratings (p=0.989) (see also).
*In summary*, our pain and disgust inductions produced similar levels of
negative valence and arousal because even though the pain-disgust pictures
elicited higher levels of valence and arousal compared to pain and “pure”
disgust stimuli, the mean difference in descriptive values between the three
types of stimuli was always lower than 1-scale-point on the SAM scales. In
contrast, pain and disgust inductions produced clear differences in pain and
disgust VAS ratings and thus, participants appeared to be completely aware of
the particular origin of the affective state. Accordingly, we produced
conditions that will allow us determining whether facial responses only indicate
general valence and arousal of affective states or whether they reflect the
specific type of affective state.
## Facial expression of pain and disgust
### Step 1a: Which single facial actions are displayed during pain and disgust experiences, respectively?
lists all those AUs that occurred above a critical occurrence level of 5% during
pain, pain-disgust and “pure” disgust induction, respectively. As can be seen,
pain induction elicited more single facial actions compared to the disgust
induction procedures. However, when focusing on those facial actions that proved
to be relevant for each of the three types of affective states (medium effect
size for the difference to the respective control conditions) there is a great
overlap between pain and disgust. As can be seen in, the facial actions “brow
lowering” (AU4), “orbit tightening” (AU6/7) and “levator contraction” (AU9/10)
encoded pain as well as the two types of disgust experiences. The only
difference between affective states was that pain experience was additionally
accompanied by “smiles” (AU12) and “mouth opening” (AU25/26/27) whereas “pure”
disgust also led to “brow raising” (AU1/2). Thus, pain and disgust experiences
were accompanied by mostly the same facial actions (see also were examples of
facial expressions are given).
### Step 1b: Are these single facial actions that encode pain and disgust displayed with similar strength during pain and disgust experiences?
The MANOVA revealed that those single facial actions that encode both pain and
disgust are displayed with different strength during pain and disgust induction
(F(6,234)=9.39; p\<0.001). As the univariate findings showed, AU4
(F(2,118)=19.63; p\<0.001) as well as AU6/7 (F(2,118)=9.11; p\<0.001) were
displayed with different strength depending on the type of affective state,
whereas AU9/10 did not differ between affective states (F(2,118)=0.95; p=0.390).
When conducting post-hoc analyses for single comparisons we found that AU 4 was
displayed less strongly in response to pain induction compared to the two
categories of disgust pictures (both p-values \<0.010, see also). Moreover, AU
4 was displayed more strongly in response to the pain-disgust compared to the
“pure” disgust pictures (p\<0.001). With regard to AU 6/7, this facial action
was displayed most strongly in response to pain induction compared to disgust
induction (pain-disgust and “pure” disgust pictures; both p-values \<0.050, see
also) and did not differ between pain-disgust and “pure” disgust (p=0.686) (see
also were examples of facial expressions are given).
Given the small but nevertheless significant differences in arousal ratings
between affective states (with pain-disgust pictures being rated as being more
arousing compared to the other two affective states), we wanted to ensure that
our findings on facial responses are indeed not affected by these differences in
arousal. To control for this, we divided the group of subjects into those who
rated the pain-disgust pictures as more arousing than the other two affective
states and those who rated them as equally arousing (median split of the
averaged difference scores). We found that the findings on facial responses (as
displayed) were not affected by differences in arousal ratings between affective
states, given that the same findings were obtained in those individuals who
rated the affective states as being differently arousing and in those
individuals who rated the affective states as being equally arousing.
### Step 2: Which combinations of single facial actions occur during pain and disgust induction?
As can be seen in, we found that in more than half of the segments (both in the
pain and in the disgust block) facial expressions were only composed of one
single facial action. Especially lowering of the brow (AU4) and orbit tightening
(AU6) were often displayed alone (indicated by being combined with “∅”). The
most frequent facial action combination was the lowering of the brow together
with orbit tightening (combination AU4, AU6/7) both during pain and disgust
induction. Despite these similarities, chi-square tests (goodness of fit)
revealed significant differences in facial action combinations between pain and
“pure” disgust (χ<sup>2</sup>(6)=20.64; p=0.002) as well as between pain and
pain-disgust (χ<sup>2</sup>(6)=22.75; p\<0.001) induction. In contrast, the two
disgust categories (“pure” and pain-disgust) elicited a similar distribution of
facial action combinations (χ<sup>2</sup>(6)=5.72; p=0.46). The standardized
residuals (stand. res.) - which indicate the importance of each cell to the chi-
square value - revealed that brow lowering (AU4) by itself (stand. res.
3.1/3.1), levator contraction (AU9/10) by itself (stand. res. 2.5/1.3) and the
combination of these two facial actions (stand. res. 1.6/2.3) occurred markedly
more frequently during disgust (“pure” disgust/pain-disgust) compared to pain
induction. In contrast, orbit tightening (AU6/7) by itself (stand. res. 3.3/5.7)
and the combination of orbit tightening and brow lowering (AU4) (stand. res. 2.8
for the difference to “pure” disgust) occurred markedly more frequent during
pain compared to “pure” disgust/pain-disgust induction.
*In summary*, pain and disgust experiences seem to elicit the same single facial
actions. However, the strength, with which these single facial actions are
displayed during pain and disgust differs significantly, even when controlling
for differences in arousal ratings. Moreover, when considering combinations of
single facial actions, we also found clear differences between pain and disgust.
Whereas brow lowering (AU4), levator contraction (AU9/10) and the combination of
these two facial responses are more pronounced during disgust, pain seems to be
encoded by a more pronounced orbit tightening (AU6/7) (often in combination with
brow lowering (AU4)).
# Discussion
The aim of our study was to investigate whether facial displays occurring during
the experience of pain and of disgust overlap substantially and
indistinguishably or whether they are distinct enough to communicate different
states. Pain and disgust were selected for this comparison because both are
elicited by actual or potential harm to the body, are characterized by strong
feelings of unpleasantness, result into defensive behavior and facial
expressions of both states are often confused by observers. Facial responses to
pain and disgust are ideal models to test the distinctiveness of facial
expressions because of their expressive similarity, which guarantees - in case
of proven differences in facial encoding - differential encoding also for most
of the other emotional states. It has to be kept in mind that the strong
similarity between the two states exists, although the subjective experience
clearly reflects the specific type of threat and allows for differential self-
reports.
Our main findings were that nearly the same single facial actions were elicited
during the experience of pain and disgust. However, these facial actions were
displayed with different strength and were differently combined depending on
whether pain or disgust was experienced. We will discuss these findings in more
detail below.
As stated above, we found that nearly the same single facial muscle movements
were elicited during the experience of pain and disgust. More precisely, the
contraction of the eyebrows (orbicularis oculi muscle; AU 6/7), contraction of
muscles surrounding the eyes (corrugator muscle; AU4) and lifting the upper lip
(levator muscle; AU9/10) were observed both during pain and disgust induction.
This overlap in facial actions is well in line with previous findings which have
also reported similar facial actions in response to pain and disgust induction
(although this is the first study to directly compare facial responses to pain
and disgust in one sample). Given that we induced both pain and disgust in one
sample we were also able to directly compare the strength to which each of these
facial actions was displayed during pain and disgust experiences. (For note, the
ratings of valence and arousal were very similar during disgust and pain.). We
found that although contraction of the muscles surrounding the eyes (AU6/7)
occurred both during pain and disgust, this facial action was more relevant for
pain than it was for disgust (see also examples displayed). This is in line with
previous assumptions that the orbicularis oculi activation is the most prominent
facial response to pain. In contrast, the contraction of the eyebrows occurred
more strongly during disgust compared to pain induction. Moreover, facial
expressions of pain and disgust could be even better differentiated when
considering how these single facial actions were combined. The most frequent
facial response that can be observed when an individual is experiencing pain
seems to be the contraction of the muscles surrounding the eyes (AU6/7), either
by itself or in combination with contraction of the eyebrows (AU4), which
accounted for nearly 2/3 of all facial responses to pain in the present study.
Facial expressions of disgust, on the other hand, seem to be more variable than
pain expressions. Most often contraction of the eyebrows (AU4) is displayed by
itself. In contrast to pain, disgust is also more often encoded by raising the
upper lip/wrinkling the nose (AU9/10) by itself as well as by the combination of
eyebrow contraction and upper lip raise (see also for examples of facial
expressions). Consequently, although the same facial actions are used to
facially encode pain and disgust, these facial actions are differently combined
and displayed with different strength during pain and disgust experiences, thus,
suggesting that the facial displays can indeed signal categorically distinct
states although similar muscles are involved.
The literature on decoding of affective states by observers has introduced two
perspectives a discrete-category and a dimensional view, which are worthwhile
being considered here; although our experiment dealt with the encoding of facial
displays by senders and not with the decoding by receivers. According to the
dimensional view facial expressions mainly convey values on the dimensions of
valence and arousal. Given that pain and disgust elicit very similar valence and
arousal ratings one would expect – based on the dimensional view – also very
similar facial responses to pain and disgust. In accordance with this
expectation (namely that facial responses mainly convey valence and arousal
information), we indeed found a great overlap of single facial actions that
encode both pain and disgust. In other words, the type of single facial actions
being displayed might convey mainly information on valence and arousal to the
observer. However, we also found that despite the great overlap of single facial
actions, facial responses to pain and disgust are also able to signal
categorically discrete states. This is in line with the discrete-category view
hypothesized that reading the facial expression of emotions results into clearly
differential classifications because each emotion has its specific facial
readout. Our data support such a hypothesis with regard to the affective states
pain and disgust. When we took into consideration not only the type of single
facial actions but the strength to which they are displayed as well as how they
are combined, facial expressions of pain and disgust were clearly distinct.
Thus, the types of single facial actions being elicited seem to convey
information on valence and arousal (dimensional view) whereas the strength and
the combination in which they are displayed seem to form the basis for
communicating differentially emotional states, allowing for their identification
by an observer (discrete-category view). Given that the differential
classification of disgust and pain is amongst the most difficult ones, our data
also corroborate a more general conclusion.
We used two different picture categories to induce disgust, namely pictures
showing pain-related content (e.g. mutilation) as well as pictures without any
pain-related content (e.g. body waste products) to investigate whether the
previously reported overlap between facial responses to pain and disgust might
be simply due to facial expressions of disgust having been elicited by pain-
confounded stimuli. We found no clear differences between facial expressions in
response to pain-disgust and “pure” disgust pictures. Moreover, facial responses
to the pain-disgust pictures were not more similar to pain than the “pure”
disgust responses. Thus, our data clearly suggest, that this confound of disgust
with pain-related content is not responsible for the great overlap in facial
actions elicited during the experience of pain and disgust.
As a limitation of the present study, it has to be mentioned that the methods
used to induce pain and disgust were quite different. Whereas for the pain
induction procedure, a physical stimulus was used to directly induce the
affective state “pain”, we induced the affective state “disgust” by use of
picture stimuli, which are only icons of the actual threat to the body. The
reason for using pictures to induce disgust was that we wanted to include both
“pure” disgust and pain-confounded disgust stimuli, which cannot be all made
available as physical stimuli. As explained above, we did this to control - for
the first time - for this confound of pain and disgust contents in some of the
IAPS designated disgust pictures.
In summary, investigating facial and subjective responses to pain and disgust
induction in one sample of participants revealed that the single facial actions
elicited during the two states are rather similar than different. This
similarity is paralleled by the similarity of the valence and arousal ratings.
However, when considering the strength with which each of these single facial
actions are displayed and how these single facial actions are combined during
the experience of pain and disgust, significant differences occurred.
Consequently, facial expressions of pain and disgust seem distinct enough to
also encode the specific type of the threat to the body. Thus, facial
expressions seem to be able to signal both, the general valence as well as
arousal and the specific threat to the body. This implies that the differential
decoding of these two states by an observer is possible without additional
verbal or contextual information. This is of special interest for clinical
practice, given that raising awareness in observers about these distinct
differences could help to improve the detection of pain in patients who are not
able to provide a self-report of pain (e.g. patients with dementia). Moreover,
in future studies the question should be answered how the decoding of the facial
activity during pain and disgust is affected when systematically varying the
strength and combination of single facial actions (e.g. by using animated avatar
facial expressions). The findings of these studies may confirm that the features
we found in the present study are indeed crucial for the identification of pain
and disgust by observers.
[^1]: The authors have declared that no competing interests exist.
[^2]: Conceived and designed the experiments: MK SL. Performed the
experiments: JP SH. Analyzed the data: MK JP. Contributed
reagents/materials/analysis tools: MK. Wrote the manuscript: MK JP SH SL. |
# Introduction
Accurate partners or research followers are imperative in scientific research.
Mining deeper author relationships in the academic network involving various
significance is achievable, which can help scholars establish potential
cooperative or reference relationships. The research visual field can also be
expanded, and the research content can be deepened. The establishment of a
citation relationship among scholars is mainly based on the correlation of their
research contents. If this relationship is deeply mined, potential partners
could be found. Given that the citation data were preserved completely and
accurately in document database, the processes and results of relationship
mining would be feasible and reliable. As a mature quantitative research method
in bibliometrics and scientometrics, citation analysis is extensively used in
scientific evaluation, scholarly communications, academic behavior analysis, and
information retrieval. Author citation analysis mainly includes three types:
author co-citation (AC), author bibliographic coupling (ABC), and author direct
citation (ADC), which is collectively called “tripartite citation analysis” in
this study. For example, in a field, both papers of Authors A and B were cited
by the same paper; thus, A and B have a co-citation relationship marked as AC
(A, B). Authors C and D both cite the same paper in their respective articles; C
and D thus have a bibliographic-coupling relationship marked as ABC (C, D). In
addition, Author D cites a paper written by A in his bibliography, or vice
versa; thus, D and A have a direct-citation or cross-citation relationship
marked as ADC (A, D).
On mining author relationship in scholarly networks based on tripartite citation
analysis, two key questions should be addressed.
1. Which bibliographic-coupled or co-cited authors did not collaborate yet
or do not cite each other regularly? If we called these relationships as
potential communication relationship (PCR), and the latter two as actual
communication relationship (ACR), could the discovery and usage of PCR
contribute to the achievement of the ACR? Furthermore, how is the
quantitative relation of PCR and ACR? This concern is the first point to be
investigated in this study.
2. In view of the similarities or diversity among tripartite citation
relationships at the author level, how can tripartite relationships be
synthetically used in discovering deeper author relationships serving for
broader scholarly communication and relevant recommendations? According to
these primary relationships, deducing the integrated relationships between
Authors A and C, or B and D, even B and C, the association strength in these
potential relationships is the second point to be answered in this study.
# Related studies
## Separate study of tripartite citation analysis
AC analysis is the most commonly used method for the empirical analysis of
disciplinary paradigm, and has been frequently studied and improved. Some AC
analyses have been conducted since Small introduced document co-citation
analysis and White and Griffith developed AC analysis. Bibliographic coupling
was proposed as early as 1963. However, author coupling relationship has not
gained considerable attention until it was formally proposed and empirically
studied by Zhao and Strotmann; the authors named this method ABC analysis, which
can be used to complement AC analysis in comprehensively viewing the
intellectual structure by mapping the research activities of active authors for
a realistic picture of the current state of research in a field.
In comparison with co-citation and bibliographic coupling, direct citation
(sometimes also called inter-citation or cross-citation) is a direct citation
relationship without a third-party paper. Paper-level direct citation has been
used in different scenarios, such as research front detection, domanial
historiography mapping, and publication classification. Boyack and Klavans found
that bibliographic coupling slightly outperforms co-citation analysis and that
direct citation is the least accurate science mapping. Shibata et al. revealed
that direct citation could detect large and newly emerging clusters earlier,
indicating that the research front detection exhibited the best performance,
whereas co-citation showed the worst. Numerous studies have focused on journal
direct citation; several key research achievements have shown that journal
direct citation can reveal the academic influence of journals, as well as the
theme evolution and field division of periodicals. Direct citation can also be
used at the macrolevel, such as citation between subject categories, to build
the global map of science. Wang et al. extensively studied ADC analysis, which
can be used to determine author relationship from another angle and reveal the
knowledge communication and disciplinary structure in scientometrics. This
process was then named as ADC analysis by Yang and Wang.
## Comparative study of tripartite citation analysis
The three types of citation analysis methods can reveal author relationships in
a field in various ways. Some studies have focused on the comparative analysis
of these methods, even comparing them with other author co-occurrence network
analysis methods, such as co-authorship (CA), word-based author coupling (WAC),
and journal-based author coupling (JAC). Related studies are especially
represented by Lu, Yan, and Qiu et al. Lu and Wolfram conducted a comparative
study of word-based, topic-based, and author co-citation approaches to measure
author research relatedness. Findings show that two word-based approaches
produced similar outcomes, except in the case in which two authors were frequent
co-authors for the majority of their articles, and that topic-based approach
produced the most distinctive map. Yan and Ding explored the similarities among
six types of scholarly networks (bibliographic-coupling, direct citation, co-
citation, topical, co-authorship, and co-word networks) aggregated at
institution level; they also detected high similarity between co-citation and
direct citation networks. Moreover, the authors recommended the use of hybrid or
heterogeneous networks to study research interaction and scholarly
communications. Qiu and Dong constructed five types of author co-occurrence
networks in the field of information library sciences, such as CA, WAC, JAC, AC,
and ABC. In their research, the capabilities of different types of author co-
occurrence relationships in revealing scientific structure are compared through
hierarchical clustering and correlation analysis by quadratic assignment
procedure (QAP) test. ABC analysis also exhibited a significant advantage in
revealing discipline structure and presented the highest correlation with other
networks. The idea of combining different author co-occurrence networks in
scholarly communication and intellectual structure analysis was also proposed.
## Combined study of tripartite citation analysis
The combination of these tripartite citation analysis methods (including AC,
ABC, and ADC) has been extensively studied. Small proposed a method for
effectively combining them; however, only few researchers have adopted this
combined linkage technique at a large scale. Persson and Gómez-Núñez et al. have
attempted to combine these citation measures in a normalized manner to weigh
existing direct citation relationships between articles or journals. According
to Persson’s research, direct citations weighted with shared references
(bibliographic coupling) and co-citations at the article level could be better
applied to domain intellectual structure detection. In addition, citation-based
measure calculation and integration (involving co-citation, bibliographic
coupling, and cross citation) at the journal level was also proposed and proven
in the application of refining the journal classification, improving journal
ranking, and further updating the subject classification structure proposed by
Gómez-Núñez et al. At the author level, Wang proposed a comprehensive and
comparative approach by combining CA, AC, ABC, ADC, and author keyword coupling
(AKC), supplemented by social network analysis (SNA), to evaluate the academic
impact of the core authors in the field of scientometrics. Existing studies are
focused on intellectual structure detection and optimization according to
tripartite citation analysis. The assessment of the author scholarly impact by
combining various citation analysis is also paid attention in few studies.
## Mining author relationship in scholarly networks
Practical research on the discovery of potential author relationships in
communication networks by tripartite citation analysis is limited. Currently,
approaches for identifying potential collaboration mainly involve machine-
learning techniques, link-prediction techniques, and SNA. Zhang and Yu proposed
supervised machine-learning approaches to predict research collaborations by the
semantic features in the field of biomedicine and author network topological
features, including co-authorship network connectivity, research profile
similarity, collective productivity, and seniority. Chen and Fang developed a
latent collaboration index model for evaluating the collaboration probability
among patent assignees by incorporating two network-related factors (i.e.,
degree and network distance) and complementary factors (i.e., assignees types,
geographical distances, and topic similarities). Guns and Rousseau introduced a
method for predicting or recommending high-potential future collaborations based
on a combination of link prediction and machine-learning techniques. Daud et al.
used discriminative and generative machine-learning techniques for predicting
the emerging scholars in a co-author network based on three classes of features
(i.e., author, venues, and co-authorship). These studies have focused on the use
of combined relationships of direct citation and co-authorship in scholarly
networks without considering other relation networks in discovering potential
collaboration.
# Data and methodology
## Basic data
Scientometrics is an international journal, launched in 1978. The journal covers
all aspects of scientometrics and published 46.31% of scientometrics research
paper of the world. Given the Scientometrics Journal as the representative
communication channel in the field of scientometrics, the characteristic trends
and patterns of the past decades in scientometric research become evident.
Bibliographic data from the journal of Scientometrics have formed the main data
object in some of recent empirical studies focusing on mapping the intellectual
structure or detecting social network community in the field of
scientometrics. Therefore, this study also employed bibliographic data that
cover all types of documents published in Scientometrics in 1978–2011 and
2011–2015 as representative experimental data object in the field of
scientometrics. All data were retrieved from Web of Science (WOS). Data
retrieval in the first period was completed in the middle of 2011; the data will
be used in deduction and mining. Meanwhile, data retrieval in the second period
was completed in the middle of 2015; the data will be used in verifying the
results obtained by the first sample.
The first retrieval recalled a total of 2,989 documents, of which 2,982 include
author information, 2,815 include references, and 2,812 include both the author
information and references. The most prominent contributors are pioneers in most
research studies. For example, when evaluating author influence levels, only
pioneer authors were considered, which was done by most studies only considered
(such as, in uncovering knowledge communication, and in revealing implicit
relationship) because the cited references only contain the first listed author
of the cited document in the database of WOS. Moreover, a complex contribution
allocation problem existed when considering all authors in relation analysis.
This problem has not been fully solved, which is beyond the scope of this
research topic. Thus, only the first authors of each paper were considered in
the current study. In counting only the first author in the citation data, the
results include 35,796 citations, 16,057 cited authors, and 1,484 citing authors
(as first signature identity in the publications). Each author’s name is
identified by his surname and first initial only. The second dataset covers
1,318 documents in total, all of which include the author information and 1,308
include references (involving 27,083 valid citations).
## Methodologies
Thus far, a uniform standard for identifying core authors in scientometrics has
not been developed. Lotka and Price identified excellent scientists according to
the number of their published papers during the study on scientists’
productivity and activity patterns. Garfield treated those authors with high-
cited frequency from SCI as excellent scientists. Some scholars also adopted
different approaches to evaluate core authors in information science; however,
they all considered both the number of published papers and the cited frequency.
Therefore, the present study identified 94 authors who have published 5 or more
papers and received 10 or more citations as core authors from the first dataset.
AC, ABC, and ADC analyses are used in discovering author relationships with co-
citation, bibliographic coupling, and direct citation in scientometrics,
respectively. CA and AKC analyses were also complementarily used to discover or
verify author relationships in this study. AKC analysis was introduced by Liu et
al. and was formally proposed by Liu and Zhang. This method was re-introduced
and compared with CA and ABC analyses by Qiu and Wang, Liu and Wang, Song and
Wu, and Yang et al.. The AKC analysis is supposed to expand the keyword co-
occurrence relationship at the author level; it can also be used to establish
author relationships through the keyword coupling strength of authors’ oeuvres.
The oeuvres can be used to discover PCRs among authors bound by the same
research themes and then describe the knowledge structure of a field or
discipline.
The networks of CA and AC were directly constructed by their co-occurrence
relationships in the same records. The network of ADC comprised two-way direct
citing network between author pairs (i.e., symmetrized by summing the two
directional citation values as the total correlation score). The citing and
cited links should be equally treated as the direct relationship between author
pairs. Thus, the summing symmetrization was selected instead of the lowest or
highest value method or even an asymmetrical matrix. However, the symmetrizing
processcould be improved by involving the total number of citations and
references of authors’ publications to eliminate the effect of the absolute
value. Although the original value can reflect an actual situation, the direct
citation frequencies must be normalized. However, such normalization could only
be done in another study, because researchers have not reached a consensus on
which measure is most appropriate for normalization purposes. For ABC and AKC,
basic matrixes, including authors\*cited reference matrix and authors\*keywords
matrix, were initially generated and then transformed into ABC and AKC networks
via formulas, selecting the minimum method to calculate the coupling strength as
suggested by Ma. All of the original co-occurrence matrixes including AC, ADC,
ABC, CA and AKC are supplied in the Supporting Information (– Tables
corresponding to the period of “Before 2011”; – Tables corresponding to the
period of “After 2011”.)
Co-occurrence analysis and deductive reasoning methods are used in mining deeper
and more potential author relationships based on the original tripartite
citation analysis. VBA program can process all types of citation analysis data.
The final results of author relationship mining will be visualized by the
Network Workbench Tool software with the analysis of MST-PathFinder Network
Scaling. The use of PathFinder can simplify the network and highlight its
important structural features and core associated nodes. This method was used in
this study to highlight the visualization of the network and improve map
readability.
The QAP is a unique method of measuring relationships in relational data. It
compares the value of various corresponding elements in two (or more) squares
and gives the Pearson correlation coefficient between two matrixes by comparing
the corresponding grid values in each square. A non-parametric test is performed
on the coefficients based on the replacement of the matrix data. A comparison on
proximity results in this study was conducted using QAP, and the statistics
process (including centrality measurement) was annotated in the documentation of
Ucinet software.
# Study process
## Discovery of PCRs
In this study, five original relation matrixes (including CA, AC, ABC, ADC, and
AKC) were first developed. These five matrixes were compared by QAP, and the
result was saved and marked as QAP1. Excluding ACR (including CA and ADC
matrixes) from PCR (including AC and ABC matrixes), the AC′ and ABC′ matrixes
could be obtained. The AC′, ABC′, and AKC matrixes were re-compared, and the
result was marked as QAP2. Then, a comparison between QAP1 and QAP2 was
performed. AKC is based on the similarity of research themes (can be called
“inherent connection”), whereas AC and ABC are based on citation relationships
(can be called “exterior connection”). When the inherent and exterior
connections are highly consistent with each other, the PCR is assumed to convert
into ACR. To test this assumption, the results obtained by PCR from the ACR
relationships were compared with the AKC matrix (2011–2015) and ACR matrix
(2011–2015).
## Deep relationship mining between author pairs
In this study, the tripartite citation analysis could be applied in deep
relationship mining at the author level. To make these relationships comparable,
original relation matrixes should be normalized. The normalization method was
based on Salton’s cosine similarity measures, which results in similarity values
ranging between 0 and 1.
The following five steps (some of the steps are shown) aid in determining author
relationship mining based on tripartite citation analysis, such as “A–C,” “B–D,”
and “B–C,” which has been discussed earlier. These steps could also be regarded
as the algorithm in relation mining. The implication of each variable A, B, C,
and D refers to the author of the matrix, L; Q refers to the relationship
between the authors in the adjacency list O; and P refers to the relationship
between the authors in the adjacency matrix.
1. *First step: Obtaining the fundamental citation relationship with
strength (\>0) among the core authors from the original matrixes*.
Tripartite adjacency matrixes were transformed into the corresponding
adjacency list: AC list {L<sub>1i</sub>, Q<sub>1i</sub>} versus matrix
{O<sub>1i</sub>, P<sub>1j</sub>}; relational degree X<sub>i</sub> (i stands
for the ID of the author pair) in the list can replace X<sub>ij</sub> (i/j
stand for different authors in the matrix). ABC list {L<sub>2i</sub>,
Q<sub>2i</sub>} versus matrix {O<sub>2i</sub>, P<sub>2j</sub>}, and
relational degree Y<sub>i</sub> versus Y<sub>ij</sub>. ADC list
{L<sub>3i</sub>, Q<sub>3i</sub>} versus matrix {O<sub>3i</sub>,
P<sub>3j</sub>}, and relational degree Z<sub>i</sub> versus Z<sub>ij</sub>.
We used the adjacency list for the calculation.
2. *Second step: Filtering no-explicit-relationship author pairs*.
The no-relationship author pairs (X<sub>i</sub> = 0, Y<sub>i</sub> = 0,
Z<sub>i</sub> = 0, and no cooperation) were filtered as {O<sub>4i</sub>,
P<sub>4j</sub>} in the adjacency matrix and {L<sub>4i</sub>, Q<sub>4i</sub>}
in the adjacency list, which formed the basic object in the subsequent
analysis.
3. *Third step: Mining the relationship of A–C from {L<sub>1i</sub>,
Q<sub>1i</sub>} {L<sub>3i</sub>, Q<sub>3i</sub>} {L<sub>4i</sub>,
Q<sub>4i</sub>}*.
{L<sub>4i</sub>, Q<sub>4i</sub>} was remarked as {A<sub>k</sub>,
C<sub>k</sub>} (k stands for the number of author pairs) to find the
D<sub>k</sub> with the relations {A<sub>k</sub>–D<sub>k</sub>,
C<sub>k</sub>–D<sub>k</sub>}. Looking for the synchronous relations with
strength between A<sub>k</sub> and D<sub>k</sub>, and C<sub>k</sub> and
D<sub>k</sub> from {L<sub>1i</sub>, Q<sub>1i</sub>} {L<sub>3i</sub>,
Q<sub>3i</sub>}, and matching the author pairs in {A<sub>k</sub>,
C<sub>k</sub>}, the pseudocode is as follows:
If one author in the pair of {A<sub>k</sub>, C<sub>k</sub>} = one author
in a pair of {L<sub>1i</sub>, Q<sub>1i</sub>}, and another one in the pair
of {A<sub>k</sub>, C<sub>k</sub>} = one author in a pair of {L<sub>3i</sub>,
Q<sub>3i</sub>}, and another one in the pair of {L<sub>1i</sub>,
Q<sub>1i</sub>} = another one in the pair of {L<sub>3i</sub>,
Q<sub>3i</sub>}
then the “one author in the pair of {A<sub>k</sub>, C<sub>k</sub>}” (so
as the “one author in a pair of {L<sub>1i</sub>, Q<sub>1i</sub>}”) as
C<sub>α</sub>, the “one author in a pair of {L<sub>3i</sub>,
Q<sub>3i</sub>}” (so as the “another one in the pair of {A<sub>k</sub>,
C<sub>k</sub>}”) as A<sub>α</sub>, and the “another one in the pair of
{L<sub>1i</sub>, Q<sub>1i</sub>}” (so as the “another one in the pair of
{L<sub>3i</sub>, Q<sub>3i</sub>}”) as D<sub>α</sub> are marked.
Finally, A<sub>α</sub> and C<sub>α</sub> could be connected according to
D<sub>α,</sub> and the final relationship strength of A<sub>α</sub> and
C<sub>α</sub> would be the top value in all of the correlation scores
(respectively equaling to the products of Y<sub>k</sub> and Z<sub>k</sub>).
4. *Fourth step: Mining the relationship of B–D from {L<sub>2i</sub>,
Q<sub>2i</sub>} {L<sub>3i</sub>, Q<sub>3i</sub>} {L<sub>4i</sub>,
Q<sub>4i</sub>}*.
{L<sub>4i</sub>, Q<sub>4i</sub>} was remarked as {B<sub>k</sub>,
D<sub>k</sub>} (k stands for the number of author pairs) to find the
A<sub>k</sub> with the relations {A<sub>k</sub>–D<sub>k</sub>,
A<sub>k</sub>–B<sub>k</sub>}. The synchronous relationship with strength
between A<sub>k</sub> and D<sub>k</sub>, and A<sub>k</sub> and B<sub>k</sub>
were searched from {L<sub>2i</sub>, Q<sub>2i</sub>} {L<sub>3i</sub>,
Q<sub>3i</sub>}, and the author pairs in {B<sub>k</sub>, D<sub>k</sub>} were
matched. This process is similar to the process of A–C. Thus, the pseudocode
was omitted. Finally, connected author pairs {B<sub>β</sub>, D<sub>β</sub>}
with relationship strength X<sub>k</sub> multiplied by Z<sub>k</sub>
could be acquired according to their co-connection with A<sub>k</sub>.
5. *Fifth step: Mining the relationship of B–C from {L<sub>1i</sub>,
Q<sub>1i</sub>} {L<sub>2i</sub>, Q<sub>2i</sub>} {L<sub>3i</sub>,
Q<sub>3i</sub>} {L<sub>4i</sub>, Q<sub>4i</sub>}*.
The remaining (no relationship such as A–C and B–D) of {L<sub>4i</sub>,
Q<sub>4i</sub>} were remarked as {B<sub>k</sub>, C<sub>k</sub>} (k stands
for the number of author pairs) to find the A<sub>k</sub> and D<sub>k</sub>
with the relationship {A<sub>k</sub>–D<sub>k</sub>,
A<sub>k</sub>–B<sub>k</sub>, and C<sub>k</sub>–D<sub>k</sub>}. The
synchronous relationship with strength between A<sub>k</sub> and
D<sub>k</sub>, A<sub>k</sub> and B<sub>k</sub>, and C<sub>k</sub> and
D<sub>k</sub> were searched from {L<sub>1i</sub>, Q<sub>1i</sub>}
{L<sub>2i</sub>, Q<sub>2i</sub>} {L<sub>3i</sub>, Q<sub>3i</sub>}, and the
author pairs in {B<sub>k</sub>, C<sub>k</sub>} were matched. The
pseudocodes are as follows:
If one author in the pair of {B<sub>k</sub>, C<sub>k</sub>} = one author
in a pair of {L<sub>2i</sub>, Q<sub>2i</sub>}, and another one in the pair
of {B<sub>k</sub>, C<sub>k</sub>} = one author in a pair of {L<sub>1i</sub>,
Q<sub>1i</sub>}, and another one in the pair of {L<sub>2i</sub>,
Q<sub>2i</sub>} = one author in the pair of {L<sub>3i</sub>,
Q<sub>3i</sub>}, and another one in the pair of {L<sub>1i</sub>,
Q<sub>1i</sub>} = another one in the pair of {L<sub>3i</sub>,
Q<sub>3i</sub>}
then marking the “one author in the pair of {B<sub>k</sub>,
C<sub>k</sub>}” (so as the “one author in a pair of {L<sub>2i</sub>,
Q<sub>2i</sub>}”) as B<sub>χ</sub>, “another one in the pair of
{B<sub>k</sub>, C<sub>k</sub>}” (so as “the one author in a pair of
{L<sub>1i</sub>, Q<sub>1i</sub>}”) as C<sub>χ</sub>, one author in the pair
of {L<sub>3i</sub>, Q<sub>3i</sub>} (so as the “another one in the pair of
{L<sub>2i</sub>, Q<sub>2i</sub>}”) as A<sub>χ</sub>, and another one in the
pair of {L<sub>1i</sub>, Q<sub>1i</sub>} (so as the “another one in the pair
of {L<sub>3i</sub>, Q<sub>3i</sub>}) as D<sub>χ</sub>
Finally, B<sub>χ</sub> and C<sub>χ</sub> could be connected according to
A<sub>χ</sub> and D<sub>χ</sub>, and the final relationship strength of
B<sub>χ</sub> and C<sub>χ</sub> would be the top value in all of the
correlation scores (respectively equaling to the products of X<sub>k</sub>,
Y<sub>k</sub>, and Z<sub>k</sub>).
Thus far, all relationships among author pairs in {L<sub>4i</sub>,
Q<sub>4i</sub>} had been established.
According to the above algorithm, potential relationships among no-direct-
relationship core author set could be generated by the VBA program and Access
databases. Finally, the comparison of new relationships and direct correlations
(including CA, AC, ABC, ADC, and AKC) in 2011–2015 would be performed to
identify the effectiveness of citation mining applied in the detection or
promotion of more potential communications.
# Results and discussion
## Analysis results of AC, ABC, and ADC
According to the tripartite citation analysis of AC, ABC, and ADC, we obtained
three original relation matrixes and the corresponding normalized matrixes.
These tripartite matrixes and the AKC matrix could be visualized by the Network
Workbench Tool (Figs –).
The core nodes in each network are different, as shown in Figs. In the AC
network, “(Moed HF, Narin F, Vlachy J)–Garfield E–Braun T–Schubert A–Glanzel
W–Egghe L–Rousseau R–Thelwall M” are core associated scholars, all of whom form
the main path in the network. In the ABC network, the main associated scholars
include “Schubert A–Glanzel W–Meyer M–Leydesdorff L,” in which new core nodes,
such as Garg KC, Bar-ilan J, Guan JC, and Zitt M, also emerge. In the ADC
network, Leydesdorff L becomes the superior core node, and the associated path
of “Abramo G–Glanzel W–Leydesdorff L–Bornmann L–Garfield E” becomes the main
path. In the AKC network, “Schubert A–Rousseau R–Vinkler P–Glanzel W–Leydesdorff
L” becomes the major associate scholars, and the key associations of Thelwall M,
Guan JC, Zitt M, and Glanzel W are also reflected. These scholars are also at
the heart of correlation formation among other scholars. Generally, the main
connected path of “Garfield E–Schubert A–Glanzel W–Leydesdorff L” is more
consistent in these four networks. However, the difference is also distinct,
such that Garfield E is the main supporter for most core paths in the AC
network, while the main supporter in the ABC and ADC networks are Glanzel W and
Leydesdorff L, respectively. In view of the ADC revealing a more direct
relationship, Leydesdorff L is more likely to be the builder of the potential
connection.
As shown in Figs, author relevance was preliminarily identified by different
citation methods. For example, *Leydesdorff L* has the strongest correlation
with other authors in the direct citation networks, whereas the core correlation
of the three types of indirect network (PCR) is relatively low, and some are
even weak in the cooperative correlation. Meanwhile, *Narin F* presents the
highest correlation degree in the AC network, and the correlation degree is
comparatively lower in other networks. *Bornmann L* and *Sooryamoorthy R* are
strongly correlated with the ADC and AKC networks, respectively. However, their
correlations are low in other networks. In terms of partnership, these two
authors do not establish any cooperative relationship with other co-authors.
*Glanzel W* has the core link status and high centrality degree in all networks,
followed by *Schubert A* and *Braun T*. In addition, *Bonitz M*, *Nagpaul PS*,
*Mccain KW*, *Eto H*, *Stefaniak B*, *Krauskopf M*, *Wagner-Dobler R*, and
*Macias-Chapula CA* have established indirect relationship in AC/ABC/AKC
network. However, no direct relationship is observed in CA/ADC. Ucinet software
was used to calculate the centrality measurement of the five types of network.
The top ten authors are shown in. In, ND is the abbreviation of NrmDegree, which
represents the normalization of degree.
## Results of PCR discovery
In this comparative study, AKC analysis was applied to produce the AKC matrix,
in which the implementation process is similar with that of the ABC analysis
(i.e., the authors are correlated with one another by indexing the same
keywords). presents the result of the QAP correlation test of CA, AKC, AC, ABC,
and ADC matrixes. The results show that the correlation between the CA and ABC
matrixes is the strongest, followed by the AC and ADC matrixes, and the AC and
ABC matrixes. These findings indicate that the current study and the topic
structure revealed by these three pairs are perhaps the most similar, or can be
mutually complementary. In addition, the ABC matrix generally has the highest
degree of correlations compared with all other relationship matrixes, which to
some extent shows that the application of the ABC analysis can more accurately
reveal the scientific structure of the disciplines. One of the possible reasons
for using this analysis method is to divide into research groups and discover
the subject structure by numerous scholars. This result supports the findings of
Yan and Qiu, both of which revealed that ABC nearly has the highest similarity
with other networks at the author level. Furthermore, the correlation
coefficient of AKC and ABC matrixes is at a middle level, which indicates that
AKC and ABC analyses share similarities to a certain extent. This finding is
consistent with the conclusion of Yang.
Excluding ACR connections (including CA and ADC) from PCR (including AC, ABC,
and AKC), the matrixes of AC′, ABC′, and AKC′ were obtained. The QAP correlation
test for the new three matrixes was performed, with results shown in. The two
groups of correlation strengths in successive QAP results were compared (marked
in color red). The comparison shows that the relation degrees among AC′, ABC′,
and AKC′ could also maintain significant correlations, especially AKC′ and ABC′,
which share higher relevancy than the original matrixes. This condition
indicates that these authors connected by PCR are likely to produce academic
exchanges based on similar themes.
The three new matrixes with ACR connections from 2011 to 2015 were further
compared in. According to the QAP analysis, the correlation coefficient between
the new ACR (after 2011) and the old PCR (before 2011) is 0.225 (p\<0.001). This
result can also sustain the assumption about applying PCR in the detection of
new academic exchanges. We converted the new relationships into author pairs and
analyzed them with Pearson correlation. We found that the three previous
relationships showed a more apparent correlation with the new PCR and ACR. Among
those relationships, ABC has the strongest correlation and the highest
predictability. In addition, the previous and new PCRs are consistent, and the
correlation between the new PCR and ACR is significant. Meanwhile, ABC also
reflects the highest correlation, followed by AC.
Further analysis of the author pairs before and after 2011 demonstrates that
several author pairs have strong PCR correlations, such as *Bordons M*–*Glanzel
W*, *Katz JS*–*Leydesdorff L*, and *Braun T*–*Rousseau R*. In addition, the new
ACR correlations appeared after 2011, which suggests that the PCR relationship
promotes the occurrence of the ACR relationship. The new main ACR author pairs
are listed in.
## Results of author relationship mining
According to various analyses of AC, ABC, ADC, and CA, we found that *Glanzel W*
had direct relationship with others, whereas most of the core authors have not
related to all of the other authors. Following the five steps described above,
new relationships between Authors A and C, B and D, and B and C were discovered,
with respective author relation pairs: 1,793, 1,916, and 10. Subsequently, the
final results among A–C, B–D, and B–C were acquired and visualized by the
Network Workbench Tool, as shown in Figs. It is needed to point out that
although the networks generated by PathFinder are sparse (in this network, only
core node and connections are preserved, in order to concentrate on crucial
points), the identification of the implicit associated author pairs is based on
all the network data. Therefore, the relational mining process is still valid in
this finite data set.
As shown in Figs, *Pinto M*, *Lee YG*, *Prathap G*, and *Breimer LH* et al. had
few direct relationships with others, and even rarer relationship among them. In
particular, *Breimer LH*, whose research focused on interdisciplinary fields,
such as medical and information sciences, had no co-citation relation with the
core authors in the field of scientometrics, and few direct-citation and
bibliographic-coupling relations with others. However, in the relation mining of
“A–C” by joining ABC with ADC, indirect relationships between *Breimer LH* and
77 authors were established, and *Breimer LH* became a core node in the A–C
network to replace the independent node in the AC network and peripheral node in
other networks. Meanwhile, *Prathap G*, who was occupied with interdisciplinary
research between material and information sciences, established indirect
relationships with approximately 60 authors. *Pinto M*, an emerging scholar in
scientometrics (with 5 papers and 10 citations in the dataset), has not yet
formed a certain influence in the field. Few direct relations are observed
between *Breimer LH* and elder statesmen in the field. However, he was also
supposed to be the core node in A–C and B–D networks.
The first direct relationship mining among *Breimer LH*, *Inhaber H*, *Lee YG*,
*Sengupta IN*, *Vaughan L*, and *Pinto M* was not fully achieved. Thus, the
mining of “B–C” joining ABC, ADC, and AC were conducted with the newly
discovered direct relationships. As shown in the results presented in, the core
link status of *Breimer LH* was re-verified, and his research can be considered
gaining more attention from colleagues and that more communication and linkages
are established over time because of him.
To verify the existence of author relationship mining based on tripartite
citation analysis proposed in this study, correlation analysis between the
mining results and author direct relationship status (such as co-author and co-
citation) was recently performed to reveal the predictive and practical value of
the mining method and results. New AC, ABC, ADC, AKC, and CA matrixes from 2011
to 2015 have been investigated; and five matrixes, including CA, AC, ABC, ADC
(symmetrized), and AKC were developed. Four author pairs could be identified
according to the comparison of data mining before 2011(A–C and B–D), as well as
evident relationship after 2011 (AKC″, CA″, AC″, ADC″, and ABC″), which are
shown in. Although no co-authorship exists among these author pairs, the other
direct relationships, such as AKC, ADC, and ABC, are still evident, especially
*Leydesdorff L* and *Prathap G*.
On the normalization process, given the presence of large amounts of 0 module
caused by less amount of data within a limited time, another type of
standardized method was selected. The AC matrix was considered as an example;
the co-citation frequency between Authors A and B is x, the total frequency of A
co-cited with all authors is m, the total frequency of B co-cited with all
authors is n, and the correlation strength between Authors A and B is x/m+x/n.
This analogy indicates that standardized matrixes of CA, AC, ABC, and ADC were
obtained. Finally, a comprehensive correlation (CC) matrix was developed by
adding four types of correlation values. Pearson correlation test was performed
among author pairs of A–C, B–D, CC, and AKC, which correspond to two types of
data sets, namely, A–C and B–D. The results are shown in, and the CC matrix was
visualized using the Network Workbench Tool. Network Scaling MST-PathFinder was
performed to show the network clearly. Therefore, the previously revealed
correlations are not fully displayed.
As shown in, certain degrees of positive correlation are observed among A–C,
AKC, and CC, and B−D and AKC, which could indicate that the indirect
relationships mentioned above would turn into direct relationships to some
extent in the near future. Scholars in the field have consciously or
unconsciously paid close attention to or cite links (including co-cited,
coupled, and citing) with other scholars who shared indirect relationships
instead of the direct relationships with the former authors, and even produced
substantial cooperation among one other. Notably, the correlation between the
AKC and CC matrixes is relatively significant. This result can be compared with
previous results shown in Tables and, in which AKC is also significantly
correlated with others, even though the related values are comparatively lower
(except for ABC). Therefore, AKC analysis may help in revealing the evolution of
the existing relationships. Meanwhile, the relationship mining method proposed
in this study could aid in revealing unknown relationships that complement with
AKC analysis or other methods, such as topic analysis.
# Conclusions
Various relationships exist in academic networks, such as CA, AC, ABC, ADC, and
AKC. In a given field, the intensity and the associated attributes among
scholars may exhibit significant differences in terms of the different network
correlations. Some scholars showed strong ACR correlations, while some had key
positions in PCR association. In this study, we compared the five types of
matrixes by QAP and found that ABC has the nearly highest similarity with other
networks. This finding can demonstrate the superiority of ABC analysis in
revealing an academic community and its scientific structure. Furthermore, the
correlation coefficient of ABC and AKC is higher than the coefficients among AKC
and others, indicating that AKC and ABC can be complementarily applied in
potential communication mining.
By comparing the relationship of ACR and PCR, a particular phenomenon was
observed, in which only PCR existed among scholars without ACR in the field of
scientometrics, such as among *Bonitz M*, *Nagpaul PS*, *Mccain KW*, and *Eto
H*. By analyzing PCR with the ACR correlations excluded (i.e., including AC′,
ABC′, and AKC′) by QAP, we found that the relationship degrees among AC′, ABC′,
and AKC′ can also maintain significant correlations, especially AKC′ and ABC′,
which share higher relevancy than the original matrixes. This result indicates
that these authors connected by PCR are likely to produce more academic
exchanges and scientific innovations because of similar themes rather than
social attribute association (e.g., teacher–student or co-worker relationships).
By conducting Pearson correlation analysis, the case study confirmed that a
significant correlation existed between the PCR that appeared before 2011 and
the new ACR″ and PCR″, which appeared after 2011. Furthermore, continuity
existed between PCR′ and PCR″, and the associated relationship between ACR″ and
PCR″ were also significant. Particularly, ABC′, ABC″, and other relations have
been highly correlated, which indicated that ABC analysis shows a good
application potentiality in relationship prediction and discovery to some extent
and may reflect actual communication.
On the basis of the algorithm design and the empirical analysis, the deduction
from the analysis results from AC, ABC, and ADC to the potential author
relationships mining is probable and practicable. For example, the relationship
between *Leydesdorff L* and *Prathap G* revealed by A–C/B–D in the case study
achieved a high degree of correlation in the practice after 2011 (including AC,
ADC, and ABC, which did not exist before 2011). The author correlation between
*Breimer LH* and *Vaughan L* obtained by two-time mining was also consistent
with the new correlation in 2011, which once again confirmed the validity and
the potential value of the proposed method for revealing author relationships.
The results presented above revealed that the indirect relationships among
interdisciplinary scholars or novice researchers can be mined by the method
combined with tripartite citation analysis, which helps with specific scientific
cooperation and broader communication. In comparison with the direct
relationship presented recently by Pearson correlation, the author mining method
proposed in this study helps in revealing unknown relationships and could
complement with AKC analysis or other methods, such as topic analysis. These
methods could be applied in discovering research fellows, exploring potential
partners, as well as tracking scholars with related research and their research
direction.
In conclusion, this study attempted to discover PCRs. Through the correlations
between the measurements, the proposed method could be used to explain that the
establishment of co-citation or coupling relation may promote the production of
the actual communications. This finding suggests that these two relations could
identify potential collaboration partners for both individuals and teams. The
proposed author relationship mining method based on tripartite citation analysis
could also be an effective method for discovering future relationships among
scholars and promoting scientific communication and innovation development.
In addition, we recognized the existence of limitations in the dataset of “core
authors” by selecting only the first author as citation data and defining the
threshold of the publication number and citation frequency. We performed such
step despite the fact that only the first cited authors tallied in the database
of WOS and regardless of the difficulty of a specific time window for obtaining
a sufficient linking signal (e.g., the data in the first period of 1978–2011,
which was acquired in 2012, is difficult to be regained at present). However,
this paper was an attempt to propose an idea and process in author relationship
mining in the context of five types of scholarly networks; thus, the collection
of core authors targeted by this research was supposed to be useful in the
application of relationship mining method. Nevertheless, we are faced with the
data limitation, thus the need to present a more credible empirical study with a
sizable sample and enhance the practicality and effectiveness in PCR discovery
by tripartite citation analysis. Finally, as an attempt, the proposed method
should also be applied in various fields. However, the method was tested only in
the field of scientometrics due to computational complexity, the amount of data
obtained, and so on. In addition, some of the studies exhibit positive results,
which are applicable only in the field of scientometrics. In the context of
scientometrics, the results are easier to explain and rigorously confirmed. In
further research work, the proposed method should be applied in other fields to
further confirm its effectivity and rationality.
# Supporting information
We wish to thank the two anonymous referees for their important insightful
comments and suggestions.
[^1]: The authors have declared that no competing interests exist. |
# Introduction
There is abundant evidence worldwide on lifelong human contamination from
mixtures of environmental chemicals as persistent organic pollutants (POPs);
yet, the vast majority of studies report each pollutant individually, with
little attention to concentrations of mixtures in individual persons or social
groups. Thus, the complex features of such internal, body contamination remain
unsatisfactorily characterized. Biomonitoring surveys, for instance, do not
integrate the number of compounds detected per person and the concentration of
each compound. Possible health effects of POPs include a variety of
developmental, metabolic, neurodegenerative, and neoplastic disorders.
Reasonable concerns exist about such effects at low concentrations; such issues
can be integrated with the fact that it is common for humans to have mixtures of
POPs at low and high concentrations.
Approaches to these issues include ‘Environment-Wide Association Studies’
(EWAS), and analyses of concentrations of POPs combined, using estimates of
total body burden, or different sums of concentrations. Efforts to improve
exposure assessment must continue: not only to advance etiologic studies and
risk assessment, but also to foster knowledge on the characteristics of human
chemical contamination itself. Such knowledge is a recognized right of citizens
in democratic societies; it is also essential to evaluate the impacts of health,
industrial, and related policies. Indeed, the sources and pathways of exposure
to pollutants are socioeconomic and cultural. Thus, strong relationships exist
between concentrations of individual POPs and social factors, including income,
education, and race / ethnicity. Unfortunately, such relationships have seldom
been analyzed integrating several compounds and their concentrations.
Recently, a set of indicators that integrate the number of compounds detected
per person and their corresponding concentrations was proposed, including the
number of compounds detected at high concentrations. The analyses were based on
the general population of Catalonia, Spain. Because studies in the U.S. on
combinations of POPs and other chemicals raised relevant questions about the
levels and effects of such mixtures, and because of the relatively large size of
the U.S. population, we aimed at applying the methodology to the U.S. general
population.
Therefore, the objectives of the present study were to analyze the number of
POPs detected per person at high concentrations (nPhc) in the U.S. National
Health and Nutrition Examination Survey (NHANES), and to analyze the
associations between such indicator and main socioeconomic factors. Our main
hypotheses were that most of the U.S. population would have POPs at low and high
concentrations, and that sociodemographic factors (such as age, gender, body
mass index (BMI), parity, or income) that are often related with POP
concentrations when each compound is analyzed individually would continue to
show similar relationships when the POPs are jointly analyzed.
# Materials and Methods
## Data
Conducted by the Centers for Disease Control and Prevention’s (CDC) National
Center for Health Statistics (NCHS), the National Health and Nutrition
Examination Survey (NHANES) collects nationally representative environmental
biomonitoring data from about 5,000 annual participants in each two-year cycle.
NHANES is a publicly available data set, and all participants provide written
informed consent, consistent with approval by the NCHS Institutional Review
Board. Ethical approval for use of NHANES data is not required as it is
anonymized. We examined data from NHANES laboratory and demographic files
corresponding to 2003–2004, which is the last period with valid updated
individual data for compounds considered in the present study. Except for
perfluorinated compounds (PFCs), in NHANES 2005–2006 and 2007–2008 serum
concentrations of POPs were measured using weighted pooled-samples, and no data
for POPs have been published for NHANES 2009–2010 and 2011–2012. Therefore, it
is not possible to calculate the number of POPs detected per person at high
concentrations in more recent periods.
In each NHANES, most chemicals or their metabolites were measured in serum
samples from random subsamples of about 2,500 participants aged 12 years and
older. The chemicals’ concentrations were analyzed by CDC’s Environmental Health
Laboratory using mass spectrometry and related methods. Data for 91 POPs were
analyzed, including: 13 organochlorine compounds (OCs) and their respective
metabolites; 10 polybrominated diphenyl ethers (PBDEs); the polybrominated
biphenyl (PBB) 153; 29 non-dioxin-like polychlorinated biphenyls (PCBs); 3
dioxin-like coplanar PCBs; 6 dioxin-like mono-ortho-substituted PCBs; 10
polychlorinated dibenzofurans (PCDFs); 7 polychlorinated dibenzo-p-dioxins
(PCDDs); and 12 PFCs. Thus, serum concentrations of lipophilic chemicals (e.g.,
dioxins and PCBs) are presented per gram of total lipid (better reflecting the
amount stored in body fat); results of analyses per whole weight of serum were
similar and are not presented. Concentrations of PFCs, non-lipophilic POPs, are
shown per liter of serum. Limits of detection (LOD) for whole weight POP
concentrations were different for each serum sample of each person, while LODs
for lipid-adjusted concentrations were the same values for all samples and
individuals (values ranged from 3.8 pg/g of lipid and 7.8 ng/g of lipid).
Finally, LODs for PFCs ranged from 0.1 to 1.0 μg/L.
We considered important covariates as age, sex, race/ethnicity (non-Hispanic
white, henceforth ‘White’; Mexican American; non-Hispanic black, henceforth
‘Black’; other Hispanic; and other), education (categorized to less than high
school diploma, high school diploma, and greater than high school diploma), and
body mass index (BMI) in kg/m<sup>2</sup>. To estimate the participants' income
we used the family’s total income divided by the family size-specific poverty
threshold income ratio (PIR), with two categories: “Low” income (PIR \< 2), and
“High” income (PIR ≥ 2). In women, we also considered the number of pregnancies
resulting in live births, and the number of children breastfed ≥1 month
(henceforth, ‘breastfeeding’).
## Statistical analyses
The present study included 4,739 participants ≥20 years old (for all adults 85
years and older, age was coded at 85 years to reduce the risk of disclosure).
They came from three subsamples. There were no significant differences between
the 1,610, 1,585 and 1,544 participants of each subsample in a broad range of
sociodemographic variables (including sex, race/ethnicity, educational level,
PIR, BMI, number of pregnancies or breastfeeding).
We imputed the unmeasured POP values by the median serum concentration of each
POP according to age, sex, race/ethnicity, PIR, BMI, and, in women, number of
pregnancies. In 79 POPs the imputation was performed using concentrations
adjusted by lipids, and in 12 PFCs, in μg/L. We calculated the total toxic
equivalency (TEQ) for 26 POPs: 3 dioxin-like coplanar PCBs, 6 dioxin-like
mono-*ortho*-substituted PCBs, 10 PCDFs, and 7 PCDDs. To compare POP
concentrations in the present study and pooled concentrations in NHANES
2005–2006 and 2007–2008 we computed concentrations of POPs by sex,
race/ethnicity and age groups. We also compared PFC serum concentrations in the
present study and concentrations in NHANES of 2005–2006, 2007–2008, 2009–2010
and 2011–2012. Descriptive values for POP concentrations imputed are summarized
in, sorted from the highest to the lowest percentage of detection.
Based on previous work by Porta et al. (2012), we calculated the number of POPs
detected in each person at high concentrations (nPhc) as follows: for each
subject we added the number of POPs whose serum concentrations were equal to or
greater than a selected cutoff point. To be conservative, in the main analyses
we included only 37 POPs that had been detected (each) in \>85% of the study
subjects (henceforth called the most prevalent POPs). Such 37 POPs were: 2 OCs,
3 PBDEs, PBB 153, 23 non-dioxin-like PCBs, 3 dioxin-like PCBs, one PCDD
\[1,2,3,4,6,7,8-Heptachlorodibenzo-*p*-dioxin (HpCDD)\], and 4 PFCs. Ancillary
analyses included 50 compounds detected in \>50% of subjects. Finally, other
analyses included all 91 POPs, with quartiles, quintiles and deciles defined
after the imputation of concentrations. As usual, serum concentrations of POPs
did not follow a normal distribution, and the increment of concentrations in the
highest percentiles was very strong (e.g., for *p*,*p’*-DDE the increment of
concentrations between P75 and P90 was of 2.14 times, and between P90 and the
maximum it was 14.5 times; for PCB 153 the corresponding figures were 53% and
12.36 times, respectively).
We defined ‘high concentrations’ using compound- and population-specific
percentiles, based on actual POP distributions, as cutoff points. In the main
statistical analyses the cutoff point used was percentile 90 (P90), the upper
decile.
Univariate statistics were computed as customary. The highest correlations were
observed between PCB congeners 170 and 180, 138 and 153, 146 and 153 (all
Spearman’s *ρ* \>0.982 and *p’s* \<0.001). Fisher’s exact test for homogeneity
was applied to assess the relationship between two categorical variables. For
comparisons between continuous variables ANOVA, Kruskal-Wallis, and Mann-
Whitney’s *U* tests were used. When a tendency was observed, Mantel–Haenszel’s
χ<sup>2</sup> test and Jonckheere-Terpstra test for linear trend were used.
To estimate the magnitude of associations between the socioeconomic factors and
the number of most prevalent POPs with concentrations in the upper decile,
multivariate-adjusted odds ratios (ORs) and their corresponding 95% confidence
intervals (CI) were calculated by unconditional logistic regression with
progressive degrees of adjustment. The main effects of all predictors were
independently explored in the base models, and final models were adjusted for
age, gender, BMI, race/ethnicity and poverty income, in accordance with the
nature of the variables and the study objectives. The number of POPs with
concentrations in the upper decile was tested in different regression models
using 3 different categorizations (all dichotomous): ≥1 POP (vs. no Phc), ≥6
POPs (vs. \<6 POPs) and ≥10 POPs (vs. \<10 POPs). Categorical ordinal variables
were analyzed for a linear dose–response relation through the multivariate
analogue of Mantel’s extension test; when a linear trend was not apparent, the
probability test was used. Analyses were conducted using SPSS version 18 (SPSS,
Armonk, NY, USA, 2009).
# Results
Over 67% of the 4,739 participants (73.8% of men and 61.1% of women) had one or
more of the 37 most prevalent POPs at concentrations equal to or greater than
the 90th percentile (≥P90), while 38.0% had ≥3 POPs, and over 13% had ≥10 POPs
each in such top decile. Over 37% of subjects had ≥10 compounds each at
concentrations in the top quartile (≥P75). The number of POPs detected per
person ranged between 23 and 74, with an average of 49.7. Over 57% of
participants had ≥50 POPs detected.
In over 45% of participants who had only one POP at high concentrations (Phc)
(≥P90), this chemical was an OC, a PBDE or PBB 153. By contrast, among subjects
with numerous Phc, the majority of such compounds were PCBs. For instance, when
the nPhc was ≥3, more than 40% of these compounds were PCBs and HpCDD.
The median age of participants with ≥10 POPs at high concentrations (Phc) was 70
years, while for participants with \<10 Phc it was 45 years, and for
participants without any Phc, 39 years. Over 11% of Whites, 2.2% of Mexican
Americans, and 29.2% of Blacks had ≥10 Phc (*p* \<0.001). Subjects with ≥10 Phc
had a slightly lower median BMI than subjects with \<10 Phc (26.7
Kg/m<sup>2</sup> and 27.6 Kg/m<sup>2</sup>, respectively, *p* for trend =
0.004). Women with ≥10 Phc had a higher number of pregnancies resulting in live
births than women with \<10 Phc (age-unadjusted medians: 3.0 and 2.0,
respectively, *p* for trend \<0.001). There were significant differences in the
nPhc by sex, age, BMI, race/ethnicity, educational level, PIR, and, in women, by
number of pregnancies, and breastfeeding.
Multivariate analyses adjusted by age, gender and BMI showed that, as compared
to Whites, Blacks had an odds ratio (OR) = 10.1 of having ≥10 Phc, whilst for
Mexican Americans the OR was 0.2 (both *p’s* \<0.001). When further adjusted by
poverty income the OR for Blacks decreased to 9.2, and for Mexican Americans to
0.18 (both *p’s* \<0.001). Differences between Blacks and Whites were larger in
the older age groups / birth cohorts, and null in the younger ones (*p* for
interaction \<0.001).
For PIR ≤2, or “Low” income (vs. PIR\>2 or “High” income) the OR of having ≥10
Phc was 1.13 (*p* \>0.05) when the model was adjusted by age, gender and BMI,
and 1.24 (*p* = 0.045) when further adjusted by race/ethnicity. The OR for
obesity (vs. normal weight) was 0.74 (*p* for trend = 0.015) in the model
adjusted by age and gender, and, when further adjusting by race/ethnicity and
poverty income, it was 0.58 (*p* for trend \<0.001). In women, pregnancy halved
the probability of having ≥10 Phc when adjusting by age and body mass index (OR
= 0.48, *p* = 0.048).
In models assessing the relationship between sociodemographic factors and the
probability of having ≥1 POPs at concentrations ≥P90 (vs. not having POPs with
concentrations ≥P90), adjusted by age, gender and BMI, the OR for Blacks (vs.
Whites) was 1.10 (*p* = 0.266) and for Mexican Americans, 0.74 (*p* \<0.001).
When further adjusted by race/ethnicity, the OR for PIR≤2 (vs. PIR\>2) was 1.17
(*p* = 0.026); and for obesity (vs. normal weight), 0.73 (*p* for trend \<0.001
for BMI). The corresponding figures for the probability of having ≥6 POPs at
concentrations ≥P90 were 3.37 for Blacks and 0.51 for Mexican Americans (both
*p* \<0.001), 1.24 for PIR≤2 (*p* \<0.01), and 0.91 for obesity (vs. normal
weight) (*p* for trend = 0.353).
Because of the influence of PCBs in the previous results, we also analyzed
associations among sociodemographic factors and the likelihood of having ≥1 of 6
POPs other than PCBs (i.e., OCs, PBDEs, and PBB 153 detected ≥85% of subjects)
at high concentrations. Contrary to what was observed when all compound families
were considered, for Blacks (vs. Whites) the OR of having ≥1 of such POPs was
0.76, and for Mexican Americans, 1.41, adjusting by age, gender and BMI (both
*p’s* \<0.01). The corresponding OR for PIR≤2 was 1.19 (*p* = 0.011), and for
obesity, 0.82 (*p* for trend = 0.013). When further adjusting by race/ethnicity,
the OR for PIR≤2 was 1.15 (*p* = 0.038), and for obesity, 0.80 (*p* for trend =
0.006).
The geometric mean (GM) of nPhc doubled when the cutoff P75 was used instead of
P90. For the cutoff P75 the percentage of subjects with ≥10 Phc was 37.5 for
POPs detected in ≥85% of participants, and 45.2 for POPs detected in ≥50% of
participants. However, the percentage of subjects without any Phc decreased
slightly when the number of POPs included in the analyses increased.
In the 1,183 participants with the highest total TEQ concentrations (≥P75 of the
distribution of total TEQ concentrations \[i.e., ≥26.68 pg WHO-TEQ/g of
lipid\]), the percentage of subjects with ≥10 Phc was about twice the
corresponding figure observed when all 4,739 participants were considered. Over
90% of the 1,183 subjects had ≥1 POPs a) not included in TEQ calculations, and
b) with concentrations ≥P90. Over 30% had ≥10 such POPs, and almost 7% had ≥20
such POPs. Spearman’s *ρ* coefficient between the total TEQ concentration and
nPhc (considering the 24 POPs not included in TEQ calculations, and the P90 in
all participants for high concentrations) was 0.475 (*p* \<0.001).
Over 43% of participants had TEQ concentrations ≥21 pg WHO-TEQ/g of lipid, a
biomonitoring equivalent value. Taking into account health-based guidelines for
other compounds, less than 1% of participants had concentrations of
hexachlorobenzene ≥47 ng/g of lipid, and concentrations for the sum of
*p*,*p’*-DDT and *p*,*p’*-DDE ≥5,000 ng/g of lipid. Two subjects had
concentrations of BDE 99 ≥520 ng/g of lipid; 2 participants (aged \<40 years)
had concentrations ≥700 ng/g of lipid for the sum of 35 PCBs (without dioxin-
like coplanar PCBs), and 6 participants (aged ≥40 years) had concentrations
≥1,800 ng/g of lipid. 10% of participants had concentrations for the sum of PCBs
138, 153 and 180 ≥216 ng/g of lipid. Only 4 participants had concentrations for
the sum of these three PCBs ≥900 ng/g of lipid.
We also compared the concentrations of POPs detected in ≥85% of the participants
in the present study (2003–2004), and their respective pooled concentrations for
the NHANES periods 2005–2006 and 2007–2008. The concentrations of some POPs in
the present study were only slightly higher than in subsequent periods; they
were not higher or not statistically significant in the case of *p*,*p’*-DDE,
PBB 153 and some PBDEs compounds. For 3 PFCs, concentrations in 2003–2004 were
similar to concentrations in 2005–2010, and slightly higher than concentrations
in 2011–2012.
# Discussion
More than half of the study population had concentrations in the top decile of
≥1 of the most commonly detected POPs, 38% had ≥3, and over 13% had ≥10 POPs
each in their respective top decile. Findings are thus partly in contrast with
the notion that human POP concentrations are low in the vast majority of the
population: such view holds only when each individual compound is looked at
separately, but not when the individual human is of concern.
Median age of participants with ≥10 of most prevalent POPs at high
concentrations was 70 years, while median age of participants without any Phc
was 39 years. This could be due to biological aging effects or to birth cohort
effects. Furthermore, the median age of participants without any Phc was near
the median age of participants with 1 or 2 Phc. There were also significant
differences in the nPhc by gender, race/ethnicity, educational level, PIR, BMI,
parity, and breastfeeding. These results are in accordance with our main
hypotheses (most of the U.S. population had POPs at low and high concentrations;
sociodemographic factors related with each POP concentration showed similar
relationships for the joint analysis of POPs).
Race/ethnicity was the sociodemographic factor most associated with a higher
nPhc: Blacks had 9 times a greater chance of having ≥10 Phc than Whites, and
Mexican Americans over 4 times a lower chance. The nPhc indicator not only shows
that Blacks have higher body concentrations of POPs than Whites (or Mexican
Americans lower concentrations), but it also quantifies how many POPs are in a
specific high concentration range. The NHANES questionnaires had a large number
of sociodemographic items; in this study, we used the sociodemographic factors
that were available and related with body concentrations of POPs.
Results of unconditional logistic regression models for ≥1 Phc, ≥6 Phc, and ≥10
Phc (vs. no Phc, \<6 Phc, and \<10 Phc, respectively) in the subsample without
imputations and PCBs, PCDDs/Fs analyzed by the sociodemographic factors (to
assess the possible biases of imputations), were similar to results of models
with imputations, except in some models for gender, which was not statistically
significant, although ORs were similar.
Most studies found an inverse association between PCB levels in blood and BMI,
as in the present analyses for all participants and 37 POPs.
Also rarely if ever noted before: high percentages of subjects with TEQ ≥P75
(≥26.68 pg WHO-TEQ/g of lipid) had numerous POPs not included in TEQ
calculations, at high concentrations. Findings suggest that studies using TEQ
measures could be even more relevant if they additionally assessed subgroups
with high nPhc. Results do not imply that nPhc and related exposure indicators
are preferable to other indicators to evaluate associations between POP mixtures
and clinical outcomes; nPhc indicators just provide a different and
complementary approach to indicators such as the sum of concentrations of PCBs,
or the sum of orders of POPs.
Our goal was not to evaluate whether individuals have increased health risks due
to multiple compounds at high concentrations, nor to assess the role of modes
and mechanisms of action, but to propose a new and useful approach for exposure
assessment. However, severe adverse health effects have been reported for
concentrations similar to or lower than P90 in the present study; e.g., in an
NHANES study the OR of having diabetes for a concentration of ≥60.2 ng/g lipid
of PCB 153 was 5.9 (95% CI = 3.0–11.9); in the present study the P90 for PCB 153
was 79.8 ng/g lipid. P90 of concentrations of individual PCBs in NHANES is as
high or higher than in other countries with population-based surveys as Canada
and Australia. For *p*,*p’*-DDE and *β*-hexachlorocyclohexane (*β*-HCH), it is
also as high or higher than in Canada, Australia and Germany.
In this study, over 43% of participants had TEQ concentrations ≥21 pg WHO-TEQ/g
of lipid, which is the biomonitoring equivalent value published for dioxin TEQ,
a health-risk based screening guideline. Also, in the present study 10% of
participants had concentrations for the sum of PCBs 138, 153 and 180 equal to or
greater than the Human Biomonitoring level-I (HBM-I), which is 3 μg/L plasma or,
when accounting for lipids, 216 ng/g lipid for the present study. HBM-I is a
health-related exposure limit recommended for PCBs by the German Human
Biomonitoring Commission. For compounds considered in the present study, other
biomonitoring equivalents values are only available for hexachlorobenzene, the
arithmetic sum of *p*,*p’*-DDT and *p*,*p’*-DDE, the sum of 35 PCBs, and BDE 99;
for these compounds very few subjects had concentrations above the corresponding
biomonitoring equivalents in this study. To our knowledge, no current health-
related limit values are available for the rest of PBDEs or for PFCs. Although
there are regulations and guidelines for other pollutants (e.g. lead, mercury,
cadmium and other metals) and for POPs in air, soil, water and food (e.g.,
tolerable daily intakes), there are hardly any other guidelines for human POP
concentrations to define levels of concern than the ones mentioned above.
Beyond findings on concentrations of individual compounds, the indicators
illuminate a crucial–and usually overlooked–feature of human contamination by
POPs: the frequency of mixtures of POPs at high concentrations. The approach
could naturally be developed to integrate other pollutants of concern.
Importantly, 2003–2004 is the last period of NHANES in which the individual
concentration of each compound is available for each individual subject. In
2005–2006 and 2007–2008 serum concentrations of POPs (except PFCs) were measured
in weighted pooled-samples (not in individual samples); no data were published
for 2009–2010 and 2011–2012. Therefore, data to calculate the number of POPs
detected per person at high concentrations in more recent periods are not
available.
Concentrations of most POPs in 2003–2004 were only slightly higher than in more
recent periods (Tables in S5 File). Virtually all major contemporary health
effects of POPs will be influenced by concentrations experienced by human
cohorts during several decades, not just by recent exposures. Furthermore, the
nPhc can be fruitfully applied to analyze data from many periods and settings.
Different POPs were analyzed in participants of the three NHANES 2003–2004
subsamples; even in two of the three subsamples all selected POPs were not
analyzed in all participants. Each sample, however, is valid; and it is
efficient not to analyze all POPs in all participants. For PFCs, the LODs were
constant for each sample analyzed. For the other 55 compounds, the LOD for the
whole weight concentrations was different for each serum sample of each person.
When PCDD/Fs, PCBs, OCs, PBDEs, and PBB 153 concentrations were measured in
serum lipid, LOD calculations were performed using the chemical concentration
expressed per amount of lipid, and the LOD concentration expressed per amount of
lipid was the highest LOD among all the individual samples analyzed. LODs for
lipid adjusted concentrations were highest compared to the LODs for the whole
weight concentrations, and rates of detection were lower, than for whole weight
concentrations; as a consequence, lipid adjusted results are more conservative
(e.g., because there were less compounds detected in ≥85% of participants).
In the present study, some associations between nPhc and sociodemographic
factors are quite influenced by the predominance of PCBs and HpCDD at high
concentrations among subjects with ≥3 nPhc. The cutoff point for nPhc should be
chosen with this issue in mind, while also avoiding a too high nPhc (e.g.,
because of lower detection rates of some POPs).
Serum concentrations of POPs do not follow a normal distribution. Values for P90
can be much higher than the P75 (e.g., for *p*,*p’*-DDE the P90 value was 2.14
times greater than P75, and for PCB 153 it was 53%). Such differences between
highest concentrations minimize a possible misclassification of concentrations
in ≥P90 or \<P90 due to laboratory measurement errors. The minimum percentage of
participants with concentrations in the top decile of ≥1 POPs will be 10%, but
such percentage will not necessarily, linearly, or indefinitely increase (nor
approach 100%) as the number of compounds considered increases: the percentage
of participants with concentrations of ≥1 POPs in the top decile is only partly
positively influenced by the number of compounds considered; it is also
inversely influenced by the magnitude of the correlations between the pairs of
compounds, being highest when POPs are completely uncorrelated (for details see
Suppl. Material of Porta et al., 2012). Therefore, the nPhc follows a
distribution that is influenced by all the correlations between the pairs of
compounds, and results may not be due to chance. Figure 1 of Supplemental
Material of Porta et al., 2012 shows different values for ≥1 by the number of
POPs considered, and the values when the POPs were completely uncorrelated. For
≥10 POPs at high concentrations (rather than ≥1 POP) this situation is even more
restrictive; when we focused on ≥10 POPs at high concentrations, it was
statistically possible for the minimum percentage of participants with ≥10 POPs
at high concentrations to be 0% (i.e., it was not statistically inevitable for
that percentage to be 10%), since it is possible that highly-correlated sets of
POPs comprise 9 or less POPs. Furthermore, such minimum percentage also depends
on the number of POPs analyzed, the number of POPs in the top decile, and the
number of participants included.
# Conclusion
In summary, more than 13% of the US population may have ≥10 POPs each at
concentrations in the top decile. This finding is not to be expected just on
statistical grounds. High percentages of subjects with TEQ ≥P75 have numerous
POPs not included in TEQ calculations, at high concentrations. The nPhc is
related to race/ethnicity, age, and BMI. It is also likely to be related to
other relevant social, environmental, and individual factors. The study findings
foster knowledge on previously unknown characteristics of human chemical
contamination in the US population. Such knowledge is a right of citizens, and
could also be considered when evaluating the impacts of relevant public and
private policies.
# Supporting Information
The authors gratefully acknowledge technical and scientific assistance provided
by Natàlia Pallarès, David J. MacFarlane, Manuel Pastor, Yolanda Rovira and
Ferran Sanz.
*β*-HCH *β*-hexachlorocyclohexane
BMI body mass index
CDC Centers for Disease Control and Prevention
CI confidence interval
GM geometric mean
HBM Human Biomonitoring
LOD limit of detection
NHANES National Health and Nutrition Examination Survey
nPhc number of POPs detected per person at high concentrations
PBBs polybrominated biphenyls
PBDEs polybrominated diphenyl ethers
PCBs polychlorinated biphenyls
PCDDs polychlorinated dibenzo-*p*-dioxins
PCDFs polychlorinated dibenzofurans
PFCs perfluorinated compounds
Phc POPs at high concentrations
PIR family’s total income divided by the family size-specific poverty
threshold income
POP persistent organic pollutant
OCs organochlorine compounds
OR odds ratio
TEQ total toxic equivalency
[^1]: The authors have declared that no competing interests exist.
[^2]: **Conceived and designed the experiments:** JP DHL MP. **Analyzed the
data:** JP TL. **Wrote the paper:** JP MG MP. |
# Introduction
T2 ribonucleases are conserved nucleases found in all branches of life. In
eukaryotic cells, T2 ribonucleases affect a variety of processes including the
regulation of self-incompatibility by S-RNases in plants –, modulation of host
immune cell responses by viral and schistosome T2 enzymes, and neurological
development and tumor progression in humans. In these contexts, T2
ribonucleases can have both catalytic and catalytic-independent functions
(reviewed). For example, the effects of RNASET2 in humans on tumor progression
appear to be independent of its catalytic activity. In contrast, the catalytic
activity of the RNASET2 ortholog in *Saccharomyces cerevisiae*, Rny1, is
required for cleavage of tRNA and rRNA molecules.
Analysis of Rny1 in yeast suggests it also has properties analogous to those
seen for other RNASET2 orthologs. For example, during oxidative stress, Rny1 is
required for the production of tRNA and rRNA fragments. Similarly, expressing
human RNASET2 rescues tRNA cleavage in yeast strains lacking Rny1, and zebrafish
neurons deficient for RNASET2 and plants deficient for RNS2 accumulate rRNA. In
addition, Rny1 is a glycosylated protein that can localize to vacuoles.
Glycosylation and acidic nuclease activity are typical of T2 ribonucleases, but
whether glycosylation is required for activity is unknown. Finally, Rny1 affects
cellular growth and sensitivity to stress independently of its nuclease
activity. This non-catalytic lethality resembles the ability of RNASET2 to
suppress ovarian tumor establishment in its catalytically inactive form, and
parallels the capacity of certain catalytically mutant pestiviral T2
ribonucleases to elicit host immune cell depletion.
Since Rny1 has a signal sequence and has been reported to primarily be a
secreted or vacuolar localized protein, an unresolved issue is how Rny1 and its
RNA substrates interact in the cell. Since an Rny1-GFP fusion protein shows
reduced vacuolar signal during stress, one possibility is that Rny1 enters the
cytoplasm to engage RNAs for cleavage during stress. Precedent for this model
comes from experiments showing that a predominant mitochondrial nuclease, Nuc1,
exits mitochondria during stress to modulate nuclear and possibly cytoplasmic
RNA degradation, and in mammalian cells, lysosomal cathepsins can be released to
the cytosol during some responses leading to cell death (reviewed).
Alternatively, or in a second mechanism, cytoplasmic RNA might enter the
vacuole, where the T2 enzyme has been shown to localize during stress. Recently,
Haud *et al.* presented evidence for the accumulation of rRNA within lysosomes
with loss of RNASET2 in zebrafish neurons. Thus, an unresolved issue is how
compartmentation of Rny1 affects its function and access to RNA substrates.
Cleavage of tRNA is not unique to yeast and is conserved in eukaryotes as a
response to specific stresses, producing tRNA cleavage products mapping
primarily to the anticodon loop,. In mammalian cells, these fragments inhibit
translation and localize to stress granules, which are cytoplasmic untranslating
mRNPs that can aggregate during stress (reviewed). Coupled with the fact that
rRNA fragments accumulate during stress conditions that induce tRNA cleavage,
these data suggest the possible regulation of translation complexes and
associated translating RNAs in a stress-specific manner by ribonucleases such as
Rny1, and loss-of-function of these enzymes might impinge on cellular survival
during stresses. Interestingly, the human RNASET2 has been reported to localize
to P-bodies although the significance of this localization remains to be
determined.
To begin to understand how Rny1 functions in both catalytic and catalytic-
independent manners we have analyzed the regions of Rny1 for their functional
importance. We demonstrate that catalytic-independent inhibition of growth is a
combinatorial property of the protein and is affected by a fungal-specific
C-terminal extension, the conserved catalytic core, and the presence of a signal
peptide. Catalytic functions of Rny1 are independent of the C-terminal
extension, are affected by many mutations in the catalytic core, and also
require a signal peptide. Biochemical flotation assays reveal that in *rny1*Δ
cells, some tRNA molecules associate with membranes suggesting that cleavage of
tRNAs by Rny1 can involve either tRNA association with, or uptake into, membrane
compartments.
# Results
## Domain Organization of Rny1
Our general strategy was to make mutations in specific parts of Rny1 and examine
their effects on catalytic and catalytic-independent functions. In this light,
Rny1 possesses three domains defined in its initial characterization. At its
N-terminus it encodes a signal peptide (amino acids \#1–18), presumably for
insertion into the ER during translation. In the central region of the protein
is a conserved RNaseT2 catalytic module (amino acids \# 19–293). Within this
region are critical amino acids known to be required for activity of RnaseT2
enzymes and for the nuclease activity of Rny1. For example, substitution of two
catalytic histidine residues with phenylalanine (H87F and H160F) produces a
catalytically inactive Rny1, referred to as rny1-ci. Finally, the C-terminal
region of Rny1 is a domain that is conserved in fungal species and is not seen
in other eukaryotes.
To analyze the function of the conserved catalytic core, we desired to identify
potential surface loops of amino acids that would not directly affect the
folding of the protein, but might affect substrate interaction, or possibly
protein-protein interactions. To do this, we took advantage of the high
resolution structures of other T2 ribonucleases to predict possible loops for
mutagenesis. Using Swiss Model (<http://swissmodel.expasy.org/>) and the known
three-dimensional structure for the fungal T2 ribonuclease, ACTIBIND, we
generated a predicted structure for Rny1 based on its 39% homology to this
enzyme. In addition, using COBALT
(<http://www.ncbi.nlm.nih.gov/tools/cobalt/cobalt.cgi>), we aligned Rny1 to
ACTIBIND and another fungal T2 ribonuclease, Rh, of known structure. This
information revealed the position of eleven loops available for surface
interactions, two of which (L4 and L7) could involve RNA binding based on
previously published alignments of Rh to T2 ribonucleases with known nucleotide-
bound structures. In 10 of these loops, we generated mutants by replacing all
loop residues with alanine, generating mutations in catalytically active and/or
inactive backgrounds. Contributions of these loops were examined in assays for
growth-inhibition in wild-type and/or catalytically inactive backgrounds, and
mutations in wild-type backgrounds were assayed for the ability to cleave tRNA.
To analyze how either point mutations, or deletions to specific regions of Rny1
affect its function we used two assays. First, to assess the effect of a
mutation on non-catalytic inhibition of growth, we examined how mutations
impacted growth inhibition when Rny1 is over-expressed from the *GAL* promoter.
These experiments were done in a *hir2*Δ background, which we had seen can
increase the growth inhibitory effects of Rny1 over-expression (data not shown).
Second, to determine the effect of a mutation on nucleolytic function, we
examined how a given Rny1 variant could restore tRNA fragment production to an
*rny1*Δ strain.
## Analysis of Domains
We examined if the signal peptide, the catalytic core or the C-terminal
extension was necessary or sufficient to inhibit cell growth when over-
expressed. Moreover, to avoid any complications due to Rny1 nuclease activity,
deletions of these regions were made in the context of the rny1-ci mutation.
This led to the following key observations. First, we observed that deletion of
the signal sequence reduced toxicity. Second, we observed that deletion of
either the central conserved core, or the C- terminal extension reduced, but did
not abolish, toxicity. These effects are not due to loss of protein expression
since these variants were all expressed. We interpret these observations to
argue that nuclease-independent toxicity is a combinatorial property of both the
central conserved core and the C-terminal extensions and requires the protein to
contain a signal sequence.
We also examined the effects of these deletions on tRNA cleavage when Rny1 is
over-expressed. We observed that both the signal peptide and the central RNaseT2
domain were required for efficient tRNA fragment production, and their deletions
resemble the phenotype of the rny1-ci allele. In contrast, the C-terminal
extension is not required (ΔCTD lane). The ability to express proteins from the
mutant constructs containing catalytic sequences was not lost. We conclude that
in addition to the catalytic core domain, a signal sequence is required for
cleavage of RNA substrates by Rny1.
One possible interpretation of our results is that glycosylation might be
important for Rny1’s functions. We analyzed Rny1-GFP fusion proteins where the
GFP is either fused to the C-terminus of the protein or was inserted immediately
after the signal peptide. We observed that fusion of GFP to the C-terminal end
of the protein (Rny1-GFP) still allowed inhibition of cell growth when over-
expressed (data not shown), was able to restore tRNA fragment production in a
*rny1*Δ strain, and was glycosylated as judged by a reduction in molecular
weight when treated with the endoglycosidase, PNGaseF. In contrast, the fusion
with GFP inserted just after the signal peptide failed to inhibit growth when
over-expressed (data not shown), failed to restore tRNA fragment production to
an *rny1*Δ strain, and was not glycosylated. These observations are consistent
with the requirement for a signal peptide for function and support a model
whereby Rny1 activity requires insertion of the nascent peptide into the ER and
possibly glycosylation.
## Analysis of Mutations in the Catalytic Core Domain
We also analyzed how mutations in the loop regions of the core domain of Rny1
affected its function. We observed that none of the mutations in the loop
regions reduced the ability of Rny1 to inhibit growth when over-expressed either
in the active Rny1 or rny1-ci background. This is consistent with the
observations above that growth inhibition is a combinatorial property of the
whole protein. We also observed that mutations within loop 2, 3, 6, or 7
inhibited tRNA cleavage, while loop 10 was dispensable for this activity even
though mutant proteins were all expressed at similar levels.
A surprising observation was that mutations in loop 4, which are near a
predicted RNA binding site alter the predominant cleavage product. This is
surprising since RNAseT2 enzymes are thought to be generally non-specific in
their cleavage sites and to cleave tRNA predominantly in the anticodon loop
since this is the most exposed part of the tRNA. One possibility is that the
mutations in the L4 loop alter the positioning of the tRNA in the active site to
preferentially lead to cleave at other sites in the tRNA.
Taken together, we suggest that specific loop regions in the catalytic core are
required for Rny1’s catalytic activity and can play a role in determining the
specific site of RNA cleavage.
## Rny1 can affect tRNA Cleavage in a Vacuole or Vacuole-like Compartment
An unresolved issue is how Rny1 is exposed to its RNA substrates during stress.
One possibility, suggested by the loss of Rny1-GFP from vacuoles during stress,
is that Rny1 is released to the cytoplasm and then can cleave various RNAs.
Alternatively, or possibly in addition, RNAs might be transported into the
vacuole by an autophagy-related process. One prediction of this latter model is
that in the absence of Rny1, RNAs would be transported into vacuole or vacuole-
like compartments, but they would not be degraded due to Rny1’s absence.
Accordingly, increased accumulation of specific RNAs should be observed within
biochemical fractions containing vacuoles.
To test this possibility, we floated cell lysates from early stationary phase
cells (where tRNAs are being cleaved by Rny1) on Ficoll step gradients (process
diagrammed). In this experiment, we compared *rny1*Δ strains either expressing
Rny1 on a functional low-copy plasmid or an empty vector. RNA and proteins were
prepared from equal volumes of each fraction and resolved by urea-acrylamide
electrophoresis and SDS-PAGE, respectively. As expected for membrane
compartments, we observed that ER (Dpm1) and mitochondrial markers (Porin)
floated in the 8 and 8–12% fractions (although some remained in the input
pellet). In contrast, we observed that the vacuolar marker Cpy1 was distributed
across the gradient suggesting the presence of a diversity of vacuoles with
different densities. Consistent with this interpretation, we observed that
intact vacuoles were present in each fraction as judged by staining with a
vacuole specific dye (MDY-64) and examination of the fractions on a microscope.
Given this, we conclude that vacuoles in yeast have a range of densities,
distribute across the gradient, and the lightest fractions have the purest
vacuole fractions due to the absence of other membrane-bound compartments.
Examination of tRNAs in the wild-type strain revealed that the tRNA cleavage
products were distributed in the 8, 8–12, and 15% fractions. The presence of
these fragments in the 8 and 8–12% fractions suggests that they are either
associated with, or within, a membrane compartment. The absence of these
fragments from the lightest fractions suggests that if they are produced by Rny1
action within vacuoles, those vacuoles are of higher density. Interestingly, in
the *rny1*Δ strain, we reproducibly observed an increased level of the full-
length tRNA in the lightest fractions. This suggests that the full-length tRNA
can associate with light vacuoles, and is either fully degraded by Rny1 in that
context, or such vacuoles mature to higher densities in wild -type strains.
Thus, although we cannot rigorously determine the nature of the tRNA-membrane
interactions, these observations demonstrate that both full-length and
fragmented tRNAs can associate with membrane-bound compartments.
# Discussion
Our results reveal that multiple features of Rny1 contribute to its function in
growth inhibition. Deleting either the signal sequence, the T2 conserved region,
or the fungal C-terminal extension partially alleviated the growth defects with
Rny1 over-expression. These effects could not be attributed to a loss of protein
expression since mutant proteins were expressed similarly to the wild-type
protein.
In contrast, Rny1’s nuclease activity maps to specific loops within the
conserved T2 region. Loss of the T2 region, but not the C-terminal extension,
inhibited Rny1’s ability to cleave tRNA. Within this region, we identified
specific loops required for cleavage of tRNA (, loops 2, 3, 4, 6, and 7), two of
which (loop 4 and loop 7) align to loops predicted to be involved in nucleotide
binding.
Targeting to membrane compartments is important for Rny1’s cleavage of tRNA.
Rny1’s signal peptide, presumably inserted into the ER during translation, is
required for cleavage of tRNA. Several possibilities could be envisioned to
explain the role of the signal peptide in regulation of Rny1’s functions. One
possibility is that in the absence of ER targeting, loss of glycosylation
disrupts interactions required for function. This is supported by our evidence
that GFP-Rny1, which fails to be glycosylated, also fails to cleave tRNA and
partially rescues growth inhibition (data not shown). Another possibility is
that the signal peptide directs Rny1’s vacuolar targeting, and this localization
could enable processing of the Rny1 zymogen that renders it active. Lastly, it
is possible that loss of the signal peptide could render an expressed but
structurally impaired protein. In this case, our evidence further strengthens a
model for a veritable catalytic-independent function for Rny1, requiring
structural integrity, since a protein lacking the signal peptide partially
alleviates growth inhibition.
Our work raises the possibility that some tRNAs are taken up into vacuoles, or
associate with vacuoles, for degradation by Rny1. Our analysis of vacuoles on
Ficoll gradients revealed that specific fractions are enriched for CPY, a
vacuolar protein, but not mitochondrial or ER proteins (fractions 2 and 4, lower
panels). These same vacuolar fractions clearly displayed tRNA in an *rny1*Δ
strain (, fractions 2 and 4, upper panels), and expression of a functional form
of Rny1 nearly eliminated detection of tRNA in these fractions (fractions 2 and
4, upper panels).
The fragments of tRNA cleavage accumulated in denser fractions, containing CPY
and other organellar markers. Our microscopic analysis of Ficoll fractions from
these gradients revealed that intact vacuoles partition throughout the gradient
(data not shown) consistent with the observation that CPY signal is found
throughout the gradient. We speculate that tRNA fragments might associate with a
denser form of vacuoles, perhaps associated with additional proteins recruited
in an Rny1-dependent manner, which are not easily resolved from ER and
mitochondrial markers. In the absence of Rny1, undigested tRNA might associate
with lighter vacuoles more easily resolved from other cellular organelles.
Our evidence suggests that Rny1 might participate in a new form of tRNA
ribophagy. To our knowledge, this is first example of tRNAs being targeted for
vacuolar association or uptake. Prior work has demonstrated that rRNA
accumulates within the lysosomes of neurons deficient for RNASET2 and that
ribosomal proteins traffic to vacuoles during autophagic conditions. While tRNA
cleavage does not require autophagy proteins, our work supports a model whereby
tRNAs are turned over at, or within, vacuoles in a T2 ribonuclease-dependent
manner during nutrient-limiting conditions, perhaps utilizing novel targeting
mechanisms that do not include normal autophagy.
It is still possible that Rny1 can exit vacuoles to cleave RNA within the
cytosol, in addition to acting as a nuclease within vacuoles. In previous work,
our lab observed decreased vacuolar signal for the Rny1-GFP fusion protein
during oxidative stress, supporting a model for translocation of Rny1 to the
cytosol to contact RNA substrates. This translocation resembles that observed
for the predominant mitochondrial nuclease, Nuc1, which exits mitochondria
during oxidative stress to modulate nuclear and possibly cytoplasmic RNA
degradation. Likewise, in mammalian cells, lysosomal cathepsins can enter the
cytosol during oxidative stress and other stresses cumulating in cell death
(reviewed in). Hence, we speculate that in our studies of stationary phase
conditions, Rny1 acts within vacuoles, and oxidative stress could trigger Rny1’s
translocation similarly to Nuc1 and lysosomal cathepsins.
Our work provides two important implications for general T2 ribonuclease
functions. First, our studies suggest that glycosylation and membrane targeting
of T2 ribonucleases could be important for nuclease and toxic functions and
could regulate self-recognition and immune cell interactions. In the case of
RNASET2, whose catalytic mutants inhibit tumorigenesis by recruiting competent
immune cells to tumor sites, plasma membrane trafficking (evidenced) and
glycosylation of the T2 protein might enable immune surveillance at the cell
surface. Second, our studies suggest that tRNA, in addition to rRNA, might
accumulate within acidic organelles of cells deficient for T2 ribonuclease
function during stress. Since ribosomal proteins are also taken up into vacuoles
and degraded during similar cellular stress, it is possible that translation
complex proteins and RNA are turned over at vacuoles to control translation,
through vacuolar protease and T2 ribonuclease functions. Thus, in diseases
arising from RNASET2 deficiency, it is possible that regulation of protein
synthesis is aberrant and drives pathogenesis.
# Materials and Methods
## Yeast Strains and Growth Conditions
Yeast strains and plasmids used in this study are described in. Cells were grown
at 30°C in all experiments. For experiments over-expressing Rny1, pRP1584 and
pRP1587 were used with pRP861 as a vector control. Cells were grown in selective
synthetic media containing 2% sucrose to saturation, and these cultures were
pelleted, aspirated, and diluted (OD600 = 0.1) in selective synthetic media with
2% galactose as the sole carbon source and grown to early midlog
(OD600 = 0.3–0.5). For experiments to analyze tRNA fragment cleavage, cells were
grown to saturation in media specified in figure legends (either selective
synthetic media or yeast extract/peptone media both containing 2% dextrose),
then diluted and grown to early midlog. Early midlog times were recorded, and
cells were grown three days from this time to represent stationary phase growth.
For frog ponding, yeast strains were patched to selective media plates, diluted
to OD600 = 0.1 in selective media containing 2% sucrose in the first column of a
96-well plate, then diluted by 10-fold across into four additional columns.
These columns were directly plated without additional growth.
## Plasmids and Plasmid Construction
Plasmids used in this study are listed in. The *GAL RNY1 2µ* plasmid (pRP1584)
and its counterpart plasmid containing mutations to produce catalytically
inactive Rny1 (pRP1587) were the templates utilized to generate *cis* mutants by
PCR using primers for site-directed mutagenesis.
## RNA Analyses
Total RNA was prepared from liquid nitrogen flash frozen pellets. For all RNA
analyses excepting that of Ficoll fractions, a hot acid phenol preparation was
used. All steps prior to acid phenol addition were performed at 4°C. Samples
were suspended in TNE buffer (50 mM Tris-Cl pH7.4, 100 mM NaCl, 10 mM EDTA),
lysed with beads (two one-minute high speed vortexes interrupted by a one-minute
incubation on ice to prevent overheating), then vortexed with SDS added to 1%
and an equivalent volume of acid phenol chloroform. Vortexing was repeated, then
samples were heated to 65°C for seven minutes, followed by additional vortexing.
After acid phenol chloroform extraction, an additional acid phenol chloroform
extraction, and one chloroform extraction, RNA was precipitated, washed, dried,
and resuspended in deionized formamide. For Ficoll flotation assays, equal
amounts of RNA were prepared (400 µl) using TriZol LS reagent (Invitrogen, Grand
Island, NY, USA) following the manufacturer’s protocol, and pellets were
resuspended in deionized formamide. Equal amounts of RNA (20 µg) as determined
by A260, or equal amounts of RNA prepared from equal volumes (for analyses of
floating RNA on Ficoll gradients), were resolved on 10% acrylamide, 47% urea,
1XTBE gels next to HinfI-digested, alkaline phosphatase treated,
γ-<sup>32</sup>P-5′ end-labelled PhiX174 markers. Electrophoretic transfer to
positively charged nylon membranes was performed in 0.5XTBE buffer. Blots were
UV cross-linked twice, prewashed once at 65°C, using 0.1%SDS 0.1XSSC, then
prehybridized in 6XSSC 0.1%SDS 10X Denhardt’s at 42°C. Hybridization with
γ-<sup>32</sup>P-5′ end-labelled probes was performed in the prehybridization
buffer. Blots were washed with 6XSSC 0.1%SDS and placed against phosphor screens
to expose, and screens were scanned into a Typhoon scanner (GE Healthcare,
Piscataway, NJ, USA) and quantitated using ImageQuant software.
## Protein Analyses
Lysates were prepared from harvested pellets of cells lysed using 5 M urea,
boiled, then vortexed in glass beads for 5 minutes. A solution of 125 mM Tris-Cl
pH6.8, 2% SDS was added at 2.5x the volume of 5 M urea used, and this was
vortexed into the mixture, then samples were boiled a second time. Collected
lysate was clarified by spinning at 16000 RCF, and the supernatant was harvested
for analysis. Cleared lysates, or proteins analyzed from Ficoll fractions, were
suspended in protein loading buffer (0.05 M Tris pH6.5, 1%SDS, 0.01% bromophenol
blue, 10% glycerol), boiled, and run on a 10% Tris-SDS acrylamide gel next to
SeeBlue Plus2 protein molecular weight standards (Invitrogen, Grand Island, NY,
USA). Gels were transferred to nitrocellulose and probed using standard Western
blotting protocols. Antibodies used were supplied by Invitrogen Molecular
Probes, Grand Island, NY, USA (CPY, Dpm1, Porin), Novagen, Madison, WI, USA (His
tag detection of Rny1), and Covance, Princeton, NJ, USA (GFP) all used with an
anti-mouse secondary coupled to HRP (Sigma, St. Louis, MO, USA, \#A4416).
Signals were revealed using Pierce (Rockford, IL, USA) SuperSignal West Dura and
exposing the blots to film and developing in a film processor (Konica, Mahwah,
NJ, USA). Films were scanned into.tif format using an HP Scanjet Pro flatbed
scanner (Hewlett Packard, Palo Alto, CA, USA), and images were analyzed using a
cross-reactive band as a loading control.
## PNGase F Digests
Wild-type strains expressing Rny1-GFP or GFP-Rny1 were harvested at midlog after
continuous growth in synthetic selective media containing galactose for
expression. Total lysates were prepared and quantitated for proteins by Bio-Rad
(Hercules, CA, USA) assay. Reactions utilized NEB reagents (Ipswich, MA, USA).
Two reactions, each containing 20 µg of proteins, were prepared for each sample,
diluting into glycoprotein denaturing buffer. Digests were first denatured (10′
at 100°C), then G7 reaction buffer, NP40, PNGase F or ddH20 were added according
to the NEB protocol. Reactions were incubated at 37°C for 1 hour, then protein
loading buffer was added, and samples were resolved by SDS-PAGE and analyzed by
Western blot (performed as indicated in Protein analysis) probing for GFP
(Covance Princeton, NJ, USA).
## Preparation of Vacuoles by Ficoll Flotation Gradients
Vacuoles were floated on Ficoll gradients using a modified version of the
protocol described here. For our experiments, we diluted 40 ml saturated starter
cultures growing in synthetic media supplemented with complete amino acids and
2% dextrose (complete+dex) to 0.1 OD600/ml in 300 ml of fresh complete+dex.
Cells were grown in 1L flasks shaking at 30°C, and the time of early midlog
(0.3–0.4 OD600/ml) was noted. Twenty-four hours from this point, cells were
harvested for analysis, and the OD600/ml was noted. Equal amounts of OD600 units
(631) were harvested and processed for each sample. Cell walls were rendered
susceptible to spheroplasting by resuspending in 30 ml of DTT solution
(detailed) and incubating at 30°C for ten minutes, then harvested cells were
spheroplasted using 18 ml of spheroplasting buffer containing 15,000 units of
lyticase (Sigma) incubated at 30°C for 35′–45′. The A800 readings in water were
observed and noted to determine optimal times for harvesting spheroplasts
(10-fold reduction in A800), but all samples were harvested at the same time to
avoid differences in processing. After harvesting spheroplasts, dextran removal
of plasma membranes was performed as described suspending in 15% Ficoll and
adding 480 µl of 0.4 mg/ml DEAE Dextran (Sigma, St. Louis, MO, USA). In our
experiments, we applied 3 ml of our lysate to the bottom of a pre-chilled SW40
tube (#331374 Beckman tubes), and we layered 2.5 ml of each of the following
solutions in the following order on top: 12%, 8%, 4%, and 0%. Gradients were
ultracentrifuged for 90 minutes at 110,000*xg* using a pre-chilled SW40 swinging
bucket rotor. All steps following spheroplasting were performed in a 4°C room,
and all tubes were prechilled. Following ultracentrifugation, 1 ml fractions
were collected from the top and placed in protease inhibitors, using the 20xpic
stock diluted to 1x for RNA fractions. From each fraction, an equivalent volume
was removed to another tube for protein analysis, and this tube was pre-loaded
with chilled complete EDTA-free protease inhibitors (Roche, Basel, Switzerland)
to yield 1X, prepared from a 7X stock in PS buffer. All of the samples, except
for aliquots reserved on ice in a 4°C room for analysis of vacuoles and
ribonuclease protection assays, were flash frozen in liquid nitrogen and stored
at −80°C. Separate tubes for protein analysis and for RNA analysis ensured that
thawing was only performed once (in a 4°C room) to prepare RNA or resolve
proteins on gels. Purified vacuoles were examined using MDY-64 (Invitrogen
Molecular Probes, Grand Island, NY, USA) label and found to be intact using the
GFP filter on a Delta Vision microscope.
We thank Dr. Pamela A. Silver for providing the *GAL 2 µ* plasmid used in our
studies (pRP861), Dr. Greg Odorizzi for valuable discussions relating to our
vacuolar analyses, Dr. Carolyn Decker for aiding in the completion of
experiments, and all members of the Parker lab for their valuable feedback in
the preparation of this manuscript.
[^1]: Conceived and designed the experiments: NL RP. Performed the
experiments: NL. Analyzed the data: NL RP. Contributed
reagents/materials/analysis tools: RP. Wrote the paper: NL RP.
[^2]: The authors have declared that no competing interests exist. |
# Introduction
Since the discovery of toll-like receptors (TLRs), we have come to appreciate
their crucial role in the induction of adaptive immunity against pathogens. In
response to microbes, the engagement of pathogen associated molecular patterns
by TLRs initiates a complex process of antigen presenting cell (APC) maturation
and pro-inflammatory cytokine production. These intricately coordinated cellular
processes are instrumental to the functional differentiation of pathogen-
specific T cells. Although it is suggested that stimulation via different TLRs
is sufficient to promote distinct immune responses, very little is known about
how this program is set by mature DCs. This issue is particularly difficult to
dissect *in vivo*, since it's not possible to directly compare stepwise events
that occur upon bacterial versus viral infections.
Previous studies have shown that the maturation of DCs is a key event that
promotes autoimmunity in a variety of models. Evidence indicates that, TLR3 and
TLR7 stimulation are critical for the activation and recruitment of autoreactive
CD8<sup>+</sup> T cells and subsequent destruction of pancreatic islet β cells.
Activation of APCs via TLR4 or TLR9 can also disrupt self-tolerance and result
in the induction of EAE. Likewise, TLR3, TLR4 and TLR9 appear to play a critical
role in the development of autoimmune myocarditis. While infection is a well
recognized trigger of autoimmunity, recent studies suggest that endogenous TLR
ligands expand TLR signaling capacity and may therefore play a role in
autoimmune disorders including those arising from sterile inflammation.
Stimulation of different PRR has the potential to induce a particular cocktail
of pro-inflammatory cytokines. A given cytokine milieu may then direct the
initiation of distinct adaptive immune responses. Hirschfeld et al. along with
subsequent studies shed light on the cytokine specificity of TLR2 and TLR4 and
its impact on T helper cell differentiation. Following TLR4 engagement by
lipopolysaccharide (LPS), murine macrophages produce large amounts of tumor
necrosis factor alpha (TNFα), interleukin (IL)-1β, IL-12 and IP-10 which
selectively induced a Th1 response. In contrast, peptidoglycan (PGN) engagement
of TLR2 induces moderate production of TNFα and IL-1β, without IL-12 and IP-10,
biasing toward a Th2 response.
Evidence suggests that differential adaptor engagement and downstream signaling
by different TLRs can contribute to the production of a given cytokine milieu.
For instance, TLR3 and TLR4 can signal via TRIF and activate IRF3 dependent
production of IFNβ. TLR sensors of nucleic acids, TLR3, TLR7, TLR8 and TLR9,
have the ability of producing both IFNα and IFNβ via IRF3 and IRF7 activation.
In addition to differential adaptor engagement and downstream signaling,
qualitative parameters of the TLR ligand interaction itself may contribute to
the induction of certain cytokines, as in the case of TLR9 signaling where
distinct endosomal trafficking of A type but not B type CpG result in IRF-7 and
IFNα production.
One key cytokine that is produced upon DC maturation is the pro-inflammatory
cytokine IL-12. IL-12, a heterodimeric cytokine consisting of the IL-12p35 and
IL-12p40 subunits, is readily produced by stimuli such as double stranded RNA,
LPS, flagellin, single stranded RNA and bacterial DNA. Although evidence
suggests that IL-12 plays an important role and has been referred to as signal
3, after the TCR induced signal 1 and costimulatory signal 2. The role of IL-12
in the induction of adaptive immune responses remain controversial. In addition
to playing a significant role in the induction of Th1 responses, IL-12 also the
functional maturation of cytotoxic T lymphocytes (CTLs) by augmenting
proliferation, survival and generation of effector molecules such as perforin
and granzymes. Studies have also identified IL-12's role in modulating
lymphocyte trafficking by modulating P-SGL1 expression.
Given IL-12's involvement in promoting multiple facets of T cell immunity, it is
believed to be a key mediator of autoimmunity and has been widely used as a
marker for DC activation in cancer immunotherapy,. Indeed, studies in several
murine experimental models and clinical settings also point to IL-12 playing an
important role in the development organ specific autoimmunity. Analysis in
autoimmune prone murine strains demonstrated that predisposition to autoimmunity
may be attributed to enhanced IL-12 production by APCs due to alterations in the
induction of NF-κB. In an autoimmune myocarditis model, IL-12 facilitates the
differentiation of pathogenic CD8 T cell effectors. In the non-obese diabetic
(NOD) model, the administration of IL-12 or ablation of IL-12 mediated Stat4
signaling can markedly accelerate or completely prevent the onset of diabetes
respectively. Furthermore, blocking IL-12 in patients with active Crohn disease,
which is commonly associated with increased production of IL-12 by APCs, can
induce stable remission.
IL-12's impact on the development of organ specific autoimmunity is however not
decidedly pathogenic across all experimental settings. In some models, IL-12
appears non-pathogenic or even protective. Complicating the interpretation of
these data is the discovery of the anti-inflammatory effects of closely related
cytokine IL-35, which has redundant usage of the IL-12p35 subunit. One potential
contributing factor to IL-12's divergent role in the pathogenesis of autoimmune
disorders may be related to the different molecular mechanisms that are
necessary to initiate different autoimmune diseases. It is possible that in
certain models of autoimmunity, the induction of disease depends on certain
cytokines induced by different PRR ligands.
The current study investigated whether different APC maturation stimuli
influenced the requirement for IL-12 in the induction of a CD8 T cell mediated
autoimmune response *in vivo*. In this report, we used the previously
characterized RIP-gp transgenic model, where the lymphocytic choriomeningitis
virus glycoprotein (LCMV-GP) is expressed in the pancreatic islet β cells. In
this model, T cells specific for the LCMV-GP remain ignorant towards the LCMV-GP
expressed in the pancreas under the steady state. Upon activation, these
autoreactive CD8<sup>+</sup> T cells infiltrate the pancreatic islets and
mediate the destruction of the GP+ β-cells, resulting in diabetes. Recent work
has shown an alternate way to activate endogenous GP-specific T cells using
mature bone marrow derived dendritic cells (BMDCs). We have compared the ability
of Poly I∶C and LPS to mature DCs since TLR3 acts via the adapter TRIF, while
TLR4 uses both TRIF and MyD88. By maturing WT or IL-12 deficient BMDCs with Poly
I∶C or LPS and pulsing mature DCs with GP peptides, the impact of TLR3 or TLR4
stimuli on the requirement for IL-12 in breaking tolerance and inducing
autoimmunity can be evaluated. Unexpectedly our studies showed that autoimmunity
induced with Poly I∶C matured DCs is IL-12 independent, whereas IL-12 was
essential for the induction of autoimmunity using LPS matured DCs. Our analysis
also showed that Poly I∶C stimulated DCs produce high amounts of IFNα, compared
with LPS stimulated DCs. Accordingly, IFNα given together with LPS stimulated
DCs was able to induce autoimmunity in the absence of IL-12. Therefore, IFNα can
overcome the requirement for IL-12 in the induction of T cell mediated pathology
elicited by LPS-matured DCs. These studies clearly demonstrate that different
TLR matured DCs require different cytokines to elicit a functional
immunopathological adaptive immune response. These observations have important
implications for controlling tissue specific autoimmunity and anti-tumor
immunity.
# Results
## TLR induced DC maturation programs the requirement for IL-12 in the induction of autoimmunity
We first explored whether the p40 subunit of IL-12 plays an obligatory role in
the induction of autoimmunity in our model. RIP-gp/p40−/− mice were generated
and infected with LCMV and blood glucose was monitored for 3 weeks.
Approximately 8 to 10 days after LCMV infection, all LCMV-infected RIP-gp/p40−/−
mice were diabetic. There was no significant difference in the kinetics or
incidence of disease in comparison with LCMV-infected RIP-gp/p40+/+ mice.
Therefore, the p40 subunit of IL-12 does not appear to play an essential role in
virus induced autoimmunity.
We further evaluated the role of the p40 subunit using an alternate method to
induce autoimmunity. BMDCs were matured using Poly I∶C or LPS, then pulsed with
both class I and class II LCMV-GP peptides (gp33, gp276, and gp61 peptides) and
given i.v. to RIP-gp transgenic mice. Poly I∶C matured p40−/− BMDCs were as
effective as wild type BMDCs in inducing diabetes. However, the treatment of
RIP-gp/p40−/− mice with LPS matured p40−/−BMDCs showed a reduced incidence of
diabetes, in contrast to the treatment of RIP-gp/p40+/+ mice with LPS-matured
p40+/+ BMDCs. The induction of diabetes using this DC based model is dependent
upon CD8 T cells (Dissanayake D. unpublished). Therefore, the absolute
requirement for the p40 subunit of IL-12 is dependent upon the stimuli that
triggered DC maturation.
Since activated B cells, macrophages, and neutrophils express p40 it remains
possible that the requirement for p40 in the induction diabetes may reflect a
defect in B cell, macrophage, or neutrophil functions. To determine whether the
ability of the DC alone to produce p40 could influence diabetes induction, we
treated RIP-gp/p40+/+ mice with LPS matured p40−/− BMDCs or conversely treated
RIP-gp/p40−/− mice with LPS matured p40+/+ BMDCs. RIP-gp/p40−/− mice treated
with LPS matured p40+/+ BMDCs developed diabetes, but not vice versa clearly
demonstrating that p40 production is required by LPS matured DCs to induce CD8
mediated immune pathology.
Studies have shown that the IL-12p40 subunit can also heterodimerize with the
p19 subunit to form IL-23. IL-23 plays an important pathogenic role in several
autoimmune models via the induction and maintenance of Th17 cells. To determine
whether the lack of diabetes induction can be attributed to the lack of IL-23,
we treated RIP-gp mice with LPS matured p35−/− BMDCs or p19−/− BMDCs. Fifty
percent of RIP-gp mice treated with LPS matured p19−/− BMDCs developed diabetes,
unlike the mice given LPS matured p35−/− BMDCs, indicating that the production
of IL-12 by DCs is critically important for diabetes induction. Together, our
data demonstrate that the different pathogen-related stimuli define the
requirement for IL-12 in the induction of autoimmunity. Specifically, IL-12
production by DCs plays a critical role in diabetes induction when DCs are
matured with LPS, but not with Poly I∶C treatment or LCMV infection.
## Reduced T cell infiltration in mice treated with LPS matured DCs
To gain insights into the mechanism through which LPS matured DCs have a limited
capacity to induce diabetes, the pancreas was evaluated by immunohistochemistry.
Five days after RIP-GP mice were given WT or p40−/− LPS matured DCs, the
pancreas was taken and stained for CD8 T cell infiltration. RIP-GP mice that
were given LPS treated p40−/− DCs showed reduced CD8+ T cell inflammation in the
islets which correlated with a reduced incidence of diabetes. A series of
experiments were done to evaluate why there may be reduced CD8 inflammation.
## Induction of T cell responses in the absence of p40
IL-12 has previously been described to be important for T cell survival and the
induction of effector functions. Therefore, we examined the impact of IL-12 on
LCMV-GP specific T cell responses. CFSE labeled P14 T cells specific for gp33
peptide in the context of H-2<sup>b</sup> were stimulated with gp33 peptide
pulsed, LPS or Poly I∶C matured p40+/+ and p40−/− BMDCs for 3 days *in vitro*.
In the absence of IL-12 secretion by DCs, antigen-specific T cell proliferation
and upregulation of the activation marker CD44 was unaffected. Furthermore, the
lack of IL-12 did not enhance T cell apoptosis since the proportion of CD8 T
cells that were CFSE-7AAD+ were comparable between co-cultures with p40+/+ and
p40−/− BMDCs under all stimulation conditions.
To determine whether T cell functional maturation was impaired in the absence of
IL-12, we evaluated cytotoxic activity in DC primed mice. C57BL/6 mice were
treated with LPS or Poly I∶C stimulated p40+/+ and p40−/− BMDCs pulsed with gp33
peptide. gp33 specific cytolytic activity was assessed *in vivo* on day 5 by
quantifying the ratio of gp33 peptide pulsed targets to AV peptide pulsed
control target cells remaining in the spleen of primed C57BL/6 mice. C57BL/6
mice given Poly I∶C stimulated p40−/− BMDCs showed gp33-specific cytolytic
activity comparable to C57BL/6 mice treated with Poly I∶C stimulated p40+/+
BMDCs. Interestingly, C57BL/6 mice given LPS stimulated p40−/− BMDCs had higher
gp33 specific cytolytic activity compared with C57BL/6 mice treated with LPS
stimulated p40+/+ BMDCs. Therefore changes in CD8+ T cell activation, function
or survival was unlikely to account for differences observed in the infiltration
of the pancreas of mice treated with LPS matured p40−/− DCs.
## Evaluating DC cytokine production in the absence of p40
Studies have shown that the induction of pro-inflammatory mediators by DCs plays
a critical role in promoting the maturation and homing of T cells. To determine
whether the induction of autoimmunity by LPS or Poly I∶C matured DCs was linked
to the production of different cytokines, we compared the cytokine production
from p40+/+ and p40−/− BMDC's. 12 hours after stimulation with LPS, p40+/+ BMDCs
exhibit robust production of IL-12. Although Poly I∶C induced IL-12 production,
the response is clearly less than that observed after LPS stimulation. Notably,
the induction of intracellular levels of TNFα was similar upon LPS versus Poly
I∶C stimulation. Therefore IL-12 is differentially induced in LPS matured DCs
compared with Poly I∶C maturation, and IL-12 was essential for the induction of
autoimmunity in our model after LPS -induced DC maturation.
Cytokine production was also evaluated in the supernatant from these cultures
using a cytometric bead assay. p40+/+ and p40−/− BMDCs stimulated with LPS or
Poly I∶C showed an approximate 10,000 fold and 1,000 fold increase in the
production of IL-6 compared to unstimulated control cultures respectively, but
revealed no significant difference in the production of IL-6 in the absence of
p40. P40+/+ and p40−/− BMDCs stimulated with LPS and Poly I∶C induced an
approximate 200 fold and 20 fold increase in the production of TNFα
respectively. In the absence of IL-12, LPS stimulated p40−/− BMDCs had a slight
reduction in TNFα production detectable by the cytometric bead assay that was
not observed by intracellular cytokine staining (T-test p\<0.05). No difference
in TNFα production was observed between poly I∶C stimulated WT and p40−/− DCs.
We also examined the production of the anti-inflammatory cytokine IL-10 and
chemotactic factor MCP-1 (data not shown) but did not find a significance
difference in production with either stimulation conditions in the presence or
absence of IL-12. Similar to observations by others, there appears to be a
qualitative difference in the ability of LPS and Poly I∶C to induce IL-10
production, as only LPS stimulation induced significant IL-10 production.
Although these studies were done to examine whether the absence of p40 would
have a negative impact on cytokine production, we found that the intrinsic
ability of LPS versus Poly I∶C to stimulate cytokines had more influence on the
levels of cytokines that were produced, than the absence of p40.
## IFNα can promote autoimmunity in the absence of IL-12
Since the absence of p40 did not result in a profound reduction in cytokine
secretion by BMDCs, we decided to evaluate whether different levels of cytokines
were made upon LPS versus Poly I∶C stimulation of BMDCs. It is possible that
cytokines that are produced by DCs after Poly I∶C stimulation, are able to
bypass the requirement for IL-12. One significant qualitative difference noted
in the literature between TLR4 and TLR3 stimulation is the robust induction of
IFNα by TLR3. Considering IFNα has been described to negatively regulate IL-12
production and our findings that diabetes induction is independent of IL-12 when
autoreactive T cells are activated when IFNα is significantly represented in the
cytokine milieu (with LCMV infection or Poly I∶C stimulation), the possibility
arises that IFNα induction is critical and may overcome the requirement for
IL-12 in LPS matured DCs. Indeed, in characterizing the IFNα response in our
model, we observed robust production of IFNα, 12 hours after stimulating BMDCs
with Poly I∶C but not LPS. However, we found no evidence of reciprocal co-
regulation of IFNα by IL-12 as IFNα production was not significantly enhanced in
the absence of IL-12. Therefore, it is clear that Poly I∶C stimulation leads to
high levels of IFNα production that is independent of p40 in our culture
conditions, and thus IFNα may have a critical role in promoting autoimmunity in
this model.
To investigate whether IFNα can overcome the requirement for IL-12 in LPS
matured DCs, RIP-gp mice were given LPS matured p40−/− DCs with various
combinations of IFNα. IFNα was included in BMDC cultures to promote DC
maturation and/or administered *in vivo* 3 days after DC treatment, to evaluate
the impact on the induction of diabetes. The following combinations were tested:
LPS matured p40−/−DCs, LPS plus IFNα matured p40−/−DCs, LPS matured p40−/−DCs
with IFNα treatment 3 days later, or LPS plus IFNα matured p40−/−DCs with IFNα
treatment 3 days later. As in previous experiments, RIP-gp/p40+/+ hosts given
LPS matured p40−/−DCs did not lead to the induction of autoimmunity. Although
nearly a quarter of RIP-gp/p40+/+ hosts receiving LPS plus IFNα matured
p40−/−DCs became diabetic, the effect of IFNα on diabetes induction had the most
impact when RIP-gp/p40+/+ mice were given IFNα 3 days after treatment with LPS
matured or LPS plus IFNα matured p40−/−DC. Treatment of RIP/p40+/+ mice with
IFNα alone was not sufficient to induce diabetes. These results suggest that
IFNα is a critical cytokine that can circumvent the requirement for IL-12 and
therefore is a key mediator of autoimmune pathology in the absence of IL-12.
Furthermore, our results suggest that the effects of IFNα's may be mediated by
modifying the host's T cells or target organ in addition to directly enhancing
the immunostimulatory activities of IL-12 deficient DCs.
IFNα has previously been described to enhance the effector functions and
survival of CD8 T cells. To assess whether IFNα circumvents the LPS specific
requirement for IL-12 by enhancing host's T cell responses, we examined the *in
vivo* CTL activity of p40+/+ C57BL/6 primed with LPS plus IFNα stimulated
peptide pulsed p40−/− BMDCs or LPS stimulated p40−/− BMDCs followed by the
administration of exogenous IFNα to the host 3 days later. However, we did not
observe a change in gp33 specific CTL activity in either treatment condition
(data not shown). Furthermore, in lethally irradiated RIP-gp mice reconstituted
with IFNαR−/− bone marrow, treatment with LPS stimulated p40−/− BMDCs followed
by the administration of IFNα 3 days later induced autoimmunity with similar
kinetics and incidence of disease compared to RIP-GP/p40+/+ mice reconstituted
with IFNαR+/+ bone marrow. These data indicate that exogenous IFNα does not act
directly on host T cells or on secondary events involving DC populations (such
as cross priming) and acts by promoting inflammation.
To determine whether IFNα plays a role in promoting an inflammatory response in
the absence of IL-12, RIP-gp hosts were given LPS matured p40+/+ BMDCs or LPS
matured p40−/− BMDCs and CD8+ T cell infiltration in the pancreas was examined
five days post-treatment by immunohistochemistry. In the absence of IL-12, RIP-
gp mice treated with LPS matured p40−/− BMDCs have clearly reduced CD8 T cell
infiltration in the pancreatic islets. The CD8 T cell infiltration was both
antigen specific and dependent on the activation of BMDCs as infiltration was
not observed in C57BL/6 mice treated with LPS matured p40+/+ BMDCs (no peptide)
or in RIP-gp mice given immature peptide pulsed p40−/− or p40+/+ BMDCs.
Furthermore, when RIP-gp mice were treated with exogenous IFNα 3 days after
injection with LPS matured p40−/− BMDCs, autoreactive CD8 T cell infiltration
was enhanced and comparable to that observed in RIP-gp mice injected with LPS
matured p40+/+ BMDCs. Quantitative analysis of the degree of infiltration showed
that that percent of islets with heavy infiltration is similar to mice immunized
with WT BMDCs. Together, our results indicate that IFNα can compensate for
IL-12's important role in diabetes induced by LPS matured DCs by promoting CD8+
T cell infiltration.
# Discussion
With this model, we can evaluate the impact of different TLR signals on the
induction of adaptive immunity *in vivo*, using the same population of DCs and
the same defined antigens. We have uncovered novel insights for how different
TLR maturation signals influence the induction of autoimmunity. We demonstrated
that the transfer of TLR3 or TLR4 matured DCs presenting self-antigens promotes
CD8 T cell mediated autoimmune responses and overt organ specific autoimmunity
*in vivo*. Our findings add to the growing body of evidence implicating a wide
breadth of TLRs in the pathogenesis of autoimmune disorders and support the
rationale behind targeting TLRs to promote tumor specific immune responses. Our
investigation also revealed previously unappreciated dynamics between TLR
signaling and their pro-inflammatory mediators in the pathogenesis of
autoimmunity. Specifically, our findings highlight the crucial influence of the
type of stimuli used to promote DC maturation and the differential requirement
for IL-12 in the pathogenesis of autoimmunity *in vivo*.
## Importance of IL-12 in the induction of autoimmunity is dependent upon TLR stimulation
Within the context of the dialogue between APCs and naïve T cells, along with
antigen and co-stimulation, IL-12 produced by APCs is thought to play a crucial
instructive role in determining the cellular fate of naïve T cells and has thus
been coined the term “Signal 3”. In contrast to this work, our study has shown
IL-12 surprisingly plays little role in mediating the proliferation, activation
or survival of CD8 T cells. Furthermore, IL-12 appears to have minimal impact on
the functional differentiation of CTLs *in vivo*. Our findings suggest a rather
surprisingly minimal physiological role for IL-12 in these facets of CD8 T cell
functional maturation *in vivo*.
In spite of the normal induction of CTLs after stimulation *in vivo* with p40
deficient DCs, TLR4 mediated induction of autoimmune diabetes is completely
abrogated in the absence of IL-12. An examination of the severity of
inflammation in the pancreatic β islets cells indicated that IL-12 is an
important component of the inflammatory milieu induced by TLR4 signaling, and
promotes CD8 T cell inflammation. Previous studies have found IL-12 to be a
critical pathogenic factor in promoting local inflammation and diabetes. In
addition, during the priming stage of naïve T cells, IL-12 may play a direct or
indirect role, i.e. via regulating the expression of co-stimulatory molecules,
in modulating the expression of homing and trafficking molecules by CD8 T cells.
As our experiments are done in p40 sufficient RIP-gp hosts, our findings suggest
DC-produced IL-12 plays an important role in programming CD8 T cell trafficking
during early encounter with cognate antigen bearing DCs.
In stark contrast to the importance of IL-12 in TLR4 mediated DC maturation and
subsequent induction of diabetes, IL-12 was not required in TLR3 mediated DC
induction of autoimmunity. IL-12 was not an essential component of the cytokine
milieu induced after TLR3- induced DC maturation and was not required to
facilitate CD8 T cell proliferation, survival, functional maturation and homing
to target tissues. The emerging paradigm from studies with infectious agents
suggests that IL-12 plays a pivotal role in promoting T cell responses to
bacterial and parasitic infections, while it is not required in some viral
infections. Our model provides a direct way to evaluate the differences in
bacterial induced DC maturation (LPS) versus virus induced DC maturation (Poly
I∶C). It is not possible to directly compare the induction of immunity by
various pathogens because they may have different inherent ways to mature the DC
*in vivo*. Furthermore, these pathogens will present a multitude of antigens
that are presented in different contexts and different doses. Using our model,
we can directly compare the *in vivo* consequence of LPS versus Poly I∶C mature
DCs using the same antigens. Our data directly demonstrate that IL-12 is
required for the induction of autoimmunity using LPS matured DCs but not Poly
I∶C matured DCs, and support the hypothesis that IL-12 is required for bacterial
but not viral infections.
## Predominant role for IFNα in promoting tissue specific T cell responses
In order to explain why TLR 3 was not dependent upon IL-12 producing DCs to
induce CD8+ T cell mediated pathology, we hypothesized that other components of
the inflammatory milieu induced by TLR3 stimulation may functionally compensate
for IL-12. Although we found no conclusive evidence of compensatory upregulation
of proinflammatory mediators such as TNFα, IL-6 and MCP-1, or downregulation of
anti-inflammatory cytokines such as IL-10, we detected a clear quantitative
difference in IFNα production between TLR3 and TLR4 stimulated BMDCs. Previous
studies have suggested that there may be functional redundancy between IL-12 and
IFNα/β signaling in facilitating T cell responses against some viral infections.
Other studies have shown that IFNα but not IL-12 p40 subunit is critical for the
induction of CD4+ Th1 responses after systemic administration of Poly I∶C. Using
this model, our results indicate that in the absence of IL-12, IFNα is a
critical pathogenic mediator of autoimmunity *in vivo*.
In contrast with previous studies which have shown IFNα directly enhances T cell
survival and effector functions, IFNα does not act directly on T cells to
mediate CD8 T cell mediated autoimmune responses in our model. Neither did we
find evidence of enhanced CTL activity in mice treated with exogenous IFNα. In
SLE patients, the heightened level of IFNα in their sera is thought to
contribute to the breakdown of peripheral tolerance via the induction of DC
differentiation. In contrast, studies have shown that DC treatment with
proinflammatory cytokines are not sufficient to promote functional T helper
immunity, while other studies have shown that Poly I∶C treatment can promote
autoimmunity, only when islet cells are engineered to express CD80. Our results
suggest IL-12 and IFNα both act to promote CD8 T cell trafficking and local
inflammation to the target organ. This is supported by previous observations of
IFNα's critical role in autoimmunity by promoting inflammation and upregulating
MHC class I molecules on the target tissues. It is likely that IFNα influences
immunity in multiple ways in different models because of the variety of ways it
has been shown to impact the immune response.
Although we have not examined the relative requirements for IFNα for the
induction of diabetes in this model by performing the reciprocal experiments
with IFNα deficient DCs matured in different activation contexts, studies
suggest that IFNα is likely a required factor in the context of TLR activation
associated with viral infections. However, the absolute requirement for IFNα
production by DCs matured with different TLR stimuli has yet to be elucidated in
our model.
## Perspectives
In the classical paradigm of DC-T cell interaction, DCs exist in 2 basic
functional states, immature and mature. Naïve T cells that encounter antigen on
immature DCs become tolerized while those that encounter antigen on mature DCs
become activated. However, current evidence extends this notion and suggests
that DCs exist in ‘states’ other than simply immature or mature and that these
states can be modified by other factors or in some cases other subsets of cells.
One would predict that maturation with different pattern recognition motifs
should induce a different type of DC and a corresponding pathogen related immune
response. Our findings begin to unravel some of these intricacies, and using the
same peptides and same DC subset, show that LPS matured DCs are functionally
different from Poly I∶C matured DCs in their ability to induce autoimmunity *in
vivo*. Furthermore, our studies suggest that IL-12 may not be an appropriate
surrogate marker for functionally mature DCs and the relevance of using IL-12 to
extrapolate immunogenicity of DC vaccines in tumor immunotherapy should be
reconsidered. Furthermore, therapeutic targeting of IL-12 for treatment of
autoimmunity should also take into consideration the potential importance of
IFNα in disease progression.
# Materials and Methods
## Ethics Statement
This study was carried out in strict accordance with the recommendations made by
the Canadian Council for Animal Care. The protocol was approved by the Animal
Care Committee at the PMH/OCI institute, protocol number 929.
## Mice and diabetes monitoring
All mice used in the experiments are on the C57BL/6 background. Wild type
C57BL/6, p35−/− and p40−/− mice were purchased from the Jackson laboratory. RIP-
gp and P14 TCR transgenic mice, were maintained in our animal facility according
to institutional guidelines. P19−/− and IFNαR−/− mice were kind gifts from N.
Ghilardi and D. Pinschewer. Genotyping for all mice were performed by PCR.
Diabetes induction was monitored by blood glucose measurements prior to
treatment and then 2–3 times per week after treatment. Blood glucose levels were
measured using Accu-Chek III glucometers and Chemstrips (Roche).
## LCMV infection
LCMV WE strain was originally obtained from F. Lehmann-Grube and grown on L929
cells and titrated as previously described. RIP-gp/p40+/+ and RIP-gp/p40−/− mice
were infected with 3000 PFU of LCMV WE and monitored for alterations in blood
glucose.
## BMDC culture
Bone marrow was extracted from the femur and tibia passed through a screen and
washed in cold media. 2×10<sup>6</sup> bone marrow cells were resuspended in 10
ml of DC media (RPMI, 10% LPS free FCS, β-mercaptoethanol, L-glutamine, 40 ng/ml
GM-CSF(PeproTech)) and cultured in 10 cm petri dishes at 37°C. On day 3, 10 ml
of DC media was added to the culture. On day 6 and 8, 10 ml of culture was
removed, resuspended in 10 ml of fresh DC media and added back to the culture.
On day 9, BMDCs were resuspended in DC media<sup>−</sup> (ex GM-CSF) at
2×10<sup>6</sup>cell/ml, re-plated in 24 well plates and stimulated with LPS
(100 ng/ml), Poly I∶C (100 µg/ml), and/or IFNα (1000 u/ml). On day 10, BMDCs
were pulsed for 2 hours with gp33 peptide alone or gp33, gp276 and gp61 peptides
(10<sup>−6</sup> M) respectively, and washed prior to use in *in vitro*
proliferation assays or intravenous infusions into treated mice. For intravenous
infusions, 2×10<sup>6</sup> BMDCs prepared in 200 µl of sterile HBSS were given
to each treated mouse. For *in vivo* treatment, mice were infused with 10000 u
of IFNα.
## Flow cytometric analyses
BMDCs, cell cultures and single cell suspensions from spleens were stained with
antibodies specific for CD11c, CD11b, CD80, CD86, MHCI CD8, CD44, TNF and
IL-12p40/p70 flow-cytometry antibodies (BD and eBioscience). Flow cytometry data
was acquired on FACSCanto (BD) and anaylzed with Flowjo (Tree Star).
## In vitro proliferation assay
Single cell suspensions were prepared from the spleen of P14 transgenic mice and
CD8<sup>+</sup> T cells were purified using CD8 negative sort kit (Miltenyi
Biotech). Purified CD8<sup>+</sup> T cells were then labeled with CFSE
(5-\[and-6\]-carboxyfluorescein diacetate, succinimidyl ester). 1×10<sup>5</sup>
CFSE-labeled CD8<sup>+</sup> T cells were co-cultured with 2×10<sup>4</sup> gp33
peptide pulsed BMDCs in 200 µl of complete IMDM in 96 well round-bottom plates
and incubated for 3 days in a 37°C incubator.
## CFSE (5-\[and-6\]-carboxyfluorescein diacetate, succinimidyl ester) labeling
After washing single cell suspensions in serum-free RPMI 1640 (GIBCO BRL), cells
were resuspended in serum-free media at 10<sup>7</sup> cells per 200 µl
containing 10 µM CFSE (Molecular Probes). After incubation for 15 minutes in
37°C incubator, cells were washed in RPMI 1640 containing 10% FCS (Sigma-
Aldrich).
## In vivo CTL assay
RIP-gp mice were injected intravenously with 2×10<sup>6</sup> gp33, gp276 and
gp61 peptide-pulsed, LPS/PolyI∶C stimulated BMDCs. 5 days after the initial
treatment, single cell suspensions from spleens of C57BL/6 mice were pulsed with
gp33 or AV peptide (10<sup>−6</sup> M) and labeled with CFSE at 1.5 µM and 20 µM
respectively. 2×10<sup>7</sup> each of gp33-pulsed and AV-pulsed CFSE-labeled
splenocytes were injected intravenously into each treated RIP-gp mouse. 4 hours
later, these mice were sacrificed and single cell suspensions from their spleens
were analyzed. The percent specific lysis is calculated by dividing the number
of CFSE<sup>Hi</sup> cells by the number of CFSE<sup>Lo</sup> cells.
## Intracellular cytokine staining
Intracellular cytokine staining was performed using the BD Cytofix/Cytoperm kit
as per manufacturer's instructions.
## Cytometric bead array analysis
Media from BMDC cultures were collected after overnight stimulation with LPS or
Poly I∶C. The CBA analysis was performed using the BD CBA assay as per
manufacturer's instructions.
## Bone marrow chimeras
Bone marrow was extracted from IFNαR+/+ and IFNαR−/− donor mice previously
treated with CD4 and CD8 depleting antibodies. 4×10<sup>6</sup> bone marrow
cells were transferred intravenously into irradiated (9 Grays) sex-matched RIP-
gp recipients, and recipients were treated with BMDCs for analysis 6–13 weeks
after reconstitution.
## IFNα assay
Media from BMDC cultures were collected after a 12-hour stimulation with LPS or
Poly I∶C and assayed with mouse IFN Alpha ELISA kit from PBL Interferon Source,
as per manufacturer's instruction.
## Immunohistochemistry
Immunohistochemistry were performed as previously described.
We would like to thank N. Ghilardi and D. Pinschewer for providing mice, members
of the Ohashi lab and our families.
[^1]: Conceived and designed the experiments: AL PO. Performed the
experiments: AL DD SD AE. Analyzed the data: AL PO. Contributed
reagents/materials/analysis tools: DD SD. Wrote the paper: AL PO.
[^2]: The authors have declared that no competing interests exist. |