Datasets:

CShorten

/

CORD-19-prototype

like 0

Unraveling the structural complexity in a single-stranded RNA tail: implications for efficient ligand binding in the prequeuosine riboswitch

Single-stranded RNAs (ssRNAs) are ubiquitous RNA elements that serve diverse functional roles. Much of our understanding of ssRNA conformational behavior is limited to structures in which ssRNA directly engages in tertiary interactions or is recognized by proteins. Little is known about the structural and dynamic behavior of free ssRNAs at atomic resolution. Here, we report the collaborative application of nuclear magnetic resonance (NMR) and replica exchange molecular dynamics (REMD) simulations to characterize the 12 nt ssRNA tail derived from the prequeuosine riboswitch. NMR carbon spin relaxation data and residual dipolar coupling measurements reveal a flexible yet stacked core adopting an A-form-like conformation, with the level of order decreasing toward the terminal ends. An A-to-C mutation within the polyadenine tract alters the observed dynamics consistent with the introduction of a dynamic kink. Pre-ordering of the tail may increase the efficacy of ligand binding above that achieved by a random-coil ssRNA. The REMD simulations recapitulate important trends in the NMR data, but suggest more internal motions than inferred from the NMR analysis. Our study unmasks a previously unappreciated level of complexity in ssRNA, which we believe will also serve as an excellent model system for testing and developing computational force fields.

Single-stranded RNAs (ssRNAs), typically located at the ends of RNA hairpins and consisting of more than three unpaired residues, serve diverse structural and functional roles. They can fold onto neighboring RNA hairpins to form pseudoknots, essential architectural RNA elements involved in ribosomal frameshifting (1, 2) , hepatitis C internal ribosomal entry site (IRES) recognition (3, 4) and telomerase activity (5) . Messenger RNA (mRNA) degradation is prevented or promoted by 3 0 addition of a polyadenylated tail, which recruits essential protein cofactors (6) . Cleavage of the 5 0 transfer RNA (tRNA) leader by RNase P is a key step in tRNA maturation (7) . In riboswitches, ssRNA links the ligand-binding aptamer domain to the expression platform, providing the basis for communication between the two (8) (9) (10) . Much of our understanding of the conformational behavior of ssRNA comes from high-resolution NMR and X-ray structures of RNA, in which ssRNA directly engages in tertiary or RNA-protein interactions. However, the atomic-level structural and dynamic behavior of these elements in the absence of these interactions remains unclear, in large part due to their high degree of flexibility. Several studies suggest that ssRNA polynucleotides adopt stacked and partially helical conformations, particularly adenine-rich sequences; however, the biological relevance of these structures is unclear (11) (12) (13) (14) (15) (16) (17) . Atomic-resolution studies of ssRNA are scarce: at present only one isosequential ssRNA and ssDNA sequence has been characterized by homonuclear NMR methods and shown to possess properties reminiscent of A-form and B-form helices, respectively (18) . Few MD studies have been performed on ssRNA, the majority of which use the AMBER force field (19, 20) to explore the impact of chemical modifications such as peptide nucleic acids (PNA) and O2 0 -methylation (21) (22) (23) (24) . The class I prequeuosine riboswitch (queC), typically found in firmicute bacterial species, is commonly located in the 5 0 -untranslated region (UTR) of the queCDEF operon, which expresses proteins directly involved in the queuosine biosynthetic pathway (25) . The aptamer binds preQ 1 , an intermediate in queuosine synthesis, with high affinity to attenuate protein expression at either the transcription or translation level (25) . This class has the smallest minimal aptamer domain (34 nucleotides, nt) discovered to date, consisting of a small hairpin followed by a 12 nt ssRNA tail ( Figure 1A ). Upon ligand recognition, the highly conserved adenine-rich tail condenses into a pseudoknot, forming a host of interactions to both the hairpin and ligand, including A-minor 'kissing' interactions between the ssRNA polyadenine tract and the minor groove (26) (27) (28) (29) (30) . The activity of transcriptionregulating riboswitches, such as the Bacillus subtilis queC riboswitch, has been shown to depend on the kinetics of ligand binding as well as the rate of transcription (8) . Notably, the very small size of the queC riboswitch leaves very little time, in comparison to other switches, for ligand binding to take place prior to formation of the anti-terminator helix which, when formed, prevents terminator helix formation, thereby allowing gene expression to continue. For example, the B. subtilis FMN riboswitch, which is highly dependent upon the rate of polymerase and contains sites that locally pause polymerase to lengthen the ligand-binding window, has $70 nt between the minimal aptamer sequence and complete formation of the anti-terminator helix (8) . In comparison, the ligand-binding window for the queC riboswitch is $20 nt (26, 27) . How efficient ligand binding is achieved is unclear given that the ssRNA tail is thought to be highly disordered, and therefore capable of sampling a wide range of competing conformations. Here, we use NMR chemical shifts, spin relaxation, and residual dipolar couplings (RDCs) in conjunction with REMD simulations using the recently updated CHARMM27 nucleic acid force field (31, 32) to explore the conformational properties of the 12 nt ssRNA tail from the queC aptamer domain and the impact of a single A-to-C mutation targeting the polyadenine tract. Our study unmasks a previously unappreciated level of complexity in ssRNA and suggests that these structures can serve as excellent model systems for testing and developing computational force fields. Uniformly 13 C/ 15 N-labeled queC36 and C14U/C17U constructs were prepared by in vitro transcription as described previously (33) . Unlabeled wild-type (WT, 5 0 -AUAAAAAACUAA-3 0 ) and A29C (5 0 -AUAACAAA CUAA-3 0 ) RNAs were purchased from Integrated DNA Technologies (IDT) and purified using a C18 column (Waters) followed by lyophilization and reconstitution in NMR buffer (15 mM sodium phosphate, pH 6.4; 25 mM sodium chloride, 0.1 mM EDTA) containing 10% D 2 O by volume. 100% D 2 O samples were prepared by repeatedly lyophilizing the sample and replacing with 99.99% pure D 2 O (Sigma) three times. RNA concentrations ranged from 1.5 to 2.8 mM. AMP, UMP and CMP (Sigma) were directly dissolved into NMR buffer with no additional purification to 5 mM. For RDC measurements, samples were dialyzed into Millipore-purified ddH 2 O using 1 kDa dialysis tubing (Spectrum Labs), lyophilized, and reconstituted into 52.4 mg/ml Pf1 phage solution (34) (35) (36) in NMR buffer with 100% D 2 O (Asla Biotech). RNA concentrations in Pf1 phage ranged from 1.5 to 2 mM. UV/Vis melting RNA samples (0.25-0.5 mM) were prepared in NMR buffer and the melting profiles measured between 275 K and 368 K using a Varian Bio 300 UV/Vis instrument equipped with a Cary Temperature Controller. The absorbance at 260 nm was recorded every 0.5 with a ramp rate of 0.5 /min. The two-state helix to coil melting transition was analyzed using where A is the absorbance value at a given temperature T, A H is the absorbance of the fully helical ssRNA, A C is the absorbance of the fully random coil ssRNA, ÁS and ÁH are the entropy and enthalpy of the melting transition respectively, and R is the gas constant (37, 38) . Absorbance values were fitted to the above equation using the non-linear least squares fitting function in Origin 7 to determine thermodynamic parameters. The melting temperature (T m ) was determined by dividing the enthalpy by the entropy. All NMR experiments were performed on a Avance Bruker 600 MHz NMR spectrometer equipped with a triple-resonance 5-mm cryogenic probe. NOESY experiments were performed at 277 K and 298 K using a mixing time of 350 ms (39) . 13 C spin relaxation experiments were performed at natural abundance and 298 K (40) . Relative order parameters were calculated by normalizing (2R 2 -R 1 ) to either A31 (C8) or C33 (C6). Relaxation parameters were computed using HydroNMR (41, 42) , assuming an idealized A-form structure, to obtain diffusion tensor parameters (t m and D ratio ), and in-house written software was used to compute R 2 /R 1 values as previously reported (33, 40) . Motionally averaged bond lengths of 1.104 Å were used for both C8 and C6 moieties as previously described (40, 43) . The following experimentally derived CSAs (s xx , s yy , s zz ) were used in the analysis: (89, 15, À104); (80, 5, À85); and (98.4, 9.2, À107.5) for C2, C8 and C6 moeties (43, 44) . IP-COSY experiments were performed at 277 K and 298 K to observe relative 3 J H1 0 -H2 0 scalar coupling crosspeak intensities (45) . Base and sugar 1 H-13 C splittings were measured from the difference between the upfield and downfield components of the 1 H-13 C doublet along the 1 H component using the narrow transverse relaxation-optimized spectroscopy (TROSY) component in the 13 C dimension as implemented in 2D 1 H-13 C S 3 CT-heteronuclear single quantum correlation (HSQC) experiments (46) . 2 H splittings were 71 and 69 Hz for WT and A29C, respectively. Idealized A-form structures were constructed using Insight II (Molecular Simulations, Inc.) correcting the propeller twist angles from +15 to À15 using an in-house program, as previously described (47) . The complementary strand was removed and the resulting ssRNA used in NMR data analysis. B-form helices were constructed using W3DNA (48) . Simulation. REMD simulations were performed with the CHARMM simulation package (49) using the recently updated CHARMM27 nucleic acid force field (31, 32) and the MMTSB (50) tool set. Each REMD simulation comprised 40 replicas exponentially distributed over a temperature range from 278 K to 330 K, resulting in an average exchange acceptance ratio of 30%. Each replica was first equilibrated for 0.5 ns, restraining nucleotide heavy atoms, and subsequently run without any restraints for 10 ns, with exchange moves attempted every 0.5 ps. Both WT and A29C RNAs were initially built in an ideal A-form helical configuration and served as the starting conformation in every simulation of REMD. The RNA was solvated in an 80-Å cubic box of pre-equilibrated TIP3P water (approximately 50 000 atoms). Twelve pairs of sodium chloride with an additional 11 sodium ions were added to the box, corresponding to the experimental ionic concentration of 40 mM. Analysis. We utilized the last 5 ns of the REMD trajectory at 298 K for the following analysis. Base stacking energies were defined as the electrostatic and van der Waals interaction energies between the adjacent bases. The molecular orientation was expressed by the order parameters S 2 of the C-H bond vectors employing the model-free approach of Lipari and Szabo (51) . After a translational and rotational fit of each RNA snapshot to the ideal A-form helical structure, the order parameters were taken from the plateau phase of the correlation function, given by where P 2 is the second order Legendre polynomial and * is the unit vector along the C-H dipole. Additionally, from the atomic coordinates we constructed the RDC values by first orienting an idealized A-form ssRNA helix into the principal axis system determined from the order tensor analysis of the experimental RDCs. Each frame of the trajectory was superimposed with this ideal helix followed by calculating the average of , 3 cos 2 À1 D E , where is the angle between a given bond vector (e.g. C1 0 H1 0 ) and the z-axis. The RDC values were then scaled by À82/r 3 , in which r is the C-H bond length and a factor of À82 is applied to shift the computed RDCs to the same scale as the NMR values. The average structure of the ssRNA was calculated as the structure with the minimal root-mean-square deviations from all RNA conformations in the 5 ns REMD trajectory. Previous studies have shown that in the absence of ligand, the queC aptamer domain folds into a non-native hairpin, in which the 5 0 -strand frame-shifts to allow the first two guanine residues to base pair, with the 12 nt ssRNA tail lacking any tertiary interactions (26) . The 2D C-H NMR spectra of the 36 nt queC minimal aptamer domain ( Figure 1A ), in the absence of ligand, show severe resonance overlap and large variations in resonance intensities indicating a highly disordered conformation ( Figure 1B ). Excess imino proton resonances as well as 1 H-15 N NOE data indicate that the unbound queC aptamer domain is in equilibrium between native and non-native hairpin conformations (data not shown), consistent with previous NMR studies (26) . The NMR spectra suggest that the unbound 36 nt queC minimal aptamer domain is highly disordered and that the ssRNA tail is not involved significantly in any tertiary interactions. To test this hypothesis further, we compared NMR spectra of the isolated 12 nt ssRNA tail with the corresponding spectra of the unbound queC aptamer. Remarkably, NMR spectra of the isolated 12 nt ssRNA tail overlay almost perfectly with the queC aptamer domain and specifically onto the highly intense resonances corresponding to highly disordered residues ( Figure 1B ). The only significant deviations are observed for A25 and U26, which are located at the junction site between the hairpin and the tail ( Figure 1B ). This indicates that in the absence of ligand, the ssRNA tail is not involved in any significant tertiary interactions under the NMR conditions (1 mM RNA, 25 mM sodium chloride, 15 mM sodium phosphate, pH 6.4, 0.1 mM EDTA, 298 K). Similarly to the Fusobacterium nucleatum queC riboswitch, the B. subtilis queC aptamer forms kissing dimers, as observed in non-denaturing polyacrylamide gels (Supplementary Figure S1 ) (52) . To ensure that the dimer does not obstruct hairpin-tail interactions, we compared a mutant C14U/C17U construct characterized previously by Kang and coworkers to generate a ligandbound solution NMR structure (26) to the WT aptamer. MFold predicts the C14U/C17U mutations will reduce the dimer stability from À6.1 kcal/mol to À0.9 kcal/mol (53). While we observe removal of the kissing dimer, chemical shifts overall overlay extremely well between the WT queC aptamer and the C14U/C17U mutant (Supplementary Figure S1 ). Specifically, tail chemical shifts correspond extremely well to the 12 nt ssRNA, further suggesting that the tail does not participate in tertiary interactions in the absence of ligand under our NMR conditions. Strikingly, the spectra of the 12 nt ssRNA are well resolved, indicating that it does not adopt a completely random conformation ( Figure 1B and Supplementary Figure S2 ). This stands in stark contrast to corresponding spectra of a 12 nt polyuridine (polyU) ssRNA, well established to have a random-coil conformation (16) , which exhibits severe spectral overlap indicative of a highly disordered conformation (Supplementary Figure S2) . This structural order is observed in the ssRNA despite the lack of any observable imino protons and therefore any base pairing or secondary structure (Supplementary Figure S3) . The 2D 1 H-1 H NOESY spectrum of the ssRNA shows abundant nuclear Overhauser effect (NOE) connectivities expected for a helical conformation, allowing the near complete assignment of base and sugar (H1 0 ) protons at 298 K (Supplementary Figure S3) . Particularly noteworthy are inter-base NOEs observed between adenine H8 protons within the polyadenine tract and between C33-U34 H6 protons, indicating significant base stacking within the polyadenine core at 277 K and decreased at 298 K (Supplementary Figure S3) (54) . Sequential NOEs are only observed for A25, U26, A35, and A36 upon decreasing the temperature from 298 K to 277 K, indicating a higher level of disorder at the terminal ends ( Figure 1C and Supplementary Figure S3 ). Furthermore, homonuclear three bond scalar couplings ( 3 J H1 0 -H2 0 ) indicate that residues within the polyadenine core adopt a C3 0 -endo sugar pucker conformation, consistent with an A-form-like geometry, with the tendency to adopt alternative sugar pucker conformations increasing towards terminal residues (Supplementary Figure S2) . NMR chemical shifts are extremely sensitive probes of the local electronic environment for a given bond vector and can provide useful structural information (55) (56) (57) (58) . Highly disordered residues are expected to have chemical shifts similar to nucleotide monophosphates (NMPs). While the chemical shifts of terminal residues are similar to their NMP analogs, increasing differences are observed when approaching the polyadenine core with the greatest differences observed for A30-32 (Supplementary Figure S2 ). The directionality of the chemical shifts is consistent with increased formation of stacking interactions towards the center of the tail (57) . This is further supported by chemical shift perturbations in a trajectory toward the NMPs with increasing temperature (data not shown). Alternatively, addition of magnesium up to 4 mM results in slight chemical shift perturbations farther from NMPs, consistent with previous studies suggesting that increases in ionic strength stabilize ssRNA stacking interactions (59) (data not shown). In contrast polyU has near-identical ( 0.1 ppm) chemical shifts to UMP (Supplementary Figure S2) . Thus, consistent with NOE data, the chemical shift data suggest a comparatively stacked core with a growing level of disorder towards the terminal ends. Normalized resonance intensities (33) further support these observations, which gradually increase towards the terminal ends, consistent with a higher level of pico-to nanosecond motions (Supplementary Figure S2) . The abundance of NOEs indicates significant base stacking interactions, which likely contribute to ordering of the tail. To probe the thermodynamic stability of the tail, we performed UV/Vis melting experiments to determine the melting temperature of the helix to coil transition. Consistent with previous studies of single-stranded nucleic acids, the melting profile of the ssRNA is extremely broad, characteristic of a non-cooperative transition (Figure 2A ) (37) . Previous studies of a 7 nt polyadenine ssRNA in similar buffer conditions yield analogous melting temperatures to those observed ($35 C compared to 31.7 ± 1.90 C) (37) . We then used our REMD simulations to explore the temperature dependence of base stacking compared to the described UV/Vis melting curves. Base stacking energies from the REMD simulation between temperatures 278-330 K show a similar gradual decrease with increasing temperature and a similar, although reduced, T m value (experimental 31.7 ± 1.90 C compared to computed 20-25 C, as estimating the T 50 value from melting curve, Figure 2A) . However, the calculated base stacking energy plateaus around 320 K while the experimental slope begins to plateau around 330 K, indicating that stacking energies may be under-estimated in the REMD simulation or that additional unaccounted-for factors contribute to the ssRNA stability. Nevertheless, our data suggest that base stacking is the guiding force behind ssRNA stability, consistent with previous studies. To gain further insights into the dynamic properties of the ssRNA at pico-to nanosecond timescales, we measured longitudinal (R 1 ) and transverse (R 2 ) carbon relaxation data for the nucleobases (C2 C6 C8) using 2D 13 C relaxation R 1 and R 1r NMR experiments (40) , where R 1 and R 2 values are determined using in-house software (Supplementary Figure S4) . These measurements represent the first nucleobase 13 C relaxation measurements performed on a ssRNA. The measured R 1 and R 2 values were used to compute order parameters (51) using S 2 = (2R 2 -R 1 ) (60), and normalized to yield a relative order parameter (S 2 rel ) describing the relative degree of order within a molecule ranging from 0 to 1, where 0 and 1 represent minimum and maximum order, respectively. The S 2 rel values were normalized against central residues A31 (C8) and C33 (C6). Resonance overlap prevented the normalization of C2 spins. Again, we observe a gradual reduction in S 2 rel indicating higher levels of disorder moving from central polyadenine residues (A28-C33) towards the terminal ends ( Figure 2B ). We also computed the S 2 rel values based on the REMD simulation described above. The REMD simulations reproduce the general trends observed in the experiments; however, the simulations show significantly increased dynamics at the terminal ends compared to experimental values, with S 2 rel values approaching the dynamic limit ( Figure 2B ). Additionally, while experimental values have similar relative order parameters from A28-C33, large variations are observed in the REMD simulation, with A29-A30 being more ordered and A32 less ordered than experimentally observed ( Figure 2B ). These differences may reflect shortcomings in the force field and/or mismatch in the experimental/computational timescales since the REMD simulations likely probe fluctuations that extend beyond the picosecond timescales sensed by spin relaxation data. The high level of disorder and motional coupling in the ssRNA prevents quantitative analysis of relaxation data using the model-free formalism, which assumes that internal and overall motions are decoupled from one another (51) . This makes it difficult if not impossible to assess the absolute level of disorder in the ssRNA; one can only make qualitative assessments about the relative disorder across different residues. However, it is noteworthy that even the comparatively high R 2 /R 1 values measured in the rigid core ($2.9, Figure 2C ) remain significantly lower than values predicted for a perfectly rigid helical ssRNA ($6.4, Supplementary Figure S4 ) as estimated using the program HYDRONMR (41, 42) . If we assume an overall diffusion tensor predicted by HYDRONMR, we find that central polyadenine residues are highly flexible with an estimated average NMR spin relaxation order parameter S 2 of $0.45 ( Figure 2C and Supplementary Figure S4) . Interestingly, similar though slightly smaller absolute S 2 values are calculated from the REMD simulations (on average S 2 $ 0.36 for core residues, Figure 2D ). These data indicate that despite measurable stacking interactions and a helical-like average conformation, the polyadenine core is highly disordered with residues experiencing fluctuations on the order of a ±40 cone angle (61) at pico-to nanosecond timescales. To further probe the conformation of the ssRNA and extend the NMR timescale sensitivity to milliseconds, we measured RDCs (62,63) using 52.4 mg/ml Pf1 phage as an ordering medium. While most RNAs align optimally in $25 mg/ml of phage, a much higher concentration of phage was used for the ssRNA to ensure optimal alignment. To our knowledge, these are the first RDC measurements reported on a single-stranded nucleic acid. The RDCs measured between two nuclei depend on 3 cos 2 À1 2 D E , where is the angle between the inter-nuclear vector and the magnetic field and the angular bracket denotes a timeaverage over all orientations sampled at sub-millisecond timescales (62, 63) . RDCs were measured for base C5H5, C6H6, C8H8, C2H2 and sugar C1 0 H1 0 moieties (47) . In general, isotropic motions tend to reduce the observed RDC value, approaching zero at the limit of spatially unrestricted isotropic motions (61, 64, 65) . In the ssRNA, large base C-H RDCs are measured in the polyadenine tract residues that gradually decrease at the termini ( Figure 3A ). Although small RDC values can also arise from static placement of the bond vector near the magic angle relative to the principal direction of order, the overall trends observed are consistent with NMR chemical shift and S 2 rel data suggesting that the RDCs indicate increased dynamic averaging at the termini ( Figure 3A) . Interestingly, the near-zero RDCs measured at terminal residues ( Figure 3A and Supplementary Figure S5 ) agree more closely to the REMD simulations compared to the S 2 rel values, indicating that the discrepancy between the measured and computed S 2 rel values may be due to truncation of the S 2 sensitivity to motions faster than nanoseconds. These results add to a growing number of NMR studies on different types of RNA showing that RDC data are capable of probing motions that are incompletely sensed by spin relaxation due to truncation of the time-sensitivity by overall correlation time of the molecule (64, (66) (67) (68) . Unfortunately, severe spectral overlap, particularly pronounced in the Pf1 phage sample, prevented measurement of several C1 0 H1 0 RDCs for the polyadenine core. We subjected the RDCs (excluding RDCs for the two flexible residues from the terminal ends) to an order tensor analysis (47, 69, 70) assuming different input structures including single strands derived from idealized A-form and B-form helices, the REMD-averaged structure, and available ligand-bound X-ray and NMR structures (26, 27) . Despite the relatively small number of RDCs used in this analysis, we clearly observe a better fit with an A-form geometry (Q-factor 4.77%) as compared to all other conformations (Q-factor ! 16%) ( Figure 3B ). This is consistent with independently observed 3 J H1 0 -H2 0 scalar coupling crosspeaks, which indicate a C3 0 -endo sugar conformation for core residues in the tail, suggesting an A-form (and not B-form) helical geometry. The RDCs are in strong disagreement with preQ 1 -bound X-ray and NMR structures (PDBID: 3FU2 and 2L1V) indicating that the tail must undergo a transition from an A-form helical geometry towards the distinct helical conformation observed in the X-ray and NMR structures in which the A-form geometry is perturbed at the hairpin-tail junction, likely due to torsional strain from the ssRNA folding back upon the hairpin. The REMD-averaged structure has a Q-factor of 30%, indicating a better fit than ligand-bound structures, but is still outside the range considered to represent a good fit. Together, these data suggest that, on average, the ssRNA tail adopts an A-form like conformation. The good RDC fit to the A-form structure also suggests that averaging of the RDCs due to internal motions is largely isotropic in nature, causing a semi-uniform attenuation of the RDCs relative to values expected for an A-form structure. The dynamics could involve exchange between a stacked ordered conformation and unstacked highly disordered conformation, or local isotropic motions about the average A-form conformation. As a further check on the accuracy of the A-form structure, we compared the principal direction of alignment (S zz ) determined experimentally using RDCs assuming a ssRNA A-form structure with the orientation predicted by PALES (71) using a ssRNA A-form structure. Surprisingly, we find that the experimentally determined S zz deviates from the helix axis by $19.8 ( Figure 3C ). Interestingly, PALES predicts a principal direction of order that deviates from the helix axis by 14.4 ; the S zz orientation predicted using PALES is in good agreement from that measured experimentally (deviation $ 5 ). The deviation from the helix axis can be attributed to the absence of the complementary strand, resulting in an overall shape with a long axis that is not coincident with the helical axis, as reported previously for a quadruplex DNA topology (72) . To further test the conformational distribution from the REMD simulations, we used a number of simplifying assumptions to compute RDCs from the REMD trajectory. Snapshots from the REMD simulations were superimposed onto an idealized A-form helix oriented in the principal axis system determined using the experimental RDCs and the order tensor fit. RDCs were then arbitrarily scaled by À82/r 3 , in which r is the C-H bond length and accounts for bond length variations during the dynamics. We find excellent agreement between experimental and computed nucleobase RDCs; however, computed C1 0 H1 0 RDCs fail to reproduce observed RDCs, particularly for A32: while the magnitude is similar (18 Hz compared to À30 Hz) the sign differs, suggesting the orientation of the C1 0 H1 0 bond vector differs between experiment and simulation (Supplementary Figure S5) . C1 0 H1 0 RDCs are generally opposite in sign to base RDCs in a double-stranded A-form helix. However, back-calculated C1 0 H1 0 RDCs from the order tensor analysis assuming a ssRNA A-form helix are positive in sign, suggesting the C1 0 H1 0 orientation in the REMD simulations deviates from an A-form structure (Supplementary Figure S5) . Taken together, the data show that the polyadenine tract is relatively ordered at 298 K, with a gradual reduction in order approaching the termini and that the base stacking interactions are the guiding force behind this order. To determine whether disrupting the polyadenine tract will destabilize the global structure, we substituted A29 within the polyadenine tract with a cytosine residue (referred to as A29C). Other types of mutations involving placements of uridine were not explored as these were expected to yield partially base paired conformations. As with the WT construct, we observed no imino protons, indicating the absence of any detectable base pairing and secondary structure (Supplementary Figure S7) . The 2D C-H spectra for the A29C mutant remain highly disperse, and the chemical shift perturbations relative to WT are clustered around the site of mutation (A28 and A30) ( Figure 4A and Supplementary Figure S6) . However, small but significant chemical shift perturbations relative to WT are also observed at more distant residues, including A27, A31, C33 and U34. These perturbations diminish when moving away from the center of the ssRNA and are basically absent in the highly flexible terminal residues ( Figure 4A and Supplementary Figure S6 ). Such longer-range perturbations suggest that the mutation may have a long-range effect possibly by influencing the stacking interactions of several nucleobases. A perturbation to stacking interactions is also supported by distinct NOE connectivities in A29C, which show weakened cross peaks to C29, and new crosspeaks between A28 (H2) and A30 (H1 0 ) that indicate C29 partially loops out to allow A28 to stack onto A30 (Supplementary Figure S7) . The melting temperature of the mutant is reduced by $5 C, and the base stacking energies are computed to be $2 kcal/mol lower compared to WT, indicating that the mutation likely destabilizes the stacking interactions (Supplementary Figure S6) . Interestingly, many of the residues that experience chemical shift perturbations following the A29 to C29 mutation also exhibit a greater degree of dynamics as assessed by normalized resonance peak intensities in 2D C-H HSQC spectra (Supplementary Figure S6 ) and carbon relaxation data (R 1 and R 2 ) ( Figure 4B and Supplementary Figure S8 ). In particular, severe line broadening consistent with a slow exchange process occurring at micro-to millisecond timescales manifesting as reduced resonance intensities in 2D spectra and higher R 2 values is observed for C29 in the A29C mutant ( Supplementary Figures S6 and S8 ). This is not observed for A29 in WT. Smaller but significant line broadening is also observed for residues A31, A32 and U34 (Supplementary Figure S8 ). This line broadening across several residues may reflect exchange between stacked and unstacked conformations. Higher intensities as well as reduced S 2 rel values are observed for residues A27 and A28, indicating a greater degree of fast pico-to nanosecond dynamics ( Supplementary Figures S6 and S8) . Note that the high R 2 and weak signal intensity leads to a higher error in the R 2 /R 1 measurements, particularly for C29. Although the A29C RDCs are generally in good agreement with the WT RDCs, variations are observed for a number of residues (U26, A27, A31) that indicate differences in conformation and/or dynamic behavior ( Figure 4C ). Though an order tensor analysis of 13 RDCs shows best agreement with an A-form structure, the quality of the fit is not as good as that observed for WT (Q-factor = 8.77%, Figure 4D ). The S zz direction measured for A29C when assuming an A-form structure deviates substantially from that predicted using PALES ($11 , Supplementary Figure S9 ). These data suggest that A29C deviates from an idealized A-form structure as compared to WT. These deviations may reflect static and/or dynamic bending about the C29 pivot point, possibly arising from looping out of this residue from the helical stack. Such a conformation is observed in the REMD simulations of A29C $1% but not in WT (data not shown). In general, the REMD simulations predict the NMR data measured for A29C with reduced quality to that noted for WT. Interestingly, the computed absolute S 2 values indicate a global reduction in order for A29C, particularly for residues A27-C33 (Supplementary Figure S8) , whereas NMR relaxation parameters between WT and A29C are more similar, suggesting comparable global order parameters. The REMD simulations reveal enhanced dynamics at C29 consistent with the NMR chemical exchange data. The REMD simulation also suggests increased dynamics at A32, which is not observed experimentally: although slightly reduced, the S 2 rel is within error of A29-A31 values (S 2 rel of 1) (Supplementary Figure S8) . Computed RDCs agree reasonably with measured RDCs, although the C1 0 H1 0 RDCs are opposite in sign as observed in the comparison between WT NMR and REMD-calculated RDCs. The Q-factor comparing the average REMD structure to measured RDCs is 70%; however, removal of A28 C8H8, A30 C2H2 and A30 C1 0 H1 0 RDCs improves the Q-factor significantly. This improvement is observed only for the REMD structure ( Figure 4D ), indicating that these residues, localized about the mutation site, adopt non-Aform conformations and likely experience perturbations from the increased dynamics at C29. The difference in timescales between the REMD simulations and NMR may be another factor leading to the observed discrepancies. Nevertheless, MD and NMR data both indicate significant dynamics at the mutation site with perturbations extending toward the 3 0 end of the ssRNA. ssRNA tail conformation and dynamics optimized for ligand docking in queC aptamer One of the main questions we set out to explore during the course of our studies was how the queC aptamer manages to efficiently bind its cognate ligand despite the small commitment time available in the kinetic switch and the large conformational space that may be available to a highly disordered ssRNA, which would have to search many competing conformations before arriving at the ligand bound pseudoknot conformation. Our study reveals that the ssRNA is not entirely disordered, but rather, has the character of a stacked A-form-like helical conformation which may effectively reduce the conformational search of the ssRNA, promoting efficient docking onto the hairpin to form the pseudoknot. Moreover, our study uncovers a greater degree of flexibility towards the terminal ends, particularly the 5 0 -end which forms the pivot point for docking the ssRNA tail onto the hairpin. The NMR data clearly show the absence of any pre-existing tertiary interactions involving the ssRNA tail in the unbound queC aptamer domain. This together with our findings regarding the conformational behavior of the unbound ssRNA tail suggests the following model for ligand binding ( Figure 5 ). In the absence of ligand, the ssRNA tail is disordered but on average forms an A-form helix-like conformation, which can efficiently explore conformational space about a highly flexible junction. The ligand may transiently form encounter complexes when the tail is close in space to the P1 hairpin, and possibly with the help of divalent ions such as calcium (27, 73) , triggering the necessary conformational changes required to form the pseudoknot and binding pocket. This finding is consistent with computational modeling of the ligand binding mechanism in which A-minor tertiary interactions form first, followed by pseudoknot formation (30) and may explain the fast ligand binding rate observed in the related F. nucleatum queC riboswitch (52). Our results, including the observation of greater dynamics in the mutant, provide a framework for more rigorous testing of this proposed model with future in vitro and in vivo studies. Our study shows that ssRNA can exhibit complex conformational behavior, including variable levels of stacking and propensities to form an A-form helical conformation across the polynucleotide chain, and also, the ability to interrupt stacked residues by introducing sequence-specific kinks and/or distortions. While it has been known for some time that polyadenine stretches tend to stack and form helical conformations (13, 14, 16, 18, 37) , the details of this helical geometry were difficult to decipher based solely on NOE-based NMR data. Our RDC measurements on the ssRNA, together with scalar coupling constant measurements, strongly suggest that the polyadenine tract forms an A-form-like conformation in the WT ssRNA. Our results also unveil dynamic complexity in ssRNA, including a gradual increase in disorder occurring towards the terminal ends that is reminiscent of unfolded polypeptide chains (74) , and also, slower sequence-specific dynamics occurring at micro-to millisecond timescales that may involve transient stacking/unstacking motions that may result in kinking of the ssRNA. Altogether, our studies show that 'structured' ssRNA exhibits exquisite quality spectra and can be studied quantitatively using NMR-based structure and dynamics measurements. The REMD simulations recapitulate many of the key features and trends observed based on the melting and NMR data, including the existence of stacking interactions that are weakened by the A29C mutation, the formation of helical geometry that may be kinked in A29C at the mutation site, and an increase in dynamic disorder towards the terminal ends and localized about C29 in the mutant. However, the REMD simulations showed weaker agreement with sugar RDCs or sugar conformation, particularly in the A29C mutant, and had increased dynamics compared to the NMR data. Prior studies on HIV-1 TAR RNA noted higher levels of dynamics in CHARMM simulations compared to NMR measurements (75) . Our studies indicate that suboptimal base stacking energies may be a source of these excess dynamics. However, a quantitative assessment of the simulations requires the application of domain-elongation methods to rigorously decouple internal and overall motions, and make it possible to quantitatively predict NMR measurements (33, (75) (76) (77) . In addition, MD simulations that retain aspects of time are required to compare the rates of dynamics observed by relaxation and exchange broadening type measurements. The simplicity of ssRNA offers a much needed model system for such studies directed at rigorously examining currently used nucleic acid force fields. Finally, our results suggest that the conformational properties of the ssRNA tail are optimized to allow the queC riboswitch to efficiently bind ligands within the short commitment time available to this kinetic switch. In particular, the pre-stacked ssRNA tail can efficiently rotate about a flexible hinge against the hairpin loop, and explore conformational space efficiently for rapid ligand binding. This pre-stacking about dynamic hinges may be a general feature of many ssRNAs that can play different architectural roles in a variety of RNA contexts. Supplementary Data are available at NAR Online: Supplementary Figures S1-S9.

Soluble RAGE as a severity marker in community acquired pneumonia associated sepsis

BACKGROUND: Community-acquired pneumonia (CAP) is considered the most important cause of death from infectious disease in developed countries. Severity assessment scores partially address the difficulties in identifying high-risk patients. A lack of specific and valid pathophysiologic severity markers affect early and effective sepsis therapy. HMGB-1, sRAGE and RAGE have been involved in sepsis and their potential as severity markers has been proposed. The aim of this study was to evaluate HMGB-1, RAGE and sRAGE levels in patients with CAP-associated sepsis and determine their possible association with clinical outcome. METHOD: We evaluated 33 patients with CAP-associated sepsis admitted to the emergency room and followed in the medical wards. Severity assessment scores (CURB-65, PSI, APACHE II, SOFA) and serologic markers (HMGB-1, RAGE, sRAGE) were evaluated on admission. RESULTS: Thirty patients with a diagnosis of CAP-associated sepsis were enrolled in the study within 24 hours after admission. Fourteen (46.6%) had pandemic (H1N1) influenza A virus, 2 (6.6%) had seasonal influenza A and 14 other diagnoses. Of the patients in the study group, 16 (53.3%) had a fatal outcome. ARDS was observed in 17 (56.6%) and a total of 22 patients had severe sepsis on admission (73%). The SOFA score showed the greatest difference between surviving and non-surviving groups (P = .003) with similar results in ARDS patients (P = .005). sRAGE levels tended to be higher in non-surviving (P = .058) and ARDS patients (P = .058). Logistic regression modeling demonstrated that SOFA (P = .013) and sRAGE (P = .05) were the only variables that modified the probability of a fatal outcome. CONCLUSION: The association of elevated sRAGE with a fatal outcome suggests that it may have an independent causal effect in CAP. SOFA scores were the only clinical factor with the ability to identify surviving and ARDS patients.

Community-acquired pneumonia (CAP) is considered the leading cause of death from infectious disease in developed countries [1] . In Mexico, the annual estimated incidence is100 to 230 cases per 100, 000 inhabitants, causing an alarming impact on public health since 25% of these cases require hospitalization [2] . Severity assessment scores help identify high-risk patients that need hospital therapy; however, the lack of specific and valid pathophysiologic severity markers affects early effective interventions. The recent H1N1 influenza pandemic (p2009A H1N1 or S-OIV) was associated with an increase in cases of CAP that required hospitalization and continues to be a national public health threat [3] [4] [5] [6] . Although the mortality rate was only 1.8%, 31% of patients with severe disease were admitted to an intensive care unit, and 14%-46% died [7] [8] [9] [10] . The first 18 cases, seen from March 24 to April 24, 2009 were reported at the National Institute of Respiratory Diseases in Mexico City. More than half of the patients were between 13 and 47 years of age. Twelve patients required mechanical ventilation and seven died (38%) [3] . Increased mortality was associated with systemic manifestations and complications of CAP with sepsis being the most common and challenging. Physicians may underestimate the severity of CAP, which can lead to insufficiently aggressive interventions inpatients with a high risk of complications [11, 12] . Scoring systems have been used to calculate the probability of morbidity or mortality. The most studied scoring system, the Pneumonia Severity Index (PSI), is a 20-point score that classifies patients into five risk categories based on their percentage of risk of death within 30 days. This score was useful in patients with a low risk of death (0.1%-0.7%) and was recommended for outpatient therapy [13] . However, PSI is limited by its number of variables, making it complex for the emergency room setting [14] . The British Thoracic Society subsequently designed a simpler prediction tool, the confusion, urea, respiration, and blood pressure (CURB) score, also based on the risk of 30 day mortality [12] . In 2003, Lim and colleagues added age ≥65 years as a risk factor to create CURB-65 [15] . CURB-65 is significantly easier to use than PSI since it has only five variables with a single point awarded for each. CURB-65 is recommended together with PSI. Other severity assessments, such as the Acute Physiology and Chronic Health Evaluation II (APACHE II), are commonly used in intensive care units to determine a patient's outcome. The Sequential Organ Failure Assessment score (SOFA) on admission has also been used with results similar to APACHE II. The combination of these may improve sensitivity [16] . Current severity assessment scores only partially overcome the difficulties in identifying patients with severe disease, providing objective classifications of patients into high-risk categories [17] . Thus, there is increasing interest in improving diagnostic accuracy by measuring inflammatory mediators that participate in sepsis. In 1999, Wang et al. reported that high-mobility group box 1 (HMGB1) was detectable in plasma of mice exposed to a lipopolysaccharide. Removal of circulating HMGB1 with a specific antibody improved survival. HMGB1 has delayed kinetics and remains in circulation longer than the initial studied immunologic mediators. HMGB-1 induces the release of proinflammatory and procoagulant factors and when injected into mice, leads to the development of clinical features of sepsis and multiorgan dysfunction [18] . Angus and colleagues studied serum HMGB1 levels in a subgroup of 122 patients with CAP and observed elevated levels more than a week after presentation with high circulating levels associated with greater mortality [19] . These data differ from Sunden-Cullberg et al., who found lower HMGB1 serum levels in non-survivors of severe sepsis [20] . There have also been studies evaluating the role of HMGB-1's receptor, the receptor for advanced glycation end products (RAGE) [21] . Experimental studies demonstrate that RAGE-dependent activation of nuclear factor-kappa B (NF-B) plays a central role in modulating mortality after cecal ligation and puncture [22] . RAGE possesses a secretory isoform known as soluble RAGE (sRAGE), which maintains the extracellular ligand-binding domain but lacks the cytosolic and transmembrane domains. sRAGE has the same ligand binding specificity and competes with cell-bound RAGE, serving as a decoy that abolishes cell activation. In sepsis models, the administration of exogenous sRAGE slightly improved survival [22] . Evidence suggests that human endogenous sRAGE is generated by alternative splicing of RAGE mRNA, or alternatively, by proteolytic cleavage from membranous RAGE [23] . This former mechanism was considered to be a cell regulating mechanism, permitting restoration of homeostasis and survival. Since there is very little knowledge of the role of HMGB-1/RAGE in the clinical setting of CAP-associated sepsis, we decided to perform a pilot study to investigate of HMGB-1, RAGE and sRAGE levels in septic patients with CAP and identify if there is a correlation with severity assessment scores. This observational clinical study included patients evaluated at the UANL University Hospital in Monterrey, Mexico. The Bioethics Committee of the School of Medicine of the Universidad Autonoma de Nuevo Leon previously approved this project and written informed consent was obtained from the patient or a legal representative. Thirtythree consecutive patients, from July 2009 through August 2010, were enrolled in the study within the first 24 hours of their arrival to the emergency room with sepsis secondary to CAP. They were followed-up either in the general ward or in the intensive care unit. Patients were classified according to the Sepsis Consensus Conference of 1992 [24] and the Infectious Diseases Society of America. Clinical data, diagnosis, treatment modalities, and blood samples were collected. The severity of CAP was estimated using the following scores: CURB-65, PSI, APACHE II, and SOFA. To be enrolled, subjects had to be ≥ 18 yrs of age and have both a clinical diagnosis of pneumonia and a new pulmonary infiltrate on chest X-ray. Patients with hospital-acquired pneumonia, an episode of pneumonia in the last 30 days, pulmonary tuberculosis, pregnancy, palliative care, cancer, human immunodeficiency virus infection, chronic steroid use, acute or chronic viral liver disease, and chronic renal disorders were excluded from the study. At enrollment, blood samples were taken, and RAGE receptor was immediately detected by flow cytometry, determining its mean fluorescence intensity. Subsequently HMGB-1 and sRAGE antigen were determined in plasma by enzyme-linked immunosorbent assay (ELISA). At the same time, CURB-65, PSI, APACHE II score, and SOFA score were documented. During the patient's hospital stay we evaluated the presence of acute respiratory distress syndrome (ARDS). Also, a follow-up at 28 days was performed to distinguish between survivors and non-survivors. After enrollment of patients, data was blinded to avoid potential bias. Blood samples were obtained from each patient and sera were recovered to test HMGB1 and soluble RAGE levels. The HMGB1 ELISA kit (IBL International, Germany) and the soluble RAGE ELISA kit (R&D system, Minneapolis, Mn) were used according to the manufacturer's recommendations. A sample of whole blood, anticoagulated with EDTA, from each patient was used for flow cytometry analysis. One hundred microliters of whole blood was incubated with a rabbit anti-human RAGE antibody (Chemicon, Billerica, MA) for 15 min at room temperature. A Goat anti-Rabbit IgG FITC conjugate (Chemicon) was used for flow cytometry detection. Samples were incubated for 15 min at room temperature in darkness. Lysis solution was then added to eliminate erythrocytes and two washes with PBS (0.1 M, pH 7.2) were done centrifuging at 220-240 × g for10 min in each time. Leukocytes were recovered by centrifugation in the same condition and the samples were resuspended in 1 ml of FACS flow (BD Biosciences, Pharmingen) for cytofluorometric analysis (FACS SortCalibur, BD, San Jose, CA). Then 10, 000 cells, in which mean fluorescence intensity (MFI) was obtained and nonspecific fluorescence was deleted, were analyzed. All statistical analyses were performed in SPSS (SPSS, version 13.0), assuming a statistical significance of P ≤ .05. The general descriptive characteristics are presented as means, standard deviations, medians, and percentages. We compared the severity assessment scores and HMGB-1, RAGE and sRAGE levels in surviving and non-surviving patients at 28 days, and between ARDS and non-ARDS patients using a statistical inferential analysis with the U Mann-Whitney nonparametric test. Using the normality tests Kolmogorov-Smirnov with the Lilliefors correction and Shapiro-Wilk, we determined if the obtained values came from a normally distributed population. We present data as plots of admission day medians. We used multivariate logistic regression and Cox regression models with a backward technique to select variables that predicted a fatal outcome, including clinical severity and the inflammatory markers studied, such as age, gender, CURB-65, SOFA score, APACHE II, pneumonia severity index, HMGB-1, sRAGE and RAGE. Correlation between clinical severity scores and immunologic markers at admission, and between the markers, was detected using Spearman's correlation coefficient. Acute organ dysfunction was defined as a new Sequential Organ Failure Assessment score [25] ≥3 in any of six organ systems, following the European Society of Intensive Care Medicine sepsis occurrence in the acutely ill patient study criteria [26] . We also added patients that met the following alternate definition to the analyses: an increase of 1 Sequential Organ Failure Assessment point in any two organ systems, 2 points in one system, or an absolute score of ≥3 in the respiratory system, similar to criteria used in several large trials of antisepsis agents [27] [28] [29] . Thirty-three patients with confirmed CAP were included in the study; three were excluded (one was pregnant and two because of problems with their blood sample). Of the remaining 30 patients, 14 (46.6%) had pandemic (H1N1) 2009 influenza virus confirmed by PCR and 2 patients (6.9%) had seasonal influenza A. No etiologic agent was found in the other 14 patients. Twenty-two patients (73.3%) had severe sepsis or septic shock detected at admission; of these, 17 developed acute respiratory distress syndrome (ARDS). The mortality rate of the study group was a total of 16 patients (53.3%) at the end of the 28 days. There were eight who never developed severe sepsis and survived to hospital discharge, six who developed severe sepsis and survived to discharge, and 16 who developed severe sepsis and died in the hospital. There were no significant differences between survival and non-survival patients with respect to age, gender, ethnicity, microbiological etiology, initial CURB-65, initial PSI class, initial APACHE II score, or emergency room length of stay (P value range, .07-.99) nor between ARDS and non-ARDS patients with respect to gender, ethnicity, microbiological etiology, initial CURB-65, initial PSI class, initial APACHE II score, or emergency room length of stay (P value range, .36-.77) ( Figure 1A) . Group characteristics are provided in Table 1 . Compared with those who did not survive, those who survived had lower SOFA scores (5.5, CI: 4.9-7.7 versus 3, CI: 2.3-4.2) ( Figure 1B) . Compared with patients that did not develop ARDS, those with ARDS had higher SOFA scores (3, CI: 2.1-4.9 versus 5, CI: 4.6-7.1) and were younger ( Figure 2B and Table 1 ). There were no statistically different RAGE, sRAGE and HMGB-1 levels found during early CAP-associated sepsis in ARDS or non-surviving patients ( Figure 1C , Figure 1E , Figure 2A and Table 2 ). No difference was found between influenza A H1N1 infected patients and the rest of the study group (2767 ± 1655 vs 2174 ± 1344, P = .327). We did not find a correlation between immunological molecules and severity assessment scores using Spearman's correlation coefficient (P value range = .16-.99). Finally, none of the studied severity assessment scores correlated with each other (P value range = .18-.79). Using a logistic regression model involving age, gender, APACHE II, SOFA, HMGB-1, sRAGE and RAGE, we found that the only variables that modified the probability of the patient having a fatal outcome were SOFA (P = .013) with a relative risk of surviving of .347 (CI: .151-.797); and sRAGE (P = .05) with a relative risk of surviving of .998 (CI: .998-1) ( Table 3 ). According to multivariate Cox regression analysis we found that a high SOFA score was an independent predictor of non-survival (hazard ratio 1.53, CI: 1.2-1.97, P = .001) ( Table 4 ). We found that SOFA scores and the measurement of sRAGE levels in patients with CAP-associated sepsis helped predict survival. To date, this is the first study that analyses the levels of both of these molecules (the "HMGB-1" ligand and the "RAGE" receptor) in the inflammatory cascade of patients with CAP-associated sepsis. In Mexico, as in developed countries, CAP continues to be an important cause of death from infectious disease [1] with an elevated cost to public health. This was particularly evident with the H1N1 (2009, S-OIV) influenza virus pandemic [8] . Overall mortality is about 50% in patients with CAP that develop septic shock [25] . Although there has been intense research on the pathophysiology of CAP and its severe forms, such as ARDS, only slight improvements in new and effective treatment strategies have occurred. Despite the identification of several recent molecules in patients with infection, such as the receptor expressed on myeloid cells-1 (TREM-1), these lack specificity in sepsis pathophysiology [26] [27] [28] . Discovery of markers may add additional information, increasing the validity of clinical estimates and permitting early, aggressive, and effective sepsis therapy. This justifies every effort to further explore the paradigm of biomarkers in the area of pulmonary infections [29] . We still lack efficient tools to identify patients with CAP who are likely to develop severe complications. Current clinical severity scores partially limit these difficulties, but are far from perfect. In our study, CURB-65, APACHE II and PSI .001 demonstrated no difference between groups (fatal outcome and ARDS). Recently published studies have found that CURB-65 dose not reliably distinguish patients with pandemic influenza CAP who will have good or poor outcomes [30, 31] . In the case of PSI, this could represent its higher ability to detect mild cases; although, this could be explained by the small number of patients in our study. In contrast, we noticed that SOFA scores, although not specific for CAP, were significantly higher in non-surviving or ARDS patients. Thus, in spite of the wide variety of etiologies, this last organ dysfunction score seems to be useful in patients with CAP. It is well known that the recognition receptor "RAGE" and HMGB-1 play a central role in the innate immune system with an impact on its perpetuation and amplification [22] . RAGE stimulation results in sustained NF-B activation, which may be a predictor of severity in sepsis [32] . Conditions that induce NF-B also increase RAGE expression, which in turn produces sustained inflammation; this is seen in CAP, where RAGE ligands are abundantly present. Angus et al. found that CAP patients had higher HMGB-1 concentrations, and this correlated with mortality [19] . Gaini et al. also found higher levels of HMGB-1 in CAP [33] . Studies of severe influenza CAP demonstrated an association between excessive release of cytokines and increased mortality [34, 35] . However, Alleva et al. found in a murine model of severe influenza that HMGB-1 concentrations were not increased in plasma at the time of peak mortality, and peak levels of HMGB1 did not occur until relatively late in infection [36] . Recently, Bopp et al. demonstrated that sRAGE concentrations in sepsis patients were higher in non-survivors when compared with survivors. They concluded that larger clinical trials should study the potential role of sRAGE as a new sepsis marker [37] . However, sRAGE has been used in animal models to block HMGB-1's binding to the RAGE receptor, leading to increased survival. This data indicates that HMGB-1 and RAGE participate in sepsis, including sepsis patients with CAP. After developing multivariate regression models using backward selection techniques, we found that sRAGE and SOFA predicted survival; although the statistical significance was greater for SOFA, a limitation of our study is the small number of patients. One explanation for the elevated sRAGE levels could be an increased gene expression of RAGE in patients with sepsis [22, 38] . Since we know that RAGE participates in tissue damage [39] , it could represent a marker for cellular damage in sepsis. As mentioned previously, there were elevated concentrations of sRAGE on admission in those with a fatal outcome, but without statistical significance. The same was observed in those patients who developed ARDS. On the other hand, receptor RAGE and HMGB-1 demonstrated lower differences between groups. Larger studies will be necessary to investigate the role of these potential sepsis markers. The elevated levels of sRAGE found in our study, as in others, might represent the septic status of the patients as splice-variants of RAGE or shed variants of cell surface RAGE. In contrast to animal studies where a protective effect of sRAGE was seen, we found that sRAGE levels were higher in patients with more inflammation and in non-survivors. This finding could be related to shed variants of cell surface RAGE but this aspect was not one of our objectives. The ELISA we used did not differentiate between splicing variants and the shed variants of RAGE. To the best of our knowledge, this is the second study that finds higher sRAGE levels in plasma of sepsis nonsurvivors compared with survivors [37] . This has Variables included in equation: RAGE, sRAGE, HMGB-1, SOFA score, APACHE II score, gender, age discrepancies with mouse model studies of sepsis after CLP [22] . This could be in part explained by different kinds of sepsis, different etiologic agents, and what was difficult to determine in our study, the time of measurement after the immunologic process started. We do not know if sRAGE concentrations were enough to bind HMGB-1, after they had scavenged AGEs and other RAGE ligands. Moreover, the higher concentrations found in sicker patients could represent sRAGE modified structurally and functionally during sepsis, diminishing its binding and neutralizing capacity. Plasma sRAGE levels are elevated in CAP patients. sRAGE performed as an independent factor affecting the probability of a fatal outcome. Interestingly, the SOFA score demonstrated greater accuracy with the ability to differentiate between surviving/non-surviving and ARDS/non-ARDS groups. Abbreviations APACHE II score: Acute Physiology and Chronic Health Evaluation II; ARDS: Acute respiratory distress syndrome; CAP: Community-acquired pneumonia; CURB-65: Confusion, urea, respiration, blood pressure and age ≥65 years; EDTA: Ethylenediaminetetraacetic acid: ELISA: Enzyme-Linked ImmunoSorbent Assay; FACS flow: Fluorescence-Activated Cell Sorting, flow cytometry; FITC: Fluorescein isothiocyanate; HMGB-1: High-mobility group box 1; MFI: Mean fluorescence intensity; NF-κB: Nuclear factor-kappa B; PSI: Pneumonia Severity Index; RAGE: Receptor for advanced glycation end products; sRAGE: Soluble RAGE.

Virus Identification in Unknown Tropical Febrile Illness Cases Using Deep Sequencing

Dengue virus is an emerging infectious agent that infects an estimated 50–100 million people annually worldwide, yet current diagnostic practices cannot detect an etiologic pathogen in ∼40% of dengue-like illnesses. Metagenomic approaches to pathogen detection, such as viral microarrays and deep sequencing, are promising tools to address emerging and non-diagnosable disease challenges. In this study, we used the Virochip microarray and deep sequencing to characterize the spectrum of viruses present in human sera from 123 Nicaraguan patients presenting with dengue-like symptoms but testing negative for dengue virus. We utilized a barcoding strategy to simultaneously deep sequence multiple serum specimens, generating on average over 1 million reads per sample. We then implemented a stepwise bioinformatic filtering pipeline to remove the majority of human and low-quality sequences to improve the speed and accuracy of subsequent unbiased database searches. By deep sequencing, we were able to detect virus sequence in 37% (45/123) of previously negative cases. These included 13 cases with Human Herpesvirus 6 sequences. Other samples contained sequences with similarity to sequences from viruses in the Herpesviridae, Flaviviridae, Circoviridae, Anelloviridae, Asfarviridae, and Parvoviridae families. In some cases, the putative viral sequences were virtually identical to known viruses, and in others they diverged, suggesting that they may derive from novel viruses. These results demonstrate the utility of unbiased metagenomic approaches in the detection of known and divergent viruses in the study of tropical febrile illness.

Viral infections pose a significant global health burden, especially in the developing world where most infectious disease deaths occur in children and are commonly due to preventable or treatable agents. Effective diagnostic and surveillance tools are crucial for reducing disability-adjusted-life-years (DALYs) due to infectious agents and for bolstering elimination and treatment programs [1] . Previously unrecognized and novel pathogens continually emerge due to globalization, climate change, and environmental encroachment, and pose important diagnostic challenges [2, 3] . Dengue virus (DENV) infection is the most common arthropodborne viral disease of humans, with an estimated 50-100 million clinical infections occurring annually worldwide [4] . DENV infection manifests clinically as dengue fever or the more severe dengue hemorrhagic fever/dengue shock syndrome (DHF/DSS) [4] . The increased spread of dengue virus and its mosquito vectors in many subtropical regions over the past several decades, especially in Latin America and Asia [5] , highlights the need for additional methods of dengue virus surveillance. Diagnosing dengue relies on detecting viral nucleic acid or antigens in the blood or confirming the presence of anti-DENV IgM and IgG antibodies and therefore traditionally depends on RT-PCR, ELISA, and viral cell culture methods [5] [6] [7] . Dengue diagnostics are of crucial importance due to its broad spectrum of clinical presentations, global emergence and spread, unique disease epidemiology, and possible clinical relation to other as-yet unknown tropical febrile pathogens. Traditional viral detection methods, such as serology, virus isolation, and PCR, are optimized for the detection of known agents [2] . However, novel and highly divergent viruses are not easily detected by approaches that rely on a priori sequence, antigen, or cell tropism knowledge. PCR-based assays that employ degenerate primers may successfully target conserved regions within related virus groups, but unlike bacteria, viruses lack universally conserved genetic regions, such as ribosomal RNA, that can be exploited to amplify all viruses [8] . Metagenomic analysis enables more systemic detection of both known and novel viral pathogens [9] [10] [11] [12] and is approached through a variety of microarray and sequencing strategies [13, 14] . The Virochip is a pan-viral microarray platform that has been previously utilized in the detection and discovery of viruses from both human and animal samples [15] [16] [17] [18] [19] . Deep sequencing and shotgun sequencing of human clinical samples has been used for viral detection [20] [21] [22] [23] , novel virus discovery [24] [25] [26] [27] , and divergent virus genome recovery [28] . Viral metagenomic approaches have also been employed as a diagnostic supplement to pathogen detection as part of public health monitoring systems [22] , but have been limited to shotgun sequencing of viral-enriched libraries and have yet to utilize deep sequencing data. Currently available sequencing platforms can generate millions to billions of sequencing reads per run, far exceeding large-scale shotgun sequencing [13] . Deep sequencing of clinical samples, in which hundreds of thousands to millions of sequencing reads are generated per sample, can be incorporated into stepwise virus detection pipelines [29] . Database searches using Basic Local Alignment Search Tool (BLAST) and other alignment tools [30] can be used to identify sequences in samples that correspond to known and novel viruses, including those present at low concentrations or deriving from viruses that may be too divergent to be detected with PCR or microarray methods. Deep sequencing represents an unbiased, highly sensitive method for identifying viral nucleic acid in clinical samples. This study describes the use of the Virochip microarray and deep sequencing for the direct viral diagnosis of serum from cases of acute pediatric febrile illness in a tropical urban setting. Patient clinical data and serum samples were collected between 2005 and 2009 as part of an ongoing pediatric dengue study in Managua, Nicaragua [31] . Virochip and deep sequencing were performed on positive control samples and on 123 dengue virus-negative serum samples. Using these methods, viruses were detected in 45 of 123 (37%) previously negative samples. Sequences derived from known and apparently divergent viruses. The viruses identified in some of the cases are known to induce symptoms consistent with those observed, though the definitive causative agent of these infections remains to be determined. Acute serum samples were collected from suspected dengue cases at the Hospital Infantil Manuel de Jesús Rivera (HIMJR), the National Pediatric Reference Hospital in Managua, Nicaragua, after undergoing informed consent or the informed consent procedure. Patients were enrolled in the study if they presented with fever or history of fever less than 7 days and one or more of the following signs and symptoms: headache, arthralgia, myalgia, retro-orbital pain, positive tourniquet test, petechiae, or signs of bleeding. Patients with a defined diagnosis other than dengue, e.g. pneumonia, were excluded. Suspected dengue cases were tested for dengue virus (DENV) infection at the Centro Nacional de Diagnóstico y Referencia (CNDR) of the Nicaraguan Ministry of Health and were considered laboratory-confirmed if: 1) DENV was isolated, 2) DENV RNA was detected by reverse transcriptase-polymerase chain reaction (RT-PCR), 3) seroconversion was observed by IgM capture enzyme-linked immunosorbent assay (ELISA) of paired acute and convalescent sera, or 4) a $4-fold increase in DENV-specific antibodies was demonstrated by inhibition ELISA in paired acute and convalescent sera [32] . All patients were aged 6 months to 14 years and presented between August 2005 and January 2009. Approximately one half of the suspected dengue cases testing negative by all four dengue diagnostic assays were included in the metagenomics analysis described here. 34 cases (pools 1-4, see below) corresponded to the subset of patients who presented within 4 days of symptom onset and who reported both fever or history of fever and rash. 89 of the samples (pool 5) were selected randomly from among the remaining samples. As positive controls, seven samples (pool 5) that had been clinically diagnosed as virus positive were included. The study protocol was reviewed and approved by the Institutional Review Boards (IRB) of the University of California, Berkeley, and of the Nicaraguan Ministry of Health. Total nucleic acid from 140 ml of serum was extracted using the QIAamp Viral RNA Isolation Kit (Qiagen), which co-purifies RNA and DNA. End-tagged dsDNA libraries were created essentially as previously described [28] . RNA was reverse transcribed in reactions containing 16 reaction buffer, 5 mM dithiothreitol, 1.25 mM dNTPs, 20 pmoles primer (59-CGC TCT TCC GAT CTN NNN NN-39), 100 U Superscript III (Invitrogen), and ,20 ng template. Following reverse transcription, Sequenase reaction buffer and 2 U of Sequenase DNA polymerase (Affymetrix) were added to samples for second strand synthesis. The Sequenase reactions were performed twice so that starting DNA templates would be converted into end-tagged library molecules. The resulting libraries were amplified by PCR using primer 59-CGC TCT TCC GAT CT-39. PCRs contained 16 reaction buffer, 2 mM primer, 0.25 mM dNTPs, 2 U Taq DNA polymerase, and 2 ml library template. Thermocycling conditions were 95uC for 2 min; 25 cycles of 95uC for 30 sec, 40uC for 30 sec, and 72uC for 1 minute, with a final extension of 5 minutes. These libraries were further processed for microarray hybridization and deep sequencing as described below. For microarray hybridization, a fraction of each library was amplified by PCR as above but with a modified dNTP mixture including 5-(3-aminoallyl)-dUTP (Ambion) in lieu of 75% of the dTTP normally in the mixture. The resulting amino-allylcontaining DNA was purified using a DNA Clean and Concentrator-5 column (Zymo Research). The eluate was heat Dengue virus infection is a global health concern, affecting as many as 100 million people annually worldwide. A critical first step to proper treatment and control of any virus infection is a correct diagnosis. Traditional diagnostic tests for viruses depend on amplification of conserved portions of the viral genome, detection of the binding of antibodies to viral proteins, or replication of the virus in cell cultures. These methods have a major shortcoming: they are unable to detect divergent or novel viruses for which a priori sequence, serological, or cellular tropism information is not known. In our study, we use two approaches, microarrays and deep sequencing, to virus identification that are less susceptible to such shortcomings. We used these unbiased tools to search for viruses in blood collected from Nicaraguan children with clinical symptoms indicating dengue virus infection, but for whom current dengue virus detection assays yielded negative results. We were able to identify both known and divergent viruses in about one third of previously negative samples, demonstrating the utility of these approaches to detect viruses in cases of unknown dengue-like illness. denatured at 95uC for 2 min, cooled briefly on ice, then fluorescently labeled in reactions containing 100 mM sodium bicarbonate pH 9, 10% DMSO, and 667 mM Cy3 mono NHS ester (GE Healthcare) for 1 hour at 25uC. Labeled DNA was purified using DNA-CC-5 columns and added to hybridization reactions containing 36SSC, 25 mM HEPES pH 7.4, and 0.25% SDS. Hybridization mixtures were heated at 95uC for 2 minutes, applied to microarrays, and hybridized overnight at 65uC. Following hybridization, arrays were washed twice in 0.576 SSC and 0.028% SDS and twice in 0.0576SSC, then scanned on an Axon GenePix 4000B microarray scanner. Three analysis tools were used to analyze Virochip data: E-predict [33] , Z-score analysis [34] , and cluster analysis [35] . An array was deemed positive for a particular virus if the virus was identified by at least two of these methods. Virochip results were deposited in the NCBI GEO database (GEO accession series: GSE28142). For deep sequencing, the Illumina paired-end adapter sequences were appended to library molecules using PCR, essentially as previously described [28] . Library generation primers (Table S1) were modified from adapter A and adapter B sequences (Illumina). Samples were reverse transcribed and libraries were created and amplified as described above for the Virochip. Library molecules of approximately 300 bp were purified on a 4% native polyacrylamide gel, ethanol precipitated, and PCR amplified for 17 additional cycles using a 22-nt-long primer consisting of the 39-end of Illumina adapter A (primer 2) and the full-length 61-bp Illumina adapter B (primer 4) under the following conditions: 2 cycles of 94uC for 30 s, 40uC for 30 s, and 72uC for 1 min, followed by 15 cycles of 94uC for 30 s, 55uC for 30 s, and 72uC for 1 min. Amplicons generated with the correct adapter topology (one end with adapter A and the other with adapter B) were approximately 355 bp and were separated by polyacrylamide gel electrophoresis from adapter A/A and adapter B/B amplicons, which migrate differently (approximately 40 bp smaller or larger than the expected size). An additional 10 cycles of PCR were then performed using the full-length adapter sequences as primers (primers 3 and 4). Libraries were validated by Sanger sequencing before high throughput sequencing. Following validation, samples were combined into five pools for sequencing. For pools one through four reverse transcription primers included a three or four-nucleotide barcode sequence at the 39-end. For pool five, barcodes were located internally in the adapter sequence. Each pool was sequenced on one lane of a flowcell on the Illumina Genome Analyzer II (pools 1-4) or HiSeq 2000 (pool 5). Pools' 1-4 molecules were sequenced as 67 nucleotide paired ends, and pool 5 molecules as 97 nucleotide paired ends. Paired-end sequencing was performed for several reasons: (1) to double the overall amount of data generated, (2) to double the amount of sequence information per molecule, and (3) to provide anchors from which additional sequence could be recovered by subsequent PCR. In some cases, PCR and Sanger sequencing was used to confirm Virochip and deep sequencing calls and to recover additional sequence. Primer sequences are listed in Table S1 . PCR conditions were: 95uC for 2 minutes, 35 cycles of 95uC for 30 seconds, 50-60uC for 30 seconds (primer dependent), 72uC for 1 minute, and 72uC for 2 minutes. PCR products were sizeselected on an agarose gel, purified with the Purelink gel extraction kit (Invitrogen), cloned, and Sanger sequenced. Full-length poliovirus genomic RNA was transcribed from MluI-linearized plasmid prib(+)XpA using T7 RNA polymerase as previously described [36] . Poliovirus RNA was mixed with HeLa total RNA in a dilution series ranging from 10 22 to 10 26 poliovirus gRNA per HeLa RNA. Randomly-primed dsDNA libraries were prepared, hybridized to the Virochip, and analyzed as described above. Predicted circovirus-like replicase sequences were searched against the NCBI non-redundant protein database (BLASTx, E value 10 22 ). Aligning sequences were retrieved and consolidated using CD-HIT into a set of representative sequences [37] (CD-HIT version 4.5.4; parameters: -c 0.7). These sequences were aligned in Geneious [38] as a global alignment with free end gaps and trimmed to the 47 amino acid overlap shared by the two recovered sequences. A neighbor-joining tree was generated by Geneious Tree Builder [38] . The initial FASTQ data from each pool's lane were binned by barcode. The barcode-split reads were trimmed of non-template deriving and potentially error-prone sequence: a randomly incorporated nucleotide (N), the barcode bases, and the sequence corresponding to the random hexamer, leaving 55 (pools 1, 2, and 4), 54 (pool 3), or 90 (pool 5) bases per read. The lowest complexity fraction was identified by sequences with LZW ratios (compressed size/uncompressed size) less than 0.45 [39] . Reads were aligned to the human genome (build hg18) first using BLAT [40] with the ''-fastMap'' flag, and after filtering, the remaining reads were aligned using BLAT without the flag. Paired reads for which at least one of the reads in the pair had at least 80% identity to the database were marked as human and removed from subsequent analyses. After removal of reads identified as human by BLAT, remaining reads were aligned and filtered by mapping to the human transcriptome using nucleotide BLAST (BLASTn version 2.2.21, word size 30, E value 10 23 ). Remaining reads were next aligned to the human genome using BLASTn (word size 30, E value 10 23 ), filtered, and again aligned to the human genome by BLASTn (word size 11, E value 10). After all human filtering, we reanalyzed the distribution of the complexity of reads and observed a relative enrichment of reads with LZW ratios lower than 0.54 (pools1-4) or 0.48 (pool5; different LZW ratio distributions are an inherent property of different read lengths), and those reads were removed from further analysis. To look for reads with viral homology, we searched the non-redundant nucleotide database (nt) using BLASTn (word size 20, E value 10 23 ). Reads that did not map to nt were aligned to the non-redundant protein database (nr) using translated BLAST (BLASTx, word size 4, E value 10). In order to make specific virus-positive calls, we implemented a set of rules to minimize false positives while maintaining sensitivity. In order to reduce the number of false positive sequences that may share identity equally with both viral and non-viral genomes, we restricted our analysis to those queries whose best alignments were only to animal viral sequences. In a number of datasets, we detected human klassevirus 1, a virus identified and studied in our lab [26] , human poliovirus, used in our Virochip sensitivity experiments, sequences from mosquito densoviruses, also studied in the lab, as well as Moloney murine leukemia virus (MMLV), the polymerase of which was used in the sequence library preparation. We believe these reads represent lab contaminants, and others studies that prepared sequence libraries in the same location have reported similar findings [41] . To account for these contaminants, positive calls were only made on viruses for which there were more supporting reads than there were reads to any known contami-nant. Finally, in order to avoid making calls based on potentially spurious alignments, we considered only those viruses for which there were at least 10 reads supporting their presence. We initially screened the serum samples with the Virochip pan viral detection microarray. This was done as a complement to the deep sequencing analysis and in order to compare the sensitivity of the two approaches. We included 7 blinded positive control samples that had been previously diagnosed in the clinic as being positive for DENV-2 (n = 4), DENV-1 (n = 1), or hepatitis A virus (HAV; n = 2). The Virochip successfully identified the correct virus in all of these positive controls, and in the case of the dengue virus positive samples, the correct serotype as well (Table 1) . We also identified ten samples positive for torque teno virus (TTV). We applied in vitro transcribed poliovirus RNA diluted into HeLa cell total RNA to the Virochip as an additional positive control to quantify Virochip sensitivity. Using the E-predict analysis tool, the lowest detectable concentration of poliovirus was 1 viral RNA per 10 5 HeLa RNA molecules (approximately 10 polio gRNAs per cell equivalent of HeLa RNA; Figure S1 ). The raw reads were first separated by barcode and analyzed as individual data sets as described in the Methods. The bioinformatic filtering process consisted of removing low complexity and low quality sequences, then filtering sequences of human origin ( Figure 1 ). After the filtering steps, an average of 1.9% of the initial reads remained, with an absolute average of 60,000 reads remaining per sample ( Figure 1 and Table 1 ). A few of the barcode datasets appeared to have a larger non-human fraction. Upon further inspection, the non-human components were accounted for by known library preparation contaminants, such as E. coli and S. cerevisiae. The reads remaining after filtering were then compared to sequences in the NCBI non-redundant nucleotide and protein databases using BLASTn and BLASTx respectively. Virus-derived sequences were detected in all 7 positive control samples and in 45/123 (37%) of previously negative serum samples (Table 1 ). In 78/123 (63%) samples, we were unable to identify virus sequence by our detection criteria (Methods). We recovered virus sequences matching the expected viral genomes in all of the positive control samples. The fraction of viral sequences in the controls spanned 4 orders of magnitude, from 0.002% to 2.8% of total reads. The two HAV positive control samples (#401) were aliquots of the same serum sample and were processed and analyzed independently. The fraction of viral reads in the duplicates was within 4-fold (0.4% and 1.2%). This demonstrates that our library preparation, sequencing, and bioinformatics pipeline is capable of reproducibly detecting evidence of clinically relevant infections. In addition to the controls, two non-control samples contained evidence of RNA virus sequence. Both samples had reads deriving from GB Virus C (GBV-C, also known as Hepatitis G Virus) and were essentially identical to GBV-C database sequences. We detected no sequences that best aligned to dsRNA viruses or to retroviruses (except for human endogenous retrovirus and contaminating MLV RT-derived sequences, see Methods). Human Herpesvirus 6 (HHV-6) sequence was detected in 13/ 123 previously negative samples (10.6%). The HHV-6 positive samples had an average normalized read count of 145 HHV-6 reads per sample (range: 24-411), representing 0.002% to 0.02% of the datasets (Table 1) , and all of these reads possessed high sequence identity to the HHV-6B reference genome sequence (gi: 9633069). We generated alignments to the reference genome to investigate the depth and genomic position of the sequence coverage across the HHV-6 genome ( Figure 2) . Although the reads only constitute a relatively small fraction of each dataset, there is coverage across the entire genome and over many genes in most of the HHV-6 positive samples. In addition to HHV-6, we detected Human Herpesvirus 4 (HHV-4, also known as Epstein Barr Virus) sequences in one sample. As with HHV-6, The HHV-4 sequences were virtually identical to previously reported sequences. One sample also contained reads similar to another dsDNA virus, African Swine Fever Virus (ASFV), which has been previously detected in human serum [42] . In this case, the reads best matched ASFV capsid sequences and were relatively divergent (47-51% amino acid identity; no similarity to non-ASFV sequences by BLASTx). Attempts to recover additional ASFV sequence by PCR were unsuccessful. We also identified sequences derived from single-stranded DNA viruses in some samples. In one sample we detected Parvovirus B19-derived reads with high identity to database sequences. Sequences related to various members of the Anelloviridae virus family (TTVs) were detected in 21 (17%) samples. This frequency of detection is within the range reported previously for human serum [43, 44] . The TTV sequences ranged from 40-97% amino acid identity to their closest database matches. We did not pursue these sequences further, because TTVs are known to form a divergent family of viruses and are commonly detected in apparently healthy individuals. Sequences similar to members of the Circoviridae family of ssDNA viruses were detected in 13/123 samples (10.6%). All of the sequences aligned to circovirus or circovirus-like replicase protein sequences. The alignments ranged from 36-84% amino acid identity, and appeared to derive from the replicase genes from multiple related species (Table 1) . Circovirus-like replicase sequences have been detected in human stool, animals, and environmental samples [45] [46] [47] [48] . We detected a range of 12 to 205 circovirus-like reads per positive sample ( Table 1 ). The low sequence coverage prohibited complete genome sequence assembly but informed sequence-specific primer design, from which we were often able to recover larger continuous regions of the replicase genes by PCR and Sanger sequencing (GenBank accessions JF781513, JN837698, and see Table S2 ). We termed the extended replicase-like sequences Circovirus-like NI/2007 1-3 (Cvl-NI 1-3), and compared them to a representative set of other replicase sequences (Figure 3) . The Cvl-NI-1 sequence is most closely related to Circovirus-like virus RW-E (gi: 254688530), a circular single-stranded DNA virus previously found in reclaimed water samples in Florida [45] . The Cvl-NI-2 sequence is most closely related to a replicase sequence recovered [48] . The Cvl-NI-3 sequence did not overlap with the other sequences enough to be included in the phylogenetic analysis, but was most similar to Circovirus-like CB-A, a circovirus-like genome identified in a Chesapeake Bay environmental sample (gi: 229562105) [45] . A subset of the positive samples (Table 1) contained sequences from more than one virus, which may be evidence of co-infection. Almost all of the cases with multiple viruses involved TTV-derived sequences along with HHV-6, DENV-2, or circovirus-like sequences (samples 282, 235, 183, 270, 350, and 377). Two samples contained circovirus-like sequences with ASFV-like (sample 315) or GBV-C sequences (sample 387). In this study, we examined the virus diversity in serum samples from Nicaraguan children with unknown acute febrile illness. We performed Virochip microarray and deep sequencing analyses on 7 positive control and 123 undiagnosed samples. Both of these methods succeeded in detecting the expected virus in the positive control samples. Virochip analysis produced putative viral hits in 10/123 (8%) of the previously negative samples, whereas deep sequencing revealed virus or virus-like sequences in 45/123 (37%). This study demonstrates the utility of these metagenomic strategies to detect virus sequence in multiple human serum samples and is the first to utilize second-generation sequencing to simultaneously investigate many cases of acute unknown tropical illness. Monitoring the emergence and spread of novel human pathogens in tropical regions is a central public health concern. Metagenomic analysis enables more systemic viral detection of both known and novel viral pathogens [42] and can be employed as diagnostic supplements to pathogen detection as part of public health monitoring systems and epidemiologic surveys [9] [10] [11] [12] [15] [16] [17] 19, 21, 23] . Despite the headway, metagenomic virus detection studies will have to confront several remaining difficulties concerning diagnostic accuracy. Foremost concerns include enhancing the sensitivity and specificity of deep sequencing-based diagnostic methods and re-evaluating the evidence for disease causality in light of increasingly sensitive nucleic acid detection and pathogen discovery methods. The former will require improved strategies to biochemically enrich and computationally identify viral sequences while reducing host background sequences. The latter will require a cautious reconsideration of criteria used to establish causal links between microbes and disease, as well as extensive case-by-case follow-up studies employing classical laboratory methods, such as serological analysis and cell culture amplification. It is important to highlight that observing viral sequence in sequencing data is insufficient to establish the role of a virus in disease causality. Like other detection strategies, deep sequencing will serve to inform secondary tests, including seroconversion assays, further nucleic acid testing, cell culture amplification, and additional investigations into plausible disease mechanisms. We detected virus sequence at concentrations as low as ,2 in 10 6 reads. Virus sequence detected in a clinical sample at vanishingly low copy numbers may reflect several possible hostmicrobe scenarios. The sequence detected may be that of a pathogenic virus capable of causing illness at low copy number or through indirect effects, a ubiquitous non-disease causing microbe, a virus outside of its primary replication site, low-level contamination, an artifact of sample collection timing/processing, or remains of incomplete immune clearance. Additional evidence must be considered in each case to define the host-microbe relationship. In this study, we compared the performance of the Virochip and deep sequencing for detecting virus sequence in human serum. The limit of detection of the Virochip was approximately one part in 10 5 for the poliovirus controls, for which there are microarray probes with perfect sequence complementarity ( Figure S1 ). The sensitivity of deep sequencing is limited by the number of reads generated per sample, or read depth. In this study, we detected virus sequences down to two parts per million. Nearly every virus that was detected on the microarray was also detected by deep sequencing; additionally, in numerous samples (n = 44), sequencing revealed viruses not detected by the Virochip (Table 1) . There were two instances where Virochip analysis identified a virus (TTV) that was not detected by deep sequencing (Table 1) . Deep sequencing, therefore, is a superior method for novel virus discovery, because it is more sensitive and provides more conclusive genotypic information than the Virochip. The NCBI TaxID and name of the virus species with the highest number of hits among those viruses with BLAST hits is given. These two samples were prepared from aliquots of the same serum sample. (c) In its deep sequencing dataset, Sample 168 had 9 reads matching TTV, just below our positive identification threshold. doi:10.1371/journal.pntd.0001485.t001 the Virochip is a relatively fast and inexpensive method that is best applied to samples with expected virus copy numbers present at levels greater than 1 in 10 5 host sequences. We were unable to detect a virus in two thirds of the 123 dengue-like illness samples. These results could reflect true negative status, which would result from a non-viral infection, illness due to non-infectious agent, or complete immunologic clearance. Alternatively, the negative results could reflect failures in our diagnostic approaches due to imperfect sensitivity, unsatisfactory sample preparation, improper sample type, or failure to recognize highly divergent viral sequences. The presence of sequences that lack even remote similarities to known species also highlights the need for further development of de novo assembly methods for metagenomic data. Assembled data, increased depth, and enhanced sequenced comparison methods should enable more sensitive detection of divergent viruses in metagenomic samples. Determining the etiology of human diseases with symptoms that overlap with dengue-like illness is important for understanding the full spectrum of emerging or previously uncharacterized pathogens in tropical populations. In this study, 10% of acute serum samples negative for dengue virus from cases of pediatric dengue-like illness were positive for HHV-6. Primary HHV-6 infection causes undifferentiated febrile illness and exanthem subitum (roseola infantum or sixth disease), an acute illness with high fever and rash that typically resolves in three to seven days [49] . Exanthem subitum is a common disease of infants worldwide, and HHV-6 infection most frequently occurs between 6 and 12 months of age [50] , with seropositivity estimates of .95% in adult populations in developed countries [51] . The HHV-6 positive patients in this study were between 7-12 months old, and presented with fever and rash (Table S3) . We detected multiple kilobases of HHV-6 sequence in each positive sample, with sequence deriving from multiple viral genomic regions (Figure 2 ). After acute infection, HHV-6 can latently persist in the host quiescently, with no production of infectious virions or with low levels of viral replication. Latency is believed to endure in several cell types, including monocytes and bone marrow progenitor cells [52, 53] , and may undergo chromosomal integration that can be vertically transmitted [54] . The confounding effects of chromosomal integration make differentiating between active and latent HHV-6 infections difficult when detecting HHV-6 sequence in Sequencing Viruses in Tropical Dengue-like Illness www.plosntds.org serum DNA [55, 56] . A previous study detected integrated HHV-6 genomic sequence in ,1% of healthy blood samples [57] . Since detection of HHV-6 nucleic acid in serum alone does not prove active viral infection, we cannot definitively confirm that the HHV-6 sequences in these samples were not derived from the vertical transmission of chromosomally integrated virus. However, the clinical, epidemiological, and virus sequence data suggest HHV-6 may be the etiologic agent in these febrile illness cases. Primary HHV-6 infection is a major cause (,20%) of infant hospitalizations in the United States [58] , a clinical burden likely shared throughout the tropical world given similar seroprevalence rates [59] . The results of this study illustrate the importance of administering HHV-6 diagnostic tests to cases of suspected dengue-like illness in infants from dengue-endemic regions to differentiate between cases of exanthem subitum, a ubiquitous selflimiting childhood illness, and dengue fever, which carries a greater risk of severe clinical complications and death. Similarly, the one sample positive for Parvovirus B19 sequence may be a case of acute infection with a commonly acquired childhood virus. Parvovirus B19 can manifest as erythema infectiosum (fifth disease), a condition associated with characteristic ''slapped cheek'' rash [60] . Infection can also be subclinical or result in mild nonspecific symptoms. It is possible that Parvovirus B19 infection caused the symptoms in this case (Table S3) , though as with HHV-6, the identification of viral sequences does not definitively demonstrate causality. Histograms of HHV-6B genome coverage generated by aligning reads with minimum 90% identity over the total read length to the genome. The depth of sequence coverage was calculated as the total Kb of aligned sequence per 1 Kb bin over the HHV-6B reference genome. Genome track representation adapted from Dominguez et al [65] . The blue box represents conserved genes across the betaherpesvirus subfamily, the orange boxes represent core genes across the herpesvirus family, the green box represents the late structural genes (gp82-105), and the asterisk denotes the origin of lytic gene replication. Inset text for each histogram is the sample code. Coverage is shown for samples with greater than 80 HHV-6 reads. doi:10.1371/journal.pntd.0001485.g002 Sequencing Viruses in Tropical Dengue-like Illness www.plosntds.org Epstein Barr Virus (HHV-4) sequences were found in the serum of one patient who presented with relatively severe symptoms, and died during hospitalization (Table S3) . HHV-4 infection is a nearly universal occurrence in the first two decades of life [61, 62] . Primary infection in adolescents or adults can manifest as infectious mononucleosis, and chronic infection is associated with various malignancies later in life. Primary infection during childhood, however, is usually asymptomatic or produces only mild symptoms. It is not clear that HHV-4 infection or HHV-4 alone caused the illness in this case. In addition to the viruses for which a plausible disease association exists, many samples contained sequences from viruses with no well-established link to human disease. These included the two samples positive for GBV-C and those containing ASFV-like, TTV-like, and circovirus-like sequences. The Circoviridae family is an extraordinarily diverse group of small, single-stranded circular DNA viruses that includes cycloviruses (genus Cyclovirus) and circoviruses (genus Circovirus), which are commonly detected in human stool and blood, and also in environmental samples [43] [44] [45] [46] [47] [48] . Some circovirus species, such as beak and feather disease virus and porcine circovirus 2, have been associated with disease in bird and pig hosts, respectively, but the pathogenic potential of circoviruses in humans remains unconfirmed [63, 64] . The circovirus-like sequences reported here were detected in nucleic acid libraries prepared from acute human serum and were most closely related to circovirus-like viruses (Figure 3 ), which were first reported in environmental samples and in bats [45, 48] . We were unsuccessful in recovering a full genome sequence corresponding to any of the circovirus-like sequences, and it has not yet been possible to prove that these sequences were not an environmental artifact introduced during sample preparation. It is also possible that these sequences derive from other organisms, such as Giardia intestinalis or Entamoeba dispar, whose genomes encode proteins that share amino acid similarity with circovirus replicase proteins ( Figure 3) . Furthermore, it has yet to be established whether circoviruses are capable of replicating in humans. Pending additional screening and serologic studies, the detection of circovirus-like sequences from human serum should be interpreted with caution. Metagenomic approaches provide an effective high-throughput method to detect uncharacterized virus diversity in a tropical setting from many samples simultaneously. The findings presented in this study further our knowledge of well-characterized and previously unknown viruses present in serum collected from pediatric dengue-like illness patients and advance our understanding of the application of metagenomic approaches to human pathogen detection. Deep sequencing analysis of clinical samples holds tremendous promise as a diagnostic tool by permitting the detection of many different viruses simultaneously, including those present at low-copy numbers and of divergent origin. Major remaining barriers to high-throughput sequencing strategies becoming standard diagnostic practice include prohibitive cost, lengthy sample preparation time, and computationally intensive data analysis requirements. These challenges are magnified in resource-limited settings, such as Nicaragua, but are gradually being addressed. Industry hardware and technical advancements have steadily decreased the per-base cost of deep sequencing, and the results presented here strengthen our expectations of multiplexed sample preparation and bioinformatic data filtering within the framework of current secondgeneration sequencing platforms. Long-term bi-directional partnerships with developing country collaborators facilitate easier access to techniques not currently available on-site, such as deep sequencing, and are also important in providing training opportunities for local scientists and developing relevant pathogen tests and diagnostic policies. This study expands our understanding of the virus diversity in pediatric dengue-like illness in Nicaragua and the application of genomic detection techniques in a tropical setting, findings that are particularly valuable given the pressing need for improved global emerging pathogen surveillance. Figure S1 Virochip sensitivity using poliovirus control RNA. The Virochip can detect one poliovirus gRNA in a background of 10 5 HeLa RNA molecules. Poliovirus RNA was mixed with HeLa total RNA and analyzed on the Virochip. Eighty enterovirus Virochip oligos were found to be responsive to the poliovirus RNA and the mean fold above background of the normalized intensity of these oligos is plotted. Background is defined as the normalized intensity for each oligo in the HeLa-only control sample. The top E-predict hit in the 10 25 to 10 22 samples was human enterovirus C. (PDF) Figure 3 . (PDF)

Effects of a Non-Conservative Sequence on the Properties of β-glucuronidase from Aspergillus terreus Li-20

We cloned the β-glucuronidase gene (AtGUS) from Aspergillus terreus Li-20 encoding 657 amino acids (aa), which can transform glycyrrhizin into glycyrrhetinic acid monoglucuronide (GAMG) and glycyrrhetinic acid (GA). Based on sequence alignment, the C-terminal non-conservative sequence showed low identity with those of other species; thus, the partial sequence AtGUS(-3t) (1–592 aa) was amplified to determine the effects of the non-conservative sequence on the enzymatic properties. AtGUS and AtGUS(-3t) were expressed in E. coli BL21, producing AtGUS-E and AtGUS(-3t)-E, respectively. At the similar optimum temperature (55°C) and pH (AtGUS-E, 6.6; AtGUS(-3t)-E, 7.0) conditions, the thermal stability of AtGUS(-3t)-E was enhanced at 65°C, and the metal ions Co(2+), Ca(2+) and Ni(2+) showed opposite effects on AtGUS-E and AtGUS(-3t)-E, respectively. Furthermore, Km of AtGUS(-3t)-E (1.95 mM) was just nearly one-seventh that of AtGUS-E (12.9 mM), whereas the catalytic efficiency of AtGUS(-3t)-E was 3.2 fold higher than that of AtGUS-E (7.16 vs. 2.24 mM s(−1)), revealing that the truncation of non-conservative sequence can significantly improve the catalytic efficiency of AtGUS. Conformational analysis illustrated significant difference in the secondary structure between AtGUS-E and AtGUS(-3t)-E by circular dichroism (CD). The results showed that the truncation of the non-conservative sequence could preferably alter and influence the stability and catalytic efficiency of enzyme.

Glycyrrhizin (GL), the main constituent of licorice extract (Glycyrrhiza glabra), is a natural edulcorant as well as an important ingredient of traditional Chinese medicine [1, 2, 3] . By hydrolyzing one or two distal glucuronides, GL can be transformed into glycyrrhetinic acid monoglucuronide (GAMG) or glycyrrhetinic acid (GA) ( Figure 1 ). As an important derivative of GL, GAMG displayed stronger physiological functions than GL such as antiviral, anti-inflammatory, anti-tumor functions, and so on; and it is also 1000-fold sweeter than saccharose [4] . On the other hand, GA is the bioactive substance of GL well known for its pharmacological features [4, 5, 6] . The research on GL biotransformation catalyzed by b-glucuronidase (GUS, EC 3.2.1.31) was reported mainly in animal tissues such as duck [7] and human [8] , whereas studies on GL biotransformation in fungal species are few [9] . In our previous work, a fungal strain, Aspergillus terreus Li-20 was screened, which can use GL as a carbon source and produce GAMG and GA after catalysis by b-glucuronidase (AtGUS). The main disadvantages of AtGUS were low enzyme productivity, low catalytic efficiency, and pathogenicity, which rendered it unsafe for use in the food and medical industries. To overcome disadvantages of a natural enzyme, many methods was applied to obtain artificial evolution enzymes, and this approach is not only faster than natural evolution but also provides a deeper understanding of enzyme evolution. Several methods of designing new enzymes are available, and gene sequence truncation is also investigated for its effects on enzymatic properties. The non-conservative N-terminal domain of the protein phosphatase1 (PP1), with 1-8 residues deleted, showed higher sensitivity to three substrates and influenced the structure and properties of PP1 [10] , whereas the truncation of the Cterminal region improved the thermal stability of endo-bglucanase from Bacillus subtilis JA18 [11] . However, the loss of the C-terminal regulatory domain resulted in a loss of the ability to catalyze the aldol reaction [12] . With development of molecular biology and bioinformatics characterization, an increasing number of sequence data have been cloned and applied in the biotransformation industry. Bioinformatics characterization from the National Center for Biotechnology Information (NCBI) showed that most b-glucuronidases belong to the glycoside hydrolase family (GHF) 2, and all of them consist of sugar-binding, immunoglobulin-like b-sandwich, and TIM barrel domains (triosephosphate isomerase, TIM) [13, 14, 15] . The TIM barrel domain, which is one of the most common catalytic domains, is adopted by about 10% of the enzymes; thus, sequence modification inside or outside the domain to improve the enzymatic property and determine the catalytic mechanism was reported in many studies. The site-directed mutagenesis of seven amino acids (aa) in the TIM barrel domain was performed to investigate the importance of the residue in the catalysis of an exo-b-d-glucosaminidase from Trichodema reesei [16] . Heparanase is an endo-b-d-glucuronidase, and its C-terminal region, which is not an integral part of the TIM barrel domain, is essential for the enzymatic activity and secretion of heparanase [17] . Although b-glucuronidases from many species have been registered in Genbank, only a few genes have been published for GL biotransformation [18] . Three fungal strains, namely, A. terreus Li-20, P. purpurogenum Li-3, and A. ustus Li-62, were screened in our previous studies and represented three modes of GL biotransformation: (1) GLRGA+GAMG; (2) GLRGAMG; and (3) GLRGA [9] . The three b-glucuronidase genes were cloned in our laboratory, and the aa sequence alignment showed that the bglucuronidase from A. terreus Li-20 (AtGUS) was quite different from the other two b-glucuronidases(PGUS and AuGUS) in Cterminal non-conservative sequence. AtGUS can hydrolyze GL into two products; thus, the different modes of GL biotransformation of the b-glucuronidases from the other two fungi may be related to the natural evolution in the sequence. In the present research, Atgus and the partial sequence [Atgus(-3t)] without Cterminal non-conservative sequence behind the TIM barrel domain were amplified in order to investigate effects of nonconservative sequence on enzymatic property. No specific permits were required for the described field studies. No specific permissions were required for these locations/ activities. No location is privately-owned or protected in any way. The field studies did not involve endangered or protected species. In our previous work, A. terreus Li-20 was isolated and screened from a G. glabra planting field in Shihizi, Xinjiang. It was incubated in 100 ml liquid Czapek's medium in a 500 ml Erlenmeyer flask at 30uC and placed in a shaker incubator at 170 rpm. Escerichia coli DH5a and E. coli BL21 were used as hosts for plasmid amplification and expression, respectively. The plasmids pMD19-T (TaKaRa, Japan) and pET28a (+) (Invitrogen, U.S.) were used as vectors. The recombinant cells were inoculated in a lysogeny broth (LB) medium with kanamycin (50 mM) and operated at 37uC for 3 h. The recombinant protein was induced by adding 0.4 mM isopropyl-b-D-thiogalactopyranoside (IPTG). GL was purchased from Xinjiang Tianshan Pharmaceutical Co. (China). GA was purchased from Sigma Chemical Co. (U.S.), whereas GAMG was obtained from the Nanjing University of Technology, China. Methanol was of chromatographic grade. All other chemicals used were of analytical grade. The DL2000 marker and the protein low weight marker were purchased from TaKaRa, Japan. An intron in an AtGUS genomic sequence was removed via three-step polymerase chain reaction (PCR) to express the gene in E. coli BL21 (Figure 2A ). According to the database of A. terreus NIH2624, a primer set containing P1(59-CCGTACgTAATGCT-GAAGCCCCGACAAACACCTT-39) and P2(59-CATGCGG-CCGCTTAAGCGCCAAATAGGAAGTATAGT-39) was designed to obtain the sequence with an intron from the A. terreus Li-20 genome under the following conditions: 94uC for 10 min, 30 cycles of 94uC for 1 min, 58uC for 1 min, 72uC for 2 min, and a final extension at 72uC for 10 min with Ex Tag (TaKaRa, Japan). After ligated into PMD19-T, a primer set containing P3(59-CACTCCACCGTGTTTTCAATGTATGAGCTGCAGC-39) and P4(59-CCGGCTTCGCAGCTATGTGTCTTGAGCATC-39) was used for the second PCR, and the reaction was performed by Pfu polymerase (Shenggong, China) under the following conditions: 94uC for 10 min, 30 cycles of 94uC for 1 min, 55uC for 1 min, and 72uC for 5 min. The fragment amplified in the second PCR was ligated by T4 DNA ligase after a terminal phosphation with T4 polynucleotide kinase (Takara, Japan), and the positive clones were screened in an LB plate with 100 mg/mL ampicillin. The primer set containing P1 and P5(CATGCG-GCCGCTTAACTCCACCGTGTTTTCAATGTATG-39) was used for AtGUS(-3t) under the same conditions as those of P3 and P4. After IPTG introduction and the ultrasonication of the recombinant E. coli BL21 cells, supernatant was brought to 70% saturation with (NH4) 2 SO 4 and stored overnight at 4uC, and then again centrifuged. The enzymes expressed by pET28a(+) vector were fused to an N-terminal six-histidine tag and purified via nickel chelate affinity chromatography (GE, U.S.), which was eluted with 150 mM imidazol. The quality of the purified protein was evaluated using sodium dodecyl sulfate polyacrylamide gel electrophoresis and coomassie blue staining. HPLC for analysis of GL, GAMG, and GA GL, GAMG, and GA concentrations were measured via reverse-phase high performance liquid chromatography (HPLC) on a C18 column (4.6 mm6250 mm, 5 mm particle size, Kromasil) at 40uC. The sample (injection volume, 10 ml) was separated with a mobile phase consisting of 6% acetic acid/ methanol (19:81 v/v), and the elution was monitored via ultraviolet detection at 254 nm. The GL, GAMG, and GA amounts were calculated from the standard curve of the peak area and concentration. Determination of pH and temperature profiles The activity of b-glucuronidase was determined using GL as the substrate. The reaction mixture consisted of the enzyme and substrate (2 g/L GL) at a 1:4 (v/v) ratio. 50 Mm Na 2 HPO 4 -citric acid buffer at pH 4.0-8.0 was used for determine the pH effects of the enzyme. The catalytic activity of the enzyme was examined at 30 to 70uC in 50 mM Na 2 HPO 4citric acid buffer (pH 7.0). The enzyme activity under the optimal temperature and pH was defined as 100%. The temperature stability of the enzyme was determined by incubating the enzyme samples at different temperatures (45, 55, 65, and 75uC) for 15, 30, 45, 60, and 120 min at optimum pH without the substrate GL, and the residual activity was determined at the optimum temperature. The effect of several metal ions on the activity of AtGUS(-3t)-E and AtGUS-E was investigated. The enzyme activity was determined in the reaction mixture consisting of K + (KCl), Na + (NaCl), Mg 2+ (MgCl 2 ), Mn 2+ (MnCl 2 ), Co 2+ (CoCl 2 ), Ca 2+ (-CaCl 2 ), Ni 2+ (NiSO 4 ), Cu 2+ (CuSO 4 ), and Al 3+ (AlCl 3 ) ions at final concentrations of 1 and 5 mM. The enzyme activity was subsequently determined at the optimum temperature after incubation for 30 min. Different concentrations of the substrate GL, ranging from 0.375 to 4 mM, were prepared to determine the kinetic constants. The catalytic reactions were continuously monitored, and the initial velocities were fitted to the Michaelis-Menten equation using the Origin 7.5 software (OriginLab). The values of the Michaelis-Menten constant (Km), maximal velocity (Vmax), catalytic turnover rate (Kcat), and catalytic efficiency (Kcat/Km) were evaluated. Far-UV Circular dichroism (CD) spectra were recorded at 25uC in the range from 190 to 260 nm with a spectral resolution of 0.2 nm using a Jasco J-715 spectropolarimeter. The scan speed was 100 nm/min and the response time was 0.125 s with a bandwidth of 1 nm. Quartz cells with an optical path of 0.1 cm were used. Typically, scans were accumulated and subsequently averaged. The spectra were corrected for the corresponding protein-free control. The protein three-dimensional structural was modeled by modeler 9v7 to analyze three domains of the protein. The 2,193 bp product was amplified and sequenced using a genomic template ( Figure 2B) , and its 219 bp intron was analyzed by NCBI. After a three-step PCR, the full encoding sequence was cloned. The results show that the open reading frame of this gene was 1,974 bp ( Figure 2B) , which encodes for 657 aa. The conserved domain database (CDD) was performed to analyze domains of AtGUS, and there were sugar-binding domain, immunoglobulin-like beta-sandwich domain, and TIM barrel domains in it which all belonged to glycoside hydrolase family (GHF) 2. GHF 2 comprised b-galactosidase (EC 3.2.1.23), bmannosidase (EC 3.2.1.25), and b-glucuronidase (EC 3.2.1.31), so the phylogenetic tree was constructed according to it (Figure 3) . It showed that the gene cloned was a b-glucuronidase gene named AtGUS (Genbank accession No. JF894133), which was found very similar to PGUS(Genbank accession No. EU095019) from P. purpurogenum Li-3 and AuGUS (Genbank accession No. JN247805) from A. ustus Li-62, especially in the sugar-binding, immunoglobulin-like beta-sandwich, and TIM barrel domain ( Table 1 ). The obvious difference among them lied in the non-conservative sequence of the C-terminal behind the TIM barrel domain which may result in the difference enzymatic properties. Therefore, the 1-1,776 bp segment, named Atgus(-3t), was amplified in the Both pET28a(+)-AtGUS and pET28a(+)-AtGUS(-3t) were constructed and transformed into the E. coli BL21 strain, and the recombinant proteins AtGUS-E and AtGUS(-3t)-E were successfully expressed ( Figure 4) . The induction condition for the optimum production of the two recombinant proteins was 20uC with 0.4 mM IPTG. Both AtGUS-E and AtGUS(-3t)-E were purified through Ni-NTA sepharose (Figure 4) . The target protein was eluted with 150 mM imidazole. Furthermore, the concentrations of the soluble purified proteins of AtGUS-E and AtGUS(-3t)-E were determined as ,7 and ,12 mg/L, respectively. Both purified enzymes could hydrolyze GL into GAMG and GA. We investigated the enzymatic properties to determine the effect of the non-conserved sequence on the enzyme. The optimal pH for the bioconversion reaction by AtGUS-E was 6.6, whereas that for AtGUS(-3t)-E was 7.0 ( Figure 5A) .The optimal temperatures for AtGUS-E and AtGUS(-3t)-E were both 55uC ( Figure 5B) . Enzyme thermal stability experiments showed that the enzymes remained more than 80% residual activity at 45uC and 55uC for 120 min heat treatment, respectively ( Figure 5C and 5D) . At 65uC, the residual activity of AtGUS(-3t)-E remained almost 60% of enzymatic activity after 30 min heat treatment, which was comparatively higher than that of AtGUS-E with less than 5% residual activity after the same treatment. At a higher temperature (75uC), both enzyme residual activity rapidly vanished, and within 15 min heat treatment, almost all enzymatic activity was lost. The effect of various metal ions with different concentration gradients (from 1 to 5 mM final concentration) on the activities of AtGUS(-3t)-E and AtGUS-E was evaluated, and the results are presented in Table 2 . The enzymatic activity assayed in the absence of metal ions was taken as 100%. The effect of monovalent cations on the two enzymes was similar: the 1 and 5 mM K + and 1 mM Na + exhibited no obviously affecting effects on the activity of AtGUS(-3t)-E and AtGUS-E, while the 5 mM Na + inhibited the enzymatic activity. The divalent cations Mg 2+ and Mn 2+ discretely promoted the activities of AtGUS-E and AtGUS(-3t)-E. With increasing concentration of Co 2+ , AtGUS-E was firstly activated and then inhibited while AtGUS(-3t)-E showed an inverse effect. Ca 2+ and Ni 2+ also exhibited opposite effects on the two enzymes. Ca 2+ at 5 mM final concentration enhanced the activity of AtGUS-E by 107% but inhibited AtGUS(-3t)-E activity by 61%. In the presence of 5 mM Ni 2+ buffer, AtGUS(-3t)-E was increased by 78%, whereas AtGUS-E was decreased by 40%. Cu 2+ distinctively inhibited the enzyme activity, while Al 3+ showed activation at 1 mM concentration and inhabitation at 5 mM concentration to both enzymes. These results reveal that K + , Na + , Mg 2+ , Mn 2+ , Cu 2+ , and Al 3+ exhibited nearly similar effects on the activity of AtGUS-E and AtGUS(-3t)-E; however, Co 2+ , Ca 2+ , and Ni 2+ , showed opposite effects on both enzymes, respectively. The data reported here have been taken from three replicate samples from three independent experiments. The reaction kinetics of AtGUS-E and AtGUS(-3t)-E were determined. The Vmax of the AtGUS-E and AtGUS(-3t)-E enzymes toward GL were calculated using Lineweaver-Burk plots ( Table 3 ) and were determined as 1.84 and 0.97 mmolmin 21 mg 21 , respectively. The Km of the recombinant AtGUS(-3t)-E was 1.95 mM, which was approximately one-seventh that of AtGUS-E (12.9 mM), indicating that a higher affinity of AtGUS(-3t)-E for GL than AtGUS-E. In addition, the catalytic efficiency (kcat/Km) of AtGUS(-3t)-E (7.16 mM s 21 ) was 3.2 folds higher than that of AtGUS-E (2.24 mM s 21 ). The enzymatic activities were determined at different concentrations of the substrate GL from three independent experiments. To determine the impact of the sequence truncation on the structure of the protein, a circular dichroism (CD) spectra was amplified. The far-UV spectra for AtGUS-E and AtGUS(-3t)-E have been presented in Figure 6 . It illustrated that the curves exhibited significant difference between the two proteins. These results suggest that the secondary structure of AtGUS-E has changed after deletion of non-conservative sequence. The b-glucuronidase (GUS) gene was first cloned in 1987 [19] , and in subsequent years, many GUS genes were cloned and registered in the GenBank. However, this gene has never been cloned for the hydrolysis research of GL into GAMG or/and GA, with more valuable merits. Based on CDD analysis, the three domains of the enzyme AtGUS were well investigated, and the main aim of the present study is to modify the non-conservative sequence of AtGUS and try to obtain an artificial evolution enzyme with better enzymatic properties. The TIM barrel domain is a canonical (b/a) 8 -barrel composed of eight units, each of which consists of a b-strand and an a-helix [20] . There was a non-conservative segment behind the catalytic domain (TIM barrel domain) of AtGUS which showed low identity with PGUS and AuGUS after the primary sequence alignment. A model of the three-dimensional structure of AtGUS was presented in the current research ( Figure 7) . The deleted sequence exhibited no involvement in the TIM barrel domain, locating near the ''stability face'' rather than the ''catalytic face'' [21] . AtGUS-E and AtGUS(-3t)-E were very similar with each other at some enzymatic properties, such as optimal pH and optimal temperature. It was reported in previous studies that many modified enzymes maintained some original enzymatic properties even though some sequence has been modulated [10] . Furthermore, we could also speculate that the non-conservative sequence lied outside of the catalytic face of the TIM barrel domain which may not affect the catalytically active residues and the GL biotransformation mode. Interestingly, the stability of AtGUS(-3t)-E was slightly higher than that of AtGUS-E at 65uC. Similar results have been reported that the modification of the C-terminal region could improve the thermal stability of endo-b-glucanase from Bacillus subtilis JA18 [11] . In addition, previous study showed that ab-loops in ''stability face'' are important for the stability [22] . The truncation of the non-conservative sequence lies near ab-loops of the stability face, therefore, we can predicted that the deletion of the C-terminal region outside the TIM barrel domain has influence on thermal stability of AtGUS. The effect of nine metal ions with different concentration gradients on the activities of AtGUS(-3t)-E and AtGUS-E was evaluated, and Co 2+ , Ca 2+ , and Ni 2+ showed opposite effects on the two enzymes, respectively. In addition, AtGUS(-3t)-E showed higher affinity and catalytic efficiency than AtGUS(-3t)-E. Both of the result might suggest that the spatial structural rearrangement, and the speculation has been proved by CD spectra, which showed great difference between the secondary structure of the two enzymes. It has been reported that the loops above the catalytic face was very important for substrate hydrolysis [23] , so we can conclude that the truncation of the non-conservative domain firstly changed the secondary structure of the enzyme and then influenced the substrate affinity, catalytic efficiency and metal ions effects. Moreover, the crystal structure of human bglucuronidase was firstly reported in 1996 [24] , and the structure of bacterial b-glucuronidase has also been published recently [25] . Both b-glucuronidase structures were tetramers. The deleted region of the AtGUS non-conservative sequence lies outside the main three domains, so it was predicted that the non-conservative might change the combination pattern of each monomer. Different methods have been applied in creating new enzyme such as error-prone PCR [26] , DNA shuffling [27] and staggered extension process (StEP) [28] . Some efforts have been made to modify the enzyme by directed screening but high ratio of negative mutated forms of enzyme in the initial screening is a big hurdle and requires further screening for positive mutated forms of enzyme. Every method has its own advantages and disadvantages that determines the feasibility of a particular method so as sequence truncation [10, 11, 12] . Based on the same hydrolyzing mode, relatively higher thermal stability, and especially the enhanced affinity and catalytic efficiency for GL, deletion of the non-conservative sequence behind the TIM barrel domain was a successful evolution of AtGUS. The truncation of non-conservative region based on sequence alignment could be an effective way of artificial enzyme evolution as it can alter and influence the stability and catalytic efficiency of enzyme, and could help in understanding the relationship between the structural modulation and enzymatic properties.

Protein sequence analysis based on hydropathy profile of amino acids

Biology sequence comparison is a fundamental task in computational biology. According to the hydropathy profile of amino acids, a protein sequence is taken as a string with three letters. Three curves of the new protein sequence were defined to describe the protein sequence. A new method to analyze the similarity/dissimilarity of protein sequence was proposed based on the conditional probability of the protein sequence. Finally, the protein sequences of ND6 (NADH dehydrogenase subunit 6) protein of eight species were taken as an example to illustrate the new approach. The results demonstrated that the method is convenient and efficient.

The comparative biological sequence is one of the issues in bioinformatics when analyzing similarities of function and properties of different sequences. Similarly, evolutionary homology is analyzed by comparing DNA and protein sequences. In general, there are two types of methodologies to conduct the comparison. One is an alignment-based method, and the other is an alignment-free method. Sequence alignment is based on computeroriented and computer-intensive comparisons of sequences, and then a distance function or a score function is obtained. Using the distance function, one can compare biological sequences. However, multiple sequence alignment of several hundred sequences always produces a bottleneck, firstly due to long computational time, and secondly due to possible bias of multiple sequence alignments for multiple occurrences of highly similar sequences (Pham and Zuegg, 2004) . Therefore, the emergence of a study on alignment-free sequence analysis is obvious. Until now, alignment-free sequence analysis is still in its early development. For most alignment-free methods, a biological sequence should be transformed into an object for which a linear algebra and statistical theory already has useful analytical tools. Since 1983, DNA sequence has been represented in different dimension spaces (Hamori and Ruskin, 1983; Hamori, 1985; Nandy, 1994; 1996; Nandy and Basak, 2000; Randić et al., 2001; Randić, 2003; Randić and Balaban, 2003; Zhang et al., 2003; Liao and Wang, 2004; Liao et al., 2005; Nandy et al., 2006; Bai et al., 2007; Feng and Wang, 2008) . Each nucleotide of a given DNA sequence is a point in different dimension spaces, and these graphical representations can allow us to qualitatively analyze DNA sequences, and provide a way of viewing, sorting and comparing various genomic sequences. Based on the graphical representation, it is possible to numerically characterize DNA sequence and further quantitatively measure similarity of different DNA sequences. Although protein sequence and DNA sequence belong to symbolic sequences, compared with DNA sequence, there are fewer methods for the graphical representation of protein sequence. This is mainly because extension of DNA graphical representation to protein sequences would enormously increase the number of possible alternative assignments for the 20 amino acids. The amino acid sequence is the key to understanding protein structure and function in the cell, so analysis of amino acid sequence is an important part of post-genomic studies. Recently, several schemes have been proposed in protein graphical representation (Randić and Krilov, 1997; Vinga and Almeida, 2003; Bai and Wang, 2005; Li J. et al., 2006; Li C. et al., 2008; Munteanu et al., 2008; Yau et al., 2008; Yao et al., 2008; Wen and Zhang, 2009) . In order to plot amino acid sequence, 20 amino acids in protein sequences are divided into different types, including protein sequence regarded as a word with three, four, or five different letters. Since ordering amino acids based on their physicochemical properties may offer better insights into comparative study of protein than representation of protein based on the random ordering of amino acid, Randić (2007) and Yao et al. (2008; outlined different 2D graphical representations of protein sequence based on different physicochemical properties. The graphical representation of protein sequence cannot only describe amino acid sequence, but also measure similarity/ dissimilarity of different protein sequences. However, the methods only consider the string's information of protein, and do not consider adjacent string's information of amino acid sequence. Here, we choose conditional probability to measure adjacent string's information. In this paper, we converted a protein sequence into three-letter sequence based on hydropathy profile of amino acid and defined the three curves to represent different hydropathy features. We then selected conditional probability as a new invariant for the protein sequences. To illustrate the proposed method, we made a comparison of the sequences belonging to eight ND6 (NADH dehydrogenase subunit 6) proteins from http://www.ncbi.nlm.nih.gov/: human , rat (AP_004903), and mouse (NP_904339). According to the hydropathy profile of amino acids, the amino acids can be classified into three groups (Nei and Kumar, 2002; Liu and Wang, 2006) : internal group (F, I, L, M, V), external group (D, E, H, K, N, Q, R), and ambivalent group (S, T, Y, C, W, G, P, A). The amino acid of internal group tends to occur in the inner side of the protein's spatial structure, while the amino acid of external group tends to appear at the surface. In order to characterize the hydropathicity of a protein primary structure, we defined a primary protein sequence as a symbolic sequence including three letters according to the following rule: where S(i) is the letter in the ith position in the protein primary sequence, and F(S(i)) is the substitution for S(i). Since the hydropathy profile can detect more evolutionary relationships, in the next section, we analyzed the new protein sequence containing three letters through different mathematical methods. Given a protein primary sequence with length N, we transformed it into a new sequence according to the above definition. For example, for the protein sequence, S=MMYALFLLSVGLVMGFVGFS, then F(S)=IIAAIIIIAIAIIIAIIAIA. To obtain more information, we defined three curves of the sequence. Firstly, we let IE EA IA 1 if ( ( )) I, 0 otherwise, where i ranges from 1 to N. Then, let Y n u and n are Y axis and X axis, respectively, and then we can draw three different curves, which are named as IE, IA, and EA curves of the protein sequence. The three different curves can give us some information about the protein sequence. According to the IE curve, we can compare the numbers of the amino acids belonging to the internal group and the external group at different positions. The IA curve can then be used to compare the numbers of the amino acids belonging to the internal group and the ambivalent group at different positions. Finally, the EA curve can compare the numbers of the amino acids of the external group and the ambivalent group at different positions. According to the above definitions of three different curves, we drew three curves of ND6 proteins for the eight species (Fig. 1 ). Fig. 1 shows that the amino acids of the internal group in ND6 protein sequences are more than the amino acids of the external group, and the amino acids of the ambivalent group are more than the amino acids of the external group. Furthermore, it is evident that G. seal and H. seal have similar curves, rat and mouse's curves are almost identical, and the three curves of human, gorilla, and chimpanzee are similar, but wallaroo's curve is different from curves of other species. Protein sequence is composed of three parts, internal group, external group and ambivalent group, so we regard the random numerical sequence to be composed of three parts (+1, 0, −1). We calculated the conditional probability, which was invariant to quantity protein sequences. For example, let X i IE represents the state of the ith (i=1, 2, ..., N) moment, state space S={+1, 0, −1}. There are nine conditional probabilities as follows: ( 1 1 ), According to the above definition, we can obtain these conditional probabilities of a given protein sequence. The conditional probability of each of ND6 proteins is listed in Table 1 . Given two protein sequences, we can obtain two nine-component vectors whose elements are conditional probabilities for each protein sequence. Based on the vectors, we can compare different protein sequences. In general, similarities of the two vectors can be obtained by calculating Euclidean distance. The smaller the Euclidean distance of two vectors is, the more similar are the protein sequences. The Euclidean distance of two vectors u and v is as follows: where u i and v i denote the components of vectors u and v, respectively. k is the dimension of vectors u and v. Yao et al. (2009) proposed a new similarity measure of sequences, and coefficient of determination (r 2 ), which is defined as: r 2 can vary from 0 to 1, and represents the percent of the data, which is the closest to the line of best fit. The larger the coefficient of determination of two vectors is, the more similar are the protein sequences. In Tables 2 and 3, we give the similarity/dissimilarity matrices for the eight ND6 sequences based on Euclidean distance and coefficient of determination amongst nine-component vectors. As shown in Tables 2 and 3, it is obvious that ND6 proteins of human, gorilla, and chimpanzee are more similar to each other. In addition, ND6 proteins are more similar for (G. seal, H. seal) and (mouse, rat). However, ND6 protein of wallaroo is very dissimilar to others amongst the eight species. The results are consistent with the known fact of evolution (Yao et al., 2009) . Biology sequence analysis is a fundamental task in computational biology, whose aim is to detect similarity/dissimilarity relationships between molecular sequences. Some alignment-free methods to analyze similarities/dissimilarities of DNA sequences have been proposed. However, there are few alignmentfree methods to analyze protein sequences. The amino acid sequence of a protein is the key to understanding its structure and function in the cell, so we present a new method to analyze protein primary sequence in this paper. The method is based on the graphical representation and conditional probability taken as the numerical characterization of the protein sequence. The demonstrable significance of the new method is that it cannot only analyze similarity/dissimilarity of protein sequences, but also provide more biological information about the protein sequences. According to the IE curve, we can compare the numbers of amino acids of the internal and external groups at different positions. Also the IA curve can be used to compare the numbers of amino acids of the internal and ambivalent groups at different positions. The EA curve can be used to compare the numbers of amino acids in the external and ambivalent groups at different positions. Therefore the three curves show the distribution of the three types of amino acids. Furthermore, the conditional probability reflected the distribution of the two adjacent amino acids. The new approach was applied to ND6 protein sequences of several species and results have shown that the introduction of hydropathy profile of amino acids into protein sequence is effectual and feasible.

Filovirus Tropism: Cellular Molecules for Viral Entry

In human and non-human primates, filoviruses (Ebola and Marburg viruses) cause severe hemorrhagic fever. Recently, other animals such as pigs and some species of fruit bats have also been shown to be susceptible to these viruses. While having a preference for some cell types such as hepatocytes, endothelial cells, dendritic cells, monocytes, and macrophages, filoviruses are known to be pantropic in infection of primates. The envelope glycoprotein (GP) is responsible for both receptor binding and fusion of the virus envelope with the host cell membrane. It has been demonstrated that filovirus GP interacts with multiple molecules for entry into host cells, whereas none of the cellular molecules so far identified as a receptor/co-receptor fully explains filovirus tissue tropism and host range. Available data suggest that the mucin-like region (MLR) on GP plays an important role in attachment to the preferred target cells, whose infection is likely involved in filovirus pathogenesis, whereas the MLR is not essential for the fundamental function of the GP in viral entry into cells in vitro. Further studies elucidating the mechanisms of cellular entry of filoviruses may shed light on the development of strategies for prophylaxis and treatment of Ebola and Marburg hemorrhagic fevers.

Ebola virus (EBOV) and Marburg virus (MARV), classified as biosafety level 4 agents, belong to the Family Filoviridae. Whereas MARV consists of a single species, Lake Victoria Marburgvirus, there are four distinct EBOV species, including Zaire ebolavirus (ZEBOV), Sudan ebolavirus (SEBOV), Côte d'Ivoire ebolavirus (CIEBOV), Reston ebolavirus (REBOV), and the proposed new species Bundibugyo ebolavirus (BEBOV) Towner et al., 2008) (Figure 1 left). Among these, ZEBOV, first identified in 1976, seems to be the most virulent, killing approximately up to 90% of infected individuals, whereas REBOV, which was initially isolated from cynomolgus monkeys imported from the Philippines into the USA in 1989, is less pathogenic in experimentally infected non-human primates (Fisher-Hoch and McCormick, 1999) and has never caused lethal infection in humans . Ebola virus and Marburg virus are filamentous, enveloped, non-segmented, single-stranded, negative-sense RNA viruses (Figure 2) . The viral genome encodes seven structural proteins, nucleoprotein (NP), polymerase cofactor (VP35), matrix protein (VP40), glycoprotein (GP), replication-transcription protein (VP30), minor matrix protein (VP24), and RNA-dependent RNA polymerase (L). EBOV also expresses at least one secreted nonstructural glycoprotein (sGP). Figure 3 summarizes filovirus replication in cells. At the first step of replication, viral attachment through interaction between GP and some cellular molecules is followed by endocytosis, including macropinocytosis (Nanbo et al., 2010; Saeed et al., 2010) . Subsequent fusion of the viral envelope with the host cell endosomal membrane releases the viral proteins (i.e., NP, VP35, VP30, and L) and RNA genome into the cytoplasm, the site of replication. Transcription of the negative-sense viral RNA by the viral polymerase complex (VP35 and L) yields mRNAs that are translated at cellular ribosomes. During replication, full-length positive-sense copies of the viral genome are synthesized. They subsequently serve as templates for replication of negative-sense viral RNA synthesis. At the plasma membrane, NP-encapsidated full-length viral RNAs and the other viral structural proteins are assembled with VP40 and GP and incorporated into enveloped virus particles that bud from the cellsurface (Noda et al., 2006; Bharat et al., 2011) . Though filoviruses show broad tissue tropism, hepatocytes, endothelial cells, dendritic cells, monocytes, and macrophages are thought to be their preferred target cells, and infection of these cells is important for hemorrhagic manifestation and immune disorders (Geisbert and Hensley, 2004) . Filoviruses are known to cause severe hemorrhagic fever in human and non-human primates, but recent studies suggest that quadrupeds are also naturally susceptible to EBOV infection (Figure 1, right) . In 2008-2009, REBOV infection was confirmed for the first time in pigs in the Philippines (Barrette et al., 2009) . REBOV was occasionally isolated from the samples subjected to the diagnostic investigation of multiple outbreaks of a respiratory and abortion disease syndrome in swine, which were caused by porcine reproductive and respiratory syndrome virus, common in pigs in Asia. It is speculated that REBOV became detectable, most likely due to the coinfection with this porcine virus. Although pathogenicity of these swine REBOV strains to humans, non-human primates, or even pigs remains unclear, other EBOV species (i.e., ZEBOV) was shown to cause severe respiratory disease in experimentally infected pigs (Kobinger et al., 2011) . During the 2001-2003 ZEBOV outbreaks in Gabon and the Democratic Republic of the Congo (DRC), when large numbers of FIGURE 1 | Phylogenetic analysis of filovirus GP amino acid sequences. The phylogenetic tree was constructed using the neighbor-joining method. For construction of this tree, ten complete GP amino acid sequences were used. Infectious viruses were isolated or viral genome and/or specific antibodies were detected (in parentheses) from the animals shown on the right. gorillas and chimpanzees were infected, the viral genome was also detected in duikers, medium-sized Bovid related to antelopes and gazelles (Leroy et al., 2004) . It was also reported that several dogs in the ZEBOV-epidemic area might have been highly exposed to the virus by eating infected dead animals, as suggested by high seroprevalence, but the putative infection seems to be asymptomatic (Allela et al., 2005) . Infectious MARV was recently isolated from Egyptian fruit bats (Rousettus aegyptiacus) in Uganda, indicating that this species is susceptible to MARV infection and potentially acts as the natural reservoir of the virus . Phylogenetic analysis showed that viruses in the bats were closely related to those isolated from victims of the 2007 MARV outbreak in Uganda, providing the first evidence for an epidemiological link between viruses in bats and hemorrhagic fever outbreak in humans. On the other hand, EBOV has not been isolated from any bat species. During the 2001-2003 EBOV outbreaks in Gabon and DRC, however, fruit bats (Hypsignathus monstrosus, Epomops franqueti, and Myonycteris torquata) captured in the outbreak area were found to have EBOV genomic RNA and virus-specific antibodies , suggesting they are potential natural reservoirs for EBOV. However, it is still unclear whether these bats continuously maintain EBOV and/or MARV and act as a potential source of filovirus transmission to humans. It has been shown that laboratory animals, including mice and guinea pigs, are susceptible to filovirus infection. However, these animals infected with filoviruses obtained from patients normally develop a non-lethal illness, though the viruses have the ability to replicate in the animals. Guinea pigs have been used as an animal model for filovirus infection since serial passage of MARV and EBOV in the animals results in a substantial increase in lethality (Bowen et al., 1980; Hevey et al., 1997; Volchkov et al., 2000; Subbotina et al., 2010) . It was also demonstrated that passages of ZEBOV through young mice resulted in the selection of variants with pathogenicity associated with mutations in viral internal genes (e.g., NP and VP40) (Bray et al., 1998; Ebihara et al., 2006) . This mouse-adapted ZEBOV is highly lethal to mice. Similarly, a mouse model for MARV infection has been established (Warfield et al., 2009) . Interestingly, mutations found in the GP gene of these mouse-or guinea pig-adapted viruses were not the primary factor for efficient replication in mice and guinea pigs, suggesting the importance of some other mechanisms underlying in viral replication and/or immune evasion, as shown in the pathogenesis of influenza virus (Fukuyama and Kawaoka, 2011) . The fourth gene from the 3 end of the filovirus genome encodes the viral envelope GP (Figure 2) , which is responsible for both receptor binding and fusion of the virus envelope with the host cell membrane (Takada et al., 1997; Wool-Lewis and Bates, 1998) (Figures 3 and 4) . GP is highly glycosylated with large amounts of N-and O-linked glycans, most of which are uniformly located in the middle one-third of the GP, designated the mucin-like region (MLR) Manicassamy et al., 2007) . The amino acid sequences of the MLR are highly variable among filovirus species (Sanchez et al., 1996 (Sanchez et al., , 1998 . GP undergoes proteolytic cleavage by host proteases such as furin (Volchkov et al., 1998) , which produces two subunits, GP1 and GP2, linked by a disulfide bond. The GP1 subunit mediates viral attachment, most likely through the MLR or the putative receptor binding region (RBR; Kuhn et al., 2006; Dube et al., 2009) . The GP2 subunit has the heptad repeat regions required for assembling GP as a trimer. The hydrophobic fusion loop on GP2 is thought to catalyze fusion of the viral envelope and host cell membrane (Weissenhorn et al., 1998; Ito et al., 1999) . Although the trigger to promote the conformational change leading to membrane fusion is not fully understood, it was recently suggested that endosomal proteolysis of EBOV and MARV GPs by cysteine proteases such as cathepsins B and L plays an important role in inducing membrane fusion (Chandran et al., 2005; Schornberg et al., 2006; Matsuno et al., 2010a) . Since GP is the only viral surface GP, it is believed to have an important role in controlling the tropism and pathogenesis of filovirus infection Hoenen et al., 2006; Sanchez et al., 2007) . In the early years, studies of filoviruses were hampered by its extraordinary pathogenicity, which requires biosafety level 4 containment. To circumvent this problem, pseudotype virus systems for functional analysis of filovirus GPs have been established (Takada et al., 1997; Wool-Lewis and Bates, 1998) . The systems rely on recombinant viruses (e.g., replication-competent or -incompetent vesicular stomatitis virus and retroviruses) that contain filovirus GP instead of their own GPs (Figure 5 ). Such pseudotype virus systems enable us to investigate cell tropism mediated by simple interaction between filovirus GP and its cellular ligands. Using such a system, it was shown that pseudotyped viruses infected primate cells more efficiently than any of the other mammalian or avian cells examined, in a manner consistent with the host range tropism of Ebola virus, and that cell-surface GPs with N-linked oligosaccharide chains might contribute to the entry of Ebola viruses, presumably acting as a specific receptor and/or cofactor for virus entry (Takada et al., 1997) . Furthermore, filovirus receptor-deficient cell lines that have been used in expression cloning strategies searching for filovirus entry mediators were discovered in an early study (Wool-Lewis and Bates, 1998) . Thus, Frontiers in Microbiology | Virology FIGURE 2 | Structure of Ebola virus particle and genome organization. Electron micrograph of Ebola virus particle (A), its diagram (B), and negative-sense genome organization (C) are shown. Viral protein names and functions are described in the text. Transcribing the glycoprotein (GP) gene produces a soluble GP (sGP). Transcriptional editing accompanied by frame shifting is required to produce full-length, membrane-anchored GP, which shares its first 295 amino acid residues with sGP. pseudotype virus systems are an essential tool for recent filovirus receptor research. Although filoviruses can replicate in various tissues and cell types, the molecular mechanisms of their broad tropism remain poorly understood (Figure 6) . By using an expression cloning strategy that has been used to identify several virus receptors, human folate receptor-α was first identified as a ubiquitous cellular cofactor that mediates infection by both MARV and ZEBOV (Chan et al., 2001) . This molecule is a glycosyl-phosphatidylinositollinked protein expressed on the cell-surface. However, a human immunodeficiency virus pseudotyped with EBOV GP could not www.frontiersin.org pseudotyped with filovirus GP. The pseudotype virus relies on a recombinant virus that contains a reporter gene instead of the viral envelope protein gene responsible for receptor-binding and membrane fusion. Since filovirus glycoprotein is efficiently incorporated into VSV particles, a recombinant VSV that contains the green fluorescent protein (GFP) gene, instead of the G protein gene can be generated. This virus is not infectious unless the envelope protein responsible for receptor binding and membrane fusion is provided in trans. infect T-cell lines stably expressing this protein, suggesting that folate receptor-α is not sufficient to mediate entry (i.e., some other molecules are required) (Simmons et al., 2003b; Sinn et al., 2003) . A similar approach identified members of the Tyro3 receptor tyrosine kinase family (Axl, Dtk, and Mer) as molecules involved in cell entry of filoviruses (Shimojima et al., 2006) . Expression of these family members in lymphoid cells, which are originally non-permissive to filoviruses, enhanced infection by pseudotype viruses bearing filovirus GPs on their envelopes. These molecules are widely distributed in many types of cells throughout the body, though not on lymphocytes and granulocytes (Linger et al., 2008) . A more recent study demonstrated that reduction of Axl expression by RNAi treatment resulted in decreased ZEBOV entry via. macropinocytosis but had no effect on the clathrin-dependent or caveola/lipid raft-mediated endocytic mechanisms, suggesting that Axl enhances macropinocytosis . However, direct interactions between these cellular molecules and the GP RBR remain to be demonstrated. Recently, a bioinformatics approach, comparative genetics analysis, was used to screen the candidate genes involved in ZEBOV entry and T-cell immunoglobulin and mucin domain 1 (TIM-1) was identified as a candidate ZEBOV and MARV cellular receptor by correlation analysis between the gene expression profiles and permissiveness to viral infection (Kondratowicz et al., 2011) . TIM-1 was shown to bind to the RBR of ZEBOV GP, and ectopic TIM-1 expression in poorly permissive cells enhanced EBOV infection. In addition, reduction of cell-surface expression of TIM-1 by RNAi decreased infection of highly permissive Vero cells, which are commonly used for filovirus propagation. However, the fact that not all cell types that are naturally permissive for filoviruses express the above-mentioned molecules implies that filoviruses may utilize multiple cellular proteins for infection of a wide variety of cells. More recent studies suggest that endo/lysosomal cholesterol transporter protein Niemann-Pick C1 (NPC1) is essential for filovirus infection, providing a model of EBOV infection in which cleavage of the GP1 subunit by endosomal cathepsin removes heavily glycosylated regions to expose the putative RBR, which is a ligand for NPC1 and mediates membrane fusion by the GP2 subunit (Carette et al., 2011; Côté et al., 2011) . Available data indicate that the cellular tropism of filoviruses does not necessarily match the distribution of any cellular molecules so far identified. Importantly, it remains elusive whether these molecules act as functional receptors that mediate both viral attachment and membrane fusion or as so-called co-receptors whose interaction with viral GP is required only for membrane fusion. Both MARV and EBOV GPs contain both N-and O-linked carbohydrate chains with different terminal sialylation patterns that seem to depend on the virus strains and cell lines used for their propagation. The MLR contains a number of potential N-and Olinked glycosylation sites as mentioned above. Though the MLR is found in all known filovirus GPs, its highly variable amino acid sequences and sugar chain structure suggest different GP properties among filovirus species. Interestingly, it is well documented that deletion of the MLR does not affect the fundamental function of GP in viral entry into cells in vitro, as indicated by the observation that pseudotyped viruses bearing GP lacking the MLR infect primate epithelial cells (e.g.,Vero E6 cells) similarly or rather more efficiently than viruses with wild-type GP (Simmons et al., 2002; Takada et al., 2004; Matsuno et al., 2010a) . According to the crystal structure of ZEBOV GP in its trimeric, prefusion conformation, the MLR may restrict access of the putative RBR to virus receptors (Lee et al., 2008) . Thus, pseudotyped viruses bearing MLR-deletion mutant GP have often been used for approaches to identify filovirus-specific receptors (Shimojima et al., 2006; Kondratowicz et al., 2011) . However, the MLR plays an important role in filovirus entry into preferred target cells such as endothelial cells, hepatocytes, and antigen-presenting cells, whose infection is likely involved in tropism and pathogenesis of filoviruses, as described below. Virus particles attach to the cell-surface through the interaction between GP and some cellular molecules (e.g., putative ubiquitous receptors, C-type lectins). Following virus uptake and trafficking to late endosomes, GP is cleaved by cellular proteases such as cathepsins to remove heavily glycosylated regions including the MLR and expose the RBR of GP1. Binding of cleaved GP1 to a coreceptor (e.g., NPC1) might be necessary for the GP conformational change leading to membrane fusion. C-type lectins are a family of Ca 2+ -dependent carbohydraterecognition proteins that play crucial roles in innate immunity. It has been demonstrated that membrane-anchored cellular C-type lectins facilitate filovirus infection in vitro by binding to glycans focused on the MLR (Figure 6) . The asialoglycoprotein receptor, a C-type lectin found exclusively in hepatocytes, initially proposed as a receptor for Marburg virus (Becker et al., 1995) , recognizes GPs displaying N-linked sugar chains with terminal galactose residues on the GP molecule and enhances filovirus infectivity. It was subsequently shown that carbohydrate chains on filovirus GP, especially on the MLR, are recognized by other cellular C-type lectins such as dendritic cell-and liver/lymph node-specific ICAM-3-grabbing non-integrin (DC/L-SIGN) (Alvarez et al., 2002; Lin et al., 2003; Simmons et al., 2003a; Marzi et al., 2004; Gramberg et al., 2008) , human macrophage galactose-type C-type lectin (hMGL) (Takada et al., 2004; Matsuno et al., 2010a) , and liver and lymph node sinusoidal endothelial cell C-type lectin (LSECtin) (Gramberg et al., 2005; Dominguez-Soto et al., 2007; Powlesland et al., 2008) . Though these C-type lectins show different specificities, depending on the structures of target glycans, and thus MLR may not be the only binding site for the lectins, all have been reported to promote filovirus entry. It should be noted that C-type lectins enhance filovirus infectivity when expressed on the target cell-surface, but are unlikely to act as functional receptors mediating both attachment and membrane fusion (Simmons et al., 2003a; Marzi et al., 2007; Matsuno et al., 2010b) . The fact that interaction between the GP MLR and C-type lectins is not essential for viral entry into cells lacking C-type lectins (e.g., Vero E6 cells) may also suggest that C-type lectins facilitate viral attachment but not infectious entry. Hepatocytes, endothelial cells, dendritic cells, monocytes, and macrophages, all of which express C-type lectins, are thought to be the preferred target cells of filoviruses Geisbert and Hensley, 2004; Hoenen et al., 2006) . Indeed, primary macrophage and dendritic cell cultures transduced for C-type lectin expression greatly increased their susceptibility to virus infection (Simmons et al., 2003a; Marzi et al., 2007) . While C-type lectins do not directly mediate filovirus entry, their pattern of expression in vivo and their ability to enhance infection indicate that C-type lectins can play an important role in filovirus transmission and tissue tropism. Thus, increased infection of these cells might be directly involved in the pathogenesis of filoviruses. Accordingly, it was shown that soluble mannose-binding C-type lectin played a role in protection from lethal Ebola virus infection in a mouse model (Michelow et al., 2011) . It should be noted that the ability to utilize the C-type lectins (i.e., DC-SIGN and hMGL) to promote cellular entry was correlated with the different pathogenicities among filoviruses (Takada et al., 2004; Marzi et al., 2006; Matsuno et al., 2010a) . Interestingly, the MLR amino acid sequence does not seem to be the primary factor contributing to the difference (Marzi et al., 2006; Matsuno et al., 2010a; Usami et al., 2011) . Although there might be some distinct mechanisms of entry between MARV and EBOV (Chan et al., 2000) , the similarity of tissue tropism and pathological features of infection between these viruses suggests that C-type lectins are one of www.frontiersin.org the important molecules, likely as attachment factors, for filovirus entry into cells, and that they are directly involved in filovirus tropism at the cellular level. In addition to the common receptor/co-receptor-dependent mechanism of cellular attachment and membrane fusion, some viruses utilize antiviral antibodies for their efficient entry into target cells (Takada and Kawaoka, 2003) . This mechanism is known as antibody-dependent enhancement (ADE) of viral infection. Filoviruses utilize virus-specific antibodies for their entry into cells in vitro through interaction between anti-GP antibodies and the cellular Fc receptor (FcR) or complement component C1q and its ligand, which likely promotes viral attachment to cells (Takada et al., , 2003a (Takada et al., , 2007 Nakayama et al., 2011) (Figure 6) . FcR are expressed exclusively on the cells of the immune system such as monocytes/macrophages, neutrophils, B-cells, and granulocytes (Fanger and Guyre, 1992) , whereas C1q ligands have been identified in most mammalian cells (Eggleton et al., 1998; Nicholson-Weller and Klickstein, 1999) , suggesting a ubiquitous mechanism for ADE of filovirus infection. By using GP-specific monoclonal antibodies, several epitopes recognized by ADE antibodies were identified and these epitopes were mostly located in the MLR of the GP1 subunit (Takada et al., 2007; Nakayama et al., 2011) . It should be noted that neutralizing antibodies appear to recognize different epitopes that are not located on the MLR (Takada et al., 2003b; Lee et al., 2008) . As reflected by the high variability of the MLR amino acid sequences and limited overall cross-reactivity of anti-sera among filovirus species (i.e., ZEBOV, SEBOV, CIEBOV, BEBOV, REBOV, and MARV), ADE activities of the anti-sera to GP are virus-species-specific (Takada et al., , 2007 Nakayama et al., 2010) . Interestingly, potential viral pathogenicity is correlated with the ability to induce ADE antibodies, suggesting the possible contribution of ADE to different pathogenicity between filoviruses Nakayama et al., 2011) . More importantly, the demonstration of ADE of filovirus infection raises fundamental questions about the development of GP-based vaccines and the use of anti-GP antibodies for passive immunization. Recently, GP has been used for viral vector-based or DNA vaccines that were shown to protect animals effectively. Replicationincompetent adenovirus expressing GP, a replication-competent vesicular stomatitis virus expressing GP, and a recombinant paramyxovirus expressing GP have been shown to protect nonhuman primates from lethal infections of filoviruses (Sullivan et al., , 2003 Jones et al., 2005; Bukreyev et al., 2007; Feldmann et al., 2007) . It should be noted that these vaccines potentially induce cytotoxic cellular response (i.e., CD8+ T lymphocytes) as well as antibody production, suggesting that activating cytotoxic T-cells is a key protective mechanism (Olinger et al., 2005; Sullivan et al., 2006; Reed and Mohamadzadeh, 2007) . Since cytotoxic T-cell response cannot be fully induced by immunization with non-replicative protein antigens such as inactivated virus and subunit vaccines, viral vector-based, or DNA vaccines may be promising in preventing filovirus infection. All enveloped viruses initiate infection by attaching to host cells followed by membrane fusion via interaction between viral surface proteins and receptor/co-receptor molecules on target cells, and this interaction is often a key determinant controlling viral tissue tropism and/or host range. As described above, it has been demonstrated that filoviruses utilize multiple molecules for their entry into cells. However, it remains elusive whether these molecules serve as functional receptors mediating both viral attachment and membrane fusion or play independent roles as either attachment receptors or fusion receptors. More importantly, none of the cellular molecules identified so far explains filovirus tissue tropism and host range reasonably. It might also be hypothesized that filoviruses do not use a single common receptor to infect a broad range of cells and, unlike many other viruses, may not need a "specific" receptor. Although the overall tropism and pathogenicity of filoviruses is controlled by multiple host cell factors (e.g., interactions with the host immune system), further studies aimed at identification of cellular molecules interacting with GP are needed to fully understand the mechanisms of cellular entry of filoviruses, and may shed light on the development of strategies for prophylaxis and treatment of Marburg and Ebola hemorrhagic fevers. I thank Kim Barrymore for editing the manuscript. This work was supported by the Takeda Science Foundation, and done within the framework of the Japan Initiative for Global Research Network on Infectious Diseases (J-GRID) and the Global COE Program "Establishment of International Collaboration Centers for Zoonosis Control" of the Ministry of Education, Culture, Sports, Science, and Technology (MEXT), Japan. The work was further supported by a Grant-in-aid from the Ministry of Health, Labor, and Welfare, Japan.

Replication-Competent Recombinant Porcine Reproductive and Respiratory Syndrome (PRRS) Viruses Expressing Indicator Proteins and Antiviral Cytokines

Porcine reproductive and respiratory syndrome virus (PRRSV) can subvert early innate immunity, which leads to ineffective antimicrobial responses. Overcoming immune subversion is critical for developing vaccines and other measures to control this devastating swine virus. The overall goal of this work was to enhance innate and adaptive immunity following vaccination through the expression of interferon (IFN) genes by the PRRSV genome. We have constructed a series of recombinant PRRS viruses using an infectious PRRSV cDNA clone (pCMV-P129). Coding regions of exogenous genes, which included Renilla luciferase (Rluc), green and red fluorescent proteins (GFP and DsRed, respectively) and several interferons (IFNs), were constructed and expressed through a unique subgenomic mRNA placed between ORF1b and ORF2 of the PRRSV infectious clone. The constructs, which expressed Rluc, GFP, DsRed, efficiently produced progeny viruses and mimicked the parental virus in both MARC-145 cells and porcine macrophages. In contrast, replication of IFN-expressing viruses was attenuated, similar to the level of replication observed after the addition of exogenous IFN. Furthermore, the IFN expressing viruses inhibited the replication of a second PRRS virus co-transfected or co-infected. Inhibition by the different IFN subtypes corresponded to their anti-PRRSV activity, i.e., IFNω5 ° IFNα1 > IFN-β > IFNδ3. In summary, the indicator-expressing viruses provided an efficient means for real-time monitoring of viral replication thus allowing high‑throughput elucidation of the role of host factors in PRRSV infection. This was shown when they were used to clearly demonstrate the involvement of tumor susceptibility gene 101 (TSG101) in the early stage of PRRSV infection. In addition, replication‑competent IFN-expressing viruses may be good candidates for development of modified live virus (MLV) vaccines, which are capable of reversing subverted innate immune responses and may induce more effective adaptive immunity against PRRSV infection.

More than 20 years after initial reports [1] [2] [3] , porcine reproductive and respiratory syndrome virus (PRRSV) continues to be a global swine industry problem with losses in the U.S. approaching $6 billion over the last decade [4] . Belonging to the arteriviridae family in the order nidovirales, PRRSV is an enveloped RNA virus containing a single positive-strand RNA genome. The 15 kb viral RNA genome consists of seven open reading frames (ORF1- 7) . ORF1 comprises about 80% of the genome and encodes proteins with protease, replicase and regulatory functions. The smaller overlapping ORF2-7 encode five minor (GP2a, GP3, GP4, 5a and E proteins) and three major (GP5, M and N proteins) structural proteins [5] [6] [7] . Several studies have shown that PRRSV possesses the capacity to subvert early innate immune responses in pigs by suppressing the production of antiviral cytokines [8] [9] [10] [11] [12] [13] [14] [15] , which also contributes to ineffective B-and T-cell responses [16] [17] [18] [19] . Superimposed on this suppressive activity is a high viral mutation rate, which has made the development of vaccines challenging [20] . Modified live vaccines (MLV) used for control of PRRSV in the U.S. are based on only two virus isolates [20, 21] . Although MLV protect against some homologous field strains, their efficacy is not satisfactory due to failure to protect against infections of heterologous strains, as well as the potential risk for reversion to virulence [20, 21] . To develop successful vaccines against PRRSV infections, particularly those by heterologous strains, it is necessary to develop novel vector systems and to extensively categorize host factors critical in the virus-host interaction. Several viral vectors, including those based on pseudorabies virus, poxvirus, adenovirus and transmissible gastroenteritis coronavirus (TGEV) have been used to express PRRSV structural proteins [22] or host immune factors [23] [24] [25] for developing anti-PRRSV immunity. For example, humoral immunity against PRRSV GP5 protein was detected in pigs immunized with fowlpoxviruscoexpressing PRRSV GP5/GP3 and porcine IL-18 [23] , and in mice immunized with adenovirusexpressing GP5/GP3 fused with swine granulocyte-macrophage colony stimulating factor (GM-CSF) [25] . In this context, we and others have proposed to use PRRSV infectious cDNA clones [26, 27] or virus replicons [28] as vectors for the expression of immune effectors that potentiate innate and adaptive immunity against a broad range of PRRSV isolates. Here we show that PRRSV infectious clones are effective vector systems to express exogenous antigens and host immune effectors. Specifically, we have constructed serial replication-competent viruses from a PRRSV infectious clone-based vector expressing indicator proteins and porcine type I interferons (IFNs). The indicator protein-expressing PRRS viruses efficiently produce progeny viruses and provide an efficient means for real-time monitoring of viral replication, thus allowing high-throughput elucidation of the role of host factors in PRRSV infection. In addition, the replication of some IFN-incorporated viruses is associated with the expression of active IFN peptides, which are capable of counteracting the subverted innate immune response and with potential to induce more effective adaptive immunity against PRRSV infection. To investigate the potential of infectious PRRSV cDNA clones as a platform for gene manipulation [29, 30] , we first engineered an established infectious clone to express several indicator proteins including Renilla luciferase (Rluc), and green and red fluorescent proteins (GFP and DsRed, respectively). Coding regions of the indicator proteins were constructed in the junction regions surrounding ORF1b/ORF2a through introduced restrictive digestion sites as described [29, 30] ( Figure 1A ). The construct was designed to express the recombinant protein gene through the creation of an additional subgenomic mRNA. Plasmids of selected authentic clones were transfected into MARC-145 cells for the production of progeny viruses. As shown in Figure 1 , infectious clones efficiently produced progeny viruses with successful expression of indictor proteins ( Figure 1B ). Replication rates of these engineered viruses were similar to their parental virus as judged by monitoring ratios of virus-positive cells; and they produced comparable infectious virons shown by similar viral titers ( Figure 1C ). Further experiments with GFP-PRRSV have shown their similar infectivity as a parental strain in porcine alveolar macrophages (PAMs) as well as monocyte-derived dendritic cells (mDCs) ( Figure 1B and data not shown). The indicator-expressing viruses provide an efficient means for real-time monitoring of viral replication. Whereas GFP-and DsRed-PRRSV facilitated detection of fluorescent proteins after 16 h, the Rluc-PRRSV was useful for measuring Rluc activity from 5-20 h. These viruses allowed us to efficiently elucidate the role of some host factors in PRRSV infection. Several infectious cDNA clones have been generated from field PRRSV strains [31] [32] [33] [34] [35] [36] [37] , which provide efficient means for molecular manipulation of PRRSV genome and evaluation of molecular evolution of this RNA virus with a high mutation rate. The vector cassette used in this study was generated from an infectious clone of North American PRRSV strain P129 with two unique restriction sites and a copy of the transcription regulatory sequence of ORF6 (TRS6) inserted between ORFs 1b and 2a [26, 27, 29, 30] . In addition to GFP, which was introduced in PRRSV infectious clones in several previous studies [26, 27, 29, 31] , two other indicator proteins of DsRed and Rluc were introduced in the PRRSV cDNA clone in this study. The construction of DsRed and particularly Rluc into the PRRSV cDNA clone was intended to produce laboratory viruses that mimic their parental PRRSV with similar replication kinetics but are easily detected and quantified during the early phase of virus infection/replication. As shown in both MARC-145 cells and PAMs, all GFP, DsRed and Rluc expressing progeny viruses had similar replication kinetics as their parental PRRSV. For detection, the red fluorescence of DsRed labeling not only provided a counterstaining choice but also was more sensitive for microscopic observation in real-time rather than traditional immunostaining procedures in fixed cells (Rural Technologies, Brookings, SD, USA). In contrast to GFP and DsRed viruses, which were generally detectable after 12 h in infected cells using microscopy, the Rluc PRRSV in conjunction with an EnduRen™ in vivo substrate (Promega, Madison, WI) facilitated real-time detection of the virus replication as early as 5 h post infection in cells. This Rluc expressing PRRSV thus provides an efficient means for genome-wide examination of host factors involved in the early stages of virus infection [38, 39] . However, our attempts to clone the firefly luciferase gene (~1.6 kb) into the same PRRSV cDNA vector were unsuccessful, suggesting that the PRRSV clone vector has a limited capacity for incorporation of exogenous genes at about 2 kb. TSG101 is a housekeeping protein and has been implicated in a number of cellular functions, including mitotic spindle formation, genome stability and endosomal sorting [40] . Essential in endosomal sorting, TSG101 interacts directly with ubiquitinated proteins and internalizes them into the multivesicular body pathway for degradation. During replication, viruses require a retrograde movement from the cell interior to the outer membrane [40] . A number of reports have shown that TSG101 is involved in the virus fusion/budding process from the cellular membrane. These viruses include human immunodeficiency virus (HIV) and hepatitis C virus (HCV) [40] [41] [42] . Recently, a tentative TSG101-targeting peptide, FGI-104, was shown to have a broad-spectrum ability to inhibit infections by several viruses including HIV, HCV and PRRSV [43] . However, there is no direct mechanistic evidence that has demonstrated the involvement of TSG101 in the control of PRRSV replication. To study this possibility, we produced MARC-145 cell lines with targeted suppression of endogenous TSG101 expression. MARC-145 cells were transfected with a pGFP-V-RS vector expressing a 29 nt shRNA (OriGene, Rockville, MD, USA) against a conserved region of TSG101. Two puromycin-resistant colonies showing significant suppression of TSG101 at both RNA and protein levels, were selected. As shown in Figure 2A , cells from colony 1 had less than 10% endogenous expression of tsg101 RNA, and those from colony 9 about 20% of endogenous expression of tsg101 RNA, as well as 70-80% reduction in TSG101 protein. We then compared PRRSV replication in these TSG101-suppressed cells with control MARC-145 cells or cells transfected with scrambled shRNA constructs. Using either GFP-or DsRed-expressing PRRSV, the retarded replication of viruses in TSG101-suppressed cells was demonstrated between 12 and 48 h after infection ( Figure 2B ,C). However, by 72 h, viral replication among TSG101-suppressed and control cells was not significantly different ( Figure 2D ). The return of PRRSV replication in TSG101-suppressed cells might result from virus-stimulated expression of TSG101 and/or incomplete suppression. The earlier and more quantitative comparison of PRRSV replication among TSG101-suppressed and control cells was conducted with the Rluc-expressing PRRSV. Significantly retarded viral replication was detected as early as 5 h post Rluc-PRRSV infection with the in vivo Rluc substrate. The difference in PRRSV replication was quantitatively correlated to the TSG101 levels in different group of cells (Figure 2A and 2E), and could be monitored until 20 h post infection when the substrate was limited. Notably, the replication retardation of indicator protein-expressing viruses in TSG101-silent cells was mostly due to TSG101 suppression because these bioengineered viruses have similar replication kinetics as their parental viruses in normal cells (Figure 1 ). Using the kinetics (i.e., 12-20 h post infection) defined by the indicator protein-expressing viruses, we reproducibly detected retarded replication rates of wildtype PRRS viruses in TSG101 suppressed cells (data no shown). To further test the role of TSG101 in PRRSV infection in porcine cells or pigs, we have characterized the porcine TSG101 cDNA sequence (GenBank TM accession numbers JN882576). The full-length porcine TSG101 cDNA is 1580 bp encoding a precursor protein of 391 residues. TSG101 genes are conserved with most mammalian homologs sharing >94% identity at both RNA and protein levels. The shRNA template sequence we used for loss-of-function studies in MARC-145 cells is identical among human, monkey and porcine TSG101 cDNA. Transient transformation of the shRNA into porcine mDCs similarly suppressed GFP, DsRed-and Rluc-expressing PRRSV until 48 h post infection (data not shown). To determine the potential of host factors as a means of counteracting viral immunomodulating activity and stimulating effective antiviral immunity [44] , we bioengineered the infectious clones to express a variety of type I IFNs, which were selected because of their well-documented role in PRRSV infections and activity in mediation of antiviral immunity [9, 15, [45] [46] [47] . In addition, the suppression of type I IFN expression by PRRSV infection has been well documented [15] . Shown in Figure 3 are data from the infectious clone expressing four type I IFNs (IFNα1, IFNβ, IFNδ3 and IFNω5). At 4 d post transfection with equivalent cDNA clones, virus-positive cells were detected in cells transfected with all four infectious clones but with different densities. Where IFNδ3-PRRSV propagated similarly as the control GFP-virus, the replication rates of IFN-β-, IFNα1-and IFNω5-viruses were attenuated by approximately 70% (β-type) or 90% (α1-& ω5-types) ( Figure 3A-H) . Attenuation of IFN-expressing viruses could be caused by viral genome alteration or more likely by the replication-associated expression of active IFN peptides, which was consistent with the anti-PRRSV activity of these IFN subtypes [15] ( Figure 3C ). In contrast, IFNδ3-and GFP-type viruses replicated well. To confirm these findings, we counterstained IFNα1-virus infected cells with antibodies against porcine IFNα (R&D, Minneapolis, MN, USA) and PRRSV nucleocapsid (N) protein ( Figure 4A-D) . Co-localization of IFNα-and PRRSV-labeling ( Figure 4C ) indicated that IFN polypeptides were expressed by the engineered viruses during infection. Furthermore, the IFN expressing viruses inhibited the replication of a second PRRS virus co-transfected or co-infected; and again, the intensity of the inhibition was consistent with the anti-PRRSV activity of these IFN subtypes, i.e., IFNω5IFNα1>IFN-β>IFNδ3 ( Figure 4E ). However, because the expression of IFN inhibits virus replication [22, 44] , engineered viruses with the most active IFN subtypes against PRRSV may not be good candidates for a modified live virus (MLV) vaccine. This limitation would prevent the preparation of a high titer virus as a vaccine. Several measures may be used to overcome these limitations, such as incorporating IFN subtypes that do not inhibit replication of a MLV strain but may up-regulate B-and T-cell responses [44] , or controlled expression/activation of the incorporated IFN peptides within certain temporal windows or cell types. Pigs have at least 35 functional type I IFN genes with diverse antiviral or immunoregulatory activity [47] , which provides several candidates for balancing antiviral and immunoregulatory activity to optimize the replication-competent recombinant viruses. Type I IFNs mediate antiviral responses through induction of IFN stimulated genes, such as MxA and RNase L. Direct incorporation of some ISGs (or their functional domains) into the PRRSV infectious clone provides an attractive alternative given that PRRSV-specific ISGs have been identified [15] . In addition, incorporating an exogenous gene tag (a compliance marker) at the vector backbone of a MLV will also allow differentiation of vaccinated from non-vaccinated animals [27] . As for the stability of the bioengineered viruses, we passed viruses through MARC-145 cells and PAMs for 5-6 generations. Authentic progeny viruses were rescued after three generations, but viruses with a titer higher than 10 3 were only rescued with the indicator protein expressing group and the one expressing IFNδ3. DsRed-and IFNδ3-expressing viruses remained stable for 6 generations (Data not shown). Viruses and cells: All virus and animal procedures were approved by the Kansas State University Biosafety and Institutional Animal Care and Use committees. The wild type strains of PRRSV used here were two North American PRRSV strains: NVSL97-7895 and SDSU28983 [26, 27] . The expression cassette was created by insertion of two unique restriction sites and a copy of the transcription regulatory sequence of ORF6 (TRS6) between ORFs 1b and 2a in the infectious cDNA clone of pCMV-P129 ( Figure 1A ) [29, 30] (gift from Dr. Jay G. Calvert, Pfizer Animal Health, Kalamazoo, MI, USA). All bioengineered viruses originated from the backbone of pCMV-P129 with incorporation of an exogenous gene between the two cloning sites. All PRRSV strains were propagated in a simian kidney cell line (MARC-145) or porcine alveolar macrophages (PAMs). MARC-145 cells were maintained in minimum essential medium (MEM) with 8% fetal bovine serum (FBS) and 1× penicillin/streptomycin and fungizone (Invitrogen, Grand Island, NY, USA) PAMs were obtained from lungs of 4-to 6-week-old pigs by lung lavage with PBS and cryopreserved in liquid nitrogen until use. In use, PAMs were plated in RPMI medium with 10% FBS plus 1X penicillin/streptomycin and fungizone. After 2 d, cells were infected with the virus [47] . Monocyte-origin dendritic cells (mDCs) were induced from porcine peripheral blood mononuclear cells (PBMCs) and infected with PRRS viruses as described in Loving et al. [48] . Production and titration of recombinant PRRS viruses with expression of exogenous genes: In brief, two restriction enzyme digestion sites (Afl II and Mlu I) were introduced into 5'-and 3'-ends of the coding regions, which were amplified from authentic cDNA clones using a high-fidelity PCR [15, 47] . The cloning primers used for this study are listed in the Supplemental Table 1 . The amplified coding regions were purified and cloned into the expression cassette. MARC-145 cells were transfected with authentic plasmids purified from E. coli clones for examination of the production of progeny viruses and expression of indicator proteins or IFNs. Transfection of MARC-145 in 24-well plates was performed with 2 μg of plasmid DNA using Lipofectamine™ 2000 (Invitrogen). Subsequent virus yields were measured by end-point titration of culture media on MARC-145 cells and PAMs. Serial 10-fold dilutions of virus were placed in 96-well tissue culture plates containing confluent MARC-145 cells or PAMs. Cells were fixed in 4% formaldehyde in PBS and the virus was detected by staining for the presence of nucleocapsid antigen using a mAb (SDOW-17, Rural Technologies, Brookings, SD, USA) labeled with TRITC-conjugated secondary antibodies. In addition, the recombinant viruses could be distinctly detected by the expression of fluorescent proteins. Results were reported as log TCID50/mL [15, 47] . The replication of Rluc-expressing PRRSV was monitored with the addition of an in vivo Renilla luciferase substrate (Promega, Madison, WI, USA) at 60 μM in cell culture medium and measured the luminescence after 2 h [49] . Determination of growth kinetics and stability of the recombinant viruses in cells: Culture supernatants from cells transfected with infectious clones were harvested at 5 d post-transfection and designated 'passage (P) 1'. The P1 virus was used to inoculate fresh MARC-145 cells to collect P2 then P3 at an interval of 4-5 d between successive passages. Each passage virus was titrated, aliquoted and stored at −80 °C until use. Growth curves of the rescued viruses were evaluated by inoculating MARC-145 cells with P3 viruses at a MOI of 0.1. Aliquots of the supernatants of infected cells were collected at points with 10 h intervals until 100 h and the virus was titrated by determining log TCID50/mL to monitor growth kinetics [26] . To determine the stability of the recombinant virus, the rescued P1 virus was passaged on MARC-145 and PAM cells until P6. Expression of indicator proteins and IFN along with virus replication was detected by RT-PCR, western blotting and immunofluorescence as described above [15, 47] . Producing MARC-145 cell lines with shRNA-mediated silencing of the Tsg101 gene: Three g of pGFP-V-RS vector expressing a scrambled shRNA or a 29 nt shRNA (OriGene, Rockville, MD, USA) against a conservative region between human and simian TSG101 cDNA, was used to transfect MARC-145 cells growing in 6-well culture plates. Transfected cells were selected with 0.8 µg/mL of puromycin (Invitrogen) to obtain individual colonies. Approximately 30 puromycin-resistant colonies were picked and two of them showing significant suppression of TSG101 at both RNA and protein levels were subcultured for loss-of-function studies of the role of TSG101 in PRRSV infection. Primers used for RT-PCR detection of tsg101 were generated against consensus sequence of monkey, human and porcine tsg101 cDNAs (GenBank TM accession numbers, NM_001195481, NM_006292 and JN882576, respectively), which allowed us to detect both monkey and porcine tsg101 with the same pair of primers. The primers were 5′-ATACCCTCCCAATCCCAGTGGTTA-3′ (sense) and 5′-ATCCATYTCCTCCTTCATCCGCCA-3′ (antisense, Y = C or T). Anti-TSG101 monoclonal antibody was purchased from Sigma-Aldrich (St. Louis, MO, USA). The involvement of TSG101 in PRRSV infection was evaluated by comparing the replication kinetics of PRRS viruses between cells with or without TSG101-suppression [47] . Data analyses: Virus titrations were done with at least three repeats to report log TCID50/mL . Relative gene-expression data of real-time PCR was normalized against C t values of the housekeeping gene (GAPDH) and the relative expression index (2 −ΔΔCt ) was determined in comparison to the base levels of control samples. Growth curves were generated with Sigmaplot 11.0 (Systat, San Jose, CA, USA) and the densitometry analysis of images were done by using AlphaEase FC Software (Alpha Inotech, Santa Clara, CA, USA) as described [47, 50] . The PRRSV cDNA infectious clone, pCMV-129, has vector capacity to express exogenous genes of less than 2 kb, which allows reconstruction of the virus for bioengineering manipulation [29, 30] . Recombinant PRRS viruses expressing several indicator proteins have replication and infection dynamics similar to the parental strain in both MARC-145 and porcine cells. Therefore, they may be used to decipher the role of host factors in PRRSV infection [38, 39] . Using several indicator protein-expressing viruses, in particular the Rluc-expressing PRRSV, we showed that TSG101 was significantly involved in PRRSV infection at the early phase of the infection. Recombinant PRRS viruses expressing antiviral cytokines produce active cytokines in the infected cells and alter the replication of co-infected PRRSV. These constructs may be candidates for modified live virus vaccines, which could ameliorate subverted innate immune responses and potentially enhance adaptive immunity against PRRSV infection.

A Co-Opted DEAD-Box RNA Helicase Enhances Tombusvirus Plus-Strand Synthesis

Replication of plus-strand RNA viruses depends on recruited host factors that aid several critical steps during replication. In this paper, we show that an essential translation factor, Ded1p DEAD-box RNA helicase of yeast, directly affects replication of Tomato bushy stunt virus (TBSV). To separate the role of Ded1p in viral protein translation from its putative replication function, we utilized a cell-free TBSV replication assay and recombinant Ded1p. The in vitro data show that Ded1p plays a role in enhancing plus-strand synthesis by the viral replicase. We also find that Ded1p is a component of the tombusvirus replicase complex and Ded1p binds to the 3′-end of the viral minus-stranded RNA. The data obtained with wt and ATPase deficient Ded1p mutants support the model that Ded1p unwinds local structures at the 3′-end of the TBSV (−)RNA, rendering the RNA compatible for initiation of (+)-strand synthesis. Interestingly, we find that Ded1p and glyceraldehyde-3-phosphate dehydrogenase (GAPDH), which is another host factor for TBSV, play non-overlapping functions to enhance (+)-strand synthesis. Altogether, the two host factors enhance TBSV replication synergistically by interacting with the viral (−)RNA and the replication proteins. In addition, we have developed an in vitro assay for Flock house virus (FHV), a small RNA virus of insects, that also demonstrated positive effect on FHV replicase activity by the added Ded1p helicase. Thus, two small RNA viruses, which do not code for their own helicases, seems to recruit a host RNA helicase to aid their replication in infected cells.

All eukaryotic plus-stranded (+)RNA viruses have similar replication cycles in infected cells. After translation of their mRNA-sense genomic RNA(s), the viral RNA and the viral replication proteins are recruited to the site of viral replication in membranous compartments. After the assembly of the membranebound viral replicase complexes (VRC), the viral replicase uses the viral RNA as a template to produce complementary (2) RNA. This is then followed by (+)-strand synthesis in an asymmetric manner, producing excess amounts of (+)-strand progeny, which is released from replication for other viral processes. For efficient replication, (+)RNA viruses recruit numerous host proteins [1] [2] [3] [4] [5] . Among the identified host proteins are RNA-binding proteins, such as translation factors, ribosomal proteins and RNAmodifying enzymes [1] . The co-opted host proteins likely affect several steps in viral RNA replication, including the assembly of the replicase complex and/or viral RNA synthesis. However, the functions of host factors in (+)RNA virus replication are known only for a small number of host factors [1] [2] [3] [4] [5] [6] [7] [8] [9] . Tomato bushy stunt virus (TBSV) is a model plant RNA virus that is used intensively for identification and characterization of host factors [1, [10] [11] [12] . The single genomic RNA codes for two replication proteins, p33 and p92 pol , which are sufficient to support TBSV replicon (rep)RNA replication in yeast (Saccharomyces cerevisiae) model host [13, 14] . p33 and p92 pol are components of the membrane-bound VRC that also requires the tombusviral (+)repRNA as a platform during its assembly and activation [11, [15] [16] [17] . Recent genome-wide screens and global proteomics approaches with TBSV based on yeast revealed a large number of host factors interacting with viral components or affecting TBSV replication. A large fraction of the identified host proteins are RNA binding proteins, which might affect viral RNA synthesis [10, [18] [19] [20] . The highly purified tombusvirus VRC is known to contain at least six permanent resident host proteins, including the heat shock protein 70 chaperones (Hsp70, Ssa1/2p in yeast) [21] [22] [23] [24] , glyceraldehyde-3-phosphate dehydrogenase (GAPDH, encoded by TDH2 and TDH3 in yeast) [25] , pyruvate decarboxylase (Pdc1p) [24] , Cdc34p E2 ubiquitin conjugating enzyme [24] [25] [26] , eukaryotic translation elongation factor 1A (eEF1A) [27, 28] , eukaryotic translation elongation factor 1Bgamma (eEF1Bc) [29] and two temporary resident proteins, Pex19p shuttle protein [30] and the Vps23p adaptor ESCRT protein [28, 31, 32] . Although the functions of several of these proteins have been studied in some details, additional host proteins might be also present in tombusvirus VRC [5, 10, [21] [22] [23] 25] . One of the intriguing questions is if tombusviruses use host-coded helicases for their replication. This is because, unlike larger RNA viruses, the tombusvirus genome does not code for a protein with helicase function. Thus, can tombusviruses and other small RNA viruses replicate without the use of a helicase or they subvert a cellular helicase(s) to assist their replication? DEAD-box proteins constitute the largest family of RNA helicases that are involved in all aspects of cellular metabolism and perform RNA duplex unwinding and remodeling of RNA-protein complexes in cells [33] [34] [35] . Ded1p, which is essential for yeast growth, is among the best-characterized DEAD-box helicases [36] . It is involved in initiation of translation of every yeast mRNA [37] [38] [39] and down-regulation of Ded1p level also reduced p33 and p92 levels in yeast [20] . Therefore, not surprisingly, TBSV replication also decreased significantly in yeast with reduced level of Ded1p. Due to its essential role in translation, it is critical to separate the possible role of Ded1p in viral replication from its effect on viral translation. In this paper, a yeast-based cell-free TBSV replication assay and recombinant Ded1p was used to test the role of Ded1p in TBSV replication. We define that Ded1p plays a role in enhancing plusstrand (+)RNA synthesis by the viral replicase. We also find that Ded1p is a component of the tombusvirus VRC and binds to the 39-end of the viral minus-stranded (2)RNA. The obtained data support the model that Ded1p unwinds the 39-end of the TBSV (2)RNA, rendering the RNA compatible for initiation of (+)strand synthesis. Interestingly, we find that Ded1p and GAPDH play non-overlapping functions to enhance (+)-strand synthesis. Altogether, the two host factors enhance TBSV replication synergistically by interacting with the viral (2)RNA and the replication proteins. Reduced level of TBSV RNA replication in yeast cell-free extract depleted for Ded1p DEAD-box RNA helicase Since Ded1p is an essential translation factor for all yeast mRNAs [38, 39] , it is difficult to separate its direct versus indirect effect on TBSV replication in yeast model host. To circumvent this problem, we used whole cell extracts (CFE) prepared from yeast containing reduced level of Ded1p ( Figure S1 ) to support cell-free TBSV replication. As we have shown previously, TBSV (+)RNA can perform one complete cycle of replication in the CFE-based replication assay when purified recombinant p33 and p92 pol replication proteins are provided [23, 40] . The CFE-based replication assay showed that both (+) and (2)RNA synthesis decreased by over 4-fold when Ded1p was down-regulated as compared with the control CFE prepared from yeast with high level of Ded1p expression ( Figure S1 , lanes 4-6 versus 1-3). Thus, these data confirm that Ded1p is important for TBSV replication. However, the observed decrease in the CFE-based TBSV replication could be due to either reduced replication when Ded1p is limiting (direct effect) or lesser amounts of other critical host factors needed for TBSV replication in the CFE with reduced level of Ded1p (indirect effect due to Ded1p's role in host translation). To address these points, we first studied if Ded1p is present within the tombusvirus replicase and then, what is the mechanistic role of Ded1p during TBSV RNA synthesis. To examine if Ded1p is present within the tombusvirus replicase complex, we FLAG affinity-purified the tombusvirus replicase from yeast cells actively replicating TBSV repRNA [14, 28] . This yeast also expressed HA-tagged Ded1p from yeast chromosome based on the native promoter. We found that the tombusvirus replicase preparation, which is highly active on added templates in vitro (not shown), contained Ded1p ( Figure 1 , lane 4), while Ded1p was undetectable in the control yeast sample obtained using the same affinity purification ( Figure 1 , lane 3). The control yeast also expressed the tombusvirus replication proteins (including the 66His-tagged p33, but lacking FLAG-tag) and the HA-tagged Ded1p from yeast chromosome based on the native promoter and supported TBSV repRNA replication (not shown). We also FLAG-affinity-purified 66His/FLAG-tagged Ded1p from the membrane-fraction of yeast co-expressing HA-p33 and HA-p92 and found that the purified Ded1p preparation contained HA-tagged p33 ( Figure S2A , lane 1). The affinity-purified Ded1p preparation also showed TBSV replicase activity in vitro on added TBSV template ( Figure S2B , lane 2), suggesting that the membrane-bound Ded1p is associated with the active tombusvirus replicase. Altogether, these data support the model that Ded1p is part of the tombusvirus replicase. Figure 1 . Co-purification of Ded1p with the p33 replication protein from yeast. The FLAG/66His-tagged p33HF or 66His-tagged p33H was purified from yeast extracts using a FLAG-affinity column (lanes 3-4). Top panel: Western blot analysis of Ded1p tagged with 66HA, expressed from the natural promoter in the chromosome, with anti-HA antibody in the purified p33 preparations. Bottom panel: Western blot analysis of HF-tagged p33 or 66His-tagged p33 with anti-His antibody. Note that ''total'' represents the total protein extract from yeast expressing the shown proteins. Each experiment was repeated three times. doi:10.1371/journal.ppat.1002537.g001 Subverted host factors play a role in plus-strand RNA virus replication. Small RNA viruses do not code for their own helicases and they might recruit host RNA helicases to aid their replication in infected cells. In this paper, the authors show that the Ded1p DEAD-box helicase, which is an essential translation factor in yeast, is recruited by Tomato bushy stunt virus (TBSV) into its replicase complex. They also show that Ded1p binds to the viral (2)RNA and promotes (+)-strand TBSV synthesis when added to a yeast-based cell-free extract depleted for Ded1p. An ATPase defective Ded1p mutant failed to promote TBSV replication in vitro, suggesting that the helicase activity of Ded1p is essential for its function during TBSV replication. In addition, the authors also show that another host protein, which also binds to the (2)RNA, namely glyceraldehyde-3-phosphate dehydrogenase (GAPDH), further enhances TBSV (+)RNA when added together with Ded1p to yeast-based cell-free extract. In summary, the authors show that the major functions of Ded1p and GAPDH host proteins are to promote TBSV replication via selectively enhancing (+)-strand synthesis. Recombinant Ded1p DEAD-box helicase enhances TBSV RNA replication in yeast cell-free extract To address if Ded1p could play a direct role in TBSV replication, we added purified recombinant Ded1p to yeast CFE with reduced level of Ded1p (Figure 2A ). We observed a ,2-fold increased level of TBSV RNA replication (based on total singlestranded repRNA level) in the presence of recombinant Ded1p when compared with the MBP control ( Figure 2B , lane 2 versus 4). Interestingly, the amount of double-stranded RNA, which correlates with (2)-strand levels [23] , did not increase in the presence of Ded1p in the CFE assay ( Figure 2B, lanes 1 versus 3) . This finding suggests that (2)-strand RNA did not contribute to the 2-fold increase in total TBSV RNA levels when recombinant Ded1p was added to the CFE-based replication assay. Therefore, we conclude that the added recombinant Ded1p selectively increased TBSV (+)-strand RNA synthesis in vitro. To demonstrate if the ATPase/helicase activity of Ded1p is important for the enhancement of TBSV RNA replication in the CFE-based replication assay, we tested an ATPase defective mutant of Ded1p (D1 mutant) [41] . This mutant could not enhance TBSV replication in vitro ( Figure 2C , lanes 1-2 versus 3-4), suggesting that the ATPase activity of Ded1p is important for TBSV replication. To further test what step(s) Ded1p promotes during TBSV replication, we performed a two-step replication assay based on yeast CFE. In this assay, the first step includes the assembly of the replicase complex on the endogenous membranes present in CFE in the presence of the viral (+)repRNA, the p33/p92 replication proteins and ATP/GTP [23] ( Figure 3A , step 1). Under these conditions, the viral replication proteins recruit the repRNA to the membrane and the replicase becomes partially RNase and protease insensitive, but it cannot initiate minus-strand synthesis yet, due to the absence of CTP/UTP [23] . Then, centrifugation and washing the membranes will remove all the proteins and molecules not bound to the membrane. This is followed by addition of ATP/CTP/GTP/UTP to initiate RNA synthesis during the second step. Addition of purified Ded1p to the CFE during the first step did not increase TBSV replication ( Figure 3B , lane 4 versus 5). This suggests that Ded1p did not facilitate the assembly of the replicase complex, unlike other host factors, such as Hsp70 and eEF1A [23, 27] . Moreover, Ded1p was likely lost during the centrifugation/washing step in this assay due to its low association with the membrane. However, addition of Ded1p exclusively during the second step of the CFE assay resulted in ,2-fold increase in TBSV RNA replication ( Figure 3B , lane 14 versus 15), similar to the stimulatory effect of Ded1p during standard CFE replication assay ( Figure 2 ). Interestingly, a Ded1p mutant (D1) deficient in ATPase activity could not stimulate TBSV RNA synthesis in this assay ( Figure 3B , lane 13), while a mutant (D11, Figure S3C ) with increased ATPase activity [41] promoted TBSV RNA synthesis by ,2-fold ( Figure 3B , lane 11). Altogether, these findings indicate that Ded1p has a direct stimulatory function during TBSV RNA synthesis, while Ded1p is unlikely to affect viral RNA recruitment for replication or VRC assembly. To further test the direct effect of Ded1p on RNA synthesis, we utilized detergent-solubilized and affinity-purified tombusvirus -stranded RNA products produced in the cell-free TBSV replication assay. The dsRNA products are also shown on the right subpanel with high contrast for better visualization. Odd numbered lanes represent replicase products, which were not heat treated (thus both ssRNA and dsRNA products are present), while the even numbered lanes show the heat-treated replicase products (only ssRNA is present). The % of dsRNA and ssRNA in the samples are shown. Note that, in the nondenatured samples, the dsRNA product represents the annealed (2)RNA and the (+)RNA, while the ssRNA products represents the newly made (+)RNA products. Right replicase from Ded1p-depleted yeast ( Figure 4A ). This purified replicase can only synthesize complementary RNA products on added TBSV templates, but, unlike the above membrane-bound replicase in the CFE-based assay, it cannot perform a complete cycle of RNA synthesis [14, 15] . We found that addition of purified recombinant Ded1p to the purified tombusvirus replicase programmed with the (2)repRNA stimulated (+)-strand synthesis by ,2.5-3-fold ( Figure 4B , lanes 3-4 versus 1-2; and Figure S3D , lane 3 versus 2). Interestingly, Ded1p stimulated the production of the full-length (+)-strand RNA product, while the amount of 39-terminal extension product (39TEX; due to self-priming by the 39 end of the template [42] [43] [44] , Figure S3B ) decreased in the presence of added recombinant Ded1p. Therefore, we suggest that Ded1p facilitates the de novo initiation on the (2)RNA template by the tombusvirus replicase. The ATPase inactive mutant D1 ( Figure 4B , lanes 5-6 and S3D, lane 4), D5 and D10 ( Figure S3D , lanes 6 and 8) could not promote (+)-strand synthesis, suggesting that the ATPase activity of Ded1p is required for the above stimulatory effect on TBSV RNA synthesis. Also, two active ATPase mutants, D3 and D11 ( Figure S3C ), did facilitate (+)-strand synthesis ( Figure S3D , lanes 5 and 9), confirming that the ATPase activity of Ded1p is required during TBSV (+)-strand RNA synthesis. Interestingly, Ded1p mutant D11, albeit has increased ATPase activity, has wt-like strand displacement activity in vitro [41] . We found that D11 behaved similarly in TBSV replicase assay ( Figure S3D ) to wt Ded1p, suggesting that the unwinding activity of Ded1p is important during TBSV replication. To test if Ded1p can also stimulate the activity of the tombusvirus replicase on (+)-stranded RNA templates, we used DI-72(+) RNA in the purified tombusvirus replicase assay ( Figure S4A ). The TBSV (+)RNA is known to carry a replication silencer element (RSE) at the 39 end that inhibits (2)RNA synthesis in vitro [45] . Addition of Ded1p to the tombusvirus replicase assay did not enhance (2)RNA synthesis ( Figure S4B , lane 3 versus 1-2). Also, D1 mutant deficient in ATPase activity had not much effect on (2)RNA synthesis ( Figure S4B , lane 4 versus 1-2). Based on these data, we suggest that Ded1p does not affect (2)RNA synthesis by the tombusvirus replicase in vitro. To identify the region(s) of the TBSV RNA bound by the recombinant Ded1p, we performed electrophoresis mobility shift assay (EMSA) with purified components. Comparison of 32 Plabeled TBSV (+) and (2)RNAs in binding to purified recombinant Ded1p revealed that (2)RNA bound more readily to Ded1p in vitro than (+)RNA ( Figure 5B , lanes 2-6 versus 8-12). Additional EMSA experiments using the four regions in DI-72 (2)repRNA ( Figure 5A ) as unlabeled competitors revealed that RI(2) was the most efficient in outcompeting the 32 P-labeled TBSV (2)repRNA in binding to Ded1p ( Figure 5C , lanes [8] [9] . This is important since RI(2) is the 39 end of (2)repRNA and contains important cisacting elements, such as the promoter and a short enhancer sequence for (+)RNA synthesis [46, 47] . Further testing of Ded1p binding to (+)repRNA regions revealed that RIV(+), representing the 39 noncoding region in TBSV RNA, was bound more efficiently by Ded1p ( Figure S5 , Step-wise cell-free TBSV replication assay does not support a role for Ded1p helicase in the assembly of the TBSV VRC. (A) Scheme of the CFE-based TBSV replicase assembly and replication assays. Purified recombinant p33 and p92 pol replication proteins of TBSV and in vitro transcribed TBSV DI-72 (+)repRNA were added to the whole cell extract prepared from Ded1p-depleted yeast strain in step 1. The assay either contained or lacked the purified recombinant Ded1p (0.3 mg) or MBP during step 1. Note that the assay was performed in the presence of ATP/GTP to facilitate TBSV VRC assembly, but prevent RNA synthesis in step 1. After step 1, centrifugation was used to collect the membrane fraction of the CFE, and after washing the membranes, step 2 was performed in the presence of ATP/CTP/GTP and 32 P-UTP to allow TBSV RNA replication. In the samples presented in right panel, Ded1p or MBP were added at the beginning of step 2. (B) Denaturing PAGE analysis of the 32 P-labeled TBSV repRNA products obtained in the CFE assays in the presence of wt and various Ded1p mutants (0.3 mg each) ( Figure S3C ) or MBP. See further details in Figure 2 . Note that the various preps of D3 mutant of Ded1p showed high variation in in vitro activity (for reasons currently unknown)-see Figure 4B . Each experiment was repeated at least three times and the data were used to calculate standard deviation. To test if Ded1p can interact with the TBSV p33 replication protein, we used the membrane-based split-ubiquitin assay [31] . We found that Ded1p interacted with p33 protein in yeast ( Figure S6A ). The pull-down experiments with MBP-p33 and MBP-p92 also showed that Ded1p bound to the tombusvirus replication proteins in vitro ( Figure S6B ). This was further supported by the reverse pull-down experiments with GST-tagged Ded1p, which resulted in the co-purification of p33 and p92 ( Figure S6C ). To test if a DEAD-box helicase from plants might have similar stimulatory function on the activity of tombusvirus replicase, we have cloned and purified RH20 cytosolic DEAD-box helicase from Arabidosis thaliana, which shows high degree of similarity to yeast Ded1p ( Figure S7 ). Addition of the recombinant AtRH20 to the CFE prepared from Ded1p-depleted yeast strain increased TBSV repRNA replication by ,2-fold ( Figure 6A, lane 3) . Also, adding the recombinant RH20 to the replicase assay led to almost Yeast with depleted Ded1p co-expressing p33 and p92 pol replication proteins and DI-72 (+)repRNA were used to affinitypurify the RNA-free tombusvirus replicase. The in vitro assays were programmed with DI-72 (2)repRNA, and they also contained purified recombinant Ded1p, mutants or MBP in addition to ATP/CTP/GTP and 32 P-UTP. (B) Representative denaturing gel of 32 P-labeled RNA products synthesized by the purified tombusvirus replicase in vitro in the presence of 0.5 mg or 1.0 mg of purified recombinant Ded1p or its mutants is shown. The level of complementary RNA synthesis producing ''repRNA'' (marked as ''FL'', the full-length product, made via de novo initiation from the 39-terminal promoter) in each sample was compared to that of the replicase activity obtained in the absence of added recombinant protein (lane 1). Note that this replicase preparation also synthesizes de novo internal initiation products (''ii'') and 39-terminal extension products (''39TEX''). Each experiment was repeated three times. (C) Representative denaturing gel of 32 P-labeled RNA products synthesized by the purified FHV replicase in vitro in the presence of 1.0 mg of purified recombinant Ded1p or its mutants is shown. Note that the FHV replicase can use TBSV (2)repRNA as a template in vitro. Each experiment was repeated three times. (D) Representative denaturing gel of 32 P-labeled RNA products synthesized by the purified FHV replicase in vitro in the presence of 1.0 mg of purified recombinant Ded1p or its mutants is shown. Note that the FHV DI-634 (2)repRNA was used as a template in the FHV replicase assay. doi:10.1371/journal.ppat.1002537.g004 3-fold increase of the activity of the purified tombusvirus replicase on TBSV (2)RNA template ( Figure 6B , lanes 2-3 versus 1). To test if AtRH20 can interact with the TBSV p33 replication protein, we used the membrane-based split-ubiquitin assay [31] . We found that AtRH20, similar to Ded1p, interacted with p33 protein in yeast ( Figure S6A ). Altogether, these data strongly suggest that plants also have DEAD-box helicases that could play similar role to the yeast Ded1p DEAD-box helicase in tombusvirus (+)RNA synthesis. To test if Flock house virus (FHV) replicase is affected by Ded1p, first we have developed an in vitro FHV replication assay based on affinity-purified FHV replicase preparation that can be programmed with exogenously added RNAs ( Figure 4C , lanes 1 versus 10). The in vitro FHV replicase assay revealed that the purified recombinant Ded1p increased (+)-strand RNA synthesis on the (2)-stranded FHV template by ,3-fold ( Figure 4D Figure S4D , lane 2 versus 1). It is likely that the observed stimulation of FHV replicase activity by Ded1p is direct, since we found that Ded1p bound to the FHV repRNA ( Figure S8A ) and protein A RdRp protein in vitro ( Figure S8B ). Altogether, these data show that, similar to the tombusvirus replicase, the activity of FHV replicase is stimulated by Ded1p only on the (2)RNA templates, but not when using the (+)RNA templates. Ded1p facilitates initiation by the tombusvirus replicase on RNA/DNA duplex To gain further insights into the mechanism of Ded1p-driven stimulation of TBSV and FHV RNA synthesis, we exploited template structures, such as an RNA/DNA duplex, that are known to hinder RdRp-driven RNA synthesis [48, 49] . Since Ded1p is an RNA helicase [39, 41] and the ATPase/helicase function of Ded1p is needed for stimulation of (+)RNA synthesis by the tombusvirus and FHV replicases, we wanted to examine if Ded1p might facilitate RNA synthesis on a partial DNA/RNA duplex. We chose partial duplex for this assay, since Ded1p and other DEAD-box helicases are not processive enzymes and can only unwind short duplexes [34] . Also, we have shown previously that short DNA oligos hybridized to the promoter region of (2)RNA can inhibit (+)RNA synthesis by the tombusvirus replicase in vitro [42, 47, 48] . Interestingly, addition of purified Ded1p to the tombusvirus replicase assay containing the short RNA/DNA duplex ( Figure 7A Non-overlapping functions of Ded1p and GAPDH in promoting initiation by the tombusvirus replicase GAPDH (Tdh2p in yeast) RNA binding protein is also a host factor stimulating (+)RNA synthesis by the tombusvirus replicase [25, 50] . To test if Ded1p and GAPDH could play a complementary role during (+)RNA synthesis, we added the purified recombinant Ded1p and Tdh2p to the in vitro tombusvirus replicase assay based on the purified preparation ( Figure 8A ). While Ded1p mostly stimulated de novo (+)RNA synthesis initiated from the 39 end of the (2)repRNA up to ,3-fold ( Figure 8B , lane 6), Tdh2p enhanced both de novo (+)RNA synthesis and 39TEX by ,2-and ,3-fold, respectively ( Figure 8B, lane 7) . Interestingly, the two host proteins together had the largest (4.5-fold) effect on (+)RNA synthesis, while their effect on 39TEX was only ,2-fold. Based on these data, we suggest that Ded1p and Tdh2p have a We also performed similar experiments with a short partial RNA/DNA duplex as a template. Both Ded1p and Tdh2p alone could enhance (+)RNA synthesis on the short RNA/DNA duplex ( Figure 8C , lanes 2-3 versus 1) but the largest stimulatory effect (i.e., close to 5-fold increase) was seen when both host factors were included in the assay ( Figure 8C, lane 4) . Altogether, these data further support that Ded1p and GAPDH play synergistic roles in (+)RNA synthesis by the tombusvirus replicase. [9] . Since the viral RNA plays multiple roles during infection, it is likely that remodeling of the viral RNAs and RNP complexes during the switch from one step to another requires RNA helicases or RNA chaperones. Accordingly, the larger RNA viruses all code for RNA helicases [51, 52] . However, RNA viruses with shorter than 6 kB genomes usually do not code for RNA helicases. They could still use RNA helicases during infections if they can subvert selected host helicases for viral purposes. Indeed, we show here that Ded1p helicase is recruited for TBSV replication to aid (+)strand RNA synthesis. It is not yet known if Ded1p is the only host helicase needed for TBSV replication, since genome-wide screens and global proteomics approaches with TBSV have identified additional host helicases as well [9, 10, 28] . However, Ded1p helicase seems to be an ideal host factor to be recruited for viral replication because it is involved in mRNA and viral RNA translation, thus it is colocalized with the viral RNA prior to replication. Also, by recruiting Ded1p, viruses could affect RNA degradation (P-body formation; [53] ) and initiation of translation of new RNAs, likely affecting subsequent host translation (including virus-induced mRNAs coding for anti-viral proteins or required for anti-viral signaling). The essential nature of Ded1p for host mRNA translation, however, makes characterization of Ded1p function as a host factor difficult. Therefore, the combined use of in vivo and in vitro approaches might be necessary to dissect the function of Ded1p in RNA virus replication as shown in this paper. By using a Ded1pdepleted yeast CFE in combination with recombinant Ded1p allowed us to define that the ATPase (helicase) activity of Ded1p is required for efficient TBSV (+)-strand synthesis (see below). Ded1p also increased the RdRp activity of the FHV replicase, suggesting that recruitment of DEAD-box helicases might also be useful for additional small RNA viruses (see below). Co-purification experiments revealed that Ded1p is present in the tombusviral VRC (Figure 1 ). In addition, Ded1p affected (+)strand synthesis, but not the assembly of the VRC in vitro (Figures 2-4) , indicating that Ded1p is likely present in the VRC. Surprisingly, Ded1p, unlike other previously tested host factors Hsp70, GAPDH, or eEF1A [21] [22] [23] 27, 54] , was able to affect TBSV repRNA replication even after the replicase assembly step took place in vitro (Figure 3 ). This suggests that Ded1p can enter the membrane-bound VRC, possibly due to its interaction with p33 ( Figure S6 ) and the helicase activity of Ded1p might lead to some remodeling of VRC. It is not yet known if additional members of the large helicase family could perform similar function to Ded1p during TBSV replication. Our in vitro data with AtRH20 plant helicase protein, which is very similar to Ded1p ( Figure S7 ), suggests that this protein can likely perform similar function in plant infections. For example, AtRH20 has been shown to boost TBSV (+)RNA synthesis in vitro ( Figure 6 ) and bind to p33 replication protein ( Figure S6A ), which can facilitate its recruitment into the VRC. Since more than 50 RNA helicases are present in plants, further experiments will be needed to define if additional host helicases might also be involved in TBSV replication. The in vitro data, based on the CFE assay containing the membrane-bound VRC as well as the solubilized/purified tombusvirus replicase, showed the Ded1p can mainly stimulate TBSV (+)-strand synthesis, while its effect on (2)RNA synthesis is less pronounced. The ATPase activity of Ded1p is required for this stimulatory effect, suggesting that the helicase function of Ded1p is likely important for unwinding the secondary structure of (2)RNA template with the purified tombusvirus replicase or destabilizing the replication intermediate with the membrane-bound VRC, which might contain dsRNA structure. This function of Ded1p can explain why yeast with down-regulated Ded1p level produced small amount of (+)-stranded RNA progeny, albeit it has also shown that depleted Ded1p reduced the level of p33/p92 in yeast [20] . However, the decreased p33/p92 levels are expected to reduce both (+) and (2)RNA levels [55] ( Figure S1 ). Since the recombinant Ded1p enhanced (+)-strand synthesis by the purified recombinant tombusvirus replicase, we propose that Ded1p directly affect TBSV RNA synthesis via affecting the structure of the RNA templates. However, we cannot exclude that Ded1p could also affect the activity of the VRC due to its interaction with p33, albeit the assembly of the VRC was not affected by Ded1p in the CFE-based assay ( Figure 3) . Overall, the recruitment of a host DEAD-box helicase for replication of a small RNA virus is remarkable, since RNA viruses with less than 6 kB genomes are usually do not code for their own helicases [51] . These viruses are thought to replicate without needing a helicase by using RNA chaperones or possibly recruiting host helicases. The emerging picture with TBSV is that this virus utilizes both the viral-coded p33 RNA chaperone [48] and the host Ded1p helicase for replication by promoting (+)-strand synthesis. Both p33 and Ded1p have been shown to open up short DNA/RNA duplexes, although the activity of Ded1p was more robust than that of p33 in vitro [48] . Why would TBSV utilize both an RNA chaperone and an RNA helicase? It is possible that both proteins are needed for robust (+)RNA synthesis to make excess amount of progeny RNA. Also, Ded1p helicase could be involved in remodeling the viral RNA bound by the viral RdRp or host proteins prior or during RNA synthesis. Since Ded1p can work in both 59-to-39 and 39-to-59 directions [56] , it could be used for multiple purposes during RNA synthesis. Based on the available data, it seems that TBSV (+)RNA synthesis is not only affected by the p33/p92 replication proteins, but by GAPDH and Ded1p host proteins as well. Since the viral replication proteins were shown to bind to TBSV (2)RNA nonspecifically [57] , we propose that the above host proteins, which bind strongly to the TBSV (2)repRNA, are involved in facilitating the proper and efficient recruitment of the p92 RdRp protein to the (2)RNA template (or alternatively to the dsRNA intermediate) within the VRC as shown in Figure 9 . For example, Ded1p might unwind either local secondary structure in (2)repRNA or dsRNA region and that, in turn, could favor binding of GAPDH or p92 to the (2)RNA template. This is followed by binding of GAPDH to an AU-rich internal site and proper positioning of the p92 RdRp (bound to GAPDH, [50] ) over the (+)-strand initiation promoter, leading to (+)RNA synthesis. The synergistic effect of these host proteins could promote efficient recycling of the viral RdRp resulting in multiple rounds of (+)RNA synthesis ( Figure 9 ). Thus, the roles of these host proteins are to serve as ''matchmakers'' between the viral RNA template and the viral RdRp. It is intriguing that two host proteins, eEF1A and eEF1Bc, are proposed to serve somewhat similar functions during (2)RNA initiation for TBSV [27, 29] . Therefore, the emerging picture is that TBSV utilizes different host proteins for promoting (2) versus (+)RNA synthesis. This strategy could be beneficial for the virus by allowing asymmetric RNA synthesis, thus leading to excess amount of progeny (+)RNA. Overall, host DEAD-box or related RNA helicases have been shown to affect many different aspects of virus infections, including translation of viral proteins [58] [59] [60] ; viral RNA replication [61] [62] [63] [64] ; reverse transcription [65] ; the activity of anti-viral proteins [66, 67] , and virus-mediated regulation of host gene transcription [68] . Interestingly, Ded1p is known to affect minus-strand synthesis during replication of the L-A dsRNA virus of yeast [69] , suggesting that the use of DEAD-box helicases is wide-spread among RNA viruses. Saccharomyces cerevisiae strain BY4741 (MATa his3D1 leu2D0 met15D0 ura3D0) and TET::DED1 yeast strain (yTHC library, MATa his3D1 leu2D0 met15D0 URA3::CMV-tTA) was obtained from Open Biosystems (Huntsville, AL, USA). The plasmid pESC-HIS-Gal-His33/Gal-DI-72 expressing Cucumber necrosis virus (CNV) 66His-tagged p33 and the TBSV DI-72 repRNA was described earlier [30] . The CNV p33 protein has very high sequence identity with the closely related TBSV p33, but the CNV p33 is expressed better and more active in yeast than TBSV p33. Recombinant yeast Tdh2p protein was produced in E. coli as GST fusion using plasmid pGEX-TDH2, described earlier [50] . Recombinant Ded1p helicase proteins were produced in E. coli as maltose binding protein (MBP) fusions [48] . The expression plasmid pMal-DED1 was prepared by PCR using primers #3956 (CCAGCTG-CAGTCACCACCAAGAAGAGTTG)/ #3957(CCAGGAATT-TGGCTGAACTGAGCGAACAAG) and the yeast genomic DNA as a template. The plasmids pMal-Dx, expressing different Ded1p mutants, were prepared by PCR using #3956/ #3957 primers and plasmids containing mutated DED1 sequences [41, 53] . The PCR products obtained from DED1 wt and mutant sequences were digested with EcoRI and PstI and inserted between EcoRI and PstI sites in pMalc-26 (New England Biolab). To express recombinant FHV protein A in yeast, we generated plasmid pGAD/Cup/FHV/protein A/C-term/HA/FLAG. The following sequence was fused downstream to the full length FHV protein A coding region by repeated PCR and cloning steps: The PCR product was amplified with oligos #3629 (CCGAT-CATGACTCTAAAAGTTATTCTTGGAG) and #3716 (GG-AGCTCGAGTTACTTATCGTCATCGTC) followed by digestion with PagI and XhoI. It was cloned into NcoI and XhoI digested pCupHis92 [70] . Expression and purification of the recombinant MBP-tagged host proteins and the MBP-tagged TBSV p33 and p92 replication proteins and Ded1p helicase from E. coli were carried out as described earlier with modifications [71] . Briefly, the expression plasmids were transformed separately into E. coli strain BL21(DE3) CodonPlus. Protein expression was induced using isopropyl b-Dthiogalactopyranoside (IPTG) for 8 h at 23uC in the case of host proteins and at 16uC in the case of p33 and p92, then the cells were collected by centrifugation (5,000 rpm for 5 min). The cells were suspended and sonicated in MBP column buffer containing 30 mM HEPES-KOH pH 7.4, 25 mM NaCl, 1 mM EDTA, 10 mM b-mercaptoethanol. The extract was then centrifuged at 16,000 g for 10 min, followed by incubation with amylose resin (NEB) for 15 min at 4uC. After washing the resin 2 times with the column buffer, the proteins were eluted with column buffer containing 0.18% (V/W) maltose. Purification of GST-tagged TDH2 (pGEX-TDH2) [50] was carried out using glutathione resin and eluted with 10 mM glutathione, 10 mM ß-mercaptoethanol in the column buffer following the same protocol as MBPproteins. Eluted proteins were aliquoted for storage at 280uC. Protein fractions used for the replication assays were at least 95% pure, as determined by SDS-PAGE (not shown). We have previously shown that our Ded1p preparation has helicase activity on short RNA/DNA duplexes in vitro [48] . The 32 P-labeled or unlabeled full-length DI-72 (+) and (2)RNAs and the four separate regions (RI-IV), were generated as described [57] . Transcripts for replicase or CFE replication assays were purified as described earlier [57] . The amounts of transcripts were quantified by UV spectrophotometer (Beckman). To obtain fulllength FHV-derived DI-634 RNA [72] , we used primers #3842 (GTAATACGACTCACTATAGTAAACAATTCCAAGTTCC-AAAATGG) and #3509 (ACCTTAGTCTGTTGACTTAAA-CTGG) or #3519 (GTAAACAATTCCAAGTTCC) and #3527 (GTAATACGACTCACTATAGGGAACCTTAGTCTGTTG-ACTTAAAC) for (+) or (2)DI-634 RNAs, respectively, and pDI634 [72] as a template in PCR reactions. The RNA transcripts were synthesized on the PCR templates using T7based transcription [47] . In vitro TBSV replication assay in cell-free yeast extract CFEs from BY4741 or TET::DED1 strains capable of supporting TBSV replication in vitro were prepared as described earlier [23, 40] . Briefly, the in vitro TBSV replication assays were performed in 20-ml total volume containing 2 ml of CFE, 0. [23, 40] . Host proteins were added in different amounts as indicated in the Figure legends. The reaction was performed as described [23, 40] . The newly synthesized 32 P-labeled RNA products were separated by electrophoresis in a 5% polyacrylamide gel (PAGE) containing 0.56 Tris-borate-EDTA (TBE) buffer with 8 M urea. To detect the double-stranded RNA (dsRNA) in the cell-free replication assay, the 32 P-labeled RNA samples were divided into two aliquotes: one half was loaded onto the gel without heat treatment in the presence of 25% formamide, while the other half was heat denatured at 85uC for 5 min in the presence of 50% formamide [27] . Fractionation of the whole cell extract was done according to [23, 40] . The total extract was centrifuged at 21,0006 g at 4uC for 10 min to separate the ''soluble'' (supernatant) and ''membrane'' (pellet) fraction. The pellet was re-suspended and washed with buffer A (30 mM HEPES-KOH pH 7.4, 150 mM potassium acetate, and 5 mM magnesium acetate) followed by centrifugation at 21,0006 g at 4uC for 10 min and re-suspension of the pellet in buffer A. In vitro TBSV replication in the fractions was performed as described [23, 40] . Yeast strains (BY4741 and TET::DED1) were transformed with plasmids pGBK-HisFlagp33 (or pGBKHisp33 as a control), pGAD-HisFlagp92 (or pGAD-Hisp92 as a control) and pYC-DI72. The 66His-Flag double-tagged HF-p33 and p92 were expressed from the ADH1 promoter and DI-72 repRNA was under the Gal1 promoter. Transformed yeast were pre-grown on SC-ULH 2 media containing 2% glucose at 29uC. After centrifugation at 2,000 rpm for 3 min and washing pellet with selective media containing 2% galactose and 1 mg/ml Doxycycline, yeast were grown for 24 hours in SC-ULH 2 media containing 2% galactose at 23uC. The replicase purification was done according to a previously described procedure [24] with the following modification. Briefly, 200 mg of yeast cells were resuspended and homogenized in TG buffer [50 mM Tris-HCl [pH 7.5], 10% glycerol, 15 mM MgCl 2 , 10 mM KCl, 0.5 M NaCl, , and 1% [V/V] yeast protease inhibitor cocktail (Ypic)] by glass beads using FastPrep Homogenizer (MP Biomedicals). The membrane fraction containing the viral replicase complex was solubilized with 1 ml TG buffer containing 1% Triton X-100, 1% [V/V] Ypic as described [14, 15] . After affinity purification of HF-p33 on anti-FLAG M2-agarose affinity resin (Sigma), the resinbound replicase complex was eluted in 100 ml elution buffer , 10% glycerol, 15 mM MgCl 2 , 10 mM KCl, 50 mM NaCl, 1% Triton X-100, and 0.15 mg/ml Flag peptide (sigma)]. In vitro RdRp activity assay was performed by using DI-72(2) or (+) RNA template transcribed in vitro by T7 transcription [14] . To measure the effect of host proteins in the replicase assay with DI72(2) RNA containing short double-stranded region at the 39end, a heat denatured RNA transcript (94uC for 2 min) was annealed with a 21-nt oligodeoxynucleotide (#20) (in 1:10 molar ratio) complementary to the 39 end of DI-72(2)RNA in STE buffer (10 mM TRIS, pH 8.0, 1 mM EDTA, and 100 mM NaCl) and then slowly (in 30 min) cooled them down to 25uC . RNase ONE digestion to remove single-stranded 32 P-labeled RNA was performed at 37uC for 30 min in a 16 RNase ONE buffer containing 0.1 ml of RNase ONE (Promega). To obtain FHV replicase preparation, BY4741 yeast strain was transformed with plasmid pGAD/Cup/FHV/proteinA/C-term/ HA/FLAG and pESC-His-GAL1::FHVRNA1framshift. After selection of transformed yeast on SC-LH 2 plates, yeast were pre-grown overnight in selective media containing 2% glucose at 29uC. After centrifugation at 2,000 rpm for 3 min and washing pellet with selective media containing 2% galactose, yeast were grown 36 hours at 29uC in SC-LH 2 media containing 2% galactose and 50 mM CuSO 4 to induce FHV RNA replication. Affinity-purification was done similarly as for tombusvirus, except for using different buffers: homogenization buffer consisted of The split-ubiquitin assay was based on the Dualmembrane kit3 (Dualsystems). The bait construct, pGAD-BT2-N-His33, expressing the CNV p33 replication protein has been described earlier [26] . The prey constructs were made by PCR amplification of individual genes using gene specific primers: #3957 (CCAGCTG-CAGTCACCACCAAGAAGAGTTG) / #4602 (CCAGCCAT-GGCCACCAAGAAGAGTTG) followed by digestion with EcoR1 and Nco1 for DED1 and #4312(CCAGGGATCCATGAC-TTACGGTGGTAGAG) / #4603 (CCAGCCATGGATAGT-TTGAACGACCTC), #4318 (CCAGGGATCCATGAGTCGC-TACGATAGCCG) / #4604 (CCAGCCATGGGCTCCACCC-TCTTCTGCTC) followed by digestion with BamH1 and Nco1 in the case of RH20 gene, respectively. Digested PCR products were fused to NubG at either the 59-or 39-termini (NubG-x and x-NubG) by cloning into pPRN-N-RE or pPRN-C-RE vectors [26] , respectively, using the same enzymes. Yeast strain NMY51 was co-transformed with pGAD-BT2-N-His33 and pPR-N-RE or one of the prey constructs carrying the cDNA for a given helicase and plated onto Trp 2 /Leu 2 synthetic minimal medium plates. Transformed colonies were picked with a loop, re-suspended in water, and streaked onto TLHA 2 (Trp 2 /Leu 2 /His 2 /Ade 2 ) plates to test for p33-helicase protein interactions as described [26] . Protein co-purification with the viral replicase S. cerevisiae strain DED1::66HA-hphNT1 was generated by homologous recombination using strain BY4741. PCR was performed using plasmid pYM-16 (EUROSCARF) [73] as template and primers #2493 (GCAGAAAACGAAGAATCCT-CACCCTAGTTTGTCTGAAATCAATCGATGAATTCGAG-CTCG) / #2494 (GGCTGGGGTAACAGCGGTGGTTCAAA-CAACTCTTCTTGGTGGCGTACGCTGCAGGTCGAC). The PCR products were transformed to BY4741 and recombinant yeast colonies were selected in YPD plates supplemented with hygromycin. Recombinant yeast strains were transformed with plasmids pGBK-HisFlagp33, pGAD-HisFlagp92 and pYC-DI72 [24] . 66His/Flag-tagged HF-p33 and HF-p92 were expressed from ADH1 promoter and DI-72 transcript was under GAL1 promoter. After selection of transformed yeast on SC-ULH 2 plates, yeast were pre-grown overnight in selective media containing 2% glucose at 29uC. After centrifugation at 2,000 rpm for 3 min and washing the pellet with selective media containing 2% galactose, yeast were grown for 36 hours in SC-ULH 2 media containing 2% galactose at 23uC. 200 ml of pelleted yeast were used to affinity-purify HF-p33 and HF-p92 with anti-FLAG M2 agarose as described previously (see also Text S1) [14, 15] . HF-p33 and HF-p92 were detected with anti-His 6 antibody (1/5,000 dilution) and AP-conjugated anti-mouse antibody (1/5,000). DED1-66HA protein was detected with anti-HA antibody from rabbit (Bethyl; 1/10,000 dilution) and AP-conjugated anti-rabbit (1/10,000) followed by NBT-BCIP detection. Figure S1 Reduced TBSV replication in CFE prepared from Ded1-depleted yeast. CFEs were prepared from TET::DED1 yeast strain cultured in the absence of doxycycline (i.e., Ded1p is expressed) or in its presence (10 mg/ml) (i.e., Ded1p is depleted). The CFEs were programmed with recombinant p33/p92 and DI-72(+) repRNA as described in REF 21. Note that the dsRNA product represents the annealed (2)RNA and the (+)RNA, while the ssRNA products represents the newly made (+)RNA products. The CFEs contained comparable amounts of host proteins (not shown). Figure S3 Ded1p mutants with ATPase activity promote plusstrand synthesis by the affinity-purified tombusvirus replicase. (A) Scheme of the tombusvirus replicase assay. Yeast with depleted Ded1p co-expressing p33 and p92 pol replication proteins and DI-72 (+)repRNA were used to affinity-purify the RNA-free tombusvirus replicase. The in vitro assays were programmed with DI-72 (2)repRNA, and they also contained purified recombinant Ded1p. (B) Representative denaturing gel of 32 P-labeled RNA products synthesized by the purified tombusvirus replicase in vitro in the presence of 1.0 mg of purified recombinant Ded1p. The level of complementary RNA synthesis producing ''repRNA'' (marked as ''FL'', the full-length product, made via initiation from the 39-terminal promoter) in each sample was compared to that of the replicase activity obtained in the absence of added recombinant protein. Note that this replicase preparation also synthesizes internal initiation products (''ii'') and 39-terminal extension products (''39TEX''). RNAse One digestion was used to confirm the replicase products: the de novo products are insensitive, while the 39TEX product changes migration after RNase treatment

Transmission of Infectious Diseases En Route to Habitat Hotspots

BACKGROUND: The spread of infectious diseases in wildlife populations is influenced by patterns of between-host contacts. Habitat “hotspots” - places attracting a large numbers of individuals or social groups - can significantly alter contact patterns and, hence, disease propagation. Research on the importance of habitat hotspots in wildlife epidemiology has primarily focused on how inter-individual contacts occurring at the hotspot itself increase disease transmission. However, in territorial animals, epidemiologically important contacts may primarily occur as animals cross through territories of conspecifics en route to habitat hotspots. So far, the phenomenon has received little attention. Here, we investigate the importance of these contacts in the case where infectious individuals keep visiting the hotspots and in the case where these individuals are not able to travel to the hotspot any more. METHODOLOGY AND PRINCIPAL FINDINGS: We developed a simulation epidemiological model to investigate both cases in a scenario when transmission at the hotspot does not occur. We find that (i) hotspots still exacerbate epidemics, (ii) when infectious individuals do not travel to the hotspot, the most vulnerable individuals are those residing at intermediate distances from the hotspot rather than nearby, and (iii) the epidemiological vulnerability of a population is the highest when the number of hotspots is intermediate. CONCLUSIONS AND SIGNIFICANCE: By altering animal movements in their vicinity, habitat hotspots can thus strongly increase the spread of infectious diseases, even when disease transmission does not occur at the hotspot itself. Interestingly, when animals only visit the nearest hotspot, creating additional artificial hotspots, rather than reducing their number, may be an efficient disease control measure.

The spread of infectious diseases strongly depends on how habitat characteristics shape patterns of between-host interactions [1, 2] . In particular, habitat heterogeneity influences patterns of between-individual contacts and hence, disease dynamics [1, 3] . For example, ''habitat hotspots'', sites that attract individuals or social groups over long distances, can be visited by a large subset of a population. Around hotspots, between-individual contact rates often increase in frequency, which amplifies disease transmission. In humans, schools and working places are typical examples of hotspots and have been shown to accelerate the spread of measles, influenza and SARS [4, 5, 6] . Thus, limiting transmission at hotspots has become a promising strategy for mitigating epidemics (e.g., influenza [7] ) although the efficiency of such strategies also depends on the role hotspots plays relative to other sources of local transmission (e.g., influenza [6, 7] ) In wild animal populations, high quality feeding spots (e.g., fruit trees), breeding sites, waterholes or sleeping sites can exacerbate direct physical contacts. Empirical and theoretical studies on the epidemiological importance of habitat hotspots have mainly focused on how the spatial aggregation of animals favors disease transmission at the hotspot itself [8, 9] . For example, the aggregation of wild boar at watering sites significantly increases the transmission of tuberculosis-like lesions [8] . However, interindividual contacts may not always significantly increase at the hotspot itself. This is for example the case of habitat hotspots that some animal species only visited occasionally, such as some mineral licks [10, 11] . Also, animals present at the same time at a particularly large hotspot may not be close enough to each other to transmit infectious diseases. This is the case of large forest clearings [12, 13] or large waterholes. Finally, species such as primates and ungulates might avoid defecating in hotspots of high food resources, limiting the transmission of fecal-oral parasites at hotspots [14, 15] . When disease transmission does not occur at the hotspot, it can still occur at a certain distance from the hotspot. This phenomenon has received little attention so far. Specifically, infective contacts may be observed when infectious individuals travel to the hotspot and cross the territory of susceptible individuals and, reversely, when susceptible individuals cross the territory of infectious individuals. This second type of transmission may be prominent when the disease reduces the mobility of sick individuals (i.e., sickness behavior [16, 17, 18] ). For example, in humans, sick individuals often stay home, which alters disease dynamics [19, 20] . Sick wild animals also commonly reduce their rate of search for food or water [21] . Such transmission may particularly apply to parasites that can survive in the environment (e.g., gastrointestinal parasites) for which the spatial overlap of the home ranges of sympatric hosts favors transmission [22] . To investigate these transmission mechanisms, we developed an agent-based model exploring patterns of disease spread in a large closed population composed of territorial social groups, in which one or more hotspots influence group movement patterns, but where direct disease transmission at the hotspot itself is negligible. Our hypothesis is that terrestrial animals necessarily cross conspecifics' home ranges on their way to a hotspot, which modifies the contact network of the population and may subsequently alter disease transmission. We assumed that between-group disease transmission can occur both between groups having neighbouring territories and between groups travelling to a hotspot and groups whose territories are crossed en route. We also assumed that only groups which territory lies within a certain distance from the hotspot (further referred as ''radius of attraction'') can visit it, and that their visitation rate decreases as this distance increases. The relationship between the radius of attraction and the disease dynamics was then investigated under two scenarios: i) when groups including sick individuals do not travel to the hotspot, and ii) when these groups still travel to the hotspot. The first scenario corresponds to the case of virulent parasites that can strongly decrease the mobility of infected individuals, such as Ebola virus in western lowland gorillas [23] , whereas the second scenario applies to pathogens that do not strongly modify the behavior of their host, such as some gastro-intestinal macroparasites and bacteria [24] . Under both scenarios, we investigated the relationship between the disease attack rate and the hotspot radius of attraction, identified the groups in the population that have the highest risk of infection and explored the relationship between the number of hotspots and the magnitude of an epidemic. The model has a 51651 lattice structure, where each cell of the lattice corresponds to a group's territory. We assumed disease transmission can occur between each group and its eight neighbours (Fig. 1) . We use N i to denote the list of indices of the eight neighbours of group i. Initially, a single habitat hotspot is placed at the center of the lattice. All groups are assumed to include ten individuals. At each daily time step, each group either visits the hotspot or stays in its territory. The probability P visit i ð Þ of a visit by group i is a decreasing function of the Euclidean distance, d i , between the group's territory and the hotspot. We assume that all groups gain the same benefit from visiting the hotspot and that the travel cost is proportional to d i , leading to: where P max is the probability of a visit for the eight groups directly neighbouring the hotspot, and R is the hotspot radius of attraction. Groups occupying cells that are farther from the hotspot than R never visit it. When a group visits the hotspot, it follows a Biased Random Walk from its home cell to the hotspot (BRW [25] ) and returns to its home cell on the same day. The length of each step of the BRW (denoted S) is held constant and the direction of the step is consistently biased towards the hotspot during the approach to the hotspot and towards the group's home cell during the return from the hotspot. Each turning angle is randomly drawn from a normal distribution N (0, s 2 ), where s is a standard deviation parameter. The list of groups residing in cells encountered along each BRW to the hotspot is recorded. Groups travel to the hotspot and come back within a single time step. At each time step, each group interacts with (i) groups occupying neighbouring cells (neighbour-neighbour contact), and (ii) if the group travels to the hotspot, all of the groups it encounters along the BRW (traveler-resident contact). We model infectious disease dynamics using a simple stochastic susceptible-infectious-removed (SIR) epidemic model, with oneday time steps. Each individual moves independently through the three states: Susceptible (at risk of contracting the disease), Infectious (capable of transmitting the disease), or Removed (recovered or dead). Susceptible individuals can be infected by infected individuals from either its own group or other groups. The latter can occur during neighbour-neighbour contacts or travelerresident contacts. The local transmission probability P i is defined as the per-timestep probability that a susceptible individual in group i is infected by an infectious individual from its own group or a neighbouring group. Let I i denote the number of infectious individuals in group i. The probability that the focal individual is infected by at least one of the infectious individuals in its own group is 1{ 1{P w ð Þ Ii , where P w is the within-group transmission probability from an infected individual to a susceptible individual of the same group. Likewise, the probability of a susceptible individual being infected by an infectious individual from one of the eight neighbouring groups is 1{ P where P B is the between-group transmission probability from an infected individual to a susceptible individual of a neighbouring group. Combining these two sources of infection, the local transmission probability is given by The per-time-step probability of becoming infected during transit to a hotspot P H1 depends on the number of infectious individuals I c in each group c encountered en route and the P T ''travelling'' probability of transmission during one of these transient contacts between a resident and traveling group. Specifically, during a one-day trip to a hotspot, the probability that a susceptible individual in the group is infected along the way is given by where BRW i denotes the list of indices of the groups encountered by group i as it travels to and from the hotspot. In a second version of the model, infected groups are assumed to travel to the hotspot. Transmission from infected travelers to susceptible residents encountered en route can then occur. In this case, the per day probability that a susceptible individual in group i is infected by a traveler depends on the numbers of infected individuals in each of the groups that travels through the territory of i en route to the hotspot, and is given by where PT i denotes the set of groups passing through i's territory. At the end of each time step, each infected individual is removed with probability c. We assume that no transmission occurs between groups travelling to the hotspot simultaneously. We explored two versions of the transmission model. In the ''Sick-stay'' model, groups that included at least one infected individual -infected groups -were assumed to stop travelling to the hotspot; in the ''Sick-travel'' model, infected groups continue to travel to the hotspot as if uninfected. In the Sick-stay model, disease transmission between travelers and residents can only occur from an infected resident to a susceptible traveler, while in the Sick-travel model, transmission can be bi-directional. We also considered models with multiple hotspots. A specified number of hotspots are randomly placed on the lattice, and groups visit only their nearest hotspot (according to P visit , described above). At the beginning of each simulation, all individuals were susceptible. Epidemics were started with a single infected individual. Unless stated otherwise, the first case was introduced into one of the eight groups adjacent to the hotspot. At each time step, the number of infected and removed individuals (and groups) was recorded until no individual in the population was infected, which indicated the end of the epidemic. For each parameter combination, we ran 1000 simulations. For all simulations, the recovery rate c was set to 0.1, the maximum probability of a visit to the hotspot, P max , was set to 0.1, and the BRW step length S was set to 0.25 (i.e., 1/4 of the distance between the center of neigbouring group's territories). We also assumed that the within-group transmission probability, P w , was at least ten times higher than the between-group transmission probabilities (P B and P T ). The model was implemented in Delphi 7 (Borland Software Corporation, 2002). Table 1 summarizes parameter definition and values, and a sample run of the model is shown in Video S1. The attack rate (proportion of groups becoming infected following a single disease introduction) generally increases with the hotspot radius of attraction R, and the traveler-resident transmission probability P T (Fig. 2 ). This occurs whether or not infected groups are assumed to travel during infection ( Fig. 2 and Fig. S1 in supplemental materials). As predicted by percolation theory [26] , the attack rate also increases with both the withingroup transmission parameter P W and the between-group transmission parameter P B . Interestingly, the highest impact of the hotspot radius of attraction on the attack rate was observed for intermediate values of P W and P B . For low values of P W and P B , inter-group disease transmission primarily occurred between travelling and resident groups and was often not sufficient to sustain an epidemic. For high values of P W and P B , the disease always percolated, even in the absence of the hotspot. For intermediate values of P W and P B , groups infected en route to the hotspot then stochastically triggered small outbreaks around their territories. The epidemiological impact of the hotspot was amplified by this interaction between traveler-resident transmission and neighbour-neighbour transmission. In both models, the attack rate decreased as the distance between the hotspot and the point of disease introduction increased (Fig. 3) . The greater this distance, the lower the probability that a group visiting the hotspot encountered the group initially infected. The Sick-travel model yields higher attack rates than the Sick-stay model, particularly for groups ranging at intermediate distances between the location of the first disease case and the hotspot. In the Sick-travel model, groups ranging closer to the hotspot exhibited higher probabilities of infection ( Fig. 4 and Fig. S2 in supplemental materials), since their territories are crossed by large numbers of infected groups travelling to the hotspot (Fig. S4a) . The relationship is more complex in the Sick-stay model: groups ranging at intermediate distances from the hotspot experience the highest risks of infection ( Fig. 4 and Fig. S3 in supplemental materials). In this case, hotspot-mediated infection occurs only from infected residents to susceptible travelers. Groups ranging close to the hotspot travelled more often, but encountered only a small number of potentially infected groups. Groups ranging far from the hotspot encountered larger numbers of groups when visiting the hotspot, but did so only rarely. Thus groups ranging at intermediate distances experienced the greatest number of potentially infective contacts with resident groups encountered en route to the hotspot (Fig. S4b) . When epidemics occur, the difference in spatial pattern of disease spread observed between the models is insensitive to the parameter values ( Fig. S2 and Fig. S3 ). The fact that groups ranging at an intermediate distance from the hotspot display a higher number of potentially infective contacts can easily be understood using a simple mathematical model. Indeed, the expected number of groups encountered by a group i visiting the hotspot, per time unit, can be assumed to be approximately proportional to P visit and to the territory-hotspot distance d i : This approximation holds as long as d i is large or the turning angle is low. The derivative of this second-order polynomial has a maximum in R=2. Second, we assessed the epidemiological impact of the hotspot on groups that never visit it because they range at a distance larger than R from the hotspot. For both models, when P B is high enough to allow some between-neighbour transmission, the attack rate for these groups was found to be larger than expected under a model with no hotspot (R = 0) (Fig. 4) . Stochastic, local between-group transmission events allow the spread of the disease beyond the radius of attraction. Groups that never visit the hotspot are thus indirectly impacted by the hotspot. Finally, outside of the radius of attraction, disease spreads as expected for a lattice model [27] whereas inside the radius of attraction, disease spread rapidly among the groups, with no apparent spatial structure. For both models, the relationship between the number of hotspots and the attack rate is bell-shaped (Fig. 5) . For a low number of hotspots, adding new hotspots increases the fraction of the population ranging within the radius of attraction of these hotspots and, thereby, increases the overall attack rate. Beyond a certain number of hotspots, however, all groups are already attracted by at least one hotspot on the landscape. Under the assumption that these groups travel exclusively to the nearest hotspot, adding more hotspots decreases the distance travelled by the groups and the number of infective contacts they can have en route, and thereby lowers the attack rate. Spatial features of the landscape such as habitat hotspots can profoundly influence the spread of infectious diseases [28, 29] . Our model extends previous studies focusing on transmission at the hotspot, and reveals that hotspots can also strongly alter disease transmission by generating infective contacts between animals travelling towards or from the hotspot and animals whose territories are traversed. Our results show that even when sick groups stay in their territory, hotspots may increase the size of an epidemic. When infected animals cease to visit the hotspot, groups ranging at intermediate distances to the hotspot are the most vulnerable. We also found that the epidemiological impact of hotspots extends far beyond the subset of the population that visits it; even groups having no contact with those visiting the hotspot display elevated risks of infection. Finally, our model predicts that when groups visit their nearest hotspots, the epidemiological impact of hotspots is most severe when the number of hotspots is intermediate. Hotspots impact disease transmission via a combination of both local between-neighbour and long-range traveler-resident transmissions, which is characteristic of a small-world network [30] . Disease dynamics in our model resemble those in a classic smallworld network in several aspects. First, the attack rate increases with long-distance interactions, determined by the hotspot radius of attraction (Fig. 2) . Second, new foci of infection established by long-distance traveler-resident contacts only spread when the local transmission rate, between neighbours, is sufficiently high. This phenomenon extends the influence of the hotspot beyond the radius of attraction (Fig. 4) . Finally, as in small-world networks [31] , all groups within the hotspot radius of attraction were infected almost at the same time. Thus, habitat hotspots potentially play a significant role in fuelling disease outbreaks, much like other natural mechanisms that generate small-world networks, such as the movement of vectors between plants [32, 33] . We find that hotspots are expected to influence disease dynamics significantly, even when infected groups do not travel to the hotspot at all. However, in this case, the hotspot effect strongly decreases as the distance between disease introduction and the hotspot increases. The reduction of mobility in infected groups also generates an unexpected spatio-temporal pattern: groups ranging at intermediate distance from the hotspot have the highest risk of infection, even if the disease is introduced immediately next to the hotspot. This counterintuitive result highlights the importance of understanding the behavioral effects Figure 2 . Influence of multiple model parameters on attack rate, when infected groups do not travel (Sick-stay model). The fraction of groups infected increases with the hotspot radius of attraction, but varies with the traveler-resident transmission probability P T (four lines in each graph), within-group transmission probability P w (three different columns of graphs), and between-neighbour transmission probability P B (four different rows of graphs). Each value is based on 1000 simulations in which disease was introduced randomly in one of the eigth groups adjacent to the hotspot. doi:10.1371/journal.pone.0031290.g002 of disease in wild animal populations. For example, as in humans, predicting the impact of hotspots on disease dynamics will strongly depend on understanding whether infectious individuals still travel to hotspots because disease symptoms appear after an infectious state (e.g., influenza H1N1 [6, 19] ), or whether infectious individuals do not visit hotspots because disease symptoms appear before the infectious state (e.g., SARS [34, 35] ). Furthermore, our results suggest that when transmission does not directly occur at hotspots, disease control measures targeting groups residing around the hotspot might not necessarily be the most efficient ones. Further simulation work is needed to identify optimal disease control measures. The habitat of wild animal populations often includes more than one hotspot. For example, the habitat of terrestrial mammals can include a small number of high-value hotspots attracting dozens of groups (e.g., salt licks or forest clearings) and more numerous low-value hotspots attracting only a few groups (e.g., fruiting trees). Our model reveals that, when groups are assumed to travel to their nearest hotspot, the impact of disease outbreaks is a bell-shaped function of the number of hotspots (Fig. 5) . This result challenges the hypothesis that the number of hotspots and disease prevalence will correlate positively [8] , and could be used to optimize strategies for controlling disease in wild animal populations. Thus, wildlife managers may consider increasing, rather than decreasing [36] , the number of water holes in order to reduce the number of highly-connected individuals or social groups, and hence the impact of an outbreak. However, additional studies are needed to determine if our result still holds when each animal visits more than one hotspot. The values of the parameters of our model can be estimated from empirical data. The relationship between the distance from a group's territory and the hotspot visitation rate can be estimated using capture-mark recapture and telemetric data, between-group contact rates can be estimated from direct observation or telemetric data, and plausible distributions of disease transmission rates can be found in the literature. The step length of the biased random walk is assumed to have a fixed value (here, 0.25 times the size of a territory). This parameter does not need to be estimated accurately since it is redundant with another parameter, the traveler-resident contact rate, which is allowed to vary. Thus, the model can be applied to a broad range of host-parasite systems, from primate groups travelling to waterholes on a daily basis [37, 38] to large mammals visiting every few weeks mineral-rich Figure 5 . Number of hotspots. Each line graphs the change in attack rate as a function of the number of hotspots, for a different value of P B (from 4e-04 to 16e-04). Results are presented for hotspots ranging from 1-100 (left) and 1-500 (right) in the Sick-stay model (top) and the Sick-travel model (bottom). Each value is averaged over 1000 stochastic simulations assuming R = 30, P w = 0.06, P T = 4e-04. Each hotspot was located randomly in the population, and disease was introduced into the group ranging in the middle of the habitat. doi:10.1371/journal.pone.0031290.g005 areas [12, 13, 39] . In our model, the impact of the hotspot is particularly sensitive to the ratio between the local and the traveler-resident between-group transmissions. When the local between-neighbour transmission is high compared to the travelerresident transmission, the impact of the hotspot is minimal. We considered two discrete transmission scenarios, the Sicktravel and the Sick-stay scenarios. However, intermediate scenarios are also possible. For example, infected groups may fission such that only healthy individuals travel to the hotspot. In this case, we expect that although the overall disease transmission will increase compared to the pure Sick-stay scenario, the spatial pattern of the disease impact will be qualitatively similar to that observed for the Sick-stay model. In this study, we have shown how transmission occurring around habitat hotspots influences disease transmission patterns, while previous studies have focused on disease transmission occurring at the hotspot itself. In some ecological systems, both transmission modes may coexist. For example, some fecal-orally transmitted parasites can infect both the soil and waterholes, and spore-forming bacteria such as Bacillus anthracis can persist for extended periods of time in animal carcasses, water and soil [40] . Additional works are needed to understand such epidemiological systems. Figure S1 Influence of multiple model parameters on attack rate, when infected groups travel (Sick-travel model). The fraction of groups infected increases with the hotspot radius of attraction, but varies with the traveler-resident transmission probability P T (four lines in each graph), withingroup transmission probability P w (three different columns of graphs), and between-neighbour transmission probability P B (four different rows of graphs). Each value is based on 1000 simulations in which disease was introduced randomly in one of the eigth groups adjacent to the hotspot. (TIF) Figure S2 Group's probability of infection in relation to the distance to the hotspot, predicted by the Sick-travel model. The relationship is presented for different values of the hotspot radius of attraction (R). Each graph represents a combination of the between-neighbour (P B ) and the travelerresident (P T ) transmission. The disease was introduced randomly in one of the eight groups adjacent to the hotspot. For all simulations, P w = 0.06. (TIF) Figure S3 Group's probability of infection in relation to the distance to the hotspot, predicted by the Sick-stay model. The relationship is presented for different values of the hotspot radius of attraction (R). Each graph represents a combination of the between-neighbour (P B ) and the travelerresident (P T ) transmission. The disease was introduced randomly in one of the eight groups adjacent to the hotspot. For all simulations, P w = 0.06. (TIF) Figure S4 Traveler-resident contact patterns. Each graph shows the relationship between the distance of a group from the hotspot and (a) the number of other groups that travel through its territory when travelling to and from the hotspot, (b) the number of resident groups it encounters when travelling to and from the hotspot. Values are based on encounters occurring during 100 time steps, in the absence of disease transmission. (TIF) Video S1 Model dynamics. The model shows one simulation run corresponding to the Sick-travel model. White, red and black squares represent susceptible, infected and removed groups, respectively. Blue squares represent groups travelling to the hotspot at each time step. The hotspot, in green, is in the middle of the lattice. The disease is introduced at the periphery. (WMV)

The Dispanins: A Novel Gene Family of Ancient Origin That Contains 14 Human Members

The Interferon induced transmembrane proteins (IFITM) are a family of transmembrane proteins that is known to inhibit cell invasion of viruses such as HIV-1 and influenza. We show that the IFITM genes are a subfamily in a larger family of transmembrane (TM) proteins that we call Dispanins, which refers to a common 2TM structure. We mined the Dispanins in 36 eukaryotic species, covering all major eukaryotic groups, and investigated their evolutionary history using Bayesian and maximum likelihood approaches to infer a phylogenetic tree. We identified ten human genes that together with the known IFITM genes form the Dispanin family. We show that the Dispanins first emerged in eukaryotes in a common ancestor of choanoflagellates and metazoa, and that the family later expanded in vertebrates where it forms four subfamilies (A–D). Interestingly, we also find that the family is found in several different phyla of bacteria and propose that it was horizontally transferred to eukaryotes from bacteria in the common ancestor of choanoflagellates and metazoa. The bacterial and eukaryotic sequences have a considerably conserved protein structure. In conclusion, we introduce a novel family, the Dispanins, together with a nomenclature based on the evolutionary origin.

Membrane proteins are essential for the ability of all cellular organisms to respond and interact with their environment. Therefore they have attained large research interest and are one of the major groups of drug targets [1] . We have previously estimated that 27% of the human genes codes for alpha-helical membrane proteins and provided a comprehensive classification based on their function and evolutionary origin [2] . However, the identification and annotation of many membrane bound protein families is still being revised. We have during recent years worked on the annotation of both G protein-coupled receptors [3, 4] and solute carriers [5] and most of the genes of these large superfamilies now have a clear identity and annotation. There is however still large work to be done to clarify the identity, annotation and the evolutionary history of several families of membrane bound proteins. Establishing a rigid nomenclature based on evolutionary information and structural features of the predicted proteins facilitates prediction of the functional role of these genes that often have only have been studied in large gene or transcription consortia. In previous studies we have found that membrane proteins with few transmembrane (TM) helices are less studied than other. This is particularly true for 2TM proteins where more than 70% of the about 700 proteins remained unclassified. Interestingly, we found evidence for several uncharacterized homologues to a small group of genes known as the Interferon-induced transmembrane proteins (IFITM) family. The IFITMs constitute a group with four human members (IFITM1-3, 5) that are found in a consecutive order on chromosome 11, having two transmembrane (2TM) helices. The IFITM4 gene is not present in human, but is located in proximity to the other four genes in the mouse genome. The IFITM1-3 proteins were identified 25 years ago as being upregulated by interferons (IFN) [6] . Recently they received considerable attentions as IFITM1-3 were found to prevent infection of a growing list of viruses such as HIV-1, SARS influenza A H1N1, West Nile and Dengue fever viruses [7, 8, 9, 10] . Hence, proteins of the IFITM family mediate part of the antiviral response orchestrated by IFNs. However, the IFITM family is also involved in other processes such as oncogenesis, bone mineralization (IFITM5) and germ cell development (IFITM1 and 3) and IFITM5 has not been identified as interferon-inducible [11, 12, 13, 14] . Although the biological roles of the IFITM genes are emerging, no thorough evolutionary analysis has been performed on this group. In this study, we sought to infer the evolutionary history of the human IFITM genes and identify potential homologues. We mined 36 eukaryotic species, covering all major eukaryotic groups, and found that the IFITMs form a subfamily in a larger novel family that has ten human members in addition to the four IFITM genes. We propose Dispanins as a novel name for this family, which refers to their common 2TM structure. Further, we find that the eukaryotic Dispanins first appeared before the radiation of metazoa and that they branch out into four subfamilies (A-D). More surprisingly, we also discover that the Dispanins are found in a large range of bacteria and in brown alga. In total, we collected 87 eukaryotic IFITM homologues from H. Sapiens (14 genes) M. musculus (17 genes), G. gallus (6 genes), X. tropicalis (13 genes), D. rerio (7 genes), P. marinus (1 gene), C. intestinalis (1 gene), B.floridae (12 genes), S. manosoni (1 gene), S. purpuratus (9 genes), N. vectensis (4 genes) and M. brevicollis (1 gene). No IFITM genes could be detected in any of the remaining 24 analyzed proteomes, which covers all other major eukaryotic groups. The search of the nr database with HMMER did not get any hits outside metazoa except bacteria, choanoflagellates and the brown alga Ectocarpus siliculosus (2 genes). Nine genes were deemed as pseudogenes based on annotation and sequence analysis and removed from further analysis. In addition to the four previously identified human IFITM genes, ten novel human homologous genes were detected. These ten genes together with the four IFITM genes form a human gene family that we choose to call Dispanins based of their common 2TM structure. In UniProt we identified 65 annotated IFITM homologues from full bacteria proteome sets spread over seven different phyla (See table S1): Acidobacteria (2 genes), Actinobacteria (43 genes), Cyanobacteria (3 genes), TG1 (1 gene), Bacteroidetes (2 genes), Firmicutes (1 gene) and Proteobacteria (13 genes). Out of these, 46 bacterial sequences from 32 species were included for further analysis. No viral or Archaean genes were annotated as IFITM homologues in UniProt. The phylogeny of the vertebrate Dispanins ( Figure 1 ) allows the division of the Dispanins into four subfamilies A-D that are supported by strong confidence with respect to posterior probabilities (pp) or bootstraps (pp.0.75 and bs.90% for all nodes). We propose a common nomenclature for the Dispanins that are based on their subclass and a number (DSPA1 etc). The proposed names together with previous gene symbols and accession number can be found in Table S1 . The finding of two Dispanin homologs in the brown alga E. siliculosus, which is evolutionary distant to metazoa, and the single Dispanin in the close metazoan relative M.brevicollis are the two only non-metazoan eukaryotic Dispanins. A BLAST search gives that the E. siliculosus proteins have a higher similarity to metazoan family members (best hit E-value,10 210 ) than bacterial Dispanins and M. brevicollis. The M. brevicollis Dispanin share a conserved splice site with all metazoan family members, which suggest that the eukaryotic Dispanins first emerged in a common ancestor of M. brevicollis and the metazoan lineage. Within the metazoan lineage the Dispanins have been lost in at least two separate occasions, i.e. T. adhaerens and in the ecdysozoan lineage (D. melanogaster and C. elegans). The vertebrate Dispanins sort into subfamilies A-D ( Figure 1 ). The DSPA subfamily has six human genes (DSPA1, DSPA2a-d and DSPA3) of which the DSPA2a-c corresponds to IFITM1-3 and DSPA1 to IFITM5. DSPA2d (AC023157) and DSPA3 (AC068580) are two novel identified genes, closely related to the IFITM family. The phylogenetic tree indicates that the DSPA2 genes have undergone an independent duplication in H. Sapiens and M. musculus and these were given a species specific nomenclature, e.g. DSPA2a-d in human. The Dspa4, Dspa5, Dspa6 genes in the phylogenetic tree do not have any clear human orthologs. The DSPB subfamily is only found in tetrapoda and contains three human genes called DSPB1 (TUSC5), DSPB2 (TMEM233) and DSPB3 (PRRT2) whereas Dspb4 is only present in G. gallus. DSPC1 (TMEM90A), DSPC2 (TMEM90B) and DSPC3 (TMEM91) make up the human DSPC subfamily, which is represented in all investigated vertebrates. The DSPD1 (PPRT1) gene is found in all the vertebrate species except the basal organism P. marinus whereas DSPD2 (AL160276) is mammalian specific. Five vertebrate genes and all invertebrate were excluded from the phylogenetic analysis and instead classified into subfamilies by using a BLAST approach (Table S1) . Some genes could not unambiguously be classified into the vertebrate subfamilies: C. intestinalis (1 gene), S. mansoni (1 gene), B. floridae (2 genes), S. purpuratus (3 genes), N. vectensis (1 gene) and M. brevicollis (1 gene). By combining the results of the phylogenetic analysis and BLAST classification, we created a schematic overview of the organisms' gene repertoire and a schematic picture of the Dispanin family's evolutionary history, which suggests that the invertebrate Dispanins share more similarity towards the DSPC and D subfamilies than DSPA and B ( Figure 2 ). All the members of the human DSPA are located on chromosome 11 except for DSPA2d which resides on chromosome 12. The genes of the other subfamilies are not enriched on any chromosome. Several features are common to all the eukaryotic Dispanin proteins ( Figure 3 ). They comprise two transmembrane helices that are predicted between 20 and 30 amino acids in length with the second helices often being slightly longer. The Dispanins are rich in both glycosylation-and phosphorylation sites that predominantly are found on the Nterminus. The N-terminus is often long (.100 amino acids) compared to the C-terminal (,10 amino acids) and both are always oriented towards the outside of the cell. The Dispanins contain several conserved motifs ( Figure 3 and 4), which are found both among eukaryotes and bacteria. The most conserved pattern is the G-D motif and the A-X(6)-A motif, both situated in the intracellular loop between the transmembrane helices that also is frequently rich in positive amino acids (K and R). The first helix is the most conserved with an alanine (A) residue and double cysteine C-C (C-F-C in the DSPB family) motif whereas a glycine (G) residue is the most conserved in the second helix. Another highly conserved motif, though only amongst the eukaryotes, is the single aspartic acid (D) on the N-terminus, flanking the first helix. Analysis of the exon structure in the protein sequences was made for all eukaryotic Dispanins except E. siliculosus where no such information was found. The eukaryotic Dispanins have a conserved splice site in the intracellular loop that separates the two transmembrane helices into different exons ( Figure 4 ). This site is only missing in the S. mansoni Dispanin and the mouse Dspa2f genes. The vertebrate proteins of the DSPA and DSPB subfamilies only have these two exons whereas the whole DSPC subfamily and the DSPD1 proteins have an additional exon that codes for their N-terminus. The DSPD2 proteins that only are found in mammals seem to have lost their N-termini exon. All the classified B. floridae and S. purpuratus sequences has the corresponding splice site in the N-terminus, whereas the N. vectensis and M. brevicollis proteins has 3-6 and 7 exons respectively. We provide evidence that the four IFITM genes together with ten additional human genes, known as TUSC5, TMEM233, PRRT2, TMEM90A, DSPC2, TMEM90B, TMEM91, AC023157, AL160276 and AC068580, form a novel gene family that we call the Dispanins, which refers to the 2TM membrane topology that is common to all identified members. This family is the second largest 2TM family in the human genome, superseded only by the Inwardly rectifying potassium channel family that has 15 members [2] . Except for the 2TM memebrane topology the Dispanins are not homologous or share domains with any other 2TM proteins in the human genome and constitute a distinct gene family. We have discovered that this family is found in metazoan, the choanoflagellate M. brevicollis and the brown alga E. siliculosus, but not in other eukaryotes. Surprisingly it is widely present in bacteria where it is found in several different phyla such as Actinobacteridae, Acidobacteria, Cyanobacteria, Bacteriodetes, Firmicutes and Proteobacteria. The highest number of bacterial Dispanins is detected in Actinobacteria and Proteobacteria, which diverged around three billion years ago [15] . We find that Dispanins in eukaryotes and bacteria have high sequence similarities and share several conserved sequence motifs (Figure 4) , which is strong evidence for a common evolutionary origin and possibly a functional relationship. As the family is found in several bacterial phyla we suggest that it first emerged in bacteria to later be introduced in eukaryotes through a horizontal gene transfer event. However, we were not able to construct a stable phylogenetic tree including bacterial and eukaryotic Dispanins. As the eukaryotic Dispanins only is widespread in metazoa it was unexpected to find the family in the evolutionary distant brown alga E. siliculosus (Figure 2 ). Our sequence analysis supports that all metazoan Dispanins have their origin in the common ancestor of M. brevicollis and metazoa as the choanoflagellate share a conserved splice site in the intracellular loop with nearly all metazoan Dispanins (Figure 4) . However, the finding of the family in E. siliculosus suggests that the family has undergone two horizontal gene transfer events. As the E. siliculosus Dispanins are more similar to metazoan family members than M. brevicollis and bacteria we propose that the first horizontal gene transfer event was from bacteria to a common ancestor of choanoflagellates and metazoa followed by a second transfer between metazoa and brown alga. During the course of metazoan evolution the Dispanins have expanded and diverged into four distinct subfamilies. However, it has also been lost in the basal metazoa T. adhaerens and the ecdysozoan lineage, which show that it is not essential for all metazoan life. Although we were unable to create a stable phylogenetic tree that include both vertebrate and invertebrate sequences BLAST searches suggest that the DSPC and D subfamilies are the oldest of the vertebrate subfamilies as the invertebrate sequences has higher resemblance to these two subfamilies ( Figure 2) . Moreover, the DSPC and D subfamilies forms a separate cluster from DSPA-B (Figure 1) . Hence, the phylogenetic analysis suggests that the DSPA and B subfamilies have their origin close to the radiation of teleost, although DSPB have been lost in D. rerio (Figure 1 and 4) . The DSPC family is found in two to three copies in all vertebrates and is the most widespread family as the BLAST classification suggest that it is present in two invertebrate species and E. siliculosus. In mouse, the family members are expressed predominantly in brain tissues (Dspc1/Tmem90a, Dspc2/Tmem90b) or ubiquitously (Dspc3/ Tmem91). DSPC1 (TMEM90A) has been proposed to have a role in striatial functioning and the pathophysiology of Huntington's disease and is localized to the Golgi apparatus [16] . The DSPD family has been lost at several occasions, both in vertebrates and invertebrates ( Figure 2 ). The mouse Dspd1 (Prrt1) gene is ubiquitously expressed with the highest expression in B-cells according to BioGPS [17] . However, no previous studies have been performed on the DSPD subfamily. The DSPB subfamily is found in tetrapoda and has three members in human and mouse. Mice expression profiles from BioGPS shows that Dspb3 (Prrt2) is exclusively expressed in brain tissues and that Dspb1 (Tusc5) is expressed in dorsal root ganglia and adipose tissues. In agreement with this expression data, Dspb1 (Tusc5) has been suggested to be involved in neural regulation of adipocyte differentiation and is regulated by PPARc [18, 19] . The DSPA/IFITM subfamily is the most numerous and the mouse and human genes are all clustered in a consecutive manner on chromosome six and eleven, respectively. These regions share a conserved synteny (http:// cinteny.cchmc.org/) and are flanked by the ATHL1 and B4GALNT4 genes on each side, which is strong evidence for the genes to have their origin in common evolutionary gene duplications. This is supported by the phylogenetic analysis ( Figure 1 ) for DSPA1 (IFITM5) and DSPA3 (AC068580), which have orthologs in all tetrapoda. However, for the DSPA2 group (DSPA2A-D/IFITM1-3 and AC068580) the phylogeny suggests that M. musculus and H. sapiens have undergone independent expansions of the group. Rather than being created by independent gene duplications in the two species, it is possible that these genes are subject to concerted evolution, where paralogous genes within a species are more conserved towards each other than towards orthologs in other species. This phenomenon is most common in tandemly repeated genes, such as the DSPA2 group, and is believed to primarily be the result of recombination mechanisms [20] . Interestingly, also the Dspa4a-f genes of X. tropicalis seem to have undergone and independent expansion. However, the phylogeny is not strong enough to prove that the Dspa4a-f genes are orthologous to the mammalian DSPA2 genes (Figure 1 ). The DSPA/IFITM subfamily is the most well studied and is a multifunctional family of which its antiviral properties are best understood [8] . The family is expressed in many mouse tissues with the highest expression in mast cells, macrophages and osteoblasts according to BioGPS. We add two novel human members to this subfamily: DSPA2D, which is closely related to DSPA2C (IFITM3) and DSPA3, which forms a distinct cluster (Figure 1 ). Both these genes are poorly characterized. Microarray data from Array Express (www.ebi.ac.uk/ arrayexpress/) shows that DSPA3 is upregulated by interferon after exposure of macrophages to interferon-gamma in a study (E-GEOD-5099) where DSPA2a-c (IFITM1-3) also were induced [21] . The mouse Dspa3 gene is like the other genes of this subfamily situated on chromosome seven, but is 1.5 Mbp away from the DSPA (IFITM) cluster where the other genes reside. Hence, Dspa3 could explain the mild phenotypes and be responsible for the suggested functional redundancy that was found when deleting the whole DSPA (IFITM) loci [22] . The Dispanin family has several conserved motifs across subfamilies that are also detected in bacteria (Figure 4) . One of the most prominent is the double cysteine motif (C-C) in the first transmembrane helix. This motif has recently been shown to undergo post-translational modification through S-palmitoylation in DSPA2C (IFITM3), which increases hydrophobicity [23] . Further, Yount and colleagues shows that the antiviral activity of DSPA2C (IFITM3) is dependent on this modification, which induces clustering of the proteins. As this motif is highly conserved, it is likely that S-palmitoylation is an important regulatory mechanism also among the other subfamilies. Intriguingly, this motif is also found in the bacterial Dispanins even though bacterial proteins do not undergo S-palmitoylation. Hence, the cysteine motif of the may have other means of structural and functional importance apart of from the S-palmitoylation. In this study, we introduce the Dispanin family, of which the IFITM genes constitute a subfamily. In addition to the IFITM genes we identify 10 novel human Dispanins and investigate the family's evolutionary history and suggest that the eukaryotic members are descending from bacteria through a horizontal gene transfer. Thus, the expansion and diversification of Dispanins in vertebrates may reflect the evolution of a larger functional repertoire, which is a supported by the distinct expression profiles Figure 3 . The protein features and topology of the Dispanin subfamilies. The picture shows the membrane topology and sequences features of a representative human member of each subfamily. Conserved motifs and residues are shown and those which have a sequence identity of more than 90% are framed in black and those with 80-90% sequence similarity are framed in blue. Predicted phosphorylation (Green) and glycosylation (orange) sites are shown. doi:10.1371/journal.pone.0031961.g003 of the subfamilies. By identifying homologs to the IFITM genes and establishing the Dispanins as a family together with a solid detailed and evolutionary based nomenclature for the vertebrate genes, we provide a fundament for future functional characterization these genes. The whole proteome dataset for the following eukaryotic species was included in the analysis: Homo sapiens, Mus musculus, Gallus gallus, Xenopus tropicalis, Danio rerio, Petromyzon marinus, Drosophila melanogaster, Caenorhabditis elegans, Saccharomyces cerviciae, Schistosoma mansoni, Apis mellifera, Anopheles gambiae, Pediculus humanus, Ixodes scapularis, Daphnia pulex, Oryza sativa, Pristionchus pacificus, Acyrthosiphon pisum, Trypanosoma brucei, Leishmania braziliensis and Ciona instestinalis were downloaded from Ensembl; Strongylocentrous purpuratus was downloaded from Spbase (www.spabase.org); Branchiostoma floridae, Nematostella vectensis, Trichoplax adhaerens, Phytophtera soyae, Thalassiosira pseudonana, Naegleria gruberi and Monosiga brevicollis were downloaded from the Joint Genome Institute; Dictyostelium discoideum was downloaded from dictyBase (www.dictybase.org); Arabidopsis thaliana was downloaded from TAIR (http://www.arabidopsis.org/); Entamoeba histolytica was downloaded from amoebaDB (http://amoebadb.org); Paramecium tetraurelia was downloaded from NCBI; Tetrahymena thermophila was downloaded from UniProt; Trichomonas vaginalis was downloaded from TrichDB (http://trichdb.org); Giardia lamblia was downloaded fromGiardiaDB (http://giardiadb.org). All proteomes were searched against a local installation of the Pfam database (v.23) [24] using HMMER3 [25] and the script pfam_scan.pl, which was obtained from the Pfam ftp-site (ftp://ftp. sanger.ac.uk/pub/databases/Pfam/Tools/), with Pfam's default settings. In Pfam, the IFITM family is represented by a specific hidden Markov model [Pfam: P04505]. All the proteins that were assigned to this model in the Pfam-search were considered to be homologous to the IFITM family and were therefore included for further analysis. The script pfam_scan.pl uses the homology criterion set by the Pfam database, which is based on a manually curated gathering threshold for each model. The gathering threshold for PF04505 is a score of 20.6. The sequence datasets were controlled for annotated pseudogenes and transcript variants from the same gene. In the case of multiple transcript variants, the longest sequence was kept. The resulting non-redundant datasets were used for the analysis. The bacterial sequences were obtained by querying Uniprot (www.uniprot.org) for the Pfam ID [Pfam: PF04505]. Thereafter, the sequence set was downloaded by browsing by taxonomy and restricting it to species with a full proteome set. To assure that no lineages were missing in the selection of proteomes the nr protein dataset from NCBI was downloaded. The nr datasets contained 15,322,545 (20-09-11) protein sequences from a wide range of organisms. The dataset was searched against the PF04505 Pfam model using HMMER3 with default settings and sequences with a score above the Pfam gathering threshold (20.6) were deemed as homologous to IFITM. Mafft-einsi was used, with default settings, to create a multiple sequence alignment (MSA) for the vertebrate protein sequences [26] . The MSAs were thereafter examined and refined in Jalview 2.5.1 [27] , i.e. the sequences were trimmed and well conserved and aligned regions were kept, which included 82 aligned amino acid columns. Phylogenetic analysis was performed with a Bayesian approach implemented in MrBayes [28] . The following settings for the eukaryote proteins were adjusted: The analysis was run using a gamma shaped model for the variation of evolutionary rates across sites (rates = gamma) and the mixed option (aamodelpr = mixed) was used to estimate the best amino acid substitution model. We generated 5 000 000 trees and the Markov chain Monte Carlo analysis reached well below a standard deviation of split frequencies of 0.01. Each hundred tree was sampled from the mcmc run and the first 25% of the sampled trees were discarded (burnin = 0.25) to reassure a good sample from the posterior probability distribution. A consensus tree was built from the remaining 37 500 trees with the MrBayes sumt command using the 50% majority rule method. The sump command was used to assure that an adequate sample of the posterior probability distribution was reached during the mcmc procedure. To validate the phylogenetic inference with MrBayes a maximum likelihood method implmemented in RAxML was used [29] . The combined rapid bootstrapping and search for the bestscoring ML tree option (-f a) in RAxML was used to create 1000 bootstraps (-# 1000) using a gamma model of evolutionary rates and the JTT substitution model (-m PROTGAMMAJTT). The JTT substitution model was identified as the most suitable model in the Bayesian analysis and therefore selected for the maximum likelihood phylogeny. The consensus phylogenetic tree found with MrBayes was drawn in Dendroscope 3.0 and the bootstrap support values from RAxML were annotated on the corresponding nodes [30] . The phylogenetic tree was used to determine subfamilies by identifying clusters with a high posterior probability and bootstrap support. Invertebrate sequences were excluded from the phylogenetic analysis as they induced highly unstable topologies together with 1 M. musculus, 2 D. rerio and 3 X. tropicalis sequences. These excluded sequences were categorized into their respective subfamilies by using a BLAST search towards the categorized sequences. The top five hits were examined to classify the invertebrate and excluded sequences into subfamilies. A sequence was assigned to the subfamily if four out of the five top hits are from the same subfamily. Several resources were used to identify the protein sequence features of the Dispanins. NetPhos 2.0 [31] identified potential phosphorylation sites. NetNGlyc 1.0 and NetOGlyc 3.1 [32] were used to find possible N-and O-glycosylation sites respectively. Transmembrane helix prediction was made using TMHMM 2.0 [33] . Motifs were found manually through Jalview 2.5.1 and the server MEME 4.4.0. Finally, EMBOSS:cons was used to create consensus sequences of the different Dispanin families that emerged from the phylogenetic analysis and the bacterial sequences. The consensus sequences were aligned together with the M. brevicollis and the invertebrate sequences. The resulting MSA was viewed and trimmed in Jalview and the conserved region around the TM helices was kept. Splice sites were detected in the eukaryotic Dispanins by studying their annotation in the respective databases and align them to their genome using BLAT at the UCSC Genome Browser website (http://genome.ucsc.edu). Table S1 This is a record of all identified Dispanins together with their accession numbers, nomenclature and species belonging. (XLS)

Health System Resource Gaps and Associated Mortality from Pandemic Influenza across Six Asian Territories

BACKGROUND: Southeast Asia has been the focus of considerable investment in pandemic influenza preparedness. Given the wide variation in socio-economic conditions, health system capacity across the region is likely to impact to varying degrees on pandemic mitigation operations. We aimed to estimate and compare the resource gaps, and potential mortalities associated with those gaps, for responding to pandemic influenza within and between six territories in Asia. METHODS AND FINDINGS: We collected health system resource data from Cambodia, Indonesia (Jakarta and Bali), Lao PDR, Taiwan, Thailand and Vietnam. We applied a mathematical transmission model to simulate a “mild-to-moderate” pandemic influenza scenario to estimate resource needs, gaps, and attributable mortalities at province level within each territory. The results show that wide variations exist in resource capacities between and within the six territories, with substantial mortalities predicted as a result of resource gaps (referred to here as “avoidable” mortalities), particularly in poorer areas. Severe nationwide shortages of mechanical ventilators were estimated to be a major cause of avoidable mortalities in all territories except Taiwan. Other resources (oseltamivir, hospital beds and human resources) are inequitably distributed within countries. Estimates of resource gaps and avoidable mortalities were highly sensitive to model parameters defining the transmissibility and clinical severity of the pandemic scenario. However, geographic patterns observed within and across territories remained similar for the range of parameter values explored. CONCLUSIONS: The findings have important implications for where (both geographically and in terms of which resource types) investment is most needed, and the potential impact of resource mobilization for mitigating the disease burden of an influenza pandemic. Effective mobilization of resources across administrative boundaries could go some way towards minimizing avoidable deaths.

Recent experience from the 2009-H1N1 pandemic highlights how health system capacities, even in developed countries, can be stretched by relatively mild pandemic scenarios [1] [2] [3] . Indeed, the vast majority of previous health system analyses in relation to pandemic influenza have focused on developed countries [1, [3] [4] [5] [6] [7] [8] [9] [10] [11] , while the capacity of health systems in low and middle income countries remains largely unstudied. Paradoxically, understanding outbreak response capacity in low and middle-income countries is arguably of greater importance than that in developed countries, not only because health systems are weaker [12] , but also because many of these countries are in regions where the risk of emerging infectious diseases is highest [13] . Moreover, these countries may suffer disproportionately because of associations between morbidity, pandemic influenza and poverty [14] . Proposed strategies for pandemic preparedness in many countries frequently focus on development and acquisition of pandemic vaccines and stockpiling and distribution of antiviral drugs [15] [16] . In the Southeast Asia region, while surveillance and outbreak response capacities have been strengthened in hope of early detection and control of outbreaks, there has been much less investment into preparing health systems for pandemic mitigation [17] . Modeling studies have been used to inform optimum intervention strategies for responding to pandemic influenza, but often neglect to take into account feasibility of health systems to implement such a response and the potential impact of resource shortages on the pandemic burden. Investigation of health system capacity in East and Southeast Asia is of particular interest, not only given the fertile conditions for the emergence and spread of new diseases [13, 18] , but also the wide socio-economic inequalities within the region, and focus of investment by the international community into pandemic influenza preparedness [19] . Resource gaps for a pandemic response are likely to be wide and vary greatly between and within countries in Asia [20] . But exactly how wide are these gaps, what are the consequences of the gaps in terms of the pandemic disease burden, and to what extent could these consequences be mitigated by improving resource allocation and mobilization? To address these questions, we conducted a health systems analysis across six Asian countries and territories with widely varying socioeconomic conditions: Cambodia, Indonesia, Lao PDR, Taiwan, Thailand and Vietnam. In this analysis, mathematical modeling and health system resource data collected across the six territories were used to estimate and compare, within and across countries, the resource gaps, and potential consequences of those gaps in terms of expected mortalities, for a hypothetical pandemic influenza scenario. This study was conducted as part of the AsiaFluCap project (www.asiaflucap.org), the overall aim of which is to conduct health systems analyses to support capacity development for responding to pandemic influenza across six countries and territories in Asia, specifically: Cambodia, Indonesia, Lao PDR, Taiwan, Thailand and Vietnam. For this comparative analysis we focus on four key health system resources: antiviral drugs (specifically oseltamivir), hospital beds, mechanical ventilators and healthcare workers (doctors and nurses), chosen due to their critical importance for responding to pandemic influenza. These resources, along with over 50 other resource items relating to health system infrastructure, equipment, materials, and human resources, were selected through a systematic literature review and a Delphi consensus process by a panel of 24 experts, as described in [21] . Quantities of these resource items were enumerated during March to September 2009 through questionnaires administered to hospitals and health offices in all districts of each of the six study countries (except Indonesia, where data were collected only from districts in Jakarta and Bali due to the vast geographic scale of the country). Additional questionnaires were sent to ministries of health to capture central stockpiles. We received 100% response rates from hospitals and district health offices in Cambodia. Overall response rates from district health offices were more than 95% in Viet Nam and Indonesia (Jakarta and Bali), 86% in Lao PDR, 72% in Taiwan, and 59% in Thailand. The response rates for hospital questionnaires were slightly lower at 93% for Indonesia and Vietnam, 70% for Lao PDR, and approximately 46% for Thailand and Taiwan. Data from the questionnaires were double-entered into an Excel database. Missing values, due to non-responses or incomplete questionnaires, were extrapolated using linear prediction models, specific to each country and resource item, based on a number of district characteristics such as total number of hospital beds or public hospital beds, population size, and geographic location (region/province). For oseltamivir and ventilators, two-step models were used to first estimate the likelihood of having any oseltamivir or ventilators, and then to predict the number of these items. Extrapolation of missing data was carried out in STATA version 11. In order to estimate health system resource needs and gaps for a pandemic influenza scenario, we used a mathematical model previously developed as part of the AsiaFluCap project to simulate the transmission dynamics of a pandemic influenza outbreak [22] . Full details and equations of the model can be found in [22] , and are summarized in Text S1. Briefly, the model is based on a deterministic SEIR (Susceptible-Exposed-Infectious-Recovered/removed) model described by differential equations tracking number of people in each compartment over time. Given that the primary aim of this analysis was to provide relative estimates of resource gaps within and across countries, rather than to accurately simulate the spread of pandemic influenza throughout each country, the model structure was kept relatively simple, with homogeneous mixing patterns and no age-structure assumed within the modeled population. However, novel complexity is incorporated through making parameters describing the clinical course of infected individuals conditional upon the availability of certain key health system resources (antivirals, beds, and ventilators). Thus we could obtain relative indications of the consequences of resource shortages on the pandemic disease burden, specifically in terms of ''avoidable deaths'', which we define as deaths that would not have occurred in the presence of sufficient resources. The infectious compartment of the model was subdivided into three groups based on clinical severity: asymptomatic, mild and severe infections. All asymptomatically and mildly infected patients were assumed to recover, while severe cases were at risk of death and were assumed to need antiviral treatment and hospitalized care, although whether they received either of these depended upon the availability of oseltamivir and hospital beds, respectively. Treatment with antivirals and hospitalization were assumed to reduce the infectious period and the probability of death for severe cases. Furthermore, a proportion of severe cases are assumed to require mechanical ventilation, which they will receive as long as ventilators are available, otherwise these cases would die. The model parameters describing transmissibility and clinical severity were chosen based on data from the 2009-H1N1 pandemic, with a basic reproduction number of 1.32 [23] [24] . Under the parameter values chosen (see Table S1 and [22] ), in a population with sufficient resources (i.e. when there are no shortages of oseltamivir, hospital beds, or ventilators), the scenario predicts an overall attack rate of 35.6%, a clinical attack rate of 24.9%, a peak prevalence (of symptomatic cases) of 0.94%, and a case fatality rate of 0.018%. In the absence of any resources, the case fatality rate is substantially higher at 0.029%, while attack rates and peak prevalence remain very similar. Estimating resource needs, gaps, and associated mortality In our baseline scenario, resource gaps were estimated assuming that 12% of ''general'' hospital resources (beds, ventilators and human resources) are available for care of pandemic influenza cases, with the remaining 88% required for maintaining essential healthcare services, as in a previous pilot study for Thailand [20] , and based on previous reports [25] [26] . We also assumed that, in the event of a pandemic all available oseltamivir doses would be dedicated to severe influenza cases. These assumptions regarding resource spare capacity and oseltamivir usage were relaxed in a multivariate uncertainty analysis (described below). The resource data were aggregated at provincial level for all countries except Taiwan, where they were aggregated at county level, and Indonesia, where the data were kept at district level. (Counties in Taiwan and districts in Indonesia are of comparable population size to the provinces of the other four countries.) We then ran the model separately for each province (Lao PDR, Cambodia, Thailand, Vietnam), county (Taiwan) and district (Indonesia), with the appropriate resources for each of these administrative areas, assuming a closed population and that resources could not be shared between these areas in a timely manner. For the purposes of narrative flow, we henceforth use the term ''province'' for counties in Taiwan. All simulations started with one mild case entering a completely susceptible population. In addition to running the model with the available resources, based on the survey data, we also ran the model with unlimited resources, in order to calculate resource needs, and thus also the resource gaps (or indeed surpluses) by comparing with the resource availability data. The needed number of hospital beds, ventilators and humans resources were estimated from the peak number of cases requiring hospitalization and ventilation, while the needed number of oseltamivir doses was calculated from total number of severe cases occurring over the duration of the outbreak (full details on assumptions of resource depletion rates are detailed in [22] , and summarized in Text S1). By comparing the number of deaths predicted by simulations with sufficient resources with those from simulations using actual resource data, we also estimated the number of deaths due to resource gaps, which we term as ''avoidable deaths''. A multivariate uncertainty analysis was conducted to approximate uncertainty surrounding avoidable deaths in light of uncertainty in resource spare capacity and effectiveness, which may vary between settings. Specifically, the proportion of ''general'' healthcare resources within each province/district that would be available to care for pandemic influenza patients was allowed to vary between 5-20%, and parameters describing the effectiveness of each resource for reducing the risk of death in cases requiring those resources were allowed to vary independently between a wide range of 20-80%. We also explored the impact of relaxing the assumption that oseltamivir administration is restricted to severe influenza cases, by allowing between 0-5% of mild cases to be treated. One hundred combinations of values were chosen randomly from these ranges using Latin hypercube sampling, and simulations were run using each combination. The medians, interquartile ranges (IQR) and 95 th percentile ranges of model outcomes were then calculated across the simulations. Since the aim of this study was to compare resource capacities across geographic areas and resource types, rather than to evaluate how transmission dynamics may vary across geographic areas, epidemiological parameters of the model were kept fixed in the multivariate uncertainty analysis to ensure comparability of resource capacity outcomes. However, due to the unpredictability of pandemic scenarios, we also explored model outcomes for a range of values for R 0 and for the severe clinical attack rate in a separate univariate sensitivity analysis. The model was coded and run in R version 2.10.1, using the ''simecol'' package [27] with the Runge-Kutta 4 th order algorithm for numerical integration of the differential equations. ArcGIS version 10 was used to map the calculated resource gaps and avoidable mortalities at provincial level. Figure 1 presents the geographical distribution of estimated resource gaps across provinces (or districts in the case of Indonesia) in each study country for the modeled pandemic influenza scenario, under our baseline assumptions and point estimates for parameter values. The corresponding statistical distributions of resource capacities across areas within each country can be found in Figure S1 . A summary of overall resource gaps for each country is presented in Table 1 . There was substantial variation in resource gaps both between and within countries, and across resources types (Figures 1 and S1) . Overall, the biggest gaps were generally seen in Cambodia and Lao PDR, particularly when standardized by population size, with almost all provinces in these countries displaying gaps in all resources, with the exception of nurses which were estimated to be sufficient in approximately half of the provinces in these countries. In contrast, relatively few provinces in Taiwan were estimated to have gaps, at least in general health system resources (beds, ventilators, and human resources), with quantities of these resources often considerably above those predicted to be needed for this scenario. Nevertheless, almost half of provinces in Taiwan were predicted to have insufficient oseltamivir supplies to treat all severe cases (although it should be noted that the results in Figures 1 and S1 do not account for central stockpiles which might be mobilized in the event of a pandemic, as discussed later). Thailand, Indonesia and Vietnam generally displayed a more mixed picture. Results were comparable between Vietnam and Indonesia, with relatively few provinces of Vietnam (7?9%) and only one district of Jakarta (and none in Bali), estimated to have insufficient oseltamivir to treat all severe cases, with supplies of this antiviral drug comparably high across most other areas in these countries. Healthcare workers were also predicted to be mostly sufficient in Vietnam, with only 3?1% and 1?6% of provinces predicted to have a shortage of doctors and nurses, respectively, for this scenario. However, gaps in hospital beds were observed in over half of provinces in Vietnam and districts of Jakarta and Bali, and all of these provinces displayed a shortage of mechanical ventilators. Indeed, of all the resources, the largest gaps were observed in ventilators across all countries except Taiwan (Table 1) . Thailand had the second highest number of ventilators (absolute and per capita) after Taiwan, but a shortage of this resource was nevertheless predicted in over 80% of Thai provinces. A very heterogeneous pattern was observed in Thailand, with around 50% of provinces showing a shortage of hospital beds and oseltamivir, while many other provinces showed a clear ''surplus'' of the latter. Meanwhile, a shortage of medical doctors was predicted in over 80% of Thai provinces, with gaps in doctors comparable to those Lao PDR and Cambodia ( Table 1 ). The number of nurses in Thai provinces was estimated to be somewhat more sufficient, however. A fairly distinct geographical pattern of resource gaps was evident in Thailand, particularly for oseltamivir and ventilators, with north-eastern (and some southern) provinces showing shortages more comparable with those in neighboring Lao PDR and Cambodia than with other Thai provinces ( Figure 1 ). It should be noted that estimates of gaps in beds, ventilators and human resources were highly dependent on the spare capacity of resources assumed to be available to care for pandemic influenza patients. For example, if spare capacity was assumed to be 5%, rather than 12%, then even in Taiwan most areas would suffer gaps in these resources for the modeled scenario. Furthermore, use of oseltamivir on even a fairly small proportion (5%) of mild cases, resulted in much faster depletion of this resource, such that all countries are predicted to experience a shortage of oseltamivir for treating all severe cases. The geographic distribution of avoidable deaths, estimated by calculating the number of deaths that would be prevented by filling all resource gaps in each province, and standardized by population size, is presented in Figure 2 . (A corresponding map showing absolute numbers of estimated avoidable death is given in Figure S2 .) Figure 3 shows estimated avoidable death rates attributable to gaps in each resource type (antivirals, beds and ventilators) aggregated across all provinces in each country, and accounting for uncertainty in resource effectiveness and spare capacity. Avoidable deaths for a given resource gap were estimated by calculating the number of deaths that would be prevented by filling that resource gap only. Figures 2 and 3 highlight how resource gaps could have a substantial impact on mortality rates during an influenza pandemic. A combination of the large population size and shortage of ventilators results in the estimation that, out of the five countries for which nationwide data were collected, Vietnam would have the highest total number of avoidable deaths. However, the results for Jakarta and Bali suggest that Indonesia would have the highest avoidable death toll if the data is extrapolated across the entire population of this country. When standardized by population size, the highest rates of avoidable deaths were estimated in Cambodia and Lao PDR, accounting for over half of all pandemic-associated mortalities in these countries. The median avoidable death rates for these countries were over 15 times higher than that for Taiwan, where a relatively low proportion (median: 7.6%, IQR: 5.5-10.7%) of total deaths was estimated to be due to resource gaps. Almost all avoidable deaths in Taiwan were predicted to be due to local shortages of oseltamivir. In all other countries shortages of ventilators were estimated to be the biggest cause of avoidable deaths (Figure 3 ). In Indonesia, Lao PDR, and Vietnam, this result was largely robust to uncertainty surrounding resource effectiveness and spare capacity. For Cambodia and Thailand, however, the uncertainty analysis suggested that gaps in oseltamivir might also be a main cause of avoidable deaths (Figure 3 ). When adding an additional layer of uncertainty to the model assumptions, by allowing for up to 5% of mild cases to be treated with oseltamivir, the increased shortages of the latter further increased uncertainty surrounding the relative importance of gaps in oseltamivir ( Figure S3 ). Given such uncertainties and the sensitivity of results to model assumptions, the proportion of mortalities that can be attributed to gaps in specific resources should be interpreted with some caution. A clear negative correlation was observed between estimated avoidable mortality rates and GDP per capita at country level ( Figure 4A ). Total funds per capita committed by donors towards avian and human influenza for each country, up to December 2009 [28] were positively correlated with avoidable mortality rates ( Figure 4B ). Within many countries it was evident that, while at least some provinces displayed resources gaps, other provinces were estimated to have more than sufficient resources for responding to the modeled scenario, with an overall ''surplus'' of some resources in several countries (Table 1) . Furthermore, central stockpiles of oseltamivir were present in all countries from which data on this could be obtained. Thus we also investigated the proportion of avoidable deaths that might be averted in each country if the total available resources were equitably distributed across provinces according to provincial population size ( Figure 5 ). When accounting for central stockpiles of oseltamivir, the overall supply of this drug was estimated to be sufficient to treat all severe cases in all countries (Table 1) . Thus it was estimated that, in each country (except for Vietnam, and Jakarta and Bali, where provincial supplies of oseltamivir are already relatively high), effective mobilization of oseltamivir across administrative areas could potentially avert a significant proportion of the avoidable deaths estimated under current resource distributions (up to 100% of avoidable mortalities in Taiwan; Figure 5 ). The (less feasible) scenario of redistributing available beds and ventilators according to provincial need within each country was generally estimated to have less of an impact on the number of avoidable deaths, compared to mobilization of oseltamivir ( Figure 5 ). In the case of ventilators, this highlights how the large numbers of deaths attributed to gaps in this resource (Figure 3 ) are mostly due to overall nationwide shortages of ventilators, rather than an inequitable distribution of ventilators within most countries. In Thailand, however, if all ventilators were distributed in proportion to provincial population sizes, the model predicts around 30% (IQR: 21-41%) fewer avoidable deaths than the number predicted under the observed ventilator distribution. Estimates of resource gaps, and thus also avoidable mortalities, were very sensitive to the severity of the modeled pandemic scenario in relation to transmissibility and proportion of cases requiring hospitalization ( Figure S4) . A sensitivity analysis showed that under more severe (yet still plausible) pandemic scenarios, even Taiwan could experience substantial deaths due to shortages of hospital resources ( Figure S4A and S4C) . Furthermore, as the severity of the scenario increased, so too did the proportion of avoidable deaths that were attributable to gaps in hospital bed capacity (shown for Cambodia in Figure S4B and S4D) . It is important to note, however, that for the ranges of values explored for the basic reproduction number and the proportion of cases that become severely ill, consistent patterns were observed when comparing relative magnitudes of avoidable mortality rates across countries (and also across provinces within countries). Our results indicate that health system resource gaps for responding to a mild to moderate pandemic influenza scenario are wide and vary greatly, both within and between countries in Southeast Asia, and that these gaps could have a profound impact on pandemic-associated mortalities. Our estimates of resource gaps and avoidable mortality rates at country level show a clear association with national GDP. This result is consistent with a previous analysis of data from the 1918 influenza pandemic, which found that per capita income explained a large proportion of the variation in mortality across countries during the pandemic period [14] . Moreover, extrapolation of these mortality rates to the 2004 world population suggested that around 96% of deaths from pandemic influenza would occur in developing countries [14] . Our results suggest that, due to inequitable distribution of resources, the variation in pandemic burden is likely to be profound within, as well as between, countries. Countries which have experienced the highest burden of Highly Pathogenic Avian Influenza (H5N1), namely Vietnam and Indonesia, appear to be most prepared in terms of the availability and geographical distribution of oseltamivir. In the other study countries, we estimated that central stockpiles of oseltamivir would be sufficient to cover any provincial gaps for treating all severe cases from the modeled scenario, and thus mobilization of this resource could potentially avert a large number of avoidable mortalities in these countries. Indeed, in all countries except Vietnam, we estimated that optimum mobilization of resources across administrative boundaries could save more than 10% of avoidable deaths. While timely mobilization of resources may be possible in Taiwan, with its small geographical size and relatively developed infrastructure, the feasibility of this scenario is questionable in the poorer, and larger, countries of the Mekong region, where it might be prudent to disburse central stockpiles of antiviral drugs to provincial and district health facilities prior to an outbreak. Gaps in mechanical ventilators were predicted to be a major cause of avoidable deaths, with almost all provinces across all countries estimated to have severe shortages of this of this resource, with the exception of Taiwan and some Thai provinces. This pattern likely reflects the relatively high cost and human resource skills associated with acquisition and operation of ventilators, and highlights the importance of developing robust triage criteria as part of pandemic preparedness plans to ensure that this resource is allocated to the patients who are most likely benefit [29] [30] . A previous analysis similarly suggested that a dire shortage of mechanical ventilators would be a major limiting factor in responding to a pandemic influenza outbreak in the United States [31] . We found particularly wide variation in the availability of ventilators, and indeed other hospital resources, in Thailand, where our results suggest that inequitable distribution of health system resources [32] , rather than simply an overall nationwide shortage, could lead to a high number of avoidable deaths from pandemic influenza. Of course, hospital equipment such as beds and ventilators are useless unless sufficient and qualified human resources are available to treat influenza patients, and our results suggest that gaps in healthcare workers would also be an important limiting factor for responding to pandemic influenza in many countries, particularly for Cambodia, Lao PDR and Thailand. It is encouraging that total donor funds committed to avian and human influenza broadly correspond to avoidable mortality rates estimated at country level. A recent paper on Financial and Technical Assistance from the 2010 International Ministerial Conference on Animal and Pandemic Influenza (IMCAPI) reports that over 50% of total donor funding committed towards avian and human influenza worldwide between 2005 and 2009 was allocated towards ''human health and pandemic preparedness'' (with other funds committed towards sectors such as animal health; monitoring, information, and internal coordination; and information, education and communication) [28] . However, the extent to which these funds have been, or will be, allocated towards mitigating the resource gaps identified in our study is unknown to us and beyond the scope of this analysis. This study is subject to several limitations, many of which relate to assumptions that were necessary for the modeled scenario. For example, in our baseline scenario we assumed that 12% of hospital capacity, across all provinces and countries, would be available to care for the surge of patients with influenza infections. In reality, surge capacity is likely to vary substantially between and within countries (and over time), but few data on this are available. Robust analytical frameworks are urgently needed to define and measure health system surge capacity in order to inform analyses of resource gaps for emergency response scenarios. The effectiveness of resources such as antiviral drugs, ventilators, and general hospital care for improving the survival rates among severe influenza cases is also surrounded by considerable uncertainty and may vary between settings. Our results show that, even if epidemiological parameters describing transmission and pathogenicity are kept constant, uncertainties in spare capacity and resource effectiveness lead to considerable uncertainties in estimates of avoidable mortalities rates. Nevertheless, the distributions of model outputs from our multivariate uncertainty analysis still showed some significant differences when compared across countries and across resource types (Figures 3 and 5) . Furthermore, it seems likely that spare capacity and resource effectiveness would be higher in more resource-rich settings, which would only strengthen the findings of this study in terms of the geographic distribution of resource gaps and avoidable mortalities. Estimates of resource gaps and avoidable mortality rates were also very sensitive to parameters describing pandemic severity, although similar patterns were observed across geographic areas for the range of values explored. However, the same cannot be said for the relative importance of gaps in different resource types. Thus, although our results generally suggest that shortages of ventilators could be a major cause of avoidable deaths in low-and middle-income countries in Southeast Asia, investment in this resource should not necessarily be prioritized over other healthcare resources. A natural extension of this study would be to investigate the cost-effectiveness of investing different types of health systems resources for mitigating the burden of an influenza pandemic. However, more data on the effectiveness of different resources for managing severe influenza cases is needed before such assessments can be made. Other limitations relate to the simplicity of the model structure. We assumed homogenous mixing and a constant basic reproduction number across all populations. In reality, heterogeneities in factors such as age-structure, geographic structure, population density, human behavior, and the underlying health of the population are all likely to play a role in transmission dynamics and burden of influenza. A previous modeling analysis, for example, has shown that higher levels of population heterogeneities, such as in age and spatial structuring of contacts, result in lower overall attack rates and peak prevalence for a given basic reproduction number [33] . However, there is a lack of data on such heterogeneities and how they might affect patterns of pandemic progression for our study region. Ongoing studies, such as contact pattern surveys in Asia similar to those undertaken in Europe [34] , are attempting to rectify this. Another limitation is that the resource data were collected between May and September 2009, which includes the first wave of the H1N1-2009 pandemic; thus some resource data (particularly for antiviral stockpiles) may be influenced by the time point within this period at which the data were recorded. Given the above caveats, it is important to emphasize that we do not advocate these results to be accurate quantitative reflections of resources shortages or deaths that are likely to occur in any given pandemic scenario. Rather, they highlight the scale of health system inequalities within and across countries in Asia, and the considerable impact such inequalities could have on the pandemic disease burden. By indicating the relative disparities in resource availability within and across countries, and the potential consequences of resource shortages, these results could help guide investment decisions in scaling up resources to mitigate the burden Figure 5 . Estimated impact of resource mobilization/redistribution across provinces on avoidable mortality rates within each territory. Data were calculated by estimating the number of avoidable deaths if available resources (including central stockpiles for oseltamivir) within each territory were geographically distributed in proportion to provincial population size, and comparing with the total number of avoidable deaths predicted given actual resource distribution. Boxplots show medians, interquartile ranges, and 95 th percentile ranges derived from a multivariate uncertainty analysis. Data are aggregated across provinces for Cambodia, Lao PDR, Thailand, Vietnam, and across counties for Taiwan. Data for Indonesia are aggregated across districts of Jakarta and Bali only. doi:10.1371/journal.pone.0031800.g005 of future pandemics. As many of these resources have a generic healthcare function beyond pandemic influenza, they may also be useful to guide health system strengthening. Figure S1 Variation in estimated resource capacities across provinces within each territory for the modeled pandemic scenario. (DOCX) Figure S2 Geographical distribution of estimated avoidable deaths due to resource gaps for a modeled pandemic influenza scenario.

First Dating of a Recombination Event in Mammalian Tick-Borne Flaviviruses

The mammalian tick-borne flavivirus group (MTBFG) contains viruses associated with important human and animal diseases such as encephalitis and hemorrhagic fever. In contrast to mosquito-borne flaviviruses where recombination events are frequent, the evolutionary dynamic within the MTBFG was believed to be essentially clonal. This assumption was challenged with the recent report of several homologous recombinations within the Tick-borne encephalitis virus (TBEV). We performed a thorough analysis of publicly available genomes in this group and found no compelling evidence for the previously identified recombinations. However, our results show for the first time that demonstrable recombination (i.e., with large statistical support and strong phylogenetic evidences) has occurred in the MTBFG, more specifically within the Louping ill virus lineage. Putative parents, recombinant strains and breakpoints were further tested for statistical significance using phylogenetic methods. We investigated the time of divergence between the recombinant and parental strains in a Bayesian framework. The recombination was estimated to have occurred during a window of 282 to 76 years before the present. By unravelling the temporal setting of the event, we adduce hypotheses about the ecological conditions that could account for the observed recombination.

The mammalian tick-borne flavivirus group (MTBFG) includes viruses associated with important human and animal diseases such as encephalitis (Tick-borne The corresponding polyprotein is proteolysed and processed into structural, Capsid (C), Pre-Membrane (PrM), Envelope (E) and nonstructural proteins NS1 (glycoprotein), NS2A, NS2B (protease component), NS3 (protease, helicase, RNA triphosphatase activity and NTPase activity), NS4A, NS4B and NS5 (methyltransferase, RNA-dependant RNA polymerase) [1] . Several viruses bear close evolutionary relationships to TBEV [2] [3] [4] [5] [6] viz. LIV, Spanish sheep encephalomyelitis virus (SSEV), Turkish sheep encephalitis virus (TSEV) and Greek goat encephalitis virus (GGEV). These four lineages have recently been assigned to a single species dubbed Tick-borne encephalitis virus [4] , whose members are primarily associated with ixodic hard-tick vectors. Within this species, the TBEV lineage is further divided into three evolutionary distinct subtypes, the Western European-(W-), the Far Eastern-(FE-) and the Siberian-(S-) TBEV [1, 7] . In this contribution, our attention is mainly focused on the evolutionary relationships between W-TBEV, SSEV and LIV. W-TBEV is widely distributed throughout continental Europe and Russia, SSEV is endemic to Spain [2] [3] [4] [5] [6] 8] , and LIV, initially considered to be restricted to the British Isles and Ireland [2] [3] [4] [5] [6] , has now been reported from Norway [2] and Denmark [9, 10] . The ecology and pathogenesis of both W-TBEV and LIV have been intensively investigated [11, 12] , whereas studies dedicated to SSEV are scarce. In contrast to mosquito-borne flaviviruses where recombination events are frequent [13, 14] , evolution in the MTBFG was considered to be clonal. This perception changed recently with reports of several putative recombinations in Tick-borne encephalitis virus [15, 16] . We aim to investigate the strength of the recombination signals reported by Yun et al. [16] , since if proved valid their discovery would lead to a radical departure from the classical understanding of the evolutionary dynamic in MTBFG. Although, we could not confirm the previously described recombinations, we did identify a strong recombinant signal in the LIV lineage. Putative parents, recombinant strains and breakpoints were further tested for statistical significance using phylogenetic methods. The second aspect of this study pertains to date the recombination event. We used the available full length coding genomes for dating, but this small sample may limit the power of the molecular-clock analysis. There are a large number of Esequences available from molecular epidemiological studies. Unfortunately, E-sequence cannot be used directly to date recombination events, as we estimated that the substitution rates for the E-gene is significantly lower than for other viral genes. This means that dates obtained from E-sequences alone tend to be younger and do not represent accurately the temporal dynamic of this viral lineage. We suggest that a large dataset that includes sonly E-sequences could nevertheless be used to date additional divergence events by specifying informative priors on the ages of some important nodes. We describe an incremental analytical strategy that bases these priors on posterior distributions derived from the analysis of full-length coding sequences following removal of the E-sequences. Alignments were generated from GenBank sequences retrieved in January 2011, aligned using Muscle [17] , rechecked and improved manually in the UTR regions. Sequences were numbered from the start of the ORFs using Neudoerfl (U27495) as reference. Details on the included sequences are provided in Table S1 . ALN1 contains 41 complete nucleotide sequences of Tick-borne encephalitis virus and three out-groups selected among LGTV and OHFV. This initial alignment was scanned for recombination events and then down sampled to an alignment (ALN2) of 28 complete sequences of known collection dates (from 1937 to 2008), with the deletion of out-groups and strains with unusual sampling locations. UTRs and gap columns were deleted. ALN2 was further partitioned by individual genes resulting in alignments ALN2_C, ALN2_PrM, ALN2_E, ALN2_NS1, ALN2_NS2A, ALN2_NS2B, ALN2_NS3, ALN2_NS4A, ALN2_NS4B and ALN2_NS5. Next, we produced ALN3 from ALN2 with the deletion of the E gene and the region of NS3 identified as a possible recombinant fragment. Finally, E_161 was compiled from the 161 longest Esequences available in Genbank (1033 to 1491 nt in length) endowed with sampling dates (from 1931 to 2008). An analysis of the entire species (ALN1) was conducted with split networks using the neighbor-net method [18] . Evolutionary distances were estimated using maximum likelihood (ML) with a GTR+C 4 +I as the best-fit substitution model as determined by MODELTEST v.3.7 [19] , according to the Akaike Information Criterion. Several methods were used to extract recombination signal from ALN1 with the RDP3beta36 package [20] , because inspection of the split network had established the possibility of recombination within the species (see results). All analyses were carried out with Bonferroni correction (P-value,0.05) and signals reported by more than one method were retained. RDP [21] , GENECONV [22] , BootScan [23] , MaxChi [24] , Chimaera [20] , and SiScan [25] were used for screenings the alignment. For this initial phase, the following settings were modified to balance sensitivity and statistical significance: RDP: window size 25, detect recombination between sequences sharing 90% to 100% identity; GENECONV: G-scale 5; BootScan: windows size 100, use NJ trees, 200 bootstrap replicates, cutoff percentage at 95% and Jin and Nei 1990 model; Chimaera: 40 variable sites per window; SisScan: window size 80, slow exhaustive scan. As all methods detected the presence of significant recombinant signals in the NS3 gene, the dataset was further evaluated for phylogenetic evidence of recombination based on an alignment of NS3-sequences derived from ALN1. For the phylogenetic analysis, the NS3 partitions 59 and 39 of the putative recombinant fragment were concatenated. Trees were inferred separately for the recombinant region alone and for the concatenated region. Maximum likelihood analyses were performed with RAxML VI-HPC v.2.2. [26] via the RAxML Web server [27] . The proportion of invariable sites and the number of bootstrap runs were automatically determined. Bayesian phylogenetic trees were constructed with a GTR+I+G nucleotide substitution model for the concatenated alignment of NS3 and a GTR+G model for the recombinant partition. Model selection was based on the corrected Akaike information criterion in MrAic [28] . For each alignment, two separate analyses were run simultaneously with MrBayes v.3.2-cvs [29] (source code accessed with CVS 22 January 2009) for 5000000 generations using the default settings for priors and MCMC proposals. Trees were sampled every 1000 th generation, and standard deviation of split frequencies was below 0.01 at the end of each analysis. For all Bayesian analyses (i.e. MrBayes and BEAST), mixing of the MCMC chains and effective sample size (ESS) for each parameter estimate were investigated using Tracer v.1.5 [30] which showed convergence and larger than 200 ESS for each summary statistic. For both MrBayes analyses, the first 2500 trees where discarded as burn-in and the 7500 remaining trees were summarized in a majority-rule consensus tree. For each of the two partitions, we tested alternative topological placement for the putative recombinant strain. Constraining the topology in ML analyses yielded likelihoods for alternative placements that were compared with the likelihood of the best ML tree using the approximately unbiased (AU) test [31] in CONSEL [32] . For this step, ML analyses were performed with PAUP* v.4.0b10 [33] and best trees were sought by heuristic searches (10 random addition replicates, TBR branch swapping, Multrees in effect). Throughout the study, node support was estimated by nonparametric bootstrap (BS, bootstrap support) in ML and with multiple samples from the posterior distribution (PP, posterior probability) in BI. Each separate gene alignment (ALN2_C, ALN2_PrM, ALN2_E, ALN2_NS1, ALN2_NS2A, ALN2_NS2B, ALN2_NS3, ALN2_NS4A, ALN2_NS4B and ALN2_NS5) was investigated for signs of positive selection. To that end, the dN/dS ratio for the whole gene, and for each codon in the alignment, was inferred using the M3 model [34] implemented in MrBayes, otherwise using default settings. Mixing of the MCMC chains, as well as the ESS of each estimated parameter was assessed by analyzing the resulting parameter files with Tracer. Each analysis was run until the ESS exceeded 200 for all parameters, after which the probability for the whole gene, or individual codons in the sequence, to have evolved under positive selection was analyzed with Tracer. Substitution rates and dates of ancient divergence were estimated with Bayesian MCMC in BEAST version 1.5.3 [35] , with collection times in years used as calibration points in the clock model. The youngest strain was collected in 2008, which sets this year as the origin for past time estimates. Each dataset was evaluated individually for best fitting substitution model, which ranged from HKY+C 4 +I to GTR+C 4 +I. However, analyses performed under GTR family models neither converged nor mixed well, possibly due to an insufficiency of data to estimate these highly parametric substitution models. Hence, the simpler, less parameter rich, HKY+C 4 +I model was used throughout the BEAST investigation. We tested the impact of using a GTR model by running an analysis for 20610 6 generations. Estimates for the parameters of interest were largely concordant (data not shown), albeit the analyses returned very low ESS and much wider confidence intervals. Pairwise comparisons of Bayes factors calculated in Tracer selected the uncorrelated lognormally distributed relaxed-clock (UCLN) and the Bayesian Skyline coalescence model [36] as the best fitting clock and demographic models following the procedure in Hon et al. [37] . We defined two partitions that separated first and second positions from third codon positions. For each analysis, four independent MCMC chains were run for 20610 6 generations and their log output combined with 10% burn-in samples discarded. Tracer was used to determine the degree of mixing, shape of the probability density distribution, median and highest posterior density regions at 95% (HPD) for the relevant parameters. The modes and parameters of the posterior distributions were estimated using the distribution fitting software EasyFit 5.3 (MathWave Technology). For all analyzed parameters, we modeled the posterior distributions with gamma distributions. The analytical framework of the BEAST analyses is presented in Figure 1 and the details are explained below. We compared the mean substitution rates derived from BEAST analyses for ten individual genes obtained from ALN2. Settings were as described above with an additional uniform prior distribution on the time interval [360-10000] fitted to the root height. This prior captures the background knowledge that crown radiation of flaviviruses occurred after the end of the last glaciations (placed 10,000 years ago) and that the Tick-borne encephalitis virus emerged before the divergence of two of its inclusive clades namely LIV and W-TBEV whose split was estimated to be earlier than 360 years ago [38] , placing the species divergence within this rather wide interval. To estimate the time of the recombination event (tRE), as well as the time of the most recent common ancestor (tMRCA) for each parental strain, we studied separately the genomic partition spanning the recombinant element from nt 5787 to 5991 (ALN2-1) and the partition covering the rest of the ORF (ALN2-2) that is the 59 region (nt 1 to 5786) together with the 39 region (nt 5992 to 10245) flanking the recombinant portion. The same uniform prior was fitted on the root height as before. For individual genes in ALN2-2, prior distributions for the MeanRate parameter were derived from posteriors in BEAST inference 1 with substitution and clock models unlinked during the analysis. This step was designed to provide posterior distributions for the BEAST inference 4. ALN3 (28 full length ORFs with both E-gene and the recombinant fragment omitted) was analyzed the same way as ALN2-2. The mode and parameters of posterior distributions for the root height, tMRCA(Neudoerfl-Hypr) and tMRCA(LIV & SSEV) were estimated in order to be incorporated as priors in the following step. BEAST inference 4: refining estimates for tMRCA(Neudoerfl-Hypr) Due to its sampling, the E_161 alignment allows access to the antiquity of additional divergence events. Posteriors obtained from BEAST inference 3 were included as priors, with an additional uniform prior distribution over [7.28610 25 -6.29610 24 substitutions/site/year] set on the meanRate parameter. This value reflects previously observed substitution rates for the E gene in the Tick-borne encephalitis virus: the lower bound comes from the value of 7.28610 25 substitutions/site/year estimated for nonsynonymous substitutions [38] , while the upper bound comes from an estimation of 4.78610 24 substitutions/site/year with a standard error of 1.51610 24 [39] for synonymous substitutions. Because the analysis of selection pressure (see results) inferred that a strong purifying selection acts on the proteins, we expect to see higher rate of synonymous substitutions than of nonsynonymous substitutions. As the mean rate takes both types of substitutions into account, its estimate should be intermediate between their two values. All alignments, xml-files for the BEAST analyses and all phylogenetic trees have been deposited at Dryad Repository: doi:10.5061/dryad.504636cd. On the inferred network (Figure 2 ), the region of the split-graph separating the four main clusters exhibits a significant ''tree-like'' structure that rules out frequent recombination between the clusters. Nevertheless, a prominent split associated with SSEV (DQ235152) and LIV (Y07863) indicates a marked conflicting and/or ambiguous signal that could be associated with a recombination event. This hypothesis was first examined with RDP3, wherein all methods identified the LIV strain as displaying signs of homologous recombination between the SSEV strain as the major parent and a strain belonging to W-TBEV as the minor parent ( Figures S1 a-b) . All methods recognized with significance that an insert within the NS3 gene of LIV originated from a W-TBEV strain, but they were not consistent with respect to the precise location of the two recombination methods. When run simultaneously, all methods bar, Chimaera and MaxChi, identified Neudoerfl (U27495) as the minor parent and estimated the breaking points at nt 5787 and 5991. When the data were analyzed with Chimaera or MaxChi as single primary detection methods, they instead proposed, with significance (P-value ,3.10 22 ), slightly different breakpoints (Table 1) . No significant evidence for recombination was found in the other strains or genes. We compared our result to the outcome of the screening performed by Yun et al. [16] that identified 11 recombinations within the 39UTR and the 39 end of the NS5 region, but did not include a LIV strain. Their observations could only be replicated when we used exactly the same settings, i.e. when detection was performed on ClustalW [40] aligned sequences, without Bonferroni correction for multiple comparisons. This suggests that the previously reported signal was not strongly supported and could have been caused by alignment problems, as UTRs are notoriously difficult to align due to spontaneous variations in length during laboratory passages [41, 42] . To evaluate phylogenetic evidence of recombination, trees were constructed for the putative recombinant region and for the concatenated regions of NS3 from both sides of the crossover points. As shown in Figure 3 , ML and Bayesian reconstructions contrast the placement of LIV in the two partitions: In the nonrecombinant partition, LIV groups with SSEV with maximum support and falls outside the highly supported W-TBEV clade (BS 99%, PP 1.00). In contrast, LIV is well embedded within the W-TBEV clade and is placed together with Neudoerfl for the recombinant partition. Although the two most supported nodes that identify close evolutionary relationships between LIV and a strain from W-TBEV display moderate BS and PP (78% and 0.92 for the branching with Neudoerfl, 89% and 0.99 for the inclusion of LIV within W-TBEV), they are among the most significantly supported nodes in this tree. We tested the three putative recombinant fragments obtained by different methods in RDP3 and found that the shorter segment branched together with Neudoerfl with higher support values. Hence, we proceed with further characterization of this mosaic history under the assumption that crossovers occurred at nucleotides 5787 and 5991, which places the 204 nt long recombination in the highly conserved helicase domain of NS3 (subdomain 3). At the nucleotide level, the comparison of the daughter with its parental strains revealed 23 variable sites within the putative recombinant element, while the rest of the NS3 gene contained 274 variable sites. A comparison of genetic distances based on nucleotide sequence is reported in Table 2 . Phylogenetic discrepancies were assessed statistically with the AU test. For the combined (non-recombined) NS3 partition, the topological constraints forced LIV and W-TBEV into a monophyletic group with SSEV as sister taxon. Conversely, for the recombinant partition we imposed the grouping of SSEV and LIV outside the W-TBEV clade. Both alternative placements were rejected by the AU test (see Table 3 ), confirming that the different placements of LIV between partitions expresses genuine phylogenetic information rather than mere stochastic effects. Results of BEAST inference 1 are summarized in Figure 4a , showing that under the same set of priors, the posterior substitution rate (meanRate) varies up to five-fold between the different genes. The estimates distinguish the E-gene; it is both the most clearly separated and narrowly distributed, with a median of 3. In order to increase precision, dating the recombination was carried out using the recombinant region ALN2-1 and the recombination free ORF (ALN2-2). Overall, when compared to the outcome of ALN2-2, divergence times for ALN2-1 were younger and less precise, probably due to the low amount of informative data. For the sake of studying the recombination event, ideally three nodes should be scrutinized: Node ''r'' on Figure 3 estimates the actual recombination. It refers to the clustering of the LIV recombinant segment with the Neudoerfl strain which places tRE at a median of 76 (HPD: 45-160) years before origin. Accordingly, this time point is paramount and constitutes the lower bound of the estimate, but caution is advised when interpreting it as the definitive estimate. Indeed, it suffers from being inferred from a dataset comprising very short sequences. Moreover this analysis, carried out with a low level of prior enforcement, demonstrates a systematic bias towards younger antiquity. ''M'' and ''m'' are time points that refer respectively to the oldest estimate for the emergence of the major and minor parental lineages. Point ''M'' corresponds to the split of SSEV and LIV lineages, dated at a median of 1017 (HPD: 664 to 1510). Point ''m'' refers to the emergence of Neudoerfl, which corresponds to its split with the most closely related strain Hypr. However, few substitutions among the nine W-TBEV strains leads to poor phylogenetic resolution. Hence, a more conservative estimate for the onset of the minor parent would coincide with the divergence of the W-TBEV clade, placed at a median of 307 (HPD: 208 to 444) years. The next step aimed to retrieve divergence times for ''M'' and ''m'' with both increased accuracy (better locate the events in time) and increased precision (achieve narrower confidence intervals). We analyzed the largest available dataset for TBEV strains with collections dates (161 sequences); unfortunately, it only covers the Envelope glycoprotein (E) obtained from epidemiological studies. This brings on two problems: Firstly, this dataset is unable to target the actual recombination that occurred within the NS3 gene. Secondly, inference 1 has demonstrated that the E-gene presents the lowest rate of substitution among the viral genes, therefore it estimates older divergence dates than other portions of the genome. Although the former issue cannot be avoided, Esequences can nevertheless pinpoint which lineages would carry the recombinant element in a much larger tree. The latter issue can be tackled in a Bayesian framework by incorporating posterior information on divergence times derived from full-length coding sequences as prior distributions in an E-sequence analysis. The underlying rational is that by injecting information that pertains to all genes, bar E and the recombinant segment, we would be able to downplay the influence of the low substitution rate, while still combining all available evidence and avoiding circularity. We used BEAST inference 3 to calculate prior distributions for the root age and tMRCA(W-TBEV). The prior on the substitution rate was derived from the literature and not from BEAST inference 1 in order to avoid circularity in the use of data. Figure 5 depicts the outcome of BEAST inference 4, wherein the general tree summarizes the entire TBEV species evolutionary history and the enlarged chronogram gives median divergence dates within the W-TBEV, LIV, SSEV, GGEV, TSEV cluster. Dates for the principal nodes are indicated in Table 4 . Within the cluster concerned with the recombination (Figure 5b tMRCAs estimated from ALN2-2 and E_161 are consistent, suggesting that the appropriate priors have successfully counterbalanced the influence of a lower substitution rate in the E-gene. Inference 2 placed tRE at 76 (HPD: 45-160) years before origin, which localizes the recombination within the Neudoerfl lineage and after the split with the Scharl lineage. The upper (older) bound for the tRE corresponds to the youngest of the parental divergence times in the tree Figure 3 . As the estimate for M is much older than time point m, the latter can be considered as the theoretical upper bound for the observed recombination event. The lower bound leaves open the possibility that recombination occurred after the LIV 369/T2 -LI/G divergence, whereas the upper bound sets it within a clade comprising LI/G, LIV 69/T2, LI/ 261, LI/K, LI/A, LI/NOR and LI/917. Based on previous phylogenetic dispersal reconstructions [39] , the first bound places the event in Scotland, whereas the second allows a wide range of locations within the UK, after the initial virus emergence in Ireland. The possibility of recombination within tick-borne flaviviruses was raised by Twiddy et al. [14] , but given the low amount of genetic variation in this group, they pointed out that detection would prove difficult. A recent report [16] would indicate that tickborne flaviviruses have the potential to obtain and spread advantageous traits, and to remove deleterious genes [43] by homologous recombination. Alas, re-analysis of the published data did not recover that signal using a more accurate alignment method and more stringent detection conditions, but found evidences for a different event. Therefore our study shows for the first time that demonstrable recombination (that is, with sufficient statistical support and with strong phylogenetic evidences) has occurred in the mammalian tick-borne flavivirus group. Substitution rates are compound products of at least four factors: generation time, effective population size, underlying mutation rate and mutation fitness [44] . The last factor can be assessed indirectly by studying the level of selection pressure on the variable sites. The low positive selection is a well documented aspect of the mode of evolution of vector-borne RNA viruses [45] [46] [47] , which demonstrate a lack of immune-driven positive selection [46] and a very effective purifying selection [48] . Our analyses did not identify any site under positive selection. Moreover, the substitution rate analysis yielded a median of 3.3610 25 (HPD: 1.5610 25 -6.1610 25 ) subs./site/year for E, significantly lower than the previously reported rates of 1.6610 24 , within S-TBEV [49] and FE-TBEV [50] and the 8.0610 24 found W-TBEV [21] . The main difference with the previous studies can be pinpointed to our use of a relaxed clock, which was chosen because Bayes factor comparisons indicated that the strict clock performed significantly worse than relaxed models. It is known that incorrect clock assumption may lead to spurious rate estimates [51] and dating analyses effectuated under a strict clock and the same set of priors as in inference 1, yielded a mean rate estimate twice as high as under a relaxed clock and, consequently underestimated all divergence times by about a factor two (data not shown). Woelk et al. [47] suggested that the reduced positive selection in vector borne RNA viruses, results from three possible trade-offs associated with the life cycle carried in both mammalian and The third addresses the differences in immune response in the two host types: Mutations facilitating immune escape or tolerance in the first host might cause the opposite effect in the second. In the present analysis the Envelope gene displays the lowest substitution rate. As it encodes the protein responsible for the induction of protective antibody response in mammals [52] , the reported rate could be explained by the third trade-off mechanism. Accordingly, the other surface-exposed structural proteins do not interact with hosts environment as strongly as the E protein (the M protein is buried under a scaffold of E dimmers and the Capsid is covered by the Envelope) and could accumulate more mutations. We conjecture that the C-gene reaches the highest substitution rates because the Capsid is not directly involved in the replication or in the mounting of an anti-viral immune response. Rate differences for nonstructural proteins could in turn be explained by the first and second trade-offs. Finally, it has been proposed that rate of replication governs the long-term substitution rate; for instance in dsDNA viruses very high replication rates may inflate the observed substitution rate [53, 54] . Within tick-borne arboviruses, the tempo of replication is the compound of phases of high replication rates following mammalian infection and phases of low to very low rates in the arthropod environment, with the phase transition commanded by a putative termo sensitive ribo-switch [55] . It is unclear how this rate shift would impact our estimate of a global rate and, as opposed to dsDNA viruses, whether the long periods of latency could deflate the observed rates. The core idea of Bayesian approaches consists in updating our degree of belief in the truth of a hypothesis in light of new pieces of evidence pertaining to it. It is a form of incremental induction wherein the belief at the end of an investigative step is injected as a prior belief for the next step. This new belief will in turn be modified by conditionalizing upon new evidence. In order to reach credibility interval for drawing conclusions about the temporal setting of the RE, we were compelled to apply several informative priors on our final BEAST analysis. In the first step it is beneficial to place a weakly informative prior on the root [56] . This prior obtained from the literature had the effect of concentrating the probability density around its mean so it could be captured by a narrow shaped gamma distribution. In the following steps, formal probability distributions were retrieved from posteriors in the previous step and used as prior assumptions about rates and node antiquity. Although, the overlap of datasets between iterations was kept minimal, our strategy imposes to maintain some sequences across datasets in order to identify the nodes to which the derived prior should be applied. The reduction of the credibility intervals for the date parameters indicates that our approach succeeded to improve the accuracy of the time estimates by combining different lines of information coming from informative data and from the literature. Our use of relaxed-molecular clocks is the main cause for discrepancies between our estimates and previously published divergence dates. Using a strict clock Zanotto et al. found four to five times younger divergences than those presently reported (Table 5 ) [38] . Due to the rejection of the strict clock model, we argue that our approach provides a better estimate of divergence times given the data at hand, although some notes of caution should be raised. Our molecular dating could be hampered by sequencing errors, especially since sequence variation is low. In addition, the low substitution rates, could lead to inaccurate rate estimations [57] . Indeed, our estimates for individual genes approaches the limit of 1610 25 substitutions/site/year below which the temporal signal for heterochronous sampled virus begins to break down [58] and tend not to converge on the true rate when analyzed with BEAST. On that account, the least reliable time estimates are produced by the shortest alignment, which casts doubts on the tRE lower bound that was derived from 204 nt long sequences from the recombinant region. Therefore, although our dating estimates are more accurate than those relying on a poorly fitted molecular clock, more full-length genomes with a wide temporal sampling are required for a definitive assessment of divergence events in the Tick-borne encephalitis virus. Our dating locates the tRE after LIV's colonization of the British Isles. Little is known about the modes of Tick-borne encephalitis virus dispersal over long distance. Birds on a longitudinal migrating route have been found to carry infected ticks through Scandinavia [59] . However, phylogenetic analyses have not shown any clear admixture of Northern and Southern strains that would point towards bird distribution. Therefore, livestock importation from central Europe to the UK seems a more likely explanation for the footprint of past W-TBEV presence observed in the LIV genome. It is not clear why W-TBEV strains did not form stable foci in the British Isles; possibly the number of continental strains was too small to find its way from infected sheep to the small rodents that are their natural vertebrate hosts. The ecology of the tick vector, which feeds only occasionally and is relatively immobile, the rarity of infected ticks, implying that the probability of multiple strains co-infecting the same tick must be low, the short mammalian viraemia and high mortality rate, are all plausible factors that would explain that no recombination has hitherto been reported in TBEV [14] . For recombination to occur, Table 4 . Times of origin (in years before 2008) for selected clades in the phylogenetic tree of Tick-borne encephalitis virus, obtained from the BEAST inference 4 based on a large Esequences dataset. a vector can become infected with multiple strains during cofeeding in close proximity on the host skin with other ticks carrying different strains. Co-infection is then mediated via the tick saliva [60] . Alternatively, ticks can engage in multiple feeding on viraemic hosts that have been previously infected with different strains [14] . For both situations, sheep are an ideal milieu for recombination to occur when they are fed upon by several vectors carrying both W-TBEV and LIV strains. Indeed, unlike TBEV, LIV can induce a high-titer viraemia in sheep which enables tick re-infection during bloodsucking [8, 11] . Given the high similarity between strains within a sub-type, recombinant sequences in Tick-borne encephalitis virus species can probably only be detected between sub-types. Dating recombination events is challenging, due to high sequence similarity, low substation rate and condensed temporal sampling. In order to refine this analysis, additional full-length genomes of LIV strains are necessary. Now that the recombining fragment has been identified, it can readily be researched in LIV genomes. Finally, although sequencing the E-gene in order identify strains is a standard practice, the low substitution rate observed in this gene does not supply enough information for robust phylogenetic/ phylogeographic studies. We would therefore recommend to sequence, together with E, a faster evolving marker such as the Capsid-gene. Figures S1 a-b RDP3 analyses results. The x axis shows genome length in nucleotides, numbered form the start of ORFs after alignment with Neudoerfl (U27495) as reference. The y axis represents the metric used by each method for detecting recombination. Detected recombination signals appear as colored rectangles. (TIF)

Isothermal Amplification Using a Chemical Heating Device for Point-of-Care Detection of HIV-1

BACKGROUND: To date, the use of traditional nucleic acid amplification tests (NAAT) for detection of HIV-1 DNA or RNA has been restricted to laboratory settings due to time, equipment, and technical expertise requirements. The availability of a rapid NAAT with applicability for resource-limited or point-of-care (POC) settings would fill a great need in HIV diagnostics, allowing for timely diagnosis or confirmation of infection status, as well as facilitating the diagnosis of acute infection, screening and evaluation of infants born to HIV-infected mothers. Isothermal amplification methods, such as reverse-transcription, loop-mediated isothermal amplification (RT-LAMP), exhibit characteristics that are ideal for POC settings, since they are typically quicker, easier to perform, and allow for integration into low-tech, portable heating devices. METHODOLOGY/SIGNIFICANT FINDINGS: In this study, we evaluated the HIV-1 RT-LAMP assay using portable, non-instrumented nucleic acid amplification (NINA) heating devices that generate heat from the exothermic reaction of calcium oxide and water. The NINA heating devices exhibited stable temperatures throughout the amplification reaction and consistent amplification results between three separate devices and a thermalcycler. The performance of the NINA heaters was validated using whole blood specimens from HIV-1 infected patients. CONCLUSION: The RT-LAMP isothermal amplification method used in conjunction with a chemical heating device provides a portable, rapid and robust NAAT platform that has the potential to facilitate HIV-1 testing in resource-limited settings and POC.

HIV-1 diagnostic tests are held to a high standard of performance, as diagnosis has a direct impact on patient care and reduction of transmission. Despite technological advances in the field of HIV diagnostics and the high sensitivity and specificity associated with most HIV diagnostic tests that are currently available, it is estimated that approximately 20% of HIV-infected individuals living in the United States remain undiagnosed [1] . Furthermore, testing sites have reported as many as 35 to 50% of individuals with an initial positive test result will not return for a confirmatory diagnosis if follow-up laboratory testing is required [2] . Rapid HIV antibodybased tests, which can be performed with minimal training and typically provide results in under 30 minutes [3] , have facilitated HIV testing at the point-of-care and subsequently increased the numbers of individuals aware of their serostatus [4] . Rapid tests are currently a key component of HIV screening at the point-of-care (POC), significantly expanding the diagnostic capabilities of testing sites in developed countries, as well as resource-limited settings. Despite the advances made by the widespread availability of rapid tests, all antibody-based tests for the detection of HIV exhibit some limitations. HIV-specific antibody typically begins to appear around three weeks post-infection, allowing for detection by most antibody-based assays within 3-6 weeks [3, 5] . The window of time prior to or during early seroconversion may lead to false-negative test results in recently infected individuals. Additionally, accurate diagnosis of infants born to HIV-infected mothers can be challenging if based solely on antibody positivity, since vertically transferred maternal antibodies may persist for 12-18 months after birth [6, 7] . For confirmatory diagnosis of early HIV infection or infant diagnosis, nucleic acid amplification tests (NAAT) are preferred, as HIV-1 RNA can be detected as early as 10-12 days post infection and HIV-1 DNA and/or RNA are definitive indicators of active infection [5] . In their current form, however, NAAT's are not feasible for POC testing, because they are timeconsuming, expensive, and technically complicated. To date, the Aptima HIV-1 RNA assay (Gen-Probe, Inc., http://www.fda.gov/ BiologicsBloodVaccines/BloodBloodProducts/ApprovedProducts/ LicensedProductsBLAs/BloodDonorScreening/InfectiousDisease/ UCM080466) is the only FDA-approved NAAT for the diagnosis or confirmation of HIV-1 infection and it is only suitable for laboratory testing. To meet the needs of HIV-1 diagnosis at the POC, a rapid NAAT that can be performed with minimal training, limited equipment, and a relatively short turnaround time (,1 hour)is desirable [8] . The development of a rapid NAAT has proven to be especially challenging since the technology involved in simplifying the test procedure often equates to increased equipment and material costs [8] . Additionally, the reduction in technical complexity should not compromise test sensitivity and specificity. For increased applicability at the POC, an increasing number of novel isothermal amplification techniques have been developed [9] . Isothermal amplification is an attractive alternative to traditional PCR or RT-PCR since thermalcycling is not required, allowing for greater versatility in terms of heating or amplification devices. One such amplification method, termed Loop-Mediated Isothermal Amplification (LAMP) [10] , has been optimized for the detection of DNA and/or RNA (RT-LAMP) from a wide range of bacterial and viral pathogens [11, 12, 13, 14, 15, 16, 17, 18, 19] , including HIV [20, 21] . LAMP or RT-LAMP exhibits several characteristics that are ideal for integration into a rapid nucleic-acid based diagnostic test. The amplification reaction requires six primers specific for eight separate regions within the target sequence, contributing to the high specificity of the amplification method. Amplified material can typically be detected within 15-60 minutes when incubated at a constant reaction temperature of 60-65uC [22] . LAMP has also proven to be less sensitive to biological inhibitors than PCR [23, 24] , which enables direct amplification from clinical specimens, thereby eliminating the need for an additional nucleic acid extraction step. Direct amplification from plasma, whole blood, and oral fluid has previously been demonstrated for HIV-1 [20, 21, 25] . Lastly, immediate visual detection of amplified products is facilitated by the large amount of DNA that is generated by each reaction. Several groups have incorporated fluorescent detection methods into the LAMP assay for real-time or immediate naked-eye detection [15, 17, 21, 22, 26] . The simplicity and isothermal nature of the LAMP procedure opens the door for the evaluation of low-tech integrated devices or novel heating elements, which are appropriate for low-resource settings, where costly equipment and electricity cannot be obtained. In this study, the HIV-1 RT-LAMP assay was evaluated using portable, non-instrumented nucleic acid amplification (NINA) devices that generate heat from the exothermic reaction of calcium oxide and water [27, 28] . We demonstrated the temperature stability of the NINA heating devices and feasibility for POC testing of whole blood specimens from HIV-1 infected individuals. Prototype NINA heaters were designed and provided by Program for Appropriate Technology in Health (PATH, Seattle, WA), as described [27, 28] . Briefly, an amplification temperature of approximately 60uC was provided by the exothermic reaction of calcium oxide (CaO; Sigma-Aldrich, St. Louis, MO) and water. The heating devices, containing the chemical reaction, were designed using thermally insulated, stainless-steel canisters with plastic screw-top lids (Fig. 1) . The lids were modified to contain three sample wells that fit standard 200 ml PCR tubes and were filled with a proprietary phase-change material (PCM) that was used to buffer the heat derived from the exothermic reaction, thereby providing a constant temperature. Lastly, plastic caps containing foam insulation were designed to fit on the top of the canister lids. The thermal profiles of the sample wells were measured and recorded using a digital thermometer (DaqPRO 5300 Data recorder; OMEGA Engineering, Inc., Stamford, CT). DNA and RNA linearity panels were prepared to determine the sensitivity of the HIV-specific RT-LAMP assay. A DNA panel was generated from DNA extracted from the human monocytic cell line OM-10.1 [29] , using a QIAamp DNA blood mini kit (QIAGEN, Valencia, CA). Cell count was used to quantify the input DNA copy number, as a single integrated provirus is contained in each cell [29] . The extracted DNA was diluted tenfold in RNase-free water to create a linearity panel, ranging from 10 5 copies/ml to 10 3 copies/ml. An RNA linearity panel was obtained commercially (PRD801; SeraCare Life Sciences, Mil- ford, MA) and ranged from 2.9610 6 copies/ml to 8 copies/ml, as determined by Roche AMPLICOR HIV MONITOR TM v 1.5, Bayer VERSANT HIV-1 RNA bDNA 3.0 Assay, bioMerieux NucliSensH HIV-1 QT, and Abbott Real Time HIV-1 m2000 TM . RNA was extracted from the panel members using a Viral RNA mini kit (QIAGEN). Negative controls included DNA extracted from PBMC infected with HIV-2 SLRHC [30] and RNA extracted from HIV-2 NIH-Z purified virus (Advanced Biotechnologies Inc., Columbia, MD). Whole blood from HIV-1 infected individuals was collected as part of a separate, IRB-approved study [31] , or obtained commercially (SeraCare Life Sciences). All HIV-positive samples were confirmed using the following tests: Genetic Systems HIV-1/ HIV-2 plus O EIA (Bio-Rad Laboratories, Redmond, WA), GS HIV-1 Western blot (Bio-Rad Laboratories), Aptima HIV-1 RNA assay (Gen-Probe, Inc., San Diego, CA), and Amplicor HIV-1 DNA assay (Roche Diagnostics, Branchburg, NJ ). Viral and proviral loads are unknown, since the samples were tested with qualitative, nucleic acid-based assays. All clinical specimens evaluated in this study were obtained from individuals infected with subtype B HIV-1 virus. As a negative control, HIV-1 seronegative blood samples (SeraCare Life Sciences) were included in every experiment involving whole blood. A positive control included HIV-1 seronegative blood spiked with 5610 6 virus particles/ml of HIV-1 BaL (Advanced Biotechnologies Inc.). HIV-1-specific RT-LAMP primers were designed to recognize a conserved sequence within the reverse transcriptase (RT) gene. The six primers required for the RT-LAMP reaction, forward outer (F3), backward outer (B3), forward inner (FIP), backward inner (BIP), and the loop primers (LoopF and LoopB), were designed using the PrimerExplorer V4 software (Eiken Chemical Co. Ltd.; http:// primerexplorer.jp/e/). The LAMP primers and amplification cycle have been described in detail by Nagamine et al. [32] . Additional modifications included a linker sequence of four thymidines inserted between the F2 and F1c sequences of the FIP primer, as described [20] , and the addition of the fluorescent molecule HEX to the 59 end of the LoopF primer. The labeled primer, along with a quencher probe, allowed for immediate visual detection of amplified products [21] . The quencher probe consisted of the complementary sequence of the LoopF primer with Black Hole Quencher-1 (BHQ-1) added to the 39 end. The HIV-1 HXB2 sequence (GenBank accession number AF033819) was used as the reference for generating the RT-LAMP primers. The sequences of the HIV-1 RT-specific primers and quencher are listed in Table 1 . The RT-LAMP reaction was performed using the following reaction mix: 0.2 mM (final concentration) of each F3 and B3 primers, 1.6 mM of each FIP and BIP primers, 0.8 mM of each LoopF and HEX-LoopB primers, 0.8 M betaine (Sigma-Aldrich), 10 mM MgSO 4 , 1.4 mM dNTPs, 16 ThermoPol reaction buffer (New England Biolabs, Ipswich, MA), 16 U Bst DNA polymerase (New England Biolabs) and 2 U AMV reverse transcriptase (Invitrogen, Carlsbad, CA). The reaction was carried out in a total volume of 25 ml for amplification of extracted nucleic acid, 10 ml of which constituted the sample. For amplification of whole blood specimens, a 100 ml reaction volume was used to facilitate visual detection of amplified products. Whole blood was added directly into the reaction at a total volume of 40 ml, following a 1:4 dilution with red blood cell lysis buffer (2.5 mM KHCO 3 , 37.5 mM NH 4 Cl, and 0.025 mM EDTA), as previously described [21] . The reaction mixture was incubated at 60uC for 60 minutes, using a GeneAmpH PCR System (Applied Biosystems, Foster City, CA) or the NINA heaters. For reactions amplified in the thermalcylcer, an additional two minute heating step of 80uC was added at the end of the amplification cycle to terminate the reaction. The reaction tubes were evaluated for the presence of amplification, following addition of the quencher probe at a 2:1 ratio of quencher to labeled-primer, as previously described [21] . Amplification was determined visually by observing fluorescence in the reaction tubes, using the UV lamp from a ChemiDoc XRS system (Bio-Rad Laboratories, Hercules, CA). Amplification was confirmed by electrophoresis using a 1.2% agarose gel containing SYBRH Safe gel stain (Invitrogen), which was subsequently visualized using the ChemiDoc XRS system. To compare temperature and amplification consistency, three NINA heaters were tested in parallel. The heating reaction was initiated by adding 18 g of CaO to each NINA canister, followed by 6 ml of water. The lid of each canister was then sealed to contain the exothermic reaction. After adding 200 ml of water to each of the sample wells, temperature recording was initiated. Reaction tubes were added to the sample wells once each reaction chamber reached a temperature of 58.5uC. For all samples incubated in the NINA heater, 15 ml of mineral oil was added to the reaction tube during the reaction mix preparation. The samples were incubated in the heaters for a total of 60 minutes. All reactions were carried out in a temperature-controlled laboratory with an ambient temperature of 28uC, unless otherwise stated. Following the amplification reaction, the samples were incubated for two minutes in a heat block set to 80uC. After each amplification cycle, the temperature profile of each device was analyzed by calculating the temperature mean, standard deviation, median, minimum, and maximum from the data provided by the DaqPRO 5300. The stability of the NINA heaters at extreme low and high temperatures was evaluated by placing the canisters in a refrigerator set to 4uC or a 37uC incubator during the length of the amplification reaction. The temperature profiles were recorded and compared to those of reactions that occurred at the laboratory room temperature of 28uC. To determine the sensitivity of RT-LAMP reaction using RTspecific primers, DNA and RNA linearity panels were tested in a thermalcycler. The limit of detection for HIV-1 DNA was 10 copies/reaction. For the RNA linearity panel, the sample containing 1700 copies/reaction was detected in all of the three replicates, while the sample containing 140 copies/reaction was detected in three out of five replicates (60%). For both DNA and RNA linearity panels, the two samples nearest the limit of detection were chosen to further evaluate the performance consistency between the thermalcycler and NINA heaters. In terms of positivity, the amplification results were consistent between all three heaters and the thermalcycler ( Table 2) . Since the RT-LAMP assay requires a constant temperature of 60uC for the length of the amplification reaction, the temperature profiles of the sample wells were compared over the course of the incubation and between all three NINA heaters. A representative temperature profile is displayed in Figure 2 , showing a steady reaction temperature at or close to 60uC for length of amplification reaction. During the 60 minute incubation, the average temperature for each device was 60.2, 59.8, and 59.7 (Table 3 ). The minimum temperature achieved during the reaction reflects the fact that the temperature of the sample port dropped temporarily after the sample tubes are added to the device, as shown in Figure 2 . The maximum temperature of the devices deviated from the desired reaction temperature of 60uC by less than one degree. The ability of the NINA heaters to maintain a steady reaction temperature in a wide range of ambient temperatures is essential for POC testing, whether referring to an air-conditioned laboratory or high-temperature field site. To evaluate the performance of the NINA heaters at extreme low or high temperatures, the canisters were placed in a 4uC refrigerator or a 37uC incubator for the length of the amplification reaction. The limit of detection for the DNA and RNA linearity panels was similar to the results obtained in our temperature-controlled laboratory (28uC; Table 2 ). The greatest degree of temperature variation of the sample wells was observed at the ambient temperature of 4uC ( Table 3 ). The average temperature was approximately two degrees lower than the desired reaction temperature of 60uC. Additionally, the temperature of the devices tended to decline from their steady state during the last 20 minutes of the reaction (data not shown). The temperature profiles at the ambient temperature of 37uC, however, were similar to those at 28uC. Whole blood samples from HIV-1 infected individuals were added directly into the RT-LAMP reaction and tested in the NINA heaters. Positivity of the clinical specimens was consistent between the thermalcycler and devices (Table 4 ). Amplification consistency was most evident with two of the patient samples (patient #4 and #5) that were only positive in one of the three replicates, regardless of the heating device that was used. All HIVnegative blood samples, included in each reaction, were negative (data not shown). A representative experiment using the NINA heaters is displayed in Figure 3 , showing detection by agarose gel and visual identification of fluorescence in the reaction tubes. In this study, we demonstrate the performance of portable, inexpensive, non-instrumented nucleic acid (NINA) heaters for amplification of HIV-1 using RT-LAMP. The isothermal amplification reaction coupled with a device that generates heat from an exothermic chemical reaction, as opposed to grid electricity or battery power, comprises a point-of-care NAAT that is practical for use in resource-limited settings. The heating devices require minimal training and technical expertise to operate and take approximately 10-15 minutes to reach a reaction temperature of 60uC once the chemical reaction has been initiated [27, 28] . Furthermore, the temperature of the sample wells remain relatively stable at the desired reaction temperature of 60uC throughout the amplification reaction, as demonstrated by the heating profiles and the consistency in amplification between the devices and thermalcycler. Since point-of-care testing may refer to an air-conditioned laboratory or a field site with high temperatures and humidity, the stability of the temperature generated by the heating devices must be reliable. Though the temperature profiles at a representative cold temperature of 4uC indicated a loss in reaction temperature towards the end of the 60 minute incubation, the temperature fluctuations were not significant enough to affect the amplification reaction. Regardless, this thermal effect could be mitigated with small modifications to the device to reduce heat loss at lower temperatures. It should be possible to extend the temperature range of the NINA heaters to 4uC and below by either adding a larger quantity of heating mixture, better insulation, or both. Of greater concern is the performance of the NINA heaters in hightemperature field sites, where temperature control is not an option. We demonstrate no difference in the temperature stability of the NINA heaters and amplification consistency at an ambient temperature of 37uC as compared to our temperature-controlled laboratory. For increased applicability for use at the POC, several modifications can be made to the NINA heaters. The prototype devices evaluated in this study contained only three sample wells; however, up to 16 sample wells can be added to the lid of the insulated canisters for a larger testing volume. In this study, samples were removed from the NINA heaters after the amplification reaction and heated for an additional two minutes in an 80uC heat block to terminate the reaction. While the additional heating step is not necessary to observe the amplified products from extracted nucleic acid, the short, high-temperature incubation facilitates the visual observation of the fluorescent label in the whole blood samples. Modifications may be made to the whole blood sample preparation method to eliminate the need for the heating step. Alternatively, a second temperature-moderating compartment can be added to the alternate end of the NINA canisters, so the samples can be removed from the amplification compartment and reinserted into the 80uC compartment. Lastly, the DaqPRO data recorder was used in this study for validation purposes only and would not be necessary for the final POC product. The feasibility of using LAMP as a diagnostic method in resource-limited settings has been demonstrated for tuberculosis [33] . To reduce hands-on time and preparation error, the authors describe the use of reaction tubes pre-prepared with lyophilized reaction mix. For POC use, limited sample manipulation and reagent preparation is desired and, therefore, it is anticipated that the test procedure of the end product will include reconstituting the amplification reagents in water and adding the sample directly into the reaction tube. We demonstrate the use of the NINA heaters for amplification directly from whole blood specimens, eliminating the need for a time-consuming, nucleic acid extraction procedure and reducing the volume of sample needed for the amplification reaction. A total volume of 10 ml of whole blood was added to each reaction tube, which can easily be obtained by finger-stick in settings where venipuncture is not feasible. Additionally, our fluorescent detection method enables immediate visualization of amplified products in the absence of specialized equipment. To avoid cross-contamination of amplified material, it is preferred that the reaction tubes remain closed post-amplification. Future modifications will include optimizing the labeledprimer/quencher sequences so that all components can be added into the reaction mix prior to amplification. Due to availability, the Bio-Rad ChemiDoc system was used as the UV source in this study; however, an inexpensive keychain light would be more suitable for naked-eye detection at the POC. For sensitive and specific detection of diverse HIV-1 isolates, including non-B subtypes, identification of the optimal primer set/sets is a key step in the development of the RT-LAMP assay. Although all experiments performed in this study involved subtype B standards and specimens, ongoing research involves the continued development and optimization of RT-LAMP primers based on regions of the HIV-1 genome that are conserved among diverse subtypes. Future studies will include large-scale evaluation of clinical specimens with the optimized RT-LAMP assay and NINA device. In summary, the RT-LAMP isothermal amplification method used in conjunction with a simplified, chemical heating device exhibits characteristics that are ideal for a rapid NAAT for POC testing. The simplified, portable assay has the potential to fill an important gap in HIV-1 diagnostics, providing immediate knowledge or confirmation of HIV-1 infection status at the POC.

Design and management of an orthopaedic bone bank in the Netherlands

The design and management of an orthopaedic bone bank is a complex process in which medical organisation and legislation intertwine. Neither in the Netherlands, nor in any other European country, there are official guidelines for the organisation and management of an orthopaedic bone bank. In the Netherlands, the recently modified ‘law of security and quality for using human materials’ (WVKL) dictates requirements for technical and organisational aspects for the use of human tissue and cells. The bone bank procedures include a thorough questionnaire for donor selection, extensive serological, bacteriological and histopathological examination, as well as standard procedures for registration, processing, preservation, storage and distribution of bone allografts. This article describes the organisation of an accredited bone bank and can be used as a proposition for an official guideline or can be useful as an example for other orthopaedic bone banks in Europe.

For reconstruction of bone defects, donor bone from orthopaedic bone banks is often necessary. Bone grafts are used in bone defects that arise from trauma (Friedlaender 1987) , infection, resection of bone tumours or it is used in spinal fusion (Raizman et al. 2009 ) and as impaction grafting in revision of total joint arthroplasty (Slooff et al. 1996) . Autologous bone is preferred because of its osteoconductive and osteoinducive activity, but it is often not sufficiently available and it repeatedly involves donor site morbidity (Summers and Eisenstein 1989) . Allogenic bone exclusively has osteoconductive activity; it serves as a frame against which newly formed bone gets deposited (Elves and Pratt 1975; Urist 1953) . Allogenic bone is provided by an orthopaedic bone bank. In Leiden, the Netherlands, the Dutch Bone Bank Foundation (NBF) was founded in 1988 (Veen et al. 1990) . In this central bone bank, bone-and tendon transplant material of deceased donor patients is stored (Veen et al. 1991) . When needed, hospitals may order such material from the NBF. A number of hospitals manage their own bone banks, such as the VU university medical center in Amsterdam, where an orthopaedic bone bank has been established. This VUmc orthopaedic bone bank contains only femoral heads of suitable patients who underwent total hip replacement surgery. The advantage of possessing a bone bank is that the hospital always has its own supply of donor bone material; this may also be a financially viable strategy for hospitals carrying out many procedures for which donor bone material is required. Up to now, nationally recognized guidelines for maintenance and management of bone banks do not yet exist in the Netherlands. In this paper we describe the VUmc orthopaedic bone bank procedure, which recently gained official approval and recognition and could serve as a potential outline for other hospitals. From October 2008, the Ministry of Health, Welfare and Sport (VWS) officially recognized the orthopaedic bone bank of the VUmc (Inspectie voor de Gezondheidszorg 2008). A biannual inspection is performed by the Health Care Inspectorate (IGZ) as a requirement to maintain this recognition. The bone bank procedure has to meet the requirements of the adjusted 'law of security and quality for using human materials' (Wet Veiligheid en Kwaliteit Lichaamsmateriaal 2003) . This law became effective from mid-2007 in the Netherlands as a result of European guidelines 2004/23/EC and 2006/86/EC. These guidelines state the technical requirements for coding, processing, preserving, storing, and distributing of human tissue and cells. Human tissue should be traceable, and serious side effects and incidents with human tissue and cells should be reported. The procedure of our orthopaedic bone bank is based on guidelines of The American Association of Tissue Banks (AATB), and the criteria of the Council for Blood Transfusion of the Netherlands Red Cross (Richtlijn Bloedtransfusie 2004), together with the recently merged Netherlands Bone Bank Foundation (NBF) and Bio Implant Services (BIS) (NBF-BIS Foundation 2010). Previously, we followed the guidelines of the European Association of Musculoskeletal Transplantation. As a result of diverging European legislations, this organization has been discontinued as an European umbrella organization; currently only national associations prevail. To date, the Netherlands has not possessed such an association; consequently there is no national guideline with regards to maintenance and management of an orthopaedic bone bank. The bone bank procedure should be carefully described in an extensive protocol consisting of the following five components: organization, donor selection, documentation, storage and processing, and implementation. The HOD (Head of Department) of the Department of Orthopaedics and the bone bank administrator compose this protocol. In an organization chart we describe the responsibilities of different stakeholders: the HOD of the Department of Orthopaedics, a bone bank administrator, a theatre nurse, a medical microbiologist, an anatomic pathologist, a clinical chemical analyst, a haematological laboratory technician, and a trainer. The HOD is the main responsible of the bone bank, whereas the bone bank administrator is responsible for the daily management. The knowledge and skills concerning surgical techniques and clinical hygiene are guaranteed by the orthopaedic surgeon and theatre nurse. The bone bank administrators' responsibilities include administration as well as storage and allocation of donor bone. Additionally, the administrator takes care of the maintenance and cleaning of the storage facilities (freezers, etc.), and verifies the registration forms of femoral heads meeting the requirements for storage in the bone bank. Both the bone bank administrator and the trainer are responsible for training of bone bank employees. Apart from an orientation module for new employees, the training program consists of regular refresher courses for all members of the staff, in order to keep the knowledge of the procedures updated. Preceding the hip replacement procedure, the attending orthopaedic surgeon requests the patient for his permission to store any removed tissue for donation. It concerns patients whose femoral head grafts will be retrieved in order to be replaced by a total hip prosthesis. Corticospongious bone tissue can not be sufficiently obtained in knee or shoulder arthroplasty; therefore patients undergoing such procedures cannot be taken into consideration for donor bone tissue donation. The attending orthopaedic surgeon informs the patient both orally and in written. In case the patient grants permission he or she signs the consent forms, and fills out a standard survey (see Table 1 ). The orthopaedic surgeon now decides whether the patient is suitable for being a donor; he uses general and specific exclusion criteria (see Tables 2, 3 ). All criteria must be met; if not, exclusion necessarily follows. The orthopaedic surgeon examines the patient thoroughly: blood samples are collected to determine blood type, Rh-factor and erythrocyte sedimentation rate (ESR) ( Tables 4, 5) . During surgery, bacterial culture swab samples from hip ligament are collected and a biopsy of corticospongious bone is sent off for histopathological analysis. Serological screening for infectious diseases is performed 6 months after surgery. Once all requirements are met (Tables 1, 2 In the past 3 months, did you have any vaccination or inoculation, or have you been injected with narcotic drugs? In the past 6 months, did you have a malaria attack or did you use anti-malarial medication? Have you ever been infected with a sexually transmitted disease? Have you ever been diagnosed with jaundice or liver illness? In the past 6 months, have you been in contact with patients diagnosed with jaundice/hepatitis? Are you a sexual partner of an individual for which any of the abovementioned questions can be answered with 'yes'? Have you been actively involved in prostitution after 1977, or have you been a sexual partner of a person involved in prostitution in the past 6 months? Have you ever been diagnosed with a haematological disease or any malignant disorder? Have you ever been treated for diabetes mellitus? Have you ever been treated for chronic brain-or neurological diseases? Have you ever received radiation therapy? Have you ever been diagnosed with rheumatoid arthritis? Have you ever been diagnosed with tuberculosis? Have you ever been diagnosed with any disease, other than the abovementioned? Have you ever received hormonal treatment? Do you use any prescribed medication? Have you ever used any narcotic drugs? Have you recently been exposed to hazardous or toxic materials? If yes, please specify. What is your alcohol consumption per week? Have you recently been in surgery? If so, when? Did you receive blood from a blood transfusion? In the past 14 days, have you been traveling through or staying in a region exposed to a SARS epidemic, or have you been in contact with patients infected with SARS? In the past 6 months, have you tattooed yourself or did you get a piercing? Cell Tissue Bank (2012) 13:63-69 Documentation Accurate documentation and coding are a necessity for a well functioning bone bank. A unique registration code is allocated to each femoral head. Only the bone bank administrator is able to trace the donor based on this code. Of every registered femoral head, a file, containing the consent forms and results of ESR, bacteriological and histopathological examination, is kept updated. Other relevant data, such as the size of the femoral head and the allocation date are also documented and stored in this file. When the file is completed (which takes at least 6 months due to the serological examination), and no abnormalities are recorded, both bone bank administrator and the responsible orthopaedic surgeon sign the forms. The femoral head is now available for transplantation. In case a file cannot be completed in full, or any Individuals who stayed in a SARS epidemic area or individuals who had face-to-face contact with a SARS patient abnormal values are recorded, the femoral head will be destroyed according to hospitals' protocol. The femoral head is surgically removed under sterilized conditions. The ligament and synovial tissue are cultured on aerobic and anaerobic bacteria. In order to exclude malignancies, auto-immune processes, or infections, a biopsy of 1 cm 3 corticospongious bone and ligament is collected for histopathological examination. After determining its size, the femoral head is wrapped in a sterile plastic bag and in three layers parcelled in sterile packing material, labelled and stored in the freezer within 30 min. The freezer has a temperature of -80°C, and has a continuous temperature registration device installed. Should the temperature fall outside the acceptable range of -90 and -70°C, an alarm system gives off a warning signal to the Technical Service, guaranteeing a 24-h security against temperature-induced damage to the tissue. A nitrogen tank is fitted onto the freezer, as a backup cooling mechanism in case of mechanical breakdown of the freezer. In deep frozen condition, the allogenic bone tissue can be preserved for a maximum of 5 years. The temperature data is stored and managed by the bone bank administrator for a period of at least 5 years. If during surgery a surgeon decides to use a femoral head as an allograft, a femoral head from the freezer together with its documents are handed over to the orthopaedic surgeon and surgery team. The orthopaedic surgeon and theatre nurse verify the file and expiration date of the femoral head. The femoral head is thawed in physiological saline; after being defrosted the theatre nurse takes a bacterial culture swab. The hospital or care institution warrant fulfilment of the traceability requirements, which implies storing the file of the femoral head and records of the receiving patient for 30 years post implantation. Macewen first describes the use of allogenic human bone tissue in 1881 (Macewen 1881). From that year onwards, the use of allogenic bone transplantation has been increasingly applied and is nowadays a standard orthopaedic procedure (Tomford et al. 1987) . However, much has changed in the past decade: donor selection, clinical hygiene, storage and processing, allocation, implantation, and documentation are bound to strict rules. Donor selection takes place by a thorough broad survey and supplementary physical examination. The survey should be regularly revised to comprise the latest knowledge and developments, and as a response to the spread of new infectious diseases. For instance, after the outbreak of the SARS epidemic in certain countries, a question was added to the survey; prospective donors were asked whether they had been in SARS-infected regions or whether they had been in contact with an infected person. It is not unthinkable that newly arising infectious diseases will be included in the survey and henceforth become exclusion criteria. Laboratory examination is performed before surgery, including ESR determination. Elevated values are often encountered, often without clinical implications. However, the exclusion criteria are strictly enforced. Preoperatively, blood type and Rh-factor are determined. The Rh-factor only becomes an issue if the receiving patient is a young woman. Additionally, the donor is serologically examined, to exclude possible transmission of infections, such as HIV and hepatitis (Strong et al. 1996; Shutkin 1954; Patel and Trampuz 2004; Karcher 1997) . The serological tests are carried out at least 6 months after surgery to avoid type II error (false negative): if the patient was infected at the time of surgery, the test will provide conclusive evidence of this. Normally, the donor will not be informed of the results, unless specifically requested by the donor. In addition to existing national and international guidelines and procedures of the AATB, NBF-BIS, and former EAMST, histopathological examination is added to the current bone bank procedure. Previous studies have found pathological abnormalities in 8%, and B-cell lymphomas in 2,2% of donor femoral heads (Palmer et al. 1999; Zwitser et al. 2009; Sugihara et al. 1999) . Though transmission of malign cells following transplantation has never been shown, femoral heads with histopathological abnormalities are excluded from allogenic bone transplantation. Apart from an orthopaedic bone bank, many hospitals often have more organ banks, such as a haematological bone marrow bank for stem cell transplantation or the IVF laboratory of the department of Gynaecology and Obstetrics. Therefore, in every hospital a central system for tissue vigilance should be established and all departments that require cells and/or tissue for transplantation purposes should participate. Thus, highest cell and tissue security levels within the hospital are warranted by controlling the complete transplantation chain from donation to transplantation, including appropriate reporting and follow-up of incidents and side effects. The design and management of an orthopaedic bone bank as an organ bank is a very complex process in which aspects of medical organization and legislation intertwine. In this paper we describe our bone bank procedure, which is approved and recognized by the Ministry of Health, Welfare and Sport (VWS) of the Netherlands. This procedure could serve as a provisional Dutch protocol for the setup and maintenance of other orthopaedic bone banks.

Dectin-1 and DC-SIGN Polymorphisms Associated with Invasive Pulmonary Aspergillosis Infection

The recognition of pathogen-derived structures by C-type lectins and the chemotactic activity mediated by the CCL2/CCR2 axis are critical steps in determining the host immune response to fungi. The present study was designed to investigate whether the presence of single nucleotide polymorphisms (SNPs) within DC-SIGN, Dectin-1, Dectin-2, CCL2 and CCR2 genes influence the risk of developing Invasive Pulmonary Aspergillosis (IPA). Twenty-seven SNPs were selected using a hybrid functional/tagging approach and genotyped in 182 haematological patients, fifty-seven of them diagnosed with proven or probable IPA according to the 2008 EORTC/MSG criteria. Association analysis revealed that carriers of the Dectin-1 (rs3901533 T/T) and Dectin-1 (rs7309123 G/G) genotypes and DC-SIGN (rs4804800 G), DC-SIGN (rs11465384 T), DC-SIGN (7248637 A) and DC-SIGN (7252229 C) alleles had a significantly increased risk of IPA infection (OR = 5.59 95%CI 1.37–22.77; OR = 4.91 95%CI 1.52–15.89; OR = 2.75 95%CI 1.27–5.95; OR = 2.70 95%CI 1.24–5.90; OR = 2.39 95%CI 1.09–5.22 and OR = 2.05 95%CI 1.00–4.22, respectively). There was also a significantly increased frequency of galactomannan positivity among patients carrying the Dectin-1 (rs3901533_T) allele and Dectin-1 (rs7309123_G/G) genotype. In addition, healthy individuals with this latter genotype showed a significantly decreased level of Dectin-1 mRNA expression compared to C-allele carriers, suggesting a role of the Dectin-1 (rs7309123) polymorphism in determining the levels of Dectin-1 and, consequently, the level of susceptibility to IPA infection. SNP-SNP interaction (epistasis) analysis revealed significant interactions models including SNPs in Dectin-1, Dectin-2, CCL2 and CCR2 genes, with synergistic genetic effects. Although these results need to be further validated in larger cohorts, they suggest that Dectin-1, DC-SIGN, Dectin-2, CCL2 and CCR2 genetic variants influence the risk of IPA infection and might be useful in developing a risk-adapted prophylaxis.

Invasive Pulmonary Aspergillosis (IPA) is a life-threatening infection caused mainly by Aspergillus fumigatus, an opportunistic fungal pathogen that frequently colonizes respiratory tracts and rapidly spreads to blood vessels and tissues [1] . The incidence of IPA infection has increased in the last years due to the use of immunosuppressive and immunomodulatory drugs and is still causing significant morbidity and mortality worldwide, especially in immunosuppressed and hematopoietic stem cell transplantation (HSCT) patients [2] . Early recognition of A. fumigatus by myeloid leukocytes is a crucial for down-stream immune response and conidia clearance and, therefore, in the development and progression of IPA infection [3] . Myeloid leukocytes are activated by patternrecognition receptors (PRRs), which directly recognize fungal cell wall structures and pathogen-associated molecular patterns (PAMPs) [4] . Among the PRRs expressed in neutrophils, pulmonary macrophages and dendritic cells, C-type lectin family members such as DC-SIGN (Dendritic Cell-Specific Intercellular adhesion molecule-3-Grabbing Non-integrin), Dectin-1 (Dendritic cell-associated C-type lectin-1) and Dectin-2 (Dendritic cellassociated C-type lectin-2) seem to be important in the effectiveness of innate immune response against A. fumigatus [5, 6] . Recent studies have reported that the C-type lectins mediate A. Fumigatus recognition, capture and internalization [7] and that their binding to fungal cell wall carbohydrates is highly specific and selective [8] . DC-SIGN and Dectin-1 recognize b-(1-3)-glucans and galactomannans (GM) in the cell wall of A. fumigatus [7, 9] while Dectin-2 interacts with a-mannans [10] . However, numerous studies have shown that C-type lectins are not only involved in the recognition of fungal pathogens but also in the induction of antifungal Th1 and Th17 immune responses [5, 11] . Although the mechanisms underlying DC-SIGN-mediated immune response are still highly speculative, it has been suggested that DC-SIGN promotes dendritic cell (DC) migration and T-cell activation through the ICAM-3 binding [12, 13] and modulates TLR signalling by targeting Raf-1, which regulates NFkB p65 activation and, consequently the production of pro-inflammatory cytokines [14] . Likewise, Dectin-1 and Dectin-2 induce Sykdependent canonical and noncanonical NFkB pathways [15, 16] promoting the production of some pro-inflammatory cytokines (IL1a, IL1b, IL12 and IL8), chemokines (CCL2/MCP-1, CCL3/ MIP-1, CXCL2/MIP-2 and CXCL10) [17] and the IL17-mediated neutrophil recruitment [18] . It has also been described that Dectin-1 and Dectin-2 are able to work in collaboration with TLRs (mainly TLR2, TLR4 and TLR6) modulating immune responses through the production of IL6, IL12p70 and TNF [3, 16] . These findings support the hypothesis that C-type lectins, cytokines or chemokines are important mediators during infection with A. fumigatus. Several polymorphisms in human PRRs [19] [20] [21] [22] [23] [24] as well as in some cytokines [25] [26] [27] , chemokines [28] and their receptors [29, 30] have hitherto been associated with an increased risk of invasive aspergillosis in susceptible hosts. Based on these observations, the objective of the present study was to investigate the role of tagging and potentially functional single-nucleotide polymorphisms (SNPs) located within the DC-SIGN, Dectin-1, Dectin-2, MCP-1/CCL2 and CCR2 genes on IPA susceptibility. All participants enrolled were Caucasian and recruited in the University Hospital Virgen de las Nieves (Granada, Spain) or in the University Hospital of Salamanca (Salamanca, Spain). All determinations and genetic analyses in hematological patients were performed with fully informed written consent, and anonymity of the data was guaranteed. The study protocol was approved by the Ethical Review Committee of Virgen de las Nieves University Hospital, Granada, Spain. The population included 182 hematological patients recruited between January 2004 and January 2011. All hematological patients in this study received a prolonged chemotherapy treatment or underwent HSCT and were therefore considered susceptible to develop IPA infection. Demographic information and clinical data were obtained by detailed review of hospital records. Data were gathered on: site of infection; host factor criteria (severe neutropenia for .10 days, persistent fever for .96 h refractory to appropriate broad-spectrum antibacterial treatment, signs and symptoms indicating graft-versus-host disease [GVHD], corticoid therapy [.0.3 mg/kg per day], and invasive fungal infection during a previous episode of neutropenia), microbiological criteria (positive result for Aspergillus antigen in $2 consecutive blood samples when considering an index of 0.5 or in only 1 sample when the index was higher than 0.8), and clinical criteria of lower respiratory tract infection [major criteria: any of the following new infiltrates on computed tomography (CT) imaging: halo sign, aircrescent sign, or cavity within area of consolidation; minor criteria: cough, thoracic pain, hemoptysis, pathologic pulmonary sound, and radiological evidence suggestive of invasive infection]. Laboratory data were also recorded. Proven and probable IPA was diagnosed based on the updated criteria (2008) reported by the European Organization for Research and Treatment of Cancer/Invasive Fungal Infections Cooperative Group (EORTC/ IFICG) [31] . Serum GM detection has been shown to be a useful test for the early diagnosis and follow-up of IPA and is now included in IPA diagnosis criteria [31] . In the present study, serum GM antigen was determined twice weekly during the hospital stay and at each outpatient visit until the end of their immunosuppressant or chemotherapeutic treatment. Serum GM concentrations were determined by Platelia Aspergillus ELISA kit (Bio-Rad, Marnes-la-Coquette, France) according to the manufacturer's instructions. This commercial kit has proven to offer good sensitivity to detect GM [32] , and GM concentration was found to correlate with the fungal tissue burden [33, 34] . A test sample was classified as positive when the optical density ratio was $0.5 in two consecutive positive samples or .0.8 in a unique serum sample. A careful review of concomitant treatments (piperacillin-tazobactam or amoxicillin-clavulonic acid) in each patient was necessary to detect possible false-positive GM determinations. Likewise, tests were performed on the same day to avoid sample contamination and to ensure accuracy of results. Twenty-seven tagging/functional SNPs within DC-SIGN, Dectin-1, Dectin-2, CCL2 and CCR2 were selected to genotype the entire panel of individuals ( Table 1 ). The aim of the SNP tagging was to identify a set of SNPs that efficiently tags all the known SNPs while the functional approach was used to determine the net impact of potentially functional variants within DC-SIGN, Dectin-1, Dectin-2, CCL2 and CCR2 genes on IPA risk. Tagging SNPs were selected using Haploview Tagger program (http://www.broad.mit.edu/ mpg/haploview/; http://www.broad.mit.edu/mpg/tagger/) and a pairwise tagging with a minimum r2 of 0.8. In this selection we forced the inclusion of the DC-SIGN rs4804803 , CCL2 rs1024610 and CCL2 rs1024611 polymorphisms as their functionality has been suggested [35] [36] [37] . Genomic DNA was extracted from peripheral blood mononuclear cells (PBMCs) Qiagen Mini Kit (Qiagen, Valencia, CA, USA). Genotyping of DC-SIGN, Dectin-1, Dectin-2, CCL2 and CCR2 polymorphisms was carried out using KASPar assays (KBiosciences, Hoddesdon, Hertfordshire, UK) in a 384well plate format (Applied Biosystems, Foster City, CA, USA) where hematological patient samples were randomly distributed. KASPar reactions were performed using KASPar assay mix (containing probes) and KASPar kit containing 26 Reaction Mix and MgCl2 (50 mM). PCR conditions were: denaturation at 94uC for 15 min, 10 cycles of denaturation at 94uC for 10 sec, annealing at 57uC for 5 sec and elongation at 72uC for 10 sec and 20 cycles of denaturation at 94uC for 10 sec, annealing at 57uC for 20 sec and elongation at 72uC for 40 sec. Recycling conditions were 94uC for 10 sec, annealing and elongation at 60uC for 60 sec. PCR products were analyzed with the ABI Prism 7900HT detection system using the SDS 2.4 software (Applied Biosystems). For internal quality control, about 5% of samples were randomly selected and included as duplicate. Concordance between the original and the duplicate samples for the 27 SNPs analyzed was $99.5%. Call rates for all SNPs were $97.8% with the exception of the Dectin-1 rs11053599 SNP with a call rate of 94.5%. The Hardy-Weinberg Equilibrium (HWE) tests were performed in the control group (non-IPA patients) by a standard observedexpected chi-square (x 2 ) test at each polymorphic site (http://ihg2. helmholtz-muenchen.de/cgi-bin/hw/hwa1.pl). Unconditional logistic regression was used to assess the main effects of the genetic polymorphisms on IPA risk using co-dominant, dominant and recessive inheritance models. For each SNP, the more common allele in the control group was assigned as the reference category. All analyses were adjusted for age, gender, hematological malignancy and established risk factors for IPA infection (HSCT, neutropenia, GVHD and corticoid therapy use) and were conducted using the statistical software SSPS (version 14.0, SPSS Inc., Chicago, USA). All tests were considered to be statistically significant with a p value of ,0.05. Adjustment for multiple testing was carried out following a conservative threshold for statistical significance, based on a revised version of the Bonferroni method: we calculated for each gene the ''number of effective independent variables'' (M eff ) by use of the SNP Spectral Decomposition approach (http://gump.qimr. edu.au/general/daleN/SNPSpDsuperlite/) [38] . We obtained a study-wise M eff value by adding up the gene M eff 's. SNPtool (http://www.dkfz.de/de/molgen_epidemiology/tools/ SNPtool.html) [39] and the Haploview v4.2 software were used for LD blocks reconstruction and haplotype association statistics. Block structures were determined according to the method of Gabriel et al. [40] . We used the Web-based tool FastSNP [41] available at http:// fastsnp.ibms.sinica.edu.tw for predicting the functional significance of the SNPs associated with IPA infection. FastSNP utilizes information from another online program PolyPhen (http://www. bork.embl-heidelberg.de/PolyPhen/) and from four different web resources (TFSearch, ESEfinder, Rescue-ESE and FAS-ESS) to determine whether SNPs are located at exonic splicing regulatory sites, cause a non-conservative amino acid change or whether they alter the transcription factor-binding site of a gene (for instance, acting as intronic enhancer). The score was given by this tool on the basis of levels of risk with a ranking of 0 (no effect), 1 (very low), 2 (low), 3 (medium), 4 (high), or 5 (very high). Whole blood samples from 21 healthy donors were collected into PAXGENE RNA tubes and stored at 280uC until use. Total RNAs were extracted using a PAXGENE Blood RNA Isolation Kit (PreAnalytiX) and reverse transcribed to cDNA using QuantiTect Reverse Transcription Kit (catalog number: 205311; Qiagen). Real-time quantitative PCR was carried out using an ABI PRISMH 7500 HT Sequence Detection System (Applied Biosystems) according to manufacturer's instructions. Briefly, 2 ml of the cDNA were loaded in a PCR reaction containing 12.5 ml of 26 QuantiTect SYBR Green PCR Master Mix with an appropriate concentration of MgCl 2 , 2.5 ml of primers (Hs_CLE-C7A_1_SG QuantiTect Primer Assay, catalog number: QT00024059; Geneglobe, Qiagen) and 8 ml of RNase-free water. PCR cycling conditions were as follows: 95uC for 15 minutes, followed by 40 cycles of denaturation at 95uC for 15 seconds combined with annealing at 60uC for 30 seconds and extension at 72uC for 30 seconds. All samples were run in duplicate. Relative quantification of Dectin-1 mRNA expression was calculated with the 2 2DDCt method. We obtained the fold changes in gene expression normalized to an internal control gene (GAPDH, Hs_GAPDH_2_SG QuantiTect Primer Assay, Catalog number: QT01192646; Qiagen) and relative to one calibrator (First- We also analyzed high-order interactions between SNPs using the multifactor dimensionality reduction (MDR) constructive induction algorithm. A detailed explanation on the MDR method has been described elsewhere [42, 43] . SNPs in LD (cut-off of r 2 ,0.8) were excluded from the MDR analysis. Cross-validation and permutation testing were used to identify the best models. All possible two-, three-and four-way SNP interactions were tested using 100-fold cross-validation and the exhaustive search. The model with the highest testing balanced accuracy (TA) and cross validation consistency (CVC) was selected as ''best model''. Statistical significance was evaluated using a 1.000-fold permutation test to compare observed testing balanced accuracies with those expected under the null hypothesis of no association (using the MDR permutation testing module 0.4.9 alpha). MDR results were considered statistically significant at the 0.05 level. Finally, interactions were visualized by constructing an interaction dendrogram according to the method described by Moore et al [43] . MDR software and MDR permutation testing module are open-source and freely available from http://www.epistasis.org. Population characteristics are described in Table 2 . Compared to non-IPA, IPA cases were more likely to have cough and pathologic pulmonary sound (p,0.001 and 0.011, respectively) and presented more often pathological chest radiographies and CT scans (p,0.001). Established risk factors for IPA infection such as HSCT, neutropenia, GVHD and corticoid therapy use were homogenously distributed between IPA and non-IPA patient groups. Fifty-seven patients were diagnosed with proven or probable IPA infection according to the revised EORTC/MSG criteria (2008). The association of risk of IPA infection with the individual 27 SNPs in the C-type lectin and chemokine genes is shown in the Table S1 . All analyzed polymorphisms fulfilled Hardy-Weinberg expectations for the control group (non-IPA patients). Several polymorphisms were found to be associated with IPA infection ( Table 3) Table 3 and Table S1 ). After correction for multiple testing using SNPSpD (number of independent marker loci, 21; p = 0.05/21 = 0.002), none of the SNPs retained significance although Dectin-1 rs7309123 showed significance when the carrier status was analyzed (CC+CG vs. GG, P = 0.001; Table 3 ). In order to assess the degree to which the selected SNPs and the positivity of GM correlated, we also performed lineal regression analysis. In the whole population, 3784 assays were performed in duplicate and 531 were considered as positive (14.03%). We found a significantly higher percentage of positive GM among patients carrying the Dectin-1 rs3901533_T allele and among those patients bearing the Dectin-1 rs7309123_G/G genotype suggesting a role of these polymorphisms in determining a defective recognition and clearance of Aspergillus conidia (Figure 1 ). No correlation was observed for the polymorphisms within DC-SIGN that were associated with the infection. We subsequently assessed whether SNPs associated with IPA infection showed capacity to change putative transcription factor binding sites using FastSNP. The predictive functional analysis suggested an intronic enhancer function for the Dectin-1 rs7309123 SNP due to its location in transcription factor Cdxa (caudal type homeo-box transcription factor 1) binding site (GAAAGAC; score 1-2). These data suggest a central role of the rs7309123 in the susceptibility to IPA infection. None of the remaining SNPs associated with IPA infection was predicted to affect transcription factor binding sites or splicing or to introduce a damaging amino acid change. To explore the potential consequences of Dectin-1 rs7309123 SNP, Dectin-1 mRNA expression was measured in 21 healthy donors and was correlated with genotypes. Our results showed that subjects harbouring GG genotype showed a significantly decreased level of Dectin-1 mRNA expression compared to C-allele carriers (p,0.001; Figure 2 ). These results further supported our hypothesis suggesting that Dectin-1 rs7309123_G allele may disrupt binding sites for potential transcription factors. The MDR analysis of the SNPs tested revealed statistically significant interactions. Two SNPs (rs1024611 and rs4264222) were excluded from the MDR analysis as they were in LD with other SNPs using a cut-off value of r 2 ,0.80 in our sample set. The best interaction models selected by the TuRF filter algorithm along with its testing accuracy and cross-validation consistency are shown in Table 4 . The overall best model with the highest crossvalidation consistency (CVC) consisted of a model that included the CCL2 rs4586 , Dectin-1 rs3901533 , CCR2 rs3918358 and Dectin-2 rs7134303 SNPs. This model had a significant testing accuracy of 0.7735 (permutation p,0.001) and a cross-validation consistency of 100/ 100. Of note is that two SNPs showing genetic interaction in this model were not significantly associated with an increased risk of IPA infection in the univariate analysis (CCR2 rs3918358 and Dectin-2 rs7134303 ). The best two-locus model consisted of the Dectin-1 rs3901533 and DC-SIGN rs4804800 SNPs, two variants that showed also association in the single-locus analysis. This model had a testing accuracy of 0.6409 and a cross-validation consistency of 76/100. This model was not any more significant after 1.000-fold permutation testing. However, the entropy based information gain calculated for this pair of SNPs indicated strong synergy, which may be interpreted as the two SNPs acting together to increase the risk of IPA infection. The best three-locus model included the CCL2 rs4586 , Dectin-1 rs3901533 and Dectin-2 rs7134303 SNPs. In this model, testing accuracy raised to 0.7085 (permutation p = 0.025) whereas the cross-validation consistency was of 68/100. Figure 3 illustrates an interaction dendogram that summarizes the estimates of interactions. The marked differences in susceptibility to IPA infection among hematological patients (with or without allo-HSCT) suggest that the effective immune response against Aspergillus is determined by both environmental and host genetic factors [44, 45] . Studies on genetic polymorphisms in genes coding for components of the innate immunity have supported this hypothesis [19, [23] [24] [25] [26] [27] [28] [29] [30] 46] . In this report, we studied the influence of the tagging and potentially functional polymorphisms of DC-SIGN, Dectin-1, Dectin-2, CCL2/MCP-1 and CCR2 in susceptibility to IPA infection in a Spanish population. Polymorphisms in these genes have been reported to influence a number of infectious diseases including HIV-1 [47] [48] [49] , HTLV-1 [50] , CMV [51] , Tuberculosis [52] [53] [54] [55] , HCV [56, 57] , HBV [58] , Dengue [35] and SARS [59] among others, revealing their potential role in host defense against pathogens. We found that polymorphisms in Dectin-1 (rs3901533 and rs7309123) and DC-SIGN (rs4804800, rs11465384, rs7248637 and rs7252229) were associated with an increased risk to develop IPA infection, which points towards their critical involvement in the pathogenesis of this invasive fungal infection. The highest risk of IPA infection was found for carriers of the Dectin-1 rs3901533_T/T and Dectin-1 rs7309123_G/G genotypes and the DC-SIGN rs4804800_G allele. Patients carrying these genotypes/alleles had from 2 to 6 times increased risk of IPA infection. Additionally, patients carrying the DC-SIGN rs11465384_T , DC-SIGN 7248637_A and DC-SIGN rs7252229_C alleles showed a 2-fold increased risk in comparison with patients harboring the wild-type allele. Although it was not statistically significant, we also found that the DC-SIGN rs2287886 SNP may be associated with a reduced risk of IPA infection. Interestingly, this latter result was in agreement with our previously reported findings using the former EORTC/MSG classification criteria, 2005 [60] . The apparent effect of these SNPs on IPA susceptibility persisted even after adjustment for age, gender, hematological malignancy and several known risk factors (HSCT, neutropenia, GVHD and corticoid therapy use), indicating that Dectin-1 and DC-SIGN variants contribute independently to the risk of infection. Another interesting finding of this study was the significantly greater positive GM percentage of patients carrying the Dectin-1 rs3901533_T allele than those with the wild-type allele. Additionally, patients harboring the G/G genotype for the Dectin-1 rs7903123 SNP showed an increased percentage of positive GM compared to those carrying the C allele (C/C+C/G). No differences were found when positive GM determinations were correlated with DC-SIGN polymorphisms. These data along with the remarkable degree of association of Dectin-1 and DC-SIGN variants with risk of IPA infection provides a compelling evidence for a critical role for these PRRs in immune response to IPA infection. In this regard, numerous studies have shown that Dectin-1 and DC-SIGN are not only involved in the recognition of fungal pathogens but also in the induction of anti-fungal Th1 and Th17 immune responses [5, 11] . Mezger et al. also demonstrated that Dectin-1 is involved in the induction of several pro-inflamatory cytokines, chemokines and immune receptors [18] while Werner et al. showed that Dectin-1 is also regulating Th17-mediated immune response in the lungs [61] . Furthermore, Dennehy and Brown suggested a role of Dectin-1 mediating its own signaling, as well as synergizing with TLRs to trigger NFkB-mediated immune response against fungal pathogens [62] . Although it is now well recognized that SNPs in genes modulating immune response are likely to be determinants of host susceptibility to fungal infections, so far, little is known regarding the biological significance of these variants. In order to shed light into the potential functionality of Dectin-1 (rs3901533 and rs7309123) and DC-SIGN (rs4804800, rs11465384, rs7248637, rs7252229 and rs2287886) variants, we investigated whether they were involved in disruption of a binding site for critical transcription factors that might influence transcription level of these genes. Our predictive analysis showed that the carriage of the C allele for the Dectin-1 rs7309123 SNP creates a putative binding site for Cdxa, a relatively unknown transcription factor, which might be involved in the control of Dectin-1 gene expression. To assess whether the Dectin-1 rs7309123 polymorphism might be associated with a decreased expression of Dectin-1, we correlated Dectin-1 mRNA expression levels with Dectin-1 rs7309123 genotypes. Interestingly, we observed that individuals harbouring the GG genotype showed a relatively lower expression than those carrying the C allele (CC+GC). These results further supported our hypothesis suggesting that Dectin-1 rs7309123 SNP may have an effect on the Dectin-1-mediated recognition of Aspergillus conidia and subsequent immune responses. However, this predicted change in transcription activity is only suggestive at this stage and will need further validation using in vitro functional assays. Several lines of evidence point to the relevance of epistatic effects in the etiology of complex diseases but, up to now, no studies have been carried out to analyze the presence of SNP-SNP interactions in IPA infection. For this reason, we decide to assess interactions among genetic polymorphisms within DC-SIGN, Dectin-1, Dectin-2, CCL2 and CCR2, genes and the risk of IPA infection. The MDR approach used in this study was able to determine two multilocus combinations associated with high risk to develop IPA infection. Of the interactions identified, MDR indicated that the type of interaction in the two significant models was synergistic. These results support the hypothesis that multiple SNP-SNP interactions may play a role in determining the risk of IPA infection. This hypothesis is biologically plausible since the immune system would warrant prevention of fungal infection even when some genetic variants were present. Recent population-based studies have even led to the identification of several SNPs involved in the early recognition of Aspergillus and associated them with an increased risk for invasive fungal infection. It has previously been suggested that SNPs within C-type lectin genes are associated with fungal infections. Platinga et al. (2009) and Cunha et al. (2010) suggested that patients carrying the Y238X (rs16910526) polymorphism in the Dectin-1 gene were more likely to be colonized with Aspergillus and Candida species [19, 63] , compared with those harboring the wild-type allele. However, these results were not replicated in a very recent study [24] . In our study, such findings were neither evidenced even when HSCT and non-HSCT patients were analyzed separately (data not shown). This puzzling finding suggests that, although intronic polymorphisms within Dectin-1 and DC-SIGN are indeed themselves a strong indication that these genes play an important role in the susceptibility to IPA infection, we cannot rule out the Figure 3 . Interaction dendrogram generated by the MDR software. The interaction dendrogram was used to confirm, visualize, and interpret the interaction model. The MDR analysis was performed by using the open-source MDR software package. The colors used depict the degree of synergy, ranging from red (highest information gain) to blue (highest information redundancy). Note that the interaction between Dectin-1 (rs3901533) and DC-SIGN (rs4804800) SNPs showed the highest degree of synergy (gain of information). doi:10.1371/journal.pone.0032273.g003 possibility that these SNPs are part of a bigger haplotype containing important other genetic variants in the neighboring genes. In any case, because all these population-based studies have been conducted using relatively small cohorts, additional studies in larger set of patients are needed to definitely establish the role of these variants in the susceptibility to invasive fungal infection. In summary, this study provides evidence of association between Dectin-1 and DC-SIGN polymorphisms and the risk of IPA infection. By the inclusion of functional prediction analyses, the correlation of genotypes with positive GM determinations and Dectin-1 mRNA expression levels, the study strongly supported the role of Dectin-1 gene variants in determining susceptibility to IPA infection. Epistatic analyses also suggested the presence of a gene-gene interaction involving Dectin-1 with CCL2 and CCR2 variants to determine IPA infection. Despite all these evidences, additional studies using larger cohorts will be necessary to confirm the effect of these polymorphisms on the susceptibility to IPA infection. Table S1 Associations of polymorphisms involved in the phagocyte-immune related response against Aspergil-lus. 1 Models adjusted for age, gender, hematological malignancy, HSCT, neutropenia (defined as absolute neutrophil count ,500 cells/mm3 for a period of more than 10 days), GVHD and corticoid therapy use (.0.3 mg/Kg/day). { Assuming a recessive model of inheritance. Abbreviations: OR, odds ratio; CI, confidence interval. Differences in samples numbers are due to failures in genotyping. (DOC)

Lethal Nipah Virus Infection Induces Rapid Overexpression of CXCL10

Nipah virus (NiV) is a recently emerged zoonotic Paramyxovirus that causes regular outbreaks in East Asia with mortality rate exceeding 75%. Major cellular targets of NiV infection are endothelial cells and neurons. To better understand virus-host interaction, we analyzed the transcriptome profile of NiV infection in primary human umbilical vein endothelial cells. We further assessed some of the obtained results by in vitro and in vivo methods in a hamster model and in brain samples from NiV-infected patients. We found that NiV infection strongly induces genes involved in interferon response in endothelial cells. Among the top ten upregulated genes, we identified the chemokine CXCL10 (interferon-induced protein 10, IP-10), an important chemoattractant involved in the generation of inflammatory immune response and neurotoxicity. In NiV-infected hamsters, which develop pathology similar to what is seen in humans, expression of CXCL10 mRNA was induced in different organs with kinetics that followed NiV replication. Finally, we showed intense staining for CXCL10 in the brain of patients who succumbed to lethal NiV infection during the outbreak in Malaysia, confirming induction of this chemokine in fatal human infections. This study sheds new light on NiV pathogenesis, indicating the role of CXCL10 during the course of infection and suggests that this chemokine may serve as a potential new marker for lethal NiV encephalitis.

Nipah virus (NiV) is a recently emerged zoonotic pathogen of the family Paramyxoviridae that is distinguished by its ability to cause fatal disease in both animals and humans. NiV along with Hendra virus (HeV) comprise the new genus Henipavirus (reviewed in [1] ). NiV was first identified during an outbreak of severe encephalitis in Malaysia and Singapore in 1998-99, with pigs serving as the intermediate amplifying host [2] . Since 1998 NiV has caused regular outbreaks, primarily in Bangladesh and India, with the most recent occurrences at the beginning of 2011 [3] . In the majority of subsequent spillover events, the mortality rate among humans has been higher (,75%) along with evidence of multiple rounds of person-to-person transmission [4, 5, 6] . The endothelial cells represent one of the major targets of NiV infection, which is characterized by a systemic vasculitis and discrete parenchymal necrosis and inflammation in most organs, particularly in the central nervous system (CNS). The high pathogenicity of NiV infection appears to be primarily due to endothelial damage, syncytia and vasculitis-induced thrombosis, ischemia and vascular microinfarction in the CNS, allowing the virus to overcome the blood-brain-barrier (BBB) and to subsequently infect neurons and glia cells in the brain parenchyma [2, 7] . Pathogenesis of NiV infection is still poorly understood. As the endothelium forms the primary barrier of the circulatory system, endothelial dysfunction during infection could broadly affect immune cell function by regulating cytokines, chemokines and cell receptors and influencing vascular permeability. To gain new insights into virus-host interaction, we analyzed the transcriptome signature of NiV infection in primary human umbilical vein endothelial cells (HUVECs). Among the ten most strongly upregulated genes 8 h after NiV infection, we identified CXCL10 (interferon gamma-induced protein 10, IP-10), a chemokine that promotes leukocyte trafficking by acting on T lymphocytes, NK and dendritic cells via its receptor CXCR3 [8] . In addition to its protective role during the viral infection, CXCL10 may enhance the severity of virus infection and cause neuronal apoptosis and calcium dysregulation [9, 10] and enhance autoreactive lymphoproliferation and brain injury [11] . We have confirmed secretion of CXCL10 by NiV-infected HUVECs by ELISA and showed that increased production of CXCL10 follows NIV replication in experimentally infected hamsters. Finally, we demonstrated production of CXCL10 in brain endothelial cells of patients with fatal acute NiV encephalitis. Altogether, these results suggest that CXCL10 may be an important regulator of NiV-induced pathogenesis as well as a potential marker for NiV infection. As endohelial cells are the first target of NiV infection, we have initially studied a permissivity of HUVEC cultures to NiV infection. We established the HUVEC cell culture and assessed their purity at passage 2, either by cell immunostaining or flow cytometry using two endothelial-specific markers, Von Willebrand Factor (VWF) stored in cytoplasmic Weibel-Palade bodies, and CD31 (PECAM-1) localized in homophilic intercellular contacts, which gave us reproducibly satisfactory results (Fig. S1 ). We then analyzed the expression of NiV receptors, production of viral specific RNAs, generation of the cytopatic effect and production of viral particles (Fig. 1) . The expression of mRNA specific for NiV entry receptors ephrinB2 and B3 [12, 13, 14] was readily detected in these cells, at levels close to those in the astroglioma cell line U373, which was also highly permissive to NiV infection [15] (Fig. 1C ). Production of NiV structural genes coding for nucleoprotein, matrix, fusion protein, glycoprotein and polymerase increased rapidly during the infection, following a gradient characteristic of paramyxoviruses, with highest expression of the first gene represented in the genome, N, and lowest for the gene coding for the polymerase (Fig. 1D ). Formation of large syncytia was rapidly detected in HUVECs (Fig. 1B ) and the highest level of the production of infectious particles was obtained 24H after infection, as most of the cells were lysed 48 h p.i. in this cell type (Fig. 1E) . Therefore, NiV infects rapidly HUVECs and induces a significant cytopathic effect. To better understand virus-host interaction, we have analysed the effect of NiV infection at the level of gene expression in HUVECs, using Codelink microarray (see Materials and Methods). Cells were infected with either NiV or treated with mock preparation and RNA was taken 8 h p.i., to obtain the information about the early changes in NiV-induced gene expression in HUVEC in the conditions when cell viability was still preserved. Among 55.000 analyzed genes, pair-wise comparisons between infected and uninfected samples revealed 807 deregulated genes (fold change cut-off = 1.3). NiV infection down-regulated galectin 3 gene expression, a member of the lectin family, involved in cell adhesion, cell activation and chemoatraction [16] , but most of the other down-regulated genes were not found associated with any known function (table 1) . The 538 up-regulated genes were classified according to their Gene Ontology (GO) biological processes and molecular functions. This analysis revealed that NiV-infection up-regulated 34 genes implicated in the immune response, particularly those associated with the interferon pathway, including MxA, RIGI, MDA, 2959-OAS 1 and 2 ( Fig. 2) . Interestingly, among the top ten up-regulated genes was CXCL10 (interferon gamma-induced protein 10, IP10), an important chemokine secreted by endothelial cells (table 1). These results have been further confirmed by RT-qPCR at 8 and 24 h post-infection for several genes linked to interferon pathway, including MXA, OAS1, IFN beta and CXCL10 (Fig. 3A ). In addition to stimulation of CXCL10, which was observed at different MOI of infection, MOI of 1 as well as 3, the induction of the closely related chemokine CXCL11 (Interferon-inducible Tcell alpha chemoattractant, I-TAC or IP-9), being at 19 th position among up-regulated genes, was also confirmed by RT-qPCR (Fig. 3A) . However, the production of the third closely related chemokine CXCL9 (MIG), which shares the same receptor CXCR3 with CXCL10 and CXCL11, was not detected. Finally, As CXCL10 is known to have an important immunobiological function in the organism, which received a great deal of attention in recent years, we focused further studies on this chemokine. Thus, we analyzed whether NiV-infection induces production of CXCL10 in vivo, using the hamster animal model, which closely reproduces human infection and induces lethal outcome starting from day 5 p.i. [17] . Hamsters were infected by NiV and sacrificed on a daily basis for analysis of RNA in different organs (Fig. 4) . The baseline expression of CXCL10 was observed in all organs and increased during the course of infection, particularly in brain and kidney. In lung and spleen the level of CXCL10 initially increased, but decreased on the last day that preceded lethal outcome of the infection. The most rapid and highest induction of CXCL10 mRNA expression was observed in the spleen (16 times more than the baseline expression, already 24 h p.i.), which may correlate with the abundance of cells capable of secreting CXCL10 in the spleen (leukocytes, endothelial cells and splenic stromal cells [18] . The highest NiV replication was observed in lungs of infected hamsters. The level of CXCL10 expression significantly correlated to the level of NiV N expression in analysed organs (p,0.001), suggesting that production of this chemokine closely follows the NIV replication. We have next performed immunohistochemical analysis of brain tissues from patients that succumbed to NiV-infection during the outbreak in Malaysia in 1999. Widespread vasculitis and perivascular infiltration were regularly detected after hematoxylin (Fig. 5A) . Altogether, these results indicated that NiV-infection activates cellular pathways leading to an increase of CXCL10 expression that may play a role in the pathogenesis of this highly lethal emergent infection. The inflammatory cellular infiltrates found in the CNS of patients with acute NiV encephalitis include neutrophils, macrophages, lymphocytes and reactive microglia [7] , which may all play an important role in NiV pathogenesis. Recruitment of these cells indicates a possible action of chemokines, known as main regulators of leukocyte trafficking. Our analysis of the transcriptome profile of NiV-infected primary endothelial cells revealed the induction of several genes involved in the interferon type I pathway as well as the overexpression of CXCL10, an important chemoattractant chemokine with proinflammatory activity [19] . Although expression of IFN alpha and beta was shown to be inhibited in human cell lines after NiV infection [20] , human primary endothelial cells seem to be resistant to this inhibition [21] . Our study demonstrated that a few other members of interferon type I pathway, including MxA, RIGI, 2959-OAS 1 and 2, are highly expressed early after NiV infection, suggesting a functional interferon pathway in NiV-infected human endothelial cells. CXCL10 is a secreted polypeptide of 10 kDa that was first identified as an early response gene induced after interferon gamma treatment in a variety of cells, and was thus named interferon gamma-inducible peptide, IP-10 [22, 23] . Furthermore, it was shown that interferon alpha could also induce CXCL10 in primary cultured neurons [24] and seminiferous tubules [25] . In addition to interferons, HIV envelope glycoprotein gp120 was shown to induce expression of CXCL10 in brains of mice [26] . HIV Tat protein can also induce CXCL10 in astrocytes [27] . Furthermore, CXCL10 is expressed early in the CNS in response to a wide variety of viruses [11, 13, 28, 29, 30] . Our results demonstrate that NiV-infection induces rapidly production of CXCL10 in primary endothelial cells. As production of interferon gamma by these cells is contested, it is possible that the generation of CXCL10 is either a direct effect of NiV proteins or is induced via NiV-activation of interferon beta, following infection. Moreover, our results demonstrate induction of CXCL10 in vivo in different organs of NiV-infected hamsters, as well as in patients that succumbed to lethal NiV-infection, thus revealing an important association between NiV-infection and CXCL10 production. Although endothelial cells were clearly secreting this chemokine, the perivascular inflammatory cells and some neurons are most probably participating in the production of CXCL10 as well. Individual chemokines, including CXCL10, play opposing roles in neuroinflammation in different experimental models of infectious disease, making it difficult to predict whether they have a protective role by contributing to immune eradication of the microbial attack or they cause inflammatory damage and disease [19] . Therefore, potential effect of CXCL10 in NiV infection could be either beneficial or harmful and the balance between these two effects may be critical for the final outcome of the disease. Infection with RNA viruses that may cause encephalitis in humans, such as HIV or lymphocytic choriomeningitis viruses, or in rodent models, such as mouse hepatitis virus and Theiler's virus, can directly induce the expression of chemokines by astrocytes and microglia and establish chemokine gradients that promote leukocyte trafficking within the CNS [11, 31, 32, 33] . In addition, neutralization of CXCL10 decreased leukocyte migration to areas of infection in measles virus-infected brain tissue [34] . Although several of these viruses directly infect neurons, these cells have not been observed to participate in the inflammatory response, suggesting the indirect induction of inflammatory response via cytokines. Nevertheless, in a transgenic mouse model of measles virus encephalitis, neuronal expression of CXCL10 was associated with T-cell recruitment, suggesting that neurons may play a role in the induction of immune responses to viral invasion [35] . In recent study Lo et al. showed that NiV infection could induce production of CXCL10 and several other inflammatory chemokines in microvascular, but not in macrovascular endothelial cells [21] . However, in contrast to our study, they cultured endothelial cells in the presence of hydrocortisone, known to have important antiinflammatory activity and to reduce the expression of proinflammatory cytokines [36] , which could account to the differences with our results. In accord to our results, NiV was shown to induce CXCL10 expression as well as several other inflammatory chemokines in brain and lungs of NiV infected hamsters [37] suggesting altogether that the other cytokines may contribute to the CXCL10 proinflammatory effect. CXCL10 has been detected in the cerebrospinal fluid of individuals with HIV-1 infection [38, 39] and in the brains of individuals with HIV-associated dementia [33] , but was absent in uninfected control individuals. These authors also reported that CXCL10 levels were closely associated with the progression of HIV-1-related CNS infection and neuropsychiatric impairment. Moreover, CXCL10 and its receptor CXCR3 were shown to be present in the brains of macaques with SIV/SHIV-E [40, 41] and to elicit apoptosis in fetal neurons [9, 42] . The mechanisms of neuronal injury mediated by the CXCL10 was suggested to be associated to calcium dysregulation during CXCL10 mediated apoptosis [10] , and we hypothesize that overexpression of CXCL10 in Nipah-infected patients may be directly involved in NiV-induced neuropathology. Lack of the appropriate regaents for hamster animal model prevented us from further in vivo analysis of the role of CXCL10 during Nipah infection. CXCL10 is the substrate for serine protease CD26/dipeptidylpeptidase 4, which cleaves its aminoterminal part and alters its receptor binding and signaling, producing therefore an antagonistic protein with dominant negative function [43] . This antagonist form of CXCL10 was very recently shown to play an important role in patients chronically infected with hepatitis C virus [44] and to present an important negative prognostic marker for the response to therapy [45, 46] . Whether this cleaved antagonist form of CXCL10 is generated during NiV-infection remains to be elucidated. If produced, its action on inhibition of the physiological role of CXCL10, may play an important role in the pathogenesis of NiV-encephalitis. These results suggest that NiV-infection in endothelial cells induces CXCL10 production both in vitro and in vivo and highlight the use of molecular profiling of virus-infected cells as a powerful tool to define novel mechanisms of virus-host cell interaction. As equilibrium in cytokine production is essential in the generation of the adequate immune response, CXCL10 overexpression may be crucial for the development of NiV-associated encephalitis and could be a target for therapeutic approaches. Umbilical cords were obtained between 2006 and 2008 from healthy full-term newborns with written parental informed consent according to the guidelines of the medical and ethical committees of Hospices Civils de Lyon and of Agence de Biomedecine, Paris, France. Ethics Committee approval for this study was not required according to institutional guidelines and French law Nu2004-800, from 6 August 2004 -art 12 ORF 7.08.2008, Article L1245-2, allowing the utilization of the placenta and umbilical cords in scientific and therapeutic purpose when their donors do not express any opposition. All animals were handled in strict accordance with good animal practice as defined by the French national charte on the ethics of animal experimentation. Animal work was approved by the Regional ethical committee (Comité Régional d'Ethique pour l'Experimentation Animal de la Région Rhone-Alpes, CREEA, protocol Nu 220) and experiments were performed in the INSERM Jean Mérieux BSL-4 laboratory in Lyon, France (French Animal regulation comittée Nu A69 387 05 02). Autopsies of human brain tissue were performed after receiving written patient's relatives consent for autopsy studies at the Pathology Department of University of Malaya, Kuala Lumpur, Malaysia and their analysis was approved by review board of Faculty of Medicine of University. Primary HUVECs were isolated from human umbilical cords of 22 donors by treating the umbilical vein with 0.1% collagenase for 30 min at 37uC as described previously [47] . Cell cultures were pooled by sets of three donors for experiments. Cells were cultured in flasks coated with 0.2% gelatin in complete medium containing M199 medium (Gibco), 20% of fetal calf serum (Gibco), 100 mg/ml bovine brain extract [48] , 14 mM Hepes (Gibco), 10 UI/ml heparin (Pfizer), and a cocktail of antibiotic/antimycotic (Gibco). At passage 4, cells were seeded at 30 000 cells/cm 2 for 8 h, then serum deprived for 16 h prior to NiV-infection in complete medium. Immunofluorescent staining and analysis of HUVEC culture was performed as described in Methods S1. U373 astroglioma and Vero cells were maintained in DMEM medium (Invitrogen) supplemented with 10% fetal calf serum, 100 U/ml penicillin, 0.1 mg streptomycin, 10 mM HEPES and 2 mM L-Glutamine at 37uC in 5% CO 2 . Nipah virus (isolate UMMC1, Genebank AY029767) [49] was prepared on Vero-E6 cells as described previously [50] . HUVECs were infected with a multiplicity of infection (MOI) of 1 and harvested for RNA isolation 8 and 24 hours post infection (p.i.). At indicated times p.i. 150 ml of cell culture supernatant were collected and frozen and viral titration was performed as detailed elsewhere [50] . Eight-week-old golden hamsters (Mesocricetus auratus, Janvier, France) were anesthetized and infected intraperitoneally (i.p.) with 0.4 ml of wild-type NiV (10 000 PFU) in the BSL-4 laboratory in Lyon. Each day post infection (p.i.), one hamster was euthanised and organs were frozen at 280uC. Total RNA was extracted from uninfected or NiV-infected HUVECs 8 h p.i., both prepared from two different pools, each containing 3 donors and cultured at same conditions, using RNeasy kit (Qiagen) according to the manufacturer's protocol. Quality of total RNA was checked on nanochips with the Agilent 2100 Bioanalyzer 2100 (Agilent Technologies,). Amplified and biotin-labeled RNAs were obtained from 2 mg of total RNA, by a round of in vitro transcription (dIVT) using the Message Amp a RNA kit version II (Ambion), following the manufacturer's protocol. Different quantities of positive synthetic mRNA controls (spikes, corresponding to 6 bacterial RNAs, used to control sensitivity, quality of hybridization and data normalization) were added to all samples, during the first step of reverse transcription of total RNAs. Hybridization was performed on Codelink Uniset Human Whole Genome bioarrays (http://www. codelink bioarrays.com). 10 mg of biotin-labeled RNA were fragmented using 5 ml fragmentation buffer in a final volume of 20 ml, then mixed with 240 ml Amersham hybridization solution and injected onto Codelink Uniset Human Whole Genome bioarrays, containing approximately 55.000 30-mer probes based on the NCBI/Unigene RefSeq database that permit the expression analysis of 57,347 transcripts and ESTs (GE Healthcare Europe GmbH). Arrays were hybridized overnight at 37uC, then washed in stringent TNT buffer at 46uC for 1 h before performing a streptavidin-cy5 detection step. Each array was incubated for 30 min in 3,4 ml streptavidin-cy5 solution, washed four times in 240 ml TNT buffer, rinsed twice in 240 ml water containing 0.2 M Triton X-100, and then, dried by centrifugation at 650 g. Arrays were then scanned at 635 nm using an Axon Genepix 4000B Scanner (Axon). Data extraction and raw data normalization were performed using the CodeLink Gene Expression Analysis v4.0 software. Normalization was performed by the global method. The threshold was calculated using the normalized signal intensity of the negative controls supplemented by 3 times the standard deviation. Spots with signal intensity below this threshold were referred to as ''absent''. Obtained datasets were deposited in GEO database in accord to MIAME guidelines (Accession number: GSE 32902: GSM813064, GSM813065, GSM813066 and GSM813067). Finally, data were converted to the excel format and data analysis was performed by using the Gene Spring v7.0 software from Agilent and pariwise comparisons were performed between infected and uninfected samples. Total RNA was extracted from mock and NiV-infected HUVECs 8 and 24 hours post-infection using RNeasy Mini Kit according to the manufacturer's instructions including additional step with RNase-free DNase (Qiagen). RNA was isolated from hamster organs 10-30 mg using a tissue lyser (Qiagen) in RLT buffer containing 1% ß-mercaptoethanol, according to manufacturer recommendations. Reverse transcriptions were performed on 0.5 mg of total RNA using the Transcriptor first strand cDNA synthesis kit (Roche) and run in BiometraH T-GRADIENT PCR devise. Obtained cDNAs were diluted 1:10. Quantitative PCR was performed using PlatinumH SYBRH Green qPCR SuperMix-UDG with ROX kit (Invitrogen TM ). qPCR was run on the ABI 7000 PCR system (Applied biosystems) using the following protocol: 95uC 59, and 40 cycles of 95uC 150, 60uC 19, followed by a melting curve up to 95uC at 0.8uC intervals. All samples were run in duplicate and results were analyzed using ABI Prism 7000 SDS software available in the genetic analysis platform (IFR128 BioSciences Lyon-Gerland). Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was used as housekeeping gene for mRNA quantification and normalization. GAPDH and standard references for the corresponding genes were included in each run to standardize results in respect to RNA integrity, loaded quantity and inter-PCR variations. Primers used were designed using Beacon 7.0 software, and validated for their efficacy close to 100% in control PCR amplification: EFN B2 for: TCGGGCTAGTTAAGGTGTGC, EFN B2 rev: ATGAGTGTTCCATGAGTGATGC, EFN B3 for: TCACCCTCTTGGCTTCTTATCC, EFN B3 rev: GGGGA-GTGGTTGGTATGAGAG, NiV N for: GGCAGGATTCT-TCGCAACCATC, NiV N rev: GGCTCTTGGGCCAATT-TCTCTG, NiV L for: ATGGTGCTGTGCTGTCTCAGG, NiV L rev: AGCCGACATTTCTTGACAACCC, IFNß for: CTCCTAGCCTGTGCCTCTGG, IFNß rev: TGCAGTACAT-TAGCCATCAGTCAC, MxA for: AGCCACTGGACTGAC-GACTTG, MxA rev: AAATCACCACGGCTAACGGATAAG, OAS1 for: AGAACTTACCTCTTGCCAAAGG, OAS1 rev: GGACAAGGGATGTGAAAATTCC, CXCL10 for: GGAAG-GTTAATGTTCATCATCCTAAGC, CXCL10 rev: TAGTAC-CCTTGGAAGATGGGAAAG, CXCL11 for: GGATGAAAG-GTGGGTGAAAGGAC, CXCL11 rev: AACGTGAAAGCAC-TTTGTAAACTCC, GAPDH for: CACCCACTCCTCCACC-TTTGAC, GAPDH rev: GTCCACCACCCTGTTGCTGTAG. Primers used for hamster study: hamster CXCL10 for: AGA-CAACAGTAACTCCAGTGACAAG, hamster CXCL10 rev: A-GTGTAGCACCTCAGCGTAGC, murine GAPDH for: GC-ATGGCCTTCCGTGTCC, murine GAPDH rev: TGTCAT-CATACTTGGCAGGTTTCT. The relative expression represents the ratio of the number of copy of mRNA of interest versus mRNA of GAPDH and expressed in relationship to the quantity of RNA analyzed All calculations were done using the 2 DDCT model of [51] and experiments were performed according to the MIQE guideline [52] . Supernatants of mock and NiV-infected HUVECs were collected 8, 24, 48, 72 h p.i. and used to quantify the induction of CXCL10 using human IP-10 Immunoassay (Invitrogen TM ), according to manufacturer's instructions. All tests were performed on 2 sets of 3 donors, in duplicates. Autopsy brain tissue was obtained from 3 fatal acute Nipah cases from the outbreak in Malaysia in 1999, fixed in 10% buffered formalin and paraffin-embedded and from two control brain tissues from patients that succumbed to non-neurological diseases. Sections (4 mm) were deparaffinized with xylene and rehydrated through graded alcohol and distilled water. Antigen retrieval was performed in 10 mM trisodium citrate buffer, pH = 6, using waterbath (99uC, 20 min). Slides were then washed 3 times in PBS 16. Endogenous peroxidase was blocked using H2O2, 3% in PBS, for 20 min at Room temperature (RT). Slides were then washed 2 times in PBS 16 and additional blocking was performed using goat serum (1:20 in PBS) for 20 min at RT. Primary rabbit anti-human CXCL10 (Peprotech), was applied diluted at 4 mg/ml, overnight at 4uC. After 3 washes in PBS, slides were incubated with secondary goat anti-rabbit conjugated with HRP (Promega) 1:500 for 1-2 hours at RT. Sections were then developed with AEC kit (VECTOR) for 25 min, washed in distilled water and counterstained with Harris hematoxylin solution (Sigma-Aldrich) 1:3 in PBS, 30 s. After one wash in water (3 min) sections were mounted with DakoCytomation faramount aqueous mounting medium and coverslipped. Slides were analysed using Axioscope microscope equipped with Zeiss Axiovision software (Zeiss Geramny). Data were expressed as mean6standard deviation (SD). Statistic analyses were performed using Mann-Whitney U-test and Pearson correlation test. Figure S1 Expression of the endothelial-specific markers. Von Willebrand Factor (VWF) and CD31 (PECAM-1) were analyzed in 2 nd day HUVEC cultures by immunostaining with anti-CD31 (green) and anti-VWF (red). Nuclei were stained with DAPI (blue). Analysis was performed as described in Methods S1. (TIF) Methods S1 Supplementary methods. (DOCX)

Influenza Virus Respiratory Infection and Transmission Following Ocular Inoculation in Ferrets

While influenza viruses are a common respiratory pathogen, sporadic reports of conjunctivitis following human infection demonstrates the ability of this virus to cause disease outside of the respiratory tract. The ocular surface represents both a potential site of virus replication and a portal of entry for establishment of a respiratory infection. However, the properties which govern ocular tropism of influenza viruses, the mechanisms of virus spread from ocular to respiratory tissue, and the potential differences in respiratory disease initiated from different exposure routes are poorly understood. Here, we established a ferret model of ocular inoculation to explore the development of virus pathogenicity and transmissibility following influenza virus exposure by the ocular route. We found that multiple subtypes of human and avian influenza viruses mounted a productive virus infection in the upper respiratory tract of ferrets following ocular inoculation, and were additionally detected in ocular tissue during the acute phase of infection. H5N1 viruses maintained their ability for systemic spread and lethal infection following inoculation by the ocular route. Replication-independent deposition of virus inoculum from ocular to respiratory tissue was limited to the nares and upper trachea, unlike traditional intranasal inoculation which results in virus deposition in both upper and lower respiratory tract tissues. Despite high titers of replicating transmissible seasonal viruses in the upper respiratory tract of ferrets inoculated by the ocular route, virus transmissibility to naïve contacts by respiratory droplets was reduced following ocular inoculation. These data improve our understanding of the mechanisms of virus spread following ocular exposure and highlight differences in the establishment of respiratory disease and virus transmissibility following use of different inoculation volumes and routes.

Despite reports of conjunctivitis following infection with numerous respiratory pathogens (including influenza, adenovirus, respiratory syncytial virus, and others), research investigating the role of ocular infection in virus pathogenicity and transmissibility has been underrepresented [1] [2] [3] [4] . Influenza virus represents a highly transmissible respiratory pathogen, resulting in .200,000 hospitalizations in the United States annually [5] . While ocular disease is generally rare following influenza virus infection in humans, viruses within the H7 subtype have demonstrated an apparent ocular tropism, with the majority of human infections with H7 influenza viruses associated with conjunctivitis [6] . Moreover, ocular complications have been sporadically documented following seasonal, 2009 H1N1 pandemic, and avian H5N1 virus infections in humans [7] [8] [9] [10] [11] [12] [13] . Numerous properties allow the eye to serve as both a potential site of influenza virus replication as well as a gateway for the establishment of a respiratory infection. Similar to epithelial cells within the human respiratory tract, human ocular tissue and secreted mucins express sialic acids, the cellular receptor for influenza viruses [14] [15] [16] . The anatomical proximity between the eye and nasal passages, notably the linkage of both systems via the nasolacrimal duct, facilitates aqueous exchange and provides shared lymphoid tissue between these sites [17, 18] . Influenza virus can rapidly spread between ocular and respiratory tissues, as was demonstrated in a recent study which detected by RT-PCR live attenuated influenza vaccine (LAIV) in nasal washes in humans within 30 minutes of experimental ocular exposure to LAIVcontaining aerosols [19] . Well-characterized mammalian models to study extraocular spread following ocular inoculation with influenza virus have been limited to the mouse [20] . The ferret, widely used to study influenza pathogenesis and transmission following intranasal inoculation, has also been recognized as an appropriate experimental model for studies involving the visual system [21] [22] [23] . A previous study demonstrated H7N3 virus replication in nasal, ocular, and rectal tissue following ocular inoculation in ferrets, but did not comprehensively examine the ability of multiple subtypes to use the eye as a portal of entry or examine virus transmissibility following inoculation by this route [24] . It is clear from epidemiological and laboratory data that ocular exposure to influenza virus can manifest in both ocular and respiratory disease. However, the properties that contribute towards the ocular tropism of select influenza virus subtypes, and the mechanisms of virus spread from ocular to respiratory tract tissue following ocular exposure to influenza virus, are not well understood. Here, we present a ferret model where influenza virus in a liquid suspension is placed on the surface of the eye and massaged across the surface of the eye within the conjunctival sac (ocular inoculation) to study the ability of human and avian influenza viruses to cause disease and transmit to naïve animals. We found that both human and avian influenza viruses can mount a productive virus infection following ocular inoculation, attributable to replication-independent drainage of virus inoculum from the site of inoculation to respiratory tract tissue. The viral infection following ocular inoculation was capable of causing severe and fatal disease (in the case of H5N1 virus), but was less transmissible by respiratory droplets (in the case of seasonal influenza viruses) compared with infection following inoculation by the traditional intranasal route. Due to a high degree of similarity in lung physiology and distribution of sialic acids in respiratory tract tissues, the ferret model is frequently utilized to model the kinetics and severity of respiratory disease following administration of human and avian influenza viruses by the intranasal route [23, 25] . To determine if this homology extends to ocular tissue, we examined the composition of sialic acids on the ferret cornea, which represents a potential site of influenza virus replication following ocular inoculation. Staining of ferret corneal epithelial sheets with biotinylated lectins specific for a2-3 and a2-6 sialic acids revealed a predominance of a2-3-linked sialic acids with relatively weak expression of a2-6 sialic acids on the epithelial surface, an expression pattern similar to human corneal and conjunctival tissue (data not shown) [2, 20] . To assess the pathogenicity of influenza viruses of multiple subtypes following ocular inoculation (i.o.) in ferrets, we administered 10 6 EID 50 of each indicated virus in a volume in 100 ml on the corneal epithelial surface of the right eye of an anesthetized ferret and massaged the inoculum across the surface of the eye with the eyelid (Table 1) . Ferrets were observed daily for two weeks for clinical signs of illness (including fever, weight loss, nasal or ocular discharge). Nasal wash (NW) and rectal swab (RS) samples were collected on alternate days post-inoculation (p.i.) and were titered for infectious virus, while conjunctival wash (CW) samples were collected alternate days p.i. to measure the incidence and kinetics of infectious virus replication and levels of viral RNA from inoculated eyes (Tables 1 and 2 , Figures 1 and 2 ). All virus subtypes tested replicated in ferrets following ocular inoculation, as measured by detectable virus in NW samples as early as day 1 p.i. (Table 2 and Figure 1 ). The duration of virus shedding from NW samples and transient fever and weight loss generally mirrored that seen following intranasal (i.n.) inoculation for each virus [26] [27] [28] [29] . However, in comparison to i.n. inoculation, the incidence of nasal discharge was reduced following i.o. inoculation with all influenza viruses tested, and Most infections with influenza virus result in respiratory disease. However, influenza viruses of the H7 subtype frequently cause ocular and not respiratory symptoms during human infection, demonstrating that the eye represents an alternate location for influenza viruses to infect humans. Using a ferret model, we studied the ability of influenza viruses to cause disease following ocular inoculation. We found that both human and avian influenza viruses could use the eye as a portal of entry to establish a respiratory infection in ferrets. Influenza viruses were also detected in ocular samples taken from ferrets during virus infection. We identified that influenza viruses spread to different tissues in ferrets when inoculated by ocular or respiratory routes, and that these differences affected the transmissibility of influenza viruses in this model. This study is the first to confirm that virus can spread from the eye to the respiratory tract in a replication-independent manner, and offers greater insight in understanding the ability of influenza viruses of all subtypes to cause human infection by the ocular route. Figure 3 ). Ocular inoculation with all H7 influenza viruses tested resulted in elevated peak mean viral titers in NW samples and a higher incidence of detection in CW samples compared with H5N1 viruses (45% positive among H7 virus samples compared with 25% of H5 virus samples) (Table 2, Figure 2 ). However, despite reduced viral replication in NW samples, H5N1 viruses were capable of causing .20% weight loss and lethal disease in ferrets following i.o. inoculation (Table 1) . Interestingly, all seasonal H1N1 and H3N2 viruses evaluated were detected at high titer in both NW and CW samples; i.o. inoculation with Mex/4108, Brisbane, and Panama viruses, along with the H7N7 NL/230 virus, resulted in the highest peak mean titers in CW samples (.10 3 EID 50 /ml) compared with all viruses examined (Table 2, Figure 2 ). All viruses with the exception of Thai/16 virus were also detected at low titer in RS samples, with peak titers 10 1.8 -10 2.7 EID 50 /ml observed days 3-7 p.i. (Table 2 ). In summary, we found that both avian and human influenza viruses were capable of mounting a productive infection in ferrets following i.o. inoculation, with virus replication observed in samples collected from both ocular and respiratory tract locations regardless of virus subtype. H7 influenza viruses replicated to peak titers .2 logs higher compared with H5 influenza viruses in NW samples following i.o. inoculation, yet H5N1 influenza viruses were capable of maintaining a lethal phenotype following introduction by the ocular route. Seasonal and 2009 H1N1 pandemic influenza viruses efficiently used the eye as a portal of entry to replicate efficiently in the upper respiratory tract as well as ocular tissue. To examine the capacity of influenza viruses to cause severe disease following ocular inoculation, and to better identify those features specific to ocular inoculation, we inoculated ferrets by either the traditional i.n. route (using a 1 mL inoculation volume) or the i.o. route (using a 100 ml inoculation volume) with 10 6 EID 50 of NL/219, NL/230, or Brisbane virus and collected systemic tissues on day 3 p.i. (Table 3 and 4). While i.n. inoculation with the H7N7 viruses tested in this study results in high virus titers throughout the respiratory tract of ferrets, H7N7 virus dissemination following i.o. inoculation was generally restricted to the upper respiratory tract, with a .3 log reduction in titers in nasal turbinates (p,0.05) and only sporadic virus isolation in trachea and lung samples compared with intranasal inoculation (p,0.005) ( Table 3) . A similar pattern of virus dissemination following H7N7 i.o. virus infection was observed when ferrets were inoculated by the i.n. route using a 100 ml and not 1 mL inoculation volume ( Table 3 ). The H1N1 virus Brisbane replicated with comparable efficiency in nasal turbinate samples regardless of the inoculation route or volume chosen, but similar to H7N7 virus ocular infections, did not consistently replicate to high titers in lower respiratory tract tissues. Unlike virus dissemination to the respiratory tract, virus spread to the intestinal tract was not contingent on the route or volume of inoculation. Despite restriction of virus following i.o. inoculation to upper respiratory tract tissues compared with traditional 1 mL intranasal inoculation through day 3 p.i., virus introduced by the ocular route was still capable of causing lethal disease, as ferrets inoculated with the HPAI H5N1 virus Thai/16 by the ocular route required euthanasia days 7-8 p.i. due to development of neurological signs (Table 1) . Ferrets which succumbed to Thai/16 virus infection following ocular inoculation exhibited pronounced lymphopenia in peripheral blood and systemic spread of virus to all regions of the respiratory tract and brain comparable to 1 mL intranasal inoculation, albeit with reduced lethargy and a delayed time-to-death ( [32] and data not shown). These data suggest that ferrets inoculated by the ocular route succumb to a similar course of disease as intranasally inoculated ferrets, however following i.o. inoculation there is a delay in both the kinetics of virus dissemination and the development of neurological signs and severe disease, potentially owing to differences in virus inoculum reaching lower respiratory tract tissues at the time of ocular inoculation. Ocular tissue is not routinely titrated following i.n. inoculation of influenza virus in ferrets, making it difficult to elucidate if viral titers in ocular tissue are a function of i.o. inoculation or are detected regardless of the inoculation route. Therefore, we collected both left and right whole ferret eyes and all surrounding conjunctiva/eyelid for virus titration from ferrets inoculated by the intranasal or ocular route with NL/219, NL/230, or Brisbane viruses (Table 4) . Surprisingly, sporadic viral titers from both left and right eye and conjunctival tissue were detected following HPAI H7N7 virus infection by both i.n. (using either a 1 mL or 100 ml inoculation volume) or i.o. inoculation routes (Table 4) . While the magnitude of viral titers and viral RNA was generally similar between intranasal and ocular routes of inoculation, realtime RT-PCR detected CW-positive samples with a greater sensitivity compared with viral culture. Isolation of virus from ocular tissue may be a reflection of the ability of these HPAI viruses to spread to extra-pulmonary tissues post-inoculation as previously described [26] . However, virus was also detected in left and right conjunctival tissue following i.n. or i.o. inoculation of the H1N1 virus Brisbane, a virus which lacks a high capacity for systemic spread [28] . Comparable levels of viral RNA were isolated from CW samples from ferrets inoculated with Brisbane virus by either intranasal or ocular routes, although infectious virus was only detected in CW samples collected from the eyes of ferrets inoculated by the ocular route. To confirm that virus detected in the eye and conjunctiva was associated with tissue-specific virus replication, immunohistochemistry (IHC) was performed to visualize the presence of influenza A nucleoprotein (NP) in ferret ocular tissues. As shown in Figure 3 , influenza virus antigen was detected in epithelial cells from both the lacrimal glands in the conjunctiva and the ciliary processes in the eye collected day 3 p.i. from ferrets inoculated by the ocular route. These results indicate that the route of virus inoculation in ferrets can affect the extent of virus dissemination in respiratory tract tissue, but extra-pulmonary spread, notably to ocular tissue, is present regardless of the point of entry once an infection is established. The detection of high viral titers in NW samples as early as day 1 p.i. following i.o. inoculation suggests replication-independent Figure 1 . Comparison of mean titers of influenza viruses recovered from nasal wash following ocular inoculation of ferrets. Ferrets were inoculated by the ocular route with 10 6 EID 50 /ml of each virus shown. Viral titers were measured in nasal washes collected on indicated days following serial titration in eggs; endpoint titers are expressed as mean log 10 EID 50 /ml plus standard deviation. The limit of virus detection was 10 1.5 EID 50 /ml. {, ferrets did not survive to day 9 p.i. doi:10.1371/journal.ppat.1002569.g001 spread of virus from the eye to the respiratory tract ( Figure 1 ); this has been similarly hypothesized in previous studies, but has yet to be proven experimentally [20, 33, 34] . Reduced viral titers in the lungs of ferrets on day 3 p.i. following ocular compared with intranasal administration further indicates differential patterns of virus spread following inoculation (Table 3) . To visualize the deposition of virus immediately following different routes of inoculation, we labeled NL/219 virus with an AF680 fluorescent tag (NL/219-FL) and inoculated ferrets with equal quantities of NL/219-FL virus by the ocular (100 ml total volume) or intranasal (1 ml total volume diluted in PBS) route ( Figure 4 ). Ferrets were euthanized 15 minutes following virus inoculation for ex vivo imaging. In ferrets inoculated by the traditional intranasal route, the majority of virus was deposited in the nasal turbinates and lungs, consistent with a previous study demonstrating virus dissemination throughout upper and lower respiratory tract tissue following this route of inoculation [32] . In contrast, virus deposition in ferrets inoculated by the ocular route (right side only) was primarily localized in the nasal turbinates and right conjunctiva. Lower relative quantities of virus inoculum were present in the upper trachea and esophagus following either intranasal or ocular inoculation. These findings demonstrate that, following i.o. inoculation in ferrets, influenza virus rapidly spreads to the nasal turbinates and upper trachea in a replicationindependent manner, but in contrast to i.n. inoculation, does not immediately deposit in peripheral lung tissue. Furthermore, initial deposition of virus inoculum following ocular inoculation occurs not on the corneal surface of the eye but is rather concentrated in the surrounding conjunctival tissue. To determine if ocular exposure to influenza virus results in a transmissible respiratory infection, we inoculated ferrets by the Figure 2 . Comparison of influenza virus recovery in conjunctival wash samples following ocular inoculation of ferrets. Ferrets were inoculated by the ocular route with 10 6 EID 50 /ml of each virus shown. Viral titers were measured in conjunctival washes (CW) collected on indicated days following serial titration in eggs; endpoint titers are expressed as mean log 10 EID 50 /ml plus standard deviation (left y-axis and bars). Relative viral RNA copy number in conjunctival washes was determined by real-time PCR using a universal M1 primer and extrapolated using a standard curve based on samples of known virus (right y-axis and lines). The limit of virus detection was 10 1.5 EID 50 /ml. {, no ferrets survived until day 9 p.i. R-squared values are shown for those viruses where a statistically significant (p,0.05) correlation between viral titer and viral RNA copy number exists. NS, not significant. doi:10.1371/journal.ppat.1002569.g002 PLoS Pathogens | www.plospathogens.org ocular route with selected influenza viruses known to transmit following traditional i.n. inoculation to naïve contacts in the presence of direct contact or by respiratory droplets ( Figure 5 ). Transmission was assessed by the detection of virus in NW samples and seroconversion of contact ferrets. To assess virus transmission in the presence of direct contact, ferrets were inoculated by the ocular route with the H7N2 virus NY/107 or the H7N7 virus NL/ 230, both viruses which transmit efficiently by this route following i.n. inoculation in ferrets [27] . Twenty-four hours later, a naïve ferret was placed in the same cage as each inoculated ferret to assess transmission. Both NY/107 and NL/230 viruses replicated efficiently in the inoculated ferrets following ocular inoculation as expected, and spread to 2/3 and 3/3 contact ferrets by day 7 postcontact (p.c.), respectively ( Figure 5A ). To assess virus transmissibility by respiratory droplets in the absence of direct contact, ferrets were inoculated by the ocular route with the H1N1 virus Brisbane or the H3N2 virus Panama, both viruses which transmit efficiently by this route following traditional i.n. inoculation [28, 30] . Twenty-four hours following i.o. inoculation, a naïve ferret was placed in an adjacent cage with modified side walls, so that air exchange was permitted between inoculated and contact ferrets in the absence of direct or indirect contact. Unlike the efficient transmission observed with these viruses following traditional i.n. inoculation, ferrets inoculated by the ocular route with either Brisbane or Panama only transmitted virus by respiratory droplets to 1/3 contact ferrets ( Figure 5B ). Virus was not detected in CW or RS samples from the infected Brisbane contact ferret; the infected Panama contact ferret had a peak CW titer of 10 2.75 EID 50 day 5 p.c. and peak RS titer of 10 2.5 EID 50 day 7 p.c. While RD contacts with detectable virus in NW samples seroconverted to homologous virus at the end of the experimental period, contact ferrets which did not have detectable virus in NW samples did not exhibit seroconversion (data not shown). Ferrets inoculated intranasally with 10 6 EID 50 of Brisbane virus in a reduced 100 ml volume and tested for their ability to transmit virus by respiratory droplets exhibited a similar pattern of virus transmissibility as ferrets inoculated by the ocular route, with virus shedding and seroconversion detected in only 1/3 contact ferrets (data not shown). These findings suggest that despite high titers of virus in NW samples, the respiratory infection resulting from inoculation of ferrets with a reduced volume, by either ocular or intranasal inoculation routes, is distinct from that following traditional i.n. inoculation, characterized by a diminished incidence of sneezing and nasal discharge and resulting in reduced transmission of virus by respiratory droplets. While the ferret has proved essential for the study of influenza virus pathogenesis and transmission, the use of this species to examine alternate inoculation methods has been limited [32, [35] [36] [37] [38] . Characterizing the progression of disease following alternate routes of inoculation with influenza virus will assist in the better understanding of unique features and the relative severity and risk associated with different exposure routes. In this study, we established an in vivo model using the ferret to assess the ability of influenza viruses of multiple subtypes to use the eye as a gateway to establish a productive infection. Both human and avian influenza viruses were capable of mounting a respiratory virus infection in ferrets following i.o. inoculation. The detection of virus in ocular samples collected from ferrets inoculated by either ocular or intranasal routes demonstrates the importance of studying ocular involvement in respiratory virus infection. Divergent patterns of virus transmissibility by respiratory droplets following use of different inoculum volumes and routes of inoculation highlights the complexity of properties which govern virus transmission. The high similarity of respiratory tract tissue between humans and ferrets makes the ferret model an attractive one for modeling human respiratory disease and investigating the role of receptor specificity in influenza virus pathogenesis, providing an advantage over murine models [25] . We found that the sialic acid composition of ferret corneal epithelial sheets more closely mimics that of humans compared with a mouse model, demonstrating another physiological parallel between ferrets and humans [20] . Bridging the a2-3 rich corneal and conjunctival epithelial surfaces with a2-6 rich upper respiratory tract tissue is the lacrimal duct, which expresses both a2-3 and a2-6 linked sialic acids [14, 34] . Characterization of the distribution of sialic acids in the ferret conjunctiva and lacrimal duct, in addition to the composition of ferret ocular mucins, will allow for a better understanding of virus attachment and replication in these locations. However, in a previous study, we demonstrated that the ability of influenza viruses to bind to or replicate in ocular tissue cannot be explained by sialic acid binding specificity alone [20] . Our detection in ferret ocular tissue of both human and avian influenza viruses with distinct binding specificities further underscores this point (Table 4 , Figure 3 ). Macroscopic signs of ocular disease in ferrets were not observed during the course of infection with any virus tested, similar to prior observations in mouse and rabbit models following deposition of influenza virus on the corneal surface [20, 39] . A previous study in ferrets reported mild conjunctival inflammation following i.o. inoculation with an H7N3 virus, however this may be attributable to strain-specific differences or the use of younger (3-5 month old) ferrets [24] . Despite the absence of visible ocular complications, virus was consistently detected in CW samples from ferrets inoculated by the intranasal or ocular route (Table 2, Figure 2 ). Levels of viral M1 RNA generally correlated with the magnitude of virus isolation, and were a more sensitive detection method compared with virus isolation in CW samples, similar to that observed in human eye swabs (Table 4, Figure 2 ) [40] . The presence of virus in RS samples following both i.n. and i.o. inoculation with influenza virus has been previously reported and likely originates from virus swallowed during inoculation [24, 32, 41] , as indicated by deposition of virus in the esophagus following initial virus inoculation by both inoculation routes (Table 3, Figure 3 ). Unlike in a murine model, the ferret model supported virus replication of both human and avian influenza viruses following i.o. inoculation [20] . In this ferret model, H7 viruses were detected at higher titer in NW samples and with a higher frequency in ocular CW samples compared with H5N1 viruses, suggesting a recapitulation of the tropism of the H7 virus subtype observed in humans (Table 2) . However, the permissiveness of multiple virus subtypes to cause a productive infection following i.o. inoculation in the ferret points to a greater capacity of influenza viruses to use the eye as a portal of entry in an experimental in vivo setting, just as previous in vitro studies have demonstrated that numerous human ocular cell types distributed throughout the ocular area support infection and replication with both avian and human influenza viruses [42] [43] [44] . Cumulatively, these previous in vitro studies suggest that the apparent ocular tropism associated with viruses of the H7 subtype is not due to a superior ability to replicate in ocular cells compared with other virus subtypes. Future studies evaluating potential fine receptor specificity differences on the ocular surface and the composition of ocular mucins which may restrict exposure to the ocular epithelial surface to selected virus subtypes may provide a greater understanding of this property. Consistently high titers of human influenza viruses in ocular samples following both i.n. and i.o. inoculation indicates that these H1N1 and H3N2 viruses are not exhibiting a preferential tropism for ocular infection but more likely are a reflection of the high titers observed in the upper respiratory tract in these tissues independent of the initial inoculation route. Specifically, the nasolacrimal duct which links the ocular lacrimal sac to the nasal meatus could serve as a conduit for virus-containing fluid exchange between ocular and respiratory tract tissue [18] . Numerous reports have documented drainage of vaccine or immunizing agents to nasal tissue following topical ocular administration as well as the spread of intranasally administered solutions to the conjunctival mucosal surface [18, 45] . However, spread of virus from respiratory tract tissue to ocular tissue following i.n. inoculation with human or avian influenza viruses has not been observed previously in the ferret and only sporadically reported in the mouse, possibly due to relatively low titers of virus in nasal tissue or other anatomical differences [20, 24, 26, 46] . Detection of both human and avian influenza viruses in ferret eyes and conjunctival tissue following i.n. inoculation indicates that virus can circulate proximal to the nasal cavity and nasolacrimal canal more readily than previously considered and that more routine collection of ocular tissue during standard virus pathotyping in mammalian models is warranted to better understand the extent of viral ocular dissemination (Table 4 ). While it is unlikely that ferret grooming practices are solely responsible for virus spread between these locations, human studies with numerous respiratory viruses have demonstrated the ability to self-inoculate between ocular and respiratory sites, and it is possible that self-inoculation could be further contributing to virus spread in this model [47, 48] . Inoculation of ferrets by the ocular route with 10 6 EID 50 of selected human and avian influenza viruses in a 20 ml volume resulted in comparable results to those obtained employing a 100 ml inoculation volume, indicating that replication-independent drainage of virus to respiratory tract tissue and subsequent virus detection in NW and CW samples reported in this study was not contingent on the inoculum volume (data not shown). Visualization of virus deposition using fluorescently-tagged virus has allowed for a new understanding of dissemination patterns following in vivo inoculation. Using this technique, we confirmed previously hypothesized reports of replication-independent drainage from ocular to respiratory tract tissue [20, 33] . Viral load measured in respiratory tract tissues day 3 p.i. following i.n. or i.o. inoculation largely mirrored initial virus deposition patterns, with tissues possessing the greatest quantity of virus reflecting those sites of greatest initial virus deposition during virus inoculation (Table 3 -4, Figure 4 ). Both i.o. and i.n. inoculation using a 100 ml volume resulted in detection of virus at high titers in upper and not lower respiratory tract tissues day 3 p.i., indicating that inoculation with a reduced volume leads to limited initial virus deposition in respiratory tract tissues regardless of inoculation route. The delay in high virus titer recovery from lower respiratory tract tissues during HPAI virus infection after i.o. inoculation and the delay in onset of severe disease and lethal outcome with Thai/ 16 virus is likely due to replication-dependent spread (and not deposition) of virus to the lower respiratory tract, suggesting that during inoculation, the majority of virus was retained in conjunctival tissues, drained to the nasal turbinates, or swallowed and diverted away from the lower respiratory tract; future study of ferret lacrimal tissue in this role is warranted. Comparable delays in onset of severe disease compared with traditional i.n. inoculation were observed in a murine model of i.o. inoculation and a ferret model of aerosol inoculation, despite ultimately similar lethal outcomes [20, 32] . The finding here of H5N1 subtype viruses using the eye as a portal of entry to initiate a lethal infection, shown previously in a mouse model, underscores the risk of ocular exposure to influenza viruses, even those subtypes not typically considered to have a tropism for this tissue [20] . Despite efficient replication of seasonal H1N1 and H3N2 viruses in the upper respiratory tract of ferrets following i.o. inoculation, these viruses did not result in frequent detection of sneezing and nasal discharge, and did not transmit efficiently to naïve contacts by respiratory droplets (Table 1, Figure 5 ). Infrequent sneezing is commonly observed among influenza viruses which do not transmit efficiently by respiratory droplets and could be contributing to the reduced transmissibility seen here [28, 30, 31] . Further research is needed to better understand the virologic and immunologic properties which confer the incidence of sneezing and nasal discharge and the role of these properties in virus transmissibility [49] . Additionally, the efficiency of virus transmission by respiratory droplets following i.o. inoculation was likely influenced by the reduced initial virus deposition and subsequent limited replication in the ferret trachea, as reduced virus transmissibility by respiratory droplets was observed following i.n. inoculation when using a 100 ml but not 1 mL volume (Table 3 , Figure 4 , and data not shown). Despite similarly high virus titers and duration of virus shedding in NW samples, the presence of expelled virus particles originating from tracheal replication which was present during traditional i.n., but not i.o. or i.n. inoculation using a reduced volume, may have contributed to differing transmission efficiencies between inoculation routes and may represent a previously unrecognized role for virus replication in tracheal tissue in virus transmissibility by respiratory droplets. In contrast, virus transmission in the presence of direct contact did not differ between inoculation routes. The reduced transmissibility of virus following i.o. inoculation is in agreement with epidemiological studies which demonstrate that the majority of human cases of conjunctivitis following H7 influenza virus exposure are self-limiting [50] . Further study evaluating the shedding of virus into the environment among persons infected with influenza viruses which cause respiratory or ocular disease will shed light on potential differences in transmission dynamics independent of virus subtype. The diversity of potential exposures to influenza virus underscores the importance of studying the development of respiratory disease resulting from alternate exposure routes. This knowledge is critical for both a greater understanding of the establishment of influenza virus respiratory disease as well as differences in virus transmission dynamics following differing exposure routes. The facile dissemination of virus inoculum from ocular to nasal tissue, and the detection of virus in both NW and CW samples throughout the acute phase of ferret infection, highlights the ability for concurrent ocular and respiratory disease following influenza virus infection; not surprisingly, reports of conjunctivitis and influenza-like illness in the same individual have been documented during H7 outbreaks resulting in human infection [40, 51] . While much regarding the properties which regulate the ocular tropism of influenza viruses remains to be determined, our results highlight the potential for a range of influenza A subtypes to initiate infection through the eye and support the use of eye protection during occupational exposure to aerosols containing influenza viruses [19, 52, 53] . This study was carried out in strict accordance with recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All ferret procedures were approved by Institutional Animal Care and Use Committee (IACUC) of the Centers for Disease Control and Prevention and in an Association for Assessment and Accreditation of Laboratory Animal Care International-accredited facility. Animal studies were performed in accordance with the IACUC guidelines under protocol #2195TUMFERC-A3: ''Studies on the Pathogenesis and Transmission of Recombinant Influenza Viruses in Ferrets''. Influenza A viruses of the H7, H5, and H1 subtype used in this study are shown in Table 1 . Virus stocks were propagated in the allantoic fluid cavity of 10 day old embryonated hens' eggs as previously described [26] ; virus stocks were confirmed by sequencing to be free of mutations. The 50% egg infectious dose (EID 50 ) for each virus stock was calculated by the method of Reed and Muench [54] following serial titration in eggs. Fluorescenttagged virus (NL/219-FL) was generated using formalin-inactivated NL/219 virus and a SAIVI Antibody Alexa Fluor 680 Labeling kit (Invitrogen) per manufacturer's instructions as previously described [32] . All experiments with HPAI viruses were conducted under biosafety level 3 containment, including enhancements required by the U.S. Department of Agriculture and the Select Agent Program [55] . Male Fitch ferrets (Triple F Farms), 7 to 10 months old and serologically negative by hemagglutination inhibition to currently circulating influenza viruses, were used in this study. Ferrets were housed in a Duo-Flow Bioclean mobile clean room (Lab Products) for the duration of each experiment. Intranasal (i.n.) inoculations were performed under anesthesia as previously described using 10 6 EID 50 of virus diluted in PBS in a 1 ml or 100 ml volume [29] . Ocular (i.o.) inoculations were performed under anesthesia using 10 6 EID 50 of virus diluted in PBS in a 100 ml or 20 ml volume. Virus inoculum was administered dropwise to the surface of the right eye of each ferret and massaged over the surface of the eye by the eyelid. Ferrets were monitored daily post-inoculation for morbidity and clinical signs of infection as previously described [29] . Any ferret which lost .25% of its pre-inoculation body weight or exhibited neurological dysfunction was euthanized. Virus shedding was measured on alternate days post-inoculation (p.i.) in nasal washes (NW), conjunctival washes (CW), and rectal swabs (RS). NW and RS samples were collected as previously described [28, 29] . CW were obtained by bathing the inoculated (right) ferret eye three times with 500 ml wash solution (PBS containing Pen/Strep, Gentamycin, and BSA) and collecting the run-off in a small petri dish, then swabbing the surface and surrounding conjunctival tissue of the right eye with a pre-wettened cotton swab for 5 seconds, and placing the swab in a collection tube containing the run-off liquid. All samples were immediately frozen on dry ice and stored at 270uC until processed. To assess virus dissemination following i.n. or i.o. inoculation, three ferrets per group were inoculated with indicated viruses and euthanized 3 days p.i. for postmortem examination and collection of tissues for virus titration as previously described [29] . Tissue specimens were collected for virus titration were immediately frozen on dry ice and stored at 270uC until processed. Blood samples collected during necropsy were subjected to complete blood counts (CBCs) and serum chemistry analyses performed per manufacturer's instructions as previously described [56] . Virus transmissibility following i.o. inoculation was assessed by inoculating ferrets by the ocular route with indicated viruses and, 24 hrs p.i., placing a naïve ferret in the same cage as an inoculated ferret [to assess transmission in the presence of direct contact (DC)] or in an adjacent cage with modified side-walls to allow air exchange between inoculated and contact animals via perforations but inhibiting direct or indirect contact between animals [to assess transmission by respiratory droplets (RD)] as previously described [30] . For each i.o. transmission experiment, an aliquot of each virus stock used to characterize transmissibility in previous publications by the i.n. route was tested. NW, CW, and RS samples were collected on alternate days p.i./post-contact (p.c.) to assess kinetics of virus shedding. Serum was collected days 17-21 p.i./p.c. to measure seroconverison. Animal research was conducted under the guidance of the Centers for Disease Control and Prevention's Institutional Animal Care and Use Committee in an Association for Assessment and Accreditation of Laboratory Animal Care International-accredited animal facility. Sample titration and processing NW, CW, and RS samples were serially titrated in eggs, starting at a 1:10 dilution (NW, RS; limit of detection, 10 1.5 EID 50 /ml) or 1:2 dilution (CW; limit of detection, 10 0.8 EID 50 /ml). Virus infectivity for all samples was calculated by the method of Reed and Muench [54] . At the time samples were thawed for virus titration, RNA was extracted from CW samples using a QIAamp Viral RNA kit (Qiagen). Real-time RT-PCR was performed with a QuantiTect SYBR Green RT-PCR kit (Qiagen) using an influenza A virus M1 gene primer set to determine viral load [32] . Viral RNA copy numbers were extrapolated using a standard curve based on samples of known virus as previously described [32, 57] . Baseline levels were determined by collecting CW samples from uninfected ferrets. Tissue specimens were homogenized in 1 ml cold PBS using disposable sterile tissue grinders and clarified by centrifugation before serial titration in eggs, starting at a 1:10 dilution. Ferret eye and conjunctival tissues were rinsed with PBS prior to virus titration. Eye, conjunctival, and nasal turbinate tissues are expressed as EID 50 /ml, while all other tissues are expressed as EID 50 /g. Uninfected ferret corneal epithelial sheets were dissociated from excised whole ferret eyes following incubation in tetrasodium EDTA for determination of expression of surface sialoligosaccharides as previously described [20, 58] . To assess virus dissemination, ferrets were inoculated with NL/219-FL virus either i.o. (100 ml) or i.n. (1 ml) as previously described [32] . Fifteen minutes p.i., ferrets were euthanized and respiratory and ocular tissues were excised for ex vivo imaging using a Spectrum in vivo imaging system and Living Image 4.0 Software (Caliper Life Sciences). All ex vivo imaging was performed in triplicate. To quantify the presence of NL/219-FL virus in excised tissues, regions of interest were drawn around each tissue using Living Image 4.0 Software to obtain maximum relative efficiency values for each tissue, expressed as photons/second/cm 2 /steradian, the mean of which was generated from three ferrets per tissue as expressed in Figure 3 . Tissues for immunohistochemistry (IHC) were collected day 3 p.i with the viruses indicated. or from uninfected ferrets, fixed by submersion in 10% neutral buffered formalin for 3 days, routinely processed, and embedded in paraffin. Immunohistochemical detection of influenza A virus nucleoprotein was performed as described previously [59] . The Pearson product-moment correlation coefficient was generated to measure the correlation between viral titer and viral RNA copy number in CW sample using GraphPad Prism 5.0 (GraphPad Software, Inc.). Statistical significance for all other experiments was determined using Student's t test.

Placement of Leucine Zipper Motifs at the Carboxyl Terminus of HIV-1 Protease Significantly Reduces Virion Production

Natural HIV-1 protease (PR) is homodimeric. Some researchers believe that interactions between HIV-1 Gag-Pol molecules trigger the activation of embedded PR (which mediates Gag and Gag-Pol cleavage), and that Gag-Pol assembly domains outside of PR may contribute to PR activation by influencing PR dimer interaction in a Gag-Pol context. To determine if the enhancement of PR dimer interaction facilitates PR activation, we placed single or tandem repeat leucine zippers (LZ) at the PR C-terminus, and looked for a correlation between enhanced Gag processing efficiency and increased Gag-PR-LZ multimerization capacity. We found significant reductions in virus-like particles (VLPs) produced by HIV-1 mutants, with LZ fused to the end of PR as a result of enhanced Gag cleavage efficiency. Since VLP production can be restored to wt levels following PR activity inhibition, this assembly defect is considered PR activity-dependent. We also found a correlation between the LZ enhancement effect on Gag cleavage and enhanced Gag-PR multimerization. The results suggest that PR dimer interactions facilitated by forced Gag-PR multimerization lead to premature Gag cleavage, likely a result of premature PR activation. Our conclusion is that placement of a heterologous dimerization domain downstream of PR enhances PR-mediated Gag cleavage efficiency, implying that structural conformation, rather than the primary sequence outside of PR, is a major determinant of HIV-1 PR activation.

Human immunodeficiency virus type 1 (HIV-1) gag encodes a polypeptide Pr55 gag that can self-assemble into virus-like particles (VLPs) [1] . During or soon after virus release from cells, Pr55 gag is cleaved by viral protease (PR) into four major products: matrix (MA, p17), capsid (CA, p24), nucleocapsid (NC, p7), and p6 domains [1] . PR is encoded by pol, which is initially translated as a Pr160 gag-pol polyprotein by a ribosomal frameshift event that occurs at a frequency of 5%, resulting in the expression of Pr160 gag-pol to Pr55 gag at a ratio of approximately 1:20 [2] . Pr160 gag-pol is incorporated into virions via interactions with assembling Pr55 gag [3, 4, 5, 6, 7, 8] . Pr160 gag-pol cleavage by PR yields reverse transcriptase (RT) and integrase (IN) in addition to Gag products. The PRmediated proteolytic cleavage of Pr55 gag and Pr160 gag-pol , known as virus maturation, is essential for the acquisition of viral infectivity [9, 10, 11, 12, 13] . How PR is activated to mediate virus maturation is not completely clear. One proposal is that interaction among Pr160 gagpol molecules triggers the activation of embedded PR, which in homodimeric form mediates Gag and Gag-Pol cleavage following PR autocleavage from Pr160 gag-pol . Maintenance of the Pr55 gag / Pr160 gag-pol expression ratio is critical to virus assembly; the artificial overexpression of Pr160 gag-pol or PR drastically reduces virion production as a result of enhanced Gag processing by overexpressed PR activity [14, 15, 16, 17, 18, 19, 20] . Equally important is the Pr160 gag-pol sequence and structure, since sequence mutations upstream or downstream of PR often result in defective virus maturation or Gag cleavage [4, 21, 22, 23, 24] . Impaired Gag cleavage is assumed as being due, at least in part, to impaired PR activation, which is likely secondary to inadequate PR dimer interaction. Since natural RT is heterodimeric [25, 26] , there is speculation that RT in the Gag-Pol context facilitates Pr160 gag-pol -Pr160 gag-pol interaction via RT-RT interaction, which in turn influences PR activation. Consistent with this scenario, RT deletion mutations can lead to severely impaired PR-mediated Gag processing [23] . In addition, efavirenz (EFV), a nonnucleoside reverse transcriptase inhibitor that enhances RT dimerization in vitro [27, 28] , reduces virus production as a result of greatly enhanced Gag and Gag-Pol cleavage [29, 30] . Furthermore, a single amino acid substitution in RT (W402A) leads to significantly reduced virus production due to markedly enhanced PR-mediated Gag cleavage [31] . Combined, these data suggest that the RT domain plays an important role in PR activation by influencing PR dimer interaction. It is likely that altered conformation induced by the RT mutation significantly impacts PR dimer interaction, resulting in premature or impaired PR activation. Accordingly, structural conformations rather than specific sequences may be major determinants of the PR activation process. A protein sequence unrelated to HIV-1 but possessing dimerization capacity may therefore promote PR activation by facilitating PR dimer interaction when fused to the end of PR. To test this possibility, we removed the RT and IN sequences and placed a leucine zipper (LZ)-coding sequence at the C-terminus of PR. Results indicate that LZ placement significantly reduced virion release due to enhanced Gag cleavage, similar to observations for RT W402A mutations. These results support the hypothesis that the placement of heterologous protein dimerization sequences downstream of PR can significantly enhance Gag processing efficiency by promoting PR activation. To determine whether forced PR dimer interactions affect virus assembly and processing, we fused a LZ protein dimerization domain either singly or in tandem repeat to the C-terminus of an HIV-1 Gag-Pol truncated construct (Gag/PR), which is virusassembly competent but processing-defective [23] . The resulting constructs were designated PRWz and PRWWz (Fig. 1) . We used PRKz and PRKKz constructs containing the dimerizationdefective LZ mutant version (Kz) as controls. Kz fusion to PRWz at the wt LZ C-and N-termini yielded constructs PRWKz and PRKWz, respectively. Each mutant was transiently expressed in 293T cells. Virus particle assembly and processing were analyzed by Western immunoblotting. The results shown in Figure 2A indicate that Gag/PR transfectants produced substantial quantities of VLPs at levels that were near wild-type. Unprocessed Gag (the Gag precursor Pr55) and incompletely processed Gag (the intermediate p41gag) represent two major Gag products compared to wt in our supernatant and cell samples ( Fig. 2A , lanes 1 vs. 2). This is consistent with a report stating that a deletion in a downstream pol sequence significantly impairs PR-mediated virus maturation [23] . PRWz and PRWWz transfectants expressed readily detectable Gag, but produced barely detectable virus-associated Gag, suggesting a severe defect in virus assembly or release ( Fig. 2A , lanes 3 and 4). In contrast, cells transfected with PRKz or PRKKz released readily detectable (though incompletely processed) Gag (lanes 5 and 6), similar to the Gag/PR scenario. VLP levels produced by PRWKz and PRKWz (lanes 7 and 8) were inbetween those produced by PRWz and PRKz. Since enhanced or premature Gag cleavage by PR can lead to significantly reduced virus release, and since the wt LZ fusion-containing constructs exhibited higher ratios of cellular p24 gag to Pr55 gag compared to those found in Gag/PR cell lysates ( Fig. 2A , lanes 3-4 vs. lane 2), we suggest that the LZ-associated virus production defect was largely due to enhanced Pr55 gag cleavage efficiency. Since cell samples were collected between 48 and 72 h posttransfection, it is possible that Gag processing reached a level of stability that prevented us from detecting any differences in efficiency between the wt form and mutants. To test this possibility, and to confirm the effect of LZ domain placement on Gag processing efficiency, we collected samples at 24 and 48 h following the transient expression of wt and mutants. We observed that both PRWz and PRWWz showed significantly higher cellular p24 gag /Pr55 gag ratios compared to those of wt or Gag/PR (Fig. 2B) . Although PRWKz and PRKWz showed cellular Gag processing profiles similar to those of PRKz and PRKKz at 48 h, they displayed higher p24 gag /Pr55 gag ratios compared to Gag/PR at 24 h post-transfection ( Fig. 2B upper panel, lanes 15-16 vs. lane 10) . This suggests that LZ domain placement significantly enhanced Gag processing efficiency. The virus-associated Gag precursor that we detected may reflect, at least in part, the release of assembled Gag molecules that escaped PR-mediated cleavage (Fig. 2B, lanes 7-8) . To determine whether LZ placement affected virion production by wt or assembly-competent mutants in trans, we coexpressed PRWz or PRWWz with the wt or the HIV-1 protease-defective mutant D25, and observed that virus-associated Gag was markedly reduced when D25 was cotransfected with either PRWz or PRWWz at a 1:1 ratio (Fig. 2C ). Similar results were observed when the PRWz or PRWWz was coexpressed with a wt HIV-1 expression vector (data not shown). Combined, these results suggest that (a) PRWz and PRWWz both provided functional PR, and (b) the LZ-triggered virion assembly defect was primarily due to a higher Gag cleavage efficiency. The PRWWz transfectant frequently expressed a lower Gag level compared to other constructs (Fig. 2B, lane 12) , likely the result of increased proteolytic degradation of Gag mediated by the PR. To determine whether the PRWz or PRWWz assembly defect is directly associated with viral PR activity, we treated PRWz and PRWWz transfectants with Saquinavier, an HIV-1 PR inhibitor (designated as PI). As expected, virus-associated PRWWz Gag (Pr55 gag and p41 gag ) that was previously undetectable (or barely detectable) became readily detectable when PI concentrations were gradually increased (Fig. 3A) . We noticed that wt cellular p24 gag was still easily detected, even though PRWWz p24 gag was undetectable under the same treatment conditions (Fig. 3A , lanes 3 vs. 7). This suggests that PRWWz is more susceptible to a protease inhibitor than wt, despite having higher Gag cleavage efficiency. Since we centrifuged the culture supernatant through 20% sucrose cushions, we assumed that the recovered Gag would be present in pelleted particles. To confirm that the recovered Gag was from VLPs, we observed supernatant samples ( Fig. 3B ) with a transmission electron microscope, and found spherical wt and mutant Gag particles with electron-dense cores in PI-treated transfectant samples (Figs. 3D and 3F). However, mature virions with cone-shaped cores were only detected in non-PI-treated wt transfectant samples (Fig. 3C ). Some vesicles lacking cores were noted, but virion-size particles containing electron-dense cores were not detected in mock-transfected samples, or barely detected in PRWWz transfectant supernatant that had not been treated with PI (Figs. 3E and 3G). These data support the hypothesis that the LZ-incurred assembly defect is PR activity-dependent. We looked for correlations between enhanced PR-mediated Pr55 gag cleavage efficiency and increased Gag-PR-LZ multimerization capacity. Believing that the potent Gag assembly domain might determine chimera multimerization status, we predicted that the contribution of LZ to enhanced chimera multimerization would be barely (if at all) detectable. We therefore assessed chimera multimerization capacity in a Gag assemblydefective context. After constructing an assembly-defective mutant (designated MoGag) and confirming that the mutation significant-ly impaired Gag assembly (Fig. 4B ), we cloned PR-LZ chimeras into MoGag. To block the effect of PR activity on Gag-PR-LZ chimera assembly assays, all chimeras were introduced into a PRinactivated HIV-1 Pr160 gag-pol -expression plasmid GPfs [4] , with gag and pol in the same reading frame (Fig. 4A ). Results from repeated independent experiments indicate that chimeras with predicted molecular weights were detected in both supernatants and cell lysates following transient expression in 293T cells, suggesting that all chimeras were capable of assembly and release to some extent. However, we noted that MoGagfsWz transfectants produced more chimeric VLPs than MoGagfsD or MoGagfsKz (Figs. 4C and 4D). To confirm that MoGag was multimerizationdefective and that LZ did enhance assembly-defective Gag multimerization, we subjected wt Gag and each mutant to velocity sedimentation analyses. A non-myristylated (myr-) Gag mutant [32] known to be severely defective in both membrane binding and multimerization served as a negative control. Our data indicate that most of the wt Gag was recovered at fractions 3 to 5; in contrast, most myr-Gag and substantial amounts of MoGag were recovered at fractions 1 and 2. Portions of MoGagfsD and MoGagfsKz were detected at lower sucrose density fractions, whereas MoGagfsWz was almost completely recovered at higher sucrose density fractions (Fig. 4E ). Unlike MoGagfsWz, which mostly sedimented at fractions 4 and 5, considerable amounts of MoGagfsWKz and MoGagfsKWz were also recovered at fraction 3, and low but detectable amounts were observed in fraction 2. This sedimentation pattern was similar to that of MoGagfsKz (Fig. 4E , three bottom panels). Also similar to MoGagfsKz, both MoGagfsWKz and MoGagfsKWz were incapable of efficiently assembling into chimeric VLPs (data not shown). We observed this result in repeated independent experiments. This finding suggests that when fused to the PR C-terminus, a LZ tandem repeat containing a wt and mutant LZ (WKz or KWz) does not enhance Gag-PR multimerization as effectively as a wt LZ tandem repeat (Wz). This may partly explain why PRWKz and PRKWz showed relatively lower Gag cleavage efficiency compared to PRWWz (Fig. 2) . Compared to MoGag, which had significant amounts of Gag detected at lower sucrose density fractions (1 and 2), MoGagfsD molecules were mostly sedimented at fractions 3 to 5, a difference that may be explained in part by p6 pol -PR contributing to MoGagfsD multimerization via PR dimer interaction. Although GagfsD presented an efficient multimerization profile, it produced VLPs at a relatively lower level compared to WtGag (Figs. 4C and 4D). This may have been due to its lack of p6 gag , which is required for efficient virus budding [33, 34] . Our next question was whether LZ enhanced the cleavage efficiency of the assembly-defective mutant MoGag. MoGag with an LZ fusion was expressed in a Gag/PR context, thus expressing both Pr55 gag and PR containing Gag-Pol or Gag-PR(-LZ) fusions (Fig. 5A ). As expected, the insertion of LZ into MoGag/PR at the PR C-terminus resulted in significantly enhanced Gag cleavage efficiency (Fig. 5C, lanes 11 vs. 15 ), suggesting that the LZ enhancement of Gag-PR multimerization is correlated with increased PR-mediated Gag cleavage efficiency. Inefficient Gag cleavage, likely due to either impaired PR activation as a result of Pol truncation or to inhibited PR activity due to a protease inhibitor, resulted in improved MoGag VLP assembly (Fig. 5C , 12 and 15-18 ). This suggests that the MoGag assembly defect is PR activity-dependent, at least in part. The MoPRWWz had significantly higher cellular p24 gag /Pr55 gag ratios compared to MoGag/PR, but slightly lower compared to PRWWz (Figs. 5B and 5C), suggesting that the multimerization-defective Gag mutation reduced the LZ-mediated enhancement of Gag cleavage. It also supports the proposal that the Gag assembly domain plays a role in PR activation. Although myristylation is not required for Gag-Pol viral incorporation [7, 35] , it is essential for Gag multimerization and virus assembly [11, 36] . To determine if a myristylation signal is necessary for the LZ enhancement effect on Gag cleavage, we introduced myr-into wt, Gag/PR, PRWWz, and PRKKz, and measured the Gag processing efficiency of each mutant. Our results indicate an efficient Gag processing profile for myr-PRWWz, similar to that of its myristylation-positive counterpart, PRWWz (Fig. 6 ). Since myristylation is essential to Gag membrane binding and virus assembly, we failed to detect virus-associated Gag products in any of the myr-supernatant samples (data not shown). This suggests that the LZ insertion made a significant contribution to enhanced Gag cleavage, regardless of the presence or absence of a myristylation signal. Despite a lack of direct evidence, it is generally accepted that Gag-Pol molecule dimerization or multimerization triggers HIV-1 PR activation, which mediates Gag and Gag-Pol cleavage. Here we demonstrated that the insertion of LZ at the C-terminus of PR triggers markedly enhanced PR-mediated Gag cleavage efficiency, which is associated with increased Gag-PR multimerization capacity. This suggests that an HIV-1-unrelated protein sequence capable of self-association can enhance Gag cleavage efficiency when fused to the PR C-terminus. It also provides evidence in support of the assumption that Gag-Pol/Gag-Pol interaction triggers PR activation. In an assembly-competent Gag/PR context, the wt LZ insertion resulted in markedly reduced, PR activity-dependent virus production (Fig. 2) . Cellular Pr55 gag was barely detected in PRWWz at 12 and 24 h post-transfection, while most Pr55 gag remained unprocessed in wt transfectants 12 h post-transfection (data not shown). Combined, these results suggest that LZenhanced Gag cleavage is associated with premature Gag processing due to premature PR activation. We suggest that when fused at the PR C-terminus, the LZ domain facilitates PR-PR interaction via enhanced interactions among MA-CA-NC-PR-LZ chimeras. Gag is subsequently cleaved by PR in trans, either by PR embedded in the chimera, or by a mature and fully processed PR dimer. Several researchers have suggested that the PR-mediated initial cleavage occurs via an intramolecular mechanism [37, 38, 39, 40, 41] . Although we did not search for the presence of a PR dimer, PR was theoretically capable of being released from the chimera as a fully processed dimer, since the cleavage site at the PR/LZ junction remained intact. Compared to the wt LZ fusion, the placement of a mutant LZ (Kz) at either the N-or Cterminus of the wt LZ (Wz) resulted in reduced Gag cleavage efficiency (Fig. 2) . This may be attributable to the inability of WKz or KWz to promote Gag-PR multimerization (Fig. 4E) . Although myristylation is required for Gag membrane binding (which may in turn promote efficient Gag multimerization [32] ), we found evidence that myr-Gag/Pol or MoGag/Pol were capable of mediating Pr55 gag processing (Figs. 5 and 6 ). This finding suggests that neither membrane association nor an assembly-competent Gag domain is essential for the activation of PR embedded in Gag-Pol. It is likely that myr-Gag-Pol can still undergo dimerization to a level that is sufficient to trigger PR activation. Previous studies have shown that myr-Gag-Pol can efficiently cleave Pr55 gag in trans and be packaged into Pr55 gag VLPs [4, 7, 35] . The multimerization defect as a result of membrane binding apparently does not significantly compromise the LZ enhancement of PR-mediated Gag cleavage. The next question is why premature PR-mediated Gag and Gag-Pol cleavage do not occur during virus assembly, given that multiple assembly domains outside of protease promote Gag-Pol multimerization. One possibility is that the transframe peptide p6 pol may play a role in modulating PR dimer interface interaction, thus preventing premature PR activation. Due to a blocking mutation at the p6 pol /PR cleavage, p6 pol -retaining PR is defective in mediating virus maturation [42, 43, 44, 45] , suggesting that a fully functional PR requires the removal of p6 pol . However, to our knowledge there are no reports on the role of p6 pol in PR dimer interaction in a Gag-Pol context. We found that a Gag-Pol mutant with deleted p6 pol was incapable of efficiently processing coexpressed Pr55 gag , which argues against the possibility of p6 pol playing a role in suppressing PR activation [46] . According to one recent study, the insertion of a larger reporter sequence in the p6 pol region markedly impairs virus maturation, whereas partial substitution with a heterologous sequence does not [47] . This suggests that structural conformation, rather than a specific sequence upstream of PR, is important for determining PR activation. Although the RT domain is essential for proper PR-mediated Gag cleavage, the RT homodimerization domain has no enhancement effect on Gag cleavage [48] unless EFV (an HIV-1 RT-dimerization enhancer) is added to culture medium [23, 49] . In contrast, we found that LZ is capable of triggering Gag cleavage enhancement when placed at the PR C-terminus. The RT domain apparently plays a role in preventing premature PR activation. Previous studies suggest that the downstream structural conformation of PR is important in terms of modulating PR activation. First, a single RT amino acid substitution (W402A) that is not known to have major impacts on in vitro RT dimerization [50] markedly enhances Gag processing [31] . An additional partial deletion at the C-terminus of p66RT not only reverses W402Atriggered Gag cleavage enhancement, but also markedly impairs virus maturation. In contrast, truncated Gag-Pol mutants that retain intact p66 or p51 RT domains still respond to the enhancement effect of W402A on Gag processing [31] . Second, substitution mutations in RT are capable of neutralizing the enhancement effect of EFV on Gag processing [49] . Combined, these data suggest that conformational changes in Gag-Pol may significantly affect PR dimer interface interactions, leading to either premature or insufficient PR activation. Our finding that the fusion of HIV-1-unrelated dimerization protein sequences at the C-terminus of PR sharply enhanced Gag cleavage strongly supports the hypothesis that structural conformation, rather than specific sequences, largely determines PR activation status. The RT structure domain in the Gag-Pol context may help prevent PR from premature activation via a conformation that prevents the Gag and Gag-LZ chimeras. HIV-1 Gag domains, pol-encoded PR, and the transframe peptide p6 pol are indicated; fs denotes a frameshift mutation that forces gag and pol into the same reading frame. The fused leucine zipper (LZ) domains wt (Wz) or mutant (Kz) at the PR C-terminus are indicated as described in the Figure 1 legend; x denotes substitution mutations in CA, NC (NC15A), and PR that blocked either Gag assembly or PR activity. (B-E) Assembly and multimerization of HIV-1 Gag mutants. 293T cells were transfected with designated constructs. At 48-72 h post-transfection, culture supernatants and cells were collected and subjected to Western immunoblotting (panels B and C). (D) Gag-associated proteins from medium or cell samples were quantified by scanning immunoblot band densities (C). Ratios of Gag in media to Gag in cells were determined for each construct, and compared with wt release level by dividing the release ratio for each chimera by the wt ratio. (E) Velocity sedimentation analysis of cytoplasmic Gag precursor and Gag-LZ chimeras. Homogenized and extracted cytoplasmic lysates were centrifuged through consecutive 25%, 35%, and 45% sucrose gradients at 130,0006g for 1 hour. Fractions were collected from the top of each gradient. Aliquots of each fraction were subjected to SDS-PAGE (10%) and probed with a monoclonal antibody directed against HIV-1 CA. doi:10.1371/journal.pone.0032845.g004 PR dimer interface from interacting efficiently. This conformation may change during Gag-Pol packaging, thus supporting more efficient PR dimer interaction. Pettit et al. [41] have demonstrated that inactivated PR dimer interface mutations can be compensated for to some extent by extra PR sequences in the Gag-Pol context. Altered PR dimerization kinetics or activity has been identified in several studies of PR-containing C-or N-terminal extensions into Gag-Pol [21, 23, 24, 38, 40, 43, 51, 52] . In most cases, it is not feasible to assay the impacts of these constructs on virus assembly or PR-mediated virus maturation, due to overlapping gag and pol reading frames. Our approach provides a convenient system for analyzing the impacts of mutations on PR dimer interactions by assessing virus particle assembly and processing. Studies are underway to determine if LZ sufficiently compensates for the inactivation of PR dimer interface mutations. To place a leucine zipper (LZ) in frame into the HIV-1 PR Cterminus, we engineered a pBRCla-Sal plasmid cassette containing an HIV-1 coding sequence (from ClaI-nt.831 to SalI-nt.5786) and a pcDNA3.1-myc/hisA polylinker inserted at the IN Cterminus. DNC(wtZip) and DNC(Kzip) [53] served as templates for amplifying the respective wt and mutant LZ domains of human CREB [54] using the forward primer 59-CGGGATCCTGGAG-GAGGACGAGAGTGTCGTAG-AAAGAAG-39 and reverse primer 59-CCAAGCGGCCGCGATTTGTGGCAG-TAT-39. Plasmids containing the wt and mutant LZ (provided by E. Barklis [55] ) were used to construct the DNC(wtZip) and DNC(Kzip). The human CREB LZ sequence is 284-RE-CRRKKKEYVKCLENRVAVLENQNKTLIEELKALKDLYC-HKSD-327. The underlined amino acid residues E298, R300, E305, Q307, I312, E314, and L321 were mutated to Lys, and N308 was mutated to His, yielding the mutant LZ [54] . Amplified fragments were digested with BamHI and NotI and ligated into the pBRCla-Sal cassette, yielding constructs PRWz and PRKz, respectively. To add the LZ copy, PCR-amplified wt and mutant LZ fragments were purified, digested with a restriction enzyme, and ligated into the PRWz and PRKz, yielding the constructs PRWWz, PRWKz, PRKKz, and PRKWz. The Gag assembly-defective mutant MoGag was constructed by recombining the CA mutant M39A/W184A/M185A with NC mutant NC15A, which was kindly provided by P. Spearman [56] . NC15A has 15 NC-basic residues replaced with alanine. M39A/ W184A/M185A was created by overlapping PCR with the following mutagenic primers: for M39A, 59-CTGATAGCGCT-GAAAATGCGGGTATCA-39, and for W184A/M185A, 59-CAACAACGTTTCTGTAGCCGCATTTTTTAC-39. The Mo-Gag mutation was cloned into the indicated PR-leucine zipper constructs. As described previously, Gag/PR has deleted RT and IN coding sequences [23] . In D25, Arg is substituted for the PR catalytic residue Asp [20] . The backbone of all expression constructs is the HIV-1 proviral plasmid HIVgpt [57] . 293T cells were maintained in DMEM supplemented with 10% fetal calf serum. Confluent 293T cells were trypsinized, split 1:10 and seeded onto 10-cm dish plates 24 hours before transfection. For each construct, 293T cells were transfected with 20 mg of plasmid DNA by the calcium phosphate precipitation method (18) , with the addition of 50 mM chloroquine to enhance transfection efficiency. Culture media from transfected 293T cells were filtered through 0.45 mm-pore-size filters, followed by centrifugation through 2 ml of 20% sucrose in TSE (10 mM Tris-HCl pH 7.5, 100 mM NaCl, 1 mM EDTA) plus 0.1 mM phenylmethylsulfonyl fluoride [PMSF]) at 4uC for 40 min at 274,0006g (SW41 rotor at 40,000 rpm). Viral pellets then were suspended in IPB (20 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.1% SDS, 0.5% sodium deoxycholate, 1% Triton X-100, 0.02% sodium azide) plus 0.1 mM PMSF. Cells were rinsed with ice-cold PBS (phosphate-buffered saline), scraped from the plates, collected in 1 ml of PBS and pelleted at 2,500 rpm for 5 min. The cell pellets were resuspended in 250 ml of IPB plus 0.1 mM PMSF, and then subjected to microcentrifugation at 4uC for 15 min at 13,7006g (14,000 r.p.m.) to remove cell debris. Either supernatant or cell sample was then mixed with equal volumes of 26 sample buffer (12.5 mM Tris-HCl pH 6.8, 2% SDS, 20% glycerol, 0.25% bromophenol blue) and 5% b-mercaptoethanol and boiled for 5 min. Samples were subjected to SDS-PAGE and electroblotted onto nitrocellulose membranes. Membrane-bound Gag proteins were immunodetected using an anti-p24 gag (mouse hybridoma clone 183-H12-5C) monoclonal antibody at a 1:5,000 dilution from ascites. The secondary antibody was a rabbit anti-mouse (HRP)-conjugated antibody at 1:15,000 dilution as appropriate and the procedures used for HRP activity detection followed the manufacturer's protocol (Pierce). Immunodetected bands on film were quantified by using AlphaImager 2000 (Alpha Innotech Corp.) and Image J software. Cells were rinsed twice with PBS, pelleted and resuspended in 1 ml TEN buffer containing Complete protease inhibitor cocktail followed by homogenization. The cell lysates then were centrifuged at 3,000 rpm for 20 min at 4uC. Five hundred ml of the postnuclear supernatants were mixed with an equal amount of TEN buffer, and were then applied to the top of a pre-made 25-45% discontinuous sucrose gradient. This gradient was prepared in TEN buffer containing 1 ml of each of 25%, 35%, and 45% sucrose. The gradient was then centrifuged at 130,0006g for 1 hour at 4uC. Five 0.8-ml fractions were collected from the top of the centrifuge tubes. The proteins present in aliquots of each fraction were precipitated with 10% TCA and subjected to western blot analysis as described in the membrane flotation assay. Virus-containing supernatants were centrifuged through 20% sucrose cushion. Concentrated viral sample was placed for 2 min onto a carbon-coated, UV-treated 200 mesh copper grid as described [58] . Sample-containing grids were rinsed 15 s in water, drained off water with filter paper, and stained for 1 min in filtered 1.3% uranyl acetate. Staining solution was drained off by applying filter paper to the edge of the grid. Grids were left to dry before viewing in a JOEL JEM-2000 EXII transmission electron microscope. Images were collected at 20,0006 and 60,0006.

Using Support Vector Machine and Evolutionary Profiles to Predict Antifreeze Protein Sequences

Antifreeze proteins (AFPs) are ice-binding proteins. Accurate identification of new AFPs is important in understanding ice-protein interactions and creating novel ice-binding domains in other proteins. In this paper, an accurate method, called AFP_PSSM, has been developed for predicting antifreeze proteins using a support vector machine (SVM) and position specific scoring matrix (PSSM) profiles. This is the first study in which evolutionary information in the form of PSSM profiles has been successfully used for predicting antifreeze proteins. Tested by 10-fold cross validation and independent test, the accuracy of the proposed method reaches 82.67% for the training dataset and 93.01% for the testing dataset, respectively. These results indicate that our predictor is a useful tool for predicting antifreeze proteins. A web server (AFP_PSSM) that implements the proposed predictor is freely available.

Antifreeze proteins (AFPs) are functional proteins in a cell. With special antifreeze activity, AFPs make the organisms less sensitive to cold temperatures. AFPs bind to small ice crystals to inhibit growth and recrystallization of ice that would otherwise be fatal [1] . By contributing to both freeze resistance and freeze tolerance, AFPs have helped to increase species diversity in some of the harshest and most inhospitable environments. Freeze resistance involves the inactivation or removal of ice-nucleating agents in freeze-avoiding species, whereas freeze tolerance involves the activation or synthesis of ice-nucleating agents in winter in freeze-tolerant species [2, 3] . AFPs have been found in various insects, fish, bacteria, fungi, and overwintering plants such as gymnosperms, ferns, monocotyledonous, angiosperms, etc. [4] [5] [6] [7] [8] [9] [10] [11] [12] . Relational analyses show that there is low sequence or structure similarity for an ice-binding domain, and lack of common features among different AFPs [7] [8] [9] [10] . One reason for this phenomenon is that ice can present many different surfaces with different arrangements of oxygen atoms [8] . So it is difficult to establish powerful prediction methods to identify AFPs. However, AFPs play important roles in different fields, such as freeze-resistant transgenic plants and animals, food technology, preservation of cell lines, organs and cryosurgery [13, 14] . How to discriminate AFPs from other proteins is important in understanding protein-ice interactions and creating new ice-binding domains in other proteins. Many lines of evidences have indicated that computational approaches can provide useful information for both drug discovery and basic research in a timely manner [15] , such as protein subcellular location prediction [16, 17] , structural bioinformatics [18] , identification of proteases and their types [19] , identification of membrane proteins and their types [20] , molecular docking [21] [22] [23] , identification of enzymes and their functional classes [24] , and signal peptide prediction [25, 26] . Up until now, there are few studies using computational approaches to discriminate AFPs and non-AFPs. Kandaswamy et al. [27] investigated this problem using the predictor of Random Forest. That is the first and the only method utilizing machine learning technique to deal with the prediction of AFPs. With the model AFP-Pred, they obtained 81.33% accuracy from training and 83.38% from testing. Although high accuracy has been achieved, the problem is worthy of further investigation because the performance of the aforementioned method is still not fully satisfactory and they do not provide an online web server for predicting antifreeze proteins. In this study, we focus on developing a new antifreeze protein predictor by seeking a more informative encoding scheme. After a preliminary evaluation of different encoding schemes, we found that the evolutionary information in the form of PSSM profiles is suitable for representing the antifreeze protein sequence. Then a predictor called AFP_PSSM is established using the feature PSSM-400 as the input of support vector machine (SVM). AFP_PSSM yields 82.67% accuracy from training dataset and 93.01% accuracy from test dataset. This indicates that our predictor is very promising and may at least play an important complementary role to existing methods. The proposed predictor is freely available at the web server AFP_PSSM [28] . For a query protein sequence of 500 amino acids, it will take about 20 s for the web server to yield the predicted result; the longer the sequence is, the more time it needs. According to a recent review [29] , to establish a really useful statistical predictor, the following four procedures need to be considered: (i) construct or select a valid benchmark dataset to train and test the predictor; (ii) formulate the protein samples with an effective mathematical expression that can truly reflect their intrinsic correlation with the attribute to be predicted; (iii) introduce or develop a powerful algorithm to conduct the prediction; (iv) properly perform cross-validation tests to objectively evaluate the anticipated accuracy of the predictor; (v) establish a user-friendly web server for the predictor that is accessible to public. Below, let us describe how to cope with these procedures one by one. The datasets used in this paper is retrieved from Kandaswamy et al. [27] which consists of 481 antifreeze proteins and 9493 non-antifreeze proteins. To get rid of redundancy and homology bias, the sequences with ≥40% sequence similarity have been removed using program CD-HIT [30] . Then the training dataset contains 300 antifreeze proteins randomly selected from the 481 antifreeze proteins and 300 non-antifreeze proteins randomly selected from the 9493 non-antifreeze proteins. The test dataset contains the remaining 181 antifreeze proteins and 9193 non-antifreeze proteins. These datasets can be freely downloaded from [31] . To develop a powerful predictor, one of the keys is to formulate the protein sequences with an effective mathematical expression that can truly reflect their intrinsic correlation with the attribute to be predicted [17] . To realize this, some popular sequence-based encoding schemes have been investigated to represent each protein sequence. Evolutionary information, one of the most important types of information in assessing functionality in biological analysis, has been successfully used to encode protein in many applications, such as our previous work of lysine ubiquitylation site prediction [32] , transmembrane protein topology prediction [33] and malaria parasite mitochondrial protein prediction [34] . To extract the evolutionary information, the profile of each protein sequence is generated by running Position Specific Iterated BLAST (PSI-BLAST) program [35, 36] . Then this information can be represented as a two dimensional matrix which is known as the PSSM of the protein. In this paper, the PSSM of each protein sequence in the constructed dataset is generated against the non-redundant Swiss-Prot database [37] (version 56, released on 22 July, 2008) using the PSI-BLAST program with three iterations (−j 3) and e-value threshold 0.0001 (−h 0.0001). This matrix is composed of L  20 elements, where L is the total number of residues in a peptide. The rows of the matrix represent the protein residues and the columns of the matrix represent the 20 naive amino acids. Each element represents the probability of the occurrence of each 20 amino acid when it's mutated to the others at one position during the evolution process. In the view of the fact that SVM requires the fixed length feature vectors as their inputs for training, we generate a vector of dimension 400, called PSSM-400 from the PSSM. PSSM-400 is composition of occurrences of each type of amino acid corresponding to each type of amino acids in protein sequence [38] . Thus for each column we have a vector of dimension 20. Figure 1 shows the schematic representation of transformation each protein sequence into PSSM-400. The purpose of calculating composition of proteins is to transform the variable length of protein sequence into fixed length feature vectors [33] . This is a necessary step during classification of proteins using SVM. The transformation of each protein sequence into a vector of 20 dimensions using amino acid composition will encapsulate the information of protein. Besides amino acid composition, dipeptide composition is also utilized, which gives a fixed pattern length of 400. The advantage of dipeptide composition compared with amino acid composition is that it encapsulates both the fraction information of amino acids and the local order information of protein sequence. PROTEIN A R N ∷ V T -4 -1 0 ∷ 0 V -1 -3 -3 ∷ 4 V 0 -3 -3 ∷ 4 F -2 -3 -3 ∷ -1 C 0 -4 -3 ∷ -1 G 0 -3 0 ∷ -3 K -1 5 0 ∷ -3 L -2 -3 -4 ∷ -1 S 1 -1 1 ∷ -2 G -1 -3 0 ∷ -3 K -1 2 0 ∷ -3 P -1 -2 -2 ∷ -3 ∷ ∷ ∷ ∷ ∷ ∷ R -2 6 -1 ∷ -3 The Chou's pseudo amino acid composition (PseAAC) encoding scheme feature has been widely used to predict various properties of proteins [39] [40] [41] [42] [43] . It can be calculated as following: 20 Where ω is a weighting factor (default ω = 0.1). Support vector machine (SVM) [45] belongs to the family of margin-based classifier and is assumed to be a very powerful method to deal with prediction, classification, and regression problems. SVM look for optimal hyperplane which maximizes the distance between the hyperplane and the nearest samples from each of the two classes. Formally, given a training vector x i R n and their class values y i {−1, 1}, i = 1, …, N, SVM solve the following optimization problems: where w is a normal vector perpendicular to the hyperplane and ξ i are slake variables for allowing misclassifications. Here C (>0) is the penalty parameter which balances the trade-off between the margin and the training error. In this study, LIBSVM package [46, 47] with radial basis kernel function is used. Two parameters, the regularization parameter C and the kernel width parameter γ are optimized based on 5-fold cross-validation using a grid search strategy. Ten-fold cross validation [48] is used in this work. The dataset is randomly divided into ten equal sets, out of which nine sets are used for training and the remaining one for testing. This procedure is repeated ten times and the final prediction result is the average accuracy of the ten testing sets. To reduce the computational time, we also adopt the independent testing dataset cross validation in this study as done by [49] to evaluate our model. Three parameters, sensitivity (S n ), specificity (S p ), and accuracy (Acc) are used to measure the performance of our model. They are defined by the following formulas: where TP, TN, FP and FN stand for true positive, true negative, false positive and false negative, respectively. Moreover, we create ROC (receiver operating curve) for all of the models in order to evaluate the performance of models using different encoding schemes. The detailed flowchart of our work is shown in Figure 2 . First, sequential evolution information in form of PSSM profiles for the input sequence is generated by PSI-BLAST. Second, the obtained PSSM is further transformed into PSSM-400 vector. Finally, the predictor AFP_PSSM is applied to output the test results. For the convenience of experimental scientists, we give a step-by-step guide on how to use it to get the desired results as follows: (i) Open the web server AFP_PSSM [28] and you can see the prediction page on your computer screen, as shown in Figure 3 . You must input your email address since the prediction process may take a long time; (ii) Input your query protein sequence to the text box in Figure 3 . Note that the input protein sequence should be in the FASTA format. The FASTA format sequence consists of a single initial line beginning with a greater-than symbol (">"), followed by lines of amino acid sequence. You can click on the "example and note" button to see the example protein sequence; (iii) Choose a threshold value in the drop-down list. For prediction with high confidence (less probability of false positive prediction), high threshold should be chosen; (iv) Click on the submit button to see the predicted result. For example, if you use the first sequence in the example page, the predicted result will be "0.847538, yes" as can be seen in Figure 4 , which means that the protein is an antifreeze protein with the probability of 0.847538. It takes about 15 s for a protein sequence of 300 amino acids before the predicted result appears. In this section, four SVM models based on amino acids composition, dipeptides composition, Chou's PseAAC and PSSM-400 are constructed respectively. The accuracies and receiver operating characteristic (ROC) curves for these four SVM models are shown in Table 1 and Figure 5 . One can see that PSSM-400 encoding scheme performs better than the others with accuracy of 82.67% and AUC (Area Under Curve) of 0.926. Thus we use it as our final encoding scheme to represent antifreeze protein sequences. In order to further examine the prediction of power of the current classifier, we compare our predictor AFP_PSSM with the recent work of Kandaswamy et al. [27] on the testing dataset. The number of antifreeze proteins and non-antifreeze proteins in the testing dataset are highly imbalanced, and this situation is close to reality. The compared results are shown in Table 2 . As can be seen from the table, the predictor proposed in this study obtains accuracy of 90.17%, higher than the accuracy of 83.38% gained by [27] . The better prediction performance may be credited to the appropriate protein sequence encoding scheme adopted in our prediction model. Accurate identification of new antifreeze proteins is important in understanding ice-protein interactions and creating novel ice-binding domains in other proteins. Though some researchers have focused on this problem, the accuracy of prediction is still not satisfied, and there are few online web servers for predicting antifreeze protein sequences. In this paper, a highly accurate method is developed for predicting antifreeze proteins using support vector machine and evolutional profiles. This is the first paper in which evolutionary information in the form of PSSM profiles has been utilized to predict antifreeze proteins. The proposed predictor is freely available at the web serve AFP_PSSM [28] .

Reliability and External Validity of AMSTAR in Assessing Quality of TCM Systematic Reviews

Objective. The aim of this study is to measure the reliability and external validity of AMSTAR by applying it to a sample of TCM systematic reviews. Study Design and Methods. We tested the agreement, reliability, construct validity, and feasibility of AMSTAR through comparisons with OQAQ. Statistical analyses were performed by using SPSS 13.0. Results. A random of sample with 41 TCM systematic reviews was selected from a database. The interrater agreement of the individual items of AMSTAR was moderate with a mean kappa of 0.50 (95% CI: 0.26, 0.73). The ICC for AMSTAR against OQAQ (total score of 9 items, excluding item 10) was 0.87 (95% CI: 0.76, 0.93). Conclusions. Although there is room for improvement on few items, the new tool is reliable, valid, and easy to use for methodological quality assessment of systematic reviews on TCM.

Traditional Chinese medicine (TCM) is one of the rarely existing traditional medicines that hold systematic theories as well as preventative and therapeutic methods for diseases in practice [1] . Since the 1950s, research methods in modern medicine (Western medicine) have been gradually introduced to TCM studies; the first TCM systematic review appeared in medical journals in the late 1990s, and now hundreds to thousands of reviews have been published on the area of TCM [2, 3] ; systematic reviews have become the standard approach in assessing and summarizing primary studies. However, even though systematic reviews proved to be useful, serious consideration must be given to how they were conducted; high methodological quality of systematic reviews is a prerequisite for recommendation of the use or avoidance of a TCM intervention. At present, either researches or methods for assessing quality of systematic reviews are still not fully developed in TCM and there is substantial room for improvement [2, 3] ; even in Western medicine current instruments for assessing methodological quality of systematic reviews are still suboptimal and need revision and updating [4, 5] . A new tool termed AMSTAR, an acronym for "a measurement tool to assess systematic reviews," was developed strictly upon the OQAQ (Overview Quality Assessment Questionnaire) [6] , the Sack checklist (quality assessment checklist) [7] , and three additional items. This tool is an 11-item questionnaire requiring assessors to answer "yes," "no," "cannot answer," or "not applicable"; and a recent study reported that AMSTAR has good agreement, reliability, construct validity, and feasibility to assess the quality of systematic reviews [8] . However, these psychometric properties within AMSTAR were tested by having it apply to only a limited set of systematic reviews from Western medicine; a further step is needed to assess its validity and reliability, by a broader range of assessors and more samples of reviews in diverse circumstances [8] . Will the results be reproducible when assessing methodological quality of systematic reviews on traditional chinese medicine (TCM)? The answer is not clear yet, further validation is necessary. The aim of this study is to validate the reliability and external validity of AMSTAR by applying it to a sample of TCM systematic reviews. We have adopted the definitions used by the Cochrane 2 Evidence-Based Complementary and Alternative Medicine Collaboration: a systematic review is a review of a clearly formulated question, that which uses systematic and explicit methods to identify, select, and critically appraise relevant researches, and to collect and analyze data from the studies included in the review [9] . Then a search strategy for locating systematic reviews on TCM was formulated by using the definitions; search terms included "Chinese medicine", "Chinese herb", "plant preparations", "Chinese medical formula," "meta analysis," "meta-analysis," "meta-analyses," and "systematic review." We performed a comprehensive search on CNKI (China National Knowledge Infrastructure), CBM (Chinese Biomedical Database), VIP (Chongqing VIP periodicals Database), Medline and EMbase databases (January 1, 1999 to the end of 2008) in addition to hand search on Chinese Journal of Evidence-Based Medicine (incept to the end issue of 2008), Chinese Journal of Integrated Traditional and Western Medicine (from the first issue in 1999 to the end issue in 2008), and Journal of Chinese Integrative Medicine (from the first issue in 1999 to the end issue in 2008). Two reviewers screened the titles and abstracts of identified studies independently, disagreement was resolved by discussion. A total of 165 systematic reviews on TCM were included, full detail of evaluations has been reported separately [10] . We used a computer-generated random sample (approximately 25% of the 165 systematic reviews) as a test set for validation. We tested the agreement, reliability, construct validity, and feasibility of AMSTAR through comparisons with OQAQ, the latter was a validated scale (overview quality assessment questionnaire, OQAQ) developed by Oxman and Guyatt in 1991 [6] . The OQAQ scale measures across a continuum using nine questions (items 1-9) designed to assess various aspects of the methodological quality of systematic reviews and the tenth item requires assessors to assign an overall quality score on a seven-point scale [6] . In order to adhere more faithfully to the guidance provided by AMSTAR and OQAQ, two assessors (KDY, WY) performed a separate translation and conducted a pilot test independently. Each translator prepared a separate translation and the difficulty in obtaining conceptually equivalent expressions in Chinese was assessed too; subsequently, a sample of 2 systematic reviews on TCM was taken to perform pilot test by two authors independently; based on the results and consensus from reviewers, the final Chinese versions of AMSTAR and OQAQ were then developed for formally evaluating methodological quality of Chinese systematic reviews; all inconsistencies identified either in translation or in application were resolved by discussion. On the basis of this, we constructed a data extraction form in Chinese for this study, in which 11 items of AMSATR and 10 items of OQAQ were adopted directly. In addition, publication status and reporting characteristics of systematic reviews, such as publish language, number of pages, funding source, update or not, Cochrane systematic review or not, target disease, and institution of first author, were also incorporated; besides, the time required to complete an assessment while applying AMSTAR and OQAQ was recorded too. Assessors were required to answer "yes" score or any other scores for each of items (AMSTAR and OQAQ). If an item was scored "yes," it would be given one point, otherwise, 0 point. We added up these to calculate a total score, the reliability of this total score was assessed through calculating intraclass correlation coefficients; the agreement for each item and the overall tool was explained by percentage of actual agreement as well as Kappa coefficient [5, 8] . Kappa coefficient is a popular measure for chance-corrected nominal scale agreement between two raters. We adopted the Kappa values of <0 rates as less than chance agreement, 0.01-0.20 as slight agreement, 0.21-0.40 as fair agreement, 0.41-0.60 as moderate agreement, 0.61-0.80 as substantial agreement, and 0.81-0.99 as almost perfect agreement [8] . The OQAQ was selected as a criterion tool because it had been rigorously developed, its face validity was strong, and its validity had been thoroughly tested [6] . We assessed construct validity by comparing AMSTAR, OQAQ, and a self-developed global assessment scale. Construct validity was showed by intraclass correlation coefficient (ICC); for the purpose of calculating ICC, we adopted methods used by the AMSTAR group [8] , and converted the mean total scores (mean of two assessors) of per review to the percentage of maximum score (11 points in AMSTAR and 9 points in OQAQ accordingly); in addition, we developed a 100-point rating scale for overall quality assessment based on answers to the eleven questions in AMSTAR, in which two assessors indicated his or her judgments by checking tick-marks on a horizontal line (0 to 100 point), an SR without any flaws would be scored 100 points. Meanwhile, we also adopted the item 10 in OQAQ as a validated global assessment instrument. The overall mean scores (mean of two assessors), either using the self-developed 100point rating scale or using the item 10 from OQAQ, were also taken to verify the construct validity of AMSTAR. The feasibility of AMSTAR was assessed by recording the time it took to complete scoring, and paired t-test or nonparametric test was applied when comparing with OQAQ. Database was established by using an electronic form on Microsoft Excel 2003 (Microsoft Corp., Redmond, WA); the data set extracted contained two quality ratings for each review, yielding a total of four ratings per review. Data analysis was performed by SPSS 13.0 (SPSS, Chicago, IL). P < 0.05 was considered significant. A random sample with 41 TCM systematic reviews was selected from a database developed in a previous study [10] . Of which, only 9 reviews were written in English, and the majority (78%) was published in Chinese journals . The sample included 35 paper-based reviews and 6 Cochrane reviews; there was only one updated Cochrane review. According to the International Classification of Diseases 10 (ICD-10), the topics of the reviews ranged across 9 systems, and mainly focused on diseases of circulatory system Evidence-Based Complementary and Alternative Medicine 3 (15 SRs or 37% of the sample, such as stroke and other cardiovascular diseases), infectious and parasitic diseases (7 SRs, like HBV, SARS), genitourinary system (5 SRs, such as ectopic pregnancy), digestive system (4 SRs, like ulcerative colitis), nervous system (3 SRs, such as Parkinson's disease), and musculoskeletal system and connective tissue (3 SRs, osteoporosis). The number of pages of included TCM SRs ranged widely from 2 to 80 with a median of 6 pages, of which, Cochrane systematic reviews had more pages than non-Cochrane reviews (P < 0.001), with medians of 31 (range: 16-80) and of 5 (range: 2-11), respectively. Less than half of the reviews (41.5%) were presented by clinicians. Total mean scores on AMSTAR ranged from 2 to 10 (out of a maximum score of 11) with a mean percentage score of 55.1%. The total mean quality scores on OQAQ ranged from 3 to 8 (out of a maximum score of 9) with a mean percentage score of 63.6%. The overall scores for the global assessment instrument (item 10 in OQAQ) ranged from 1 to 6 (out of a maximum score of seven) with a mean of 3.3 (95% CI: 2.9, 3.6), and overall scores using the self-developed 100-point rating scale ranged from 15 to 73 with a mean of 47.6 (95% CI: 43.4, 51.7). 3.1. Agreement and Reliability. Substantial agreements (>80% or nearly 80%) were observed in majority of the individual items (item 1, 2, 3, 4, 6, 7, 8, 10, 11) on AMSTAR, with a mean percentage of 84% (95% CI: 71.1%-91.9%); agreements on item 5 and item 9 were mild at a percentage of less than 60%. Kappa ranged widely from −0.03 to 1.00 with a mean of 0.50 (95% CI: 0.26, 0.73); five items (item 2, 3, 4, 10, 11) scored a kappa of >0.70 (0.70 to 1.00); the highly agreements in four items (item 1, item 6, item 7, item 8) ranged from 80% to 95% and inversely low kappa (−0.034, 0.40, 0.36, 0.36) may be explained by a skewed distribution of responses, that is, approximately 80% (in item 6) to 90% (in item 1, 7, 8) of reviews in which the assessors agreed on the score "yes." However, items 5 (list of included and excluded studies) and 9 (appropriate method to combine studies) scored relative low either in agreement percentage or in kappa parameter (Table 1) . Agreements on individual items of OQAQ were inferior to AMSTAR, with a mean percentage of 79% (95% CI: 64%, 89.5%), and the mean of kappa was relatively low too, with a mean of 0.35 (95% CI: 0.08, 0.62). The interobserver ICC to the total score was excellent for AMSTAR at 0.84 (95% CI: 0.70, 0.91), and superior to OQAQ, 0.67 (95% CI: 0.38, 0.82). The interrater agreement (kappa) between two assessors for the global assessment (item 10 in OQAQ) was 0.72 (95% CI: 0.48 to 0.85), and better than the self-developed 100-point rating scale: 0.50 (95% CI: 0.07-0.74). Total mean score was converted into the percentage of the maximum score for each of the instruments, the ICC for AMSTAR against OQAQ (total score of 9 items, excluding item 10) was 0.87 (95% CI: 0.76, 0.93), that is, the results of AMSTAR were highly convergence with the results of OQAQ. Besides, both overall scores were converted into the percentage of maximum score. ICC obtained when comparing AMSTAR with the item 10 in OQAQ was 0.84 (95% CI: 0.69, 0.91), and when comparing with the 100-point rating scale, ICC was at 0.81 (95% CI: 0.65 to 0.90) respectively; thus AMSTAR showed well convergence with global assessment instruments too. In addition, we compared the total scores obtained by applying AMSTAR on Cochrane reviews (n = 6, 8.42 ± 1.02) with that of non-Cochrane reviews (n = 35, 5.66 ± 1.31), and the former had higher quality score than the latter (mean difference = 2.76, P < 0.001, 95% CI: 1.62, 3.90). The relationship between quality scores and publish year was explored too, reviews published after 2005 had similar AMSTAR scores comparing to earlier reviews (5.98±1.51 versus 6.18 ± 1.76, P = 0.70). As the methodological quality and the reporting quality were not mutually exclusive addressing in AMSTAR [8] , we explored whether the number of pages had a positive or negative effect on the AMSTAR score, the result showed there was a statistical association between AMSTAR score and the number of pages (Spearman's rho = 0.67, P < 0.001). It took 13.2 (95% CI: 12.2, 14.2) minutes to complete use of AMSTAR for each review, while it took less time to complete scoring of OQAQ, averagely 9.3 (95% CI: 8.8, 9.9) minutes per review (paired difference = 3.9, P < 0.001). Besides, a linear regression analysis was performed (time = 6.35 + 2.94 × langrage + 1.75 × log(pages), P < 0.001), revealed the time needed to complete using AMSTAR had significant associations with logarithm of the number of pages (unstandardized coefficients = 1.75, 95% CI: 0.61 to 2.90) and langrage (unstandardized coefficients = 2.94, 95% CI: 0.73 to 5.15); that is, systematic reviews with more pages or written in English need more time to complete scoring. The two assessors found there's difficulty in approaching a final decision on item 9 "were the methods used to combine the findings of studies appropriate" and item 5 "was a list of studies (included and excluded) provided"; for the latter, more detailed guidance for scoring "yes" are required. A considerable amount of systematic reviews on traditional Chinese medicine have been conducted since the first TCM systematic review was published in the late 1990s [2, 3] . We selected a random sample of TCM systematic reviews from a database developed in a previous study [10] , the sample covered a wide variety of health topics, and thus, we believe that we had a representative sample of TCM reviews of which the AMSTAR was ready to apply. Our findings in this research revealed that the AMSTAR is a good choice for evaluating quality of TCM systematic reviews. The AMSTAR showed satisfactory interrater agreement, convergent validity, and feasibility in assessing methodological quality of TCM systematic reviews. Interrater reliability was evaluated by assessing the degree to which different individuals agreed on the scientific quality of a set of reports [7, 8] , the performance of AMSTAR in terms of agreement and reliability was better than that of OQAQ; overall agreement and kappa of items in AMSTAR ranged from moderate to perfect, the reliability of its total score was excellent. However, fair agreement and relatively low kappa were observed in item 9 "were the methods used to combine the findings of studies appropriate" and item 5 "was a list of studies (included and excluded) provided," indicated that there is a room for improvement on AMSTAR when applying this new tool to assess methodological quality of systematic reviews on TCM. In the absence of a gold standard, we assessed construct validity by comparing AMSTAR with OQAQ as well as two global assessments. Construct validity was shown by intraclass correlation coefficient (ICC); this statistic reflects the extent to which the results of AMSTAR converge with the results of other "criterion" instruments. The analysis revealed that the construct validity was excellent, that is, the AMSTAR is a reliable and valid tool. Given the extremely strict implementation of Cochrane systematic review, such reviews conducted with high methodological quality have been widely recognized [52] ; the AMSTAR revealed Cochrane systematic reviews have higher quality scores than non-Cochrane systematic reviews, that is, the AMSTAR has an ability to discriminate methodological quality, so it is sensitive when applying it to a sample of systematic reviews in diverse quality. The relationship between AMSTAR quality score and the number of pages can be explained by the fact that Cochrane reviews always have considerable amount of pages, these extreme outliers determined the direction and strength of the association. Compared with the results reported by Shea et al. [8] , considerable differences exists either on ICC for AMSTAR contrast to the OQAQ, or on items with low kappa values. Shea et al. reported a lower ICC of 0.66 (95% CI: 0.28, 0.84), and different items (item 4 and item 7) with relatively low kappa parameter. Possible explanations for these differences include (i) the different samples of systematic reviews being evaluated, most SRs in our sample were short, published in Chinese, and more likely to be rated as low quality, such less conversely evaluations could produce a higher ICC value in our study; (ii) the extra procedure of translation and the conducts of evaluations by applying two tools simultaneously may act as a kind of consensus training to make the evaluations of AMSTAR more likely convergent with the results of OQAQ, that lead to a higher ICC value too in the present study; (iii) the background, skills, and expertise of the assessors were different from that of Shea et al. study. Regarding the feasibility of AMSTAR, the time needed to complete scoring showed this new tool is feasible in assessing quality of TCM systematic reviews too; it took about 13 minutes on average to complete an assessment and showed well applicability. The statistical analysis revealed that the AMSTAR was slightly more time consuming in contrast to OQAQ; there may be several explanations for this. First, the AMSTAR has 11 items, longer than the 10 items in OQAQ; second, the AMSTAR was developed based on two instruments, including OQAQ itself, so items in the instruments may be overlapped, and it would take less time to complete scoring when filling replicated items; third, the sequence of applying tools may be another explanation, as we conducted the assessment in the order of first AMSTAR and then OQAQ, assessors need more time to look through a systematic review to facilitate it in the first round of assessment by using AMSTAR, while in the second round by applying OQAQ, such time would be saved. The reliability analysis revealed that the Kappa was poor to fair in some items on AMSTAR. As kappa coefficient shows the proportion of agreement beyond that expected by chance alone, it is a popular measure for chance-corrected nominal scale agreement between two raters [5, 8] ; however, if distribution of item responses is skewed or over concentrated on either the "yes" or the "no" category, the kappa coefficients will become unstable and invalid, and no longer suitable for measuring agreement between two raters, so new methods for calculating valid agreement coefficients are needed to explore in future study. Such items shown in our study included item 1 "was an 'a priori' design provided," item 6 "were the characteristics of the included studies provided," item 7 "was the scientific quality of the included studies assessed and documented," and item 8 "was the scientific quality of the included studies used appropriately in formulating conclusions"; however, the relatively low kappa in item 5 and item 9 cannot be explained by the limitation of kappa statistic. AMSTAR proved to be feasible to apply in quality assessment of TCM systematic reviews; the main problems emerged were the absence of guidance for certain item response, such as item 5 "was a list of studies (included and excluded) provided," to get "yes" score, four situations may be encountered: list of included studies provided, list of excluded studies provided, both lists of included and excluded studies provided in the same time, and the characteristics of the included studies presented; it is difficult to reach a final conclusion without more detailed directions regarding its use. Those items (4 and 7) with relatively low kappa values identified by Shea et al. [8] , on the contrast, presented more precisely guidance, were easy to apply and easily reached consensus among raters. As the assessment was undertaken by two assessors, one assessor was with expertise in clinical epidemiology and clinical research methods, and the other was a novice user to these quality assessment instruments, thus it could possibly result in underestimation of the reliability of AMSTAR. Another limitation in this study is lack of backward translation for the adapted tools, the translation into Chinese may produce a different measurement instrument with different properties. The current Chinese version should be translated back into English by a third party, and the back translations would be compared with the original tools to ensure the conceptual equivalence. However, the absence of back translation may offset somewhat in present study by a check of accuracy with a previous Chinese version of two instruments tools [10] . Although both instruments proved to be useful in this study, the performance of AMSTAR in terms of reliability and validity was better than OQAQ; the new tool is reliable, valid, and easy to use when applied to assess methodological quality of systematic reviews on TCM, although there is room for improvement on a few items.

Innate Immune Response of Human Alveolar Macrophages during Influenza A Infection

Alveolar macrophages (AM) are one of the key cell types for initiating inflammatory and immune responses to influenza virus in the lung. However, the genome-wide changes in response to influenza infection in AM have not been defined. We performed gene profiling of human AM in response to H1N1 influenza A virus PR/8 using Affymetrix HG-U133 Plus 2.0 chips and verified the changes at both mRNA and protein levels by real-time RT-PCR and ELISA. We confirmed the response with a contemporary H3N2 influenza virus A/New York/238/2005 (NY/238). To understand the local cellular response, we also evaluated the impact of paracrine factors on virus-induced chemokine and cytokine secretion. In addition, we investigated the changes in the expression of macrophage receptors and uptake of pathogens after PR/8 infection. Although macrophages fail to release a large amount of infectious virus, we observed a robust induction of type I and type III interferons and several cytokines and chemokines following influenza infection. CXCL9, 10, and 11 were the most highly induced chemokines by influenza infection. UV-inactivation abolished virus-induced cytokine and chemokine response, with the exception of CXCL10. The contemporary influenza virus NY/238 infection of AM induced a similar response as PR/8. Inhibition of TNF and/or IL-1β activity significantly decreased the secretion of the proinflammatory chemokines CCL5 and CXCL8 by over 50%. PR/8 infection also significantly decreased mRNA levels of macrophage receptors including C-type lectin domain family 7 member A (CLEC7A), macrophage scavenger receptor 1 (MSR1), and CD36, and reduced uptake of zymosan. In conclusion, influenza infection induced an extensive proinflammatory response in human AM. Targeting local components of innate immune response might provide a strategy for controlling influenza A infection-induced proinflammatory response in vivo.

Alveolar macrophages (AM) reside at the air-tissue interface in the lung and are one of the first lines of defense that interact with inhaled microorganisms and particles [1] . They play a critical role in homeostasis, host defense, and tissue remodeling [2] , and they are readily infected by influenza [3] . AM express many pattern recognition receptors (PRRs) to help recognize the pathogenassociated molecular patterns (PAMPs) on the surface of microorganisms [4, 5] . They are important in initiating response to influenza, regulating the inflammatory response, and potentially limiting secondary bacterial infections [6] . Influenza A virus causes seasonal and pandemic flu, both of which pose significant public health burdens. Influenza viral antigens have been detected in AM from humans and many animal species [7] [8] [9] [10] [11] [12] [13] [14] [15] , and AM are critical for controlling viral replication in vivo [11, 14] . Recently, several groups have explored the responses of human monocyte-derived macrophages to avian and/or seasonal flu viral infection using a genome-wide approach [16] [17] [18] . Avian H5N1 and human H1N1 and H3N2 viruses induce increases in similar groups of genes despite the stronger response induced by pathogenic avian viruses compared to seasonal flu viruses in human monocyte-derived macrophages [16] [17] [18] [19] . However, the genome-wide response of resident human AM to influenza infection has not been reported. Our previous study showed that cultured primary human AM support a productive infection with H5N1 but not H1N1 and H3N2 influenza viruses though AM express both avian and human influenza receptors [3, 19] . However, human monocyte-derived macrophages support productive infection with both human and avian viruses [19] [20] [21] . These results suggest that the response of human AM to influenza might be different from the response of human macrophages derived from peripheral blood [22] . The purpose of our study was to use a genome-wide approach to define the innate immune response of human AM to influenza. Using H1N1 influenza virus PR/8, we performed gene profiling of virus-infected human AM at 4 and 24 h post inoculation (hpi) and verified the alterations in IFN-related genes by real-time RT-PCR and cytokine response by ELISA. We investigated the kinetics of infection-induced cytokine response in human primary AM infected with both live and UV-inactivated PR/8 and the contemporary H3N2 virus A/New York/238/2005 (NY/238) [23] . We also determined if the cytokine response was amplified by paracrine proinflammatory cytokines, TNF-a and IL-1b. In addition, we explored whether influenza infection diminishes gene expression of macrophage scavenger receptors, which could contribute to the impaired ability of AM to clear other pathogens after influenza. Viral infection resulted in significant alterations of mRNA levels in 1,347 transcripts at 4 hpi and 2,152 transcripts at 24 hpi; these transcripts mapped to 1,077 (4 hpi) and 1,493 (24 hpi) known genes. Tables 1 and 2 show the top 25 genes that were up-regulated or down-regulated by influenza virus. The complete list of altered genes is listed in Data S1. To identify the cellular functions and pathways affected by the infection, the array data were processed by Ingenuity Pathway Analysis (IPA) using IPA version 8.0 (IngenuityH Systems, Redwood City, CA), which associates differentially regulated genes with known specific biological pathways based on information from published literature (www.ingenuity.com). The results from IPA indicate some functional groups of genes were changed at both time points. These genes are involved in antimicrobial and inflammatory responses, cell death, cancer, infection mechanisms, cellular growth and proliferation, cell-mediated immune responses, and immune cell trafficking. Interferon regulatory factor (IRF) activation and PRR signaling were the most prominent pathways activated by viral infection at both time points. In addition, retinoic acid-induced gene-1 (RIG-I) and interferon (IFN) signaling were dominant at 4 hpi, whereas homeostasis-related pathways such as IL-10 and IL-6 were activated at 24 hpi. As shown in Table 1 at 4 hpi, seven out of the top ten PR/8 upregulated genes were type I IFN family members, and another two up-regulated genes, IFN-stimulated gene (ISG) 20 and CXCL11, were IFN-stimulated genes [24] . At 24 hpi, IFN-stimulated genes CXCL9-11, and IFITM1 were among the top ten genes upregulated by PR/8 (Table 2) . Therefore, we verified the microarray data with a focus on IFN-associated genes by real-time RT-PCR ( Figure 1 ). As shown in Figure 1A , PR/8 infection induces an early response in type I IFN genes IFNA1 and IFNB as well as type III IFN genes IL-29 and IL-28A, although the degree of increase was slightly less than that of most type I IFN genes (Tables 1 and 2) . Along with the increased IFN gene expression, the infection also increased expression of well-known PRR genes associated with IFN production. mRNA levels of RIG-I and melanoma differentiation associated protein-5 (MDA-5) were mainly increased at 4 hpi, whereas TLR3 and 7 were mainly stimulated at 24 hpi ( Figure 1B ). The infection also significantly increased mRNA of IFN-stimulated anti-viral genes myxovirus (influenza virus) resistance 1 (MX1), 2959 oligoadenylate synthase (OAS), and IFN-stimulated gene 56 (ISG56) ( Figure 1C and Data S1), as compared to control cells. In addition to IFN related genes, PR/8 significantly increased the expression of many cytokine genes and cytokine-regulated genes. These included the proinflammatory cytokines TNF-a (7.3-fold at 4 hpi and 25-fold at 24 hpi), IL-1a (6.9-fold at 24 hpi), and IL-1b (2.3-fold at 4 hpi and 16.3-fold at 24 hpi). We verified the alteration in IL-1a and IL-1b by real-time RT-PCR (data not shown). The infection also upregulated expression of TNF-a induced proteins 2, 3, 6, and 8, as well as TNF receptor family members 9 and 10. Expression of IL-1 family members interleukin 1 receptor 1 (IL-1R1) and the IL-1 receptor antagonist (IL-1Ra) was also increased (Data S1). In addition, PR/8 infection up-regulated mRNA expression of many chemokine genes including CC chemokines CCL2-5, and CCL20, as well as CXC chemokines CXCL9-11 (Data S1 and Tables 1 and 2). CXCL9-11 were markedly increased when compared to controls both in the microarray studies and in additional verification studies (Tables 1 and 2 and Figure 1D ). CCL5 was the most increased CC chemokine (232-fold at 4 hpi, 234-fold at 24 hpi) (Tables 1 and 2) . Besides the increased mRNA expression of proinflammatory mediators, PR/8 also increased mRNA expression of the antiinflammatory cytokine IL-10 (1.9-fold at 4 hpi and 2.8-fold at 24 hpi), its receptor (8-fold at 4 hpi and 6.8-fold at 24 hpi), and suppressor of cytokine signaling (SOCS)1 (9.4-fold at 4 hpi and 12.1-fold at 24 hpi) and SOCS3 (5.4-fold at 4 hpi), which have been shown to be important in turning off inflammatory responses and dampening a robust innate immune response [25] . PR/8 also upregulated expression of IL-6 (36.9-fold at 4 hpi and 66.6-fold at 24 hpi), another important cytokine responsible for homeostasis, and several cytokines that activate and regulate adaptive immune response, especially IL-15 and its receptor, IL-23A, and IL-27 [26] [27] [28] (Data S1). We verified the putative increases in secreted cytokines and chemokines at the protein level by ELISA in 8-14 additional donors. As shown in Figure 2 , PR/8 infection significantly increased secretion of cytokines of TNF-a, IL-6, IFN-a, and IL-29, and CXC chemokines CXCL8-11 as well as CC chemokines CCL2, 4 and 5. Consistent with the mRNA data, CXCL10, CXCL11 and CCL5 were the main chemokines induced by the virus, and AM secreted slightly more IFN-a than IL-29 ( Figure 2 ). Secretion of cytokines and chemokines occurs without the release of a significant amount of infectious viral particles From our previous studies we knew that AM do not release a significant amount of infectious virus particles after infection with human influenza viruses [3, 19] . To investigate whether the infected macrophages were synthesizing viral proteins, we performed a time-course infection experiment in AM from an additional 4 donors using both live ( Figure 3A -D) and UVinactivated PR/8 viruses ( Figure 3E ), examined the kinetics of viral antigen synthesis by staining hemagglutinin (HA) or nucleoprotein, and measured secretion of selected cytokines by ELISA. As shown in Figure 3F , there was a slight increase in viral production at 6 hpi, when about 20% of the cells expressed viral antigens and then no more net increase in viral release as up to 80% of the cells expressed viral proteins by 48 hpi. The viral antigen staining was due to viral replication, since there was no signal with UV-inactivated virus ( Figure 3E ). Despite the abortive release of infectious virus, PR/8 infection induced a time dependent cytokine and chemokine response in human AM ( Figure 3G -K). Viruses triggered an early and rapid secretion of IFN-a and CXCL10 at 6 hpi. Secretion of CCL5 and CXCL8 followed the pattern of the viral protein synthesis increasing with time. The virus-induced increase of TNF-a peaked at 24 hpi and then declined. UV-inactivation abolished the virus-stimulated TNF-a production, significantly decreased secretion of IFN-a, CXCL8, and CCL5. However, the inactive virus was able to stimulate a strong CXCL10 response, although the degree was slightly smaller than that from live PR/8 ( Figure 3I ). The different patterns of the induction suggest that the cytokine response may involve different regulatory mechanisms. In addition, we compared the alterations in mRNA levels of selected innate immune response genes at 3 and 24 hpi for both UV-inactivated PR/8 and live PR/8 infections. Consistent with the protein data, both live and UV-inactivated PR/8 stimulated a large increase in CXCL10 mRNA at both time points. UV-inactivated PR/8 stimulated an up to 4 fold increase of CCL5 and IFNA1. UV-inactivated virus did not alter mRNA levels of RIG-I, TLR7, or ISG56 at either time point (data not shown). These results indicate that viral replication is required for most selected innate immune responses but not required for the CXCL10 response. To investigate whether the results observed with PR/8 can be extended to contemporary human influenza virus infection, we performed a time-course experiment using a H3N2 virus NY/238, a influenza virus rescued by reverse genetics technology based on a swab sample from a patient from New York during the winter of 2005 [23] . Consistent with the results from PR/8 infection, human AM do not support a productive NY/238 infection as verified by no increase in infectious viral particles released from infected culture as measured by plaque assay (data not shown). As shown in Figure 4A , NY/238 infection markedly stimulated CXCL10 mRNA, NY/238 virus also triggered an early increase in the expression of RIG-I and IFNA1 genes and increased mRNA levels of antiviral gene ISG56 and CCL chemokine CCL5. Inoculation with the same amount of UV-inactivated NY/238 virus was able to stimulate an IFNA1 and CXCL10 response. However, the response was smaller than that observed with live virus. Unlike PR/8, NY/238 virus did not induce a significant increase in TLR7 mRNA ( Figure 4A) . At the protein level, NY/238 virus induced a similar response as PR/8 virus in terms of cytokine and IFN production ( Figure 4B ). Consistent with the finding with PR/ 8, viral replication was required for most chemokine and cytokine response and but was not requisite for CXCL10 release. Because PR/8 infection increased secretion of TNF-a and increased gene expression of IL-1 family members, well-known proinflammatory mediators that cause release of inflammatory chemokines, we were interested in the impact of these proinflammatory mediators on the overall chemokine response during the infection in human AM. Our hypothesis was that TNF and IL-1 signaling would augment chemokine secretion in a paracrine manner [29] . As shown in Figure 5 , neutralization of TNF pathway by its soluble receptor significantly decreased secretion of CXCL8 by 65% (P,0.001) and CCL5 by 53% (P,0.05), but did not alter secretion of IFNs, CXCL10, or TNF-a itself. Blockade of the IL-1 receptor by its naturally occurring receptor antagonist IL-1Ra [30] had a similar effect. When the activity of both cytokines was inhibited, there was no further reduction in chemokines greater than that of a single inhibitor, although there was a slight decrease in CXCL10 response in the presence of both inhibitors, the response was not statistically significant. AM are important phagocytes and express many scavenger receptors. The microarray experiments indicated that PR/8 infection also significantly decreased mRNA levels of many macrophage receptors especially at 24 hpi (Table 2 and Data S1). We, therefore, investigated the impact of influenza infection on expression of scavenger receptors by real-time RT-PCR. Consistent with the results from microarray experiments, PR/8 infection significantly decreased the mRNA levels of CLEC7A (Dectin 1), macrophage scavenger receptor 1 (MSR1), CD36, and the mannose receptor C type 1 (MRC1) but did not change the expression of MRC2. However, we were not able to confirm the decrease of MARCO, due to the large variation in responses among different donors ( Figure 6A ). To further investigate if the decrease in macrophage receptor expression was associated with functional consequences, we evaluated the uptake of zymosan, which are yeast walls recognized by CLEC7A, and heat-killed S. aureus. As shown in Figure 6B , PR/8 infection reduced uptake of zymosan by AM at 24 hpi in a dose dependent manner. We did not observe a significant cell loss or cytopathic effect at 24 or 48 hpi, although most cells were infected as seen in Figure 3A -C. In addition, PR/8 infection did not affect uptake of heat-killed S. aureus until 72 hpi, when the infection induced a significant cytopathic effect (data not shown). Alveolar macrophages produce a robust innate immune response to influenza. This includes a significant induction of cytokines and chemokines, pathogen recognition, and apoptotic responses, which are similar to the responses of human monocyte derived macrophages [16, 17] . Consistent with other studies of avian or human influenza infections in humans and animals [16, 17, [31] [32] [33] , PR/8 stimulated an early and prominent IFN response in human AM despite of the failure to release infectious viral particles. Human AM produce both type I and type III interferons (Figures 1 and 2) . In contrast, alveolar epithelial cells do not produce any type I interferon IFN-a in response to influenza [34] . These results indicate a cell-specific pattern in producing IFN in response to viral infection. It is well known that RIG-I like RNA helicases (RLHs) and TLRs are the two main PRRs responsible for IFN production against RNA viruses including influenza. RLHs (RIG-I and MDA-5) recognize cytoplasmic viral double-stranded RNA, whereas TLRs (TLR3 and TLR7) sense viral nucleic acid in the endosomal compartment [35, 36] . In the current study, PR/8 infection up-regulated mRNA levels of RIG-I and MDA-5 mainly at 4 hpi, but the mRNAs of TLR3 and 7 mainly at 24 hpi ( Figure 1B) , which suggests that RLHs might be the early sensors and TLRs might be the late sensors for PR/8 in human AM. These results correlate well with those reported by Takeuchi and Thompson that RLHs were responsible for local production of IFNs, whereas TLRs were mainly involved in the late stages of systemic infection [35, 36] . At early times PR/8 triggered mainly pro-inflammatory responses, whereas at later times PR/8 also activated pathways involved in the maintenance of homeostasis such as the activation of IL-10 and IL-6, as well as up-regulation of SOCS genes (Data S1). Therefore, therapeutic regulation of the inflammatory response in acute lung injury should consider both strategies to inhibit secreted cytokines but also strategies to dampen the innate immune response by stimulating IL-10 and SOCS genes. We were able to confirm the results found with PR/8 in contemporary influenza virus NY/238-infected human AM with the exception of an increase in TLR7 mRNA. This might be due to a lower MOI of virus used in the experiments because of the limitation of the viral titer, but it could also be due to differences in the natures of these two viruses or the difference in methods for propagating these two viruses. CXCL9-11 were the most highly induced chemokines by influenza viruses as verified at both mRNA and protein levels (Figures 1 and 2) . These three chemokines bind to a common receptor CXCR3, and the importance of CXCR3 signaling has been shown in the pathogenesis of several viruses including influenza [32, [37] [38] [39] . CXCL10 is highly induced in avian flu (H5N1)-infected ferrets, non-human primates, and human cells including alveolar epithelial cells and monocyte-derived macrophages [16] [17] [18] 32, 33, 40] , and has been viewed as a prognostic marker for several viral infections [37, 39, 41, 42] . In mice, the peak level of CXCL11 mRNA coincides with the peak of the viremia [43] , and the CXCL11 protein has been reported to have antiviral activity [44] . In addition, all three CXCR3 ligands can induce epithelial cell chemotaxis and proliferation and perhaps accelerate epithelial wound repair during the resolution of viral infections [45, 46] . The robust induction of CXCL9, 10, and 11 in both AM (Figures 1 and 2) and human alveolar type II cells [34] as well as the distinct CXCL10 response induced by both live and UV-inactivated influenza virus PR/8 and contemporary virus NY/238 (Figure 3 and 4) suggest that this family of proteins likely plays an important role in the human lung alveolar defense against influenza infection, which will require further study. The response of alveolar macrophages was different in a several ways from that reported for human monocyte derived macrophages. The major difference is that alveolar macrophages infected with human influenza viruses do not release much infectious virus, whereas human monocyte-derived macrophages do ( [19, 20] and Figure 3 ). The mechanism for the non-productive infection was not investigated in this study and is likely complicated. One of the possible mechanisms might be related to the lack of gene expression of transmembrane protease serine S1 member 2 (TMPRSS2) and human airway trypsin-like protease (HAT) by human AM (microarray data not shown). Both TMPRSS2 and HAT are type II transmembrane serine proteases [47] possessing trypsin-like activity and are known to be important for cleaving influenza HA required for productive infection [48] . In recent studies Bottcher et al suggest that TMPRSS2 is mainly responsible for cleavage of newly synthesized HA, whereas HAT cleaves both endocytosed and newly synthesized HA [49] . Therefore, lack of these two gene products in human AM may partially explain the lack of released infectious virus by these cells. In addition, both PR/8 and NY/238 viruses induced an early activation of type I IFN, especially IFN-a (Table 1 and Figures 1, 3, and 4) . The strong anti-viral property of type I IFN [50] may also contribute to the non-productive infection in these cells. Further studies will be required to understand the mechanism for the failure of release of infectious viral particles by human AM. In addition, inactivation of influenza by UV did not abolish the influenza viruses-stimulated CXCL10 secretion by AM (Figures 3 and 4) , which is different from studies with human monocytederived macrophages [21, 51] and with human alveolar type II epithelial cells isolated from the same donors ( [34] and data not shown). In those studies, release of CXCL10 is totally dependent on viral replication. The mechanism for the distinct CXCL10 response in human AM will require additional and carefully designed studies. The differences between human AM and monocyte-derived macrophages indicate the importance of investigating the response of AM to influenza infection during the initial phases of infection in the lung because AM are main targets for both human and avian influenza viruses [19] . Chemokine and cytokine responses are required for protection of the host against viral infection. However, an exuberant response contributes to the influenza-induced morbidity and mortality, especially in severe pandemic and avian influenza infections [16, 52] . In the current study, PR/8 infection induced an increase in TNF-a and IL-1b, well-known paracrine proinflammatory factors. Therefore, we hypothesized that inhibiting these factors might reduce the influenza-induced-inflammatory response. Since the contemporary virus NY/238 induced a similar cytokine and chemokine response as PR/8, it would be reasonable to expect that the regulation of chemokine and cytokine in contemporary influenza infection might also be similar to PR/8 infection. As shown in Figure 5 , inhibiting TNF and/or IL-1 decreased more than 50% of the PR/8-induced secretion of inflammatory chemokines CXCL8 and CCL5 but did not truly affect type I interferon or CXCL10 response, although we observed a decrease of CXCL10 in the presence of both inhibitors ( Figure 5 ). TNF and IL-1 signaling are known to be regulated by NF-kB and there are several NF-kB binding sites in the promoter of CXCL10 [53] , despite of the fact that CXCL10 is an IFN-induced protein [24] . This may explain why inhibiting both pathways slightly decreased the amount of CXCL10 from infected AM. Our results suggest that short term targeting the critical paracrine factors might be beneficial for controlling the excessive infiltration of inflammatory cells and acute lung injury during pandemic or avian flu infection in vivo. Of course, this would require careful consideration of time and dose so as not to increase secondary bacterial infections. Influenza infection significantly decreased mRNA level of macrophage receptors CLEC7A, MSR1, CD36, and MRC1 ( Figure 6A and Table 2 ). CLEC7A belongs to the C-type lectin family and functions as a PRR that recognizes a variety of beta-1, 3-linked and beta-1, 6-linked glucans from fungi. A decrease of CLEC7A in infected AM suggests that these cells might not efficiently recognize and engulf fungi after influenza infection. As shown in Figure 6B , the uptake of zymosan, a yeast cell wall component containing beta-1-3-glycosolic linkeages, was decreased in a dose-dependent manner in PR/8-infected human AM. This effect was not associated with cell loss or cytopathic effect because we did not observe a significant cytopathic effect ( Figure 3B ) even at a MOI of 1 (data not shown). However, the explanation of the decreased uptake might be more complicated than simply the loss of this receptor. In addition, other macrophage receptors MSR1, MARCO, CD36, as well as mannose receptor MRC1 are important for bacterial and particle uptake [54] [55] [56] . Mice with deletions of MSR1 or CD36 have increased susceptibility to pneumococcal or staphylococcal pneumonia [57] [58] [59] . Although impairment of macrophage phagocytosis of bacteria after influenza in mice is well recognized [60, 61] and secondary bacterial infection after influenza is a common clinical problem, we were not able to detect a significant decrease in uptake of heat-inactivated S. aureus in human AM until 72 hpi, at which time the cytopathic effect was significant. We did not observe a consistent decrease of MSR1 protein by flow cytometry in PR/8-infected AM, which might explain why the infection did not impair the bacterial uptake (data not shown). We were also not able to verify the decrease of mRNA level of MARCO, another important macrophage scavenger receptor for influenza infections in mice and human cells [54, 56, 58, 62] . Nine of 11 donors showed a decrease in mRNA levels of MARCO after infection with PR/8 ( Figure 6A ). Two other donors had an increase in levels of MARCO mRNA. Therefore, changes of bacteria-related receptors in human AM after influenza require additional studies, and there may be variations in response among individuals. In summary, we performed a global profiling of innate immune response and regulation with a focus on chemokine and cytokine response in influenza-infected human AM. Human AM are apparently different from human monocyte derived macrophages in their ability to release infectious virus and the CXCL10 response to UV inactivated virus. Future studies should compare these responses in peripheral and alveolar macrophages from the same donors. In addition, during acute lung injury, short term targeting of paracrine inflammatory factors such as TNF and IL-1 as well as targeting IL-10 and SOCS genes might decrease the acute injury and allow for better gas exchange. Isolation and culture of human alveolar macrophages AM were isolated from deidentified human donor lungs, which were not suitable for transplantation and donated for medical research. We obtained the donor lungs through the International Institute for the Advancement of Medicine (Edison, NJ) and the National Disease Research Interchange (Philadelphia, PA) [3] . The Committee for the Protection of Human Subjects at National Jewish Health approved this research and has designated this research as non-human project. The isolated AM could be frozen and recovered in 90% FBS and 10% DMSO. There was no apparent difference in response with frozen or freshly isolated macrophages in terms of the level of infection and virus-induced TNF-a secretion (data not shown). AM were plated in DMEM/ 10% FBS with antibiotics, and cultured at 37uC in 10% CO 2 overnight. The cells were then washed and cultured for another day in DMEM and 1% charcoal stripped FBS with antibiotics before viral infection. Their purity was measured by staining for CD68 (Dakocytomation, Carpinteria, CA) [3] . Influenza A virus A/PR/8/34 (PR/8) was grown in 10-day-old SPF Premium Eggs (Charles River SPAFAS. North Franklin, CT) and prepared as described previously [3] . Contemporary influenza H3N2 virus, A/New York/238/2005 (NY/238), was created by reverse genetics using plasmids that corresponded to the consensuses sequence obtained from a human swab specimen collected in New York State in the winter of 2005 [23] . NY/238 was passaged in Madin-Darby Canine Kidney (MDCK) cells and the viral titer was measured by plaque assay on MDCK cells as described previously [34] . Briefly, stocks of purified virus was serially diluted in DMEM with 1 mg/ml TPCK trypsin (Sigma-Aldrich, St. Louis, MO) and used to inoculate triplicate wells of near confluent MDCK cells. After a 1 h inoculation, the inoculum was removed and the cells were overlaid with MEM with 4% FBS and 0.5% SeaKem LE agarose (Cambrex, Rockland, ME). Plaques were stained and counted after 72 h incubation at 37uC, with the agarose overlay medium containing 10% neutral red (Sigma-Aldrich). For UV-inactivation of PR/8 or NY/238, 500 ml diluted virus was placed in a 35-mm 2 petri dish on ice and irradiated twice in a UV Stratalinker (Stratagene, La Jolla, CA) at a cumulative dose of 120 mJ/cm 2 . Viral inactivation was demonstrated by plaque assay on MDCK cells as described above. On the day of infection, AM were inoculated with live PR/8 at a designated multiplicity of infection (MOI) or with the same amount of UV-inactivated PR/8 for 1 h. After inoculation, cells were washed and then cultured until harvest. Influenza infection was verified by immuno-fluorescent staining with goat antibody to the hemagglutinin of PR/8 (kindly provided by BEI Resources, Manassas, VA). For NY/238 infection, AM was inoculated with a MOI of 0.1 instead of 0.5 due to the limitation of viral titer and infection was confirmed by immuno-fluorescent staining with mouse antibody to influenza nucleoprotein (Millipore, Billerica, MA). For the inhibition experiments, cells were treated with 10 mg/ml human IL-1 receptor antagonist (IL-1Ra) [63] and extracellular TNF neutralization was achieved by treating cells with 10 mg/ml recombinant human soluble TNF receptor (sTNFR) [64] . Both IL-1Ra and/or sTNFR were added to the cells for 45 min before virus inoculation. DMEM alone was used as vehicle control for both inhibitors. After inoculation, cells were washed and cultured with the inhibitors for an additional 24 h. At 4 and 24 hpi, total RNA from virus-infected and noninfected AM from three donors was extracted and purified using RNeasy kit (QIAGEN, Valencia, CA). The samples were run on Affymetrix HG-U133 Plus 2.0 chips (Affymetrix, Santa Clara, CA) and processed as indicated by the manufacturer in the Microarray Core of the University of Colorado Denver. All data is MIAME compliant and the raw data had been deposited in a MIAME compliant database Gene Express Omnibus (GEO). The GEO accession numbers are GSM762686, GSM762687, GSM762688, GSM762689, GSM762694, GSM762695, GSM762696, GSM762697, GSM762702, GSM762703, GSM762704, GSM762705. Analyses of microarray data were performed using R statistical package from Bioconductor open source software for bioinformatics. Prior to statistical analyses, raw data from array scans were processed using the Robust Multi-chip Average (RMA) normalization method to subtract a background value [34] . After normalization, data were filtered to exclude all probe sets with an ''absent'' call in all samples and to remove transcripts that demonstrated little variation across all arrays by comparing the variances of the log-intensities for each gene with the median of all variances for the entire array. The filtered gene list was generated using the Student's T test to select statistically significant genes and corrected using the False Discovery Rate approach. Genes that had at least a 2-fold change in comparison to the uninfected controls for all three subjects were used for further analyses. Real-time RT-PCR mRNA expression of selected genes that were significantly upregulated by PR/8 or NY/238 were validated by quantitative realtime RT-PCR [34] . These genes include IFNs, PRRs, chemokines, and SOCSs. Except for IFN-b and IL-29 genes whose probes were synthesized in house [34] , the specific probes for other genes were purchased from Applied Biosystems (Applied Biosystems Inc. Foster City, CA). The expression level of each specific gene was normalized to the level of a constitutive probe cyclophilin B [34] . Supernatant from PR/8 OR ny/238-infected and non-infected cells were harvested at designated times after inoculation for the measurement of chemokine and cytokine secretion by ELISA. The ELISA kits for human CXCL9, CXCL10, CXCL11, CCL5, CXCL8, and IL-29 were purchased from ELISA Tech (ELISA Tech, Aurora, CO). The ELISA kit for IFN-a was purchased from Invitrogen (Invitrogen, Carlsbad, CA). Human AM were cultured and infected with PR/8 at the designated MOI. Uptake of zymosan or heat-killed S. aureus were performed according to manufacturer's instructions. For uptake of zymosan, cells were incubated with fluorescent-labeled zymosan A Bioparticles (Invitrogen) at a ratio of 10 particles per cell for 2 h, then cells were washed and fixed with 4% paraformaldehyde for 10 min. The uptake was analyzed by fluorescent microscopy. For uptake of heat-killed S. aureus, cells were incubated with pHrodolabeled, heat-killed S. aureus (pHrodo-SA) (Invitrogen) at a ratio of 20 particles per cell for 2 h. The cells were then washed to remove non-internalized particles, collected, and fixed with 4% paraformaldehyde. Uptake of the pHrodo-SA was analyzed on the LSR II flow cytometer (BD Biosciences) in the National Jewish Health Flow Cytometry Core, and the data were analyzed using FlowJo software (TreeStar, Ashland, OR). In addition to the PR/8 infected cells, positive control uninfected cells and negative control paraformaldehyde fixed cells were also used. Statistical analyses were conducted in GraphPad Prism version 5.0 (GraphPad Software, San Diego, CA). Pair-wise comparisons were tested for significance using Wilcoxon matched pairs test or Paired T test. Comparison among three or more groups was performed using one-way ANOVA with Tukey's post test analysis. Data S1 Human AM from 3 non-smoking donors were isolated, cultured, and infected by PR/8 virus at MOI of 0.5. The gene profiling of infected and non-infected cells at 4 and 24 hpi from each donor was examined by microarray experiments using Affymetrix HG-U133 Plus 2.0 chips (Affymetrix, Santa Clara, CA). The filtered gene list was generated as described in the Section of Methods and Materials. The data show probe ID, gene symbol, gene name, and fold change at 4 and 24 hpi. Red indicates similar results from multiple probes for the same gene, and the probe ID is the representative probe ID from several probes. (XLS)

Successful treatment of Chlamydophila pneumoniae acute respiratory distress syndrome with extracorporeal membrane oxygenator: a case report and diagnostic review

INTRODUCTION: Chlamydophila pneumoniae is a respiratory pathogen known to infect the upper and lower respiratory tracts. Infection severity can range from sub-clinical pulmonary infection to acute respiratory distress syndrome. CASE PRESENTATION: A previously healthy 62-year-old Caucasian man was admitted to our hospital for acute respiratory failure. Serum samples obtained every week starting from the day of admission showed clear-cut seroconversion for C. pneumoniae antibodies. All other cultures obtained during the first days of hospitalization were negative. Despite maximal ventilatory support (high positive end expiratory pressure, fraction of inspired oxygen of 1.0, nitric oxide inhalation, neuromuscular blocking agents and prone positioning), our patient remained severely hypoxemic, which led us to initiate an extracorporeal membrane oxygenation treatment. Extracorporeal membrane oxygenation and hemodiafiltration were withdrawn on day 12. Our patient was extubated on day 18 and discharged from our Intensive Care Unit on day 20. He went home a month later. CONCLUSION: We describe the first published case of acute respiratory distress syndrome due to C. pneumoniae infection successfully treated by extracorporeal membrane oxygenation, a very useful tool in this syndrome. A quick and specific method for the definite diagnosis of Chlamydophila infection should be developed.

Chlamydophila pneumoniae is an obligate intracellular Gram-negative bacterium. The spectrum of disease, in addition to pneumonia and influenza-like illness, includes pharyngitis, sinusitis, bronchitis, exacerbation of chronic obstructive pulmonary diseases and reactive arthritis [1] [2] [3] [4] [5] . C. pneumoniae accounts for 6% to 20% of cases of community-acquired pneumonia (CAP) [1, 2] . Many of these cases have few symptoms and don't require hospitalization ('walking pneumonia'). However, more severe cases may occur, with up to 18% requiring hospitalization [6] and even mechanical ventilation, especially in elderly, immunocompromised hosts and patients with coexisting cardiopulmonary disease [7] , but also, rarely, in previously healthy adults [8] . Old and new macrolides are effective against C. pneumoniae and have been recommended as first-line treatment. New fluoroquinolones are also effective in vitro against C. pneumoniae and can be used. Studies have shown that 35% to 47% of C. pneumoniae pneumonia is mixed with other pathogens, the most common being Streptococcus pneumoniae [9, 10] . We describe the case of severe CAP due to C. pneumoniae infection in a previously healthy adult patient, with acute respiratory distress syndrome (ARDS) necessitating extracorporeal membrane oxygenation (ECMO). A previously healthy 62-year-old Caucasian man was admitted to our hospital for acute respiratory failure. Our patient developed a fever of up to 40°C seven days earlier and a non-productive cough three days later. He had not received any antimicrobial treatment prior to his hospitalization, the diagnosis of his primary care physician being influenza (A/H1N1v), given the ongoing outbreak. His medical history was remarkable for possible viral pericarditis without any consequence in 2007 and a gastric ulcer 30 years earlier. He had no drinking habits. He did not smoke cigarettes. He had not travelled abroad recently. He did not have any bird or pet. On hospital admission, our patient was in acute respiratory distress. His respiratory rate was 40 breaths/ minute, his temperature 38.3°C, his pulse 98 beats/minute and his blood pressure 114/60 mmHg. Auscultation revealed crackles over his whole left lung and over his right lower lung field. A computed tomography scan showed diffuse alveolar type infiltrates in his two lung fields with air bronchograms (Figure 1 ). Arterial blood gas analysis (under 100% oxygen through a non-rebreathing mask) showed pH 7.54, a partial pressure of carbon dioxide (PaCO 2 ) 44 mmHg, partial pressure of oxygen (PaO 2 ) 38 mmHg and an arterial blood oxygen saturation of 84%. His white blood cell count was 5780 cells/μL (86% neutrophils) and the erythrocyte sedimentation rate was 92 mm/h. Laboratory values showed serum creatinine at 1.7 mg/ dL, potassium at 2.8 mEq/L, creatine phosphokinase at 644 IU/L, liver test alterations (alanine transaminase at 87 IU/L), lactate dehydrogenase elevation (1708 IU/L) and D-Dimers at 7420 ng/mL, activated partial thromboplastin time of 72 seconds, normal international normalized ratio and blood platelets at 166.000/μL. His urine output was 0.4 mL/kg/h over six hours. An electrocardiogram showed a sinus tachycardia with a complete right bundle branch block. Serum samples obtained every week as from the day of admission showed a clear-cut seroconversion for C. pneumoniae antibodies (the course of the antibody titers shown in Table 1 ). Paired serum samples and antigens against the most common microorganisms, including atypical bacteria and common viruses such as A/H1N1v, were negative. Blood, sputum and urine tests for bacterial cultures obtained during the first day of hospitalization were negative. Our patient was treated by amoxycillin-clavulanic acid, moxifloxacin and oseltamivir. His respiratory status necessitated endotracheal intubation and mechanical ventilation. Severe arterial hypotension prompted norepinephrine infusion and the insertion of a pulmonary artery catheter. The initial hemodynamic pattern was typical for sepsis (hemodynamic values are shown in Table 2 ). Metabolic data showed a mixed venous saturation of 56%. Despite maximal ventilatory support (high positive end expiratory pressure, an inspired oxygen fraction (FiO 2 ) of 1.0, nitric oxide inhalation of 20 ppm, neuromuscular blocking agents and prone positioning), our patient remained severely hypoxemic (PaO 2 /FiO 2 = 38) which led us to initiate ECMO treatment. Venovenous ECMO (a Sorin Revolution centrifugal pump, a Sorin ECCO oxygenator and a Sorin Satcrit console from Sorin Group, Milano, Italy) was put in place on the fifth day of hospitalization, with a left femoral 22-Fr drainage cannula and a right femoral 23-Fr return cannula, inducing a drastic improvement of our patient's oxygenation parameters (PaO 2 = 120 mmHg). The mixed venous oxygen saturation (SVO 2 ) increased from 56% to 86%. Continuous veno-venous hemodiafiltration (CVVHDF) renal replacement therapy was also initiated on day three because of acute renal failure. There were no severe complications of the ECMO treatment except for hemorrhagic suffusion of the two femoral catheter insertion points, requiring a blood transfusion. ECMO and CVVHDF were withdrawn on day 12. Our patient was extubated on day 18 and discharged from our Intensive Care Unit on day 20. He went home a month later. He is now in good physical condition and has returned to work and to a normal social life. This case illustrates the polymorphism in the presentation of C. pneumoniae infection, which can cause severe CAP complicated by ARDS, even in immunocompetent patients. Pneumonia and bronchitis are the most common clinical infections associated with C. pneumoniae. The classical pulmonary presentation is a single subsegmental infiltrate, even though lobar consolidation or bilateral infiltrates can also be seen. Two major aspects are discussed below: the rationale for the use of ECMO in ARDS patients as well as its different techniques, and the difficulties of diagnosing C. pneumoniae pneumonia. There are two main ECMO techniques according to the type of vascular access that is used: venovenous and venoarterial. Each one has a specific indication. Venoarterial ECMO gives cardiac and respiratory support. Indications for venoarterial ECMO include postcardiac surgery (heart surgery or heart transplantation), cardiogenic shock due to acute myocardial infarction or acute myocarditis and intoxication. For cardiac support, the goal is to optimize organ perfusion (by obtaining a SVO 2 greater than 70%, which usually needs an output index of about 3 L/ min/m 2 ). This is achieved by choosing the appropriate sizes of cannula according to the patient's body surface area. In isolated respiratory failure, a venovenous access is preferred. The objective is CO 2 removal at least equal to the patient's metabolism (roughly 3 cm 3 /kg/min in adults). The purpose of venovenous ECMO use in ARDS patients is lung protection (reducing ventilatorinduced lung injury) through a decrease in alveolar distention permitted by the reduction in ventilator conditions. Even though a lot of progress has been done in technical issues, the risk-benefit ratio must be taken into account when ECMO is proposed to a patient with severe respiratory failure. As in all extracorporeal devices, anticoagulation is mandatory. Many patients will experience bleeding, which can be very severe. Contraindication for anticoagulation remains the most important limitation of this technique. Previous severe disability or poor prognosis due to underlying disease constitutes the other main contraindication for ECMO initiation. The main complications are bleeding, thromboembolism, cannula-related complications, pulmonary embolism or infarction, aortic thrombosis and coronary or cerebral hypoxemia; the latter three being more frequent in a venoarterial montage type. Discussion still stands in the literature as to whether mechanical ventilation of more than seven days is or is not a relative contraindication for ECMO because of ventilation-induced lung injury [11] . The first randomized clinical trials failed to demonstrate beneficial effect of ECMO for severe respiratory failure in the 1970s and 1980s. Since then, technical improvements for ECMO on one hand, and better treatment of the ARDS (protective ventilation and others) on the other hand, have renewed interest in ECMO [12] . The recent CESAR trial (Conventional Ventilation or ECMO for Severer Adult Respiratory Failure [13] ) demonstrated a possible beneficial effect of ECMO. Of those patients referred for ECMO, there was 63% survival rate at six months without disability, compared to 47% in those who were assigned to conventional management. This translates to one extra survivor without disability for every six patients treated. A recent Italian experience, in patients with ARDS due to the influenza A/H1N1 virus, based on pre-emptive patient centralization showed a 77% survival rate if ECMO was started within seven days of initiation of mechanical ventilation [11] . Indication for ECMO in this study was refractory hypoxia or an oxygenation index below 30 despite a PaO 2 /FiO 2 ratio greater than 100 mmHg. Treatment of critically ill patients affected by the 2009 Influenza A (H1N1v) outbreak in Australia and New Zealand [14] included ECMO, with a 71% survival rate at Intensive Care Unit discharge, an excellent result. These three studies emphasize the renewed place of ECMO in the treatment of severe ARDS with very good survival rates, considering the severity of the initial insult. The limitation of plateau airway pressures and the low tidal volume used in ARDS patients have been at the cost of an increased PaCO 2 . There has therefore been an increase in interest for extracorporeal CO 2 removal techniques. Different techniques have been developed but most of them have not gone to relevant, sufficiently powered clinical trials due to technical problems or insufficient CO 2 removal or oxygenation [15] . The pumpless extracorporeal lung assist (PECLA), or Novalung ® [Novalung GmbH, Heilbronn, Germany], is a compact pumpless device driven by the pressure gradient between arterial and venous blood. A femorofemoral setting is most commonly used. The main advantage of a pumpless device is a reduction in mechanical blood trauma, bleeding and hemolysis. The Novalung ® has been widely studied in ARDS. Three studies have demonstrated the clinical efficacy of this PECLA device [16] [17] [18] . CO 2 removal is efficient but oxygenation may be insufficient in these critically ill patients. Diagnosis of C. pneumoniae pneumonia is still debated in the current literature. According to Grayston et al. and Saikku et al. [19, 20] , who first described this Chlamydia species, almost everybody is infected and reinfected with C. pneumoniae throughout his or her life. Unspecific symptoms of C. pneumoniae infections make the diagnosis even more difficult, with a possible underestimation of its frequency. Undetected infections may lead to chronic disease with serious consequences, such as atherosclerotic cardiovascular disease [21] or asthma. Its incidence in CAP ranges from 3% to 22%, varying according to the diagnostic method used [22] . The most frequent routinely used diagnostic techniques are serological, including complement fixation test, immunofluorescence assay, microimmunofluorescence (MIF) and genus-and species-specific enzyme-linked immunosorbent assay (ELISA) systems. No currently available serologic tests of a single serum specimen will provide conclusive evidence of a current infection with C. pneumoniae. Since the publication of the Centre for Disease Control (CDC) 2001 recommendations for Nucleic Acid Amplification Tests (NAATs) [23] , a multitude of inhouse polymerase chain reaction (PCR) tests have been described, though very few of them have been validated by the CDC. Results of NAATs may be unreliable because of cross-contamination, inappropriate treatment of the clinical samples (leading to the loss of the target nucleic acid) or the presence of inhibiting substances [24] . Additionally, validation of these tests is primarily analytical and not against clinically obtained specimens. Bacterial culture has traditionally been considered the 'gold standard' for diagnosis; however, its sensitivity, even in excellent laboratories, seldom exceeds 90% and is typically between 75% and 85%. The culture is technically difficult to implement and is only available in a few laboratories worldwide [25] . Even though MIF is still actually considered as the 'serological gold standard' technique for the detection of species-specific antibodies, there is a large discrepancy between MIF testing for C. pneumonia and detection of these organisms by culture or PCR. The MIF test is reliable for detection of a prior exposure to Chlamydiae by the presence of immunoglobulin G (IgG) antibodies and is relatively sensitive for the detection of IgM. IgM is an unreliable marker of acute infection in adolescents and adults since it is often not present, presumably because of previous infection by a chlamydial species [24] . Additionally, MIF is quite a laborious and subjective technique that requires much experience, is not standardized and has significant laboratory-to-laboratory variations. In our institution, in order to get better sensitivity (for example, after reinfections), better reproducibility, greater objectivity and less cross-reactivity in the serology tests for Chlamydiaceae, two-level serological testing has been implemented for C. pneumoniae. The first level test is a search by ELISA-technique for anti-major outer membrane protein (MOMP) IgG and IgA antibodies (C. pneumoniae-IgG & IgA-ELISA plus, Medac, Hamburg, Germany) which are species-specific. This permits us to screen patients. The disadvantage of anti-MOMP antibodies is their late appearance in acute primary infection (three weeks for IgA, six to eight weeks for IgG) and their considerable persistence in human serum afterwards, independent of the bacterial eradication. In general, the prevalence of C. pneumoniae IgG antibody reaches 50% in adults above 20 years of age, and 80% in the elderly population (> 70 years old). The second level test is another ELISA assay, rELISA (Medac), for the detection of genus-specific anti-lipopolysaccharide(LPS) antibodies. Chlamydiae contain, as a common immunodominant antigen, the LPS to which the first immune reaction is directed. These anti-LPSs have the main advantage of rising quickly during acute infection (five to ten days for IgA), allowing early diagnosis, and returning to normal a few weeks after the infection and are thus associated to the eradication of the bacterium [26] . The use of paired sera enables the discrimination of current infections, reinfections and reactivations (defined by titer increase) from chronic persistent ones (constant LPS, IgG and IgA antibody titers, present in 5% to 10% of the adult population) for which no antibacterial treatment is needed. The latter is due to the chronic presence of bacteria in the organism in a latent phase (monocytes). Criteria for acute infections are a four-fold increase in IgA or IgG or a doubling of IgA and IgG at 10 to 15 days. Moreover, the described consecutive sera of our patient underwent an additional retrospective examination by a MIF assay (Chlamydia MIF IgG & Chlamydia MIF IgM, Focus Diagnostics, Cypress, CA, USA). A straightforward IgG seroconversion could also be observed (Table 1 ). Thanks to the detection of IgA and of IgG antibodies to C. pneumoniae in various combinations, serology allows a classification of the state of infection. In our patient, reinfection was diagnosed because MIF were positive on admission; there was a four-fold increase in anti-MOMP IgG and, finally, an important rise and a rapid decline in anti-LPS IgA. C. pneumoniae can induce very severe ARDS. We describe the first published case of ARDS due to C. pneumoniae infection successfully treated by ECMO. Definite Chlamydophila diagnosis remains a challenge but should be sought for in severe ARDS patients without evidence of other infectious cause. A quick and specific method for the definite diagnosis of Chlamydophila infections should be developed. Written informed consent was obtained from the patient for publication of this case report and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal.

Environmental Sampling for Avian Influenza Virus A (H5N1) in Live-Bird Markets, Indonesia

To identify environmental sites commonly contaminated by avian influenza virus A (H5N1) in live-bird markets in Indonesia, we investigated 83 markets in 3 provinces in Indonesia. At each market, samples were collected from up to 27 poultry-related sites to assess the extent of contamination. Samples were tested by using real-time reverse transcription–PCR and virus isolation. A questionnaire was used to ascertain types of birds in the market, general infrastructure, and work practices. Thirty-nine (47%) markets showed contamination with avian influenza virus in >1 of the sites sampled. Risk factors were slaughtering birds in the market and being located in West Java province. Protective factors included daily removal of waste and zoning that segregated poultry-related work flow areas. These results can aid in the design of evidence-based programs concerning environmental sanitation, food safety, and surveillance to reduce the risk for avian influenza virus A (H5N1) transmission in live-bird markets.

F ood markets that offer both poultry meat and live birds either for sale or for slaughter are collectively referred to as live-bird markets (LBMs). LBMs are part of the supply chain and are essential for maintaining the health and nutritional status of rural and urban populations, especially in developing countries (1, 2) . However, LBMs provide op-timal conditions for the zoonotic transfer and evolution of infectious disease pathogens because they provide major contact points between humans and live animals (3, 4) . Studies in Hong Kong Special Administrative Region, People's Republic of China; other areas of China; Indonesia; and the United States have shown that LBMs can harbor avian infl uenza viruses (AIVs), including highly pathogenic infl uenza virus A (H5N1), and have been associated with human infection (4) (5) (6) (7) (8) (9) . Continual movement of birds into, through, and out of markets provides opportunity for the introduction, entrenchment, and dissemination of AIVs. Most studies have focused on testing live birds rather than environmental sites in the LBMs (6, 7, 10) . However, a study in New York, NY, that tested environmental sites for AIV (H7N2) found that virus could be isolated from samples from fl oors, walls, and drains from the poultry areas of LBMs (8) . The study also found that despite the ongoing infl ux of infected birds into LBMs, the level of environmental contamination decreased with routine cleaning and disinfection. Another study in Hong Kong LBMs showed that AIV (H9N2) could be isolated at higher rates from poultry drinking water than from samples of bird fecal droppings (11) . Environmental aspects of LBMs are needed for an avian infl uenza control program for 2 reasons. First, a contaminated environment can provide a continuing source of virus transmission, in which healthy birds coming into the market may become infected and persons working in or visiting the market may also be exposed. Second, ongoing surveillance programs in LBMs based on environmental sampling are more likely than those based on invasive bird testing to be acceptable to traders and stall vendors. Environmental sampling is also safer for public health offi cers and veterinary health offi cers than handling and sampling live birds that may be infected with AIV. Markets, Indonesia In this study, we aimed to identify the environmental sites commonly contaminated by AIV (H5N1) in LBMs in Indonesia. Identifying these sites is the fi rst step in the design of evidence-based environmental sanitation, food safety, and surveillance programs to reduce the risk for virus transmission and to develop environmental surveillance programs to monitor LBM contamination status. Three provinces in the western part of Java Island in Indonesia participated in the study: Jakarta, Banten, and West Java (Figure) . Eighteen districts in these provinces were selected on the basis of their proximity to the laboratory, high levels of avian infl uenza activity in farmed birds (Ministry of Agriculture, unpub. data), and high number of LBMs available for study (n = 300). The required sample size was 73 markets based on an estimated disease prevalence of 50% and a maximum error of 10% at 95% confi dence. We based our assumption that 50% of LBMs would be contaminated with AIV (H5N1) on results from a previous study in US LBMs in 2001 (12) . This study found that 60% of markets tested positive for AIV (H7N2) virus in areas in which the virus was endemic. To account for nonresponse, we increased the total sample size to 83 LBMs. We selected markets for inclusion in the study using systematic sampling. On the basis of a sampling frame of 300 markets, every fourth market (the sampling interval) was selected from a list of all the markets. A random numbers table was used to determine the starting point for selection of the 83 markets from the list. Diagnostic specimens and data were collected during October 2007-March 2008. These months have high rainfall and high AIV transmission according to data gathered during 2005-2007 about AIV (H5N1) outbreaks in farmed birds (Ministry of Agriculture, unpub. data). A structured questionnaire containing 42 questions to assess risk factors for AIV (H5N1) contamination was developed. Responses to questions were obtained through visual inspection of each LBM and through an interview with the manager of the participating LBM. The questions sought information about volume of poultry in the LBM and the infrastructure in the delivery, holding, slaughter, sale, and waste-disposal zones of the market. These 5 zones refl ect general demarcation of work fl ow and activities relating to poultry in LBMs (13) . Questions about the sanitation and slaughtering practices were also included. Questionnaire validation was conducted by members of a study advisory team. The team comprised 2 food safety/environmental health offi cers from the Ministry of Health, a communicable disease epidemiologist from the World Health Organization, a veterinary epidemiologist from the Food and Agriculture Organization, and 2 virologists from the Ministry of Agriculture in Indonesia. The questionnaire was tested in 3 LBMs in West Java province to ensure coherence, appropriate use of terminology, and high face validity. The same markets were also inspected to ensure that the questionnaire addressed all aspects of the poultry-related work fl ow in the 5 poultry zones and relevant infrastructure. Members of the study advisory team trained 3 study data collection teams in questionnaire administration and sample collection procedures. To select the environmental sites to be sampled in each LBM, the study advisory team visually inspected 3 markets and reviewed the literature to identify LBM sites commonly contaminated with AIVs or similar pathogens. Sites sampled in previous studies for AIV included fl oors, drains, and water troughs (8, 11, 12) . In this study, 27 sites were selected for environmental sampling ( Table 1) . The sites represented different poultry-related work activities: 3 sites related to delivery of birds into LBMs, 7 in the birdholding zone, 9 in the slaughter zone, 6 in the sale zone, and 2 in the waste-disposal zone. Because of variation in LBM infrastructure and processes, each LBM did not necessarily have all 27 sites. Samples were collected from as many of the 27 sites as were available in each LBM. For each of the 27 sites, 6 swab specimens were collected and pooled. Each pool (vial) consisted of a maximum of 3 swabs. The data collection teams were instructed to increase the representativeness of the samples by swabbing different locations for each environmental site. For example, if the market had 6 poultry stalls, each with its own scale for weighing poultry, then teams collected 1 swab from each scale and pooled them into 2 pools of 3 swabs each. Swab specimens were pooled in the market, and swabs remained inside the vials until testing. The data collection teams were instructed to focus on visibly dirty, moist, or diffi cult-to-clean surfaces in an effort to increase the sensitivity of the sampling. Sample collection, pooling, transportation, and storage were based on techniques used in previous studies (10, 12) . Each data collection team comprised 3 persons, 2 of whom collected samples and 1 administered the questionnaire. To reduce the risk for cross-contamination during sample collection, teams changed disposable gloves and shoe covers between each of the 5 LBM poultry zones. Sterile cottontipped swabs were used to collect all samples, and samples were placed in viral transport media and transported immediately back to the laboratory on frozen gel packs. The viral transport media consisted of Dulbecco modifi ed Eagle medium (Sigma-Aldrich, St. Louis, MO, USA) with 1,000 IU penicillin and gentamicin, and 1% fetal buffer serum (14) . Samples were stored in the laboratory at -70°C until tested. RNA extraction, cDNA synthesis, and real-time reverse transcription-PCR (RT-PCR) were used as described (15) . Virus isolation methods have also been described (16) but in general involved supernatants from a 1,000-μL sample homogenized by vortex and centrifuged at 2,500-3,000 rpm into 9-to 10-day-old specifi c pathogen-free eggs. Those positive in the hemagglutination assay were tested by hemagglutination-inhibition test with reference antiserum (A/chicken/West Java/Hamd/2006). The degree of association between AIV (H5N1) positivity in the 5 LBM poultry zones was determined by using Spearman rank correlation. To assess risk factors for environmental virus (H5N1) contamination, we estimated odds ratios (ORs) using multivariable logistic regression analyses, where variables with p<0.1 from the univariate analyses were included in the initial model. A backward stepwise variable-selection strategy was used to construct a fi nal model with a signifi cance level of p<0.05. The Hosmer and Lemeshow test and the residual χ 2 goodness-of-fi t test were used to assess model stability. Microsoft Excel (Microsoft, Redmond, WA, USA), Epi Info (Centers for Disease Control and Prevention, Atlanta, GA, USA), and All 83 LBMs selected participated in the study; 62 (75%) were located in urban and 21 in rural areas. LBMs were from 16 districts in 3 provinces: 31 (38%) from Jakarta province, 11 (13%) from Banten province, and 41 (49%) from West Java province (Figure) . Most (49 [59%]) LBMs were retail markets, 10 (12%) were wholesale only, and 24 (29%) were a combination of retail and wholesale. Most Forty-eight (58%) LBMs reported monthly or more frequent visits from animal/human health personnel to inspect the poultry zones. Eight (10%) LBMs reported that live birds were tested periodically (less frequently than weekly) for AIV infection. For cleaning and sanitation, 80 (96%) LBMs reported washing poultry zones daily, and 55 (66%) applied detergent or disinfectant daily. Thirty-nine (47%) LBMs had evidence of contamination. For 17 (44%) of these, <5 environmental sites were positive for AIV (H5N1) by real-time RT-PCR. For each of 22 (56%) LBMs, >6 environmental sites were positive. The environmental sites most heavily contaminated were in the slaughter and sale zones (Table 1 ). In the slaughter zone, the most contaminated sites were the poultry-processing tables (21%), baskets holding poultry meat (20%), and chopping boards (20%). In the sale zone, the most contaminated sites were the tables for carcass display (24%) and scales (21%). Another commonly contaminated site was the waste-disposal bin in the waste-disposal zone (19%). In most cases, this bin is not an enclosed bin but rather was a dedicated uncovered fl oor space where remnants are dumped daily and collected weekly by the local government rubbish collection team. Thirteen viruses were isolated from LBMs, most frequently from the slaughter zone (7 of 13 viruses isolated, Table 1 ). All isolated viruses came from 6 LBMs, from which 1-4 viruses were isolated per LBM. From the zones contaminated in each LBM (Table 1) , we calculated correlations between different zones. Contamination in preceding LBM poultry zones correlated with contamination in the subsequent zones ( Table 2 ). Correlations were high between holding and slaughter zones, slaughter and sale zones, and sale and waste-disposal zones. We assessed risk factors for AIV (H5N1) contamination in LBMs. We compared exposures in 39 LBMs with a minimum of 1 contaminated environmental site to 44 LBMs with no contamination. From the univariate analyses, several exposures predicted AIV (H5N1) contamination in LBMs (Table 3) . LBMs with wooden tables, Muscovy ducks, or >200 ducks other than Muscovy were at greater risk for AIV (H5N1) contamination, as were LBMs in West Java province. Six other exposures approached signifi cance, either as protective factors or as risk factors. LBMs that disposed and removed solid waste daily (OR 0.41, 95% confi dence interval [CI] 0.16-1.09); had zoning that clearly segregated poultry delivery, holding, slaughter, sale, and wastedisposal areas (OR 0.28, 95% CI 0.07-1.11); or stacked poultry cages vertically rather than side by side (OR 0.38, 95% CI 0.13-1.10) had less risk for avian infl uenza virus (H5N1) contamination. LBMs with pigeons (OR 3.06, 95% CI 0.96-9.81), mixed bird species in the same cages (OR 2.92, 95% CI 0.98-8.70), or slaughtered birds in the market (OR 3.53, 95% CI 0.89-13.93) were more likely to be contaminated. None of the 9 other variables considered in the study were associated with AIV (H5N1) contamination in LBMs (data not shown). These included the LBM trading category (wholesale, retail, or combination), days operational per week, chicken population in LBM, source of chickens (small-scale backyard farmers, commercial farms, or com- bination), inspection from authorities, use of detergent during cleaning, mixing poultry arriving on different days in the same cages, average length of poultry stay in LBM, and whether poultry were removed from stalls before cleaning. From the univariate analyses, 10 variables were signifi cant at p<0.1. However, the ducks other than Muscovy variable was removed from the multivariate analyses because of its collinearity with another variable (presence of Muscovy ducks, r>0.4). Nine variables were considered for the multivariate analyses. The fi nal multivariable logistic regression model had 4 variables, of which 2 were independent risk factors for subtype H5N1 contamination in LBMs (Table 3) . They were location in West Java province (adjusted OR [aOR] 6.83, 95% CI 2.01-23. 19 ) and bird slaughtering in the LBM (aOR 6.43, 95% CI 1.01-40.82). Two variables were independent protective factors: zoning of poultry activities in LBMs (aOR 0.16, 95% CI 0.03-0.86) and daily disposal of solid waste (aOR 0.2, CI 95% 0.06-0.69). We have demonstrated extensive environmental contamination in LBMs with the AIV (H5N1) in Indonesia. Nearly 50% of LBMs in AIV (H5N1)-endemic districts were positive, with all 5 poultry zones affected. The study identifi ed environmental points of contamination and protective and risk factors for contamination. This study provides baseline information for 2 aspects that can aid in control of AIV (H5N1) in LBMs: 1) development of routine monitoring and surveillance programs and 2) structural interventions and work fl ow modifi cations to minimize risk for contamination. Our fi ndings provide further evidence that environmental contamination with AIVs is not uncommon (8, 14) . Poultry water, drains, tabletops, cages, tablecloths, utensils, bins, and fl oors were all contaminated. Environmental sites most commonly contaminated were located in slaughter zones and zones where carcasses were taken after slaughtering, such as the sale and waste-disposal zones. This contamination can be expected because slaughtering generates droplets that may contain viral particles and exposes internal organs with potentially high viral loads. Even if slaughtering is conducted in a separate zone, contamination can spread to the sale and waste-disposal zone through the carcasses and through the process of evisceration usually conducted in both slaughter and sale stalls. We found rates of contamination in water from poultry feeding bottles similar to those from the study in Hong Kong on AIV (H9N2) (11% and 7% markets with contamination respectively, p = 0.12) (11) . Even though AIVs were detected from poultry drinking water, our study suggests that other environmental sites are more effi cient for monitoring AIV (H5N1) in markets. Processing tables and baskets holding freshly cut poultry meat in the slaughter area, as well as display tables and scales in the sale area, were positive in 20 (24%) LBMs surveyed. The risk and protective factors we identifi ed complement fi ndings from previous studies. Daily disposal and removal of waste from the market is part of routine environmental cleaning and sanitation and eliminates AIV reservoirs (8) . Segregating poultry-related activities into zones limits virus spread (17) . Vertical stacking of cages can limit transmission because trays between layers of birds prevent the scatter of fecal matter. These results add evidence to the World Health Organization current recommendation that waste trays should be used to segregate stacked cages in markets to prevent cross-contamination (13) . LBMs in West Java province had a higher risk for contamination than did other provinces. This risk probably is due to greater AIV (H5N1) disease activity in the province. Surveillance activities during 2006-2008 showed that West Java had a 4.7% outbreak detection rate compared with rates in Banten (4%) and Jakarta (0.2%) (18) . Furthermore, in West Java province chicken density is high: 14,000 birds/km 2 compared with densities in the neighboring provinces Banten and Jakarta (3,900 birds/km 2 and 400 birds/km 2 , respectively) (19) . Poultry density data are commonly used as a proxy for disease activity where areas of high poultry density have the highest risk for an outbreak (20, 21) . Several issues need to be considered regarding our fi nding of low virus isolation rates compared with realtime RT-PCR-positive rates. Virus isolation detects viable virus, whereas real-time RT-PCR detects small stretches of nucleic acid, even if the larger genomic RNA is inactivated. This makes real-time RT-PCR a more sensitive detection tool but does not provide information about virus viability. Samples obtained from the environment may be less suitable than animal samples for virus isolation techniques. Organic matter, duration and temperature of exposure, and humidity can all affect virus survival outside the animal host (22) . Three studies conducted in LBMs tested environmental samples and bird samples by using virus isolation (8, 10, 23) . Only 1 of these studies stratifi ed the avian infl uenza detection rates by type of sample (bird vs. environment) (8) ; that study found that from 12 LBMs, 11 were positive for avian infl uenza in bird samples compared with only 5 positive in environmental samples. These results were based on a small sample of LBMs, and real-time RT-PCR was not conducted. Therefore, to determine the suitability of virus isolation for environmental samples, we recommend that future studies compare real-time RT-PCR-positive rates to virus isolation rates in both environmental swab and bird samples. Risk and protective factors identifi ed in this study, together with fi ndings from other studies, can assist in developing environmental or behavioral interventions to reduce AIV transmission in LBMs. Previous studies have shown that regular cleaning with detergents, including free chlorine concentrations typically used in drinking water treatment, can rapidly decontaminate surfaces from AIVs (8, 24) . Previous studies also have shown that periodic market rest days coupled with thorough cleaning can minimize the reservoir of AIV in LBMs (4, 12, 25) . These messages have been disseminated to LBMs throughout Indonesia and formed the basis of the Ministry of Health Decree in 2008 on building healthy food markets (26) . For a more systematic food safety monitoring system, this study will be used to develop a risk-based approach for AIV risk reduction in LBMs in Indonesia (27) . The contamination sites and risk factors will be used to determine critical control points and critical limits for intervention. LBM operators, stall vendors, and other stakeholders (e.g., sanitarians and public health offi cers) will need to be provided with simple monitoring plans to reduce the risk for contamination. Such monitoring plans are expected to have an impact not only on AIV (H5N1) but also on other viruses and bacteria commonly associated with food safety for poultry products. In addition to tools for disease control, the study fi ndings can aid AIV (H5N1) surveillance activities in LBMs. Commonly contaminated environmental sites in LBMs can form the basis of an environmental sampling strategy for detection of AIV (H5N1) in LBMs. Environmental sampling is more benefi cial than live-bird sampling because it is less time and labor intensive and eliminates the need to handle and restrain live birds. Environmental sampling reduces the potential for virus aerosolization and the risk for infection for persons collecting the samples or standing nearby. Further work is needed to assess the adequacy of environmental sampling for surveillance in LBMs under different conditions, especially because detection sensitivity will vary by AIV (H5N1) prevalence in farms supplying the birds. A limitation of this study is that the observation of environmental contamination was based on a cross-sectional survey in which LBMs were sampled only once. We recommend that future studies observe persistence of the virus over time in the various environmental sites. Reports from market managers and vendors about inspection and cleaning practices in the LBMs were not verifi ed during the course of the study. These activities may have been overreported because respondents may have wanted to report what they perceived interviewers wanted to hear. Be-cause of the high cost associated with the fi eld and laboratory work for such studies, studies should focus on a small number of markets and collect in-depth information about contamination trends and associated risk factors, as well as data on other indicator organisms, such as Escherichia coli or Enterobacteriaceae, that provide information about general market hygiene. Future work also should evaluate the effects of interventions in markets especially in lowresource settings because this would be of most benefi t to low-income and middle-income countries.

Influenza in Refugees on the Thailand–Myanmar Border, May–October 2009

We describe the epidemiology of influenza virus infections in refugees in a camp in rural Southeast Asia during May–October 2009, the first 6 months after identification of pandemic (H1N1) 2009 in Thailand. Influenza A viruses were detected in 20% of patients who had influenza-like illness and in 23% of those who had clinical pneumonia. Seasonal influenza A (H1N1) was the predominant virus circulating during weeks 26–33 (June 25–August 29) and was subsequently replaced by the pandemic strain. A review of passive surveillance for acute respiratory infection did not show an increase in acute respiratory tract infection incidence associated with the arrival of pandemic (H1N1) 2009 in the camp.

We describe the epidemiology of infl uenza virus infections in refugees in a camp in rural Southeast Asia during May-October 2009, the fi rst 6 months after identifi cation of pandemic (H1N1) 2009 in Thailand. Infl uenza A viruses were detected in 20% of patients who had infl uenza-like illness and in 23% of those who had clinical pneumonia. Seasonal infl uenza A (H1N1) was the predominant virus circulating during weeks 26-33 (June 25-August 29) and was subsequently replaced by the pandemic strain. A review of passive surveillance for acute respiratory infection did not show an increase in acute respiratory tract infection incidence associated with the arrival of pandemic (H1N1) 2009 in the camp. P andemic (H1N1) 2009 emerged in April 2009 and subsequently spread around the globe. The World Health Organization issued a pandemic declaration on June 11, 2009 (1,2). By October 25, 2009 , >440,000 laboratory-confi rmed cases, including >5,700 deaths, had been reported to WHO (3) . The fi rst case of pandemic (H1N1) 2009 infection was diagnosed in Thailand on April 28, 2009 , and subsequently the virus was detected in all provinces. The Thailand Ministry of Public Health reported 27,639 confi rmed cases and 170 deaths as of October 10, 2009 (4) . Myanmar (Burma) reported its fi rst confi rmed case of pandemic (H1N1) 2009 infection during the week beginning July 5, 2009 , and by the end of October 2009 had reported <100 confi rmed cases with no deaths (5) . Although most infections caused by this new virus have been mild, severe disease has been reported, particularly in young adults (6) . Data regarding the effect of infl uenza in rural areas of the developing world are scarce, as are etiologic data from refugee populations (7) (8) (9) . A recent review of published reports from Southeast Asia concluded that infl uenza infection may be identifi ed in up to 26% of outpatients with febrile illness and in 14% of hospitalized patients with pneumonia (10) . In Thailand, seasonal infl uenza virus activity peaks during the rainy season (June-September), with smaller peaks occurring during the cold months (January and February) (11) . Incidence of infl uenza infections in Thailand was 64-91 cases/100,000 persons per year during 1999-2002; the infl uenza-related hospitalization rate was 21/100,000 persons during 1999 (11) . Infl uenza infections in Myanmar are also seasonal; cases are documented predominantly in the rainy season (May-October) (12) (13) (14) . Incidence data for infl uenza virus infections in Myanmar are not readily available. Of 15.2 million refugees worldwide, approximately one third live in camps (15). These refugees often live in crowded conditions and have contact with populations from the host country and the country of origin, where public health infrastructure and surveillance may be poor (16, 17) . Approximately 150,000 refugees from Myanmar are housed in several camps on the Thailand-Myanmar border. Maela Temporary Shelter (Maela, Thailand) is the largest of these camps, with a population of >40,000, predominantly of the Karen ethnic group, housed in a 4-km 2 area (18) . This camp is located in the hills adjoining the Myanmar border, ≈500 km northwest of Bangkok, and has been in operation since 1984. Primary health and sanitation services are provided by nongovernmental or- From May 1 through October 31, 2009, trained local fi eld workers visited the hospital in Maela daily (Monday-Saturday). Patients whose illnesses met clinical case defi nitions for infl uenza-like illness (ILI) or pneumonia (Table 1) were identifi ed by clinic staff at the time of examination, and these patients were asked to complete an additional clinical interview. Inpatient and outpatient department cases were included in the surveillance. From July 27 through October 31, 2009, original clinical case defi nitions were modifi ed to capture each patient who had a history of fever during the current illness but who was not febrile at the clinic visit (either because of the intermittent nature of fever or self-administration of antipyretics). A nasopharyngeal aspirate (NPA) was collected from each patient; a sterile 8-French infant feeding tube was inserted into the nasopharynx and then withdrawn while suction was applied with a 20-mL syringe attached to the feeding tube. The nasopharyngeal secretions and the tip of the feeding tube were transferred to a 1-mL tube of viral transport medium and stored in a cool box until transfer, within 24 h, to a -80°C freezer before analysis. All NPA specimens were subjected to a panel of realtime reverse transcription-PCR (rRT-PCR) assays for the following viruses: infl uenza A (separate primer/probe sets for infl uenza A [universal], pandemic [H1N1] 2009, seasonal subtype H1N1, and seasonal subtype H3N1 detection) (20) ; infl uenza B (CDC in-house assay [details available on request]); respiratory syncytial virus (RSV; CDC in-house assay [details available on request]); and human metapneumovirus (HMPV) (21) . An internal control PCR specifi c for the human RNAseP gene was used to monitor sample adequacy and to detect the presence of PCR inhibitors (22) . Positive and negative controls were included in each PCR run. A Rotorgene 6000 real-time PCR thermocycler (Corbett Life Science, Mortlake, New South Wales, Australia) and SuperScript III One-Step RT-PCR Kits (Invitrogen, Carlsbad, CA, USA) were used throughout. All laboratory work was conducted at the Shoklo Malaria Research Unit microbiology laboratory in Mae Sot, Tak Province, Thailand. To compare virologic results from 2009 with our surveillance data from 2008, we subsequently restricted the 2009 dataset to match data collected in 2008 (i.e., we included only patients whose illnesses met the strict case defi nitions and who were sampled on either Monday or Tuesday in the outpatient department). Clinical and laboratory data collected in 2008 were identical to data collected in 2009. To estimate the incidence of infl uenza-associated illness, we reviewed passive disease surveillance data collected by the hospital in Maela and collated by the Com- mittee for Coordination of Services for Displaced Persons in Thailand. This surveillance system captured data only on patients visiting the hospital for treatment. The number and incidence rate (calculated by using monthly camp population census data) of clinically diagnosed upper respiratory tract infections (URTIs) and lower respiratory tract infections (LRTIs) were reported by month. No information was available to determine the number of ILI cases; therefore, we could not estimate the proportion of URTIs caused by infl uenza viruses in Maela. However, because most LRTIs reported are likely to be clinical pneumonia, we estimated the incidence of infl uenza-associated pneumonia as the incidence of LRTI multiplied by the percentage of pneumonia patients with specimens positive for infl uenza A. To determine the effect of pandemic (H1N1) 2009 on overall case numbers, we compared 2008 data with 2009 data. The Human Studies Oversight and Review Team of CDC reviewed the surveillance project and declared it to be a nonresearch activity, as defi ned by US 45 CFR 46.102(d). Therefore our study was exempt from the need for full review by an institutional review board. All statistical analyses were performed by using STATA version 10.1 software (StataCorp, College Station, TX, USA). Categorical variables were analyzed by using the Fisher exact test; continuous variables were analyzed by using the Wilcoxon rank-sum test (because none were normally distributed). Two-tailed p values <0.05 were considered signifi cant. Epidemiologic week numbers were calculated by using standard criteria (23) . During May 1-October 31, 2009, a total of 324 patients were included in the surveillance. Of these, 19 were excluded from further analysis; 18 patients did not meet the clinical case defi nitions, and no NPA specimen was received for 1 patient (Figure 1) . Pneumonia was diagnosed for 234 (77%) of the 305 eligible patients, and ILI was diagnosed for 71 (24%). For patients with pneumonia, median age was 2.0 years (range 0.1-68 years) and 55% were male; for those with ILI, median age was 1.4 years (range 0.2-10 years) and 54% were male. Fifty seasonal infl uenza A infections and 17 pandemic (H1N1) 2009 infections were detected by rRT-PCR. Fortynine of the 50 seasonal infl uenza A infections were subtyped as H1N1; one was subtype H3N1 (Figure 2 ; with dual virus infection were not signifi cantly more ill than those with infl uenza A infection alone (3/7 vs. 26/60; p = 1.0). Illnesses for 205 (67%) patients met the strict case definition for ILI or pneumonia; 100 (33%) met only the expanded case defi nitions. Age distribution and proportion of infl uenza A viruses did not differ signifi cantly between the strict and expanded case defi nition groups. However, a signifi cantly higher proportion of patients with ILI (18/25 vs. 6/46; p<0.001) or pneumonia (99/180 vs. 5/54; p<0.001) whose illnesses met the strict case defi nition were hospitalized, which suggests that the expanded case defi nitions captured patients with milder illnesses. Overall, at least 1 virus was detected in 175 (57%) patients (37/71 ILI, 138/234 pneumonia). HMPV and RSV accounted for 120/187 (54%) viruses detected. These viruses were detected in ILI cases (HMPV 23%,; RSV 17%) and pneumonia (HMPV, 21%; RSV 18%). RSV was detected signifi cantly more often in children <5 years of age (48/221 vs. 6/84; p = 0.003) and was more age restricted than all other viruses. In Figure 4 ) (R. Sedhain, pers. comm). Our study demonstrates that infl uenza virus infections are common etiologic agents of respiratory infection in a Southeast Asian refugee population living in crowded conditions. During the 6 months of surveillance in 2009, infl uenza A viruses were detected by rRT-PCR in 23% of clinical pneumonia and 20% of ILI cases sampled, representing a considerable impact that this vaccine-preventable disease has among patients with ARI. Maela is an overcrowded and relatively closed refugee camp and therefore might be considered an ideal location for a novel infl uenza virus to cause an explosive outbreak. However, the number of confi rmed cases indicated that no major outbreak occurred in 2009. After the fi rst case of pandemic (H1N1) 2009 was identifi ed in August, these cases increased modestly in September, then substantially declined during October. Overall, only 25% of all infl uenza A viruses were determined to be the pandemic strain. However, supportive data show a change of the predominant infl uenza virus. In late August 2009, seasonal infl uenza A (H1N1) was the predominant circulating virus; during the subsequent 2 months, only cases of pandemic (H1N1) Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 16 2009 were detected. During May-August, the incidence of LRTI and URTI in cases captured by the passive surveillance system was higher each month in 2009 than in 2008. The rates of URTI were similar in September and October of both years, whereas the LRTI rate was higher in October 2008 than in October 2009. Pandemic (H1N1) 2009 did not clearly increase in case-patients with ARI after its fi rst detection in the camp in August 2009. However, surveillance did not capture mild infections that did not result in visits to the outpatient department. The occurrence of most infl uenza A infections in patients who had pneumonia most likely refl ects a sampling bias, although infl uenza is a generally underrecognized cause of pneumonia in the tropics (24) . ILI is not a routinely used diagnosis for the clinic staff at Maela, so most of the ILI case-patients likely were not interviewed and sampled. However, when infl uenza A was identifi ed, pandemic (H1N1) 2009 case-patients were less likely than seasonal infl uenza case-patients to have been hospitalized. This information suggests that, in this population, illness caused by pandemic (H1N1) 2009 was no more severe than illness associated with seasonal infl uenza A. Several confounding factors, unrelated to the innate pathogenicity of the viruses, may account for this fi nding: 1) the timing of the modifi cation of case defi nitions in relation to the appearance of pandemic (H1N1) 2009; 2) differences in age distribution; and 3) presence of underlying illnesses in the patient groups. Data regarding underlying medical conditions were not collected as part of this surveillance so the effect of other conditions cannot be assessed. To prevent spread of infection, public health systems may request persons with ILI to self-quarantine, which might result in underestimation of the number of cases identifi ed in clinic-or hospital-based surveillance systems. During the 2009 infl uenza season, announcements regarding infl uenza and the need for good hygiene were made on the Maela public address system; healthcare workers reinforced these messages by home visits. Whether this intervention had any effect on healthseeking behavior remains unclear. An infl uenza triage system was in operation at the hospital, but our surveillance staff had access to patients seen and treated in this area. Our study has several limitations. Most importantly, not every patient eligible for sampling was included, frequently because the patient refused or clinic staff failed to identify patients with illnesses that met the case criteria. These data were not recorded, so the effect of this bias cannot be estimated. As previously discussed, ILI is not a frequently used diagnosis outside this surveillance program, and most cases with this clinical syndrome were diagnosed as common cold. Many of the ILI cases documented were miscategorized in the clinic as pneumonia but were subsequently found not to meet the case defi nition, explaining the presence of persons hospitalized with ILI. Overall, these factors may bias toward sampling of case-patients who had more severe symptoms. Also, screening took place in only 1 of the 2 hospital outpatient clinics. However, because both are general clinics, the impact of this screening is likely to be refl ected in the absolute number of cases detected rather than in the proportion of ILI and pneumonia cases caused by infl uenza viruses. Regarding laboratory data, the likelihood of confi rmation of infl uenza infection is associated with the clinical case defi nitions in use: the strict ILI case defi nition used in our surveillance has a sensitivity of 98.4%-100% but a specifi city of only 7.1%-12.9% (25) . In another study, the probability of having a positive infl uenza virus PCR was directly related to magnitude of fever (26) . Therefore, given the bias toward severe cases, we may have considerably underestimated the impact of infl uenza in Maela. As a result of the limitations noted above, we could (27) . Given the likely health inequalities between our refugee population and rural provinces in Thailand, direct comparison of these datasets is diffi cult. However, the incidence of infl uenza-associated pneumonia in Maela was ≈5× higher than in the Thai provinces (27) . Population structure, such as the number of young children and elderly persons, may account for some of this difference, because the incidence of infl uenza infection is highest in these age groups. As with ILI, the case defi nitions used may have affected the data or the use of different laboratory confi rmation tests for infl uenza infection may have resulted in considerable variation in disease rates between studies; the study in Thailand used RT-PCR for laboratory confi rmation. Although the rates of infl uenzaassociated pneumonia were different in the refugee camp, the proportions of pneumonia cases associated with infl uenza were similar (23% vs. 18%). Methods of preventing or mitigating infl uenza outbreaks in a community include vaccination; use of antiviral drugs; and basic infection control measures, particularly good respiratory etiquette, hand washing, and social distancing (28) . The World Health Organization has devised a specifi c infl uenza pandemic preparedness and mitigation plan for refugee and displaced populations, but implementation requires the coordinated efforts of healthcare providers (frequently nongovernmental organizations) and governments to ensure that control measures are available and used effectively (29) . Because resources are likely to be strained during an infl uenza pandemic, refugee and displaced populations might not be adequately represented in a country's pandemic preparedness plan. Availability of items required to control infl uenza transmission (personal protective equipment, vaccines, and antiviral medication) may be limited for this population without robust planning at the local and national levels. In addition to pandemic preparedness, camp administrators and donor agencies should consider routine vaccination for seasonal infl uenza in these populations. Continuation and refi nement of this surveillance as the pandemic continues may provide further insight into the epidemiology of infl uenza in resource-poor rural Asian populations. Work such as this solidifi es the need of inclusion of refugee populations in infl uenza vaccine strategies and pandemic planning.

Early Clinical Features of Dengue Virus Infection in Nicaraguan Children: A Longitudinal Analysis

BACKGROUND: Tens of millions of dengue cases and approximately 500,000 life-threatening complications occur annually. New tools are needed to distinguish dengue from other febrile illnesses. In addition, the natural history of pediatric dengue early in illness in a community-based setting has not been well-defined. METHODS: Data from the multi-year, ongoing Pediatric Dengue Cohort Study of approximately 3,800 children aged 2–14 years in Managua, Nicaragua, were used to examine the frequency of clinical signs and symptoms by day of illness and to generate models for the association of signs and symptoms during the early phase of illness and over the entire course of illness with testing dengue-positive. Odds ratios (ORs) and 95% confidence intervals were calculated using generalized estimating equations (GEE) for repeated measures, adjusting for age and gender. RESULTS: One-fourth of children who tested dengue-positive did not meet the WHO case definition for suspected dengue. The frequency of signs and symptoms varied by day of illness, dengue status, and disease severity. Multivariable GEE models showed increased odds of testing dengue-positive associated with fever, headache, retro-orbital pain, myalgia, arthralgia, rash, petechiae, positive tourniquet test, vomiting, leukopenia, platelets ≤150,000 cells/mL, poor capillary refill, cold extremities and hypotension. Estimated ORs tended to be higher for signs and symptoms over the course of illness compared to the early phase of illness. CONCLUSIONS: Day-by-day analysis of clinical signs and symptoms together with longitudinal statistical analysis showed significant associations with testing dengue-positive and important differences during the early phase of illness compared to the entire course of illness. These findings stress the importance of considering day of illness when developing prediction algorithms for real-time clinical management.

Dengue virus (DENV) causes the most prevalent mosquitoborne viral disease affecting humans, with 2.5-3 billion people at risk for infection and approximately 50 million cases of dengue each year [1, 2] . The four DENV serotypes are transmitted to humans by Aedes aegypti and Ae. albopictus mosquitoes, primarily in urban and peri-urban areas in tropical and subtropical countries worldwide. Most cases present as classic dengue fever (DF), a debilitating but self-limited illness that manifests with high fever, retro-orbital pain, severe myalgia/arthralgia, and rash. However, in some cases, mainly children, illness progresses to life-threatening dengue hemorrhagic fever/dengue shock syndrome (DHF/DSS), characterized by vascular leakage leading to hypovolemic shock and a case fatality rate up to 5% [1, 3, 4] . Currently, no licensed vaccine or antiviral therapy exists for dengue. Early identification of patients at risk of developing severe dengue is critical to provide timely supportive care, which can reduce the risk of mortality to ,1% [1, 2] . However, distinguishing dengue from other febrile illnesses (OFIs) early in illness is challenging, since symptoms are non-specific and common to other febrile illnesses such as malaria, leptospirosis, rickettsiosis, and typhoid fever [5] [6] [7] in dengueendemic countries. In addition, many distinguishing clinical features of DHF/DSS generally emerge only after 4-5 days, at defervescence, when the patient is already critically ill. Although the World Health Organization (WHO) has recently established new clinical guidelines to classify dengue severity [1] , serological, virological, and molecular biological tests are required to definitively diagnose DENV infection. In many endemic countries, laboratory diagnosis of dengue is often problematic due to lack of reagents, expense, or delay in obtaining results. Patients with suspected dengue are often hospitalized for close monitoring to ensure proper treatment if they begin to develop severe dengue; however, up to 38-52% are later diagnosed with OFIs [8, 9] and thus were hospitalized unnecessarily at great financial cost to their family and society [10] . New tools are therefore needed to distinguish dengue from OFIs to prevent deaths from severe dengue and to mitigate the economic burden of excess hospitalization. Recent approaches using multivariable logistic or linear regression models have shown that petechiae, thrombocytopenia (platelet count #100,000 cells/mm 3 ), positive tourniquet test, rash, and other signs and symptoms can distinguish dengue from OFIs [11] [12] [13] [14] [15] [16] [17] ; however, results were not consistent across studies. Only two studies considered clinical and laboratory features according to day of illness [18] [19] [20] , but as these were hospital-based studies, the results likely reflect patients with more severe symptoms and not the clinical spectrum of all symptomatic cases in dengue-endemic populations. Furthermore, none of these studies analyzed data using longitudinal statistical methods, which account for correlations between repeated measures on individuals over time. The use of longitudinal statistical methods to analyze cohort data is essential to utilize all of the data available for analysis and appropriately estimate the within-person and betweenperson variance in measures over time. In this study, we used five years of data from an ongoing prospective cohort study of approximately 3,800 children aged 2-14 years in Managua, Nicaragua, to examine the frequency of clinical signs and symptoms by day of illness and to generate models for the association of signs and symptoms during the early phase of illness and over the entire course of illness with testing dengue-positive. In order to account for the longitudinal structure of the data, odds ratios (ORs) and 95% confidence intervals were calculated using generalized estimating equations (GEE), adjusting for age and gender. In August and September 2004, a community-based pediatric cohort was established in District II of Managua, a low-to-middle income area with a population of approximately 62,500 [21] . Study activity was based in the Health Center Sócrates Flores Vivas (HCSFV), a public facility that is the primary source of health care for District II residents. Briefly, participants aged 2-9 years were recruited through house-to-house visits, and additional two year-olds were enrolled each year to maintain the age structure of the cohort [21] . Children were eligible to remain in the study until age 12 or until they moved from the study area. The parent/legal guardian of each participant signed an informed consent form, and children $6 years old provided verbal assent. In 2007, participants #11 years old were given the opportunity to continue for an additional 3 years, and a second informed consent was performed. The study was approved by the Institutional Review Boards of the University of California, Berkeley, the Nicaraguan Ministry of Health, and the International Vaccine Institute in Seoul, Korea. Parents or legal guardians of all subjects in both studies provided written informed consent, and subjects 6 years of age and older provided assent. Upon enrollment, parents/legal guardians of all participants were encouraged to bring their child(ren) to the HCSFV at first sign of illness or fever. Study physicians and nurses, trained in identification of possible dengue cases, provided medical care for study participants. Febrile illnesses that met the WHO criteria for suspected dengue (Table 1 ) and those without other apparent origin (undifferentiated febrile illnesses) were treated as possible dengue cases and followed daily while fever or symptoms persisted through visits with study medical personnel ( Figure 1 ). Complete blood counts (CBCs) were completed every 48 hours or more frequently as necessary, as indicated by the physician. Cases were monitored closely for severe manifestations and were transferred by study personnel to the Infectious Disease Ward of the Manuel de Jesús Rivera Children's Hospital, the national pediatric reference hospital, when they presented with any sign of alarm (Table 1 ). In addition, an annual healthy blood sample was collected to identify all DENV infections during the previous year and for baseline CBC values. Study physicians in both the hospital and HCSFV completed systematic data collection forms that contained approximately 80 variables (Table 1 ). In the hospital, additional clinical data, including fluid balance and treatment, were collected daily during hospitalization or through ambulatory follow-up visits by a team of study physicians and nurses. Data were also recorded on medical tests ordered and treatments prescribed. A case was considered laboratory-confirmed dengue when acute DENV infection was demonstrated by: detection of DENV RNA by RT-PCR; isolation of DENV; seroconversion of DENVspecific IgM antibodies observed by MAC-ELISA in paired acuteand convalescent-phase samples; and/or a $4-fold increase in anti-DENV antibody titer measured using Inhibition ELISA [22] [23] [24] [25] in paired acute and convalescent samples. DENV serotypes were identified by RT-PCR and/or virus isolation. Laboratory-confirmed dengue cases were further classified by severity. DHF and DSS were defined according to the traditional WHO criteria (Table 1 ) [26] . Additional categories of severity were included for those cases presenting with shock without thrombocytopenia and/or hemoconcentration (dengue with signs associated with shock (DSAS)) [23] or dengue fever with compensated shock (DFCS) [27] (Table 1) . Laboratory-confirmed Dengue virus causes an estimated 50 million dengue cases and approximately 500,000 life-threatening complications annually. New tools are needed to distinguish dengue from other febrile illnesses. In addition, the natural history of pediatric dengue early in illness in a community-based setting has not been well-defined. Here, we describe the clinical spectrum of pediatric dengue over the course of illness in a community setting by using five years of data from an ongoing prospective cohort study of children in Managua, Nicaragua. Day-by-day analysis of clinical signs and symptoms together with longitudinal statistical analysis showed significant associations with testing dengue-positive and important differences during the early phase of illness compared to the entire course of illness. These findings are important for clinical practice since outside of the hospital setting, clinicians may see dengue patients toward the beginning of their illness and utilize that information to decide whether their patient has dengue or another febrile illness. The results of these models should be extended for the development of prediction algorithms to aid clinicians in diagnosing suspected dengue. Early Clinical Features of Pediatric Dengue www.plosntds.org cases were defined as primary DENV infections if acute-phase antibody titer, as measured by Inhibition ELISA, was ,1:10 or if convalescent phase antibody titer was ,1:2560, and as secondary infections if the acute titer was $1:10 or convalescent titer was $1:2560 [22] [23] [24] [25] . Data from the first five years of the study (August 30, 2004-June 30, 2009) were used for analysis. The first three days after onset of fever were considered the early febrile phase of illness. Day of illness at presentation was determined by the date of fever onset, which was defined as the first day of illness as reported by the parent/guardian. Variable definitions are described in Table 1 . Positive tourniquet test was examined using cut-offs of $10 petechiae/in 2 and $20 petechiae/in 2 . Platelet count was dichotomized using a cut-off of #150,000 cells/mm 3 to enable comparisons during days 1-3. Only data from days 1-8 of illness were included for analysis. The frequency of dengue testing results (laboratory-confirmed dengue-positive versus dengue-negative) and disease severity (DF versus severe dengue) was examined by year, demographics, serotype and immune response. The frequency of the WHO case definition for suspected dengue was examined by dengue testing results and age, and a chi-square test for trend was performed. The frequency of clinical signs and symptoms by day of illness and dengue severity was also examined using chisquare tests. To examine the association between clinical signs and symptoms and the odds of testing dengue-positive versus dengue-negative, odds ratios (ORs) and 95% confidence intervals (CIs) were calculated using GEE models assuming an exchangeable correlation structure with robust standard errors to account for the correlations between repeated measures on the same patients over time. First, ORs were calculated using bivariable models that included only dengue testing results and each of the signs or symptoms. All signs and symptoms were then examined in multivariable models that adjusted for age and gender. Data from the first three days of illness and from all days of illness only were analyzed separately. Finally, for comparison, we used traditional logistic regression models to analyze the association between signs and symptoms and testing dengue-positive with data collapsed by illness episode to disregard repeated measures on the same From August 2004 to June 2009, 22,778 episodes of febrile illness were evaluated, of which 1,974 episodes were suspected dengue or undifferentiated fever (Figure 1 ). Of the 1,974 possible dengue cases, 1,793 (91%) tested negative and 181 (9%) were laboratory-confirmed as dengue-positive, of which 161 (89%) were classified as DF, 9 (5%) as DHF, 4 (2%) as DSS, 3 (2%) as DSAS and 4 (2%) as DFCS (Table 1) . Nearly all (95%) of the severe dengue cases but only 116 (72%) of the DF cases met the WHO case definition for dengue. The proportion of laboratoryconfirmed DENV infections that met the WHO case definition significantly increased by age (chi-square test for trend 5.977, p = 0.01), while younger children experienced significantly more undifferentiated febrile illness due to DENV infection ( Figure 2 ). The median age for cases meeting the dengue case definition was 8 years (range 2-13) and that of undifferentiated febrile illness due to DENV infection was 6 years (range 2-10). The number of confirmed dengue-positive cases varied by year, as expected (Table 2 ) [28] . Both genders were equally represented, with a slightly higher percentage of females experiencing severe dengue, though this difference was not statistically significant. The majority of DF cases were DENV-2 (58%), followed by DENV-1 (21%) and DENV-3 (9%), while 60% of severe dengue cases were DENV-2, followed by DENV-3 (25%) and DENV-1 (10%). In addition, there were nearly equal proportions of primary and secondary immune responses among DF cases, whereas the majority (70%) of severe dengue cases were secondary DENV Figure 2 . Frequency of dengue-positive episodes that met the WHO classification criteria for suspected dengue by age (n = 181). Upon presentation to the health center or hospital, children with a febrile illness were classified according to whether or not they met the WHO classification criteria for suspected dengue. One patient had two dengue virus infections over the course of the study and is represented twice. n = 6 for age 2, n = 10 for age 3, n = 18 for age 4, n = 23 for age 5, n = 21 for age 6, n = 16 for age 7, n = 21 for age 8, n = 23 for age 9, n = 24 for age 10, n = 19 for age 11+. Chi-square test for trend 5.977, p = 0.01. WHO, World Health Organization. doi:10.1371/journal.pntd.0001562.g002 Table 2 . Characteristics of study participants by dengue testing results and disease severity (n = 1,974). Table 2 ). The median day of illness at presentation was day 2 for all patients, and almost all presented on days 1-3 of illness (90%). The total follow-up time of all children in the cohort was 17,931 person-years with a median follow-up of 3.9 years per child. As shown in Figure 3 , several signs and symptoms appeared to differentiate OFIs from DF cases, and DF cases from severe dengue cases, according to day of illness. In particular, higher proportions of DF and severe dengue cases experienced petechiae, platelets #150,000 cells/mm 3 , leukopenia, and positive tourniquet test compared to patients with OFIs. Higher proportions of severe cases experienced petechiae, platelets #150,000 cells/mm 3 , myalgia/arthralgia and abdominal pain compared to DF cases and patients with OFIs. Abdominal pain differentiated severe dengue cases from DF and OFI only beginning on day 3 of illness (for severe dengue compared to DF: chi-square 0.144, p = 0.70 for days 1-2 versus chi-square 16.910, p,0.0001 for day $3). Bivariable and multivariable analyses were performed using GEE models to examine signs and symptoms early in illness and over the course of illness (Table 3) . On days 1-3 of illness, dengue-positive cases had up to 2-fold increased odds of fever, headache, retro-orbital pain, myalgia, arthralgia, and vomiting compared to patients with OFIs. They also had from 3-fold to 9fold increased odds of rash, petechiae, positive tourniquet test with cut-offs of $10 and $20 petechiae/in 2 , leukopenia, platelets #150,000 cells/mm 3 , poor capillary refill, cold extremities and hypotension compared to patients with OFIs. In contrast, they had decreased odds of abdominal pain, likely Figure 3 . Frequency of signs and symptoms by day in patients with OFI, DF and severe dengue. Over the course of an episode of febrile illness, signs and symptoms were observed by medical personnel or reported by children and/or their parent/guardian. Selected signs and symptoms are shown here. A, Petechiae; OFI versus DF: chi-square test for trend 21.313, p,0.0001; day 1, n = 606; day 2, n = 1,243; day 3, n = 1,066; day 4, n = 876; day 5, n = 675; day 6, n = 481; day 7, n = 291; day 8, n = 175; B, Platelet count #150,000 cells/mm 3 ; OFI versus DF: chi-square test for trend 14.928, p = 0.0001; day 1, n = 604; day 2, n = 970; day 3, n = 615; day 4, n = 568; day 5, n = 348; day 6, n = 234; day 7, n = 122; day 8, n = 65; C, Myalgia/ arthralgia; OFI versus DF: chi-square test for trend 4.569, p = 0.03; day 1, n = 612; day 2, n = 1,253; day 3, n = 1,075; day 4, n = 877; day 5, n = 671; day 6, n = 477; day 7, n = 289; day 8, n = 181; D, Leukopenia; OFI versus DF: chi-square test for trend 6.449, p = 0.01; day 1, n = 604; day 2, n = 971; day 3, n = 615; day 4, n = 568; day 5, n = 348; day 6, n = 234; day 7, n = 122; day 8, n = 65; E, Positive tourniquet test; OFI versus DF: chi-square test for trend 20.124, p,0.0001; day 1, n = 256; day 2, n = 496; day 3, n = 402; day 4, n = 308; day 5, n = 202; day 6, n = 156; day 7, n = 78; day 8, n = 38; F, Abdominal pain; OFI versus DF: chi-square test for trend 9.149, p = 0.002; DF versus severe dengue: chi-square test for trend 4.127, p = 0.04; day 1, n = 609; day 2, n = 1,245; day 3, n = 1,066; day 4, n = 877; day 5, n = 675; day 6, n = 482; day 7, n = 290; day 8, n = 174; All other chi-square tests for trend comparing DF to severe dengue were non-significant. OFI, other febrile illness; DF, dengue fever; Severe dengue = dengue hemorrhagic fever, dengue shock syndrome, dengue with signs associated with shock, or dengue fever with compensated shock. Leukopenia is defined as WBC #5000 cells/mm 3 because this feature appears later in the entire course of dengue illness. On all days of illness, dengue-positive cases had increased odds of the same signs and symptoms as on days 1-3 of illness; however, the magnitude of the point estimates tended to be higher. This difference was most pronounced for rash and platelets #150,000 cells/mm 3 , which had ORs approximately double in magnitude. In addition, denguepositive cases had increased odds of three additional signs and symptoms: poor appetite, absence of cough, and increased hematocrit. When GEE analyses on data with the longitudinal structure preserved were compared to traditional logistic regression analyses on data collapsed on febrile episode, the point estimates for the ORs were similar, although the 95% confidence intervals for the logistic regression models tended to be slightly narrower (data not shown). In this study, we describe the clinical spectrum of pediatric dengue starting early in illness in a community setting. Longitudinal statistical analysis of day-by-day clinical signs and symptoms revealed significant associations with testing dengue-positive and important differences during the early phase of illness compared to the entire course of illness. These results stress the importance of considering day of illness when developing prediction algorithms for real-time clinical management. The early identification of dengue cases and particularly those at risk for severe dengue is critical for preventing severe illness and death. We found that 25% of laboratory-confirmed dengue cases did not meet the WHO case definition, suggesting that the WHO criteria are not sufficient to identify dengue at younger ages. Younger children may experience different signs and symptoms from adults or may be unable to communicate their symptoms to their parents, health care providers, or both. Previous studies demonstrated that children may experience significantly more cough, vomiting, abdominal pain, rash, epistaxis, oliguria, thrombocytopenia, hepatomegaly, and shock compared to adults, although the direction of these differences was not consistent across studies [13, 15, [29] [30] [31] [32] [33] [34] . A recent study of dengue in adults showed significant differences in clinical features and outcomes across ten-year age groups, indicating that signs and symptoms associated with DENV infection may continue to evolve past childhood [12] . If these differences are confirmed, the WHO case Table 3 . Signs and symptoms associated with testing DENV-positive among patients using generalized estimating equation models. Days Generalized estimating equation models assume an exchangeable correlation structure with robust standard errors. DENV, dengue virus; OR, odds ratio; CI, confidence interval; aOR, adjusted odds ratio. definition may need to be adjusted to be age-specific to function effectively for younger children and older age groups. Retro-orbital pain and low platelets were among the clinical features independently associated with DENV infection in this study. These results are supported by a study of dengue patients in Puerto Rico in which data were recorded at the time of initial consult rather than at hospitalization [15] , and by a study of Thai children [11] . Moreover, our results showing increased frequency of abdominal pain in patients beginning at day 3 of illness are consistent with a prospective study of adults admitted to an emergency department in Martinique [35] . A positive tourniquet test using cut-offs of $10 and $20 petechiae/in 2 was also independently associated with DENV infection. Both cut-offs were used because studies have indicated that a cut-off of $10 may improve discrimination of DENV infection [20, 36] ; however, the 1997 WHO classification scheme specified a cut-off of $20 [26] . Our results support using a cut-off of $10 petechiae/in 2 , and this cut-off has been specified in the 2011 WHO clinical guidelines [37] . A major strength of this study is the use of statistical models designed for analysis of longitudinal data. Few other prospective community-based cohort studies have analyzed early clinical features in pediatric dengue compared to OFI [20, [38] [39] [40] , and none that we are aware of were analyzed using longitudinal statistical methods that account for correlations between repeated measures on patients. Here, we preserved the longitudinal structure of the dataset by using statistical models that support repeated measurements on subjects over time and account for correlations between signs and symptoms experienced within the same individual on different days of illness and in multiple episodes. Longitudinal data have long been collected in dengue research but have rarely been analyzed using appropriate statistical methods. This may introduce bias into findings, as studies may overestimate the magnitude of association or reduce the statistical power of the study as data are lost when they are collapsed for non-longitudinal analysis. An additional strength of this study is that it is community-based [21] , enabling day-by-day capture of information on the early course of illness and on the full clinical spectrum of symptomatic dengue. In contrast, nearly all previous studies enrolled patients upon presentation to a hospital [18] , where patients present later; thus, these studies were unable to capture information on the early days of illness or on mild disease. By examining the clinical spectrum of dengue by day of illness, we were able to detect differences in the prevalence of signs and symptoms that could not be revealed by simply analyzing whether they ever occurred over the course of illness. In addition, through multivariable longitudinal models, we were able to identify distinguishing features of dengue during the early phase of illness compared to the entire course of illness. These findings are important for clinical practice since outside of the hospital setting, clinicians may see dengue patients toward the beginning of their illness and utilize that information to decide whether their patient has dengue or another febrile illness. The results of these models should be extended for the development of prediction algorithms to aid clinicians in diagnosing suspected dengue. This study was not without its limitations. Some participants migrated out of the study area or withdrew from the study; however, our retention rate was approximately 95% per year [21] , suggesting that any bias from loss to follow-up would be minimal. It is also possible that we did not capture all symptomatic dengue cases. However, in yearly participant surveys, only an average of 2-3% of participants reported having attended a health-care provider outside of the study or having an illness and not attending any medical provider [21] , and approximately 20-fold more laboratory-confirmed dengue cases were captured in the cohort study than by the National Surveillance System [41] . Unfortunately, due to the low number of severe dengue cases, this study did not have sufficient statistical power to compare severe dengue cases to DF cases using GEE models, and these low numbers may have influenced the lack of significant association of signs of severe dengue with testing dengue-positive. For this study, we used the 1997 WHO classification scheme for disease severity. In 2009, the WHO updated its guidelines for classification of dengue disease severity [1, 37] ; it would be interesting to re-analyze the data in a future study using the new classification scheme. Studies of the usefulness and applicability of the revised guidelines have been recently performed [42, 43] . In summary, this study is one of the few cohort studies to provide early data on the full clinical spectrum of pediatric dengue. Though we found significantly increased odds for association of several clinical signs and symptoms with testing dengue-positive, these increases were more modest for the early phase of illness compared to the course of illness, suggesting that caution should be taken when using the results from the entire course of illness to develop prediction algorithms. Non-parametric methods such as decision tree analysis overcome some of the limitations of traditional logistic regression models and have recently been applied to develop algorithms for prediction of dengue diagnosis and disease severity [9, 44, 45] . These and other data-adaptive approaches such as Super Learner [46] that are less subject to bias should be further explored to develop prediction algorithms for early identification of dengue cases and improved clinical management. Checklist S1 STROBE checklist for cohort studies. (DOC)

Influenza A H1N1 2009 (Swine Flu) and Pregnancy

The Influenza A H1N1 pandemic (A H1N1) occurred between June 2009 and August 2010. Although the pandemic is now over, the virus has emerged as the predominant strain in the current seasonal influenza phase in the northern hemisphere. The A H1N1 influenza is a novel strain of the influenza A virus and is widely known as swine flu. The virus contains a mixture of genetic material from human, pig and bird flu virus. It is a new variety of flu which people have not had much immunity to. Much has been learnt from the Pandemic of 2009/2010 but the messages about vaccination and treatment seem to be taken slowly by the clinical profession. Most people affected by the virus, including pregnant women, suffer a mild viral illness, and make a full recovery. The median duration of illness is around seven days. This influenza typically affects the younger age group i.e. from the ages of 5–65 years. Current experience shows that the age group experiencing increased morbidity and mortality rates are in those under 65 years of age. Pregnant women, because of their altered immunity and physiological adaptations, are at higher risk of developing pulmonary complications, especially in the second and third trimesters. In the United Kingdom, twelve maternal deaths were reported to be associated with the H1N1 virus during the pandemic and clear avoidable factors were identified (Modder, Review of Maternal Deaths in the UK related to A H1N1 2009 influenza (CMACE). www.cmace.org.uk, 2010). The pregnancy outcomes were also poor for women who were affected by the virus with a fivefold increase in the perinatal mortality rate and threefold increase in the preterm delivery rate (Yates et al. Health Technol Assess 14(34):109–182, 2010). There continues to be a low uptake of the flu vaccine and commencement of antiviral treatment for pregnant women.

Although the outbreak of influenza A H1N1 2009 appeared first in Mexico in April 2009, this was followed by a growing number of cases reported across the globe. The outbreak of the novel A H1N1 virus (swine flu) was declared a global pandemic by the World Health Organisation (WHO) from 11 June 2009 until 10 August 2010. These pandemics happen when a new influenza virus, to which the population has little or no immunity emerges and starts to spread. Unlike seasonal influenza, high rates of disease due to a pandemic virus may occur throughout the year. Swine flu is a novel strain of the influenza A virus affecting humans and contains segments of genes from pig, bird and human influenza viruses. The WHO classifies the spread of infections such as influenza in terms of phases. Phase 6 or a global pandemic is declared when the infection is characterized by humanto-human spread of the virus with community level outbreaks in at least two WHO regions, meaning that there is widespread global transmission. The influenza A virus has been responsible for three global pandemics in the last century: the Spanish Flu in 1918, Asian Flu in 1957 and the Hong Kong Flu in 1968. These pandemics were responsible for a large number of fatalities, the Spanish Flu being the most severe and caused severe pneumonias, particularly among pregnant women (Table 1) . A recently published epidemiological study about cohorts born in and around the pandemic estimated that the individuals born had a[20% excess cardiovascular disease at 60-82 years of age, relative to those born without exposure to the influenza epidemic, thus suggesting that prenatal exposure to even uncomplicated maternal influenza may have lasting consequences later in life [3] . The 1918 pandemic was associated with a high maternal mortality rate of 27% and also associated with high rates of spontaneous miscarriage and preterm labour. In the 1957 pandemic, a 20% maternal mortality rate was reported and increased incidences of birth defects such as neural tube defects and cardiac abnormalities were reported. Although we are now in the post pandemic phase, the A H1N1 virus has now emerged as the predominant strain of virus in the seasonal influenza season that is currently affecting the northern hemisphere. Current experience in the United Kingdom shows that the population below the age of 65 years are worst affected by the complications of the flu and that the deaths associated with the flu are predominantly associated with the H1N1 virus. From October 2010 to early January 2011, 50 deaths were reported by the Health Protection Agency. Forty-five of these people died with the H1N1 (2009) strain and 5 with Influenza B. The majority were under 65 years of age and five were under the age of five [4] . Every year, we experience an outbreak of the seasonal flu during the winter months. It affects around 5-15% of the population and most of the complications and deaths tend to affect the elderly population. The swine flu virus, however, typically affects the younger population, i.e. from 5 to 65 years. Most will experience a mild illness and make a full recovery. Children and young adults appear to be most susceptible to clinical infection with the highest incidence in 5-9 and 10-14 year olds, with a smaller number of cases in adults older than 50 years of age (born before 1957). The low number of cases seen in those aged over 50 is thought to be due to previous exposure in this age group to similar strains of H1N1 that circulated between 1918 and 1957. The median duration of illness is estimated at 7 days. However, people who are in the at risk group such as those with underlying medical conditions e.g. chronic respiratory disease, asthma, diabetes, chronic heart disease, chronic renal disease, chronic liver disease, morbidly obese, those who are on medication to suppress their immune system and children under the age of 5 years. Pregnant women are all at higher risk of developing complications should they catch the infection. The illness may be complicated by bronchitis, or viral or secondary bacterial pneumonia. This virus is an Orthomyxovirus. This is an enveloped virus with the characteristic spike-like viral glycoproteins called Haemagglutinin and Neuraminidase, hence their description of H1N1, H5N1 etc. depending on the type of H or N antigens. Haemagglutinin is an antigenic glycoprotein which causes erythrocytes to clump together. It is responsible for The influenza virus is known for its ability to mutate. This new strain appears to be a result of reassortment of human influenza and swine influenza viruses, in all four different strains of subtype H1N1. Reassortment is the mixing of genetic material of a species into new combinations in different individuals. This process has been responsible for some of the major genetic shifts experienced in previous pandemics. This novel strain of the H1N1 Influenza A subtype is a result of a mixing of the virus from humans, birds and pigs. The typical symptoms of swine flu are a sudden fever of at least 38°C and sudden cough with at least one other symptom of chills, lethargy, dehydration, headache, sorethroat, coryza, diarrhoea, vomiting, abdominal pain, myalgia or arthralgia. Gastro intestinal symptoms (vomiting and diarrhoea) have also been reported more often with H1N1 than with seasonal flu (Table 2A) . The likelihood of infection with pandemic H1N1 2009 influenza is strongly influenced by age; the young are at most risk and those aged [60 years at least risk. Approximately 60% of patients with confirmed H1N1 2009 influenza virus infection are aged 16-64 years. However, case fatality increases with age especially those with associated co-morbidities, the young children \5 years of age [5] and pregnant where it is also high (Table 2B) . In the UK, the first case of Influenza A H1N1 was confirmed on 27th April 2009. The first wave of the pandemic in UK started in July 2009 with the second wave from September 2009 and carrying on over the winter months. The vast majority of patients in the first wave in the UK experienced mild illness with 50% of patients recovering within 7 days of symptom onset and a further 25% within 10 days. The main symptoms reported were fever, fatigue, dry cough, sore throat and headache. However, severe gastrointestinal disease (nausea, vomiting, diarrhoea, abdominal pain) has been a feature of H1N1 infection in children and adults requiring admission [6, 7] . Complications of swine flu in the general population appear similar to seasonal influenza. Myocarditis has been observed, usually associated with a marked tachycardia. The prognosis is unclear, though influenza-related myocarditis usually has a good prognosis for recovery. As with seasonal influenza, neurologic complications such as seizures and encephalitis, with altered mental state, can occur. The prognosis for patients with isolated neurological symptoms which can be optimised without requirement for ICU admission appears good. The initial symptoms of rapidly-lethal infections such as meningitis, encephalitis and bacteraemia can resemble those of influenza [8] . Although in the majority of cases of H1N1 2009 influenza affects the upper respiratory system, the most common complication is that of influenza-related pneumonia. This has occurred in 40% of hospitalised patients in the United States, usually with bilateral infiltrates apparent in the chest X-ray [9, 10] . Amongst patients with H1N1 influenza admitted to ICUs in New Zealand and Australia, 49% had viral pneumonitis or ARDS and 20% had secondary bacterial pneumonia. The most common bacterial pathogens reported included Streptococcus pneumonia, S. pyogenes, Staphylococcus aureus, S. mitis and Haemophilus influenzae [11] . The virus is highly contagious and is contracted by aerosol transmission from inhaling the droplets from an infected person coughing and sneezing, and also touching the nose and mouth after being in contact with contaminated surfaces. The best way to reduce the risk of spreading the virus is to observe good respiratory hygiene by covering the mouth and nose with a piece of tissue paper when coughing and sneezing and disposing of it immediately. In addition, it is important to observe good hand hygiene by washing There is no evidence to show that the risk is reduced by avoiding crowded places as long as these good hygiene measures are observed. Naturally, if anyone is down with the flu, they should avoid going to crowded places themselves (Table 3) . There is no evidence that face masks are effective in reducing the risk of catching or spreading the virus and are not recommended for use by the general population. They are only effective for use in a healthcare setting. In most clinical situations, a patient admitted to hospital with suspected swine flu or its complication, they will be managed in isolation and barrier nursed. Healthcare professionals should don a plastic apron and surgical face masks, observing strict hygiene measures. The respirator masks (FFP 3 masks) should only be used when potentially aerosol generating procedures are used e.g. in ventilated patients or those using nebulisers. Patients suffering from the following symptoms are more likely to be admitted to the Intensive Care Unit These are: • dyspnoea (strongly predictive of both death and ICU requirement), respiratory rate [30 per minutes • requirement for supplemental oxygen (strongly predictive of ICU care and death) • pneumonia on admission (strongly predictive of significant complications after admission-including ICU multi-organ support and death) • tachycardia (the higher the pulse the greater the chances of ICU care being required) • altered conscious level It is commonly thought that a pregnant woman's immunity is suppressed. In pregnancy, there is modification of the immune system to prevent rejection of the fetus and placenta. It is thought that the cell mediated immunity is modified by an increased in the production of T-helper cells as opposed to T-killer cells to reduce the risk of rejection. There appears to be a tendency towards humoral immunity. As the virus replicates intracellularly, the ability to fight off an infection is reduced. It is known that infections such as chicken pox get worse in pregnancy, particularly in the later half of pregnancy, but this is due to the risk of pneumonitis from local spread to the respiratory system of the virus and also the mechanical effects of the enlarged uterus on the diaphragm and the respiratory system. There is no evidence that Human Immunodeficiency Virus (HIV) or tuberculosis get worse in pregnancy. Hence, there is no evidence to suggest that pregnant women are at increased susceptibility of catching the flu virus [13] . However, due to the modification in their immune systems to accommodate their developing fetus and adaptations in their body as a result of the hormonal and physical changes, they are at greater risk of developing complications should they acquire the illness. The enlarging uterus presses on the diaphragm and together with changes in the lungs such as reduced tidal volume, congestion and localized oedema, make the woman more prone to complications such as pneumonia and Adult Respiratory Distress Syndrome (ARDS). Recent reports have shown that the percentage of hospitalised cases requiring transfer to ICU within each gestational band is similar (27, 25, 24%); and the death rate amongst hospitalised patients within each gestational age group is also remarkably uniform (10, 6, 9%). This may imply that outcome in severely ill pregnant women is independent of gestational age after severe illness is established [14] . There is still much to learn about the virus. Studies from the USA showed that pregnant women with swine flu were four times more likely to be hospitalised for complications compared to the non-pregnant population. In separate studies pregnant women are over represented in the group of patients admitted to hospital requiring Level 2 (High Dependency Care) or Level 3 care (Critical/Intensive Care). Observations from the USA, Canada and Australasia showed that pregnant women formed between 7 and 9% of admissions to Intensive Care Units (ICU). In the United Kingdom, the United Kingdom Obstetric Surveillance System (UKOSS), working with the Department of Health and Health Protection Agency, collected information on pregnancies reported to be affected by A H1N1 from 221 of the 223 (99%) obstetric units in the UK during the pandemic. UKOSS assessed the data on morbidity in pregnancy. The Centre for Maternal and Child Morbidity UKOSS was able to obtain complete data on 241 women admitted to hospital with confirmed A H1N1. During that period, there were 314,135 maternities, representing an incidence of 7.7 hospitalised cases per 10,000 maternities. Twenty percent (1.6 per 10,000 maternities) of these women were admitted to the Intensive Care Unit. Only 2% of women admitted had received the A H1N1 vaccine but it has to be noted that the vaccination programme in the UK was rolled out after the peak of the hospital admissions in the series. Sixty percent of the hospitalised women commenced antiviral treatment within 48 h of onset of their symptoms . (Table 4) Obesity and delay in treatment in antiviral treatment were significant factors associated with admission to ICUs. Treatment with antiviral agents within 2 days of onset of symptoms was associated with an 84% decrease in the odds of admission to an ICU. As for pregnancy outcomes, a three fold increase in preterm delivery and a five fold increase in stillbirth rate was observed. Between 1 April 2009 and 13 January 2010, 12 maternal deaths in the United Kingdom and 1 maternal death in the Republic of Ireland with confirmed A H1N1 were reported to the Centre for Maternal and Child Enquiries (CMACE). Eight of the maternal deaths were investigated in detail by CMACE through their confidential enquiry panel. As in the observations noted by UKOSS, Black and ethnic Minority (BME) women were over-represented (4/8) . Clinical co-morbidities such as asthma, paraplegia, previous stroke and stroke were present. Obesity was found to be a risk factor for severe morbidity and mortality. Smoking did not appear to be a risk factor although both studies noted them to be a factor in the younger women. Clear avoidable factors were noted and they are: • Delayed admission to hospital and/or consideration of the likelihood of A H1N1 infection in the differential diagnosis; • Delayed confirmation of A H1N1; wrong swabs used or inappropriate storage/carriage to laboratory; • Lack of clear clinical leadership in overall case management (1 case); • Delay in administering antiviral medication. Pregnant women admitted with respiratory complications should be managed jointly between the obstetric and medical teams. An assessment needs to be made with respect to the best place to manage the woman. There should be early involvement of obstetric anaesthetists, respiratory physicians and haematologists and a clear management plan needs to be set out from the outset. Women requiring more respiratory support may be best managed in the Respiratory Unit with close input by the obstetric and midwifery teams. If a woman is in labour, she would be best managed in the Delivery Unit with input from the Respiratory team and the Obstetric Anaesthetic team. Following delivery, she may need to be transferred to the clinical area that would be best to provide expert care for her. It is important to bear in mind that pregnancy related pathologies such as chorioamnionitis, severe urinary tract infections, group A and B streptococcus infections and even malaria may present with similar symptoms. Careful clinical assessment is of vital importance in order not to miss an important infection. The patient needs to be carefully monitored with the modified early warning scores, taking into account the pregnancy. Complications of obstetric problems such as pre-eclampsia, venous thromboembolism and pulmonary embolism have also to be excluded. The most common complication of A H1N1 influenza is pneumonia. The plan of management in such cases has been summarized in Table 5 . Complications must be recognized and treated appropriately. In order to treat pneumonia effectively, the antimicrobial therapy should be based on bacteriological sensitivities. However, pending bacteriological confirmation, co-amoxiclav (Augmentin) is the recommended empiric antibiotic. In penicillin sensitive patients, Clarithromycin is a suggested alternative. Co-amoxiclav is not contraindicated in pregnancy. In situations where there is preterm prelabour rupture of membranes concomitant with pneumonia, the use of co-amoxiclav is not recommended because of the increased risk of necrotizing enterocolitis observed in the ORACLE study. An alternative antibiotic such as piperacillin is recommended and the microbiologist should be involved in management discussions. In addition, antiviral treatment should be started on clinical grounds whilst awaiting test results. It is also important to treat maternal pyrexia with paracetamol and not Non Steroidal Anti Inflammatory Drugs (NSAID). Epidemiological studies have linked uncontrolled maternal pyrexia to miscarriage and fetal abnormalities such as neural tube and cardiac defects. Maternal pyrexia is also a recognised risk factor for preterm delivery. The importance of control of maternal pyrexia with regular and effective doses of paracetamol and hydration should be emphasised. Current obstetric practice is to administer corticosteroids e.g. 2 doses of betamethasone or dexamethasone 12 or 24 h apart to promote fetal lung maturity in situations of threatened pre-term labour or where a decision is made to deliver the fetus prematurely for maternal or fetal reasons. The effects on the maternal immune system from a single course of corticosteroids are unclear but the evidence does not suggest that it results in sufficient immuno-suppression to cause maternal harm or exacerbation of infection. The administration of corticosteroids is important for the promotion of fetal lung maturity, and the benefits outweigh the risks. Recent evidence suggests that repeated (rescue) doses of corticosteroids may be beneficial for fetal reasons, but studies have also shown that these may lead to maternal secondary adrenal insufficiency and fetal complications. The practice of repeated (rescue) doses of corticosteroids is not recommended. The use of high dose steroids for the treatment of pulmonary complications is not recommended by ICU specialists for fear of immune suppression and prolonged viral shedding. Most mothers with symptoms of influenza in labour will be able to tolerate labour with adequate pain relief and hydration. In most cases, the decision to deliver will be made for an obstetric indication. In the event of a critically ill woman close to term, it is not unusual to deliver her baby, usually by caesarean section, to help with mechanical ventilation of the lungs to improve her recovery. This should be done once her clinical condition is stabilised and other potential complications such as coagulopathy have been excluded or corrected. As indicated above, most of the respiratory complications have been shown to occur in the second and third trimesters. There may be situations where a preterm baby needs to be delivered in order to improve the outcome for ventilation of a very ill mother. The decision is made in conjunction with the obstetric, critical care and neonatal teams. The pregnant woman should be informed but if unable to participate in clinical decision making, the partner or close relatives should usually be involved in discussions. In order to improve the outcome for the premature infant, corticosteroids (in accordance with the guidance above) should ideally be administered at least 24 h prior to delivery. It is unlikely that pregnancies in the 1st or early second trimesters will need to be terminated unless it is felt that continuation of the pregnancy will be detrimental to the woman's condition. There may be occasions where a woman who is booked for elective caesarean section becomes symptomatic at the time of the planned procedure. If possible, it would be advisable to commence her on antivirals and to delay the procedure by about 5 days, to allow her to recover, in order not to increase her risks of respiratory complications, and also to reduce the risk of spread to other patients and staff. In severe cases of respiratory complications where the woman has developed Adult Respiratory Distress Syndrome (ARDS) and where she is not responding to mechanical ventilation, she may have to be considered for Extra Corporeal Membrane Oxygenation (ECMO). This is highly Be prepared to deal with: DIC, post viremia encephalitis Anti microbial therapy should be based upon bacteriological sensitivities-be cautious with the use of Augmentin in women with ruptured membranes The anti viral treatment should be started ASAP Maternal pyrexia should be treated with paracetamol Consider antenatal steroids for preterm labour specialized treatment and the decision is made by the Intensive Care specialists. As above, it may not always be necessary to terminate a pregnancy in the first or early second trimesters as there is no benefit in doing so for maternal ventilation. It is of vital importance to maintain the maternal physiology in the best possible state in order to allow the satisfactory progress of the pregnancy. There is now more capacity for ECMO treatment in the UK and more pregnant women are being referred to ECMO units for treatment. It is vitally important that the obstetric teams in the ECMO units are jointly involved in the mother's care and good communication is essential. Women in the postnatal period are probably at lower risk of respiratory complications because the effects of the gravid uterus on the lungs have been removed. However, they may still experience similar complications if they are infected and there is a risk of transmission to the newborn infant. They should observe the same strict hygiene measures and be offered antiviral medication i.e. oseltamivir if clinically indicated. Mothers should be encouraged to breastfeed. Breast feeding is important and should be continued as long as possible. The benefits of breastfeeding are significant and are twofold: (i) it gives babies the most appropriate nutrition for health and promotes attachment between mother and baby and (ii) colostrum is rich in antibodies which will help to protect the baby from many infections. Women who are breastfeeding and have symptoms of influenza should be treated with an antiviral medicine. The preferred medicine is oseltamivir, as for other adults. However, if a baby is born and breastfeeding is started while the woman is taking zanamivir, she should complete the course of zanamivir; it is not necessary to switch to oseltamivir. If the woman is too tired to breastfeed, she should be encouraged to express her breast milk in order to feed her baby. She should get help with breastfeeding and her baby should be with her as much as possible. She should observe strict hygiene measures to avoid spread of the virus to her baby. The two types of antiviral drugs known to be effective against the swine flu virus are oseltamivir (Tamiflu Ò ) and zanamir (Relenza Ò ). They are neuraminidase inhibitors and act by preventing the virus from budding and escaping from the host cells. Oseltamivir is given in the form of oral capsules and zanamivir is given as an inhaler (Diskhaler). The safety of these antiviral drugs in pregnancy has been looked at from experience with their usage in seasonal flu in some countries and from recent data from Canada. Although oseltamivir has been shown to cross the placenta and breast milk in small amounts, no adverse effects on the fetus or pregnancy have been recorded. As zanamivir acts in the respiratory tract with no absorption into the blood stream, it is recommended as first line treatment of swine flu in pregnancy in the United Kingdom. However, if the woman is unable to tolerate the inhaler e.g. if she has severe asthma or is admitted to hospital with complications such as pneumonia, oseltamivir is the recommended drug for treatment. The benefits of antiviral treatment are most noticeable if it is commenced within 48 h of symptom onset but recent experience with hospitalised patients reveals that antivirals given more than 48 h after symptom onset also confer benefit. Treatment with antivirals should be started on clinical grounds whilst awaiting test results. Waiting for confirmatory tests will only lead to a delay in commencement of treatment. Clinical judgement should be exercised in the initiation of treatment. A negative test does not necessarily exclude H1N1 influenza as the sensitivity of rapid influenza diagnostic tests can range from 10 to 70% for 2009 H1N1 virus. Pre or post exposure prophylaxis with antivirals has been shown to be effective in preventing the development of the influenza infection. Unlike in treatment after symptoms, prophylaxis may inhibit the development of immunity and prolonged or repeated prophylaxis may predispose to development of resistance. Prophylaxis is only recommended for very high-risk individuals for whom prompt treatment may not avoid the development of severe disease. Most pregnant women do not require prophylaxis. In the United Kingdom, the vaccination programme was launched in October 2009. Healthcare workers were the first group to be offered the vaccine. Following this, at risk groups, pregnant women and children between 6 months and 5 years were invited to be vaccinated. It is of no doubt that any programme to vaccinate pregnant women will raise concerns and controversy. New vaccines cannot be tested on pregnant women but the experience on its safety has been drawn from data from seasonal flu vaccines. The swine flu vaccine is similar to the seasonal flu vaccine in many respects. The vaccine contains the inactivated (killed) H1N1 virus which is developed from the bird flu H5N1 virus-a virus to which most people have no immunity. The inactivated virus will not cause any harm to the fetus or mother and active immunity will develop. The antibody response to vaccines in pregnant women has been shown to be as effective as those who are not pregnant. The reluctance to be vaccinated is mainly due to worries about the side effects. Experience from the 2009/2010 pandemic confirmed the safety profile of the vaccine. Over 350 million doses of the vaccine were administered worldwide during the pandemic with no adverse long term effects noted in both the pregnant and non pregnant populations [15] . In the UK, the seasonal influenza vaccines for 2010 have incorporated the A H1N1 strain as part of the trivalent vaccine. In spite of the proven safety profile, until the end of December 2010, the uptake of the vaccine in the younger population who are at risk, including pregnant women, remained low with less than 50% uptake [4] . Communication strategies have been stepped up to promote the value of the vaccine and to heighten awareness of the severity of complications affecting the younger at risk population. Although we are now experiencing the post pandemic phase, the A H1N1 virus has emerged as the predominant virus in the current seasonal influenza season affecting Europe and the UK, with significant morbidity and mortality rates affecting the younger population and pregnant women. Most people who catch swine flu experience mild symptoms and make a full recovery. However, pregnant women are more likely to develop life-threatening complications should they catch the flu. It is important for pregnant women seek medical advice early and are started on antiviral treatment, ideally within 48 h, but they can also be effective up to 7 days. If, in spite of treatment, they do not get better, they must contact their doctor or midwife. It is vitally important to treat any fever in pregnancy with paracetamol and drinking plenty of fluids. Strict hygiene measures are effective in preventing the spread of the infection. Vaccines are the most effective way to avoid catching the virus and pregnant women should consider taking up the offer of vaccination against swine flu.

Use and Evaluation of Molecular Diagnostics for Pneumonia Etiology Studies

Comprehensive microbiological testing will be a core function of the Pneumonia Etiology Research for Child Health (PERCH) project. The development stage of PERCH provided the time and resources necessary for us to conduct a comprehensive review of the current state of respiratory diagnostics. These efforts allowed us to articulate the unique requirements of PERCH, establish that molecular methods would be central to our testing strategy, and focus on a short list of candidate platforms. This process also highlighted critical challenges in the general design and interpretation of diagnostic evaluation studies, particularly in the field of respiratory infections. Although our final molecular diagnostic platform was ultimately selected on the basis of operational and strategic considerations determined by the specific context of PERCH, our review highlighted several conceptual and practical challenges in respiratory diagnostics that have broader relevance for the performance and interpretation of pneumonia research studies.

Comprehensive microbiological testing will be a core function of the Pneumonia Etiology Research for Child Health (PERCH) project. The development stage of PERCH provided the time and resources necessary for us to conduct a comprehensive review of the current state of respiratory diagnostics. These efforts allowed us to articulate the unique requirements of PERCH, establish that molecular methods would be central to our testing strategy, and focus on a short list of candidate platforms. This process also highlighted critical challenges in the general design and interpretation of diagnostic evaluation studies, particularly in the field of respiratory infections. Although our final molecular diagnostic platform was ultimately selected on the basis of operational and strategic considerations determined by the specific context of PERCH, our review highlighted several conceptual and practical challenges in respiratory diagnostics that have broader relevance for the performance and interpretation of pneumonia research studies. The development of a comprehensive microbiological testing strategy has been a core principle in the conception and design of the Pneumonia Etiology Research for Child Health (PERCH) project [1] . In formulating the most effective approach for respiratory diagnosis, we determined that a multiplex molecular diagnostic platform would be an essential component in our approach. Many of the technical and operational considerations encountered through this process proved relevant to the overall design of the project. We describe here the theoretical and practical challenges encountered in the evaluation and selection of a molecular platform for the diagnosis of pneumonia. As described elsewhere in this issue [2, 3] , microbiological evidence of infection must be considered in the context of several fundamental difficulties found in respiratory diagnostics, including the frequent lack of access to the site of infection, the insensitivity of available tests, insufficient assay validation, and complexities in determining whether a detected pathogen has a causal role in the illness. The specific research-related demands of PERCH added to these constraints, requiring that our diagnostic strategy must exclude any prior assumptions regarding the likely importance of specific pathogens; must include a full range of respiratory tract specimens, including upper respiratory swab or aspirate, induced sputum, lung aspirate, bronchoalveolar lavage, and pleural fluid; must be comprehensive, yet realistic; must appropriately balance the demands of accuracy and efficiency; must account for both clinical and research ethical issues; and must be feasible for use and support at all participating field sites. To begin the selection process, the PERCH investigators conducted an extensive review of the microbiologic diagnosis of respiratory infections. Using published and unpublished data, as well as user and developer experiences, our team prepared a strategic summary of the available technologies that could detect pathogens from respiratory tract specimens. We evaluated each major assay category, including traditional bacteriology and viral culture, direct antigen and immunofluorescent antibody detection, and nucleic detection acid tests. It was evident that molecular diagnostics should be among the mix of diagnostic tools required to meet the needs of PERCH. Nucleic acid detection tests (NADTs) have a number of advantages over other diagnostic platforms for the evaluation of respiratory specimens [4] . They demonstrate superior sensitivity in detecting organisms that are fastidious, less viable, or present in only small amounts [5] . Molecular diagnostics can also be quickly adapted to detect evolving or emerging pathogens and are amenable to efficiencies of scale such as automation. They also allow the simultaneous detection of multiple targets (multiplexing), which in turn allows for testing by clinical syndrome and the detection of co-infections. NADT methods present less of a safety hazard for laboratory personnel compared with culture, typically require less time compared with bacterial culture, and require less technical capacity compared with viral culture. Given these advantages, NADTs have been extensively evaluated in the detection of several viruses and bacteria of the respiratory tract and have become the diagnostic tool of choice for many agents that are difficult to isolate [4] . Molecular diagnostic platforms are not without their disadvantages. Cost and complexity remain significant barriers to adoption in many laboratories, and NADTs often risk problems of laboratory contamination with amplified products, particularly if the assay procedure requires opening of the reaction tube prior to the target detection step [6] . Measures to limit contamination often require additional laboratory space that may not be available in resource-constrained settings. Nevertheless, NADT methods represent one of the more productive areas of diagnostics research, promising future improvements in automation and speed, smaller devices, improved cost efficiencies, and better detection of emerging pathogens [7] . The focus on molecular methods for respiratory pathogen detection yielded a large variety of potential technologies for consideration in PERCH (Table 1) . Polymerase chain reaction (PCR) technology is more common at research sites worldwide, can be adapted to various platforms, and easily allows for multiplex amplification. Multiplexing, in which several targets are assayed for simultaneously, is commonly employed in PCR-based assays and offers significant advantages over single-pathogen assays in terms of efficiency and pathogen coverage. Still, developers must overcome considerable complexities in harmonizing the reaction requirements of each individual target and limiting potential competition among the analytes. These factors may result in a measurable decrease in sensitivity compared with single-plex assays. Several techniques have been developed to address such factors, such as alterations in cycling protocols [8] , nested primer combinations [9, 10] , complex primer structures and concentrations [11, 12] , and the use of nontraditional nucleotides [13] [14] [15] . Other NADT technologies, such as nucleic acid sequence-based amplification and loop-mediated isothermal amplification, have been used for the detection of respiratory pathogens, but experience with multiplexing is limited [16] [17] [18] [19] . Technologies for target detection take on an even larger variety of formats. Older methods include agarose gel electrophoresis, reverse-transcription PCR enzyme hybridization assay [20] [21] [22] and enzyme-linked oligonucleotide capture [18] . More recently, solid-and liquid-phase array platforms have become more useful for the detection of multiple targets. Solid-phase arrays use a variety of formats, typically embedding target-specific oligonucleotides onto a glass or silicon microchip [23] [24] [25] [26] [27] [28] [29] [30] [31] to detect anywhere between dozens to hundreds of thousands of amplified sequences. Several respiratory diagnostic systems have been based on a liquid-phase technology using polystyrene microbeads (Luminex) [9, 10, [13] [14] [15] 32] or mass spectroscopy [33, 34] for amplicon discrimination. Although these approaches have greatly expanded the versatility and sensitivity of multiplex PCR, their complexity, specialized equipment, and high start-up costs have limited their widespread adoption to date. Moreover, these platforms typically require separate steps for amplification and detection, increasing both the workload and the risk of operator error or amplicon contamination. Real-time PCR assays address these issues by combining amplification and detection in one reaction tube, thus facilitating automation and reducing contamination. In addition, this technique allows for the quantification of pathogens and the assessment of replication efficiency. As with conventional PCR, multiplex real-time assays are subject to competition and inhibition among primers [5, 35] . Real-time assays are also restricted in the number of reaction products that can be detected in parallel [35] , although this problem can be partially circumvented using arrays of uniplex real-time reactions at very small volumes [36] . Successful in-house realtime assays directed against respiratory pathogens have been developed using uniplex [37] and multiplex [38, 39] approaches, but data on the performance of commercialized versions are not readily available. Much of the effort in NADT development for respiratory diagnostics has focused on the detection of viruses, given the advantages of these techniques over conventional methods in terms of speed, sensitivity, and versatility for detecting this class of pathogens. Multiplex approaches for viral detection have become more common as technologies have improved (reviewed by [5, 7, 40, 41] ). In addition, NADTs have now become the gold standard for the detection of Mycoplasma pneumoniae [42] and Chlamydophila pneumoniae [43] and a useful addition to antigen testing for Legionella species [44] . Multiplex assays for the detection of more traditional bacterial pathogens have not been studied as frequently in respiratory specimens, primarily because culture techniques are usually adequate for clinical practice. Moreover, molecular methods provide no additional advantage over culture in differentiating infection from colonization of the upper respiratory tract. Nevertheless, multiplex NADTs for bacteria such as S. pneumoniae, Haemophilus influenzae, and Streptococcus pyogenes have been evaluated in respiratory specimens [45] and will likely be incorporated into larger multiplexing assays. Having conducted our survey of the field, we narrowed our list of candidate molecular diagnostic platforms even further on the basis of the unique needs of our research study. As with clinical laboratories, we closely examined factors such as cost, feasibility, quality assurance, capital investment, platform versatility, and future utility. In contrast with clinical laboratories, we considered issues such as the rapid return of results or regulatory approval for use in patient care to be less crucial to our objectives. Moreover, our approach emphasized comprehensive pathogen detection, rather than focusing primarily on pathogens relevant for clinical management or infection control. Finally, our selected platform would be deployed in low-resource settings, where requirements for a reliable or continuous power supply, adequate access to reagents, sensitivity to extreme environmental conditions, and access to technical support would be highly relevant. As our appraisals progressed, we encountered many challenges in interpretation that are common to the field of diagnostics evaluation. Most basic among these was confusion regarding the usage of the terms ''sensitivity'' and ''specificity,'' and the evaluations needed to measure these parameters [46] . The distinction between ''analytic'' performance characteristics, as opposed to ''diagnostic'' or ''clinical'' performance characteristics, is essential for properly assessing the validation of any assay, but it is particularly true in the field of molecular diagnostics. For NADTs, analytic sensitivity refers to the lowest concentration of target that can be detected, whereas analytic specificity measures the ability of the test to exclude undesired targets despite similar genetic sequences. In contrast, diagnostic or clinical sensitivity of a nucleic acid detection test refers to the appropriate identification of all patients carrying the agent, and diagnostic or clinical specificity describes the assay's ability to exclude uninfected patients. Clinical performance characteristics are subject to a number of factors, including the patient's disease status, variations in the concentration of the target throughout the course of illness, inhibition by other substances present in the specimen, sample quality, sampling variability, and specimen degradation. Generally, assays should be tested against a reference or gold standard. For tests of microbial detection, the reference standard typically is culture, but molecular diagnostics are often much more sensitive in detecting nucleic acid than is culture for viable organisms, leading to difficulties in interpreting the clinical relevance of false-positive results. The challenges of assessing diagnostic tests have been increasingly recognized in recent years. For instance, the Standards for Reporting of Diagnostic Accuracy initiative [47, 48] offers guidelines on the reporting of diagnostic studies, whereas the Quality Assessment tool for Diagnostic Accuracy Studies provides corresponding guidance on their evaluation [49] . Nevertheless, respiratory diagnostics are particularly limited by the inability to determine whether the detection of a particular pathogen in a symptomatic patient indicates that it is causative of the illness or results from contamination, colonization, or prolonged shedding from a prior unrelated infection, particularly when testing specimens from the upper respiratory tract. This issue is not typically addressed in diagnostic evaluation studies, but it has become more relevant as molecular diagnostics have expanded the lower limits of pathogen detection by several orders of magnitude. Attempts to answer this question have suggested an additional category of test performance, the ''epidemiological'' specificity of a test, to describe the ability of an assay to assign true etiologic status to a pathogen for a specific illness. Ultimately, determination of the epidemiologic specificity of a respiratory diagnostic would require the interpretation of microbiologic results in conjunction with all other clinical and laboratory data, perhaps in the form of a predictive model. Such analyses are uncommon but will be a main focus of the PERCH study. Respiratory diagnostics are further complicated by the absence of a perfect gold standard. Culture is difficult or insensitive for some pathogens and unavailable for others (eg, human metapneumovirus, parainfluenzavirus type 4, rhinovirus group C, or Pneumocystis jiroveci). Serologic tests are often not available and usually require paired serum specimens for accurate results. Statistical methods to adjust for such alloyed gold standards, such as discrepant analysis, have been frequently employed, but they can be susceptible to significant bias [50] . Comparative evaluations of respiratory diagnostic assays must also take into account variations in which panel of pathogens is selected, which genetic sequences are targeted, what specimen sources are used [3, 51] , and even what methods are used for nucleic acid extraction [52] . The US Food and Drug Administration has recently published industry guidance that may encourage additional work in this area [53] . As the PERCH evaluation progressed, the concepts derived from our deliberations were distilled into a list of desirable and essential attributes summarizing our strategy for evaluation ( Table 2 ). This list addressed issues such as assay performance (range of targets, acceptable specimen sources, sensitivity, and specificity), operational concerns (space requirements, assay throughput, quality assurance programs, maintenance requirements, and reagent availability), and strategic issues (capacity for automation, versatility and future utility, start-up and maintenance costs, and developer engagement). For additional input, we presented our summary to the Pneumonia Methods Working Group, an expert committee formed to advise PERCH. Ultimately, this outline of key qualities and data allowed us to articulate our thoughts and communicate our strategy more effectively to collaborators, advisors, and assay developers. We applied our list of attributes to more than a dozen candidate diagnostic systems that met our initial criteria, and developed a short list of candidate platforms. We then tested these final assays in our PERCH-affiliated laboratories, using a standardized set of mock specimens. This process allowed us to engage with the assay manufacturers and their academic partners, directly compare the performance characteristics of Nucleic acid extraction procedure included in overall process (and automated) Ability to process a variety of respiratory tract specimens Small specimen volume requirements Specimen collection requirements well-characterized and suitable for field studies Readily available reagents with long expiry dates and room-temperature storage requirements the platforms, and gain essential information that could only be acquired through hands-on experience, such as capabilities for technology transfer, ease of use, and workflow. Details of this evaluation will be the subject of a separate article. By including a phase for protocol development, the PERCH investigators were able to perform an extensive literature review of respiratory diagnostics, clearly outline the major theoretical and practical concerns, and engage a group of experts for critical input. Through this process, we confirmed the suitability of molecular diagnostics for our needs and identified critical information gaps. Our evaluation highlighted numerous advantages of this technology, including excellent sensitivity and adaptability for a full range of respiratory pathogens and specimen sources, as well as clear capabilities for multiplexing and automation. We nevertheless realized that our conclusions represent but a snapshot in time, and the field of molecular diagnostics is rapidly evolving, with constant improvements in accuracy, speed, automation, and cost. Yet it can also be expected that methods for evaluating respiratory diagnostics will continue to evolve in parallel, providing new answers to the practical and conceptual challenges that shaped the development of a diagnostic testing strategy for PERCH.

Resveratrol Inhibits KSHV Reactivation by Lowering the Levels of Cellular EGR-1

In the field of herpesvirus research, the exact molecular mechanism by which such viruses reactivate from latency remains elusive. Kaposi's sarcoma-associated herpesvirus (KSHV) primarily exists in a latent state, while only 1–3% of cells support lytic infection at any specific time. KSHV reactivation from latency is an exceedingly intricate process mediated by the integration of viral and cellular factors. Previously, our lab has described early growth response-1 (Egr-1) as an essential component for the KSHV reactivation process via its ability to mediate transcription of KSHV ORF50, the gene encoding for replication and transcription activator (RTA), a viral component known to control the switch from latent to lytic infection. In here, electrophoretic mobility shift assays (EMSA) and chromatin immunoprecipitation (ChIP) experiments revealed that Egr-1 binds KSHV ORF50 promoter (ORF50P) in at least two different GC-rich binding domains. Expression profiles of cellular egr-1 and KSHV-encoded ORF50 follow a similar pattern during de novo KSHV infection. Over-expressing Egr-1, a signaling component downstream of Raf>MEK>ERK1/2, in KSHV-infected cells activates KSHV lytic replication. Through performing more physiologically relevant experiments, we analyzed the effect of a dietary supplement containing resveratrol on KSHV-infected cells. Our results, for the first time, demonstrate resveratrol to act in lowering ERK1/2 activity and expression of Egr-1 in KSHV-infected cells, resulting in the suppression of virus reactivation from latency. Taken together, these findings will undoubtedly contribute to future studies on not only combating KSHV related disease conditions, but also on other herpesviruses-induced pathogenesis.

Significant strides have been made since the discovery of Kaposi's sarcoma-associated herpesvirus (KSHV) by Chang et al [1] nearly 20 years ago that have helped to increase our understanding of this infectious agent. KSHV is a c2-herpesvirus that has been directly linked to the development of Kaposi's sarcoma (KS), primary effusion (PEL), and multicentric Castleman disease (MCD). This virus is closely related to Epstein-Barr virus (EBV), murine gammaherpesvirus-68, and herpesvirus saimiri [2] . The prevalence of KSHV infection varies depending on the geographical location with highest levels observed in Africa, where it has been reported to be greater than 40% [3] . As KSHV displays several characteristics shared among other herpesviruses, its ability to switch between latent and lytic stages of infection is of particular concern. The virus remains predominantly in a latent state, while 1-3% of cells may support a lytic infection at any given time [4] . Regulation of the switch between the two stages of infection is mediated by viral and cellular factors. Specifically, the KSHV protein, replication and transcription activator (RTA), is known to be a crucial viral component controlling the transition from latency to a lytic infection [5] . Recently, cellular early growth response-1 (Egr-1) protein was also shown to be an important factor involved in KSHV reactivation through its ability to mediate transcription of KSHV ORF50, the gene encoding RTA [6] . Egr-1 is a transcription factor that is also known as zif268, Krox-24, NGFI-A, and TIS8 [7] . It is induced by several external stimuli including growth factors, different forms of stress, and hormones. As a result of stimulation from various factors, egr-1 gene products advance to play a role in several cellular functions such as, but not limited to, growth, proliferation, and differentiation [8] . Egr-1 is part of a zinc-finger gene family that includes Egr-2, Egr-3, Egr-4, and the Wilms tumor suppressor (WT1) [9] . TPA is used to activate a lytic infection in KSHV-infected cells [10] . Egr-1 mediates the effect of TPA activation and is a downstream target of MAPK signaling [9, 11] . Furthermore, MAPK signaling is crucial for triggering KSHV reactivation from latency [12, 13] . However, despite the ability of Egr-1 and KSHV ORF50 to interact with each other, there is little information available describing this association. In a recent study, the ability of Egr-1 to bind KSHV ORF50 promoter (ORF50P) was described [6] . In this report, electrophoretic mobility shift assays (EMSA) and chromatin immunoprecipitation (ChIP) experiments were employed to determine the locations on ORF50P that have an affinity for Egr-1 binding. Our results demonstrated at least two targets that are likely crucial for mediating Egr-1 binding to ORF50P. These findings were confirmed through the use of mutation studies. In addition, we tested the ability of resveratrol, a naturally occurring product found in a variety of fruits and nuts [14] , in regulating MAPK signaling.Egr-1 expression.promoting virus latency. As such, the ramifications on the ability of Egr-1-induced transcription of ORF50 in viral pathogenesis are discussed. Egr-1 binds at least two different sites within the ORF50P Egr-1 is said to bind a GC-rich DNA template (such as GCGC(G/T)GGGCG, GCGGGGGCG, and CGCCCATGC) on the promoter and initiate gene transcription [15, 16] . Eight possible GC-rich Egr-1 binding sequences have been identified by us in the promoter region of KSHV ORF50. In order to determine the sites where Egr-1 bound ORF50P, EMSA experiments were performed using 8 different DIG-labeled probes, referring to identified locations on ORF50P, (Table 1) and Egr-1 in vitro transcribed and translated (IVT) proteins. IVT of egr-1/ pcDNA3.1(+) construct yielded a protein of roughly 78 kda (data not shown) [6] . Of all the probes tested, IVT-synthesized Egr-1 proteins were able to bind and form separate protein:DNA complexes with the ORF50P3 and ORF50P8 probes, respectively; displaying distinct band shifts when compared to controls using the probes alone ( Fig. 1A ; lanes 6 and 16). It is important to note that the sequence for ORF50P3 is the same for the probe used in earlier study [6] . Band shifts were not observed when ORF50P probes were incubated with IVT-synthesized KSHV glycoprotein L (gL) (data not shown). Additionally, experiments using the ORF50PNP probe (does not contain the GC-rich binding domain) did not form a complex with Egr-1 protein, thus confirming the specificity of Egr-1 binding (Fig. 1A, lane 18) . Finally, competition experiments using unlabeled ORF50P probes reduced band shifts by preventing Egr-1 binding to DIG-labeled probes (data not shown). A second set of EMSA studies were conducted using ORF50P probes carrying mutations (ORF50P3m and ORF50P8m; Table 1 ) in the putative Egr-1 binding region to further confirm the binding ability of Egr-1 proteins. Briefly, ORF50P3 and ORF50P8 were mutated to carry the 5 bp ATATA sequence in the GC-rich binding domain and then incubated in binding buffer alone or with IVT-synthesized Egr-1. Samples consisting of wildtype (wt) or mutated probes alone did not display a shift in the respective ORF50P3 and ORF50P8 probes ( Fig. 1B ; lanes 1, 3, 5. and 7). Following incubation of wt probes with Egr-1, complexes were formed producing separate band shifts (Fig. 1B , lane 2 and 6). Interestingly, Egr-1 did not bind ORF50P3m or ORF50P8m probes that displayed mutations in the Egr-1 binding domain, thus confirming the necessity for the consensus GC-rich binding domain mediating Egr-1/ORF50P interactions (Fig. 1B, lanes 4 and 8) . Finally, gel shift assays were performed using the nuclear extract from KSHV-infected cells to verify the ability of Egr-1 to bind ORF50P3 and ORF50P8. As expected, there was no hindrance in the migration of ORF50P probes without the addition of cell lysate ( Fig. 1C; lanes 1 and 5) . However, the presence of the lysate in the samples resulted in the formation of protein:DNA complexes indicated by a band shift (Fig. 1C , lanes 2 and 6). Egr-1 binding to ORF50P was confirmed by incubating lysates with specific Abs and performing a supershift. Samples that were incubated with nonspecific IgG Abs displayed band shifts that were similar to samples containing only Egr-1 and the respective probes (Fig. 1C, lanes 3 and 7) . Alternatively, a supershift occurred exhibiting a discrete band when nuclear lysates were pre-treated with Egr-1 specific Abs (Fig. 1C, lanes 4 and 8). Taken together, these experiments provide support for the ability of Egr-1 to specifically bind to two separate locations on KSHV ORF50P. IVT-synthesized Egr-1 binds to ORF50P probes. EMSA studies were performed using IVT-synthesized Egr-1 products and DIGlabeled ORF50P probes (see Table 1 ). (B) Mutations in the putative Egr-1 binding domain inhibit Egr-1 binding. EMSA experiments were performed using wt ORF50P3 and ORF50P8 probes as well as corresponding probes displaying mutations in the suspected Egr-1 binding domain (ORF50P3m and ORF50P8m). (C) Nuclear lysates from KSHV-infected cells formed a complex with ORF50P probes. BCBL-1 cells were synchronized in S phase of cell cycle according to earlier protocols [28] , treated with 20 ng/ml TPA for 8 h, and lysed. Nuclear extracts containing Egr-1 proteins were used to perform EMSA studies using ORF50P3 and ORF50P8 probes. Specific Egr-1 binding was confirmed by performing supershifts using specific antibodies to Egr-1 (lanes 4 and 8) or nonspecific IgGs (lanes 3 and 7). The arrowhead indicates protein/ DNA complex formation. Specific antibody/protein/DNA supershifts are denoted by the asterisk. doi:10.1371/journal.pone.0033364.g001 A semi-quantitative chromatin immunoprecipitation (ChIP) assay was performed to analyze Egr-1 binding to ORF50P in a chromatin context (in vivo) using specific antibodies. TPA-induced KSHV-infected cells were used to assess the binding ability of Egr-1 to ORF50P via ChIP assays. The presence of specific ORF50P in the IP samples was analyzed by semiquantitative PCR using specific primers covering the regions of ORF50P3 or ORF50P8. As expected, when Egr-1 was expressed in BCBL-1 cells it was recruited to the promoter of KSHV ORF50 and specifically targeted both ORF50P3 and ORF50P8 ( Fig. 2A, cycle 30 ). Recruitment of Egr-1 to the nonspecific ORF50NP region was not detectable in our experiments (data not shown). For negative controls, samples were IP with nonspecific (NS) IgG Abs and recruitment of Egr-1 to ORF50P was not observed ( Fig. 2A , cycle 30 on control gels). However, positive controls using specific Abs to histone proteins recovered ORF50P targets ( Fig. 2A) . These results help us confirm that Egr-1 binds to two separate domains on ORF50P, in vivo. To establish a critical role for these interactions between Egr-1 and ORF50P, luciferase reporter constructs were used to investigate the necessity of ORF50P3 and ORF50P8 during Egr-1-mediated activation of the ORF50P. Empty vector (pGL3) or vectors encoding a deletion series of ORF50P (Fig. 2B ) along with the downsteam luciferase gene were transiently transfected into target cells in conjunction with empty vector or egr-1/ pcDNA3.1(+). Cells transfected with pcDNA3.1(+) did not induce significant luciferase activity (Fig. 2C) . However, following incubation of cells transfected with egr-1/pcDNA3.1(+), we noticed the luciferase activity to be significantly greater in cells that were also transfected with constructs encoding the full length (FL) ORF50P compared to cells transfected with pGL3 ( Fig. 2C) , Furthermore, we observed a decrease in relative luciferase activity following deletion of the fragment containing the ORF50P3 domain (D-2922 to -2044; D-2922 to -1322; D-2922 to -894; and D-2922 to -169) when compared to the construct encoding full length ORF50P (Fig. 2C) . Although the absence of ORF50P3 contributed to a decrease in luciferase activity, this activity was never completely abolished ( Fig. 2C) suggesting the need for an intact ORF50P3 and ORF50P8 for an optimal Egr-1-induced transcription of ORF50. Taken together, these results suggest a role for Egr-1 to specifically bind and activate ORF50P to trigger a lytic infection in KSHV-infected cells. Cellular Egr-1 and virus-encoded KSHV RTA follow a similar expression pattern during de novo KSHV infection Several different viruses are known to activate Egr-1 expression upon infection [17, 18, 19, 20] . Since BCBL-1 cells already carry KSHV DNA, KSHV-infected HEK293 cells were used to evaluate the expression pattern of Egr-1 and KSHV RTA during early stages of de novo infection. Expression of Egr-1 and RTA proteins were significantly elevated by 1 hour post infection (hPI) and continued to maintain increased expression until roughly 6-8 hPI (Fig. 3A, lanes 2-4) . In contrast, a considerable decrease in the expression of these proteins was observed from 12-48 hPI (Fig. 3A , lanes 5-7). A significant difference in expression of total bactin was not observed (Fig. 3A ) during the course of KSHV infection demonstrating the specificity of the results on Egr-1 and RTA expression. To further support our findings, mRNA extracted from KSHVinfected HEK293 cells were subjected to quantitative real-time PCR (qRT-PCR) analysis in order to evaluate egr-1 and ORF50 transcriptional activity. Uninfected cells (0 hPI) did not show detectable ORF50 expression (Fig. 3B ). On the other hand, a low baseline level of egr-1 expression was observed in the uninfected samples ( Fig. 3B ). With the onset of a primary infection, expression levels of both egr-1 and ORF50 increased up to 6hPI (Fig. 3B ). These elevated levels of egr-1 and ORF50 decreased substantially by 12-24 hPI (Fig. 3B ). No significant changes in the expression of the internal control gene encoding M6PR was observed indicating specificity of the results (data not shown). Taken together, the expression profiles of Egr-1 and RTA seem to follow an identical pattern during primary infection of cells. Elevated Egr-1 expression activates lytic genes in KSHVinfected cells BCBL-1 cells have turned out to be a blessing in disguise for this study as they harbor KSHV DNA in a predominantly latent state [21] . BCBL-1 cells were transiently transfected using egr-1/ pCDNA3.1(+) for 24, 48, and 72 h ( Next, qRT-PCR studies were performed to identify changes in egr-1 and virus-encoded ORF50 gene expression. We did not observe any noticeable alterations in egr-1 and ORF50 transcription in target cells that were untransfected (UT), mock transfected, or transfected with empty vectors (Fig. 4B ). As expected, levels of egr-1 were significantly increased in cells transfected with egr-1/pCDNA3.1(+) over controls (Fig. 4B) . Furthermore, elevated egr-1 expression coincided with an increase in KSHV ORF50 transcription (Fig. 4B ). Peak expression for both genes was observed by 48 h post transfection (Fig. 4B) . Incidentally, we also observed an elevated expression of ORF8 (Fig. 4B ), encoding the late structural virus protein termed as gB, in response to an enhanced egr-1 and ORF50 expression. All these changes implicate Egr-1 expression to significantly induce virus reactivation. Immunofluorescence assay (IFA) was conducted to determine a possible role for elevated Egr-1 on the expression of KSHVencoded lytic proteins in the above transfected cells (Fig. 4A, B ). KSHV-encoded ORF59, a processivity factor for KSHV DNA polymerase, is expressed in the nucleus of infected cells during early stages of virus reactivation [22] . Target cells transfected with empty vectors showed low levels of ORF59 expression (Fig. 4C) . However, transfection cells with egr-1/pcDNA3.1(+) augmented the level of ORF59 protein expression in KSHV-infected cells (Fig. 4C ); clearly implicating a critical role for Egr-1 in inducing KSHV reactivation. Finally, we analyzed MAPK signaling in the above cells relative to Egr-1 expression levels. Our results suggest that cells transfected with egr-1/pCDNA3.1(+) displayed elevated levels of Egr-1 (Fig. 4D, lane 7) . Transfection of cells with pCDNA3.1(+) (Fig. 4D , lane 5) or mock transfection (Fig. 4D, lane 3) of cells did not significantly alter the expression levels of Egr-1 and phosphorylation state of ERK1/2 compared to untransfected cells (Fig. 4D, lane 1) . Interestingly, treatment of cells with a known inhibitor of MEK1/2 (10 mM of U0126) significantly lowered both the expression levels of Egr-1 and ERK1/2 activity (Fig. 4D, lane 8) . We observed U0126 to dose dependently inhibit ERK1/2 activity and Egr-1 levels in the above cells confirming the specificity of U0126 in targeting MAPK.Egr-1 signaling. The inhibition of ERK1/2 activity and Egr-1 levels was greatest following treatment of infected cells with 10 mM of U0126 (Fig. 4E, lane 4) . These results suggest Egr-1 to be downstream of the MAPK signaling cascade. [28] and treated with 20 ng/ml of TPA for 8 h. ChIP assays were performed using 2 mg of specific antibodies to Egr-1 or nonspecific IgGs. Primers specifically targeting ORF50P3 or ORF50P8 (see Table 1 ) were used to perform semi-quantitative PCR experiments on 1% of total DNA (input) and IP samples. Respective cDNA at 10, 15, 20, 25, and 30 cycles were removed and resolved on a 2% agarose gel. IP of BCBL-1 DNA using specific antibodies to histone H3 was used as positive controls. (B) A schematic representation of ORF50P used to make the deletions of the luciferase reporter constructs. The nucleotide locations correspond to the old KSHV genome sequence NC_003409 which has since been updated to NC_009333.1. Asterisks refer to the ORF50P3 and ORF50P8 locations, respectively. (C) Overexpression of Egr-1 activates ORF50P via interacting with ORF50P3 and ORF50P8 domains. HEK293 cells were co-transfected with a combination of three vectors, one from the following groups: (i) pcDNA3.1(+) or egr-1/pcDNA3.1(+), (ii) the control vector, pRL-TK, and (iii) empty pGL3 vectors or pGL3 vectors encoding FL ORF50P or one of several deletions (D-2922 to -2044; D-2922 to -1322; D-2922 to -894; and D-2922 to -169). After 48 h post-transfection, the cells were lysed, and relative luciferase activity was monitored. Firefly luciferase was normalized to the corresponding Renilla luciferase activity. The luciferase activation of pGL3 by egr-1/pcDNA3.1(+) was represented as 1-fold. Each point denotes the average 6 SD of three experiments. Columns with different alphabets are statistically significant (P,0.05) by least significant difference (LSD). doi:10.1371/journal.pone.0033364.g002 Resveratrol inhibits Egr-1 and ORF50 during early and late stages of infection Due to the vital role of MAPK signaling on Egr-1 expression and KSHV reactivation, the effect of resveratrol on KSHV replication was analyzed. We chose to use resveratrol because: (i) it is a naturally occurring product; and (ii) it is a known regulator of MAPK signaling and Egr-1 expression [14] . Furthermore, resveratrol inhibits ERK1/2 activity in virus-infected cells [23] . It is important to understand that even though KSHV-encoded ORF50 is a gene crucial for reactivation, it is also expressed during early stages of KSHV infection and may play a role in the establishment of virus latency [6, 24] . Therefore, it was necessary to determine the expression pattern of cellular egr-1 and virusencoded ORF50 during both early stages of infection as well as during virus reactivation (late stages). In this study, effect of resveratrol on early stages of infection was analyzed in HEK293 cells while its effect on late stages (reactivation using TPA) was analyzed predominantly in BCBl-1 cells, for convenience. In this study, resveratrol was able to inhibit egr-1 expression in a dose dependent manner during early stages of de novo KSHV infection of HEK293 cells (Fig. 5A) . The doses tested in this study were confirmed by trypan blue test to be non-toxic to cells (data not shown). The resveratrol doses used by us are also those that have been published previously [25, 26, 27] . Resveratrol (100 mM) was able to suppress the expression of phospho-ERK1/2 and Egr-1 proteins during de novo infection of cells (Fig. 5B, lanes 3, 6, and 9 ). Additionally, resveratrol was able to inhibit the expression Egr-1 and phophorylated ERK1/2 in mock-infected cells ( Figure S1, lanes 2, 4, and 6 ). On the other hand, it was not able to significantly alter the expression of endogenous ERK1/2 and actin controls (Fig. 5B, lanes 3, 6, and 9 ; Figure S1, lanes 2, 4, and 6) . DMSO, the vehicle for resveratrol, did not significantly alter the phosphorylation of ERK1/2 and the expression of Egr-1 (data not shown). In order to present more physiologically relevant studies, an over-the-counter resveratrol dietary supplement (RDS) was used to treat KSHV-infected BCBL-1 cells under TPA-induced conditions. RDS containing 100 mM resveratrol did not significantly induce cell death as monitored by the lactate dehydrogenase assay (data not shown). These results were confirmed by the conventional trypan blue test. More than 95% of the target cells were found to be viable when the target cells were treated with RDS (data not shown). As shown in earlier reports [12] , TPA treatment augments phospho-ERK1/2 expression (Fig. 6A, lane 2) . The effect of TPA also resulted in an increase in Egr-1 and KSHV RTA expression when compared to uninduced cells (Fig. 6A, lane 2) . Unfiltered RDS successfully inhibited phospho-ERK1/2, Egr-1, and RTA expression in TPA-induced KSHV-infected cells (Fig. 6A, lane 3) . A slightly lesser inhibitory effect was observed in cells that were treated with RDS that had particulates removed using a 0.2 mm filter (Fig. 6A, lane 4) . We did not discover either form of RDS to have noticeable effect on endogenous ERK1/2 and actin controls (Fig. 6A, lanes 3 and 4) . The data from Western blotting (Fig. 6A ) was further confirmed in HEK293 cells by IFA (Fig. 6B) . To authenticate the results from monitoring the effect of RDS on protein levels of Egr-1 and RTA, we analyzed the effect of RDS on (i) uninduced cells transfected with vectors encoding egr-1, and (ii) TPA-induced KSHV reactivation in BCBL-1 cells by performing qRT-PCR. Our results clearly demonstrate the ability of RDS to lower the expression of both egr-1 and ORF50 under both circumstances (Fig. 6C, D) . Finally, ChIP assay was performed to discern the specificity of RDS on virus reactivation using primers specific to ORF50P8 region. Under TPA-induced conditions, Egr-1 specifically targeted ORF50P8 (Fig. 6E . cycles 25 and 30). However, under RDS treated conditions the binding of Egr-1 to ORF50P8 was significantly decreased (Fig. 6E, cycles 25 and 30 ). For negative controls, samples were IP with (NS) IgGs and recruitment of Egr-1 to ORF50P was not observed (Fig. 6E) . However, positive controls using specific Abs to histone proteins recovered ORF50P targets (Fig. 6E) . These results suggest resveratrol in its chemical form and RDS may lower Egr-1 expression to inhibit KSHV reactivation from latency. Egr-1 regulates expression of several viral genes and plays a crucial role in the replication of different viruses Our results from employing the EMSA and ChIP assay ( Fig. 2A) demonstrate that Egr-1 may bind ORF50P via at least two different GC-rich binding domains: at positions between 2 100 -2 76 bp (ORF50P8) and 2 2173 -2 2149 bp (ORF50P3). The results from employing the ChIP assay ( Fig. 2A) demonstrate that Egr-1 may bind ORF50P with a greater affinity at positions between 2 100 -2 76 bp (ORF50P8) compared to 2 2173 -2 2149 bp (ORF50P3). However, we did not observe any such differences in the binding affinity of Egr-1 to ORF50P3 and ORF50P8 by EMSA using IVT Egr-1 (Fig. 1B, C) . All the more, our data supports the need for Egr-1 to bind both ORF50P3 and ORF50P8 for an optimal transcriptional activity in luciferase reporter assays (Fig. 2C) . This difference observed in Egr-1 binding to both these domains could be due to one or both the reasons: (i) IVT synthesized Egr-1 was used in EMSA experiments; and (ii) the design of ChIP assay conducted in this study was not to decipher the relative binding affinity of Egr-1 to these domains; instead was performed to just confirm if Egr-1 bound these domains, in vivo. Although we previously noticed that the egr-1 and KSHVencoded ORF50 followed a similar expression pattern, the experiments were conducted in TPA-induced cells to evaluate their expression during the reactivation process [6] . The present study discovered that the transcriptional activity of egr-1 and ORF50 and their subsequent translation is comparable and followed a similar pattern during de novo KSHV infection (Fig. 3) . However, enhanced cellular Egr-1 and viral RTA expression during early stages of primary infection (Fig. 3) is not sufficient to trigger a lytic infection [6] . These results suggest the following: (i) the role of Egr-1.RTA signaling in initiating a lytic cycle of infection during the course of initial infection of cells is limited; and (ii) there is a missing element in the Egr-1.RTA driven cellular events critical for inducing a productive replication. In this study, transfection of cells with egr-1/pCDNA3.1(+) resulted in a significant increase in virus-encoded ORF50 transcription followed by the expression of early-lytic ORF59 protein and late-lytic gene (ORF8) encoding gB (Fig. 4B, C) ; all of which are indicators of an active lytic replication of KSHV [12, 28, 29] . MAPK signaling was observed to regulate Egr-1 expression in cells transfected with egr-1/pCDNA3.1(+) (Fig. 4D) . Interestingly, treatment of KSHV-infected cells with TPA induces a lytic replication via MAPK signaling [12, 13] . In addition, Egr-1 is a downstream target of Raf/MEK/ERK signaling (Fig. 4D ) [6, 30] . It is unclear at this time if the effect of egr-1/pcDNA3.1 over-expression resembles the milieu supporting a lytic infection in vivo. It is important to remember that KSHV reactivation can be regulated by other cellular factors including STAT6, NFkB, and XBP-1 [31, 32, 33] . Thus, activation of a lytic infection may require unique cellular factors under specific conditions or a combination of factors. Further investigation is required to unravel the environment(s) supporting virus reactivation under physiologically relevant conditions; especially in terms of the different transcription factors. In order to support these findings, more physiological relevant studies were performed by analyzing the effect of resveratrol on KSHV-infected cells. Resveratrol, or trans-3,5,49-trihydroxystilbene, is a phytoalexin that is produced in various plants such as grapes, berries, and peanuts in response to attacks by pathogens [14] . Several reports provide evidence for resveratrol to exhibit antiviral effects [27, 34, 35] . On the other hand, resveratrol has also been shown to induce virus replication [36, 37] . We have demonstrated that resveratrol, in its chemical form, inhibits 1-3) were incubated at 37uC for 48 h while the cells transfected with egr-1/pcDNA3.1 were incubated for 24, 48, and 72 h, respectively. At the end of incubation, the cells were lysed and the lysates were used to perform Western blotting. (B) Effect of elevated Egr-1 on KSHV ORF50 and ORF8 expression. BCBL-1 cells were untransfected or transfected as described above. RNA was extracted, cDNA was synthesized, and the expression of cellular egr-1; and KSHV encoded ORF50 and ORF8 were analyzed by qRT-PCR. Baseline expression of genes was considered as 1-fold for comparisons. Each point denotes the average6S.D. (error bars) of three experiments. (C) Expression of lytic proteins in BCBL-1 cells transfected with Egr-1. KSHV-infected cells were untransfected, transfected with pcDNA3.1, or egr-1/pcDNA3.1 for 48 h. The stained cells examined using a confocal microscope (magnification 662). The average number of fluorescent cells were counted over five random fields and used for comparisons. (D) Enhanced egr-1 activates MAPK signaling in BCBL-1 cells. KSHV-infected cells were untransfected, mock-transfected, transfected with pcDNA3.1(+), or egr-1/pcDNA3.1(+) for 48 h. Each group of cells was left untreated or they were treated with 10 mM of U0126 1 h prior to transfection and remained throughout the incubation period. Cell lysates were resolved on a 10% SDS-PAGE, transferred to a PVDF membrane, and Western blotting was performed using specific antibodies. (E) U0126 inhibits phosho-ERK1/2 and Egr-1. Briefly, BCBL-1 cells were treated with different concentrations of U0126. Following 24 h incubation, the cells were lysed and the lysates were used to perform Western blotting as per earlier protocols using specific antibodies. doi:10.1371/journal.pone.0033364.g004 Egr-1 and phospho-ERK1/2 in KSHV-infected cells (Fig. 5B) . RDS significantly inhibited KSHV reactivation in uninduced and TPA-induced cells (Fig. 6A , B, C, D). While performing these experiments we noticed that unfiltered RDS was able to inhibit KSHV reactivation to a greater extent when compared to cells treated with RDS that was passed through a 0.2 mm filter. However, both treatments were able to significantly inhibit gene products associated with KSHV reactivation (Fig. 6A, D) . These differences are likely due to the presence of unknown factors in unfiltered RDS that may act in combination with resveratrol. Incidentally, the decrease in the expression of RTA by RDS coincided with a sharp decline in the ability of Egr-1 to bind ORF50P as shown by the semi-quantitative ChIP assay (Fig. 6E) . Taken together, this is the first report to describe the role of physiologically relevant RDS on KSHV infection. Several cellular pathways are regulated by resveratrol including apoptotic, NFkB, and all forms of MAPK signaling [14, 38] . We found resveratrol to inhibit expression of Egr-1 and phosphorylation of ERK1/2 resulting in suppression of KSHV reactivation ( Fig. 5 and 6 ). Very recent studies have established the fact that resveratrol significantly lowers phosphorylation of ERK1/2 (directly upstream of Egr-1) in target cells, in vivo and in vitro [25, 26, 39, 40] . At this point in time, we are certain about the ability of RDS to block TPA-induced virus reactivation. However, further research is required to confirm if this ability of RDS to promote viral latency is by its direct inhibitory effect on the expression Egr-1 or the upstream MAPK signaling component(s), namely ERK1/2 activity. KSHV reactivation from latency is a complicated process which is regulated by an intricate relationship between viral and cellular factors. The method in which resveratrol may regulate KSHV reactivation has yet to be fully understood. However, we propose that resveratrol may inhibit KSHV reactivation by altering the interactions between cellular Egr-1 and viral ORF50P in a Raf.MEK.ERK-dependent manner. All three MAPKs (ERK, p38, and JNK) have been shown to positively regulate Egr-1 expression [41] . However, the role for active p38 MAPK is not fully understood as it has been shown to reduce Egr-1 expression in B-lymphocytes unlike ERK and JNK [42] . Further studies are required to better understand the involvement of different MAPKs on Egr-1 expression during KSHV infection. The findings presented in this study open a Pandora's Box of questions pertaining to treating/managing a variety of viral infections using RDS. Future studies are aimed at appreciating the cellular milieu critical for the effectiveness of the MAPK associated signaling in inducing virus reactivation. These findings may provide for more useful applications to combat a variety of viral lytic infections. Cells HEK293 cells and BCBL-1 cells were cultured in DMEM and RPMI (Invitrogen, Carlsbad, CA), respectively, as per earlier studies [6, 43] . Rabbit antibodies to gB [44] , RTA [6] ; and mouse antibodies to ORF59 were used in this study. Rabbit antibodies to phospho-ERK1/2, total ERK1/2, actin, and Egr-1 (15F7; monoclonal antibodies) purchased from Cell Signaling Technology, Beverly, MA were used in this study. Mouse (S-25) and rabbit (15F7) antibodies to Egr-1 purchased from Santa Cruz biotechnologies, Inc. (Santa Cruz, CA) were used in Immunofluorescence assay (IFA) and Western blotting experiments, respectively. Additionally, rabbit polyclonal antibodies (588) to Egr-1 was used in gel shift and chromatin immunoprecipitation assays (Santa Cruz Biotechnology, Santa Cruz, CA). We used egr-1/pCDNA3.1(+) and gL/pCDNA3.1(+) vectors in this study. Both these vectors have been described elsewhere [6] . IVT of egr-1/pCDNA3.1(+) and gL/pCDNA3.1(+) was conducted as per earlier studies [45] using the TNT-coupled rabbit reticulocyte lysate system (Promega). HEK293 cells were infected as per earlier procedures (25) . Figure 6 . RDS reduces the Egr-1/ORF50 association in vivo. (A) RDS lowers Egr-1 and KSHV RTA expression. KSHV-infected BCBL-1 cells were synchronized in S phase and untreated or treated using 20 ng/ml of TPA for 2 h. Each group of cells was left untreated or was further treated using filtered or unfiltered RDS containing 100 mM of resveratrol. The cells were incubated at 37uC for 6 h and lysed. The lysates were resolved on a 10% SDS-PAGE, transferred to a PVDF membrane, and Western blotting was performed using specific antibodies. (B) RDS reduces the number of KSHVinfected cells undergoing reactivation in HEK293 cells. Mock-infected, KSHV-infected, and KSHV-infected cells in the presence of RDS containing 100 mM resveratrol were stained using monoclonal mouse anti-Egr-1 IgGs and rabbit peptide antibodies targeting KSHV RTA and examined under a fluorescent microscope (magnification 6100). (C) Overexpression of Egr-1 is unable to overcome RDS-mediated inhibition of virus reactivation. BCBL-1 cells were transiently transfected using pcDNA3.1(+) or egr-1/pcDNA3.1(+) and subsequently treated with unfiltered RDS containing 100 mM of Resveratrol for 6 h. RNA was extracted, cDNA was synthesized, and egr-1 and KSHV ORF50 were analyzed by qRT-PCR. Each point denotes the average6S.D. (error bars) of three experiments. (D) RDS lowers egr-1 and KSHV ORF50 transcriptional activity. BCBL-1 cells were synchronized in S phase, treated with 20 ng/ml of TPA, and treated using filtered or unfiltered RDS containing 100 mM of resveratrol as before. RNA was extracted, cDNA was synthesized, and egr-1 and KSHV ORF50 were analyzed by qRT-PCR. Each point denotes the average6S.D. (error bars) of three experiments. Columns with different alphabets are statistically significant (P,0.05) by LSD. (E) RDS inhibits the ability of Egr-1 to bind KSHV ORF50 in vivo. BCBL-1 cells were synchronized and treated with TPA as before. The cells were further treated using unfiltered RDS containing 100 mM of resveratrol and incubated for 6 h. ChIP assays were performed using 2 mg of specific Egr-1 Abs. Semi-quantitative PCR experiments were performed using samples from 1% input DNA and IP samples in order to determine the expression of ORF50P8. Respective cDNA at 10, 15, 20, 25, and 30 cycles were resolved on a 2% agarose gel. Specific antibodies to histone H3 and nonspecific IgGs were used to IP sample chromatin and served as positive and negative controls, respectively. doi:10.1371/journal.pone.0033364.g006 In this study, we synchronized BCBL-1 cells in S phase of cell cycle as per earlier protocols [46] . Equal amounts (20 mg) of protein was used in Western blotting experiments as per earlier studies [47] . The qRT-PCR was performed using the synthesized cDNA in a 25 ml reaction volume to analyze the expression of ORF50, egr-1, and M6PR as per earlier protocols [47] . Target cells were transfected with pCDNA3.1(+), egr-1/ pCDNA3.1(+), gL/pCDNA3.1(+) using GeneJammer transfection reagent (Stratagene, La Jolla, CA) as per earlier studies [12] . Target cells were fixed for 10 min in ice cold acetone and washed thrice in phosphate buffered saline (PBS). These cells were sequentially stained with mouse anti-ORF59 antibodies and goat anti-mouse FITC as per earlier studies [12] . The stained cells were further incubated for 20 min on ice with 5 mM SYTO Red (a nuclear stain; Invitrogen) before being analyzed with a laserscanning LSM 510 Carl Zeiss confocal microscope. In another set of experiments, acetone fixed cells were incubated with a combination of mouse anti-Egr-1 IgGs and rabbit anti-RTA for 45 min at room temperature (RT), and incubated with a combination of goat anti-mouse FITC and goat anti-rabbit TRITC) for 30 min at RT. Stained cells were washed in PBS, mounted by using an anti-fade reagent containing DAPI (4,6diamidino-2-phenylindole; Molecular Probes) and examined under a Nikon fluorescent microscope with appropriate filters. IVT products of Egr-1 or KSHV gL were evaluated by EMSA for DNA binding using several 25 bp digoxygenin (DIG)-labeled probes containing sequences from the ORF50P (Table 1) as per earlier studies [6] . For a supershift, the cellular lysate was incubated with rabbit monoclonal antibodies to Egr-1 or nonspecific IgG at 37uC for 30 min prior to the addition of the DIG-labeled probe. All samples were run on a 4% nondenaturing gel for approximately 1.5 h and transferred to a PVDF membrane. The protein:DNA interaction was detected using the CSPD detection system (Roche Applied Science). Target cells were transiently co-transfected using appropriate pGL3 and internal control pRL-TK contructs (Promega) and pcDNA3.1(+) vectors (Invitrogen). The total amount of DNA used per sample was approximately 2 mg. Following 48 h post-transfection, cells were harvested and Firefly and Renilla luciferase activities were analyzed using the dual luciferase system (Promega). Luciferase activity was monitored using a Turner Systems Luminometer (Sunnyvale, CA) as per earlier protocols [6] . The relative luciferase activity was calculated by normalizing ORF50P luciferase activity to control Renilla luciferase activity. The results were plotted as a percentage of the activity of the empty pGL3 vector. BCBL-1 cells were treated with a final concentration of 1% formaldehyde and crosslinked for 10 min at RT. Crosslinking was stopped by addition of glycine at a final concentration of 0.125 M for 5 min. The cross-linked cells were washed in 16 PBS and counted so that approximately 10 7 cells were used in each immunoprecipitation (IP) reactions. Nuclei from the cells were purified and lysed to collect chromatin. Chromatin was sheared to approximately 500 bp using a Bioruptor sonicator (Diagenode, Sparta, NJ). Lysates containing the chromatin were pre-cleared using 35 ml of Protein A sepharose beads (Amersham Biosciences) in pre-IP dilution buffer for 30 min at 4uC. The samples were centrifuged to remove beads and the lysate was recovered. After setting aside input controls, primary antibodies were added to the samples and incubated overnight at 4uC. The DNA/protein complexes were IP using protein A beads for 4 h at 4uC and then washed using various ChIP wash buffers. Following elution, proteinase K was added to the complexes and incubated at 65uC overnight in order to reverse the crosslinks. The DNA samples were purified using phenol/chloroform extraction, resuspended in 16 TE buffer, and finally analyzed by polymerase chain reaction (PCR). PCR was performed using platinum pfx polymerase (Invitrogen, Carlsbad, CA) as per standard protocols. PCR amplification of the precipitated DNA was performed using primers that targeted ORF50P3, ORF50P8, and ORF50PNP regions ( Table 1) Figure S1 Resveratrol inhibits Egr-1 expression in the absence of KSHV infection. HEK293 cells were mockinfected by incubating with growth medium for 2 h at 37uC. These cells were washed and cultured in growth medium in the presence or absence of 100 mM of resveratrol for 48 h. The cells were lysed using gold lysis buffer (GLB) and the lysates were resolved on a 10% SDS-PAGE, transferred to a PVDF membrane, and Western blotting was performed using specific antibodies. (TIF)

Common Variants in CDKN2B-AS1 Associated with Optic-Nerve Vulnerability of Glaucoma Identified by Genome-Wide Association Studies in Japanese

BACKGROUND: To date, only a small portion of the genetic variation for primary open-angle glaucoma (POAG), the major type of glaucoma, has been elucidated. METHODS AND PRINCIPAL FINDINGS: We examined our two data sets of the genome-wide association studies (GWAS) derived from a total of 2,219 Japanese subjects. First, we performed a GWAS by analyzing 653,519 autosomal common single-nucleotide polymorphisms (SNPs) in 833 POAG patients and 686 controls. As a result, five variants that passed the Bonferroni correction were identified in CDKN2B-AS1 on chromosome 9p21.3, which was already reported to be a significant locus in the Caucasian population. Moreover, we combined the data set with our previous GWAS data set derived from 411 POAG patients and 289 controls by the Mantel-Haenszel test, and all of the combined variants showed stronger association with POAG (P<5.8×10(−10)). We then subdivided the case groups into two subtypes based on the value of intraocular pressure (IOP)—POAG with high IOP (high pressure glaucoma, HPG) and that with normal IOP (normal pressure glaucoma, NPG)—and performed the GWAS using the two data sets, as the prevalence of NPG in Japanese is much higher than in Caucasians. The results suggested that the variants from the same CDKN2B-AS1 locus were likely to be significant for NPG patients. CONCLUSIONS AND SIGNIFICANCE: In this study, we successfully identified POAG-associated variants in the CDKN2B-AS1 locus using a Japanese population, i.e., variants originally reported as being associated with the Caucasian population. Although we cannot rule out that the significance could be due to the differences in sample size between HPG and NPG, the variants could be associated specifically with the vulnerability of the optic nerve to IOP, which is useful for investigating the etiology of glaucoma.

Glaucoma is a neurodegenerative ocular disease and one of the leading causes of irreversible blindness worldwide [1] . It is characterized by the progressive loss of retinal ganglion cells and optic nerve axons, resulting in visual field defects [2] . One of the well-known major risk factors for glaucoma is elevated intraocular pressure (IOP) [2] . Thus, the measurement of IOP is routinely involved in the diagnosis of glaucoma. In fact, the IOP level has been applied to subdivide the most common form of glaucoma, primary open-angle glaucoma (POAG), into two subtypes [3] : POAG with high ($22 mmHg) IOP (POAG/HPG, high pressure glaucoma; hereafter referred to as ''HPG'') and with normal (,22 mmHg) IOP (POAG/NPG, normal pressure glaucoma; hereafter referred to as ''NPG''). Interestingly, ,92% of the Japanese POAG patients are categorized into the NPG subtype [4] , whereas ,41% of Caucasian POAG patients are categorized as NPG [5] , thus showing a unique epidemiological distribution of Japanese patients compared with other ethnic groups. Aside from IOP measurements, the diagnosis of glaucoma is commonly made by observing optic nerve degeneration and visual field defects by means of fundus examinations and visual field tests, respectively. In the case of HPG, early drug treatment to lower IOP immediately following the onset of visual field damage has been shown to be quite effective in slowing the irreversible progression toward blindness [6, 7] . In contrast, since NPG patients show normal IOP, they are often misdiagnosed. Therefore, both fundus examinations of the optic nerve and visual field tests are critical for the proper diagnosis of NPG patients. However, due to the restriction of the healthcare expenditure to include those examinations into a person's regular medical checkup, especially in the preclinical state of glaucoma, it would be of great benefit if the risk of developing glaucoma could be ascertained based on a simple blood test to assess the genetic markers for the disease. Since glaucoma shows familial aggregation and its prevalence varies between individuals of different ethnicities, it has been theorized that genetic factors play a significant role in the pathogenesis of glaucoma [8, 9] . Therefore, several institutions are making profound efforts to discover single-nucleotide variants for glaucoma by conducting a genome-wide association study (GWAS) [10] [11] [12] [13] [14] , and there are a few published reports of the association of particular loci, such as the CAV1/CAV2 locus (7q31.1) [13] and the TMCO1 or CDKN2B-AS1 loci (1q24.1 or 9p21.3, respectively) [14] , with POAG using Caucasian subjects. However, it appears that a controversy still exists as to determining the authentic variants associating with POAG, even within the same ethnicity of European descent [14, 15] . We previously reported a GWAS and the subsequent follow-up study focused on the high-ranked variants identified in the initial population using in total of 1,575 Japanese subjects [10] . We identified six variants located in three loci, 1q43, 10p12.31, and 12q21.31, on the chromosome which were modestly associated with POAG, although the association results were not reproducible with a different population of different ethnicities [13, 16, 17] . In that previous study, we combined the patients from both HPG and NPG subtypes as a single case group in order to increase the statistical power [10] . Therefore, the genetic loci that were identified are most likely to be components of the molecular mechanism shared by both subtypes. Another Japanese group performed a GWAS [12] and discovered variants in SRBD1 on 2p21 and ELOVL5 on 6p12.1 that were associated with NPG (in that study, NPG was referred to as ''normal tension glaucoma'') [18] . However, since the findings in that study were derived from a single population, replication studies to support those findings still need to be conducted. In this present study, we examined two independent GWAS data sets, and then performed a meta-analysis by combining the data sets in order to discover authentic genetic markers for POAG, HPG, or NPG. We succeeded in identifying some significant variants in the CDKN2B-AS1 locus using a Japanese population, as that locus has also been shown to be significantly associated with POAG in Caucasians [14] . Moreover, the variants that passed the Bonferroni correction seemed to be associated with POAG and POAG/NPG, thus suggesting that the locus potentially affects the NPG, rather than the HPG, subtype of POAG. Although we still need to confirm our findings by use of a larger HPG cohort in order to increase the statistical power, the variants identified in this study could be associated specifically with the vulnerability of the optic nerve to IOP, thus enabling us to investigate the etiology of glaucoma. We first performed a GWAS in 833 POAG patients and 686 control subjects (hereafter referred to as ''Present GWAS''; Figure 1A ) who were selected and divided into case and control groups based on our strict diagnosis criteria. After genotyping, 653,519 SNPs passed the quality controls and were used for the association study ( Figure 1A) . In a quantile-quantile (Q-Q) plot ( Figure S1A ), the genomic inflation factor (l) showed 1.021, suggesting that the population substructure should not have any substantial effects on the association analysis. Under these conditions, we obtained 8 genome-wide significant SNPs (Table S1 ) that passed the Bonferroni correction threshold (0.05/ 653,519 = 7.65610 28 ), of which 5 SNPs were located in CDKN2B-AS1 on chromosome 9p21.3 ( Figure 2 ). A few SNPs that showed modest to strong association were distributed across a single linkage disequilibrium (LD) block located in the locus ( Figure S2 ). The remaining genome-wide significant SNPs identified from different chromosomes turned out to be genotyping errors due to the poor 2D clusters ( Figure S3B-D) . Next, we updated the case and control groups of the previous GWAS population [10] with the latest clinical information and ended up with 411 POAG patients and 289 controls (hereafter referred to as ''Previous GWAS''; Figure 1A ). One SNP (rs8181047) of 5 genome-wide significant SNPs in CDKN2B-AS1 identified from the Present GWAS was not able to be assessed in the Previous GWAS because a DNA probe of that particular SNP was not designed for the Affymetrix 500 K platform. Consequently, we combined the two sets of GWAS genotype data for the remaining 4 SNPs by use of the Mantel-Haenszel test [19] , and the level of significance of all 4 SNPs increased from P,5.2610 29 to P,5.8610 210 (Table 1 ). These 4 SNPs were highly correlated with one another (r 2 .0.98), and were considered to have a similar effect for POAG based on the conditional analysis. These results suggested that it was effective to combine the data since the population stratification between the two data sets seemed to be ignorable, which was also supported by the results of the heterogeneity test. Moreover, we assessed the confounding effects for these SNPs with respect to age and gender, and none were found to be statistically significant (Table S2) , thus suggesting that the obtained association results were specific to the case-control comparison. In addition, the two sets of GWAS data for 40 SNPs surrounding the significant 4 SNPs in the 9p21.3 locus were also combined and analyzed (Table S3, Table S4 ). Since these 44 SNPs, in total, were genotyped in both the Present and Previous GWAS, and all of the SNPs passed the genotyping quality controls, these SNPs should prove useful in providing a broader overview of the significance of the locus. The results showed that the significance of combined SNPs was generally high between 22.0-22.1 Mb ( Figure 3A ). Within the particular locus, there seemed to be two distinct LD blocks; one in the side including genome-wide significant SNPs (''LD Block 1'' in Figure 3D ) and one in the other side with modestly associated SNPs (''LD Block 2'' in Figure 3D ). For the subtype analyses ( Figure 1B) , the POAG patients from both the Present and Previous GWAS populations were divided into two subtypes based on the clinical record of IOP measurement: 1) POAG patients with HPG (IOP$22 mmHg) and 2) those with NPG (i.e., patients who consistently showed an IOP of less than 22 mmHg) [3] . The number of subjects in each subtype was found to be 330 HPG and 503 NPG patients and 215 HPG and 196 NPG patients in the Present and Previous GWAS, respectively ( Figure 1B) . When we performed the analysis separately using the Present GWAS data set, the Q-Q plots of HPG vs control and NPG vs control appeared to be quite different from one another ( Figure S1B , C). In particular, the HPG results indicated non-deviated Q-Q plots ( Figure S1B ). In fact, although none of the SNPs were genome-wide significant for HPG ( Figure S4A ), we obtained 4 genome-wide significant SNPs that passed the Bonferroni correction threshold for NPG (Table S1, Figure S4B ). As for NPG, the significance level of all 4 SNPs also increased when the two data sets were combined ( Figure 3C , Table S3 , Table S4 ), however, the significance was still far from the Bonferroni threshold of significance in relation to HPG ( Figure 3B , Table S3, Table S4) . Although the significance level of SNPs residing in LD Block 2 showed slight differences among the different subtypes, the level in LD Block 1 seemed to be determining the difference of significance between the two subtypes, thus suggesting that the variants on LD Block 1 are closely associated with the susceptibility to glaucoma in the 9p21.3 locus ( Figure 3D) . Surprisingly, all of the significant SNPs from NPG overlapped with those obtained from POAG (Table S1) . However, when we performed the heterogeneity test between the HPG and NPG groups, the results were not significant (P = 0.21-0.25, Table S3 ), suggesting that the significant differences between the two subtypes could be due to the small sample size of the HPG group. Although we need to confirm the above results using a larger HPG cohort, these results suggested that the significant SNPs identified in CDKN2B-AS1 on the 9p21.3 locus probably serve as genetic markers of glaucoma, which could be useful for investigating the etiology of glaucoma with respect to the vulnerability of the optic nerve to IOP. In this study, we successfully identified genetic markers in CDKN2B-AS1 strongly associated with POAG and POAG/NPG, but not with HPG, by analyzing two GWAS data sets using independent study populations totaling 2,219 Japanese subjects. For the case-control study of the Present GWAS, we excluded the samples without hesitation that didn't meet our strict quality control (see Supporting Information S1). Since we succeeded in excluding the samples possessing a fair amount of no-call or missed-call genotype data, our more stringent filters certainly improved the actual genotyping results for the association studies performed later. Thus, we were able to obtain an increased level of significance without any substructure effects of the two populations based on the definite diagnosis when combining the genotyping results of the Present and Previous GWAS data sets (Figure 3 ). Using our polished data sets derived from distinct case and control subjects (see Methods and Supporting Information S1), we succeeded in identifying a cluster of genome-wide significant SNPs associated with POAG in CDKN2B-AS1, a non-coding gene with an unknown function, on chromosome 9p21.3 (Figure 2A, B) . Several variants in chromosome 9p21 have been found to be associated with a variety of common diseases, and 9p21 was initially identified as a locus for coronary artery disease (CAD) [20] [21] [22] [23] and type-2 diabetes [24] [25] [26] . However, little is known about the biological meanings underlying the locus, as 9p21 is a ''gene desert'' locus and most of the variants identified were noncoding. Recently, Harismendy et al. [27] reported that they have identified 33 enhancers in 9p21, some of which being within CDKN2B-AS1, and it turned out to be the second densest gene desert for predicted enhancers, thus suggesting the regulatory role of sequences residing within non-coding loci. They finally determined a few adjacent, as well as distant (.45 kb), target gene regions relevant to CAD biology physically interacting with the enhancer by 3D-DSL (also known as ''4C''), a chromatin conformation capture technology, in human vascular endothelial cells. Overall, their study has provided an excellent example of a solution to link the unknown meanings of statistical association to a biological function. Since the POAG variants were also identified in CDKN2B-AS1, the variants would probably affect the expression level of the downstream genes CDKN2A and CDKN2B. In fact, Burdon et al. reported up-regulated CDKN2A and CDKN2B expression in response to the elevated IOP [14] , suggesting the involvement of the locus in molecular pathways leading to glaucoma development. Moreover, the possibility that the variants would also affect the distant unidentified target genes in the context of the complex etiology of glaucoma, as well as the nature of the identified locus [27] , cannot be ruled out. To date, several institutions have attempted to discover variants for glaucoma by conducting a GWAS, and a few published studies have reported the association of particular loci with POAG using Caucasian subjects (Table S5) . Thorleifsson et al. reported the association of common variants near CAV1/CAV2 on 7q31.1 using a population of European ancestry [13] . In contrast, Kuehn et al. reported that they failed to replicate their results in a United States cohort [15] . Moreover, meta-analysis of the association results derived from several institutions of Northern Europe, including the data from Thorleifsson et al. [13] , identified a few new loci associated with POAG, including the CDKN2B-AS1 locus on 9p21.3 [17] . Recently, the new loci at TMCO1 on 1q24.1 and CDKN2B-AS1 were reported to be associated with Australian populations [14] . Thus far, only the association of the CDKN2B-AS1 locus was replicated, even within the same ethnicity of European descent. Interestingly, Thorleifsson et al. also showed a modest association with the CAV1/CAV2 locus using Chinese subjects [13] , although the allele frequency of the particular variant in Asian subjects was quite low when compared with that in Caucasians, suggesting etiological differences due to the genetic background. In fact, according to the results of our Present GWAS for POAG, we were unable to replicate the association with the CAV1/CAV2 and TMCO1 loci ( Figure S5F , G, Table S5 ). Moreover, in the Present GWAS, we were unable to replicate the association results of the 6 SNPs identified in the previous study [10] (Table S5) , even though the populations were of the same ethnic background, thus suggesting that we still need to discover authentic variants, irrespective of the difference in ethnicities, to elucidate the complex etiology of glaucoma. Consequently, it should be noted that the variants identified in the CDKN2B-AS1 locus in this study using a Japanese population seemed to be shared with the Caucasian subjects (Table S5) , thus showing that we have successfully obtained one of the authentic variants for POAG that is not ethnicity related. In this study, we also subdivided the POAG subjects into two subtypes based on the IOP level measurements in an attempt to discover subtype-specific variants, which would be useful for investigating the different mechanism of each pathogenesis. Since the clinical states of both subtypes overlap almost completely, they are usually categorized as a single disease. However, as a unique epidemiological distribution compared with other ethnic groups, ,92% of the Japanese POAG patients fall into the NPG category, as determined by the Tajimi study [4] , a robust epidemiology study. In fact, another Japanese group performed a GWAS focused on discovering NPG-specific variants [12] . They reported that the SNPs in SRBD1 on 7q31.1 and ELOVL5 on 6p12.1 were associated with NPG, although our group was unable to replicate those results ( Figure S5D, E) . On the contrary, we obtained an unexpected result that the variants associated with NPG were completely identical to those associated with POAG in CDKN2B-AS1 identified in this study (Table S1 ). In contrast, none of the variants were significant for HPG ( Figure 3B ). Although we cannot rule out that the significance could be due to the differences in sample size between HPG and NPG cases according to the heterogeneity test (P = 0.21-0.25; Table S3 ), the results suggested that the genetic loci identified are most likely components of the molecular mechanism specific for NPG. It has been reported that there are differences in vulnerability of the optic nerve to IOP between HPG and NPG [18] . It has also been theorized that CDKN2B-AS1 affects the susceptibility of the optic nerve [14, 28] . Since the variants identified in this study seemed to be shared among different ethnicities in functional aspects as well, the other variants should be contributing to the unique epidemiology of NPG in Japanese. By continuing to build upon the detailed investigation, such as obtaining in-depth sequencing data of the non-coding 9p21 locus, it might be possible to reveal not only the molecular mechanism of glaucoma pathogenesis but also the genetic diversity that resides within the locus among different ethnic backgrounds. This study was approved by the Institutional Review Board of Kyoto Prefectural University of Medicine and all procedures were conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent after an explanation of the nature and possible consequences of the study. All participants were interviewed to determine their familial history of glaucoma and other ocular or general diseases. A total of 2,126 Japanese participants, including POAG patients, healthy volunteers, and patients with other ocular diseases, were recruited between March 2005 and December 2010 at the University Hospital of Kyoto Prefectural University of Medicine (Kyoto, Japan) to give peripheral blood samples for this study. A third person who was blind to both the blood sampling and genotyping assigned an anonymous code to each blood sample. Genomic DNA used for the genotyping experiments was isolated from the blood, and Epstein-Barr virus (EBV)-transformed lymphocytes were prepared from the remaining blood as previously described [29] to serve as the future resource of genomic DNA. POAG patients and controls suitable for this study were precisely selected based on the strict diagnosis as previously described [10] . In particular, subtype selection was performed by dividing POAG into two categories according to peak IOP without treatment; POAG/High Pressure Glaucoma (POAG/HPG, defined as $22 mmHg) and POAG/Normal Pressure Glaucoma (POAG/ NPG) [3] . All of the diagnostic procedures, including case-control selection and subtype classification, were performed by three ophthalmologists (YI, MU, and KM) from a single institution. The age and female/male ratio of all of the subjects used in the Present and Previous GWAS are shown in Table S6 . To examine the possible confounding effects of age and gender, we assessed the correlations between the values and the genotype data from the case and control samples by one-way ANOVA or chi-square test (Table S2) . First, 906,600 SNPs were genotyped for 2,126 Japanese subjects by Genome-Wide Human SNP Array 6.0 (Affymetrix, Santa Clara, CA) according to the manufacturer's instructions. As for the Present GWAS population, 839 POAG patients and 708 controls were selected after performing the quality control (QC) as described (Supporting Information S1). To exclude the potential inclusion of genetically related subjects into the population, identity-by-descent (IBD) estimates were performed for all possible combinations by PLINK v1.07 (http://pngu.mgh.harvard.edu/ ,purcell/plink/). In total, 6 POAG patients and 22 controls were assumed to be in first-degree relationships or in relationships with a few more-distant relatives, and were thus excluded from the Present GWAS population. Finally, the Present GWAS population ended up with 833 POAG patients, comprised of 330 HPG and 503 NPG patients, and 686 controls. Population stratification for the present GWAS population was examined by principal component analysis using EIGENSTRAT software v3.0 (http:// genepath.med.harvard.edu/,reich/Software.htm). As for the reference genomic population, the four HapMap populations (CEU, YRI, JPT, and CHB) were simultaneously applied to EIGENSTRAT. The generated cluster plots indicated that our POAG and control population was genetically clustered within the JPT population, and there was no outlier sample ( Figure S6 ). We performed SNP quality control for the population on the autosomal SNPs based on the following QC filters: (i) call rate per SNPs in case and control samples $95%, (ii) minor allele frequency (MAF) in case and control samples $1%, and (iii) Hardy-Weinberg equilibrium (HWE) in control samples P$0.001. Consequently, we analyzed the remaining 653,519 SNPs for the Present GWAS population. Since 6 years had passed since the patients and healthy volunteers who participated in our Previous GWAS [10] were first recruited, we updated the case (n = 418) and control (n = 300) groups based on the latest clinical information. In total, 7 samples from the case group and 11 samples from the healthy control group were removed. The population for the reanalysis finally ended up with 411 POAG patients, comprised of 215 HPG and 196 NPG patients, and 289 controls. The Previous GWAS data was obtained by GeneChipH Human Mapping 500 K Array platform (Affymetrix) containing 500,568 SNPs. The quality control for the reanalysis was performed by using the same QC filters as used in the Present GWAS. Genotype data derived from the particular locus for combining with the Present GWAS data was extracted from these filtered SNPs. To manage and analyze all of the genotype data, we used our in-house Genoika Server System (SASA Plus Co., Ltd., Fukuoka, Japan), which was well improved from the system used in the Previous GWAS [10, 30] by the same system engineers. In addition, to manage the Previous GWAS data effectively, the Labo Server System (World Fusion Co., Ltd.) was used simultaneously. The Genoika Server System comes with PLINK v1.07 (http://pngu.mgh.harvard.edu/,purcell/plink/), the R program v 2.9.2 and 2.10.1 (http://www.r-project.org/), EIGEN-STRAT software v3.0 (http://genepath.med.harvard.edu/ ,reich/Software.htm), and Haploview 4.2 (http://www.broad institute.org/scientific-community/science/programs/medical-andpopulation-genetics/haploview/haploview) built in, and all the analyses were performed by use of this system. In addition, Microsoft Office Excel 2003 (Microsoft Corporation, Redmond, WA) was used for preparing the data sets and statistical analysis. The frequency of alleles in the case and control samples was compared by use of the basic allele test. The odds ratio (OR) and the upper and lower limit of the 95% confidence interval (CI) of each SNP were calculated for the allele possessing a higher frequency in the case samples than in the control samples. The HWE was evaluated by the chi-square test. Q-Q plots were generated by ranking the observed values from minimum to maximum and plotting them against their expected chisquare values using the ''snpMatrix'' package ver 1.14.6 in the R program (http://www.r-project.org/). We applied the Mantel-Haenszel test [19] to combine the data derived from the two data sets in order to reduce potential negative effects arising from the biases in age and female/male ratio among the populations (Table S6 ). Figure S1 Q-Q plots for the Present GWAS. Quantilequantile (Q-Q) plots for the Present GWAS of POAG (A), HPG (B), and NPG (C) were generated by ranking the observed chisquare values from minimum to maximum and plotting them against their expected values. Genomic inflation factors (l) are also shown. These plots were created using the R-package snpMatrix. (PDF) Figure S2 LD plots in the 9p21 locus. LD plots were generated from the Present GWAS data. The SNPs applied to these plots and the span of the region are the same as shown in Figure 2B . Upper and lower plots indicate the value of pairwise D9 and r2, respectively. The positions of 5 SNPs, which passed the Bonferroni correction threshold in the Present GWAS, are drawn in vertical dashed lines. These LD plots were generated using Haploview v4. Figure S6 Population stratification analysis. Population stratification analysis by EIGENSTRAT in the Present GWAS data set. POAG samples separated with HPG and NPG groups were applied. The version of HapMap reference data is release22 (Build36). Since Control, NPG, HPG, and HapMap JPT samples were tightly overlapped, only the plots for NPG (yellow cross) are visible. (PDF) Table S1 Genome-wide significant SNPs that passed the Bonferroni correction threshold in the Present GWAS. (PDF) Supporting Information S1 Quality control for genotyping. (PDF)

Interleukin-18 expression and the response to treatment in patients with psoriasis

INTRODUCTION: The aim of the study was to demonstrate Interleukin-18 (IL-18) expression in keratinocytes from psoriatic lesions in comparison to keratinocytes from uninvolved skin and to study the change of expression after therapeutic interventions. MATERIAL AND METHODS: This study included 16 patients of different clinical subtypes of psoriasis. IL-18 gene expression analysis was performed using real-time quantitative PCR. Three biopsies were obtained from each patient. Two were taken from the lesional psoriatic skin and from uninvolved skin before starting treatment. A third lesional skin biopsy was taken at the end of two months' treatment course. The treatment was in the form of topical steroids or oral systemic methotrexate. RESULTS: Of all 16 studied patients significantly increased IL-18 expression was noted in keratinocytes from psoriatic lesions before and after treatment when compared to keratinocytes from uninvolved skin (P = 0.001 and 0.002 respectively). The IL-18 expression in the skin lesions after treatment was significantly lower than lesional skin before treatment (P = 0.023). In psoriatic skin lesions of all studied patients IL-18 expression was significantly correlated with disease duration (r = 0.40 and P = 0.01) and clinical severity of psoriasis (r = 0.72 and P = 0.001). CONCLUSIONS: Increased IL-18 expression in keratinocytes from psoriatic lesions of our patients and its correlation with disease duration and severity supported the concept which views psoriasis as a T-cell-mediated autoimmune disease. This could establish therapeutic and preventive approaches for psoriasis that ultimately lead to improved outcomes for patients.

Psoriasis is a chronic inflammatory condition which is the result of persistent stimulation of T cells by antigens of epidermal origin [1] . This persistent stimulation of T cells is the result of an interaction between T cells, antigen-presenting cells, i.e. Langerhans cells, and antigens, and comprises the following steps: primary signals (signal 1) including T-cell receptor stimulation by peptide antigen, co-stimulation (so-called accessory signals) (signal 2), and T helper type 1 (Th 1) differentiation and proliferation [2] . In 1986, it was proposed that cytokines released by activated T-lymphocytes initiate and maintain the psoriatic process by stimulating keratinocyte proliferation [3] . Since that time it has become clear that, upon stimulation, keratinocytes are also able to secrete an array of different cytokines with a variety of functional effects on both themselves and other cell types, including T-lymphocytes [4] . Interleukin-18 (IL-18) is related to the IL-1 family in terms of both structure and function. At the receptor level, the activity of IL-18 takes place through a signalling chain of a putative IL-18 receptor (IL-18R) complex [5] . Interleukin-18 was first described as an interferon-γ (IFN-γ) inducing factor. In general, IL-18 induction of IFN-γ is similar to that of IL-12, when it is a sole cytokine. IL-18 induces low levels of IFNγ, but in the presence of co-stimulants, IFN-γ production is greatly enhanced [6] . However, because antibodies to IL-18 also reduced the hepatotoxicity of endotoxaemia, IL-18 was considered to possess other biological properties beyond that of inducing IFN-γ [7] . IL-18 activates T cells to synthesize IL-2, GM-CSF, and TNF-α. There are also reports that IL-18 suppresses the production of IL-10 and does not induce the production of IL-1Ra. In general, the ability of IL-18 to induce different cytokines depends on the cellular targets [8] . Flisiak et al. confirmed an association between plasma IL-18 concentration and psoriasis severity. Moreover, it was shown that combined measurement of IL-18 and TGF-beta1 in plasma can be considered as a possible biomarker of psoriasis activity [9] . Kato et al. indicated that a single nucleotide polymorphism (SNP) in the promoter of the interleukin-18 gene is associated with the presence of psoriasis, but not atopic dermatitis. This suggests that the SNP is associated with susceptibility to psoriasis vulgaris, presumably by affecting the production of IL-18 [10] . A few studies have analysed IL-18 expression in psoriasis vulgaris [11] [12] [13] [14] . Much less information is available on the ability of different therapeutic interventions to modulate ongoing changes in expression in chronic disorders, especially in a human disease like psoriasis. The aim of this study was to demonstrate expression of IL-18 in keratinocytes from psoriatic lesions in comparison to keratinocytes from nonlesional skin and to study the change of IL-18 expression after therapeutic interventions. This study included 16 patients. The patients were selected from the outpatient clinic of Benha University Hospital, Al-Haud Al-Marsoud and The National Research Center, Cairo, Egypt. The patients included 4 (25%) females and 12 (75%) males, and their ages ranged from 21 to 62 years with a mean of 40.87 ±13.17 years. They had different clinical subtypes of psoriasis and exhibited various degrees of severity. History of present illness included the onset, course, duration, previous treatment and date of last topical application or systemic therapy. Inclusion criteria: those included in the study were adult psoriatic patients ≥ 18 years who should not have received a topical therapeutic modality apart from petrolatum for the last four weeks or systemic therapy for the last eight weeks. Presence or absence of any coexisting noncutaneous conditions was established. Three patients were hypertensive, three had psoriatic arthropathy, one had hepatitis C virus (HCV) with anaemia of chronic disorder type and one had diabetes mellitus with hypertension. A positive family history was encountered only in one case. Two patients had upper respiratory tract infections; cold exposure worsened five patients' clinical presentation; sweating aggravated one patient's condition; three patients had psychogenic factors precipitating their cases. Only five of the studied patients were not affected by predisposing factors (31.3%). The psoriasis area and severity index (PASI) score sheet was adopted according to Fredriksson and Pettersson and Marks et al. [15, 16] for all patients as an indication of degree of severity. In mild cases, the extent of the disease does not exceed 10, while in severe cases, the extent of the disease is more than 30 based on the PASI score [17] . Informed consent was obtained after complete description of the study from each participant, according to the guidelines of the local ethical committee of the National Research Center. Two biopsies were taken from each patient in the first set. The first biopsy was taken from the lesional psoriatic skin and the second from nonlesional skin. All patients were subjected to continuous treatment (in the form of topical steroids and oral systemic methotrexate) and followed up for 2 months. Six patients with moderate to severe degrees of psoriasis and two with palmoplantar subtype of psoriasis (who did not respond to topical steroids) received oral systemic methotrexate. The dosage used was: 15 mg per week in three divided doses taken at 12-hour intervals during a 24-hour period with folate supplementation. Patients receiving systemic methotrexate were subjected to monitoring liver blood tests monthly. Eight patients with mild to moderate degrees of psoriasis were maintained on topical steroid creams and ointments in the form of betamethasone dipropionate, twice daily. At the end of the course of two months, a third psoriatic lesional skin biopsy was taken from all patients. To prevent degradation by intracellular RNases, tissues were embedded soon after excision in RNA stabilizing reagent RNAlater provided from QIAGEN worldwide companies according to the manufacturer's instructions to be stored at -70°C before sample processing. Frozen tissues were not allowed to thaw during handling to be ready for Interleukin-18 expression and the response to treatment in patients with psoriasis quantitative PCR determination of IL-18. Each biopsy was weighed before processing. Biopsies weighed up to 20 mg. Efficient disruption and homogenization of the starting material was performed using a rotor-stator homogenizer. Total RNA was extracted using QIA amp RNA extraction kit provided from QIAGEN worldwide companies. The concentration and purity of RNA were determined by measuring its absorbance at 260 nm (A 260 ) and 280 nm (A280) in a UV visible spectrophotometer. Total RNA (2 μg) was reversetranscribed to cDNA with High-Capacity cDNA Archive kit provided from Applied Biosystems. cDNAs were stored at -20°C until real-time quantitative PCR was performed. Real-time quantitative PCR was performed using Applied Biosystems Perkin Elmer 7300 sequence detection system. Primers were IL-18 sense primer 5'-AGG AAT AAA GAT GGC TGC TGA AC-3' and antisense primer 5'-GCT CAC CAC AAC CTC TAC CTC C-3'. Each PCR reaction (in a 50 uL volume) contained 1 x of TaqMan universal PCR master mix, (2 x) 50 to 900 nM of forward primer, 50 to 900 nM of reverse primer, 250 nM of TaqMan probe labelled with 6-carboxyfluorescein (FAM) and 10 to 100 ng of cDNA. The amplification protocol was 2 min at 50°C, 10 min at 95°C, then 40 cycles for 15 sec at 95°C and 1 min at 60°C. Relative gene expression was determined for each sample. Amplification of the gene for glyceraldehydes-3-phosphate dehydrogenase (GAPDH), a constitutively expressed housekeeping gene, was performed on all samples. All genes were subsequently normalized against GAPDH levels. In this study mRNA levels were expressed in picograms/microlitre (pg/μl). All data were collected from the patient charts and entered into a computerized spreadsheet. The fit of the data to the normal distribution was tested with the Kolmogorov-Smirnov test. Since the distribution of the data was significantly different from normal, nonparametric statistics were further used. Comparisons were made using Mann-Whitney U, Kruskal Wallis and Wilcoxon signed ranks tests between variables adjusted for appropriate covariates, and Spearman rank correlation coefficients were calculated between the assessed variables. The null hypothesis was rejected with a two-sided P value of < 0.05. All analyses were performed with SPSS 11.0 for Macintosh (Statistical Package for the Social Sciences. SPSS Inc., Chicago, IL, U.S.A.). Demographic and behavioural characteristics of included patients are presented in Table I . PASI score used to evaluate the severity of psoriasis varied from 1.4 to 39 with a mean of 14.96 ±11.82. According to Ramsay and Lawrence [17] , the PASI score grades in the different subtypes of psoriasis are demonstrated in Table II . All patients had detectable levels of IL-18 mRNA in the uninvolved skin. Table 3 shows the range, mean ± SD and median levels of cytokine gene expression in each skin biopsy at the different times before and after receiving medication whether topical or systemic. Comparisons between median IL-18 expression levels with percentiles in all studied patients are presented in Figure 1 . Significantly lower IL-18 expression in uninvolved skin than lesional skin before treatment (P = 0.001) and lesional skin after treatment (P = 0.002) was observed in all studied patients as one group. The IL-18 expression in the psoriatic skin lesions after treatment was significantly lower than lesional skin before treatment (P = 0.023). IL-18 mean expression levels in non-lesional and lesional skin (before and after treatment) in the different subtypes of psoriasis are shown in Figure 2 . The highest expression level was noted in plaque and nails clinical subtype of psoriasis. Among patients receiving topical steroids significantly lower IL-18 expression was detected in non-lesional skin than lesional skin before treatment (P = 0.012) and after treatment (P = 0.012). No statistical significance was obtained when comparing the change of expression before and after topical treatment (P = 0.069). Individual variations of the IL-18 expression levels in non-lesional and lesional skin before and significantly lower IL-18 expression was detected in non lesional skin than lesional skin before treatment (P = 0.018). The change of expression levels in lesional skin after treatment was not statistically significant when compared to non lesional skin (P = 0.069) and lesional skin before treatment (P = 0. 12) after receiving topical steroids are presented in Figure 3 . In patients receiving methotrexate, significantly lower IL-18 expression was detected in non-lesional skin than lesional skin before treatment (P = 0.018). The change of expression levels in lesional skin after treatment was not statistically significant when compared to nonlesional skin (P = 0.069) and lesional skin before treatment (P = 0.12). Individual variations of the IL-18 expression levels in non-lesional and lesional skin before and after receiving systemic methotrexate are shown in Figure 4 . In the correlation between IL-18 expression and different demographic and behavioural characteristics of the studied patients as one group, a significant correlation was obtained between IL-18 expression levels in psoriatic lesions and the severity of psoriasis expressed by PASI score (r = 0.72 and P = 0.001) and disease duration (r = 0.40 and P = 0.01). No significant correlation was revealed between IL-18 expression levels in psoriatic lesions and patients' age, smoking status, predisposing factors or coexisting diseases. In the present study our patients with psoriasis showed significantly increased IL-18 expression in keratinocytes from psoriatic lesions before and after receiving medication whether topical or systemic in comparison to keratinocytes from non-lesional skin. This supports the concept viewing psoriasis as a T-cell-mediated autoimmune disease. Krueger [18] reported, in his study on a large series of skin biopsies, increased protein expression in lesional as opposed to non-lesional skin samples. Other investigators stated that lesional skin might be the source of the elevated plasma levels of IL-18 [19] . In the present work quantitative PCR using mRNA taken from uninvolved skin revealed low but detectable levels of IL-18 mRNA expression. This finding is consistent with previous studies showing that human keratinocytes are capable of synthesizing low levels of IL-18 mRNA even under non-stimulated conditions [20] . Chang et al. [21] found that genes specifically modulated in uninvolved skin play a major role in triggering disease progression, and in being the first candidate for early disease diagnosis or development of new therapies. This finding could allow selection of the treatment regimen for individual patients. In the present study the IL-18 expression in patients receiving topical steroids was significantly lower in non-lesional skin than lesional skin before and after treatment. After 2 months of treatment, IL-18 expression levels returned back to low values but not significantly lower than the lesional skin levels of expression before treatment. In patients receiving methotrexate non-lesional skin expression was significantly lower than lesional skin before treatment. Also the expression levels decreased after treatment but still not statistically significantly when compared to lesional skin before treatment. Otkjaer et al. [22] determined IL-19 and IL-20 in nonlesional and lesional psoriatic skin after 28 days of treatment using topical calcipotriol ointment for mild cases and oral cyclosporine for moderate and severe cases. They found that in patients treated with calcipotriol, IL-19 and IL-20 mRNA expression levels in lesional psoriatic skin were reduced after 14 days, but the reduction became statistically significant only after 28 days. So the timing for mRNA expression analysis could affect the degree of significance. This is further supported by studies demonstrating a decrease of IL-18 levels after narrowband UVB therapy together with other parameters' characteristics for the T helper 1 (Th1) response [23] . Apart from stimulating the Th1 response, IL-18 can regulate the Th2 response depending on the local cytokine network. IL-12 enhances IFN-γ production induced by IL-18, whereas IL-18 alone induces IL-4 and IL-13 production [24] . Classic systemic treatments for psoriasis have not fully met the needs of patients for permanent improvement. Antibody-based or fusion proteinbased selective targeting of key mediators of inflammation has been added to the treatment approaches for psoriasis. Kong Our study demonstrated that IL-18 expression in the psoriatic lesions before treatment was significantly correlated with the severity of psoriasis and disease duration. No correlation was noted between IL-18 expression and patients' age, smoking status, predisposing factors or coexisting diseases. Flisiak et al. [9] found a significant correlation between plasma IL-18 levels and PASI values. But they did not find any correlation between IL-18 concentration in scales and PASI score. Trinquet et al. [28] studied the influence of disease duration on gene expression profiles in psoriasis. They did not reveal any correlation. However, Craven et al. [29] studied IL-10 in early and late onset psoriasis and confirmed that IL-10 expression was correlated significantly with disease duration. Sampogna et al. [30] studied 380 patients with psoriasis and did not find any significant difference between older and younger patients regarding IL-18 levels. On the other hand, Chen et al. [31] found a statistically significant correlation between IL-18 expression levels and age, with a P value of 0.015. In conclusion, based on the increased IL-18 expression levels in psoriatic skin lesions relative to uninvolved skin, this cytokine appears to be crucial in the development of the active psoriatic lesion itself, where it is produced locally by a step in the evolution of the psoriatic lesion. This could be supported by its correlation with clinical severity and disease duration, and reduction of its expression after treatment. The decreased expression after treatment was not to an extent indicating that the skin had returned to normal. The small size of the study population and the heterogeneous group of patients participating in the study preclude the drawing of firm conclusions. Future studies on genes involved and array-based technologies should allow confirmation of the results obtained. Also further studies are recommended to try other lines of treatment for psoriasis. One of these lines of therapy is to try r-h IL-18 BP to neutralize IL-18 activity on a wide scale that will ultimately lead to improved outcomes for patients.

Replication enhancer elements within the open reading frame of tick-borne encephalitis virus and their evolution within the Flavivirus genus

We provide experimental evidence of a replication enhancer element (REE) within the capsid gene of tick-borne encephalitis virus (TBEV, genus Flavivirus). Thermodynamic and phylogenetic analyses predicted that the REE folds as a long stable stem–loop (designated SL6), conserved among all tick-borne flaviviruses (TBFV). Homologous sequences and potential base pairing were found in the corresponding regions of mosquito-borne flaviviruses, but not in more genetically distant flaviviruses. To investigate the role of SL6, nucleotide substitutions were introduced which changed a conserved hexanucleotide motif, the conformation of the terminal loop and the base-paired dsRNA stacking. Substitutions were made within a TBEV reverse genetic system and recovered mutants were compared for plaque morphology, single-step replication kinetics and cytopathic effect. The greatest phenotypic changes were observed in mutants with a destabilized stem. Point mutations in the conserved hexanucleotide motif of the terminal loop caused moderate virus attenuation. However, all mutants eventually reached the titre of wild-type virus late post-infection. Thus, although not essential for growth in tissue culture, the SL6 REE acts to up-regulate virus replication. We hypothesize that this modulatory role may be important for TBEV survival in nature, where the virus circulates by non-viraemic transmission between infected and non-infected ticks, during co-feeding on local rodents.

Tick-borne encephalitis virus (TBEV) is a human pathogen that causes about 16 000 human cases of tick-borne encephalitis (TBE) across Europe and Asia annually (1) (2) (3) . Taxonomically, TBEV is a species within the mammalian tick-borne flaviviruses (mTBFV). Together with the seabird tick-borne flavivirus group (sTBFV), they comprise one ecological group of tick-borne flaviviruses (TBFV) within the genus Flavivirus, family Flaviviridae. Two other ecological groups within the genus Flavivirus are the mosquito-borne flaviviruses (MBFV) and flaviviruses with no-known vector (NKV) (4) . A fourth group including Kamiti River virus (KRV) (5) , cell fusion agent virus (CFAV) (6) and Culex flavivirus (CuFV) (7) have been isolated only from mosquitoes with no demonstrated capacity to replicate in mammals and are under consideration by the ICTV Committee for classification as 'probably arthropod-borne viruses' (PABV). Flavivirus virions are $50-nm particles with a nucleocapsid composed of capsid (C) protein surrounding a positive-sense single-stranded RNA genome of $11 kb. The capsid is enclosed in a lipid membrane within which the viral membrane (M) and envelope (E) proteins are embedded. The genome encodes a single polyprotein of approximately 3400 amino acids from which the three structural (C, M and E) and seven non-structural (NS1, NS2A, NS2B, NS3, NS4A, NS4B and NS5) proteins are processed by cellular and viral proteases (8) . Flavivirus genome replication involves synthesis of a negative-sense template strand by the RNA-dependent RNA polymerase (RdRp; NS5 pol ) from which additional genome-sense strands are transcribed. This process is controlled by numerous RNA-RNA and RNA-protein interactions determined by virus RNA sequence motifs and secondary structures, called cis-acting replication *To whom correspondence should be addressed. Tel: +44 0118 378 6368; Fax +44 (0)118 378 6537; Email: t.s.gritsun@reading.ac.uk elements (CRE), mapped to the 5 0 -and 3 0 -untranslated regions (UTR) that flank the single open reading frame (ORF) of the genome (9) (10) (11) (12) (13) (14) (15) . The concept of promoter and enhancer function during replication has been introduced recently in relation to the flavivirus CREs (16) . The promoter has been identified as a complex of highly conserved interacting RNA structures recruited from the 5 0 -and 3 0 -UTR to assemble viral and cellular proteins into a functional RdRp complex. In evolutionary terms, the 3 0 -UTR of the TBFV group is formed by four conserved long imperfectly repeated sequences (LSRs), genetic remnants of which are revealed in the MBFV, NKV and PABV groups (17) . It has been proposed that the 5 0 -UTR may have evolved from a trans-terminal duplication of the archival flavivirus 3 0 -UTR (16) . An additional complexity in flavivirus replication is the presence of replication enhancer elements (REEs) in the 3 0 -UTR that, while not obligatory for replication of laboratory-maintained viruses, are likely essential for virus circulation and transmission in nature (16, 18) . Engineered deletions or modifications of the REEs enable the recovery of viable viruses that are attenuated as a result of reduced RNA synthesis (10, (19) (20) (21) (22) . The cumulative effect of several REEs enhances the assembly of the RdRp complex and is probably critical to the survival of flaviviruses in nature (23) . The REEs identified for MBFV have become an important target for the development of a live attenuated vaccine for dengue virus (24, 25) . The relatively compact nature of the flavivirus genome, together with constraints imposed by the need to replicate in vertebrate and invertebrate hosts, means that additional CRE sequences may be present in parts of the genome other than the non-coding regions. Indeed, RNA secondary structures have been predicted within the coding region of several flaviviruses (26) (27) (28) . Here, using bioinformatic and reverse genetic analysis we demonstrate that the capsid-encoding region of TBEV contains an REE which we designate SL6 (26, 27) . Phylogenetic evidence suggests that the MBFV group also contains at least a partial SL6-like structure, though it is absent in the NKV or PABV groups. The significance of these findings in the context of flavivirus evolution and adaptation to transmission is discussed. Genbank accession numbers for sequences from all four groups of flaviviruses (TBFV, MBFV, NKV and PABV) used for in silico analysis are listed in Supplementary Table S1 . RNA nucleotide sequences were aligned using ClustalX (29) and then edited manually. Nucleotide and dinucleotide scans and analysis of suppression of synonymous site variability (SSSV) were determined by mean pair-wise distance comparison at each codon within the ORF using the Simmonics 1.6 package (http://www.picornavirus.org/), as previously described (30) . SSSV was calculated only at aligned codon positions in which over 40% of sequence comparisons were synonymous and averaged over a sliding window of 21 codons; consequently, data point are only produced from codon 11. RNA secondary structures were predicted using the MFold 3.2 and DINAMELT packages (http://mfold .bioinfo.rpi.edu/) with default settings (31, 32) . Phylogenetically conserved RNA structures were predicted using STRUCTURE_DIST (http://www. picornavirus.org/) to analyze connect files generated using hybrid-ss-min from the UNAFold suite of programs (32) . Porcine embryo kidney cells (PS) have been used in experiments with TBEV strain Vasilchenko (Vs) and its infectious clone (pGGVs) as described previously (33) (34) (35) . The construction of the infectious clone pGGVs for Vs virus and methods of mutagenesis have been described (34, 36) . Briefly, the pGGVs was subcloned into two plasmids; one, pGGVs 660 contained the first 660 nt of the virus genome and the second pGGVs 660-10927 included the remainder. Site-directed mutagenesis was accomplished by PCR (details of primers are available on request). Mutated PCR products were cloned into the pGGVs 660 between MluI and EcoRI sites followed by sequencing. The recovery of virus from the two plasmids representing the infectious clone has been described previously (34) (35) (36) . Briefly, plasmid pGGVs 660 (or mutated derivatives) was digested with PspAI, dephosphorylated with Shrimp Alkaline Phosphatase (SAP; USB) and, after heatinactivation of SAP, digested with AgeI. Similarly, pGGVs 660-10927 was digested with NotI, dephosphorylated and then digested with AgeI. The excised linker DNA fragments from pGGVs 660 and pGGVs 660-10927 were removed using MicroSpin TM S-400 columns (Pharmacia Biotech) and ligated at the AgeI site generating full-length cDNA which was linearized with SmaI and used as a template for SP6 transcription (34) . In vitro-synthesized RNA was inoculated intracerebrally into suckling mice to recover the mutant viruses which were not passaged further prior to phenotype evaluation (35) . Recovered viruses were amplified by RT-PCR between nucleotides 1-940 (5 0 -UTR-C-prM region of the TBEV genome) and 10206-10927 (3 0 -UTR), and sequenced to validate the presence of the introduced mutations and to exclude extraneous mutations at the 5 0 -UTR and 3 0 -UTR (36) . For growth curves, monolayers of PS cells in 96-well plates were infected with viruses at a multiplicity of infection (moi) of 1 PFU/cell, in quadruplicates. The inoculum (30 ml) was removed after 1 h, the monolayer washed thoroughly and replaced with 200 ml of media containing 2% serum. Media (10 ml) was collected at different time-points (8, 12, 16 and 24 h post-infection) and stored frozen at À70 C, before virus quantification by plaque assay. For cytopathic effect (cpe), PS cell monolayers were infected in 96-well plates at an moi of 1 PFU/cell, in quadruplicates, and stained with naphthalene black after 72 h. Statistical analysis was performed on the data obtained from the virus growth curve studies and the evaluation of cpe in PS cells. For growth curves, the data were plotted to include the standard error of the mean (SEM) for each data set. At any given time point divergence by at least 2 SD from the mean, between wild-type and mutant viruses, was taken as significant. Measurement of cpe was done visually by three independent evaluators in a 'blind' manner. The cpe of viruses were estimated on a scale of 1-4 corresponding to 20-40, 40-60, 60-80 and 80-100% of monolayer destruction following microscopic examination. The interevaluator consistency error was verified using F-test which revealed no one evaluation was significantly different from that of the others. Previous in silico studies have predicted a stable RNA structure designated SL6, in the C protein-encoding region, for a limited number of viruses within the mTBFV subgroup (16, (26) (27) (28) . Structural RNA elements were also revealed in the C region of some MBFV (28) although their homology to SL6 of TBFV had not been established. Here, we utilized a variety of independent structure prediction methods and a much larger sample of viral sequences to analyze whether or not the SL6-like structure was conserved throughout the entire genus Flavivirus. In silico analysis of SL6 in the TBFV subgroup It was found that the overall folding of the first 333 nt of TBFV was highly conserved among several members of the mTBFV subgroup (16, (26) (27) (28) , with six stable SLs (enumerated 1-6 in Figure 1A ). This analysis was extended to investigate the conservation of SL6 in the larger group of distantly related mTBFV, sTBFV and KADV (37) . A nucleotide alignment of the C region was generated and optimized by the introduction of numerous gaps (Supplementary Figure S1A) ; it shows that divergent RFV, GGYV and KSIV (distant virus species of the mTBFV) maintained homology in the SL6 region. However, some nucleotide perturbations in the SL6 region were observed between mTBFV, sTBFV (MEAV, TYUV and SREV) and KADV proving that the region between the initiation codon and SL6 had evolved with frame shifts as we previously demonstrated (16) . We conducted MFold analysis to investigate the presence of SL6-like structures in the distantly related mTBFV (RFV, GGYV and KSIV), sTBFV and KADV groups ( Figure 1B) . Despite sequence divergence, all viruses in the mTBFV group formed similar SL6-like structures when the 333 nt or a longer nucleotide region (up to 1000 nt) was used for MFold analysis (data not shown). The SL6-like folds contained a remarkably high number of co-variant and semi-covariant substitutions which maintained the general conformation across divergent viruses ( Figure 1B ). The minimum free energy dG of folding for SL6 varied between À32.3 and À17.2 kcal/mol with RFV and LIV/GGYV as extremes in this range. Although KADV had a shorter SL6 compared with other TBFVs, the energy of folding was À17.32 kcal/mol, within the range found for mTBFV. In comparison to SL6 of the mTBFV, the SL6 of sTBFV was shorter and less stable, with a dG in the range À12.5 to À10.6 kcal/mol ( Figure 1B) . However, the SL6-like structures of sTBFVs were observed as elements of longer and branched RNA conformations (data not shown). A smaller terminal loop was revealed in the KFDV/ AHFV and KSIV sequences resulting in the formation of the tetraloop U(GCCA)A ( Figure 1B ). The presence of U:A as a loop-closing base pair has been shown to decrease tetraloop stability considerably; in combination with some intraloop sequences this results in intermolecular tensions that prevent the folded tetraloop from achieving a global thermodynamic minimum (38, 39) . Thus, despite the MFold-mediated predictions, a tetraloop may not form for KFDV/AHFV and KSIV or at least not be sufficiently stable for biologically significant (RNA-RNA or RNA-protein) interactions. The conservation of a UGCCAA hexanucleotide motif in the terminal loop of SL6 in all the divergent TBFVs was striking. Both TYUV and KADV showed one substitution in the hexanucleotide UGCCUA; TYUV has also lost the first nucleotide (Supplementary Figure S1A ). In the minus-sense orientation, the conservation of an SL6-like structure was not as robust as in the positivesense. Although most of the TBFVs formed a structure in the minus-sense RNA, the number of hydrogen bonds, the lengths of the stems and free minimal energy of folding varied significantly even between closely related viruses (data not shown). Consequently, the formation of SL6 is likely to be biologically significant only in the positive-sense RNA. Structure predictions correlated with evidence for SSSV in TBFV genomes ( Figure 2) . A remarkable drop in SSSV was observed in the SL6 region between positions 209 and 254 of the Vs sequence. The most extreme drop in variability was observed in a window centred on position 221 within the apical stem of SL6. The levels of SSSV within the remainder of the structural protein-encoding region (positions 295-2435) were higher than the upstream portion. Similarly, high levels of SSSV were observed across the non-structural portion of the genome between positions 322 and 2425 (data not shown). We excluded the possibility that SSSV in the C-coding region was due to codon bias by analyzing nucleotide composition at each position within the codons. Comparison of SL6 between TBFV species. Numeration in brackets corresponds to SL6 numbered from the start codon of each virus (abbreviated in Table S1 ). Free energy dG values of folding are shown in kcal/mol. Covariant and semi-covariant substitutions are underlined on Vs virus. No unusual variation of G/C or purine/pyrimidine composition was observed at the third codon position or at positions one or two of the codon (not shown). Likewise, we analyzed the dinucleotide composition at all three possible positions. Although there was a general under-representation of CpG and UpA, and overrepresentation of CpA and UpG, there was no correlation between areas of SSSV and regions of unusual dinucleotide frequencies (data not shown). These results indicated that evolutionary constraints restrict nucleotide variation within the 5 0 -coding regions of flavivirus genomes. The phylogenetic conservation of thermodynamically stable RNA structures across all TBFV group ORFs was further analyzed using the program STRUCTURE_DIST ( Figure 2 ) (40) . This method quantifies phylogenetically concordant structures predicted using the widely accepted MFold or UNAFold algorithms, which can then be aligned and overlaid with SSSV results (31, 32) . Analysis of the entire ORF showed the most striking evidence for conserved base-pairing between the initiation codon at position 133 and position 318, after which a large drop in the frequency of base-paired nucleotides was observed. Within this region SL6 was predicted to be the most significant structure, with conserved pairing between 209 and 254 centred on a region with a conserved lack of base pairing between positions 228 to 236, representing the unpaired apical loop of SL6. The base-paired stem of SL6 contained conserved short single-stranded regions between positions 218-220 and nucleotides 244 and 245 consistent with the unpaired bulge, either side of the paired stem. This corresponds exactly to the position and structure of SL6 predicted by MFold (Figure 1 and Supplementary Figure S1A ). An annotated nucleotide alignment of the C-coding region between TBFV and three MBFV groups (JEV, DENV and YFV) was constructed based on a previously presented alignment (16) but modified to include newly sequenced distantly related mTBFV, sTBFV and KADV isolates (Supplementary Figure S1A ). The C protein TBFV/MBFV alignment (available on request) was used to anchor the divergent nucleotide sequences. The annotations include the 5 0 -CYCL of MBFV, an 8-nt long cyclisation domain highly conserved between all MBFVs (16) . The 5 0 -CYCL interacts with a complementary sequence 3 0 -CYCL in the 3 0 -UTR to form a dsRNA panhandle, a vital element of the replication promoter that initiates viral RNA synthesis (16) . For the TBFV the 21-nt long 5 0 -CYCL is located in the 5 0 UTR (i.e. outside the alignment in Supplementary Figure S1A ; highlighted in Figure 1A ). The 5 0 -CYCL for MBFV mapped to the capsid gene and, among the TBFV, aligns optimally with a region that is identified only in TUYV (Supplementary Figure S1A) . Nucleotide sequence homology was observed between the TBFVs and MBFVs particularly in the SL6 region of some JEV group viruses. For example, WNV was observed to share both the stem and loop sequences of TBFV SL6 (Supplementary Figure S1A) . It is of note that the SL6-like region of MBFV maps directly downstream of the highly conserved 5 0 CYCL (Supplementary Figure S1A) . MFold was used to test the ability of these regions to form SL6-like structures within each MBFV group and the stem and loop elements of these SL6-like structures were superimposed onto the TBFV/MBFV alignment (Supplementary Figure S1A) . This comparison revealed that structures predicted within each MBFV group show not only sequence but also structural homology with SL6 of the TBFV group. This alignment was further annotated with RNA structures predicted by the ALIDOT-based analysis of entire flavivirus genomes of 11 000 nt (28), i.e. JE2, JE3 and JE4 for JEV; DV2 and DV3 for DENV and YF4 for YFV (Supplementary Figure S1A) . For all MBFVs with the exception of YFV the MFold predictions were somewhat different from those made using ALIDOT, most likely due to the shorter length of the regions (60-80 nt) used for the MFold analysis. Additional statistical methods, SSSV and STRUCTURE_DIST were used to assess the conservation of the SL6 homologous structures for each of the major MBFV groups (Figure 2) . For the JEV group the mean SSSV between positions 117-358 (start codon at position 97) was consistent with ALIDOT-predicted RNA structures JE2, JE3 and JE4 (Supplementary Figure S1A) (28) . However, the SL6-like structure for the JEV group was clearly predicted by STRUCTURE_DIST analysis (brown box in Supplementary Figure S1A ), in accordance with MFold and alignment analysis. For the DENV subgroup, a marked region of SSSV was revealed in the C-coding region between positions 155-257 (start codon at position 95) when compared with the rest of the structural coding region (Figure 2) , consistent with RNA structure DV3 previously predicted between nucleotides 163-183 (28) (Supplementary Figure S1A) . Both ALIDOT-predicted DV2 and DV4 (28) fall immediately either side of the region of maximum SSSV suggesting that they are less conserved than DV3 (Supplementary Figure S1A) . STRUCTURE_DIST also predicts the formation of the DV2 and DV3 but not the SL6-like structure (Supplementary Figure S1A and Figure 2 ). However, a truncated SL6-like structure was predicted to form in all DENV serotypes, albeit at a suboptimal energy level, when the SL6-like region was folded independently from neighboring regions that form more stable overlapping structures (Supplementary Figure S1A) . Taken together, these data indicate that the DENV SL6-like structure was the least stable conformation among the MBFVs, potentially preventing its prediction by statistical approaches used here and elsewhere (28) . Despite this, the short-stem region of putative DENV SL6-like structures is highly conserved within the DENV group (DENV serotypes [1] [2] [3] [4] and also between DENV and JEV (Supplementary Figure S1A) suggesting that a linear or conformational signal at this location might have some functionality. A similar restriction in SSSV was observed in the YFV C-coding region, with maximum SSSV corresponding to the ALIDOT-predicted structure YF4 ( Figure 2 ) (28). Among the MBFVs, only the YFV SL6-like structure was predicted by both thermodynamic and phylogenetic methods. In summary, a proximally truncated SL6-like structure was predicted in all MBFV groups, although it was less stable in the DENV group, particularly the DENV3 serotype. In silico prediction of SL6 in the NKV and PABV groups In contrast to TBFV and MBFV, the NKV and PABV groups are not arboviruses and their replication is limited to only one natural host, i.e. rodents/bats (NKV) or mosquitoes (PABV). The high nucleotide divergence (Supplementary Figure S1B and S1C) and limited number of complete published sequences for members of the NKV and PABV groups precluded the use of both phylogenetic and thermodynamic approaches to RNA structure prediction. When MFold analysis was performed with available sequences, no thermodynamically stable RNA structures were observed in the region corresponding to the TBFV SL6 region. However, an SL6-like structure, with a similar apical loop CCAA motif was observed in KRV (PABV), upstream of the analogous TBFV SL6. Strategy of mutagenesis on stem-loop 6. Initial design of mutations focused on synonymous codon positions. However, in all but a few instances, this was limited due to the distinctive sequence organization of the apical loop and base paired stem. The first and third codons of the conserved MPN tripeptide (loop region) are limited in respect of variation; M could not be changed and N has two possible silent variations both of which are outside the apical loop ( Figure 3) . Consequently, when mutating the terminal loop sequence UGCCAAAU, silent substitutions could only be introduced into the P codon. Similar difficulties were encountered with mutagenesis of the stem, in which the vast majority of possible synonymous and non-synonymous mutations resulted in no significant conformational changes. The MFold-simulated folding of numerous SL6-mutants revealed a high level of evolutionary 'protection' of SL6 against spontaneous single mutations (not shown) and provides additional evidence for the maintenance of SL6 functionality. In order to resolve the difficulties with design of mutations, three different approaches were adopted ( Figure 3) . First, we introduced all possible silent substitutions, to target the conserved hexanucleotide and the stem (mutants C12, C13, C14, C16 and C33). Second, we introduced mutations (C10, C15, C17, C19 and C34) that mimicked 'natural' amino acid substitutions observed in this region of other mTBFV spp. Third, as a control for mutations that changed amino acids we also introduced compensatory substitutions encoding the same mutated amino acids while restoring the SL6 structure. Accordingly, mutations R32, S31, N 28, V 39, V 39 and P 28 were designed as controls for non-synonymous mutants C22, C23, C27 and C34 (Figure 3) . The predicted impact of each substitution (Figure 3 ) on the secondary structure of SL6 is shown in Figure 4 . The plaque characteristics, cpe and growth dynamics of each mutant compared with those of original pGGVs virus (Table 1 and Figure 5 ). Single-step growth curves revealed differences of $1 log 10 between the mutants early after infection (12-16 h p.i.) which were reproducible and statistically significant ( Figure 5) . To exclude the effect of spontaneous mutations in the 5 0 -and 3 0 -UTRs which contain TBFV promoter and enhancer elements (16) that might compensate for the effect of the SL6-mutations, rescued virus was not passaged prior to phenotype evaluation and key regions of the genome (1-940 and 10206-10927) were sequenced following recovery of each SL6-mutated virus. Only the intended substitutions were present, with no reversions or other compensatory mutations were observed. The effect of each mutation (reduction from large wild-type plaques of the pGGVs virus to medium, small or pin pointed) was scored if the SL6-mutated strain contained >90% of plaques with the altered morphology. The presence of a minor plaque population (between 1 and 10%) was considered as the inevitable result of the variation inherent R 1 4 5 3 3 3 6 2 4 5 2 6 3 2 8 2 2 9 0 2 | | | | | | | | TBEV L40361 ACG CGU CAA UCC AGA GUC CAA AUG CCA AAU GGA CUC GUG UUG AUG CGC TRQSRVQMPNGLVLMR TBEV C10 . in all RNA viruses, the consequence of a high error rate in the virus RdRp (41) . Sequence changes in the apical loop of SL6. In mutants C12, C13, C14, C16 and C19 substitutions within the apical loop changed the nucleotide sequence without altering the overall conformation (Figures 3 and 4) . Four of these mutations C12, C13, C14 and C19 were introduced into the conserved hexanucleotide UGCCAA. Silent mutations C12, C13 and C14 changed plaque morphology; the C13 and C14 mutants that contain purine-to-pyrimidine substitutions also showed reduced growth characteristics and cpe (Table 1 and Figure 5 ). Silent substitution C16, located outside the conserved hexanucleotide in SL6, did not affect the virus phenotype. Two purines were changed for two pyrimidines in mutant C19, one in the conserved hexanucleotide. This mutant was highly attenuated in cell culture producing no cpe and a small turbid plaque phenotype (Figures 3-5 and Table 1 ). These two purine-pyrimidine substitutions resulted in the amino acid substitution M 33 !L that mimicked the corresponding natural amino acid in KFDV and AHFV (Figure 3) . Nevertheless, M 33 !L, mutant C15, produced by different nucleotide substitutions had only a moderate effect on virus replication (below). Therefore, the biological consequences of mutation C19 may be attributed, at least partially, to the nucleotide substitutions. Conformational changes to the apical loop of SL6. Mutations C10, C11, C15, C21, C22 and C23 changed the shape of the loop and base-paired stem within SL6 (Figure 4) . Replication of mutant C10 (with an enlarged loop and shortened stem) and C17 (restored wild type conformation due to the second compensatory mutation) was delayed in the early stage of the infection cycle; C17 caused slightly reduced cpe but the plaque morphology of both was equivalent to that of the parental pGGVs virus ( Figure 5 and Table 1 ). The minor phenotypic changes resulting from these mutations could be explained by the accompanying amino acid substitutions N 35 !K and Q 32 !P imitating POWV ( Figure 3 ). However, a silent substitution A 234 !G that also enlarged the apical loop of mutant C11 (Figure 4 ) Table 1 . Affect of mutations within the SL6 on TBEV phenotype Plaque size for each mutant was defined as large (5-6 mm), medium (3-4 mm), small (1-2 mm) or pinpointed (>1 mm). Some plaques, in comparison with parent Vs virus, were described as turbid. The cpe produced by each mutant in comparison to the wild-type virus was evaluated on a scale of 0-4 where 0 indicates no cpe and 4 is maximum cpe (i.e. 80% cell lysis as observed for the control pGGVs virus) in five repeated experiments, each in quadruplicates. Nt/AA* -Nucleotide/amino acid substitutions. caused similar biological effects; it did not affect virus plaque size or level of cpe, but reduced virus replication rate early after infection ( Figure 5 and Table 1 ). Mutation C15, which shortened the apical loop (Figure 4) , did not affect virus growth but changed the plaque morphology and delayed the development of cpe (Table 1 and Figure 5 ). The C15 mutation altered the amino acid M 33 !L, which imitates the KFDV/AHFV group (Figure 3) , potentially contributing to the observed biological effect. Mutation C21 that reduced the apical loop size (Figure 4 ) interfering with exposure of the hexanucleotide, also had a moderate affect on virus growth although the accompanying effect of amino acid substitution Q 32 !H (Figure 3 ) cannot be excluded (Table 1 and Figure 5 ). Mutant C22 contained three substitutions that considerably increased the size of the apical loop thus shortening the base paired stem. Three nucleotide substitutions present in mutant C23 had the opposite affect in shrinking the apical loop (Figures 4) . Both C22 and C23 had altered growth dynamics, plaque morphology and cpe (Table 1 and Figure 5 ). The nucleotide substitutions of both led to amino acid substitutions Q 32 !R and V 31 !S, respectively. To exclude their influence on virus growth, counterpart 'control' mutants R 32 and S 31 were analyzed, with the same amino acid substitutions but without alteration of the SL6 conformation (Figures 3 and 4) . Both of these control mutants exhibited wild-type plaque morphology and cpe characteristics (Table 1) . Substitutions in the stem of SL6. Three mutants were designed to investigate the influence of SL6 stem length. Most attempts to design synonymous substitutions had little effect on the stem folding conformation. Only silent mutant C33 (C 253 !A and C 255 !G) exhibited a significantly shortened duplex stem, with a corresponding elevated level of dG folding energy. These positions are highly conserved among the mTBFV (Figure 3 ) and, as expected, had a profound effect on virus replication; C33 displayed pinpoint plaques, reduced growth characteristics and almost no cpe (Table 1 and Figure 5 ). Two other mutants C27 and C34 had shortened stems due to the formation of a large internal bulge (Figure 4) , and exhibited profoundly altered biological characteristics (Table 1 and Figure 5 ). However, C27 and C34 included amino acid substitutions Q 28 !N and Q 28 !P, respectively, the latter resembling POWV (Figure 3 ). To rule out the amino acid change as influential, two mutants were designed as a control for C27; double mutant N 28 V 39 and single mutant V 39 , neither of which affected SL6 conformation (Figure 4 ). Similarly mutant P 28 , a control for C34, contained the same amino acid substitution Q 28 !P but maintained SL6 conformation (Figures 3 and 4 ). All three control mutants, N 28 V 39 , V 39 and P 28, displayed wild-type large plaque phenotype and cpe (Table 1) . In a previous study using MFold-simulated RNA structures for a limited number of mTBFV species, we predicted the existence of SL6 in the C-coding region of TBFV. A conserved hexanucleotide UGCCAA in the apical loop and compensatory mutations in the duplex stem of the SL6 implied the formation of the stable RNA structure in ORF of the TBFV genomes (26, 27) . However, contradicting these findings a deletion within the C-coding region, which included SL6, did not prevent recovery of viable, albeit attenuated, TBEV (42) . In this study, we employed a variety of complementary phylogenetic and thermodynamic methods to examine the evolutionary conservation of SL6 using a much larger sample of significantly divergent TBFVs, including new members of the mTBFV, sTBFV and Kadam subgroups (37) . The viruses in the other ecological groups, namely MBFV, NKV and PABV were also included in this analysis to trace the evolution of SL6 throughout the entire genus Flavivirus. In addition, we used a reverse genetic system (34, 36) to engineer TBEV strains with mutated SL6 to reveal the biological significance of this structure. Thermodynamic and phylogenetic analysis of large sequence data sets indicated that all TBFVs including even the distantly related mTBFV, KADV and sTBFV form an SL6-like structure with an exposed conserved hexanucleotide although the molecular details of the predicted stem-loop varied among the mTBFV, sTBFV and KADV subgroups ( Figure 1B) . In similar manner, an SL6-like structure has been predicted for MBFV although with less stability in comparison to SL6 in TBFV. Two other flavivirus groups NKV and PABV demonstrated no significant sequence homology with the TBFV SL6 region although the genome of KRV (PABV group) formed a thermodynamically stable structure in close vicinity to the TBFV SL6 with a similar terminal loop motif CCAA (TBFV-UGCCAA) (Supplementary Figure S1) . To test the biological significance of SL6 in the TBFV group we engineered 21 mutant viruses with point mutations that altered the linear sequence of the unpaired apical loop or destabilized the base-paired stem. Substitutions within the conserved hexanucleotide loop down-regulated virus growth kinetics whereas changes in the terminal loop outside the hexanucleotide sequence did not alter the observed phenotype. The most significant changes of virus phenotype resulted from substitutions that distorted the stem of SL6; mutations that influenced the length or stability of the stem resulted in the recovery of viruses that formed small and/or turbid plaques. Increasing or decreasing the size of the apical loop had a minor biological effect on virus replication although this could also be interpreted as an effect of the altered stem length. However, the changes in replication kinetics from all modifications of SL6 were moderate and manifested themselves predominantly during the early stage of the virus replication cycle (Table 1) . Previous analysis of RNA secondary structure across the Flavivirus genus led to the concept of promoter and enhancer elements that initiate assembly of the virus polymerase complex (16) (17) (18) 23, 27, 43, 44) . Enhancers were identified as RNA structures that individually produce only small biological effects on virus replication. However, the significance of enhancers as targets for the attenuation of flaviviruses to engineer live vaccines is evident from the example of dengue virus (24, 25) . Moreover, sequence and structural conservation of flavivirus enhancers is consistent with a role as key players in virus survival in the natural environment. We previously proposed that the cumulative action of several enhancer elements could contribute significantly to the overall rate of assembly of polymerase complexes, thereby enhancing virus survival across a range of natural hosts (17, 18, 23, 27, 43, 44) . In this respect, the presented experimental data indicate that SL6 belongs to the category of REEs, i.e. RNA structures that accelerate the replication of viruses (45) (46) (47) (48) (49) (50) (51) (52) (53) (54) (55) (56) . This eliminates the apparent contradictions between extremely high levels of SL6 conservation across divergent TBFV virus species and the redundancy of this element for the replication of laboratory-maintained TBEV strains (45) (46) (47) (48) (49) (50) (51) (52) (53) (54) (55) (56) . However, the specific mechanism by which SL6 functions to enhance virus replication remains to be elucidated. It has recently been demonstrated that a short but highly conserved RNA hairpin (sHP) localized in the 3 0 -UTR of DENV2 RNA regulates the transition from a circular (required for the initiation of RNA replication) to linear RNA form during the progress of viral RNA synthesis (57) . The SL6-like structure of MBFV is localized immediately downstream of the 5 0 -CYCL (i.e. within the capsid gene, Supplementary Figure S1A ) suggesting it could also contribute to genome circularization. It is possible that in accord with the 3 0 sHP (highly conserved throughout the genus Flavivirus), it contributes to the unpairing of the 5 0 -3 0 -CYCL panhandle, to promote RNA elongation on the linear template. In contrast, the 5 0 -CYCL of TBFV is mapped to the 5 0 -UTR (i.e. upstream of the capsid gene, Figure 1A ) and therefore other tentative functions of the TBFV SL6 are not excluded, such as enhancing virus translation, RNA replication or playing a role in regulation between these processes; the possibility of a kissing-loop enhancer of genome circularization was previously discussed (17, 18, 23, 27, 43, 44) . The C protein of flaviviruses is highly basic at the N-terminus, specifically binding virus genomic RNA during encapsidation and plausibly acting as an RNA chaperone as shown for other viruses (58) . The sequence of SL6 within the C coding region localizes to the junction of the positively charged domain and a following hydrophobic domain that interacts with the virus envelope proteins during assembly (42) . It is possible that additional synonymous codon flexibility may be accommodated in this region due to the requirement to conserve the charge or hydrophobic characteristics of the domain, rather than any specific amino acid sequence. Although our studies provide support for the REE role of SL6 in TBEV it is unclear if SL6-like structures of MBFV act similarly as functionally significant REE. However, the remarkable resemblance of the WNV SL6-like structure to TBFV SL6 suggests that it might serve a similar function, at least in one virus group. However, a final conclusion for the MBFV and also for the more distant NKV or PABV groups is not possible ahead of further functional studies. Being arboviruses, MBFVs and TBFVs are adapted for transmission between distantly related vertebrate hosts and invertebrate vectors. The requirement to adapt to different molecular environments might result in the evolution of enhancer elements essential for virus replication in one host while being redundant in another. This could explain the contradiction between strict conservation of the different flavivirus enhancers and their apparent redundancy in laboratory systems, which are largely based on mammalian cells (17, 18, 23, 27, 43, 44) . Mutations in SL6 described here have demonstrated its enhancer properties in mammalian cells and it will be interesting to evaluate SL6 enhancer activity in ticks, the major host for maintenance of the TBFV group in the environment (59) (60) (61) . In conclusion, bioinformatic analysis demonstrated the presence of a conserved RNA secondary structure in the C coding region of the divergent TBFV group. Disruption of this structure compromised virus replication implying an REE function for SL6. By homology with the TBFVs, SL6-like structures were observed in the genomes of some MBFVs and plausibly indicate a similar role as replication enhancers. Future studies using sub-genomic replicons will allow direct measurement of the influence of these sequences on RNA replication and on interaction with viral and host proteins.

Both TLR2 and TRIF Contribute to Interferon-β Production during Listeria Infection

Synthesis of interferon-β (IFN-β) is an innate response to cytoplasmic infection with bacterial pathogens. Our recent studies showed that Listeria monocytogenes limits immune detection and IFN-β synthesis via deacetylation of its peptidoglycan, which renders the bacterium resistant to lysozyme degradation. Here, we examined signaling requirements for the massive IFN-β production resulting from the infection of murine macrophages with a mutant strain of L. monocytogenes, ΔpgdA, which is unable to modify its peptidoglycan. We report the identification of unconventional signaling pathways to the IFN-β gene, requiring TLR2 and bacterial internalization. Induction of IFN-β was independent of the Mal/TIRAP adaptor protein but required TRIF and the transcription factors IRF3 and IRF7. These pathways were stimulated to a lesser degree by wild-type L. monocytogenes. They operated in both resident and inflammatory macrophages derived from the peritoneal cavity, but not in bone marrow-derived macrophages. The novelty of our findings thus lies in the first description of TLR2 and TRIF as two critical components leading to the induction of the IFN-β gene and in uncovering that individual macrophage populations adopt different strategies to link pathogen recognition signals to IFN-β gene expression.

Detection of microbial pathogens by pattern recognition receptors, such as Toll-like receptors (TLRs) triggers innate immune responses as a first line of defense against infections [1] [2] [3] . Pathogen-associated molecular patterns (PAMPs) such as bacterial cell walls and their structural components induce a vast variety of biological effects in host organisms. The innate response against infection with intracellular pathogens includes the synthesis of type I IFNs (IFN-I). Whereas this cytokine family generally protects against viruses, its impact on bacterial infections can be either detrimental or advantageous for the host organism [4] . Listeria monocytogenes is a bacterial pathogen which replicates in the cytoplasm of infected cells. Cytosolic pattern recognition receptors (PRRs) respond to cytosolic bacterial products and contribute to the induction of the innate immune response [5, 6] . Previous studies in bone marrow-derived macrophages (BMM) and epithelial cells show that in these cell types the synthesis of IFN-I in response to infection with L. monocytogenes is independent of TLRs and their adapters, relying exclusively on signals originating from cytosolic sensors [5] [6] [7] [8] . DNA as well as cyclic dinucleotides released from lysed bacteria were suggested to function as the relevant L. monocytogenes PAMPs [9] [10] [11] . Several cytosolic proteins with the ability to sense pathogen-derived nucleic acids have recently been described [11] [12] [13] [14] [15] [16] [17] [18] [19] . Cytosolic recognition of L. monocytogenes causes the activation of the serine/threonine kinase TBK1 and the phosphorylation of its substrate transcription factors IRF3 and IRF7 [7, 8] . Both IRF3 and IRF7 participate in the formation of an enhanceosome at the IFN-b promoter [20] . During uptake by host cells L. monocytogenes is exposed to plasma membrane and endosomal TLRs. Among these, TLR2 which recognizes lipotechoic acids and lipopeptides, contributes to the innate response against infection [21] [22] [23] . Reportedly, TLR2 signals through the interacting adapter proteins Mal/TIRAP and MyD88 and does not contribute to the synthesis of type I IFN in Listeria-infected BMM [7] [8] [9] . Signaling through TRIF, an adapter protein known to connect TLRs 3 and 4 with the IFN-I genes was similarly ruled out for Listeria-infected BMM [7] . In order to establish a successful infection, pathogens must survive host defense systems or else mitigate the activities of PRRs. Consequently, they have evolved to modify the structural components which normally trigger PRR responses. Bacterial PGN is a hetero-polymer consisting of alternating residues of b-1,4-linked N-acetylglucosamine and N-acetylmuramic acid to which a peptide chain is attached [24] . Interestingly, L. monocytogenes modifies its PGN, with fifty per cent of the muropeptide composition being N-deacetylated [25] . We previously reported that a PGN N-deacetylase gene, pgdA, is responsible for this modification [25] . PGN deacetylation confers resistance to the action of lysozyme, one of the most important and widespread antimicrobial agents of the innate defense system, thus preventing degradation and release of immunostimulants. A strain of L. monocytogenes mutated in its ability to alter its PGN, DpgdA, is sensitive to lysozyme and induces an enhanced IFN-b response in macrophages compared to the isogenic parental strain [25] . The aim of the present study was to decipher the signaling pathways involved in this response to DpgdA infection. We reveal that IFN-b production in peritoneal macrophages requires TLR2 signaling and the TRIF adapter protein. Listeria DpgdA mutant in a TLR2-dependent manner A L. monocytogenes pgdA mutant induced a much higher IFN-b response than the parental strain [25] . To definitively establish a role for the peptidoglycan deacetylase PgdA in the downregulation of IFN-b production, we complemented our original pgdA mutant with the wild-type gene and we measured IFN-b secretion of peptone elicited peritoneal macrophages (PEM) infected with wild-type EGDe, DpgdA and a complemented DpgdA strain (Fig. 1) . Inactivation of pgdA led to a strong induction of IFNb secretion in wild-type macrophages. In contrast, the complemented strain did not induce any massive IFN-b secretion, similar to wild-type EGDe. Thus, PgdA directly contributes to downregulation of IFN-b production. Consistent with our previous report measuring secretion of IFNb protein in PEM, IFN-b mRNA synthesis induced by L. monocytogenes infection of PEM required TLR2 ( Fig. 2A) , while TLR2-deficient BMM showed no impairment in their synthesis of IFN-b mRNA (Fig. 2B) . Moreover, IFN-b secretion was strongly reduced in tlr2 2/2 PEM infected with both the DpgdA mutant ( Fig. 2C ) and the complemented DpgdA strain (Fig. 2D) , definitively establishing the TLR2 dependence of IFN-b production. We next analyzed the pathways by which Listeria induces IFN-b. Our previous study and the above results strongly suggested the critical involvement of TLR2 [25] . TLR2 signaling depends on Mal/TIRAP and MyD88 adaptor proteins. We had previously shown that MyD88 contributed to full IFN-b induction by Listeria [25] . We then compared IFN-b production by wild-type and mal/ tirap 2/2 macrophages infected with EGDe or DpgdA (Fig. 3A) . Surprisingly, production of IFN-b was not decreased in infected macrophages deficient in Mal/TIRAP, indicating that the normal TLR2 adaptor Mal/TIRAP was not required for Listeria-mediated induction of IFN-b. The adapter TRIF is employed by TLRs 3 and 4 to signal through the TBK1-IRF3/7-IFN-b pathway. There is no previous evidence of an association or functional interaction between TRIF and TLR2. In spite of this, the link between TRIF and the IRF pathway on the one hand, and the unusual employment of TLR2 for signaling to the IFN-b gene in PEM on the other suggested the possibility of a role for TRIF. To test this hypothesis we compared induction of IFN-b expression in wild-type and trif 2/2 PEM or BMM infected with EGDe or DpgdA strains. IFN-b induction strongly decreased in TRIF-deficient macrophages infected with any of the two Listeria strains compared to wild-type PEM, showing the requirement for TRIF (Fig. 3B ). In contrast, BMM showed a TRIF-independent IFN-b production (Fig. S1 ). The PEM used in our studies are recruited to the peritoneal cavity by injection of the sterile irritant proteose peptone. Hence they differ from BMM not only regarding their anatomical location, but also their partially inflammatory character. To distinguish which of these differences was responsible for the TLR2 and TRIF signaling pathways, we examined IFN-b production by resident PEM. Figure 3C demonstrates a requirement for TLR2 and TRIF by the resident macrophage population. Thus, location to the peritoneal cavity rather than inflammatory character determines the difference in signaling to the IFN-b gene between BMM and PEM. To examine the role of TLR3, which uses TRIF to trigger IFNb synthesis, we compared induction of IFN-b in wild-type and tlr3 2/2 PEM infected with EGDe or DpgdA strains. IFN-b production was decreased in TLR3-deficient PEM infected with EGDe or DpgdA (Fig. 4A) . We also compared induction of IFN-b in wild-type and tlr4 2/2 PEM infected with EGDe or DpgdA strains, as TLR4 can mediate TRIF-dependent synthesis of IFN-b. In contrast to TLR3-deficient PEM, TLR4-deficient PEM did not show a decrease in IFN-b response to EGDe or DpgdA (Fig. 4B ). Thus, IFN-b induction in response to Listeria infection relies in part on TLR3 and does not require TLR4. Induction of IFN-b via TLR2 is no longer an exception. It has recently been shown that vaccinia virus-induced IFN-b production was dependent on TLR2 signaling and it was reported that this was occuring from late endosomes [26] . To investigate if an intracellular localization was also required in the case of Listeria, we pretreated cells with cytochalasin D to prevent internalization and measured IFN-b secretion by macrophages infected with EGDe or the DpgdA mutant (Fig. 5A ). In both cases, IFN-b induction was strongly reduced. Thus, internalization is critical for Listeriamediated IFN-b production. We also used dynasore, a dynamin inhibitor and chloroquine, which inhibits endosome acidification, and measured IFN-b induction in macrophages infected with EGDe or the DpgdA mutant ( Fig. 5B -C). IFN-b synthesis was strongly diminished by both dynasore and chloroquine treatments. Together, these results suggest that the TLR2-dependent IFN-b induction is triggered intracellularly. In BMM rapid synthesis of IFN-b is entirely dependent on IRF3, but not on IRF7, whereas in bone marrow-derived myeloid DC IFN-b synthesis requires both IRF3 and IRF7 [27] . We investigated the role of IRF3 and IRF7 in the production of IFN-b by PEM. To this end we infected irf3 2/2 and irf7 2/2 macrophages with EGDe or the DpgdA strains. Inactivation of IRF3 totally abrogated IFN-b mRNA induction in response to both strains (Fig. 6 ). IFN-b induction in IRF7-deficient macrophages was also strongly affected highlighting the important role of both transcription factors in response to Listeria infection (Fig. 6 ). PEM thus resemble bone marrow-derived myeloid DC, not BMM, in relation to their IRF requirement for Listeria-mediated IFN-b synthesis. In addition to IRF3/7, NFkB contributes to the formation of the IFN-b enhanceosome [20, 28] . We therefore examined the involvement of the NFkB pathway by measuring induced synthesis of an NFkB-dependent mRNA. IkB is an NFkB-dependent gene and thus a read-out for NFkB activation in response to Listeria infection. We measured the induction of IkB expression in PEM infected with EGDe or the DpgdA mutant. Both strains induced IkB expression and this required internalization as treatment with dynasore reduced the level of IkB induction (Fig. 7A) . Degradation of the IkB protein was examined in PEM infected with EGDe by immunoblot using anti-IkB antibodies. IkB level was reduced rapidly after infection of wild-type PEM (Fig. S2A ). In contrast, IkB degradation was not observed in tlr2 2/2 PEM infected with Listeria (Fig. S2B ). Infection of wild-type, tlr2 2/2 and trif 2/2 macrophages with EGDe or DpgdA showed that both TLR2 and the adaptor were required for full induction of IkB mRNA in response to EGDe and DpgdA strains (Fig. 7B) . These results suggest that TLR2 and TRIF contribute to NFkB activation. The comparison between EGDe and DpgdA strains showed that both caused similar magnitudes of IkB mRNA synthesis. Thus, the activation of NFkB by Listeria is independent of PgdA, suggesting that the increased IFN-b production after infection with DpgdA relies on activation of other transcription factors such as IRFs. Nucleic acids released intracellularly are critical for IFN-b induction TLR2 or TRIF deficiency strongly reduced, but did not completely shut off IFN-b synthesis. This suggested a potential contribution of intracellular, nucleic acid-dependent pathways to IFN-b synthesis, particularly after infection with DpgdA. We therefore examined whether these pathways are able to signal in PEM. Since inactivation of PgdA increases Listeria sensitivity to peptidoglycan-targeting antimicrobials such as lysozyme, and thus induces bacterial degradation, we measured the DNA and RNA released by EGDe and DpgdA strains following lysozyme exposure. As expected, DpgdA released significantly higher amounts of DNA and RNA than wild-type and complemented DpgdA strains, raising the possibility that both DNA and RNA could be involved in IFNb production (Fig. 8A) . We thus measured IFN-b induction in THP1 macrophages transfected with Listeria DNA, either undigested or treated with DNase. Intact but not DNase-treated DNA significantly induced IFN-b (Fig. 8B ). Macrophages were then transfected with lysozyme-digested EGDe or DpgdA, either untreated or digested with DNase. Treatment with DNase significantly reduced IFN-b production (Fig. 8C) . Taken together, these results show that Listeria DNA can induce IFN-b, strongly indicating that destruction of DpgdA bacteria intracellularly activates DNA sensors. We had recently reported that a PGN modification involving a N-deacetylase gene, pgdA, was playing a key role in L. monocytogenes virulence [25] . A DpgdA strain of L. monocytogenes which is unable to modify its PGN, was shown to be extremely sensitive to the bacteriolytic activity of lysozyme, normally found within macrophage vacuoles and its virulence was strongly attenuated [25] . Furthermore, this mutant induced a much higher TLR2dependent IFN-b response than the parental strain [25] . We hypothesised that this unconventional IFN-b response induced by the pgdA mutant was due to an enhanced accessibility of bacterial cell wall components to TLR2. Here we have shown that IFN-b production requires bacterial internalization and is triggered by Mal/TIRAP-independent pathways which involve TLR2, TRIF, IRF3 and IRF7. It was surprising to see a role for TLR2, as, based on results in BMM and epithelial cells, type I IFNs production is usually not known to result from TLR2 signaling [5] [6] [7] [8] . Classical TLR2 signaling leads to NF-kB-dependent production of inflammatory cytokines [21] . However, in support of an unconventional role for TLR2, recent studies reported roles for TLR2-dependent induction of IFN-b in response to vaccinia virus or synthetic ligands [26, 29] . In the vaccinia virus study, a specific inflammatory monocyte population -Ly6C hi -was shown to be the source of IFNb [26] . In the present study we show that TLR2-dependent IFN-b synthesis is a property of both resident and recruited inflammatory PEM. Furthermore, the two previous studies documented that TLR2 activation of type I IFN responses to TLR ligands occurs within intracellular compartments, and that TLR2 signals from the phagosome in response to viral infection or synthetic TLR2 ligands [26, 29] . These results challenged the view that TLR2 signals solely from the plasma membrane. In our experiments, pretreatment of PEM with either cytochalasin D, an inhibitor of actin polymerization and thus internalization, dynasore, an inhibitor of the endocytic effector dynamin, or chloroquine, which inhibits endosome acidification [30, 31] , significantly impaired the induction of IFN-b following Listeria infection, strongly suggesting that phagocytosis of L. monocytogenes and intracellular location of TLR2 trigger this response. These observations also correlate with our early hypothesis that the inflammatory response induced by DpgdA is due to an enhanced release or accessibility of bacterial cell wall components to TLR2. Induction of the IFN-b gene was independent of the TLR adapter Mal/TIRAP, but, unexpectedly required the TLR3/4 adapter TRIF. Francisella tularensis has recently been shown to signal through TLR2 from the phagosome in a Mal/TIRAP independent manner [32] , and it was shown that Mal/TIRAP is dispensable in TLR2 signaling at high concentrations of ligands [33] . Thus our study reinforces the view that TLR2 can act independently from Mal/TIRAP. In addition our report suggests a synergy between a TLR2 pathway and TRIF, an adapter previously known to trigger the synthesis of pro-inflammatory cytokines and type I IFNs upon engagement of TLR3 and TLR4. TLR3 is known to bind viral dsRNA to induce secretion of type I IFN and lead to control of viral infections [3, 34, 35] . To our knowledge Chlamydia muridarum is the only bacterium reported to induce a TLR3-dependent IFN-b response specifically in murine oviduct epithelial cells [36] . We tested whether the dual requirement for TLR2 and TRIF resulted from a functional or physical interaction between TLR2 and TLR3. In fact, IFN-b production was reduced in TLR3-deficient macrophages, but significantly less so than in trif2/2 PEM. Therefore, there is no evidence for a putative TLR2/TLR3 interaction. Another possibility to incorporate TRIF into the pathway stimulated in PEM by Listeria would be a cooperation of TLR2 and TLR4. This was ruled out by showing that Listeria-infected tlr4 2/2 PEM produced a similar amount of IFN-b as their wild-type counterparts. TRIF could possibly orchestrate an additional pathway. Along these lines, TRIF has recently been shown to be required for IFN-b synthesis by dendritic cells upon activation of the cytosolic receptor complex DDX1/DDX21/DDX36 by viral RNA [19] . Engagement of TLRs by various microbe-associated molecular patterns induces activation and translocation to the nucleus of NF-kB, IRF3, IRF7 and/or activator protein-1 (AP-1), which collaborate to induce transcription of type I IFNs [37] . We addressed the role of these transcriptional activators in the IFN-b response to wild-type Listeria and DpgdA, and revealed that inactivation of IRF3 totally abrogated this response to both strains while IFN-b induction was significantly but not totally impaired in IRF7-deficient macrophages, indicating that both of these transcription factors are required for induction of IFN-b following infection with L. monocytogenes. We also assessed the involvement of NF-kB in this response using induction of the IkB gene as a readout. We observed an induction of IkB expression in macrophages which was similar after infection with EGDe or DpgdA. Thus, activation of NF-kB by Listeria is independent of PgdA, strongly suggesting that the elevated IFN-b production by the DpgdA mutant mostly relies on IRF3. The increased IFN-b response to the DpgdA strain probably results from the fact that within the phagosome, its lysozymesensitive cell wall is degraded, releasing PAMPs able to interact with TLR2 and other PRRs, including cytoplasmic ones. As recent studies have highlighted novel DNA-sensing pathways in the induction of type I IFNs [9, [14] [15] [16] [17] 38] , we thus also investigated the involvement of bacterial nucleic acids in the IFN-b induction, Firstly, we showed that inactivation of PgdA, which confers a higher susceptibility to lysozyme, leads to increased release of DNA. We then showed that DNA from L. monocytogenes can induce IFN-b expression in PEM, suggesting that this macrophage population employs cytoplasmic nucleic acid sensing similar to macrophages or macrophage lines derived from different anatomical locations [9, 38] . Which -if any-of the recently described nucleic acid sensors are used by PEM for the recognition of Listeria DNA remains subject to future investigation. Nevertheless, other bacterial components could participate in IFN-b production upon infection with the DpgdA mutant. For example, the second messenger molecule cyclic diadenosine monophosphate (c-di-AMP), was shown to be secreted by Listeria multidrug efflux pumps triggering type I IFN response [10] and could be involved in the process. In conclusion, this study describes a novel mechanism leading to induction of type I IFNs in which intracellular sensing plays an important role, ultimately showing how these different recognition pathways can synergise to induce innate immune responses which are required to control infection. In this regard cooperation between TLR2 and TRIF may reflect the need for convergence of the NF-kB and IRF pathways at the IFN-b promoter, with TLR2 being responsible mainly for NF-kB activation and TRIF being instrumental for activation of IRF3 and IRF7. By employing the strategy of PGN modification, L. monocytogenes can avoid immune detection by TLR and evade the innate immune response, thus enabling the infectious process to occur. It is important to recall that pgdA orthologs are found in other pathogenic bacteria, such as Streptococcus pneumoniae, Bacillus cereus, Bacillus anthracis and Helicobacter pylori, strongly suggesting that PGN N-deacetylation is a general mechanism evolved by microbes to escape from pattern recognition receptor-mediated immune recognition [39] [40] [41] [42] . Bacterial strains and growth conditions L. monocytogenes EGDe (BUG1600, ATCC BAA-679), L. monocytogenes isogenic mutant DpgdA (BUG2288, [25] ) and L. monocytogenes DpgdA complemented strain (BUG2382) were grown in brain heart infusion (BHI, Oxoid), aerobically at 37uC and 200 rpm. A DNA fragment containing the pgdA gene (lmo0415) and its promoter was generated by PCR using oligonucleotides lmo0415-1 (59-AAGGATCCCACAATATGTTAGTTTTCAGGGG-39) and lmo0415-2 (59-AAGGATCCTTATTTCACCATTCTT-GAATCTG-39). The fragment was integrated into pCR-Blunt-II-TOPO (Invitrogen) and the construct was verified by sequencing. After digestion of the construct by BamHI, the fragment was purified on agarose gel and cloned into the integrative vector pPL2 [43] , previously digested by BamHI, constructing pOD98. The pOD98 was electroporated into DpgdA at 2,500 V, 200 V and 25 mF. Transformants were selected at 37uC on BHI agar containing chloramphenicol (7 mg/mL). The presence of the pgdA gene in the complemented strain was confirmed by PCR using oligonucleotides lmo0415-1 and lmo0415-2. Mice were used for obtaining peptone-elicited peritoneal macrophages, resident peritoneal macrophages and bone marrow-derived macrophages. Animal experiments were performed in accordance with protocols approved by the Animal Experimentation Ethics Committee of the Institut Pasteur (permit #03-49) and following Austrian law in accordance with protocols approved by the Ethics Committee of the University of Veterinary Medicine, Vienna (#GZ680 205/67-BrGt/2003). Isolation and culture of murine peptone-elicited peritoneal macrophages (PEM) PEM were isolated from 7 to 10 week-old C57BL/6J and genetically-matched tlr2 2/2 , tlr3 2/2 , tlr4 2/2 , mal/tirap 2/2 , trif 2/2 , irf3 2/2 and irf7 2/2 mice as previously described [44] . The percentage of macrophages was determined by flow cytometry using CD11b (1:100, eBiosciences) and F4/80 (1:100, eBiosciences) antibodies. More than 90% of the cells were macrophages. PEM were seeded onto 6-well plates at a concentration of 2610 6 cells per well in DMEM (PAA) supplemented with 10% FCS, 10% L929 conditioned medium (LCM) and 1% penicillin-streptomycin or RPMI-1640 (Gibco) supplemented with 10% FBS and 1% penicillin-streptomycin. Resident macrophages were isolated from 6 to 8 week-old C57BL/6J and genetically-matched tlr2 2/2 , trif 2/2 mice by washing the peritoneum twice with 10 mL DMEM (PAA) supplemented with 10% FBS, 10% LCM and 1% penicillinstreptomycin. Harvested cells were centrifuged at 300 g for 5 minutes and resuspended in complete medium. The percentage of macrophages was determined by flow cytometry analysis as above. Cells were seeded onto 6-well plates (Nunc) at a concentration of 2610 6 cells per well. Tibia and femur from 6 to 8 week-old C57BL/6J and genetically-matched tlr2 2/2 , trif 2/2 mice were collected in ice cold PBS. Bones were sterilized with 70% ethanol and flushed with a 25-G needle using cold DMEM supplemented with 10% FCS, 10% LCM and 1% penicillin-streptomycin. Cells were seeded onto 6-well plates (Nunc) at a concentration of 10 6 cells per well and incubated at 37uC with 5% CO 2 . After 4 days, complete medium was added and cells were split at a ratio of 1:2. After 8 days, macrophages were fully differentiated. Human acute monocytic leukemia THP-1 cells (ATCC TIB202) were maintained in RPMI-1640 supplemented with 10% FBS and 1% penicillin-streptomycin. Cells were seeded onto a 24-well plate at a concentration of 4610 5 cells per well in antibiotic-free media supplemented with 12.5 ng/mL phorbol myristate acetate and incubated for 24 h at 37uC with 5% CO 2 . Differentiation was determined to be successful upon formation of a confluent adherent monolayer. HEK-blue type I IFN cells (Invivogen) were grown in DMEM supplemented with 10% FBS and 1% penicillin-streptomycin. Cells were seeded at a concentration of 5.6610 4 cells per well onto a 96-well plate. For cytokine analysis, macrophages were infected with Listeria strains at MOI 10:1, centrifuged at 300 g for 2 min and incubated at 37uC for 15 min. Following phagocytosis, monolayers were washed twice followed by incubation in RPMI-1640 supplemented with 10% fetal bovine serum (FBS) and gentamicin (20 mg/mL). Supernatants were collected at various time points, for detection of IFN-b by ELISA. For transcript analysis, macrophages were infected with Listeria strains at MOI 20:1 and incubated at 37uC for 1 h to allow phagocytosis. Monolayers were washed and incubated in DMEM supplemented with 10% FCS and gentamicin (5 mg/mL). After 2 h, medium was changed to DMEM supplemented with 10% FCS and gentamicin (1 mg/mL). Cells were lysed at various time points and RNA collected for qPCR analysis. For inhibition of bacterial internalization, cell monolayers were pretreated either for 2 h with 100 mM cytochalasin-D (Sigma-Aldrich), or 30 min with 80 mM dynasore (Sigma-Aldrich) or 30 min with 100 mM chloroquine (Sigma-Aldrich) prior to infection assays. Listeria were grown overnight in BHI at 37uC and cultures were centrifuged at 8000 g for 5 min. Bacterial pellets were resuspended in 75 mg/mL lysozyme and incubated at 37uC for 1 h. DNA was then extracted using the DNeasy blood and tissue kit (Qiagen) and quantified by spectrophotometry (Nanodrop). For transfection assays, THP-1 macrophages were transfected with 200 ng/mL DNA with 2% lipofectamine 2000 (Invitrogen) and incubated for 24 h. Following incubation, supernatants were collected for IFN-b analysis. For pretreatment of DNA with DNase, DNase was added at final concentration of 100 mg/mL for 45 min at 37uC. Bacterial cultures were treated with 10 mg/mL lysozyme, a concentration leading to lysis of DpgdA but not EGDe, and incubated at 37uC and 200 rpm for 1 h. Following lysozyme treatment, lysed bacterial cultures were centrifuged at 5000 rpm during 10 min. Two types of experiments were performed on supernatants. First, nucleic acid release was quantified. DNA was purified using the Qiagen DNeasy blood and tissue kit omitting lysis steps and quantified by spectrophotometry (Nanodrop). RNA was purified using Qiagen RNeasy kit and quantified by spectrophotometry (Nanodrop). Data shown are representatives of at least three independent experiments. Second, 100 mL of each supernatants were treated by DNase during 30 min at 37uC. Enzymes were inactivated and treated-or untreated-supernatants were transfected in PEM. 8 h after transfection, supernatants of cells were recovered and the IFNb was quantified. Detection of type I IFN by ELISA and HEK-blue type I IFN cell assay Murine IFN-b production was detected in macrophage supernatants by ELISA according to the manufacturer's procedure (PBL Biomedical Laboratories). For the HEK-blue type I IFN assay, supernatant from THP-1 macrophage assays was collected and 20 mL added onto HEK-blue type I IFN cells plated in 96-well plates, which were incubated at 37uC overnight. Supernatant from HEK-blue cells was collected and 40 mL added to 160 mL of Quanti-blue reagent (Invivogen) for 20 min at 37uC. The colorimetric reaction was measured at 625 nm on a plate reader. Data was normalised against absorbance for the untreated cells and plotted as relative fold increases. Data shown are representatives of at least three independent experiments. PEM from WT or tlr2 2/2 C57BL/6J mice were infected with EGDe. Cells were lysed 0, 0.5, 1, 1.5, 2, 2.5, or 3 h post-infection. IkB and tubulin were detected in lysates by immunoblotting using anti-IkB (Santa Cruz, 1:100) and anti-a-tubulin (Sigma, 1:5000) antibodies. RNA preparation was performed using NucleoSpin RNA II kit (Macherey-Nagel) according to the manufacturer's instructions. Quantitative real-time PCR was performed on a Mastercycler EP realplex S (Eppendorf). Primers for HPRT (housekeeping gene control), IFNb and IkBa mRNA expression were as follows: HPRT forward GTTGGATACAGGCCAGACTTTGTTG, HPRT reverse GAGGGTAGGCTGGCCTATTGGCT, IFNb forward 59-TCAGAATGAGTGGTGGTTGC-39, IFNb reverse 59-GACCTTTCAAATGCAGTAGATTCA-39; IkBa forward 59-GCAATTTCTGGCTGGTGGG-39, IkBa reverse 59GATCC-GCCAGGTGAAGGG-39. Data shown are representatives of at least three independent experiments. Results are expressed as means of at least three values, with error bars representing standard deviations. Student's t tests were performed to determine statistical significance where * indicates P,0.05, ** indicates P,0.01 and *** indicates P,0.0001. Figure S1 TRIF is not required for IFN-b response to Listeria in bone marrow macrophages. BMM from C57BL/6J or trif 2/2 mice were infected with the parental EGDe strain (black bars) or the DpgdA mutant (grey bars). After 4 h of infection, IFN-b induction was measured by qRT-PCR. Data are mean 6 SD (NS, non significant, n = 3). (EPS) Figure S2 TLR2 is required for optimal activation of NF-kB. (A) PEM from WT C57BL/6J mice were infected with EGDe. Cells were lysed 0, 0.5, 1, 1.5, 2, 2.5, or 3 h post-infection. Activation of NF-kB was measured by determination of IkB degradation relative to tubulin following immunodetection. (B) PEM from tlr2 2/2 mice were infected with EGDe. Cells were lysed 0, 0.5, 1, 1.5, 2, 2.5, or 3 h post-infection. Activation of NF-kB was measured by determination of IkB degradation relative to tubulin following immunodetection. (EPS)

Protein Reporter Bioassay Systems for the Phenotypic Screening of Candidate Drugs: A Mouse Platform for Anti-Aging Drug Screening

Recent drug discovery efforts have utilized high throughput screening (HTS) of large chemical libraries to identify compounds that modify the activity of discrete molecular targets. The molecular target approach to drug screening is widely used in the pharmaceutical and biotechnology industries, because of the amount of knowledge now available regarding protein structure that has been obtained by computer simulation. The molecular target approach requires that the structure of target molecules, and an understanding of their physiological functions, is known. This approach to drug discovery may, however, limit the identification of novel drugs. As an alternative, the phenotypic- or pathway-screening approach to drug discovery is gaining popularity, particularly in the academic sector. This approach not only provides the opportunity to identify promising drug candidates, but also enables novel information regarding biological pathways to be unveiled. Reporter assays are a powerful tool for the phenotypic screening of compound libraries. Of the various reporter genes that can be used in such assays, those encoding secreted proteins enable the screening of hit molecules in both living cells and animals. Cell- and animal-based screens enable simultaneous evaluation of drug metabolism or toxicity with biological activity. Therefore, drug candidates identified in these screens may have increased biological efficacy and a lower risk of side effects in humans. In this article, we review the reporter bioassay systems available for phenotypic drug discovery.

have increased biological efficacy and a lower risk of side effects in humans. In this article, we review the reporter bioassay systems available for phenotypic drug discovery. Keywords: drug development; high throughput screening; reporter mice; age-related disorders Reporter assays have been widely used in biological research [1] [2] [3] . Such assays are a powerful means of investigating the signaling pathways involved in various biological functions induced by the activation of specific transcription factors. The reporter construct usually consists of a promoter (enhancer) that drives the transcription of a gene that is easily detected by its activity or expression in the assay system. The promoter region binds transcription factors that are activated in response to the stimulation of an upstream, receptor-mediated or receptor-independent signaling cascade. Binding of the transcription factor to the promoter results in the induction of reporter gene (and subsequent reporter protein) expression in response to signaling pathway activation by external stimuli, including drug candidates. It is important that the exogenous protein encoded by the reporter gene has unique enzymatic activity or another property that distinguishes it from endogenous proteins [4] . The activity of the reporter protein is typically assayed by photometry, colorimetry or fluorometry. Recently, detection systems based on bioluminescence, chemiluminescence and fluorescence have become widely used because of their signal detection sensitivity and ease of use. Improved detection equipment, such as charge-coupled device (CCD) cameras, has also enabled the development of convenient reporter assay systems for drug screening. There are two major types of reporters, i.e., intracellular and extracellular reporters. Intracellular reporter gene products are expressed and retained inside the cells. On the other hand, extracellular reporter genes encode proteins that are secreted into the medium of cultured cells or the blood stream in animals. Some intra-and extracellular reporter gene assays may detect the reporter protein's activity without killing the cells or animals in which the assay is performed. This allows time-course experimentation by sampling of the medium of cultured cells, or the blood plasma of research subjects. As such, report bioassays are potentially suitable for high throughput screening (HTS) of small molecule drug candidates. Bioassays are used to determine the concentration or biological activity of molecules such as hormones, growth factors, and enzymes. Moreover, they can be used for measuring the effects of candidate drugs on an organism, cultured cells, or recombinant receptors by comparison to gold standard molecules with known target activity. The present review provides an overview of the basic properties, efficacies and limitations of protein reporter assays, with an emphasis on their application as bioassays for drug screening. Chloramphenicol O-acetyltransferase (CAT) has, until recently, been widely used as a reporter protein [5] . CAT is a bacterial enzyme that is not expressed in eukaryotes, and it therefore provides a highly specific detection system. CAT catalyzes the reaction between acetyl-CoA and chloramphenicol to form CoA. The CAT assay has been widely used but is difficult to adapt to HTS platforms because its activity is not easy to detect directly in cells. β-Galactosidase (β-Gal) is derived from the bacterial lacZ gene and is widely used as an intracellular reporter gene. There are a number of detection options depending on which substrate is used. However, β-Gal is affected by endogenous enzymatic activity in most mammalian cells [4] . Therefore, it is important to distinguish between the endogenous and exogenous (reporter) enzymes in the assay system. This usually requires the pH of the assay mixture to be adjusted, although this may not completely eliminate endogenous enzyme activity. Therefore, the utility of β-Gal for HTS assays is limited. Genes encoding luciferase enzymes have been cloned from several species, including firefly and sea pansy [1] . Firefly luciferase catalyzes the oxidation of firefly luciferin, which produces a short-lived flash of light that decays within a few seconds. However, the sensitivity of the luciferase assay is significantly higher than the CAT assay [6] . The high sensitivity of luciferase-based bioluminescence assays is better suited to HTS platforms than CAT or β-Gal assays. Moreover, various luciferase substrates are available that have a long half-life and provide linear results over eight orders of magnitude [4] . The enzyme has also been used in a dual-luciferase reporter assay system, which allows both firefly and sea pansy luciferase reactions to be monitored independently in the same cell [7] . This enables the activation of two different signaling pathways to be measured simultaneously in response to the same stimulus. Rather than being a reporter enzyme, green fluorescent protein (GFP) (and its engineered derivatives) is particularly suited to bioimaging assays in living cells or live animals. GFP is produced by the jellyfish, Aequorea victoria. Upon excitation by UV light, GFP emits green light with an emission maximum at 509 nm [4] . The intensity of light emissions has been improved by protein engineering in the fluorophore region of the protein. GFP-based reporter assays provide a distinct advantage because they do not require cell permeabilization or the addition of exogenous substrates. Therefore, these proteins are suitable for drug screening in HTS assays. Recent improvements in GFP detection technologies allow changes in the localization of GFP-tagged proteins, in response to small molecules, to be detected at the subcellular level in living cells cultured in 384-well plates. Therefore, GFP is suitable as both a reporter of transcriptional activation, and of functional protein translocation in the cell, in response to various stimuli. The secreted alkaline phosphatase (SEAP) reporter system has been used to investigate the activity of known or putative promoter/enhancer elements [8] . As a reporter, SEAP has several important advantages over other reporters, such as high sensitivity and specificity both in vitro and in vivo [9] [10] [11] [12] . SEAP is a C-terminal truncated mutant of the membrane-anchored, placental alkaline phosphatase [13] . Removal of the anchoring domain results in secretion of the enzyme into the culture medium or blood stream, thus enabling repeated sampling from living cells or animals. Expression of the gene can be detected using chemiluminometry [14] . The secreted form of alkaline phosphatase is extremely heat stable, and its activity can be detected following the inactivation of endogenous alkaline phosphatases by heating. Therefore, it is easy to distinguish between exogenous and endogenous SEAP activity. SEAP assays can also be coupled to a luciferase reaction by incubation with the substrate, firefly D-luciferin-O-phosphate. The product of the first reaction, luciferin, then becomes a substrate for luciferase. This assay protocol improves the sensitivity of the reporter assay system, thus enabling its broader application [15] . A recently-identified secreted form of luciferase also serves as a useful reporter for HTS assays. Luciferase from the marine copepod, Metridia longa, has unique secretary features that make it suitable for extracellular reporter assays. Markova et al. reported the substrate specificity of Metridia luciferase (MetLuc) and demonstrated its suitability for use as a reporter for monitoring gene expression [16] . In common with SEAP, secreted MetLuc is well suited to monitoring gene expression in time-course reporter assays. Both SEAP and MetLuc are currently available in a commercial dual reporter system that has been successfully applied in drug screening research [17] [18] [19] [20] . Reporter assay systems are suitable for the multi-parameter phenotype screening of drug candidates [21] . In contrast to the single molecular target approach, cell-based reporter assays reflect the complexity of the living organism. Therefore, they are becoming widely used to screen the effects of small molecules on signaling pathways [22] . However, drug discovery by HTS in cell-based assays presents its own difficulties. For instance, isolation of a hit compound often does not suggest a single, particular molecular target, preventing further optimization of hit compounds by medicinal chemistry [23] . This has led to cell-based screens being termed 'black-box' screens [21] . However, in most cases, target pathways may be predicted and investigated further. Nonetheless, drug companies prefer the well-characterized molecular target approach rather than phenotypic screening to accelerate the early phase of drug development. In fact, cell-based screens tend to identify more false positives than molecular target screens, presumably because of the existence of many potential targets in the biological pathway of interest [24] . Nonetheless, hit compounds from molecular target screens can fail to generate the expected effect on cells or have undesirable side effects that are potentially harmful to humans. In the context of the ongoing paradigm shift towards pathway-driven drug discovery, particularly in the academic field, it becomes increasingly important to accomplish promising candidate isolation earlier in the drug discovery pipeline and with higher throughput. Currently, many large pharmaceutical companies have established open innovation research programs through collaboration with non-profit research organizations to overcome these problems. Academic research teams typically utilize cell-based reporter assay systems for primary drug screening. Furthermore, research is also being conducted using living animals to monitor the more complex dynamics of the hit compounds identified by screening. Animal-based reporter assays using reporter mice are a potentially powerful tool for the identification of drug candidates. To our knowledge, the ERE-Luc (estrogen responsive element-luciferase gene) mouse is the first example of a reporter animal to be used for drug development [25] , although several studies with simple reporter mice have been documented previously [26] [27] [28] . The ERE-Luc model is a unique reporter animal for drug discovery that was developed to identify ligands that bind estrogen receptors (ERs). Because ERs are ubiquitously expressed in mammals, this mouse model was required to be generally applicable to the activity of ERs [29] . Validation of the animal model is important to establish whether the reporter activity correlates with receptor activation through the targeted pathways. Nonetheless, the ERE-Luc mouse demonstrated the existence of unliganded activation of ERs [30] . Thus, animal-based reporter assays may provide new insights into the physiology of target pathways. Indeed, the use of the reporter mice revealed differential mechanisms of ER activation in reproductive and non-reproductive tissues [31] . These results indicate that reporter mice are useful for biological research. Precise toxicology studies may also be performed by crossing another mouse model, such as the humanized liver metabolism model, with reporter mice [32] . Such studies might require that the high costs of compound provision are met, and that animal welfare conditions are appropriate. Table 1 summarizes the properties of reporter proteins. A A: suitable, B: applicable, C: not applicable. 1 Time-course assay is suitable for extra-cellular reporters compared to intra-cellular reporters. 2 An in vivo imaging system is needed to detect luciferase and GFP. 3 A tissue specific transgenic mouse is needed to detect specific tissue expression of extra-cellular reporters. 4 Chemiluminescence is more sensitive than fluorescence. CAT: Chloramphenicol acetyltransferase; β-Gal: β-Galactosidase; GFP: green fluorescent protein; SEAP: secreted alkaline phosphatase; MetLuc: Metridia longa luciferase; HTS: high throughput screening; CCD: charge-coupled device. Calorie restriction (CR) extends the lifespan of many organisms [33] . The identification of CR mimetic compounds may therefore enable therapeutic retardation of the onset of various age-related disorders, including metabolic syndrome, cancers, neurodegenerative diseases, and other rare disorders [34] . The development of CR mimetics would therefore fulfill unmet medical needs. Although SIRT1-activating compounds (STACs) may offer a promising approach to the development of CR mimetics, a novel approach to finding potential CR mimetics is required [9] . For this purpose, we developed a cell-and mouse-based screening system using an extracellular reporter assay [10] . Using microarray analysis, we have identified putative regulatory elements in pro-longevity genes (dwarfism and CR-response elements: DFCR-RE) [9, 35] . Using one of the DFCR-REs, we have developed a cell-and animal-based reporter screening system, named CRISP: the CR-Imitating anti-aging agent Screening Platform. Because the DFCR-REs were isolated from long-lived mice, the unidentified transcriptional regulators of this sequence may be beneficial for longevity and delay the onset of age-related diseases. We found that reporter construct transgenic mice (CRISP mice) showed increased reporter activity upon CR ( [9] and our unpublished data). Hence, compounds that activate this reporter may be CR mimetic pro-longevity drug candidates. As revealed in the ERE-Luc mice, CRISP mice may also reveal novel longevity signaling pathways and target molecules. The activity of candidate compounds can be easily studied on cells by their addition to culture medium, or in mice by their addition to food and water. In addition to isolating the activators of pro-longevity elements, our CRISP method may also identify inhibitors or negative regulators of these elements. In this respect, CRISP is advantageous compared to single molecular target screening. Therefore, a two-step screen in which a primary HTS in CRISP cells is used to identify hit compounds, followed by a secondary screen to verify the activity, metabolism and toxicology of the hits in CRISP mice, is a potentially powerful tool for identifying promising novel drug candidates. Current reporter assay technologies enabled us to perform high throughput screening of chemical libraries. The isolation of candidate molecules is dependent on the choice of reporter protein and the construction of efficient cell-or animal-based assays. The identification of CR mimetics is attractive for the development of anti-aging therapeutics. While CR mimetics may not be a 'cure-all' for ageassociated disorders, they may suppress and/or delay the progression of various kinds of refractory diseases. Importantly, pharmaceutical companies do not have a high incentive to develop drugs for rare diseases. Therefore, academic laboratories are perfectly placed to develop drugs for rare disorders by using phenotypic cell-or animal-based reporter assay systems. Although candidate CR mimetics identified in such systems must still be validated to show that their beneficial effects are similar to those provided by CR, CR mimetics offer great potential to solve the unmet medical needs of rare diseases and common age-related disorders. We have focused on bioassay (protein reporter assay) screening systems in this review, but biosensor technologies (analytical devices consisting of biological components) are also important for drug discovery [36] . Therefore, the development of biosensors to screen CR mimetics will also be an important approach to future drug discovery efforts.

Suppression of Adenosine-Activated Chloride Transport by Ethanol in Airway Epithelia

Alcohol abuse is associated with increased lung infections. Molecular understanding of the underlying mechanisms is not complete. Airway epithelial ion transport regulates the homeostasis of airway surface liquid, essential for airway mucosal immunity and lung host defense. Here, air-liquid interface cultures of Calu-3 epithelial cells were basolaterally exposed to physiologically relevant concentrations of ethanol (0, 25, 50 and 100 mM) for 24 hours and adenosine-stimulated ion transport was measured by Ussing chamber. The ethanol exposure reduced the epithelial short-circuit currents (I(SC)) in a dose-dependent manner. The ion currents activated by adenosine were chloride conductance mediated by cystic fibrosis transmembrane conductance regulator (CFTR), a cAMP-activated chloride channel. Alloxazine, a specific inhibitor for A(2B) adenosine receptor (A(2B)AR), largely abolished the adenosine-stimulated chloride transport, suggesting that A(2B)AR is a major receptor responsible for regulating the chloride transport of the cells. Ethanol significantly reduced intracellular cAMP production upon adenosine stimulation. Moreover, ethanol-suppression of the chloride secretion was able to be restored by cAMP analogs or by inhibitors to block cAMP degradation. These results imply that ethanol exposure dysregulates CFTR-mediated chloride transport in airways by suppression of adenosine-A(2B)AR-cAMP signaling pathway, which might contribute to alcohol-associated lung infections.

Alcohol abuse is a risk factor for pulmonary infections. It is not fully understood how alcohol exposure compromises the lung host defense. Previous studies suggest that multiple pathophysiological mechanisms may be involved [1, 2, 3] . Airway mucosal immunity and mucociliary clearance are the two primary host defense mechanisms, which take place in a thin layer of liquid on the top of airway epithelia, known as airway surface liquid (ASL). ASL, composed of a gel-like mucus layer and a sol-like periciliary liquid layer [4, 5, 6] , is the ''battlefield'' for pulmonary infection and immunity. The viscous mucous blanket traps inhaled microorganisms and particles to restrict their spreading in the lung. In contrast, the watery periciliary liquid (PCL) underneath pools antimicrobial substances, antibodies, cytokines, chemokines and other immune modulators [7, 8, 9] . More importantly, PCL provides the milieu for innate and adaptive immune cells including neutrophils and macrophages to home and function. Moreover, PCL prevents cilia from being entrapped in viscous mucus and bathes them for mechanical movement for mucociliary clearance [5, 10] . ASL composition and volume are collectively regulated by epithelial chloride secretion, sodium absorption and secondarily water secretion and absorption [6, 11] . Mounting evidence indicates that paracrine/autocrine purinergic signaling is critical to airway epithelial ion transport and ASL hydration [12] #. Adenosine has been shown to be a potent regulator in the process, which can be directly released by local epithelial cells and immune cells [13] or from extracellular metabolism of ATP [12] #. It is known that ATP is constitutively released by epithelia due to various stimuli including mechanical stretch and shear stress due to respiration [14] . The released ATP is then converted to adenosine by extracellular ectonucleotidases [15] . Thus, ASL has relatively high levels of adenosine. Further studies demonstrate that adenosine largely regulates epithelial CFTR channel function by acting on A 2B AR [16, 17, 18] . Thus, the adenosine-A 2B AR signaling pathway is a crucial element in lung host defense [19, 20] . Previous alcohol studies have documented that ethanol exposure decreases cAMP signaling and protein kinase A (PKA) activation [3, 21] . Ethanol also up-regulates phosphodiesterase 4 (PDE4), which increases cAMP degradation [22] #. In spite of the clear link between alcohol exposure and alteration of adenosine signaling, no published data are currently available concerning alcohol effects on airway ion transport through this signaling pathway. The current report directly measured the adenosineinduced chloride secretion of airway epithelia under the exposure of physiologically relevant concentrations of alcohol and found that ethanol attenuates epithelial CFTR-mediated chloride transport by modulating cellular cAMP levels. No human subjects or animals were used in this study. Calu-3 cells, a human airway epithelial cell line (ATCC, Manassas, VA), were seeded on collagen-coated MillicellH-PCF membrane inserts (Millipore, Billerica, MA) at a density of 1610 6 cells per insert of 0.6 cm 2 surface area. Two days after the initial submerged culture, the apical media were aspirated off and the cells cultured at an air-liquid interface according to previously published protocol [23] . Regardless of submerged culture or air-liquid interface culture, the media used were the same, consisting of Advanced-MEM (Gibco, Carlsbad, CA) containing 10% fetal bovine serum, 1% L-glutamine, 100 U/ml penicillin, 100 mg/ml streptomycin and 0.25 mg/ml amphotericin B. After 2 weeks at 37uC in presence of 5% CO 2 , the epithelia established a dry apical surface and had a transepithelial electrical resistance greater than 1000 V/cm 2 . The cystic fibrosis (CF) airway epithelial cells CFBE41o- [24] were similarly cultured. The fully differentiated CF epithelia after 2 weeks exhibited a transepithelial resistance greater than 700 V/cm 2 . Air-liquid interface cultures were basolaterally exposed to different concentrations of ethanol (200 Proof; AAPER Alcohol and Chemical Co., Shelbyville, KY), as indicated in individual experiments. All the cultures were kept at 37uC, 5% CO 2 in incubators that had been pre-saturated with specified concentrations of ethanol. The TEER of the airway epithelial cell cultures was measured by using a ''chop stick'' epithelial ohmmeter (World Precision Instruments, Sarasota, FL), as described previously [25, 26] . Calu-3 cells, cultured at the air-liquid interface for 2 weeks, were exposed to different concentrations of ethanol for 24 hours and mounted on an Ussing chamber apparatus (World Precision Instruments, Sarasota, FL). The cells were bathed with an apical low chloride buffer (135 mM sodium gluconate, 5.0 mM HEPES, 1.2 mM MgCl 2 , 0.6 mM KH 2 PO 4 , 4 mM CaCl 2 , 2.4 mM K 2 HPO 4 3H 2 O, and 10 mM Dextrose, pH 7.4) and with a basal high chloride buffer (135 mM NaCl, 5.0 mM HEPES, 1.2 mM MgCl 2 , 0.6 mM KH 2 PO 4 , 1.2 mM CaCl 2 , 2.4 mM, K 2 HPO 4 3H 2 O, and 10 mM Dextrose, pH 7.4). Both buffers were continuously stirred, gassed with 95% O 2 and 5% CO 2 and maintained at 37uC. These buffers were to maintain a chloride gradient across the epithelial monolayer. The TEER was measured with an open circuit by applying an electrical pulse across the epithelial monolayer. Short circuit currents (I SC ) were measured by applying an epithelial voltage clamp. For all the experiments, 100 mM of amiloride was added to the apical side to block sodium channels. Anion currents were induced by apical addition of 100 mM adenosine. The A 2B AR blockade was achieved by apical application of 50 mM alloxazine. To differentiate between chloride and bicarbonate currents, 100 mM bumetanide was used basally to block chloride transport. Acetazolamide (20 mM) or DNDS (4,49dinitro stilbene-2,29-disulfonate) (100 mM) was employed apically to block bicarbonate transport. Further, to inhibit epithelial phosphodiesterases, 100 mM of IBMX (3-isobutyl-1-methylxanthine) or 50 mM papaverine was added apically. The air-liquid interface cultures of Calu-3 cells were exposed to either 0 mM or 100 mM of ethanol at the basolateral side for 24 hours. On the apical surface the cells were stimulated with 100 mM of adenosine. The cells were washed with PBS containing 100 mM IBMX to inhibit phosphodiesterases that might degrade cAMP. The cells were lysed and cAMP was measured by cAMP immunoassay (R&D Systems, Minneapolis, MN). IBMX (100 mM) was also included in the lysis buffer. The samples were read at 450 nm using a spectrophotometer (BioTEK, Winooski, VT). All the treatments were carried out in quadruplicates. Data presented represent mean of multiple experiments and error bars indicate standard deviation from the mean. Where indicated, the data points were analyzed by Student's t-test or One-way ANOVA test. The P values smaller than 0.05 were considered statistically significant. To explore if ethanol affects adenosine-activated ion transport function of airway epithelium, we employed air-liquid interface cultures of Calu-3 cells, a system widely used to investigate airway epithelial electrophysiological properties [27, 28] . The cultured epithelia were exposed basolaterally for 24 hours to different concentrations of ethanol (0, 25, 50 and 100 mM) and adenosineinduced transepithelial ion transport was assessed by measuring I SC with an Ussing chamber apparatus. Two buffers with asymmetric chloride were applied: apical low chloride (10.4 mM) and basolateral high chloride (139.8 mM). After voltage clamp, sodium channels were blocked by apical amiloride (100 mM) and I SC was stimulated by apical addition of adenosine (100 mM). As shown (Fig. 1) , the ethanol exposure decreased adenosine-activated I SC in a dose-dependent manner. Significant differences were detected by Figure 1 . Effect of ethanol pre-exposure on adenosine-induced epithelial ion transport. Calu-3 cells were cultured at an air-liquid interface on membrane filters and basolaterally exposed to 0, 25, 50 and 100 mM of ethanol for 24 hours. These Calu-3 epithelia were placed in an Ussing chamber with asymmetrical buffers of chloride (apical 10.4 mM and basolateral 139.8 mM). Following voltage clamp, stable short circuit baselines were attained in 15 to 20 min. I SC was measured after blocking Na + channels with 100 mM of apical amiloride and stimulated with 100 mM of apical adenosine. Alterations in ion transport are expressed as difference in I SC from their baseline. Ethanol preexposure decreased 100 mM adenosine mediated epithelial ion transport in a dose-dependent manner. Asterisks indicate significant differences between groups by One-way ANOVA test (p,0.05, n = 5 for each condition). doi:10.1371/journal.pone.0032112.g001 One-way ANOVA test between the control and the alcoholexposed (50 mM and 100 mM) groups (p,0.05, n = 5). It was previously documented that adenosine-stimulated anion secretion in airway epithelia is largely mediated by chloride transport [17, 29] . To explore if the alcohol-suppressed I SC , as identified above, is actually a chloride current, we first confirmed chloride transport by Calu-3 epithelia in our experimental setting. To this end, the cultured Calu-3 epithelia without any alcohol exposure were subjected to I SC measurement in the presence of various inhibitors to block different ion channels. Bumetanide (100 mM), a specific inhibitor for the Na + -K + -2Cl 2 cotransporter, was applied to the basolateral side of the Calu-3 epithelia after adenosine stimulation. This drug decreased I SC by ,58%, while acetazolamide, a carbonic anhydrase inhibitor, and DNDS, an inhibitor for Na + /HCO 3 2 cotransporters and Cl 2 /HCO 3 2 exchangers, had no effect on I SC (Fig. 2A) . These data suggest that the adenosine-stimulated anion current that is suppressed by alcohol is the chloride channel conductance. The result was further validated by measuring the adenosine-stimulated I SC with identical apical and basal chloride buffers. Without chloride gradient (Fig. 2B) , adenosine-induced I SC was dropped significantly by Student's t-test (p,0.01, n = 5). The I SC was only ,8% of that measured with the asymmetric chloride buffers. The adenosine-stimulated chloride transport is mediated through the CFTR channel CFTR has been found to be a major chloride channel in the airway epithelium responsible for adenosine-induced ion transport [30] #. To validate if CFTR is responsible for the observed adenosine-stimulated chloride transport in the Calu-3 epithelia, we applied CFTR channel inhibitor CFTR inh 172 to the apical side of Calu-3 epithelia for 30 minutes, followed by adenosine stimulation. The chloride conductance was decreased by ,76% when CFTR was inhibited (Fig. 3) , which is significantly lower than that of the no drug control (p,0.01, n = 5). To seek a second approach to confirm the result, CFBE41o cells, an airway epithelial cell line derived from a cystic fibrosis (CF) patient with homozygous DF508 mutations in CFTR, were used [24] . Strikingly, adenosine failed to elicit any chloride currents across the CF epithelia (p,0.01, N = 4). Thus, the currents under our experimental condition and drug profile are CFTR-mediated chloride conductance. These data altogether confirmed that in airway epithelial cells ethanolsuppression of chloride transport is mediated through CFTR, a cAMP-activated chloride channel. The aforementioned data suggest that the negative modulation of epithelial chloride secretion by ethanol is likely through the adenosine-adenosine receptor signaling pathway. Up to date, four different adenosine receptors have been identified: A1, A 2A , A 2B and A3 receptors [31] . In airway epithelial cells, A 2B AR predominantly regulate CFTR function [18] . Here, we wanted to examine if the same signaling pathway is involved in the alcohol-induced inhibition of CFTR-mediated chloride transport. To this end, alloxazine (50 mM), a commonly used A 2B AR specific blocker, was applied to the apical buffer after adenosine stimulation. The data ( Figure 4) demonstrate that ,75% adenosine-stimulated epithelial I SC was reduced by alloxazine, indicating that A 2B AR was largely responsible for the adenosinemediated chloride secretion. Figure 2 . Adenosine-stimulated short circuit current is chloride conductance. A) I SC of Calu-3 epithelia was measured by using the asymmetric chloride buffers. After stimulated with 100 mM of apical adenosine, the cells gave rise to I SC of ,75 mA/cm 2 . The I SC was blocked by ,58% by basolateral addition of bumetanide (100 mM), but not by acetazolamide (20 mM) or DNDS (100 mM). B) I SC of Calu-3 epithelia was measured with either asymmetric chloride buffers or symmetric chloride buffers. The adenosine-stimulated I SC with no chloride-gradient buffers was ,92% lower than that with chloride-gradient buffers. Significance of the difference was determined by Student's t-test (p,0.01, n = 5). doi:10.1371/journal.pone.0032112.g002 Figure 3 . Adenosine-stimulated transepithelial I SC was mediated by CFTR channel. Calu-3 epithelia were stimulated with 100 mM of apical adenosine, resulting in a significant increase in I SC above baseline. Such an adenosine-induced I SC was mostly absent in CFBE41o cells in which CFTR channel was dysfunctional or was significantly inhibited with 25 mM of CFTR inhibitor CFTR inh 172. Student's t-test was performed to determine the statistic significance (p,0.05, n = 5). doi:10.1371/journal.pone.0032112.g003 Ethanol exposure affects CFTR-mediated chloride secretion through modulation of cellular cAMP level Upon adenosine binding, the A 2B AR signals through G s protein to activate adenylyl cyclase and raises intracellular cAMP [13, 32] . Because CFTR is a cAMP-activated chloride channel, we hypothesized that ethanol inhibits adenosine-stimulated cAMP production to cause a reduced CFTR channel activity. To test this hypothesis, air-liquid interface cultures of Calu-3 cells were exposed basolaterally to either 0 mM or 100 mM of ethanol for 24 hours. These cells were apically stimulated with 100 mM of adenosine as described above for 10 minutes, the time point at which the epithelial I SC was at its peak levels. Then, the cells were lysed and the levels of cellular cAMP assayed. The results demonstrate that ethanol pre-treatment significantly decreased adenosine-stimulated cAMP levels (p,0.05, n = 5) (Fig. 5A ). In addition, to test whether the decrease in cAMP was in fact causing the suppression of adenosine-stimulated chloride transport by ethanol, Sp-cAMPS, a cell permeable and phosphodiesteraseresistant cAMP analogue, was used to directly activate PKA which then activates the CFTR channel. As shown in Figure 5B , the alcohol-treated Calu-3 cells in the presence of 10 mM Sp-cAMPS obliterated the ethanol suppressive effect on adenosine-induced chloride secretion (p,0.05, n = 6). Thus, we conclude that ethanol inhibits CFTR-mediated chloride secretion by directly affecting the cellular cAMP level instead of the downstream PKA enzyme. Based on the data that ethanol impairs the adenosine-activated chloride secretion by reducing the cellular cAMP level, we chose to block endogenous cAMP degradation pharmacologically to counteract the alcohol-suppressive effect on chloride secretion. Air-liquid interface cultures of Calu-3 cells were similarly exposed to ethanol for 24 hours and treated with 100 mM of non-specific phosphodiesterase inhibitor IBMX along with adenosine stimulation. The data in Figure 6 indicate that IBMX almost completely restored the ethanol suppression of Calu-3 chloride secretion (p,0.05, n = 4). Papaverine is a clinically used phosphodiesterase inhibitor. This drug also overcame the inhibitory effect of ethanol on adenosine-induced transepithelial chloride conductance (p,0.05, n = 4) (Fig. 6) . These results not only confirm that ethanol modulates adenosine-cAMP signaling but also suggest that phosphodiesterase inhibitors may be useful as the potential therapeutic agents for improving the airway epithelial ion transport and mucociliary clearance in alcoholic patients. Airway epithelial cells not only constitute the physical barrier that separates airway lumen from interstitial compartments, but also actively participate in innate and adaptive immunity to protect the host from pulmonary infections [33] . Multiple defense systems have been evolved in airways. First, the polarized airway epithelia have a mechanical clearance mechanism. Goblet cells or submucosal glands secrete mucus that entraps airborne particles and inhaled infectious agents. Synchronized ciliary movement sweeps inhaled particulate matter toward the mouth to be expectorated or swallowed [34] #. Second, airway epithelia secrete antimicrobial factors, such as lysozyme and b-defensins [35] . Third, airways and alveoli are patrolled by phagocytic cells, most notably the alveolar macrophage, which can engulf microbes [36] . Fourth, airways can recruit neutrophils and monocytes to sites of inflammation [37] . Fifth, the acquired immune system, including antigen-stimulated T and B lymphocytes, provides cellular and humoral defenses for the airways [38, 39] . It is noteworthy that all the above-mentioned pulmonary defense mechanisms are executed in ASL, the thin film of liquid on top of the airway epithelia. Therefore, ASL homeostasis is pivotal to lung host defense. Here we provide the first evidence suggesting that ethanol exposure suppresses adenosine-stimulated chloride secretion that regulates ASL. Our data demonstrate that ethanol affects adenosine-stimulated chloride secretion through CFTR. On epithelial apical surface, adenosine binds to A 2b AR, a predominant adenosine receptor present in airway epithelia [16] . This receptor is coupled to G s protein to activate adenylyl cyclase, which elevates intracellular cAMP and consequently activates PKA. Then, CFTR is phosphorylated by PKA and the channel opens to permeate chloride [40] #. More chloride in ASL will cause less Na + absorption and more water retention, thus increasing ASL height and volume [15, 20] . It is even speculated that adenosine is a sensor for ASL homeostasis. When ASL falls, adenosine levels increase in airways beyond their basal levels and activate A 2B AR and CFTR channels [16, 29] . Our data have linked ethanol-suppression of cAMP levels to attenuation of chloride secretion by airway epithelia. Previous studies have documented that ethanol decreases ciliary beating and epithelial cell migration during wound repair which are associated with compromised cAMP signaling and PKA activation [21, 41] . Moreover, alcohol is reported to upregulate the PDE4 enzyme expression as well as the enzymatic activity in epithelia [22] #, which results in accelerated cAMP degradation. Thus, our data are consistent with the data reported, indicating that alcohol modulates various cellular pathways by reducing cellular cAMP levels. Pulmonary mucociliary clearance is largely affected by three factors: 1) mucus production, 2) ciliary sweeping and, 3) ASL state. Studies on human bronchial epithelial cells have revealed that a 24-hour exposure with 100 mM of alcohol caused an 8-fold increase in trachea-bronchial mucin gene expression [42] . Interestingly, experiments using bovine bronchial epithelial cells have established that alcohol exposure at 100 mM beyond 6 hours decreases ciliary beating frequency [22, 41] . A recent publication by Allen-Gipson and colleagues reports that purinergic stimulation of CBF requires A 2B AR activation [43] . Similarly, in rats chronic alcohol administration decreased ciliary beating frequency and enhanced lung colonization of nasally administered S. pneumoniae [44] #. Our current results demonstrate that 24-hour ethanol exposure reduces chloride secretion, which consequently alters ASL composition and volume. Thus, alcohol affects all three mucociliary clearance components. Pharmacologically, blocking cAMP degradation has the potential of restoring ethanol-suppressed cellular cAMP levels and therefore epithelial functions. This finding is of clinical implications. Phosphodiesterase inhibitors such as papaverine may improve mucociliary clearance in alcoholics by enhancing airway epithelial ion secretion and ciliary beating. Moreover, enhancement of airway epithelial ion secretion alters the composition of ASL. Our laboratory has reported that extracellular chloride levels affect neutrophil microbial killing ability [45] . Thus, increasing ASL chloride level by phosphodiesterase inhibitors may also improve phagocytic innate immunity in airways of alcoholics. In summary, ethanol exposure compromises chloride ion secretion which is pivotal to maintain ASL volume and composition. Such an effect may alter the properties of airway host defenses predisposing ethanol abusers to an increased risk of infection in the lung. Restoration of cellular cAMP level with phosphodiesterase inhibitors could potentially ameliorate mucociliary clearance by improving epithelial ion secretion and hence lung host defense. Figure 6 . Phosphodiesterase inhibitors restore the ethanol suppression of adenosine-induced transepithelial chloride conductance. The air-liquid interface cultures of Calu-3 cells were exposed to 0 and 100 mM of ethanol for 24 hours. Adenosinestimulated I SC was measured, showing that ethanol-exposed Calu-3 epithelia had significantly lower adenosine-stimulated I SC than the no ethanol control. However, phosphodiesterase inhibitors, IBMX (100 mM) or papaverine (50 mM), fully restored the ethanol-suppressed adenosine-stimulated I SC . Asterisks indicate statistically significant differences between groups (p,0.05, n = 4 for each group). doi:10.1371/journal.pone.0032112.g006

High Fidelity Processing and Activation of the Human α-Defensin HNP1 Precursor by Neutrophil Elastase and Proteinase 3

The azurophilic granules of human neutrophils contain four α-defensins called human neutrophil peptides (HNPs 1–4). HNPs are tridisulfide-linked antimicrobial peptides involved in the intracellular killing of organisms phagocytosed by neutrophils. The peptides are produced as inactive precursors (proHNPs) which are processed to active microbicides by as yet unidentified convertases. ProHNP1 was expressed in E. coli and the affinity-purified propeptide isolated as two species, one containing mature HNP1 sequence with native disulfide linkages (“folded proHNP1”) and the other containing non-native disulfide linked proHNP1 conformers (misfolded proHNP1). Native HNP1, liberated by CNBr treatment of folded proHNP1, was microbicidal against Staphylococcus aureus, but the peptide derived from misfolded proHNP1 was inactive. We hypothesized that neutrophil elastase (NE), proteinase 3 (PR3) or cathepsin G (CG), serine proteases that co-localize with HNPs in azurophil granules, are proHNP1 activating convertases. Folded proHNP1 was converted to mature HNP1 by both NE and PR3, but CG generated an HNP1 variant with an N-terminal dipeptide extension. NE and PR3 cleaved folded proHNP1 to produce a peptide indistinguishable from native HNP1 purified from neutrophils, and the microbicidal activities of in vitro derived and natural HNP1 peptides were equivalent. In contrast, misfolded proHNP1 conformers were degraded extensively under the same conditions. Thus, NE and PR3 possess proHNP1 convertase activity that requires the presence of the native HNP1 disulfide motif for high fidelity activation of the precursor in vitro.

Antimicrobial peptides (AMPs) produced by animal cells provide a first line of defense against potentially infectious microorganisms. In mammals, defensins and cathelicidins are the primary AMPs expressed in professional phagocytes such as neutrophils and macrophages [1] . Mammalian defensins comprise three structural subfamilies termed a-, band h-defensins. While all defensins are arginine rich, tridisulfide containing peptides, they are distinguished from each other by the presence of unique disulfide motifs [2] . In addition, h-defensins, which are approximately half the size of aand b-defensins, are macrocyclic molecules which have only been isolated from leukocytes of Old World monkeys (reviewed in [2, 3] ). Defensins have antimicrobial activities against diverse microbial targets that include bacteria [4] , fungi [5, 6] , viruses [7, 8, 9] , and protozoa [10] . Certain a-, b-, and h-defensins function as regulators of adaptive immune cells and thus function to modulate inflammation [11, 12, 13] and bridge innate and adaptive immunity [14] . In humans, four a-defensins, human neutrophil peptides 1-4 (HNP [1] [2] [3] [4] , are expressed in neutrophils and monocytes of myeloid lineage, and a-defensins (HD5 and HD6) are produced primarily in Paneth cells of the small intestine. In neutrophils, HNP 1-4 are stored in the cytoplasmic azurophil granules which are mobilized during phagocytosis resulting in the exposure of ingested microorganisms to high concentrations of HNPs in the phagolysosomes [2] . In the setting of bacteremic sepsis, HNPs also accumulate in the extracellular compartment [15, 16] . Several studies have demonstrated an extracellular role for myeloid adefensins in regulating cellular activities including cell proliferation [17, 18, 19] , neovascularization [20, 21] , chemotaxis [22] , and regulation of cytokines [23, 24] and signal transduction pathways [17, 19, 24, 25] . To date, these activities have been attributed to the mature, folded a-defensin. Defensins of all three subfamilies are expressed as preprodefensins containing an N-terminal signal peptide followed by the propeptide that contains the prosegment and the folded/oxidized mature peptide sequence [26] . The prosegment in aand hdefensin precursors is ,45 amino acids in length, but that of bdefensin precursors is typically very short. In a-defensins, studies suggest that the anionic prosegment interacts with residues present in the mature peptide to maintain it in an inactive state until it reaches the correct cellular location, possibly to protect cellular compartments from peptide-mediated cytotoxic effects [27, 28, 29] . Processing of the human enteric a-defensin, HD5 occurs extracellularly and is mediated by trypsin [30] , whereas mouse Paneth cell a-defensins are processed intracellularly by matrix metalloprotease 7 (MMP7, matrilysin) [31] . Neutrophil elastase (NE), an azurophil granule protease, was shown to correctly process RMAD4, a rhesus macaque neutrophil a-defensin in vitro [32] . Bovine and human cathelicidins, a distinct class of antimicrobial peptides, are processed by the azurophil granule proteases, NE [33] and proteinase 3 (PR3) [34] respectively. Studies with HL-60 cells, a human acute promyelocytic leukemia cell line, show that HNPs are biosynthesized and processed to mature peptides in these cells [35] , but the proteases involved in maturation of human myeloid a-defensins have not been identified. Because HNPs co-localize to neutrophil granules containing the serine proteases NE, PR3 and cathepsin G (CG) [36] , we hypothesized that HNP maturation is mediated by one or more of these proteinases. In this study, recombinant proHNP1 was used as substrate to analyze the human a-defensin convertase activities of NE, PR3, and CG. The requirement for the native HNP1 fold was analyzed by assessing the processing of folded and misfolded preparations of proHNP1. In addition, the antimicrobial activity of the resulting HNP-1 peptides was analyzed utilizing Staphylococcus aureus as the test organism. Recombinant hexa-His-tagged pro-HNP-1 ( Fig. 1 ) was expressed at approximately 2 mg/L in E. coli cells. The imidazole eluant from the nickel affinity column (Ni-eluant) contained a major band of ,12 kDa, consistent with the molecular weight of His 6 -proHNP1 (theory = 12647 A.M.U.; Fig. 2A ). C18 reversed phase high performance liquid chromatography (RP-HPLC) of the Ni-eluant produced two major peaks (A and B; Fig. 2B ). On reducing SDStricine PAGE, peaks A and B contained the ,12 kDa band (Fig. 2C ) and both were detected with anti-His 6 antibody confirming the presence of His 6 -proHNP1 (Fig. 2D) . On AU-PAGE, His 6 -proHNP1 present in peak A (proHNP1A) migrated more rapidly than that contained in peak B (proHNP1B; Fig. 2E ), suggesting that proHNP1A and proHNP1B are differently folded forms of the recombinant protein. This was confirmed by the finding that DTTreduced proHNP1A and proHNP1B co-eluted on RP-HPLC (data not shown) and by demonstrating that folding of proHNP1B in the presence of reduced/oxidized glutathione led to the formation of proHNP1A (Text S1 and Figure S1 ). Thus, proHNP1A and proHNP1B contain folded and misfolded His 6 -proHNP1 respectively. ProHNP1A and proHNP1B-containing fractions were pooled separately and purified to homogeneity by a second round of RP-HPLC (Fig. 2F ). The presence of a Met residue at the junction of the prosegment and mature peptide in proHNP1 sequence (see Fig. 1 ) offered a non-enzymatic approach to generate HNP1 from proHNP1 utilizing cyanogen bromide (CNBr) cleavage [28] . Cleavage of proHNP1A with CNBr, followed by C18 RP-HPLC, gave rise to a major peptide product (Fig. 3A ) with a mass (Fig. 3B ) and HPLC retention time (Fig. 3C ) matching that of natural HNP1. However, CNBr digestion of proHNP1B generated multiple species (Fig. 3A ) that had masses consistent with oxidized HNP1 (Fig. 3B ), but which eluted later than native HNP1 in RP-HPLC (Fig. 3C) . On AU PAGE, HNP1 derived from proHNP1A comigrated with natural HNP1. However, HNP1 derived from proHNP1B migrated more rapidly than the natural standard (Fig. 3D) . These data demonstrate that proHNP1A contained folded HNP1 with native disulfide linkages but the mature peptide in proHNP1B, although oxidized, is misfolded and contains non-native disulfide linkages. In solution, HNP1 exists as a non-covalent dimer [37] and the dimeric form corresponds to the single band observed in AU PAGE (Selsted et al., unpublished data) . Therefore, it is likely that the higher mobility of proHNP1B-derived HNP1 (Fig. 3D ) is due to the monomeric state of this conformer. To analyze the microbicidal properties of the HNP1 conformers produced by CNBr cleavage of proHNP1A and proHNP1B, we tested the respective peptides for killing of S. aureus in vitro. HNP1 derived from proHNP1A had microbicidal activity similar to that of native HNP1 (Fig. 4) . However, HNP1 derived from proHNP1B was devoid of staphylocidal activity (Fig. 4) . Thus, formation of native disulfide linkages in HNP1 is essential for the staphylocidal activity of the peptide, and may reflect a further requirement for dimerization. Previous studies have demonstrated that NE, PR3, and CG colocalize with HNPs in the azurophil granules of neutrophils. Kamdar et al. demonstrated that NE digestion of proRMAD4, precursor of rhesus myeloid a-defensin-4 (RMAD4), converted proRMAD4 to mature RMAD4. However, digestion of pro-RMAD4 with PR3 and CG generated RMAD4 variants with Nterminal extensions (Fig. 5D ) [32] . To test for the role of these granule serine proteases in human HNP processing, proHNP1A was incubated in separate reactions with NE, PR3, or CG, and the RP-HPLC-purified cleavage products were analyzed by MALDI-TOF MS and analytical HPLC ( Fig. 5 and Table 1 ). Incubation of His 6 -proHNP1 with NE or PR3 generated peptide products that coeluted with native HNP1 in RP-HPLC (34. Table 1 ). The identities of HNP1 derivatives were also confirmed by MALDI-TOF MS of each peptide following reduction and S-carboxyamidomethylation. In each case reduction and alkylation produced a mass increase of 348 A.M.U., confirming the presence of 6 disulfide-linked cysteine residues ( Table 1 ). The NE-and PR3-derived peptides also comigrated with native HNP1 on AU-gels ( Fig. 5C ; lanes 1-3). Thus, NE and PR3 processing of His 6 -proHNP1 generated a peptide that was indistinguishable from natural HNP1. Incubation of His 6 -proHNP1 with CG did not generate mature HNP1 but instead a digestion product that eluted from C18 HPLC later than HNP1 (Fig. 5A ) and had a mass of 3685.4 A.M.U. This mass corresponds to HNP1 with a dipeptidyl (Asn-Met-) aminoterminal extension, consistent with Lys62QAsn63 cleavage by CG (Fig. 5D ). The CG liberated HNP1 product eluted as a single peak (Fig. 5A) , which was resolved into two species by high resolution RP-HPLC (Alliance analytical HPLC system; Fig. 5B ). Peptide in the second peak corresponded to the mass of Asn-Met-HNP-1 (3685. Although the junctional sequences recognized by NE, PR3, and CG in His 6 -proHNP1 are identical to those in the natural precursor, we questioned whether the 41-residue extension of the proHNP-1 sequence resulting from the hexa-His and DDDDK tags and spacer sequences might alter protease interactions with the target. To test this, we produced proHNP1 without the Nterminal tag by incubating proHNP1A with enterokinase, thereby releasing the 41-residue tag sequence. This peptide was then treated with NE, PR3, or CG which liberated peptides with monoisotopic masses of 3439.7, 3439.7 and 3684.5 A.M.U., respectively, products equivalent to those generated from His 6 -proHNP1A. Thus the 41-residue N-terminal fusion tag did not alter the specificity of cleavage of proHNP1A by NE, PR3, or CG. Protease processing of misfolded His 6 -proHNP1 NE, PR3, and CG degraded proHNP1B, the misfolded form of the HNP1 precursor. In contrast to the discrete products generated by digestion of proHNP1A by the serine proteases ( Fig. 5) , NE, PR3, or CG treatment of proHNP1B yielded no mature HNP1 or HNP1 derivatives in the digestion mixtures detectable by C18 RP-HPLC or MALDI-TOF MS (data not shown). These results indicate that the correct fold and disulfide linkages in proHNP1 are required for processing of the precursor to mature peptide and to confer resistance to these activating convertases. Microbicidal activities of HNP1 products of proHNP1A processing by NE, PR3 and CG The functional properties of HNP1 generated by serine protease processing of proHNP1A were tested in microbicidal assays against S. aureus. Natural HNP1 and NE, PR3 and CG processed preparations possessed nearly identical dose-dependent bacteri- cidal activities against this organism (Fig. 6) . These data demonstrate that HNP1 generated by NE and PR3 processing of proHNP1A is functionally equivalent to native HNP1. Of note, the Asn-Met-HNP1, produced by CG processing of proHNP1A, had staphylocidal activity equivalent to that of the fully processed HNP1 isoform. To study the processing of human myeloid a-defensins, we established an in vitro assay utilizing recombinant proHNP1. Approximately 60% of the recombinant His 6 -proHNP1 was folded with native disulfide linkages (Fig. 2) and misfolded propeptide was readily converted to the folded form in buffer containing urea and reduced/oxidized glutathione (Fig. S1 ). The expression of folded and misfolded forms of proHNP1 allowed for the analysis of the structural requirements for HNP1-mediated bactericidal activity in vitro. CNBr treatment of folded (proHNP1A) or misfolded (proHNP1B) precursor generated different HNP1 conformers (Fig. 3) . Despite the generation of oxidized peptides with masses matching that of natural HNP1, only the conformer generated by CNBr cleavage of proHNP1A had microbicidal activity against S. aureus. Processing of proHNP1A by NE or PR3 generated HNP1 peptide identical to natural HNP1, but proHNP1B was heterogeneous and was degraded by NE, PR3 and CG. These data demonstrate that the native HNP1 fold is necessary for proteolytic stability and accurate processing of the HNP1 propeptide by the three azurophil granule serine proteases investigated. The requirement for the native HNP1 fold for microbicidal activity is consistent with a previous report demonstrating that an all Cys-to-Ala analog of HNP1 was nearly 10-fold less potent in bactericidal assays than the native peptide [38] . Similarly, disruption of disulfide linkages by mutagenesis of Cys residues to Ala or Ser in the human Paneth cell a-defensin HD5 reduced in vitro bactericidal activity demonstrating the requirement for the disulfides for in vitro bactericidal activity of this peptide [39] . In contrast, disruption of disulfide linkages in mouse enteric defensin, Crp4, by Cys-to-Ala mutagenesis did not eliminate microbicidal activities of the peptide [40] . Results presented here support the hypothesis that NE and PR3, proteases that co-localize with myeloid a-defensins, participate in the intracellular maturation of HNP1. In previous studies with HL-60 and chronic myelogenous leukemia cells, Valore and Ganz demonstrated sequential proteolytic processing of proHNP with the formation of a 56-residue intermediate that is subsequently processed to mature HNP. While proHNPs were found exclusively in the cytoplasmic/microsomal fraction, the granule fraction contained only the intermediate and mature HNP [35] . Therefore it is likely that proHNP-processing by NE and PR3 occurs during the packaging of HNPs in azurophil granules. CG processing of proHNP1A produced a peptide that was extended at the amino terminus by two residues (Asn-Met) expressed at the carboxyl terminus of the HNP1 prosegment. Harwig et al. identified a minor form of HNP3 in fully differentiated neutrophils with the same Asn-Met dipeptide extension [41] reported here. Thus, CG processing of proHNP1 may also occur in vivo. In contrast to the processing of folded proHNP1 by the azurophilic proteases, misfolded His 6 -proHNP1 was degraded by these enzymes. These results are consistent with studies demonstrating that Cys substitutions in mutants of RMAD4 [32] , Crp4 [40] and HD5 [39] are protease sensitive compared to the respective native peptides. However, in contrast to the examples cited above, misfolded His 6 -proHNP1B was disulfide-linked but still susceptible to proteolysis. These data demonstrate that the native disulfide motif is essential for proteolytic stability of the proHNP1 as well as the mature peptide. Wilson et al. demonstrated that proHNP1 could be processed by matrix metalloproteinase 7 (MMP7; matrilysin) to generate a 59residue molecule with microbicidal activity, but no mature HNP1 (30 amino acids) was detected [42] . Since MMP7 is not present in neutrophils, it was proposed that MMP7 may be involved in the extracellular processing of proHNP1 released from the specific granules of neutrophils. In this regard, Borregaard and co-workers showed that proHNP1 is present in the specific granules of mature human neutrophils and that the propeptide is secreted by activated cells [43] . Since enzymatically active NE is present in inflammatory exudates [44] , we hypothesize that proHNP1, elaborated by activated PMNs, may be activated by NE (and possibly other enzymes such as PR3 and CG) in the extracellular milieu. Therefore this process may regulate the extracellular expression of mature a-defensin in the setting of inflammation. The proHNP1 cDNA sequence encoding the natural 75 amino acid residue sequence extended by 5 amino acids (DDDDK; enterokinase recognition site) at the N-terminus was amplified using oligonucleotides (59-GCGAATTCGATGACGATGACAAGGAGCCACTCCAGG-CAAGAGCTG) and (59-GCGGATCCGGTACCTCGAGTCAG-CAGCAGAATGCCCAGAGTC) with plasmid PEC2081 (proHNP1 cloned in pCR2.1 Topo vector) as template. EcoRIand XhoI-digested DNA was ligated into pET28a+ vector to generate plasmid PEC2129 which encodes proHNP1 with a 41 amino acid residue N-terminal sequence containing the His 6 -tag (His 6 -proHNP1; Fig. 1 ). The plasmid was used to transform Escherichia coli BL21(DE3)(codon+), generating strain PEC2132-1 which was used for expression and purification of His 6 -proHNP1. PEC2132-1 cells were cultured for 3 h at 37uC in 3 L of LB+Kanamycin (50 mg/ml) medium and expression of His 6 -proHNP1 was induced by addition of 0.5 mM isopropyl b-D-1thiogalactopyranoside (IPTG) for an additional 3 h. Cells were harvested by centrifugation and lysed by stirring in 60 ml guanidine lysis buffer (10 mM Tris HCl, 6 M guanidine, 100 mM NaH 2 PO 4 , pH 8.0) at room temperature for 2 h. The lysate was clarified by centrifugation and supernatant was mixed with 6 ml of Ni-resin (His-Bind resin; Novagen) overnight at 4uC. Resin was transferred to a chromatography column, washed with 10 volumes of wash buffer 2 (20 mM Tris HCl, 450 mM NaCl, 0.1% Tween, pH 8.0), and bound proteins were eluted with 3.5 column volumes of elution buffer (20 mM Tris-HCl, 450 mM NaCl, 0.1% Tween, 250 mM imidazole, pH 8.0). The eluate was dialyzed in Spectrapor 3 membrane against 2 L of dialysis buffer (4 M guanidine, 20 mM Tris, pH 8.0), overnight at 4uC. The dialysate was acidified by addition of acetic acid to 10% (vol/vol) and fractionated on a 10 mm6250 mm Vydac C18 reversed phase high performance liquid chromatography (RP-HPLC) column equilibrated in aqueous 0.1% trifluoroacetic acid (TFA). Proteins were eluted using a water-acetonitrile (ACN) gradient containing 0.1% TFA. As discussed in Results, His 6 - Protein preparations, quantified using the BioRad protein assay (BSA standard) or by measuring absorbance at 280 nm, were analyzed by sodium dodecyl sulfate (SDS)-tricine and acid-urea (AU) gel polyacrylamide gel electrophoresis (PAGE) as described [45] . For western blotting, proteins resolved by SDS-tricine PAGE were transferred to nitrocellulose membranes using a GE Novablot/Pharmacia Biotech Multiphor II system. The mem-brane was blocked with 5% milk in TTBS (100 mM Tris HCl pH 7.5, 0.9% NaCl, 0.1% Tween 20) and incubated with 0.02 mg/ml anti-His 6 mouse IgG (Qiagen). The membrane was washed with TTBS and probed with goat anti-Fc mouse IgGhorse radish peroxidase conjugate (1:20,000 dilution; Sigma) for 1 h, washed with TTBS, and developed by incubation with Supersignal West Pico chemiluminescent substrate (Pierce) and the resulting chemiluminesence was visualized on x-ray film. Matrix assisted laser desorption/ionization time-of-flight mass spectroscopy (MALDI-TOF MS) was employed for analysis of HPLC-purified peptides. Typically, a 0.8 ml sample was mixed 1:1 with sample matrix (15% w/v a-cyano-4-hydroxycinnamic acid dissolved in 50% ACN+0.1% TFA), spotted and dried on the MALDI-TOF MS sample plate. In some cases, samples were first purified using C18 Zip Tips (Millipore) as recommended by the manufacturer. MALDI-TOF MS was performed on a Microflex LRF MALDI-TOF system (Bruker Daltonics) [46] . Samples of lyophilized His 6 -proHNP1A (150 mg) or His 6 -proHNP1B (550 mg) were dissolved in 2 ml 80% formic acid containing 20 mg CNBr. The reaction mixtures were sparged with N 2 and incubated for 18 h in the dark at room temperature after which the solutions were diluted 10-fold with water and lyophilized. The reaction products were dissolved in 0.1% acetic acid and purified by RP-HPLC on an analytical C18 column and analyzed by MALDI-TOF MS. Fractions containing HNP1 were purified further by C18 RP-HPLC. Human NE, PR3, and CG were purchased from Elastin Products Company. Reactions were performed in 50 ml volumes containing 15 mg His 6 -proHNP1 and 2 mg of protease. Buffer for NE and PR3 reactions was 50 mM Tris-HCl pH 7.5, 150 mM NaCl and CG reactions were performed in 50 mM Tris-HCl pH 8.3, 150 mM NaCl [32] . Incubation was performed at 37uC for 18 h after which 5 ml samples were acidified by addition of 10 ml 5% acetic acid, and desalted with C18 Zip Tips (see above) prior to MALDI-TOF MS analysis. The remainder of the reaction mixture was subjected to further purification by RP-HPLC on an analytical C18 column using a 0-60% water-ACN gradient containing 0.1% TFA. The identity of putative HNP-containing species was confirmed by MALDI-TOF MS analysis of reduced (DTT) and alkylated (iodoacetamide) samples which produced a new species with the mass of hexa-S-carboxyamidomethylated peptide derivative (+348 amu) [47] . To generate enzymatically-processed HNP1 for functional studies, 150 mg of His 6 -proHNP1A was digested with 0.1 mg/ml PR3 or CG in 200 ml or with NE in 400 ml reactions for 22 h at 37uC and purified by C18 RP-HPLC as described above. Analytical HPLC HNP1 preparations derived by enzymatic or CNBr treatment were analyzed on an Alliance 2690 Separations Module using a C18 column (3 mm particle, 180 Å pore, 2.0 mm6150 mm, Varian Polaris 5A) with a flow rate of 0.2 ml/min using a 0-to-60% ACN linear gradient in 0.1% aqueous TFA at 1% per minute. Eluting peaks were detected at 210 nm and 280 nm using a Waters 2487 Dual l absorbance detector using Clarity software. Natural (human neutrophil-derived) HNP1 was used as standard in these analyses. The antibacterial activity of HNP1 preparations was evaluated using Staphylococcus aureus 502a as the test organism employing a microbicidal assay format described previously [46] . Briefly, log phase cultures of S. aureus (,1610 4 CFU/ml) were suspended in 100 ml buffer (20 mM Tris pH 8.0, 28 mM NaCl) plus 0.03% TSB with dilutions of HNP1 ranging from 0 to 8 mg/ml in 0.01% acetic acid for 2 h at 37uC. Aliquots were plated on TSB plates and colonies were counted after overnight incubation at 37uC. Native HNP1 from human leukocytes was used as control. The source of human leukocytes was discarded, de-identified buffy coat preparations from The University of California Irvine-Medical Center clinical labs. Figure S1 Folding of His 6 -proHNP1B. His 6 -proHNP1A and His 6 -proHNP1B were analyzed by C18 RP-HPLC and AU-PAGE. Shown are chromatograms and corresponding Coomassie stained gels of His 6 -proHNP1A and His 6 -proHNP1B analyzed A) without additional treatment, B) after 4 h incubation in TUN buffer alone, and C) after 4 h incubation in TUN buffer supplemented with reduced and oxidized glutathione. (TIF) Text S1 Folding of His 6 -proHNP1B in redox buffer system. In a previous study, Wu et al. demonstrated efficient folding of proHNP1 in vitro using reduced/oxidized glutathione to mediate thiol-disulfide exchange [29] . Employing this approach, 10 mg of HPLC purified His 6 -proHNP1 was dissolved in 500 ml 50 mM Tris HCl pH 7.5, 0.8 M urea and 100 mM NaCl (TUN buffer) with or without 2 mM reduced and 0.4 mM oxidized glutathione and incubated at 37uC for 4 h at which point the reaction was stopped by addition of 50 ml acetic acid. Reaction products were purified by RP-HPLC on a Vydac analytical C18 column (4.6 mm6250 mm) using a 20-to-55% (1% per min; 1 ml/min flow rate) linear gradient of ACN in 0.1% TFA acetonitrile. Peak fractions were collected and analyzed by AU-PAGE. In the absence of glutathione, the retention time during RP-HPLC and electrophoretic pattern was unchanged compared to the respective starting materials (Fig. S1A and B ). After incubation with glutathione, the behavior of proHNP1A was unchanged on C18 RP-HPLC (Fig. S1C) and on AU PAGE (Fig. S1C, lane 5) . However, treatment of proHNP1B with glutathione generated a new species that behaved like proHNP1A on both C18 RP-HPLC and AU-PAGE (Fig. S1C) . These data indicate that proHNP1A and proHNP1B are folded and misfolded forms of His 6 -proHNP1respectively and that proHNP1B can be folded in the presence of reduced/oxidized glutathione. (DOC)

Foxp3(+) Regulatory T Cells Control Persistence of Viral CNS Infection

We earlier established a model of a persistent viral CNS infection using two week old immunologically normal (genetically unmodified) mice and recombinant measles virus (MV). Using this model infection we investigated the role of regulatory T cells (Tregs) as regulators of the immune response in the brain, and assessed whether the persistent CNS infection can be modulated by manipulation of Tregs in the periphery. CD4(+) CD25(+) Foxp3(+) Tregs were expanded or depleted during the persistent phase of the CNS infection, and the consequences for the virus-specific immune response and the extent of persistent infection were analyzed. Virus-specific CD8(+) T cells predominantly recognising the H-2D(b)-presented viral hemagglutinin epitope MV-H(22–30) (RIVINREHL) were quantified in the brain by pentamer staining. Expansion of Tregs after intraperitoneal (i.p.) application of the superagonistic anti-CD28 antibody D665 inducing transient immunosuppression caused increased virus replication and spread in the CNS. In contrast, depletion of Tregs using diphtheria toxin (DT) in DEREG (depletion of regulatory T cells)-mice induced an increase of virus-specific CD8(+) effector T cells in the brain and caused a reduction of the persistent infection. These data indicate that manipulation of Tregs in the periphery can be utilized to regulate virus persistence in the CNS.

The role of CD4 + CD25 + regulatory T cells (Tregs) in autoimmune and pathogen-induced immune responses has been studied intensively during recent years. An important tool was provided by the discovery of the transcription factor Foxp3 (forkhead box P3) as a marker for Tregs and their suppressive activity [1, 2, 3, 4] . During acute viral infections depletion of Tregs was found to prevent the development of exhausted T cells and to improve the immune response. In addition, transient depletion of Tregs in several persistent viral infections led to reactivation of virus-specific T cells and reduction of the virus load [5, 6, 7, 8, 9] . The important protective role of Tregs against an overshooting immune response in the CNS became obvious in animal models of stroke and experimental autoimmune encephalitis [10, 11] , and human immunodeficiency virus-1 (HIV-1)-associated neuro-degeneration, where they reduce astrogliosis and microglia-mediated inflammation [12] . Interestingly, some viruses even developed the strategy to support the expansion of Tregs in order to suppress anti-viral cytotoxic T cell (CTL) responses and to limit viral immunopathogenesis [13, 14, 15, 16, 17, 18] . Defects in regulation of numbers or the activity of Tregs are also involved in a number of human autoimmune diseases such as type 1 diabetes, rheumatoid arthritis, and multiple sclerosis [19] . Viral infections of the brain mostly represent clinically important, often life-threatening complications of systemic viral infections. For example, after acute measles, CNS complications may occur early as acute post-infectious encephalitis, or after years of viral persistence as subacute sclerosing panencephalitis (SSPE). Epidemiological studies indicated that the primary MV infection of SSPE patients takes place predominantly below the age of two years, when the immune system of the host is still immature and residual maternal antibodies may be absent or not sufficient for complete virus neutralization [20, 21] . For intracerebral MV infection of mice, a transgenic human receptor for MV is not necessarily required, and various infection models exist depending on the age of the mice, the virus strain, and the infectious dose [22, 23, 24, 25, 26] . As found in genetically unmodified and MVreceptor transgenic mice, T cells and IFN-c have a critical role for protection and clearance of virus from the brain [27, 28, 29, 30] . Transient immunosuppression during MV-infection enhanced virus replication and facilitated persistence [31] . After intranasal infection of MV-receptor CD150-transgenic mice, a specific antiviral cellular immune response including an increased proportion of Foxp3 + Tregs in the spleen was observed [32] . Although the presence and activity of Tregs has been demonstrated, their actual role in viral immunosuppression or immunopathogenesis in the brain remains to be elucidated. Here we investigated the role of Tregs for virus persistence in the CNS. According to our model, two week old C57BL/6 mice (an age in which these mice survive infection, while the immune system is still not fully matured) were intracerebrally infected and virus persists in a limited number of neurons in most animals for more than 10 weeks [33] . We expanded and depleted Tregs in the periphery during the persistent phase of the viral infection, and investigated whether this can be exploited to modulate the ''hidden'' CNS infection. Our data indicate that this is indeed the case and that manipulation of Tregs can be utilized to regulate virus-specific CD8 + effector T cells and virus persistence in the brain. Regulatory T cells are present predominantly in spleen and lymph nodes, but at a low frequency also in the brain Two week old mice were intracerebrally (i.c.) infected with 10 3 PFU recombinant MV expressing the rodent brain-adapted haemagglutinin CAMH and eGFP, or, when indicated, the recombinant MV without eGFP. Both recombinant viruses (rMV Edtag EGFP-CAMH and rMV Edtag CAMH) have the same distinct tropism for mouse neurons, and infections cause similar acute and persistent CNS infections [24, 33, 34, 35] . These recombinant viruses were designated throughout the manuscript as rMV-green and rMV. To determine the number of Tregs present in our infection model in secondary lymphoid organs and the brain, C57BL/6 mice were infected with rMV-green and analyzed 3, 7, 10, 14 and 28 days post infection. Lymphocytes were isolated from 6 draining cervical lymph nodes (LN), the spleen, and the brain of MVinfected and control (i.c. injected PBS) animals. The fraction of CD4 + CD25 + Foxp3 + T cells of all lymphocytes was determined to be 1.5-3% in the spleen and 4-6% in LN ( Fig. 1 A, B ) with no significant difference between infected and control animals. This corresponds to a total number of Tregs of approximately 1610 6 in the spleen and 2610 5 in the prepared LN at day 28 post infection. Only a small statistically not significant number of Foxp3 + T cells was detected in the brain, irrespective of the infection (data not shown). To obtain statistically significant results and to reduce standard errors caused by the staining procedure, we used DEREG mice, which express GFP in Foxp3 + cells. After infection of these mice with rMV (not expressing GFP), a small but significant number of GFP + cells, approximately 800 per brain, was detected in the brains, whereas almost no GFP + cells were detected in controls ( In order to find out whether Tregs have an influence on the persistent CNS infection we first expanded and functionally activated Tregs using the superagonistic anti-CD28 monoclonal antibody clone D665 (mAb D665) [36] . This mAb, while causing a transient expansion of the total lymph node and splenic T cell population, predominantly expands functional CD4 + Tregs without causing systemic cytokine release [37] . In order to test the efficiency of mAb D665 in young mice, we measured Tregs in a preliminary experiment in uninfected mice 3 days after a single i.p. injection of 100 mg D665 by flow cytometry. The proportion of CD4 + Foxp3 + Tregs increased in the spleen and lymph nodes approximately 2fold ( Fig. 2 A; n = 4; P,0.01). This result indicated that mAb D665 can be used for expansion of Tregs also in young mice. For the following experiments, to investigate the effect of mAb D665 on the persistent viral infection, we applied 100 mg at day 14 and 21 post infection, and analyzed the brains at 28 dpi (Fig. 2 B ). Under these conditions the total number of lymphocytes increased significantly after D665 treatment in spleen and LN, but not in the brain (Fig. 2 C) . Furthermore, the proportion of Foxp3 + Tregs increased significantly (approximately 2-fold, P,0.01) in spleen in LN (Fig. 2 D) . Interestingly, histological analyzes revealed that the extent of the viral infection increased considerably after D665 treatment. Large clusters and groups of bright GFP-positive (directly reflecting virus replication) infected neurons emerged in the brains of these mice (Fig. 3 A) . This was in striking contrast to the brains of untreated mice at 28 dpi, which contain only a limited number of infected neurons (Fig. 3 B) . It is also strikingly different from brains at 14 dpi, when the treatment with D665 begins. At this time point areas with weak (vanishing) GFP expression are observed, that represent areas from which virus is being eliminated (Fig. 3 C) . The quantification demonstrated approximately 100-fold more infected cells per brain in D665treated animals in comparison to control animals at 28 dpi ( Fig. 3 D, compare lanes 3 and 5; differences were highly significant: P,0.0001). Two weeks later, at 42 dpi, the number of infected cells was reduced, but still higher than in control animals ( For the depletion of Tregs we used transgenic mice expressing the human diphtheria toxin (DT) receptor under the control of the Foxp3-promoter, which can be treated with DT to eliminate specifically Foxp3 + Tregs (DEREG-mice) [38] . To test the activity of the DT-batch used, adult DEREG mice were in a pilot experiment treated with DT under standard conditions (1 mg DT i.p. injected at 6 consecutive days) and analyzed the next day. Lymphocytes isolated from the spleen and lymph nodes were analyzed by flow cytometry to demonstrate the successful depletion of Foxp3 + GFP + Tregs by more than 90% (Fig. 4 A) . In order to assess the effect of Treg depletion on virus persistence in young mice, persistently infected mice were treated at 3 consecutive days (at day 17, 18, and 20 post infection) with 1 mg DT and analyzed at day 28 post infection ( Fig. 4 B) . In these mice, the viral infection was again quantified histologically in subsequent brain slices through the complete cerebrum as described earlier [33] . Treg depletion in DEREG 2/+ mice (compared to DT- treated DEREG 2/2 mice as control animals) led to a significant reduction of the number of infected neurons from brains of persistently infected animals (n = 6, P = 0.0098; Fig. 4 C) . In order to investigate whether this increase in virus clearance correlates with the presence of higher numbers of virus-specific CD8 + T cells, we first identified T cell receptor recognized peptides presented by MHC class I of C57BL/6 mice (H-2 b ). Two CD8 + T cells were detected in the brain at 7, 10, 14, and 28 dpi (Fig. 5 A) . Interestingly, high percentages of D b MV-H 22-30specific CD8 + T cells are present in the brain during the persistent phase of the infection (at 28 dpi approximately 18% of all CD8 + T lymphocytes). Mock-treated animals (i.c. PBS injection; ctrl) did not contain virus-specific CD8 + T cells. After treatment of DEREG mice with DT, the number of Tregs decreased by approximately 95%, while the number of CD8 + T cells in brains slightly increased (Fig. 5 B) . Interestingly, after depletion of Tregs, the fraction and absolute number of D b MV-H 22-30 -specific CD8 + T cells increased from 2,000 to 8,000 cells per brain, or 5% to 23% of CD8 + T cells (n = 3, P = 0,05; Fig. 5 C) . Thus, depletion of Tregs during the persistent phase of infection led to an increase of virus-specific CD8 + T cells and a significant reduction of the persistent infection in the brain. Our results indicate that the immune system keeps the ''hidden'' persistent viral CNS infection under permanent control. Manipulation of Tregs in the periphery had significant consequences for the fate of the viral infection in the brain, although only few infected neurons are present during the persistent phase of the infection. Expanding the number of Tregs by superagonistic CD28 antibody D665 led to an activation of virus replication and dramatic increase of the number of infected neurons, whereas transient depletion of Tregs by DT led to a significant reduction of the number of infected neurons. Interestingly, complete elimination of virus and clearance of the infection was not achieved by transient depletion of Tregs suggesting that additional means of an antiviral immune response are required for complete clearance. Looking for Tregs in the brain, we did not detect a significant number of CD4 + Foxp3 + Tregs by FACS-analysis after staining with Foxp3-specific antibodies, and only a small number of GFP- Figure 3 . Expansion of T lymphocytes with the superagonistic CD28 antibody D665 induces virus replication and spread. Consecutive coronal brain sections (100 mm sections) were prepared from complete rMV-green-infected mouse cerebra and analyzed using the UV microscope. Overviews and details of a typical section of an infected brain of a mouse treated with mAb D665 and analyzed at 28 dpi (A), and sections of infected control animals in the absence of mAb D665 at 28 dpi (B), and 14 dpi (C) are shown. The numbers of infected eGFP + cells per brain (sections through the complete cerebrum of each animal were evaluated as described [33] ) were determined microscopically in infected control C57BL/6 mice at 7, 14, 28 and 42 dpi (D, lanes 1-4) and in D665-treated mice at 28 and 42 dpi (D, lanes 5 and 6). The difference between control and D665-treated mice at 28 dpi was highly significant (P,0,0001). doi:10.1371/journal.pone.0033989.g003 expressing Foxp3 + cells in infected DEREG mice. These findings suggest that only very few Tregs, if any at all, are required within the brain for the regulation of effector immune cells, and that this regulation predominantly takes place in secondary lymphoid organs. Sellin et al. observed an increase of Foxp3 + Tregs as a consequence of MV infection of CD150-transgenic mice in spleen and brain [32] , which suggests that also the MV-infection itself supports the expansion and migration of Tregs, and that these infection-induced Tregs may be part of the multifactorial MVinduced immunosuppression [3, 4] . Tregs can restrain effector T cell responses through the production of immunomodulatory cytokines, such as TGF-b, IL-10, and IL-35, expression of inhibitory ligands, such as CTLA-4 and LAG-3, cytokine consumption, and direct cytolysis. It remains to be elucidated which of these mechanisms are involved in the persistent brain infection with MV. Several reports support the view that T cells play a decisive role in control of viral CNS infections. Using primary human lymphocytes, in vitro experiments demonstrated that CD8 + T lymphocytes control the dissemination of MV [39] . In resistant mouse strains the depletion of the CD4 + T cell subset by monoclonal antibodies led to a breakdown of resistance to the infection, whereas depletion of CD8 + T cells had no effect [23, 40] . In TAP-transporter deficient mice, which cannot present antigen on MHC class I molecules, MV was found to spread more transneuronally than in brains of normal mice [41] . These findings indicated that infected neurons are target cells of CD8 + lymphocytes, and that brain infections to some extent can be inhibited by CTL activity. Further investigations revealed that CD4 + T cells are able to protect either alone (in resistant mouse strains), or through cooperation with CD8 + T cells (in mice with intermediate susceptibility), and that CD8 + T cells are able to protect alone after immunization of the mice [42, 43] . Using CD46transgenic Rag-deficient mice, adoptive transfer of lymphocytes revealed that the combined activity of CD4 + T lymphocytes with CD8 + T cells or B cells is required to control the intracerebral infection [44] . Thus, most findings support the view that CD8 + T cells play an important role in the control of transneuronal virus spread, and our findings suggest that MV-specific CD8 + T cells are involved in maintaining the steady state and control of infection during the persistent phase of CNS infection. This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of and control (DEREG 2/2 ) mice both infected i.c. with rMV-green and treated with DT. The reduction of mean values from 50 to 8 was significant, with P = 0,0098. The number of infected eGFP + cells per brain was determined by microscopic evaluation of 100 mm sections through the complete cerebrum as described [33] . doi:10.1371/journal.pone.0033989.g004 Laboratory Animals of the National Institutes of Health. The protocol was approved by the Committee on the Ethics of Animal Experiments of the University of Würzburg (Permit Number: 55.2-2531.01-67/06). Specific pathogen free C57BL/6 mice were purchased from Harlan-Winkelmann, Germany. DEREG mice [38] were breeded in the animal facilities of the Institute for Virology and Immunobiology and the Centre for Experimental Medicine, Würzburg. Mice were anesthetized using isofluran and infected intracerebrally (i.c.) into the left hemisphere with 20 ml virus suspension containing 1610 3 PFU at an age of two weeks. To expand Tregs, mice were treated intraperitoneally (i.p.) with 100 mg superagonistic anti-CD28 monoclonal antibody (mAb) D665 [36] , and to deplete Tregs, DEREG mice were treated i.p. with 1 mg diphtheria toxin (DT; Merck) [38] . Vero cells (African green monkey; ATCC CRL 6318) were cultured in Eagle's minimal essential medium (MEM) containing 5% fetal calf serum (FCS), 100 U/mL penicillin and 100 mg/mL streptomycin. Recombinant measles viruses expressing the rodent adapted haemagglutinin of the strain CAM/RB (CAMH) and/or not the enhanced green fluorescent protein (eGFP) rMV Edtag EGFP-CAMH (rMV-green) and rMV Edtag CAMH (rMV) [35, 45] were propagated using Vero cells. For analyzes, animals were anesthetized with CO 2 and perfused with 4% (w/v) paraformaldehyde (PFA). Brains were fixed in 4% PFA at least 18 h, and free-floating sections (100 mm) were prepared using a vibratome (Technical Products International) as described [33, 35] . Slices were analyzed directly by UV microscopy. Photomicrographs were taken with a digital camera (Leica). Numbers of infected eGFP-positive neurons were counted and statistical analyzes done using the student's t test and the program Prism (GraphPad, Inc.). Isolation of lymphocytes from lymph nodes, the spleen, and the brain Draining cervical lymph nodes and the spleen were pressed through a steel sieve in 4 ml HBSS and homogenized in a total volume of 13 ml HBSS. After a centrifugation step at 310 g for 10 min the cell pellets were resuspended in an adequate volume of HBSS (approximately 10 7 /cells/ml). Spleen cells were additionally treated with erythrocyte lysis buffer (155 mM NH 4 Cl, 10 mM KHCO 3 , 0.1 mM EDTA) and washed with HBSS. Brains were pressed through a steel sieve in 5 ml HBSS 3% FCS and homogenized in a total volume of 20 ml HBSS 3% FCS. After a centrifugation step at 170 g for 10 min the cell pellet was resuspended in 1.4 ml dissociation buffer (23 mM CaCl 2 , 50 mM KCl, 42 mM MgCl 2 , 153 mM NaCl) containing 0.4 U collagenase (Serva) and 50 U Benzonase (Novagen) and incubated at 37uC for 1 h. Afterwards the cells were washed with HBSS and applied on a percoll density gradient to separate the lymphocytes from the rest like myelin debris or neuronal cells as described [46] . The lymphocytes were isolated and washed to remove the percoll for subsequent analyzes. mouse CD3 (clone 145-2C11)-, CD4 (clone RM4-5)-, CD8 (clone Ly-2)-, CD19 (clone 1D3)-and CD25 (clone 7D4)-antibodies were purchased from Becton Dickinson. PE-conjugated anti-mouse Foxp3 (clone FJK-16s)-antibody was purchased from NatuTec. Lymphocytes were stained in FACS buffer (PBS containing 0.4% BSA and 0.02% sodium azide) at 4uC for 20 min. Intracellular staining of Foxp3 was performed using the Foxp3 Staining Buffer Set (NatuTec) according to the manufacture's protocol. Briefly, cells were fixed and permeabilized in 500 ml fixation/permeabilization buffer (Concentrate/Diluent 1:4) at RT for 1 h and stained afterwards in permeabilization buffer at RT for 30 min. Flow cytometric analysis was performed on a FACSCalibur (Becton Dickinson). For identification of peptides presented by MHC class I that can be used in ELISPOT and pentamer staining experiments we used the software programs SYFPEITHI (University of Tübingen, Germany) and BIMAS (BioInformatics and Molecular Analysis Section, National Health Instituts, Bethesda, USA) to establish a ranking of potential peptides. From 12 potential H-2 K b and D bpresented peptides of MV-N and MV-H with the highest probability scores, we found that D b MV-H 22-30 and D b MV-H 446-454 (RIVINREHL and SNHNNVYWL, respectively) were most efficiently recognized. ELISPOT experiments were performed using the Mouse IFN-c ELISPOT set (BD Biosciences) according to the manufacture's protocol. MHC class I (H-2D b ) pentamers presenting the selected peptide MV-H 22-30 (D b MV-H 22-30 -pentamers) were ordered from ProImmune Ltd (Oxford, UK). Cells were washed with FACS buffer and stained with 5 mL pentamer-solution diluted in 100 ml FACS buffer at 4uC for 30 min. After one washing step the cells were analyzed using the FACSCalibur. MV-specific cells were gated as CD8 + and CD19-negative lymphocytes to exclude pentamer + CD19 + cells.

Superimposed high-frequency jet ventilation combined with continuous positive airway pressure/assisted spontaneous breathing improves oxygenation in patients with H1N1-associated ARDS

BACKGROUND: Numerous cases of swine-origin 2009 H1N1 influenza A virus (H1N1)-associated acute respiratory distress syndrome (ARDS) bridged by extracorporeal membrane oxygenation (ECMO) therapy have been reported; however, complication rates are high. We present our experience with H1N1-associated ARDS and successful bridging of lung function using superimposed high-frequency jet ventilation (SHFJV) in combination with continuous positive airway pressure/assisted spontaneous breathing (CPAP/ASB). METHODS: We admitted five patients with H1N1 infection and ARDS to our intensive care unit. Although all patients required pure oxygen and controlled ventilation, oxygenation was insufficient. We applied SHFJV/CPAP/ASB to improve oxygenation. RESULTS: Initial PaO(2)/FiO(2 )ratio prior SHFJV was 58-79 mmHg. In all patients, successful oxygenation was achieved by SHFJV (PaO(2)/FiO(2 )ratio 105-306 mmHg within 24 h). Spontaneous breathing was set during first hours after admission. SHFJV could be stopped after 39, 40, 72, 100, or 240 h. Concomitant pulmonary herpes simplex virus (HSV) infection was observed in all patients. Two patients were successfully discharged. The other three patients relapsed and died within 7 weeks mainly due to combined HSV infection and in two cases reoccurring H1N1 infection. CONCLUSIONS: SHFJV represents an alternative to bridge lung function successfully and improve oxygenation in the critically ill.

The swine-origin 2009 H1N1 influenza A virus has become the predominant influenza virus worldwide since its identification. H1N1 influenza might cause acute respiratory distress syndrome (ARDS) and potentially result in extracorporeal membrane oxygenation (ECMO) therapy [1] . Experience from Australia and New Zealand describes an incidence of mechanical ventilation in 64.6% of H1N1 patients and 11.6% ECMO (with a mortality of approximately 21%) treated within the intensive care unit (ICU) [2] . Mortality rates of ARDS patients suffering from hypoxemia and/or hypercapnia remains high [3] . In this situation, ECMO therapy represents the standard of bridging lung function [2] . However, ECMO therapy is associated with unfavorable complications and high cost. In this report, we suggest an alternative strategy to bridge lung function: the use of superimposed high-frequency jet ventilation (SHFJV) in combination with continuous positive airway pressure/assisted spontaneous breathing (CPAP/ASB). This alternative ventilation strategy is based on jet ventilation to improve oxygenation at lower plateau pressure combined with assisted spontaneous breathing by CPAP/ASB to improve CO 2 removal. In winter 2009-2010, five patients with ARDS (H1N1) were admitted to our ICU due to inadequate oxygenation by conventional mechanical ventilation from outside hospitals. ARDS was defined according to the definition of the American-European Consensus Conference on ARDS [4] . Virological and microbiological tests were performed in bronchioalveolar lavage under sterile conditions. Influenza A H1N1 infection was determined by realtime reverse transcriptase-polymerase chain reaction (RT-PCR) assay. In addition, herpes simplex virus (HSV)-DNA, cytomegalovirus (CMV)-DNA, Epstein-Barr virus (EBV)-DNA, RSV, and adenovirus were tested by TAQMAN PCR. Atypical species, such as Mycoplasma, Pneumocystis jiroveci, or Legionella were screened through PCR. BAL, swabs, and blood cultures were taken routinely on admission and depending on clinical need. Patients were ventilated with a Monsoon I or Monsoon III jet ventilator (Acutronic Medical Systems, Hirzel, Switzerland, distributed by IfM GmbH, Wettenberg, Germany) combined with a standard respirator, Evita XL (Draeger, Lübeck, Germany) or Hamilton G5 (Hamilton Medical, Bonaduz/Switzerland, distributed by Heinen & Löwenstein GmbH, Bad Ems, Germany). • Spontaneous breathing within 12 h after admission • Positive end-expiratory pressure (PEEP) primary 12-15 mbar • Ppeak (peak pressure) < 30 mbar • Pmean < 20 mbar • FiO 2 , work pressure, and frequency of jet, ASB in adaption to blood gas analyses; goal: oxygen delivery (DO 2 ) normal, lactate normal, pH 7.3-7.45 • Supine position 135°up to 12-14 h as long as oxygenation benefits • ECMO (QUADROX PLS and ROTAFLOW RF 32, MAQUET GmbH &Co. KG, Rastatt, Germany) Interventional Lung Assist (ILA Membranventilator ® , Novalung GmbH, Talheim, Germany) when CO 2 > 50 mmHg, pH < 7.3 and no progress by conservative ventilation and sedation management To achieve a RASS (Richmond agitation and sedation scale) of -3 to -4 propofol or midazolam, in two cases additionally with γ-hydroxybutanoic acid was used in combination with remifentanil or sufentanil and clonidine. All patients received oseltamivir, two patients in combination with ribavirin and amantadine up to 9 days. HSV was treated as standard with acyclovir; two patients received foscarnet-sodium when CMV coinfection was suspected or proven. Five patients with severe influenza A H1N1 infection and refractory hypoxemia were ventilated between 2 and 16 days before admission. The transfer of these patients to our ICU was due to failure to improve during conventional ventilation (PEEP 10-15 mbar, pressure-controlled ventilation with P peak up to 35 mbar, FiO 2 1.0). On admission, patients presented with a paO 2 /FiO 2 ratio of 58 to 79 (Murray score > 3.0; Table 1 ). After institution of SHFJV, oxygenation improved in all patients within 24 h (paO 2 /FiO 2 ratio 105-306) with a PEEP of 8 to 15 mbar, P peak < 30 mbar, and P mean 15-18 mbar. Therefore, the institution of SHFJV allowed improved oxygenation with adequate DO 2 and a simultaneous reduction of lung injury, inducing high peak and mean pressures at the same time. Spontaneous breathing was set during 8 h after admission. Three patients required jet ventilation for up to 72 h. One patient required prolonged jet ventilation due to concomitant mycoplasma pneumonia, and the other due to persistent H1N1 pneumonia. FiO 2 was reduced from initially 1.0 to 0.4-0.6 in four cases. However, one patient with a ventilation period of 14 days before admission due to H1N1 and concomitant Pneumocystis jiroveci infection needed a FiO 2 of 0.9 up to day 7. Jet ventilation was performed in three cases with a work pressure of 1.2-1.7 bar, a frequency of 140-200/ min (Monsoon I Jet ventilator); in two patients a novel jet ventilator was used and the work pressure was set to 0.5-0.9 bar and the frequency adjusted to 500-600/min (Monsoon III Jet ventilator). When only H1N1 infection was present, the oxygenation deficit was the leading clinical symptom. In cases with prolonged treatment (reinfection or concomitant infection with HSV, Mycoplasma) hypercapnia also was notable. Therefore, one patient received ECMO and one patient ILA therapy (Figure 1 ). The patient with ECMO therapy demonstrated improved oxygenation and normocapnia but deceased due to intracranial bleeding. This patient had histologic confirmed HSV pneumonia for 4 weeks without improvement after acyclovir/foscarnet treatment. A similar clinical pathway was observed in the two other patients who died. Despite pulmonary failure, organ dysfunction was moderate. Some patients showed liver dysfunction; two of three nonsurvivors had acute renal failure. Vasoactive drugs were needed initially in all patients. From anamnesis, four patients were obese; one patient had pre-existing medical complications (acute CMV colitis and acute Pneumocystis jiroveci pneumonia by ulcerative colitis). None of the patients had pre-existing immunodeficiency, such as HIV. During ARDS, ECMO is seen as the standard therapy to improve oxygenation. The CESAR study demonstrated that in highly specialized centers, ECMO therapy may improve long-term outcome. However, this study had limitations within the study design, because the intervention and control group were treated at different centers [5] . In a review of 62 cases with ECMO therapy, the survival rate was 55% [6] ; 17.8% of patients died as a result of ECMO-related complications, translated into a mortality of 8% due to ECMO alone. A retrospective U. S. database analysis (1986-2006) demonstrated a mortality rate of 50% in patients treated with ECMO. Although technical improvement of ECMO equipment translated into fewer cases of circuit rupture, renal insufficiency, pulmonary hemorrhage, inotropic medications, hyperglycemia, extremes of pH, arrhythmias, or hypertension became more common [7] . The primary goal for ARDS patients is to preserve an acceptable gas exchange without further injury to the lungs. High-frequency jet ventilation (HFJV) and highfrequency oscillatory ventilation (HFOV) are characterized by rapid delivery of small tidal volumes (V t ) (1-3 ml/kg predicted body weight) at high frequencies (2.5-5 Hz). As such, P mean during jet ventilation is comparable to the pressure level of PEEP in conventional mechanical ventilation due to its minimal V t . Yet, P peak and P mean are markedly reduced with a reduction of tidal hyperinflation and tidal decruitment [8] . Furthermore, oxygen toxicity may be reduced as lower FiO 2 often are required compared with mandatory ventilation [8] . This beneficial effect could be demonstrated in animal models of HFOV. In addition, pulmonary compliance improved, the infiltration of polymorph nuclear leukocytes was reduced, and tumor necrosis factor-α levels attenuated. As a result, injury to alveoli and membranous bronchioles was reduced [9] . HFOV mainly differs from HFJV by active in-and expiration, whereas HFJV only uses active inspiration and passive expiration. Thus, HFJV allows patients spontaneous breathing, preserving muscles of diaphragm and intercostal musculature. This is a major advantage, because early spontaneous breathing significantly reduces deleterious organ-organ interactions, e.g., lungliver interactions, and improves liver perfusion [10] . One major problem of jet ventilation is CO 2 clearance due to very low V t . There are two options in resolving this problem. The simplest approach is to establish spontaneous breathing. However, patients might be sedated and therefore need superimposed CPAP/ASB to improve tidal inflation until sedation is adapted. In the case of severe hypercapnia, the use of Interventional Lung Assist (ILA) can improve CO 2 clearance. We observed a high mortality rate, which can be -explained, in part, by patient selection. All of our patients presented in this report were admitted with fixed hypoxemia despite optimized conventional ventilation. Two patients presented a prolonged period of hypoxemia and mechanical ventilation before admission. One of these patients was already on mechanical ventilation 14 days due to septic CMV-colitis and Pneumocystis jiroveci pneumonia with poor prognosis. The other patient was admitted to our ICU after a prolonged period of severe hypoxemia and septic shock with hypoperfusion. All five patients had a HSV-PCR-positive BAL, confirmed via lung biopsy in one patient. Viral coinfection with HSV may cause and prolongs a status of persistent immunosuppression during severe H1N1 infection, which is associated with a high risk of death. This hypothesis is supported by findings of Monsalvo and colleagues, who found pathogenic immune complexes as previously unknown biological mechanism for the unusual age distribution of severe H1N1 infections [11] . Although jet ventilation is actually not in the focus of ARDS treatment, we were able to demonstrate that SHFJV represents an alternative of lung-protective ventilation during ARDS. Compared with ECMO, it is easier to use, associated with fewer complications, cost-effective, and can be used in secondary and tertiary centers. Therefore, SHFJV is an alternative approach to improve lung function and oxygenation in patients suffering from ARDS. Nevertheless, this report is limited by the small study size, including a heterogeneous patient collective. Therefore, a controlled study with SHFJV to treat patients with ARDS is required for an evidencebased conclusion.

Serious Invasive Saffold Virus Infections in Children, 2009

The first human virus in the genus Cardiovirus was described in 2007 and named Saffold virus (SAFV). Cardioviruses can cause severe infections of the myocardium and central nervous system in animals, but SAFV has not yet been convincingly associated with disease in humans. To study a possible association between SAFV and infections in the human central nervous system, we designed a real-time PCR for SAFV and tested cerebrospinal fluid (CSF) samples from children <4 years of age. SAFV was detected in 2 children: in the CSF and a fecal sample from 1 child with monosymptomatic ataxia caused by cerebellitis; and in the CSF, blood, and myocardium of another child who died suddenly with no history of illness. Virus from each child was sequenced and shown to be SAFV type 2. These findings demonstrate that SAFV can cause serious invasive infection in children.

M olecular biology has revolutionized the diagnostics of infectious diseases through the introduction of more sensitive and specifi c diagnostic tests. Despite these advances, the etiologic agents of many apparent infections are still unidentifi ed. For example, the etiologic agent is unknown for many cases of apparent pneumonia (1); in a study conducted in California, USA, despite extensive testing and evaluation, an underlying cause of encephalitis was unidentifi ed for 207 (62%) of 334 patients (2) . During the past few years, intensive searches for new viruses, using conventional virologic methods and metagenomics, have resulted in the discovery of several new viruses. During the past decade, the family Picornaviridae has grown as the number of recognized genera has increased from 6 to 12 (3, 4) ; the numbers of species, types, and subtypes have increased even more. However, only viruses from 3 genera (Enterovirus, Hepatovirus, and Parechovirus) have been fi rmly established as being capable of causing clinically signifi cant disease in humans. Viruses from other genera (Cardiovirus, Cosavirus, and Kobuvirus) have so far been detected only in noninvasively collected human sample material such as fecal and respiratory samples, and their clinical signifi cance remains to be fully elucidated. (Invasively collected sample material is that from tissues considered sterile, i.e., devoid of microorganisms.) The phylogenetic relationships of human picornaviruses are shown in Figure 1 . Most picornaviruses that are pathogenic to humans are ubiquitous viruses capable of causing a variety of diseases, from monosymptomatic febrile infection to severe infection in the central nervous system and myocardium. However, most infections with these viruses are asymptomatic (5) . Saffold virus (SAFV) was discovered by Jones et al. in 2007 by sequence-independent genomic amplifi cation of virus isolated from a fecal sample (6) . The sample had been obtained in 1981 from an 8-month-old child with fever of unknown origin. The genetic sequence of the virus indicated that the virus belonged to the species Theilovirus of the genus Cardiovirus, which contains 3 other members: Theiler's murine encephalomyelitis virus (TMEV), Vilyuisk human encephalomyelitis virus (VHEV), and Thera virus. In mice, TMEV is capable of causing infection in the central nervous system, and some variants of this virus cause a persistent infection and even multiple sclerosis -like disease (7) . VHEV was isolated in the 1950s from cerebrospinal fl uid (CSF) from a patient with Vilyuisk encephalomyelitis, a progressive neurologic disorder that occurs in indigenous populations of an isolated part of eastern Siberia (8) . However, the correlation between VHEV and Vilyuisk encephalomyelitis is still uncertain because VHEV has been isolated by multiple passages in mice and thus may represent a highly divergent strain of TMEV. Thera virus (previously named Theilerlike rat virus) has been isolated from rats, but the clinical signifi cance of infections with this virus is unknown (9, 10) . The genus Cardiovirus also contains a second species called encephalomyocarditis virus. Only 1 serotype is known, and it is capable of causing encephalitis and myocarditis in various animals (11, 12) . Since the discovery of SAFV, several articles have provided insight into its epidemiologic and, to a minor degree, clinical signifi cance. Saffold viruses are distributed worldwide (6, (13) (14) (15) (16) (17) (18) (19) , and 2 serologic studies have demonstrated that infection occurs early in life (14, 20) . However, fi nding an association with human disease has thus far been elusive. Most studies (13, 15, 17, (20) (21) (22) have tried to associate SAFV with gastroenteritis, but no convincing results have been produced. A few studies (16, 18, 21, 23) analyzed the clinical signifi cance of SAFV virus in the respiratory system, but no substantial association between the virus and respiratory symptoms or disease has been made. Only 1 study (21) reports having tested invasively collected sample material (CSF samples), but no fi ndings were positive. To investigate the possible invasive potential of SAFV in humans, we developed a diagnostic PCR and tested CSF samples from a group of children. SAFV was detected in 2 of these children. We tested previously submitted CSF samples for SAFV, reviewed the patients' medical records, and sequenced the viruses isolated. We selected fecal samples from 479 children <5 years of age with gastroenteritis that had been submitted for viral diagnostic testing from September 2009 through February 2010 and tested them for SAFV. Nucleic acid extracted from these samples was combined into 48 pools, with 9 or 10 samples per pool. Samples from pools with positive results were identifi ed, and new extractions from these pools were tested individually. However, enough sample material for new extractions was available for only about half of the samples. Nucleic acids were extracted from 200 μL of CSF or blood (from SAFV-positive patients) by using the QIAamp DNA Blood Mini Kit (QIAGEN, Hilden, Germany) and semiautomatic extraction on the QIAcube instrument (QIAGEN). Nucleic acid was extracted from 200-μL fecal suspension (10% in phosphate-buffered saline) by using the Total Nucleic Acid Isolation Kit (Roche Diagnostics GmbH, Mannheim, Germany) on the MagnaPure LC instrument (Roche Diagnostics GmbH). Five microliters of extracted material was used per reverse transcription PCR (total volume 25 μL) by using the OneStep RT-PCR Kit (QIAGEN). The reaction mixtures contained 1 μmol/L of each primer and 0.2 μmol/L of probe. Design of the primers and probe was based on an alignment of all available SAFV sequences in GenBank (www.ncbi.nlm.nih.gov/genbank) in July 2010 by using ClustalW (www.clustal.org) and Primer3 (http:// frodo.wi.mit.edu/primer3) software. The primers and probe are selective for a highly conserved part of the 5′ untranslated region (Table) . The Strategene Mx3005P real-time thermocycler instrument (Agilent Technologies A/S, Horsholm, Denmark) was used for amplifi cation and detection with the following settings: 50°C for 20 min, 95°C for 15 min, followed by 45 cycles of 95°C for 15 s and 55°C for 1 min. Genotyping was conducted by nested PCR and sequencing of parts of the viral protein (VP) 1 and VP2 regions of the capsid gene by using primers listed in the Table. The inner VP1 and VP2 primers amplifi ed DNA fragments of ≈599 and 577 bp, respectively. PCR products were purifi ed by using the High Pure PCR Purifi cation Kit (Roche Diagnostics GmbH) before sequencing, which was performed by using the inner PCR primers on an ABI automated sequencer and BigDye version 1.1 chemistry (both from Applied Biosystems, Darmstadt, Germany). Sequences were aligned, and phylogenetic analysis with known reference sequences was performed by using MEGA4 software (www.megasoftware.net). The sequences have been submitted to GenBank under accession nos. JF693612-23. SAFV was detected in CSF from 2 of the 319 children. Additional sample material from these 2 children was subsequently obtained and tested. From child 1, blood and CSF collected at the same time and a fecal sample collected 2 weeks later were tested; only test results for the fecal sample were positive for SAFV. From child 2, a postmortem blood sample and a myocardial biopsy sample were tested; test results for each sample were positive for SAFV. Child 1 was a 16-month-old, previously healthy boy who became ill in May 2009. The boy had a fever 6 days before hospital admission, followed 1 day later by sudden onset of monosymptomatic ataxia, with no fever. The ataxia fl uctuated from causing an insecure gait to walking into things and falling. The patient also had intermittent diffi culty controlling his arms when trying to eat. No history of recent travel was reported. The boy was in otherwise good health; he had no abnormal psychological symptoms and retained a normal degree of consciousness throughout the acute phase of the disease. Differential diagnoses at hospital admission were intracranial tumor or viral cerebellitis. The boy's 4-year-old sister remained healthy. Laboratory test results are listed in the online Appendix Table ( wwwnc.cdc.gov/EID/article/18/1/11-0725-TA1. htm). At hospital admission, CSF values (leukocyte count, protein level, and glucose level) were within reference ranges, and no microorganisms were detected. A magnetic resonance imaging scan of the brain showed a small venous anomaly in the left frontal lobe but no tumor, hemorrhage, or infl ammation. A fecal sample collected 2 weeks later yielded positive test results for parechovirus type 3 and negative results for enterovirus and adenovirus. Parechovirus was not found in the CSF or blood. During the next 2 months, the ataxia remitted completely without sequelae. The diagnosis at this time was viral encephalitis, possibly caused by parechovirus type 3. Child 2 was a 27-month-old, previously healthy girl who was found dead in her bed in August 2009; she had no Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 18, No. 1, January 2012 9 known history of disease or symptoms. During necropsy, signs of cerebral herniation were detected. A small vascular malformation surrounded by edema was found in the brain. No signs of encephalitis or hemolytic uremic syndrome were found. The following sample materials were collected: CSF, blood, feces, myocardium, pericardial aspirate, lung tissue, and respiratory secretions (online Appendix Table) . In the CSF, mononuclear pleocytosis was noted. Results of routine bacteriologic culture found coagulasenegative staphylococci in the CSF, a few nonhemolytic streptococci in the lung tissue, and a few nonhemolytic streptococci and a few Staphylococcus aureus organisms in the pericardial aspirate. Verotoxin-producing Escherichia coli was cultured from the fecal sample. The conclusion of the autopsy and laboratory fi ndings was that cerebral herniation was the immediate cause of death. Virus from each of the 2 children was characterized by sequencing part of the VP1 region of the capsid gene, and the sequences were compared with those of other SAFVs detected in fecal samples from the patients with gastroenteritis ( Figure 2 ). Of the 48 fecal sample pools, 10 were positive for SAFV by PCR. From these pools, 6 individual samples were available for further testing; the VP2 capsid region was successfully sequenced for 4 of these 6 samples, and they were all SAFV type 2. Later, an SAFV type 2-specifi c VP1 PCR was designed, which provided VP1 sequences from all 6 fecal samples and samples from the 2 children reported here: the fecal sample from child 1 and the blood sample from child 2. The phylogenetic analysis ( Figure 2) showed that all viruses were SAFV type 2 and that the sequences arranged in 2 clusters with 8% nt differences and 5% aa differences between the clusters. Several novel viruses have recently been discovered by using microarrays or mass-sequencing methods (6, 21) . Almost all of these viruses have been found in noninvasively collected patient materials, making the correlation to clinical disease diffi cult to establish. For the picornavirus group, this issue is further complicated by knowledge that this virus family can cause a wide variety of diseases with a high proportion of nonspecifi c symptoms or asymptomatic infections (5) . We report 2 cases of invasive infection with SAFV type 2 in children. For each child, SAFV was detected in at least 1 compartment other than the central nervous system. This fi nding strengthens the evidence of an acute infection as the cause of clinical disease. In child 1, SAFV virus was found in the CSF and a fecal sample. In child 2, fi ndings were even more convincing because SAFV was found in 3 invasively collected samples: CSF, myocardium, and blood. No other credible cause of infection was found for either of the 2 children. In child 1, the only other positive fi nding was parechovirus type 3 in a fecal sample taken 2 weeks after onset of disease. Parechovirus was not found in the CSF and therefore seems unlikely as the cause of the acute symptoms. From child 2, several types of bacteria were identifi ed, but these seem unlikely to be the cause of death because culture of postmortem samples often grows commensal organisms. The bacteria from this child were not found consistently in the tissue samples, thus making systemic bacterial infection less likely. Verotoxinproducing E. coli also seems an unlikely cause of death because no diarrhea or signs of hemolytic uremic syndrome were present before the child's death or at autopsy. At autopsy of child 2, a small vascular malformation surrounded by edema was found. This edema could be the cause of the herniation. However, the edema might also have been caused by cerebral infection or septicemia. This possibility is supported by the fi nding of mononuclear 10 Emerging cells in the spinal fl uid and the direct virus identifi cation in the CSF and blood. The exact cause of death and whether there is a connection between the infection and the changes surrounding the malformation are unclear. However, our investigations show that before her death, the child's blood contained SAFV. Child 1 had monosymptomatic ataxia preceded by 1 day of fever. The symptoms receded over the next few months, and the patient recovered fully. Child 2 died without any preceding symptoms or any known predisposing factors; this clinical picture is sometimes found for patients with enteroviral infections (24) . The fi nding of SAFV in invasively collected samples from the 2 children described here fi ts well with the knowledge about the picornavirus group. SAFV probably behaves similarly to the viruses in the enterovirus group, producing mainly asymptomatic infections but also producing nonspecifi c symptoms in other patients and severe disease in a few patients. Enteroviruses are known to cause many more or less organ-specifi c diseases such central nervous system infection (meningitis, encephalitis, and myelitis), myocarditis, enanthema, exanthema, and septicemia. These 2 cases fi t well within the expected range of diseases attributed to picornaviruses, but further studies are needed to determine the exact correlation of SAFV to disease in humans. In a previous study, Chiu et al. looked for SAFV in 400 CSF samples but found none (21) . Lack of detection could be explained by different assay sensitivities or by the selection of samples tested. In their study, patient selection was based on neurologic disease and patient ages were not reported. Our study has also shown that Saffold virus type 2 circulated in Denmark in 2009, the same year that the children reported here became ill. Other studies have shown type 2 to be a common type of SAFV and to be circulating worldwide (13, 15, 16, 18, 20, 21, 23) . Because SAFVs are single-stranded RNA viruses, a nucleotide variation of 8% among the restricted number of samples in this study is expected. In conclusion, we have established SAFV virus as a cause of invasive infection and a highly probable cause of severe disease in children. More studies are needed to further illuminate the role of SAFV as a human pathogen.

Pandemic (H1N1) 2009 among Quarantined Close Contacts, Beijing, People’s Republic of China

We estimated the attack rate of pandemic (H1N1) 2009 and assessed risk factors for infection among close contacts quarantined in Beijing, People’s Republic of China. The first 613 confirmed cases detected between May 16 and September 15, 2009, were investigated; 7,099 close contacts were located and quarantined. The attack rate of confirmed infection in close contacts was 2.4% overall, ranging from 0.9% among aircraft passengers to >5% among household members. Risk factors for infection among close contacts were younger age, being a household member of an index case-patient, exposure during the index case-patient’s symptomatic phase, and longer exposure. Among close contacts with positive test results at the start of quarantine, 17.2% had subclinical infection. Having contact with a household member and younger age were the major risk factors for acquiring pandemic (H1N1) 2009 influenza virus infection. One person in 6 with confirmed pandemic (H1N1) 2009 was asymptomatic.

I n early April 2009, human cases of infection with a novel infl uenza virus of swine origin, pandemic (H1N1) 2009 virus, were identifi ed in the United States and Mexico, and this virus spread rapidly across the world (1-3). On June 11, 2009 , the World Health Organization raised the pandemic level to 6, the highest level for pandemic alert (4) . Estimating attack rates is a major task in characterizing pandemic (H1N1) 2009. Some studies have reported attack rates of pandemic (H1N1) 2009 among household members and aircraft passengers (5) (6) (7) . These studies suggested that the transmissibility of pandemic (H1N1) 2009 virus was low. These studies were conducted in outbreak settings, and attack rates were calculated on the basis of clinical diseases that included infl uenza-like illness (ILI) or acute respiratory illness (ARI) of close contacts rather than confi rmed infection with pandemic (H1N1) 2009 virus. In addition, in these studies only symptomatic index and secondary cases were included. Although most infections of pandemic (H1N1) 2009 infl uenza virus produce ILI or ARI symptoms (8) (9) (10) (11) (12) , subclinical infection can occur and can change the estimate of attack rate. In addition, the infectivity of asymptomatic case-patients has not been clearly defi ned (13) . Because of the high rates of illness and death among the initial case-patients with pandemic (H1N1) 2009 (14) , the Chinese government decided to prevent and contain the rapid spread of disease through tracing and quarantine of persons who had close contact with persons with confi rmed cases of pandemic (H1N1) 2009. Beijing, the capital of the People's Republic of China, took strict containment and control measures through October 2009. The Beijing municipal government implemented border entry screening, ILI screening in hospitals, health follow-up of travelers from overseas, and quarantine and testing of close contacts to identify new introduction of cases and local transmission. Public health workers conducted epidemiologic investigation of all index case-patients (including those with subclinical infections) and traced and quarantined close contacts whose residence was within the jurisdiction of Beijing. We estimated the attack rate of pandemic (H1N1) 2009 virus infection and assessed risk factors or correlates for infection among different types of close contacts, including household members and aircraft passengers. In 2009, under the guidance of the Beijing Center for Disease Prevention and Control (Beijing CDC), a network of 55 collaborating laboratories was established to perform reverse transcription PCR testing to confi rm cases of pandemic (H1N1) 2009 (15) . The confi rmed cases included symptomatic and asymptomatic cases, and these cases were detected mainly by border entry screening, ILI screening in hospitals, health follow-up of travelers from overseas, and quarantine and testing of close contacts. Once confi rmed, index case-patients were immediately quarantined in designated hospitals to receive treatment while in isolation. All the confi rmed cases were required by law to be reported to Beijing and local CDCs. From May through October 2009, a detailed epidemiologic investigation was conducted for each confi rmed case of pandemic (H1N1) 2009 (including symptomatic and asymptomatic cases) by Beijing and local CDCs within 6 hours after confi rmation of infection. Patients with confi rmed cases were interviewed about demographic characteristics, course of illness, travel and contact history, and information about close contacts. Patients with confi rmed cases were categorized as having imported cases (travelers) and locally acquired cases (no travel history) on the basis of where the infection was acquired. Close contacts were defi ned as anyone who ever came within 2 meters of an index case-patient without the use of effective personal protective equipment (PPE) (including masks and gloves, with or without gowns or goggles) during the presumed infectious period. Trained staff from local CDCs made the determinations on the basis of fi eld investigation. The relationships of close contacts to index case-patients were categorized as 1) spouses, 2) other household members, 3) nonrelated roommates, 4) contacts at workplace or school, 5) nonhousehold relatives, 6) passengers on the same fl ight, 7) friends, and 8) service persons met at public places. A close contact on an aircraft was defi ned as a passenger sitting within 3 rows in front and 3 rows behind the index case-patient. All close contacts were traced and quarantined for 7 days after the most recent exposure to the index casepatient. All index case-patients detected between May 16 (the fi rst case, the date of confi rmation) and September 15, 2009 (before widespread transmission in Beijing), and their close contacts were included in this study. We excluded cluster or outbreak cases for which close contacts could not be determined clearly by epidemiologic investigation (the transmission chain was obscure). For each close contact, before quarantine, a pharyngeal swab specimen was collected for reverse transcription PCR testing, regardless of symptoms. A second pharyngeal swab specimen was collected for testing for pandemic (H1N1) 2009 virus if any of the following symptoms developed in a close contact during quarantine: axillary temperature >37.3°C, cough, sore throat, nasal congestion, or rhinorrhea. Data were analyzed by using SPSS version 11.5 (SPSS Inc., Chicago, IL, USA). Median and range values were calculated for continuous variables, and percentages were calculated for categorical variables. Differences in attack rates were compared between subgroups of close contacts by using the χ 2 test. For the signifi cant difference found in multiple subgroups, this test does not enable identifi cation of which multiple subgroups are signifi cantly different, only that across all the subgroups there are differences. The variables with p<0.10 in χ 2 test were included in multivariate analysis. Multivariate unconditional logistic regression analysis was conducted to determine risk factors associated with infection in close contacts. Backward logistic regression was conducted by removing variables with p>0.10. Odds ratios (ORs) and 95% confi dence intervals were calculated for potential risk factors of infection. The Hosmer-Lemeshow goodness-of-fi t test was used to assess the model fi t for logistic regression. All statistical tests were 2-sided, and signifi cance was defi ned as p<0.05. A total of 613 eligible index case-patients, detected from May 16 through September 15, 2009, were included in this study. Through fi eld epidemiologic investigations, 7,099 close contacts were traced and quarantined in Beijing. The median number of close contacts per index case per day was 7.0 persons (range 2.0-95.0 persons); the median number for an imported index case was 7.0 persons (range 1.7-95.0 persons) and for a locally acquired index case was 5.3 persons (range 1.0-25.0 persons). For the 601 symptomatic index case-patients, the median interval between illness onset and sample collection was 1.0 days (range −1.9 to 7.0 days). Among close contacts with symptomatic infection, the median interval between illness onset and sample collection was 0.5 days. More than 85% of close contacts were quarantined within 72 hours after interview of the index case-patients. The median interval between fi rst exposure and quarantine was 3.4 days for the close contacts, and it was shorter, on average, for fl ight passenger contacts than nonpassenger contacts (1.7 days vs. 3.8 days). For symptomatic close contacts infected with pandemic (H1N1) 2009, the median of generation time (i.e., the time from illness onset in an index case to illness onset in a secondary case) were 2.4 days; it was shorter for fl ight passenger contacts than nonpassenger contacts (1.6 days vs. 2.5 days) ( Table 1) . Approximately 43% of the index case-patients were women; the median age was 20 years, and 38% likely contracted pandemic (H1N1) 2009 virus locally because they had not traveled recently. Among the index casepatientss, 2% had subclinical infection. Only 18% of index case-patients had close contacts with confi rmed pandemic (H1N1) 2009 (Table 2) , and the total number of close contacts who were infected by the virus from 110 index case-patients was 167. Fifty percent (3,514 of 7,032) of close contacts were women, and the median age was 27 years. Approximately 12% of close contacts were household member of index case-patients (spouse or other household member), and aircraft passengers accounted for 44% of close contacts. Approximately 61% of close contacts were exposed to symptomatic index case-patients during their symptomatic phase. About 70% were quarantined in a quarantine station ( Table 2 ). The overall attack rate for infection among close contacts (positive test result) was 2.4% (167 of 7,099), indicating that 1 index case-patient transmitted infection to 0.27 close contacts (167 of 613) on average (reproduction number = 0.27). Among those close contacts with a positive test result, 14.4% (24 of 167) had subclinical infection; among the close contacts with positive test results at the start of quarantine, 17.2% (20 of 116) had subclinical infection. Attack rates did not differ by index case-patient's sex (p = 0.225). However, attack rates differed signifi cantly by index case-patient's age (p = 0.022), and the lower attack rate was found for older index case-patients. There was no signifi cant difference in attack rates between close contacts of patients with imported cases and those with locally acquired cases (p = 0.282). No infection was found in close contacts exposed to index case-patients with subclinical infection, and the attack rate observed in close contacts exposed to symptomatic index case-patients during their symptomatic phase was higher (p<0.001). Almost identical attack rates were found among male and female close contacts (p = 0.808). However, attack rates were signifi cantly different among different age groups of close contacts (p<0.001), and the lowest attack rate was found for those >50 years of age. The attack rates were signifi cantly different across 8 contact types (p<0.001). The attack rate was 5.3% among spouses and 6.6% among other family members in the household, and was lower among other types of close contacts ( Table 3 ).The attack rate among passengers on the same fl ight was low, 0.9% overall, and 1 index case-patient transmitted infection to 0.19 close contacts on a fl ight on average (28 of 147), and the attack rate was higher among the passengers with longer fl ight times (>12 hours, p = 0.001). The attack rate among close contacts of service persons at public places was 0.2%, and 1 index case-patient transmitted infection to 0.01 close contacts of service persons on average (1 of 113). Nonpassenger close contacts with longer exposure duration (>12 hours), compared with those with shorter duration (>12 hours), recorded the higher attack rate (p<0.001) ( Table 3) . By multivariate analysis, age and type of contact were the major predictors of infection ( (OR 3.42; p = 0.002) and 0-19 years of age (OR 7.76; p< 0.001) were at higher risk for infection. Other signifi cant independent risk factors associated with infection included being a household member of a person with an index case (OR 3.83; p<0.001), being exposed to index case-patients during their symptomatic phase (OR 1.86; p = 0.003), and exposure duration >12 hours (OR 1.83; p = 0.002). Similar risk factors were observed among aircraft passengers. We estimated that pandemic (H1N1) 2009 virus was transmitted by 18% of index case-patients to their close contacts and that 2.4% (167 of 7,099) of close contacts we traced were infected. Our data indicate that pandemic (H1N1) 2009 virus has low transmissibility in nonoutbreak settings. We found that 1 index case-patient transmitted infection to 0.27 close contacts on average, i.e., reproduction number = 0.27. This fi nding suggests that among those quarantined index case-patients, the number of persons with secondary cases who could be traced through rigorous fi eld investigation was small and far less than the number needed for the sustainable transmission of infectious disease in the population (reproduction number >1). However, the fact that the pandemic eventually spread in Beijing indicates that contact and case tracing were far from complete, especially later in the summer and early fall of 2009. The strict control measures may have worked to some extent at the beginning but were outpaced by local transmission (16) ; the percentage of locally acquired infections ranged from <10% in June 2009 to >80% in September 2009 (data not shown). In this study, the median number of close contacts per index case-patient per day was 7.0 persons. Although locating and quarantining these close contacts was done quickly, and stringent quarantine measures were used, which hindered implementation of control measures, the real number of close contacts was unknown and probably exceeded this number. Many close contacts were persons met in public places, including public transportation, theaters or cinemas, and shopping malls, and it is nearly impossible to trace all of the contacts. In addition, some persons who had worn PPE during contact with index casepatients were excluded from close contacts management (i.e., they were not quarantined), but because wearing PPE might not protect (or fully protect) against infection, some persons excluded might have become infected. In addition, many persons with mild and asymptomatic cases cannot be detected, but they may transmit the virus. Furthermore, the short generation time of pandemic (H1N1) 2009 shown in this study and in a previous study (13) could lead to the rapid accumulation of infection sources and close contacts. This rapid compounding could overwhelm response capacity and would have resulted in compromised effectiveness of containment measures. It should also be mentioned that we did not include persons with cluster or outbreak cases for whom close contacts could not be determined clearly by epidemiologic investigation to examine the basic feature of pandemic (H1N1) 2009 (e.g., generation time), and the reproduction number obtained from our data is an underestimate. Attack rates of infection differed signifi cantly by contact type. Among household members of index casepatients, the attack rate was the highest, as shown in the multivariate analysis after controlling for age and other factors. The most likely reason for this fi nding is that household members are more likely to have come into closer contact with index case-patients for a longer period with shorter distance and longer duration. Another possible reason is that household members may have some certain linkage with index cases in genetic susceptibility or living habits that would cause higher predisposition in household members than in other close contacts. This fi nding is similar to fi ndings in other investigations of respiratory infectious disease (17) . Close contacts on fl ights accounted for the highest proportion of all the close contacts, in part because of how the index cases were detected and the broad defi nition we used for close contacts. However, the attack rate was much lower than that for other close contacts; 1 index case infected only 0.19 close contacts on fl ights on average. This fi nding indicated that the possibility of transmission of pandemic (H1N1) 2009 virus on fl ights was low, and the yield of tracing and quarantining of close contacts on fl ights was limited. Tracing contacts of service persons at public places was more diffi cult than tracing other categories of contacts, and the lowest attack rate (0.2%) was recorded in this category. Despite extensive measures, on average, only 0.01 infected close contacts per index case-patient were identifi ed among service persons. Tracing the contacts of service persons at public places seems far less cost-effective. Criteria for close contacts on fl ights and those of service persons should be refi ned with respect to exposure duration and age of those exposed. Exposure to index case-patients for >12 hours was a signifi cant independent risk factor for infection in fl ight passenger contacts. This fi nding suggests that limiting the time of contact with persons with ILI on aircraft can reduce risk for transmission, and a long duration of exposure may be necessary for transmission to occur on aircraft. Younger close contacts were at higher risk for infection than older ones. The possible reason was that younger persons had much closer contact with index case-patients than did older persons; another reason may be that younger persons were more susceptible to infection with pandemic (H1N1) 2009 virus (18) . This fi nding was consistent with fi ndings reported in other studies (5, 6) . No secondary cases were found among close contacts exposed to index case-patients with subclinical infection. The attack rate among close contacts who were exposed to symptomatic index case-patients during their symptomatic phase was much higher than that among those exposed to these case-patients before their illness onset. Exposure to index case-patients during the symptomatic phase was a signifi cant independent risk factor for infection among close contacts. These fi ndings indicate that the infectivity of pandemic (H1N1) 2009 virus was higher after illness onset, and that the infectivity of symptomatic pandemic (H1N1) 2009 case-patients before illness onset was higher than that of persons with subclinical cases, although persons in each group were asymptomatic when in contact with other persons. In general, the earliest infectious time for pandemic (H1N1) 2009 was considered as 1 day before illness onset (19) . We found that index case-patients and infected close contacts shed pandemic (H1N1) 2009 virus <1 day before illness onset, which suggests that the infectious period of symptomatic persons with pandemic (H1N1) 2009 might be <1 day before illness onset. Among close contacts with pandemic (H1N1) 2009, ≈14.4% were asymptomatic. It is noteworthy that specimens from some close contacts tested negative for pandemic (H1N1) 2009 virus before quarantine, but those persons could shed the virus during quarantine without symptoms. Such infection could not be detected, and the proportion of subclinical infection was underestimated. Therefore, we calculated the proportion of subclinical infection by cross-sectional analysis of the subclinical infection of close contacts before quarantine, and we found that ≈17% of case-patients with pandemic (H1N1) 2009 were asymptomatic. This study has several limitations. We could not fi nd all close contacts of persons with pandemic (H1N1) 2009 and did not know their infection status, so the infection parameters of pandemic (H1N1) 2009 that we found in this study might not be precise, especially for reproduction number, which may be underestimated to some extent. Furthermore, we could not exclude the possibility that the infected close contacts had been infected from another unknown source before quarantine started, which might infl uence our conclusion to some extent. In summary, the attack rate among close contacts was low, even among household contacts. Household member and younger age were the major risk factors for infection with pandemic (H1N1) 2009 virus among close contacts. Approximately 17% of cases of pandemic (H1N1) 2009 were asymptomatic. Table 2 were included in multivariate unconditional logistic regression analysis. Hosmer-Lemeshow goodness-of-fit test was used to assess the model fit for logistic regression. OR, odd ratio; CI, confidence interval; NA, not available, indicating not included in the final model. †One dependent variable (infection with pandemic [H1N1] 2009 virus) and 5 independent variables (age of index case-patient, type of exposure to index case-patients, age of close contacts, relationships to index case-patients, and exposure duration of close contacts) were included in multivariate analysis. One independent variable (age of index case-patient) was removed in the stepwise regression equation. The goodness-of-fit test suggested that the logistic regression model fitted well (p = 0.631). ‡One dependent variable (infection with pandemic [H1N1] 2009 virus) and 4 independent variables (age of index case-patient, type of exposure to index case-patient, age of close contacts, and exposure duration of close contacts) were included in multivariate analysis. Two independent variables (age of index case-patient and type of exposure to index case-patient) were removed in the stepwise regression equation. The goodness-of-fit test suggested that the logistic regression model fitted well (p = 0.982). §One dependent variable (infection with pandemic [H1N1] 2009 virus) and 5 independent variables (age of index case-patient, type of exposure to index case-patient, age of close contacts, relationships to index case-patient, and exposure duration of close contacts) were included in multivariate analysis. Two independent variables (age of index case-patient and exposure duration of close contacts) were removed in the stepwise regression equation. The goodness-of-fit test suggested the logistic regression model fitted well (p = 0.751). ¶Exposed to symptomatic index case-patients before their illness onset or exposed to index case-patients who had subclinical infections.

Dengue Virus Infection Perturbs Lipid Homeostasis in Infected Mosquito Cells

Dengue virus causes ∼50–100 million infections per year and thus is considered one of the most aggressive arthropod-borne human pathogen worldwide. During its replication, dengue virus induces dramatic alterations in the intracellular membranes of infected cells. This phenomenon is observed both in human and vector-derived cells. Using high-resolution mass spectrometry of mosquito cells, we show that this membrane remodeling is directly linked to a unique lipid repertoire induced by dengue virus infection. Specifically, 15% of the metabolites detected were significantly different between DENV infected and uninfected cells while 85% of the metabolites detected were significantly different in isolated replication complex membranes. Furthermore, we demonstrate that intracellular lipid redistribution induced by the inhibition of fatty acid synthase, the rate-limiting enzyme in lipid biosynthesis, is sufficient for cell survival but is inhibitory to dengue virus replication. Lipids that have the capacity to destabilize and change the curvature of membranes as well as lipids that change the permeability of membranes are enriched in dengue virus infected cells. Several sphingolipids and other bioactive signaling molecules that are involved in controlling membrane fusion, fission, and trafficking as well as molecules that influence cytoskeletal reorganization are also up regulated during dengue infection. These observations shed light on the emerging role of lipids in shaping the membrane and protein environments during viral infections and suggest membrane-organizing principles that may influence virus-induced intracellular membrane architecture.

In the past 20 years, it has become increasingly evident that lipids are important bioactive molecules that mediate signalling cascades and regulatory events in the cell. The ability to synthesize lipids predisposes an organism to function as a host to parasites that have lost or lack this trait [1] . Viruses as obligate parasites rely exclusively on the host to fulfill their membrane and lipid requirements. This is especially true for enveloped viruses since they utilize host-derived lipid membranes to facilitate release from infected cells by budding as well as to enter cells through membrane fusion. Lipids also form an integral structural component of the virus particle. For most viruses that replicate in the cytoplasm of infected cells, lipids are essential for the replication of viral genomes. Both enveloped and non-enveloped viruses induce extensive ultrastructural changes in infected cells. Host-derived membranes are rearranged to provide extensive platforms that help to assemble arrays of replication factories [2] [3] [4] [5] [6] . Some of these factories are housed in specialized membrane compartments that assist in evading host antiviral defense mechanisms [2] [3] [4] 7] . These compartments also function to increase the local concentration of molecules necessary for efficient viral RNA replication and particle assembly. Recent advances in electron tomography techniques have been instrumental in providing a three-dimensional perspective of these virus-induced membranes [2] [3] [4] 7] . However, the metabolic cost to the host or vector and the contribution of lipid biosynthesis and trafficking to the formation of these replication factories is yet in its early stages of investigation [8] [9] [10] [11] [12] . In this study, we have focused on the importance of lipid biosynthesis on dengue virus (DENV) replication. DENV is one of the most aggressive re-emerging pathogens worldwide [13] . Over two and a half billion people in more than 100 endemic countries are at risk for contracting dengue fever. Currently 50-100 million cases of dengue fever are estimated annually [14] . Since DENV replicates within the mosquito vector as well as the human host, the spread of the virus can be greatly reduced by controlling the vector. Much effort has been placed in understanding the dynamics of virus transmission and replication in the mosquito vector, including identification of host proteins in the midgut and salivary glands that are regulated by DENV infection [15] [16] [17] . Less is known about the global impact of DENV on host metabolic pathways. Previous electron microscopy studies on DENV infected mosquito cells have shown that virus-induced membrane structures similar to those observed in human cells are prevalent [18] [19] . This extensive requirement in both host and vector, for intracellular membranes that support viral RNA replication and assembly suggest that quantifiable changes may exist in the lipid repertoire of the infected cell to assist in the formation of these membranes. Identifying these lipid changes that occur during infection is a necessary first step to discovering how DENV and its constituent proteins modify the lipid metabolism of cells. The reason(s) for such modifications has yet to be described, but can be pursued with a knowledge of which lipid changes occur. Furthermore, novel therapeutics that modify or inhibit these lipid changes and lipid-protein interactions could conceivably result in inhibition of virus replication. To investigate these possibilities, we used high-resolution mass spectrometry methods to profile the lipidome of DENV infected mosquito cells. We have identified several lipid classes that are regulated by DENV infection. Many of these lipids have characteristic roles in influencing membrane architecture as well as functioning in cellular signal transduction pathways. Specifically, we have identified differences induced in the lipid profile upon virus binding and entry alone compared to those induced by viral RNA replication, assembly and egress. We have also profiled the lipidome of cells treated with an inhibitor of de novo phospholipid biosynthesis. Through this we have identified a lipid environment that supports cell survival but is yet inhibitory to DENV replication. We had previously shown that fatty acid biosynthesis was a key target of DENV in human cells and that the rate limiting enzyme, fatty acid synthase (FAS), was both required and re-localized to sites of viral RNA replication during DENV infection [20] . In this study, using an inhibitor of FAS, C75, we determined that this requirement for fatty acid biosynthesis was conserved between the host and its vector during DENV infection. In the presence of C75 DENV replication was significantly reduced in C6/36 mosquito cells indicating that fatty acid biosynthesis is important for virus viability ( Figure 1A ). Furthermore, a time course of addition of C75 ( Figure 1B ) indicated that while pre-treatment or treatment of cells with C75 during viral adsorption reduced virus replication by ,10-100 fold, the most significant effect occurred upon addition of the drug at 4 and 8 hr post-infection (,1000 fold). This suggested that a post-entry step was affected by the inhibition of FAS. A comparison of virus released into the supernatant to intracellular virus indicated that an accumulation of intracellular virus was not occurring in the C75 treated cells (data not shown). Thus, the block in replication was not at the level of virus assembly or release. Based on the observation that DENV induces significant rearrangements in the membrane architecture of infected cells, together with its distinct susceptibility to inhibitors of FAS, it was our hypothesis that lipid biosynthesis was important to DENV replication. Therefore, to investigate whether the intracellular lipid composition was altered during DENV infection, we carried out LC-MS-based analyses of the global lipidome of mosquito cells infected with DENV. To differentiate between changes to the lipid profile that occurred upon exposure to infectious virus versus that induced by virus entry alone, we included UV-inactivated DENV (UV-DENV) in the studies. This inactivated virus is only capable of binding, entry and initial rounds of viral RNA translation but does not have the ability to replicate its genome. Figure 2 shows a principle component analysis (PCA) plot of the overall lipid abundance in C6/36 mosquito cells infected with either DENV or UV-DENV at 36 and 60 hr post-infection. The 36 hr time point was chosen to represent a peak in viral replication while the 60 hr time point represents late stages of replication as well as increased cellular stress. Our previous analysis of earlier time points (ie. 24 hr post-infection) indicated that concurrent with increasing RNA synthesis activity and virus release, there were substantial changes in the lipidome of infected cells. Some of these changes were greater (higher fold changes) than those observed at the 36 hr time point. However, the overall intensities of the expressed lipids was lower contributing to a low signal to noise ratio. Furthermore, the number of species expressed at significant levels (p,0.05) were limited. Therefore, we chose to pursue the later time points. Under conditions that ensured that all cells were infected (multiplicity of infection of 20), optimal viral RNA synthesis occurred between 24-36 hr post infection (data not shown). A total of 7217 features observed in the LC-MS analyses were included in the PCA analyses. The plot shows specific segregation of lipid profiles between DENV and UV-DENV exposed cells compared to uninfected cells (mock). A temporal regulation of the lipid profile was also observed. However, this is more discernible upon analysis of individual lipid species (Figure 3 ) rather than in the PCA analysis. Overall, in this whole cell analysis Dengue virus is one of the most aggressive human pathogens worldwide. It causes 50-100 million infections per year but there is no vaccine or antiviral that is currently effective against the disease. The virus is spread by Aedes aegyptii and Aedes albopictus mosquitoes and viral replication within the mosquito vector is required for transmission to a new human host. During this replication cycle, the virus causes significant changes to the membrane organization of infected cells. These virus-induced membrane alterations help to assemble arrays of viral replication factories and aid the virus to evade host antiviral defense mechanisms. Previously, much effort has been placed in trying to identify viral and cellular protein effectors that aid virus replication. In this study we have explored the role of lipids in the formation of these extensive membrane platforms in mosquito cells. Using high-resolution mass spectrometry we have profiled the lipid composition of dengue virus infected mosquito cells and compared it to uninfected cells. Through this we have identified several lipid classes that are differentially regulated during dengue virus replication. Using inhibitors of lipid biosynthesis we have also identified a lipid repertoire that is inhibitory to viral replication. Knowledge of how dengue virus utilizes cellular lipids and downstream signaling pathways to facilitate its replication will provide novel targets that could be utilized for developing effective antivirals. This study is also a forerunner for future comparative analyses of the human host and vector membrane environments required for viral replication. 15% of the metabolites identified were significantly (Anova p,0.05) different between virus infected cells and the mock control. Based on the total number of lipids that were significantly regulated between groups (uninfected and infected, Anova p,0.05) and subsequently structurally identified (Table S1 and Methods) the overall abundance of lipid classes is as follows: phosphotidylcholine (PC), 39.2%; phosphatidylethanolamine (PE), 31.2%; phosphatidylserine (PS), 0.8%; sphingomyelins (SM), 18.5%; ceramide (CER), 5.3%; lysophospholipids, 0.8% and ceramide phosphoethanolamine (CER-PE), 4%. This distribution is similar to the membrane lipid composition of eukaryotic cells where the most abundant phospholipid is PC [21] . It is also consistent with the membrane lipid composition of the Diptera species where PE is a predominant PL [22] . CER-PE, which is preferentially expressed in insect cells, is also observed here [23] [24] . Several of these lipids are differentially regulated upon exposure of cells to DENV or UV-DENV ( Figure 3 ). It was noticeable however, that several negatively charged lipid classes such as phosphatidylglycerol (PG), phosphatidylinositol (PI), and Cardiolipin, were absent from this list. These lipids have previously been reported to account for 6-13% of the lipidome of mosquito cells [25, 26] . However, in this whole cell analysis, these lipids were not significantly regulated during virus infection. A comparative analysis of the overall abundance of lipid species (relative to the mock control) in DENV and UV-DENV exposed cells is shown in Figure 3A and B. In all of the lipid classes with significantly regulated lipids (Anova p,0.05), DENV-infected cells have a unique expression pattern (expressed lipid molecular species) compared to UV-DENV infected cells. For a complete list of the differentially expressed lipids see Table S1 . Phospholipids (PL). In this whole cell analysis, the primary PLs that were significantly regulated (Anova p,0.05) (and subsequently structurally identified) were mostly neutral or zwitterionic. PS was the only acidic lipid significantly regulated. At 36 hr post-infection, there was an ,2 fold up-regulation of PC species in DENV-infected cells compared to UV-DENV and mock controls ( Figure 3A ). At the later time point (60 hr post-infection), the relative levels of PC remained elevated in DENV-infected cells, however the differential expression between DENV and UV-DENV-exposed cells was not as evident ( Figure 3B ). Selected PC species were up regulated only at the 36 hr time point while others were up regulated at both time points ( Figure 3C ). Interestingly, a majority (,80%) of the PC species that were up regulated had unsaturated fatty acyl chains. Analysis of the overall fold change in PE at the 36 hr time point ( Figure 3A ) does not show a difference between the virus-exposed (DENV and UV-DENV) cells and the mock. This is due to an equal number of individual lipid molecular species being up regulated as were down regulated ( Figure 3D ). At the 60 hr time point the relative levels of PE in virus exposed cells were lower than mock levels ( Figure 3B and D). There was also limited overlap between the specific PE molecular species regulated between DENV-and UV-DENV-exposed cells (Table S1 ). Given that insect cells have a high abundance of PE in their membranes (40-50%), it is interesting that there is a selective requirement for PC (over PE) in DENV-infected mosquito cells [22] . Another group of PLs that were preferentially expressed in DENV-infected cells were the lysophospholipids (LPLs) ( Figure 3A and B). These lipids result primarily from the hydrolysis of PC by phospholipase A2-type enzymes (PLA 2 ) and represent a PL that is missing an acyl chain [27] [28] [29] [30] . In DENV-infected cells, there was a preferential up regulation (,3 fold) of these lipids compared to both UV-DENV and mock controls at the 36 hr time point. This expression was slightly down regulated at the 60 hr time point to about ,1.5 fold above the mock. In UV-DENV exposed cells, LPLs were undetectable at the 36 hr time point, while at the later time point, they were similar or slightly above the mock levels. As shown in figure 3H , LPLs with C16 acyl chains were up regulated at both time points while there was a selective up regulation of LPLs containing C18 acyl chains only at the early time point ( Figure 3H ). The C16 LPLs have been implicated in pro-apoptotic signaling pathways. This observation of selected LPL expression presents an attractive hypothesis that chain length differences may dictate specific roles for LPLs during virus infection. Since LPLs result from the activity of PLA 2 -type enzymes, we investigated whether DENV infected cells displayed a higher activity of PLA 2 compared to uninfected cells. Utilizing a fluorescent PC substrate (BODIPY-PC) we used mass spectrometry to monitor the hydrolysis of PC to LPC by intracellular PLA 2 . Essentially, we monitored the production of BODIPY-LPC (resulting from PLA 2 mediated metabolism of PC) during a time course of DENV infection ( Figure S1 ) (method from [31] ). The assay indicated that following 24 hr of infection and longer, PLA 2 activity was elevated in DENV-infected cells compared to the controls, with the highest activity occurring at 48 and 72 hr postinfection. Therefore, the elevated levels of PLA 2 could be the source of the LPL detected in DENV-infected cells. Sphingolipids. Although originally known as vital components of barrier membranes, sphingolipids are also potent bioactive molecules that regulate cell death, growth, differentiation and intracellular trafficking [32] [33] . The primary sphingolipids regulated during DENV infection were sphingomyelin (SM) and ceramide (CER). In DENV infected cells, SM was up regulated by ,2-fold compared to the controls (UV-DENV and mock) and remained elevated at both 36 hr and 60 hr time points ( Figure 3A and B). Furthermore, this temporal expression varied depending on the specific molecular species being expressed ( Figure 3E and Table S1 ). UV-DENV exposed cells showed similar levels of SM compared to the mock at the 36 hr time point, but these levels decreased at the later time point. Interestingly, an analog of SM known as ceramide phosphoethanolamine (CER-PE), was preferentially expressed (up regulated by ,2 fold) in UV-DENV exposed cells at the 36 hr time point (compared to DENV and mock) ( Figure 3G ). However, this expression level dropped below the mock at the later time point. In comparison, DENV infected cells showed very low expression of this lipid at the early time point (,0.7 fold compared to mock) and its expression was undetectable at the later time point ( Figure 3A and B). This lipid is expressed more abundantly in insect cells rather than in mammalian cells [23] [24] . Another bioactive sphingolipid, CER, results from either the degradation of SM by sphingomyelinases or de novo synthesis through the condensation of palmitate and serine [32, 34] . These lipids were preferentially up regulated in DENV-infected cells at both the 36 and 60 hr time points (,2-fold compared to the mock). Since SM was also up regulated similarly in DENV infected cells, the rise in CER is possibly the result of de novo synthesis rather than SM degradation. UV-DENV-exposed cells showed undetectable CER levels at 36 hr, while at 60 hr the expression was (,2 fold) above the mock control. Since there was a concurrent decrease in SM in UV-DENV exposed cells at 60 hr, the increased levels of CER in this case could be due to the degradation of SM resulting from increased cellular stress (resulting from incubation of the cells for 60 hr). Similar to other lipid species, CER also showed temporal regulation depending on the molecular species expressed ( Figure 3F ). An overall comparison of the PLs regulated at the 36 and 60 hr time points indicated that in all lipid classes that were significantly regulated (Anova p,0.05) during DENV infection, selected molecular species were regulated only at the 36 hr time point while others were regulated at both time points. Representative examples for each lipid class are shown ( Figure 3C -H). These observations may represent lipidome differences in cells sustaining early but active DENV replication (36 hr) compared to those experiencing advanced cellular stress (60 hr). It is also interesting to note that any up regulation observed in UV-DENV exposed cells (at either time point) was only in the range of 1-1.5 fold compared to the mock control, while DENV infected cells showed much higher levels of expression. In addition to evaluating the lipid composition of whole cells, we also explored the possibility of profiling the lipid repertoire of specific membranes induced by DENV. Given the extensive interconnectivity of the membranes induced in DENV infected cells, isolation of morphologically distinct membranes proved challenging. Therefore, we carried out subcellular fractionation of C6/36 cells and isolated post-nuclear supernatants that were further fractionated to provide a total membrane fraction (16K pellet) enriched in viral replication components and a remaining cytoplasmic extract (CE) [35] . To confirm that the 16K pellet was indeed the membrane fraction enriched in the replication complex components, we used quantitative RT-PCR to measure the ratio The fold changes represent DENV-infected cells or UV-DENV exposed cells compared to the mock control. A lack of cones indicates that the expression level of those specific lipids were not significant (p,0.05). Panels C-H are representative lipid molecular species from specific lipid classes significantly regulated at the two different time points. The data are plotted as the integrated LC-MS peak abundance, in log 2 scale with standard deviation. PC, phosphatidylcholine; PE, phosphatidylethanolamine; SM, sphingomyelin; CER, ceramide; CER-PE, ceramide phosphoethanolamine; Lyso, lysophospholipids. See supplementary table S1 for a complete list of lipid features detected in this study. Four replicates were included in the lipidomic analyses. The error bars represent standard deviation of the mean. The blue dashed line separates species that remain elevated at both time points (36 and 60 hr) from species that are only elevated at the 36 hr time point. Infections were carried out using an MOI of 20 in C6/36 cells. Significantly expressed lipids species are shown denoted with an asterisk (*). doi:10.1371/journal.ppat.1002584.g003 of viral RNA to lipid in both 16K and CE fractions. Lipids were quantified by pulse-chase analysis using 14 C-acetate to label newly synthesized lipids and the results are shown in Figure 4 . In postnuclear supernatants, there was increase in the incorporation of 14 C-acetate into lipids in the DENV infected cells compared to the mock control ( Figure 4A ). This increase was observed at both 36 hr and 60 hr post-infection with the latter time point showing a greater difference between DENV and mock. A more detailed analysis of the lipid distribution in the subcellular fractions (16K and CE) is shown in Figure 4B . At the 36 hr time point, the CE fraction was enriched in newly synthesized lipids (compared to the 16K fraction) in both the DENV and mock samples, with DENV showing an almost 2-fold increase compared to mock. This distribution changed at the 60 hr time point where the 16K pellet was more enriched in newly synthesized lipids, however, the fold change (DENV compared to mock) was not as significant. To identify the primary fraction containing viral induced membranes that included the viral replication complex, we carried out a comparison of the viral RNA genome copies to labeled lipid in the 16K and CE subcellular fractions ( Figure 4C ). The results indicated that the 16K membrane fraction was enriched (a ,3fold increase) in viral RNA (compared to CE) at the 36 hr time point while at the 60 hr time point the CE fraction had slightly more viral RNA per labeled lipid (compared to the 16K fraction). Based on the 14 C-acetate labeling studies above, the CE fraction showed an increased amount of newly synthesized lipids (compared to the 16K fraction at the 36 hr time point). In lipid biosynthesis pathways, acetate is a precursor to both phospholipid as well as sterol biosynthesis. Therefore, the labeled lipids (in CE) represent not only phospholipids in membranes, but also cholesterol and triglycerides that form lipid droplets [36] . In this study, lipidomic analyses were carried out on both the 16K and CE subcellular fractions isolated from the 36 hr time point from DENV infected cells, UV-DENV exposed cells and the mock control. The primary lipids observed in the CE fraction were sphingolipids. Cholesterol and triglycerides were not successfully separated using our mass spectrometry analyses and very few phospholipids were observed to be statistically significant (p,0.05) in virus infected cells over the mock control (data not shown). Therefore, we focused our primary lipidomic analysis on lipids extracted from the 16K pellet at the 36 hr time point. Similar to the whole cell analysis, DENV infected cells displayed a unique expression profile compared to UV-DENV and mock infected cells ( Figure S2 ). Over 85% of the metabolites detected were significantly regulated in virus-infected cells compared to the controls. Analysis of the membrane fraction also highlighted several lipid species not detected in our whole cell analysis. An overall fold change (compared to mock) of the significantly (p,0.05) expressed lipid classes in the membrane fraction is shown in Figure 5 . For a complete list of differentially expressed lipids see Table S2 . Phospholipids. There was an up regulation of phospholipids, sphingolipids and several bioactive intermediates in DENV infected cells compared to UV-DENV and mock controls ( Figure 5A ). Amongst the specific phospholipids expressed, PE was selectively up regulated in membrane fractions isolated from DENV infected cells ( Figure 5B ). However, the primary PE molecular species up regulated was a lysophospholipid (PE P-16:0e). Both DENV and UV-DENV exposed cells showed similar expression of fatty acids ( Figure 5A ). However, the individual molecular species expressed differed between the two viruses ( Figure 5C ). For instance, stearic acid was up regulated in DENV infected cells while oleic acid was up regulated in UV-DENV exposed cells. Palmitic acid had a similar level of expression in both DENV and UV-DENV exposed cells. Several bioactive intermediates such as monoacylglycerol (MG), diacylglycerol (DG) and phosphatidic acid (PA) were also prominently up regulated in DENV infected cells compared to the controls ( Figure 5D ). Sphingolipids. Although the Blygh and Dyer method [37] utilized for lipidomic analyses of whole cells or membrane fractions detected the expression of sphingolipids, Merrill et al optimized a protocol dedicated to analyzing sphingolipids without the interference from co-purifying PLs [38] . Utilizing this protocol, we carried out mass spectrometry analysis of these bioactive lipids from mosquito cells fractionated to provide three subcellular fractions: nuclear (N), cytoplasmic (CE) and membrane (16K). Similar to the above analyses, UV-DENV and mock infected cells were included as controls. The specific sphingolipids monitored included CER, SM, monohexosylceramide (GlcCER) and dihexosylceramide (GalCER). As shown in Figure 6A , CER levels in C6/36 cells showed clear elevation in DENV-infected cells compared to mock-infected or UV-DENV exposed cells. Comparison of the different subcellular fractions indicated that the elevated CER was primarily concentrated in the CE fraction with the exception of two species (d18:1/ 16:0 and d18:0/16:0) which were enriched in the 16K pellet. These two species consist of the most abundant fatty acids observed in cells. The analysis of SM levels in mosquito cells ( Figure 6B ) indicated that virus exposed cells have a higher level of SM compared to mock-infected cells. This is most evident in the 16K pellet with the exception of the most abundant species (d18:1/20:0) that showed the highest levels in the CE. Comparison of infectious DENV to UV-DENV exposed cells indicates that the latter which is incapable of replication maintained higher levels of SM overall. GlcCER were primarily detected in the 16K pellet of both DENV infected and UV-DENV exposed cells ( Figure 6C ). However, the overall levels were down-regulated compared to the mock control with the exception of d18:1/16:0, which showed slight elevation in the UV-DENV exposed cells. Dihexosylceramides were not detected to significant levels in any of the samples. The FAS inhibitor, C75 is a potent inhibitor of DENV replication in both mosquito ( Figure 1 ) and mammalian cells [20] . Cells treated with non-cytotoxic concentrations of C75 (,50 mM) induce an environment that has redistributed its lipid repertoire to support cell survival in the absence of FAS activity, and yet, this environment does not support viral replication. To determine the basis for this exclusion of virus replication, we profiled the lipidome of cells treated with 25 mM C75. Membrane fractions isolated from DENV infected and mock control cells in the presence or absence of C75 were profiled. A hierarchical clustering analysis of the lipidomic data ( Figure S3 ) indicated that the lipid environment of DENV infected cells treated with C75 clustered very closely to the mock C75 treated control and was furthest from DENV infected cells suggesting that the two represent significantly different environments. A comparative analysis of the lipid species expressed in the two environments indicated that a majority of the lipids that were up regulated in DENV infected cells were down regulated upon treatment of those cells with C75. Amongst the phospholipids, several phosphatidylinositol species were up regulated 3-4 fold in the untreated DENV infected cells compared to the C75 treated cells ( Figure 7B ). However, none of these PI species were phosphorylated. A single species of PG was also up regulated by over 5 fold in the untreated cells. Since C75 disrupts the formation of lipids down stream of FAS, several fatty acids (stearic and palmitic acid) were down regulated in the drug treated cells. An enrichment of these fatty acids was observed in the untreated cells upon comparison with C75 treated cells ( Figure 7C ). In contrast, as observed in the previous comparison of DENV infected cells to uninfected cells ( Figure 5 ), oleic acid was down-regulated in untreated DENV infected cells and therefore, showed enrichment in C75 treated cells. Interestingly, C75 treated cells also showed a down regulation of metabolic intermediates such as mono-and diacylglycerol as well as phosphatidic acid. Enrichment of these lipids was observed in untreated DENV infected cells compared to C75 treated cells ( Figure 7D ). In addition to the phospholipids, de novo synthesis of sphingolipids was also disrupted by C75. Key intermediates in this pathway such as N-palmitoylesphingosine and N-stearoylesphingosine were down regulated upon treatment of cells with C75 and up regulated in untreated DENV infected cells ( Figure 7A and Table S3 ). The architecture of biological membranes is defined by the composition and distribution of lipids and proteins in the bilayer. Optimizing this composition and distribution to facilitate the formation of virus-induced intracellular membrane structures has to be a requirement for efficient DENV replication. Previously we showed that FAS, the rate-limiting enzyme in lipid biosynthesis was both recruited and activated during DENV replication in human cells [20] . In this study we demonstrated that this requirement for FAS activity was also conserved in the mosquito vector-derived cells. Furthermore, inhibition of FAS activity seemed to be most effective between 4-12 hr post-infection suggesting an early requirement for FAS activity during the replication cycle. As a next step in investigating the role of lipids in DENV induced membrane expansion we used high-resolution mass spectrometry to profile the lipid composition of DENV infected mosquito cells. We observed a very distinct segregation in the lipid composition between DENV infected and uninfected cells. Specific lipid changes that occurred upon virus binding and entry alone were also identified. In DENV infected cells, there was a selective enrichment of lipids that have characteristic functions in influencing membrane structure as well as those that have potent signaling functions. Among these lipids were bioactive sphingolipids such as sphingomyelin and ceramide, lysophospholipids and several intermediates such as mono-and diacylglycerol and phosphatidic acid. The phospholipids that were enriched in DENV infected cells were primarily unsaturated. There was also evidence for the up regulation of de novo phospholipid and sphingolipid biosynthesis as well as triacylglyceride metabolism ( Figure 8 ). Furthermore, we identified a unique lipid environment that supported cell survival but did not support DENV replication. This environment was created by the addition of C75, an inhibitor of fatty acid synthase, the rate-limiting step in phospholipid biosynthesis. Analysis of the whole cell lipidome indicated that unsaturated PC was the most up regulated phospholipid in DENV infected cells. Membranes that are similar or derived from the ER are enriched in unsaturated lipids, primarily PC [39] . Although DENV-induced membranes in mammalian systems are ER-derived, it is unknown if the same is true in mosquito cells. While PC is known to form planar membranes, unsaturated PC induces membrane curvature which may be important for maintaining the highly curved membranes observed in DENV-infected cells [40] . PC enriched membranes are also more fluidic compared to the rigid membranes enriched in SM and cholesterol, which may be a feature attractive to DENV infection [41] . Bioactive sphingolipids such as SM and CER are also key lipids that were up regulated during DENV infection. In cells exposed to replication competent virus, CER was up regulated in both the whole cell lipidome as well as replication complex membranes. This is either a cellular response to virus infection, or a direct need for CER in the virus replication cycle. The latter hypothesis is particularly attractive considering the enrichment of specific molecular species of CER in the 16K pellet. CER is a cone-shaped Figure 6 . Bioactive sphingolipid species are differentially regulated in replication complex membranes isolated from DENVinfected mosquito cells. Multiple Reaction Monitoring (MRM) analysis of sphingolipids species differentially regulated in DENV-infected cells (MOI 20) or UV-DENV exposed cells compared to the mock control (see also supplementary table S3). The data represent fold changes observed in three subcellular fractions that were analyzed in this study; 16K, replication complex membranes; CE, cytoplasmic extracts following removal of replication complex membranes and nuclei; N, nuclear fraction. Panels A-C represent ceramide, sphingomyelin and monohexosylceramide species respectively. The dashed line highlights values equal to the mock. The data represent three independent experiments. The error bars represent standard deviation of the mean. doi:10.1371/journal.ppat.1002584.g006 molecule that promotes inward budding of membranes (negative curvature) [42] . Some DENV-induced membranes show a double membrane morphology indicating that negative curvature modifying lipids such as CER may be active in their formation. As a second messenger, CER can also induce apoptosis and autophagy [34] . However in mosquito cells, DENV maintains a persistent infection and antagonizes apoptotic pathways [19, 43] . Therefore, CER up regulation may not be due to the prevalence of an apoptotic response but may be utilized to up regulate autophagy during DENV infection. It has been shown in mammalian cells that autophagy is up regulated and necessary for DENV replication. The prevalence of several intermediates in CER biogenesis, including N-palmitoylsphingosine and diacylglycerol, suggest that the observed CER could result from either de [20] . 2. Inhibition of this process with C75 disrupts the cellular lipid repertoire in mosquito cells to be unfavorable for virus replication. 3. The lipidomic analyses reveal an up regulation of fatty acids such as palmitic (C16) and stearic (C18) acid. These fatty acids are intermediates in the biosynthesis of phospholipids, which is up regulated during DENV infection. Interestingly, in DENV infected cells the prevalent phospholipids primarily consist of C16 and C18 unsaturated acyl chains. Very long chain fatty acids are not significantly up regulated during infection. 4. FAS activity also stimulates de novo sphingolipid biosynthesis. In the lipidomic analyses, the up regulation of intermediates such as N-palmitoylesphingosine suggests sphingolipid biosynthesis is activated during DENV infection. Specifically, SM and CER are enriched in DENV infected cells. Alternately, the up regulation in CER (and DG) during infection could result from the degradation of SM through the activity of sphingomyelinases (Smase). The resulting CER and DG could be redirected into several signaling pathways or be utilized for de novo phospholipid biosynthesis. The glycopshingolipids, GlcCER and GalCER are down regulated during DENV infection, which suggest that they are catabolized to produce CER. 5. Lipidomic analyses also suggest the up regulation of triacylglycerol catabolism (Lipolysis) in DENV infected cells. This pathway results in the generation of MG, DG and palmitic acid. These intermediates are all up regulated in DENV infected cells and could be utilized for downstream signaling or de novo phospholipid biosynthesis. It has also been shown that TG catabolism is necessary for mitochondrial b-oxidation during DENV infection [50] . 6. Elevated levels of LPC in DENV infected cells also suggest activation of PC hydrolysis by PLA 2 . This enzyme is activated during DENV infection. The elevated levels of other phospholipids such as PA, PI, PE, PG as well as PC suggest that the CDG-DG pathway for phospholipid biosynthesis could also be activated. FAS, fatty acid synthase; DENV, dengue virus; C75, inhibitor of FAS; SM, sphingomyelin; CER, ceramide; MG, monoacylglycerol; DG, diacylglycerol; TG, triacylglycerol; LPC, lysophosphatidylcholine; PLA 2 , phospholipase A 2 ; PA, phosphatidic acid; PI, phosphatidylinositol; PE, phosphatidylethanolamine; PG, phosphatidylglycerol; PC, phosphatidylcholine. doi:10.1371/journal.ppat.1002584.g008 novo synthesis by serine palmitoyltransferase (SPTLC) in the ER or catabolism of SM by sphingomyelinases (SMases) at the plasma membrane [44] . The lipidomic analyses also indicated a down regulation of GlcCER, which may suggest that a salvage pathway that metabolizes GlcCER to CER may also be active. Inverted cone-shaped lipids such as LPLs were also up regulated during DENV infection. Primarily, LPC was up regulated in the whole cell analysis. It results from the hydrolysis of PC by PLA 2 [45] . We have shown that this enzyme is activated during DENV infection. Many other viruses have also shown a dependency for PLA 2 in their life cycle [31, 46] . Due to its inverted cone-shaped structure, asymmetric incorporation of LPC into membranes causes positive curvature and induces vesicle fission and budding [47] . Structural analyses of DENV-induced membrane structures in mosquito cells indicate the presence of highly curved membranes and smaller vesicles [18] [19] . Therefore, vesiculation through fission and budding as well as enrichment of molecules that increase membrane curvature may be an underlying mechanism for forming these membrane structures. LPC also increases the permeability of membranes [48] . This is an attractive concept given that DENV replication complexes are encased within membrane barriers that need to facilitate the exchange of components to and from the cytosol for genome replication and virus assembly. Therefore, leaky membranes may be favored in DENV-infected cells through the selective incorporation of molecules such as LPC. Another observation that gives credence to this hypothesis is that the transmembrane permeabilizing effect of LPC is synergistically enhanced in the presence of palmitic acid, which is also up regulated in DENVinfected cells. As a signaling molecule, LPC is both a pro-survival as well as an inflammatory molecule [46] . Therefore, its expression in DENV infected cells could be a cellular response to viral infection. It has been previously reported that de novo lipid biosynthesis was up regulated during DENV infection [20] . Consistent with these observations, several primary fatty acids (palmitic, stearic and oleic acid) involved in the biosynthesis of higher order PLs were detected in our lipidomic analyses. Palmitic acid is the first fatty acid of the lipogenesis pathway and stearic acid (C18:0), is immediately downstream of palmitic acid. They are both up regulated in DENV infected cells. The latter is the precursor to the biosynthesis of oleic acid (C18:1) or long chain fatty acids (with acyl chains of C20 or greater). Interestingly, since oleic acid is down regulated in DENV-infected cells stearic acid may be stimulating long chain FA biosynthesis. However, a majority of the PLs expressed in DENV infected cells have C16-or C18-acyl chains with varying degrees of unsaturation. Therefore, it seems that desaturation of palmitic and stearic acid rather than elongation may be the chosen pathway for lipogenesis during DENV infection. Several intermediates in lipid catabolism were also detected in the replication complex membranes (palmitic acid, MG and DG). Analysis of these intermediates suggested the up regulation of pathways for PC hydrolysis and triacylglycerol (TG) metabolism. The latter pathway is responsible for regulating lipid homeostasis and energy production in the cell through the metabolism of lipid droplets [49] . Through the action of lipases the degradation of lipid droplets results in the release of palmitic acid, DG and MG [36] . These metabolic products are substrates for mitochondrial boxidation as well as intermediates (fatty acids) in the synthesis of new lipids. DENV infected cells show a high prevalence of these species compared to mock infected cells. While several acidic phospholipid species were detected in the 16K membrane fraction very few were significantly regulated during infection. Previous studies in mammalian cells have indicated the importance of acidic lipids in DENV infection [46, 50] . These negatively charged lipids are also implicated in influencing membrane structure. The most striking results were obtained by the comparison of DENV infected cells in the presence and absence of the FAS inhibitor, C75. It is well established that treatment of cells with this inhibitor disrupts de novo phospholipid biosynthesis [51] . To circumvent this inhibition, the cell redistributes its lipid repertoire to ensure cell survival. However, this redistribution does not support DENV replication. A comparison of the different lipid environments indicated that many of the lipids that were up regulated in DENV infected cells compared to cells treated with C75, were similar to those previously observed when infected cells were compared to mock (uninfected) controls. Therefore, comparison of DENV infected cells to two different controls (uninfected cells and C75 treated infected cells) highlighted the same lipids as being up regulated during an active DENV infection. In these studies we utilized UV-DENV as a control to identify lipid metabolic changes that occurred upon virus binding and entry alone. This virus is incapable of replication but immunofluorescence microscopy with anti-NS3 antibodies have indicated that there is a very low level of translation that occurs within the first 24 hr (data not shown). Analysis of the lipidome of UV-DENV exposed cells showed a significant difference in the phospholipid and sphingolipid content in comparison to DENV infected cells and the mock control. It is possible that binding, membrane fusion and entry alone, or the expression of viral proteins from the incoming viral RNA triggers the activation of lipases or lipid transport mechanisms that result in the differences observed in the lipidome of the cells. This has been shown for other viruses. Jan et al. have shown that binding and entry alone of Sindbis virus triggered the activation of sphingomyelinases that degraded membrane bound SM to Cer [52] . It has also been shown that viruses induced apoptotic-signaling cascades upon binding and entry alone [53] [54] [55] . These cascades may either result from or induce lipolysis or lipid transfer between compartments in the cell. The most pronounced changes are those observed in the 16K pellet analyses where a significant decrease in phospholipids, sphingolipids and lipid intermediates are observed (compared to mock cells, Figure 5A) . A comparison of SM and CER in the whole cell analysis at the 36 hr and 60 hr time points ( Figure 1A and 1B) indicated a reduction in SM with a corresponding increase in CER with time. This could be due to a similar scenario to SINV where a non-replicating UV-DENV is capable of inducing lipid conversion in membranes [52] . Therefore, SM and other sphingolipids may be converted to CER that is transported away from the 16K membrane fraction to other locations in the cell. This is also evident from the increased content of fatty acids compared to mock cells that may result from the degradation of lipids. UV-DENV exposed cells do not show significant regulation of MG or DG, but show an up regulation of palmitic acid (C16:0). Virus binding and entry alone could stimulate lipid droplet hydrolysis (catabolism of TG) resulting in the release of palmitic acid. These studies have shown that UV-DENV could be a versatile tool to pursue mechanistic studies on the metabolic pertabations that occur upon virus binding, membrane fusion, and entry. In summary, using high-resolution mass spectrometry, we have determined that DENV drastically alters the lipid profile of infected cells. Specifically, DENV infection elevates the expression of lipids that have the capacity to change the physical properties of the bilayer such as bilayer curvature, permeability, and the recruitment and assembly of protein complexes in the membrane. Several of the identified molecules also function as bioactive messengers that control signaling and membrane trafficking pathways in the cells. They represent molecules that result from the activation of cellular stress pathways that respond to viral infections. Based on these findings, the next steps will be to investigate the mechanisms of how these lipid species play a role in DENV replication, as well as identify the control points in these pathways that may be influenced by viral gene products. Cell culture and virus infections C6/36 (Aedes albopictus) cells (ATCC) were maintained in Minimal Essential Medium (MEM) supplemented with glutamine (2 mM), non-essential amino acids, 25 mM Hepes and 10% heat inactivated fetal calf serum. Dengue virus 2, strain 16681 (obtained from Richard Kinney, CDC. Ft. Collins) stocks were amplified in C6/36 cells in MEM (supplemented as above) and 2% heat inactivated fetal calf serum. UV-inactivated DENV was obtained by exposing the same virus stock (used for infection) to UV-light for 3 hr and then conformation of inactivation by two blind passages of the virus on cells for 60 hr per passage. Lack of infectivity was confirmed by plaque assay and immunofluorescence assay. Sample preparation. For infection, ,5610 8 C6/36 cells were infected with DENV at a MOI of 20 at 30uC. Infection of all cells (required for lipidomic analyses) was confirmed by immunofluorescence analyses. Each experiment included three biological replicates. Following infection, cells were harvested at 36 and 60 hr post-infection, pelleted and resuspended in 100 ml of 100 mM ammonium bicarbonate. Lipids were extracted from an equal number of cells using a modified Bligh and Dyer protocol. Briefly, a mixture of 2:1 Chloroform:methanol, 0.1% acetic acid and 0.01% butylated hydroxy toluene (BHT) were added to the cell suspension in ammonium bicarbonate such that there was a 4:1 ratio of organic solvent to cells. Following lipid extraction, the organic phase was separated from the aqueous phase by centrifugation and dried down under a N 2 stream in low retention microfuge tubes (Fisher). The dried lipids were resuspended in 75 ml of methanol and vortexed for 10 s. The samples were then centrifuged at 13,4006 g for 5 min to remove any particulates. chromatography/mass spectrometry. Lipid molecular species were profiled using a dual-column nanocapillary LC system equipped with 75 mm665 cm columns, each packed with 3 mm Jupiter particles (Phenomenex, Torrance, CA). The mobile phases were (A) 10 mM ammonium acetate in 50:50 water/methanol (v/v) and (B) 10 mM ammonium acetate in 50:50 methanol/acetonitrile (v/v). The LC system was equilibrated at 10,000 psi with mobile phase A prior to injecting 0.4 mL of sample. Exponential gradient elution was initiated 50 min after sample loading with an initial column flow of ,300 nL/min. After 90 min of gradient separation, the mobile phase mixer was purged with 3 mL of mobile phase B, followed by a 5 min column wash. Finally, the mobile phase mixer was purged with 10 mL of mobile phase A, which represented the end of one separation cycle. While gradient elution was performed on one column, the other column was equilibrated with mobile phase A. The LC system was coupled to a hybrid linear ion-trap-Orbitrap mass spectrometer (ThermoFisher, San Jose, CA), and the capillary temperature and electrospray voltage were 200uC and +2.2 kV, respectively. The Orbitrap was used as the mass analyzer during MS survey scans over the m/z range 200-2000 with a duty cycle of ,1.2 s. Data-dependant MS/MS was performed in the LTQ for the top 5 ions, with a normalized collision energy of ,35%. Dynamic exclusion in the LTQ during data-dependant MS/MS experiments was enabled as follows: repeat count of 2, repeat duration of 30 s, exclusion list size of 250, and exclusion duration of 60 s. Data processing. LC-MS datasets, defined as the data obtained from a single LC-MS analysis, were processed using the PRISM Data Analysis system [56] , a series of software tools freely available at http://ncrr.pnl.gov/software/ and developed in-house. The first step involved deisotoping of the raw MS data to give the monoisotopic mass, charge state, and intensity of the major peaks in each mass spectrum using Decon2LS [16] . The data were next examined in a 2-D fashion using MultiAlign to identify groups of mass spectral peaks that were observed in sequential spectra using an algorithm [57] that computes a Euclidean distance in n-dimensional space for combinations of peaks. Each group, generally ascribed to one detected species and referred to as a ''feature'', has a median monoisotopic mass, central normalized elution time (NET), and abundance estimate computed by summing the intensities of the MS peaks that comprise the entire LC-MS feature. LC-MS features were then chromatographically aligned across all datasets using the LCMSWARP algorithm [58] in MultiAlign, and the lipid identities of detected features were initially determined by comparing their measured monoisotopic masses and NETs to calculated monoisotopic masses and observed NETs for lipids in an AMT tag database [59] within search tolerances of 63 ppm and 60.03 NET for monoisotopic mass and elution time, respectively. Subsequent identifications of lipid features that did not match to entries in the AMT tag database were made by searching their molecular weights against entries in the Lipid MAPS database followed by manual confirmation based on MS/ MS spectra. Statistical analysis of processed data. Following chromatographic alignment and database matching, the abundances of all detected features (both AMT tag database matched and unmatched) were loaded into DAnTE [60] for statistical analysis. Feature abundances were transformed to log2 scale then subjected to central tendency normalization [61] . Comparative data analysis was then performed and statistically significant differences between the lipid profiles of the samples were determined using ANOVA. Partial least-squares (PLS) [62] analysis was also performed using the data matrices containing either AMT tag database matched features alone or all features (both database matched and unmatched). homogenized and centrifuged at 15006 g, at 4uC for 5 min to remove nuclei. Post-nuclear supernatants were centrifuged at 160006g for 30 min at 4uC to obtain a membrane pellet (referred to as the 16K pellet). The supernatant post nuclear and membrane fractionation is referred to as the cytoplasmic extract (CE). Total RNA was extracted from each fraction (16K and CE) using Trizol reagent (Sigma) according to the manufacturer's instructions. Following RNA extraction, total lipids were extracted using the Bligh and Dyer method mentioned above and counted using a Beckman LS 6000 scintillation counter. The viral RNA in each fraction was measured using the SuperScript III Platinum SYBR Green One-Step qPCR Kit with ROX and DENV specific primers; (forward) 59 ACAAGTCGAACAACCTGGTCCAT 39 and (reverse) 59 GCCGCACCATTGGTCTTCTC 39 on a Applied Biosystems 7300 Real Time PCR machine. The labeled lipid in subcellular fractions was standardized to the total RNA isolated from each fraction prior to determining the viral RNA genome copies per labeled lipid in subcellular fractions. Sample preparation. For profiling the lipidome of the replication complex membranes, lipids were extracted from the 16K pellet using the same modified Blygh and Dyer protocol described above. Liquid chromatography/mass spectrometry profiling analysis of lipid content. A LTQ Orbitrap XL instrument (Thermo Fisher Scientific San Jose, CA) was used for the analysis. It was coupled to an Agilent 1100 series LC (Agilent Technologies, Santa Clara, CA) equipped with a micro well plate auto sampler and binary pumping device. Reverse phase liquid chromatography was used to analyze the samples. An Agilent Eclipse XDB-C8 column with 2.16150 mm, 3.5 mm dimensions was used for the separation. Solvent A consisted of water +0.1% piperidine. Solvent B contained acetonitrile : methanol (50:50 v/v) +0.1% piperidine. The flow rate was 300 mL/minute. A volume of 10 mL was loaded onto the column. The gradient was as follows: time 0 minutes, 50% B; time 25 minutes, 95% B; time 45 minutes, 95% B; time 50 minutes, 50% B; time 60 minutes, 50% B. The MS analysis used negative polarity electrospray ionization. The source voltage was 4.0 kV, source current 100 mA, capillary voltage 230.0 V, tube lens voltage 2100.0 V. The capillary temperature was 200uC, sheath gas flow was 35, auxiliary and sweep gas were both set to 0. Data were acquired using data dependent scanning mode. FTMS resolution of 60,000 with a mass range of 70-1200 was used for full scan analysis and the ITMS was used for MS/MS data acquisition. The top three most intense ions were acquired with a minimum signal of 500, isolation width of 2, normalized collision energy of 35, default charge state of 1, activation Q of 0.250, and an activation time of 30.0. The samples were evaluated with Thermo XCalibur software (version 2.1.0) and downstream alignment done with an in-house data processing package called Omics Discovery Pipeline [63, 64] . Liquid chromatography/mass spectrometry for targeted sphingolipid analysis. The method was slightly modified from [65] . Briefly, each sample was extracted according to published protocol then dried and reconstituted in 100 mL of methanol : water : formic acid (74:25:1) containing 5 mM ammonium formate. An Agilent 6400 QQQ (Agilent Technologies, Santa Clara, CA) was used for analysis coupled to an 1100 Series LC equipped with HPLC Chip interface (Agilent Technologies Santa Clara, CA). Solvent A consisted of methanol : water : formic acid [65] with the following source conditions: gas temperature 300uC, gas flow 4 L/minute, capillary voltage 1900 V. The fragmentor voltage was set to 180 V in all cases. Data were processed with Agilent Mass Hunter software version B03.01. C6/36 cells were infected with DENV at an MOI of 3. Following adsorption, virus was removed and the cells were washed and overlayed with media containing varying concentrations of C75 [51] . The vehicle for C75 was ethanol. Virus supernatants were harvested at 24 hr post-infection and virus titer assayed by plaque assay. Cytotoxicity of C75 was simultaneously assayed using the Quick Cell Proliferation Kit (Abcam). TIME OF ADDITION OF C75: C6/36 cells were infected with DENV as described above. At the indicated time points, 6.3 mM C75 was added to the cells. Cells were harvested at 24 hr post-infection and the amount of released virus was determined by plaque assay. For the lipidomic studies, cells were infected with DENV (as described above) or left uninfected. Following adsorption of the virus at room temperature for 2 hr, C75 was added to the media in the overlay, and cells were incubated at 30uC for 36 hr. Sample preparation, lipid extraction and mass spectrometry analyses were carried out as described above for the fractionated (16K pellet) samples. Phospholipase A2 activity was monitored using a Red/Green BODIPY PC-A2 substrate (Invitrogen) using a similar method as described in [31] . C6/36 cells were infected with DENV at an MOI = 3 in 6-well plates. Following virus adsorption, the cells were overlayed with MEM (supplemented as above) and 10% heat inactivated fetal calf serum (2 ml/well). At the indicated time points, the media were removed, and new media (1 ml/well) containing 6 mM of the fluorogenic phospholipase A substrate (BODIPY-PC) were added to the cells. The cells were incubated at 30uC for 30 min. Following incubation, media was removed and the cells were washed with 16PBS. The lipids were then extracted using butanol-1 (1 ml/well). The aqueous fraction was discarded and the lipids in the butanol phase were analyzed by mass spectrometry to monitor the conversion of BODIPY-PC to BODIPY LCP by PLA 2 . Phospholipase A2 activity analysis. The lipid samples were reconstituted in 100 mL of methanol : water : formic acid (74:25:1) containing 5 mM ammonium formate and analyzed with an Agilent 6400 QQQ (Agilent Technologies, Santa Clara, CA) coupled to an 1100 Series LC equipped with HPLC Chip interface (Agilent Technologies Santa Clara, CA). Solvent A consisted of methanol : water : formic acid (74 : 25 : 1) containing 5 mM ammonium formate and solvent B methanol : formic acid (99 : 1) containing 5 mM ammonium formate. The gradient was as follows: time 0 minutes, 40% B; time 5 minute, 70% B; time 7 minutes, 100% B; time 9 minutes, 100% B; time 9 minutes, 40% B; time 13 minutes, 40% B. The flow rate was 0.4 mL/min. Data were acquired for parent ions corresponding to the BODIPY-PC (986.9 m/z) and BODIPY LCP (320.2 m/z) in single ion monitoring (SIM) mode with the following source conditions: gas temperature 300uC, gas flow 4 L/minute, capillary voltage 1900 V. The fragmentation voltage was set to 200 V in all cases. Data were processed with Agilent Mass Hunter software version B03.01. Supplemental data include three figure and three tables. carried out on 100 lipids. For each condition and time point, the following information is provided: Treatment P; p-values from the Anova analysis on minimum observations data (p,0.1), Exact mass; mass of each lipid from http://www.lipidmaps.org, [M+H] + ; Protonated molecular ion, NET; normalized elution time, total carbon/double bond; corresponds to each lipid species detected, LM_ID; identification for each lipid as displayed in LIPID MAPS, Formula, elemental composition of each lipid, PPM Error; difference in experimental mass compared to exact mass, Fold DENV/Mock; fold change of average abundance from 4 replicates of each treatment. Identity abbreviations were made for phoshatidylcholine (PC; O-fatty acid chain number means that an alkyl acyl linkage to the glycerol chain is present for the respective PC), phosphatidylethnolamine (PE), phosphatidylserine (PS), sphingomyelin (SM), ceramide (Cer), ceramide phosphoethanolamine (Cer-PE), lysophosphatidylcholine (LPC). The notation further indicates total number of carbons and double bonds however it does not discern redundancy associated with varying fatty acid composition for the same molecular weight. Accession numbers (LM_ID) were obtained from the Lipid Maps Gateway (lipidmaps.org). (PDF) Table S2 Select list of lipid species from the analysis of the 16K membrane fraction (isolated from mosquito cells treated with conditions described below) regulated across conditions (p,0.05). Conditions: Mock (uninfected cells), DENV (infectious dengue virus type 2, strain 16681), UV-DENV (UV-inactivated dengue virus type 2, strain 16681). A total of 484 features were detected in the mass spectrometry analysis following normalization of Mock, DENV and UV-DENV treated samples and 415 of these features were significantly regulated (p,0.05). Identification was successful for 68 lipids (ppm error ,10). For each condition the following information is provided: pttest and pwilcox; p-values on minimum observation data (a measurement must be present in two of three replicates to be considered present), Exact mass; mass of each lipid from http:// www.lipidmaps.org, M/Z; mass to charge ratio, RT; retention time, Formula, elemental composition of each lipid, PPM Error; difference in experimental mass compared to exact mass, Fold DENV/Mock; Fold UV-DENV/Mock; fold change of average abundance from 3 replicates of each treatment. Identity abbreviations were made for phoshatidylcholine (PC), phosphatidylethnolamine (PE), phosphatidylglycerol (PG, O-fatty acid chain number means that an alkyl acyl linkage to the glycerol chain is present for the respective PG), sphingomyelin (SM), monoacylglycerol (MG), diacylglycerol (DG), phosphatidic acid (PA). Accession numbers were obtained from the Lipid Maps Gateway (lipidmaps.org) and the Human Metabolome Database (hmdb.ca). (PDF) Table S3 Select list of lipid species from the analysis of the 16K membrane fraction (isolated from mosquito cells treated with conditions described below) regulated across conditions (p,0.05). Conditions: DENV (cells infected with DENV in the presence of vehicle only), DENV C75 (cells infected with DENV and treated with the FAS inhibitor C75). A total of 366 features were detected in the mass spectrometry analysis following normalization of DENV and DENV C75 treated samples and 314 of these features were significantly regulated (p,0.05). Identification was successful for 27 lipids (ppm error ,10). For each condition the following information is provided: pttest and pwilcox; p-values on minimum observation data (a measurement must be present in two of three replicates to be considered present), Exact mass; mass of each lipid from http://www.lipidmaps.org, M/Z; mass to charge ratio, RT; retention time, Formula, elemental composition of each lipid, PPM Error; difference in experimental mass compared to exact mass, Fold DENV/DENV C75; fold change of average abundance from 3 replicates of each treatment. Identity abbreviations were made for phoshatidylinositol (PI), phosphatidyllglycerol (PG), monoacylglycerol (MG), diacylglycerol (DG), phosphatidic acid (PA). Accession numbers were obtained from the Lipid Maps Gateway (lipidmaps.org) and the Human Metabolome Database (hmdb.ca). (PDF)

Human Cardioviruses, Meningitis, and Sudden Infant Death Syndrome in Children

Cardioviruses cause myocarditis and encephalomyelitis in rodents; human cardioviruses have not been ascribed to any disease. We screened 6,854 cerebrospinal fluid and 10 myocardium specimens from children and adults. A genotype 2 cardiovirus was detected from a child who died of sudden infant death syndrome, and 2 untypeable cardioviruses were detected from 2 children with meningitis.

Cardioviruses cause myocarditis and encephalomyelitis in rodents; human cardioviruses have not been ascribed to any disease. We screened 6,854 cerebrospinal fl uid and 10 myocardium specimens from children and adults. A genotype 2 cardiovirus was detected from a child who died of sudden infant death syndrome, and 2 untypeable cardioviruses were detected from 2 children with meningitis. Cardiovirus) are pathogens of rodents and include a murine encephalomyocarditis virus and Theiler's virus and related strains (species Theilovirus), the latter serving as laboratory models of the pathogenesis of multiple sclerosis in mice (1) . The existence of specifi c human cardioviruses was suspected in the 1960s in conjunction with a rare infectious neurodegenerative disease known as Vilyuisk encephalitis (2, 3) . Recently, human cardioviruses (hCVs) were identifi ed in archived diagnostic cell culture supernatants (4) and in clinical samples from children with diarrhea or respiratory infection (5, 6) . Up to 8 different putative hCV types have since been characterized in human feces (7) . Despite the remarkable pathogenicity of rodent cardioviruses, specifi c disease associations of hCV could not be made. An initial clinical study yielded no evidence of hCV in cerebrospinal fl uid (CSF) of 400 patients with aseptic meningitis, encephalitis, or multiple sclerosis (8) . To evaluate the pathogenetic potential of these emerging viruses, we investigated 6,854 CSF specimens from adults and children with neurologic disease and 10 myocardium specimens from infants who had died of sudden infant death syndrome (SIDS). CSF specimens were collected from 3 cohorts. The fi rst cohort comprised 2,562 specimens sent during 1998-2008 to the Institute of Virology, University of Bonn Medical Center (UBMC), Bonn, Germany, for routine investigation of meningoencephalitis (333 from the Department of Pediatrics and 2,229 from other departments). The second cohort comprised 3,960 specimens collected during 1982-2008 at the UBMC children's hospital from children with cancer and neurologic complications during chemotherapy. The third cohort comprised 348 specimens from hospitalized children with clinical meningitis or encephalitis in which no etiologic agent had been found; the specimens were sent for virologic investigation to the Institute for Hygiene and the Environment in Hamburg, ≈400 km from UBMC, during 2006-2008. Myocardium specimens were collected during 2010 at the UBMC Institute for Forensic Medicine from 10 epidemiologically unlinked children who died of SIDS. Viral RNA was purifi ed from clinical specimens by using the Viral RNA Mini and RNeasy Mini kits (QIAGEN, Hilden, Germany). Detection of hCV RNA was done in pools of 2-10 specimens by using quantitative realtime reverse transcription PCR (RT-PCR) and nested RT-PCR specifi c for the viral 5′ untranslated region (5′-UTR), as described (6) . Amplifi cation of further hCV genomic regions from individual positive specimens was conducted by using ≈20 sets of different nested RT-PCRs (primers available on request from C.D.). In 2 of 681 CSF specimens (n = 333 and n = 348 from cohorts 1 and 3, respectively) from children with meningitis (online Appendix Table, wwwnc.cdc.gov/EID/ article/17/12/11-1037-TA1.htm), hCV RNA was detected at low concentrations (1.14 × 10 4 and 9.63 × 10 2 copies/ mL). In 1 of these patients, hCV was also detectable in feces (9.50 × 10 2 copies/g). In 1 of 10 myocardium specimens, hCV was detected by nested RT-PCR, and results of quantitative real-time RT-PCR were negative. Underquantifi cation because of nucleotide mismatches below oligonucleotide binding sites and contamination of nested RT-PCR was excluded by sequence comparison (up to 5% nt divergence from other hCV strains, including the positive control). Serum and liver specimens from the patient who died of SIDS were negative according to realtime RT-PCR. No histopathologic alterations could be observed in myocardial tissue from this same patient. To evaluate whether detected hCV strains differed from previously described genotypes, amplifi cation and nucleotide sequencing of additional genomic regions was attempted. In a case of meningoencephalitis (specimen 07/03981), we sequenced a 1,297-nt fragment comprising the near complete 5′-UTR and the fi rst 489 nt of the structural protein gene (leader, viral protein [VP] 4 domain, and upstream VP2 domain, GenBank accession no. JN209931). Despite repeated trials, further sequence fragments could be amplifi ed neither from the specimen from this patient nor from that from the second patient with meningoencephalitis that showed very low virus concentrations (specimen VI1607). From the specimen from the SIDS patient (specimen 347/10), amplifi cation of the complete structural genome and partial nonstructural genome was successful (5,333 nt, GenBank accession no. JN209932). This virus belonged to hCV genotype 2 in the VP1 genomic region (i.e., the region used for the designation of genotypes) (Figure, panel A) . The CSF specimen 07/03981 was also phylogenetically related to genotype 2 viruses in the 5′-UTR and Leader-VP2 genomic regions ( Figure, panels B and C). On the basis of the 5′-UTR sequences, the closest known relative to both viruses was D/VI2229, obtained in Germany in 2004 (nucleotide percentage distance 4.7% for the SIDS specimen and 0.9% for the CSF specimen). In the structural protein gene fragment, the closest relative of both viruses was a strain obtained in the Netherlands in 2008 (Nijmegen2008, nucleotide distance 13.9% for the SIDS specimen and 3.5% for the CSF specimen). This suggested geographic rather than phylogenetic clustering of viruses detected within and beyond the respiratory and enteric tracts. However, formal and fi nal virus typing is pending because VP1 regions could not be sequenced from 2 viruses. Absence of other detectable pathogens in 1 of the meningoencephalitis case-patients (07/03981) made causation by hCV plausible (online Appendix Table) . For the second case (VI1607), an enterovirus was co-detected by real-time RT-PCR in CSF and feces. Serotyping from feces classifi ed this virus as echovirus type 30, known to cause aseptic meningitis. For the specimen from the child who died of SIDS, a rhinovirus was co-detected at low concentrations (real-time RT-PCR threshold cycle value >40), most compatible with shedding after previous respiratory infection (9) . The detection of hCVs in body compartments beyond the respiratory and enteric tracts is novel and suggests a role of these viruses in organ-related disease. A low detection rate in CSF does not contradict a general potential of these viruses to cause meningoencephalitis, as exemplifi ed by enteroviruses for which lack of detection in CSF despite clear association with disease is not uncommon (10) . Considering links between the related Theilovirus and demyelinating disease in laboratory models (1) outcomes of patients with hCV infection of the central nervous system should be followed up. Such longitudinal studies should include suffi cient numbers of patients because natural infections with Theilovirus in rodents are common and will less frequently result in multiple sclerosis-like disease than in laboratory models (1) . The rarity of hCV detection in our study suggests the assembly of such cohorts to be a diffi cult and lengthy task that could benefi t greatly from international coordination. Despite the absence of histopathologic alterations, the detection of hCV in a child who died of SIDS is remarkable because the related encephalomyocarditis virus constitutes a prototypic model for myocarditis in mammals (11) . Again, the high human seroprevalence against hCV (12) will complicate epidemiologic studies, yet investigations of links between hCV and SIDS are highly justifi ed because diarrhea is an acknowledged risk factor for SIDS (13) . A limitation of our study is that the VP1 genomic region of the viruses detected in CSF could not be obtained. In analogy to enteroviruses and parechoviruses, certain genotypes may be associated with distinct disease profi les, like polioviruses with encephalitis or parechovirus 3 with meningitis (14) . Although we were able to classify the virus detected in the child who died of SIDS as a common genotype 2, the partial hCV sequence from a patient with meningitis did not permit typing because hCVs, as all picornaviruses, recombine frequently (15) . We thus cannot exclude that the viruses detected in the meningitis cases may have acquired distinct features in their capsid protein or elsewhere that might infl uence pathogenicity.

Knowledge of Avian Influenza (H5N1) among Poultry Workers, Hong Kong, China

In 2009, a cross-sectional survey of 360 poultry workers in Hong Kong, China, showed that workers had inadequate levels of avian influenza (H5N1) risk knowledge, preventive behavior, and outbreak preparedness. The main barriers to preventive practices were low perceived benefits and interference with work. Poultry workers require occupation-specific health promotion.

In 2009, a cross-sectional survey of 360 poultry workers in Hong Kong, China, showed that workers had inadequate levels of avian infl uenza (H5N1) risk knowledge, preventive behavior, and outbreak preparedness. The main barriers to preventive practices were low perceived benefi ts and interference with work. Poultry workers require occupationspecifi c health promotion. I n 1997, a zoonosis in humans caused by a highly lethal strain of avian infl uenza virus (H5N1) was reported in Hong Kong. Live-poultry markets were the source of this outbreak (1) . As one of the world's most densely populated regions (16,000 persons/mile 2 [>6,300 persons/km 2 ]) (2), Hong Kong is a city at high risk for a large-scale outbreak of avian infl uenza caused by live poultry in large-volume wholesale markets and within neighborhood wet markets (open food stall markets). Because members of the average household in Hong Kong shop in wet markets on a habitual basis, these markets are located in the most densely populated areas ( Figure) and are commonly multistory complexes or in basement levels of shopping centers. Because poultry workers are a potential bridge population (3, 4) , the government has instigated voluntary avian infl uenza training since 2001 that reviews regulations for workplace disinfection, waste disposal, poultry storage, and personal hygiene measures (5,6). Despite occupational risk for exposure to avian infl uenza (7, 8) , there have been few studies of poultry workers (8) (9) (10) (11) (12) . Most studies were conducted in rural settings in developing countries (9) (10) (11) (12) , but fi ndings cannot be readily extrapolated to cities such as Hong Kong because of differences in food-handling practices and occupational settings. Knowledge, perceptions, and work practices of live-poultry workers in Hong Kong have not been examined. Therefore, a survey of these workers is timely and warranted, given confi rmed persistence of avian infl uenza in Asia. (13) The Study An anonymous, cross-sectional survey was conducted during June-November 2009. Interviewers approached 132 licensed live-poultry retail businesses in wet markets and 23 wholesale establishments. The fi nal sample was 360 poultry workers (194 retailers and 166 wholesalers; response rate 68.1%). Respondents were asked about their demographics, past month's work and preventive behavior, and avian infl uenza-related knowledge on the basis of a World Health Organization factsheet (14) . We asked perception questions based on the Health Belief Model and the likelihood of adopting certain behavior patterns in the event of a local bird-to-bird or bird-to-human outbreak of avian infl uenza. Summative scores were computed for avian infl uenzarelated knowledge, current preventive behavior patterns, outbreak preparedness, and various perception domains. Higher scores refl ected more benefi cial levels of each domain. Unconditional multilevel regression indicated no evidence of clustering effect by poultry market. Standard multivariable linear regression was conducted by using SAS version 9.1.3 (SAS Institute, Cary, NC, USA) with knowledge, practice, and preparedness scores as outcomes and potential predictors showing p<0.25 in unadjusted analyses as input variables. Distribution of standardized residuals and their association with predicted values were examined to assess model assumptions. Most (208, 60.1%) respondents were men 35-54 years of age, of whom 192 (55.3%) had worked a mean of 16.1 years in the poultry industry. Respondents showed low mean summative scores for knowledge of avian infl uenza (online Appendix incorrectly believed that a human vaccine for avian infl uenza was available. Most (208, 89.9%) respondents were familiar with infl uenza-like symptoms of avian infl uenza virus (H5N1) infection such as fever, but fewer workers were aware of respiratory and gastrointestinal symptoms of virus infection. The Internet and other sources (e.g., health talks) of information about avian infl uenza were strong independent predictors of greater knowledge. However, training did not result in higher knowledge levels. Poultry workers reported low-to-moderate levels of compliance with hand hygiene and other preventive measures (ranging from 7.3% [36] using eye protection to 65.2% [245] using handwashing with soap after slaughtering poultry). Working in the poultry industry ≥10 years, lower perceived barriers to preventive behavior, and retail poultry work were independent predictors of higher preventive behavior scores. With regard to avian infl uenza-related perceptions, lack of training (277, 83.4%) and the view that compliance with all infection regulations is diffi cult during peak hours (218, 64.9%) were the most frequently cited barriers to adoption of preventive behavior. A total of 154 (46.4%) workers believed that face masks reduced business, and 153 (46.1%) believed that vaccination was expensive. Low anxiety about illness was reported by 242 (76.6%) respondents. In the event of a local outbreak, workers expressed various levels of acceptance for precautionary actions, ranging from 15.8% (56) for reducing work hours to 82.4% (290) for seeking medical care for infl uenza-like symptoms. Ninety-six (27.4%) respondents anticipated taking oseltamivir. Greater perceived benefi t of preventive behavior was the strongest independent predictor of higher preparedness scores (online Appendix Table 2 , wwwnc.cdc. gov/EID/article/17/12/11-0321-TA2.htm). Similar to other regions (8) (9) (10) (11) , poultry workers in Hong Kong showed low risk perceptions for avian infl uenza, inadequate knowledge, and a wide range of compliance with preventive measures. Because training (6) was not associated with overall preventive behavior or preparedness, there may be an unmet need for occupationspecifi c health information. Higher levels of knowledge demonstrated by workers who accessed health information sources (e.g., Internet) that provide detailed information suggest that comprehensive, occupation-relevant information should be more widely accessible. However, occupational practices of animal workers might not be amenable to change solely on the basis of improvements in knowledge. Only 129 (42.1%) respondents reported that poultry workers could realistically adhere to all government guidelines (6) . Interference with work, high cost, and reduction of business were repeatedly cited as impediments to the adoption of preventive behavior. Even in the event of local outbreaks of avian infl uenza, most workers were not amenable to actions having adverse economic effects such as reducing work hours. Animal workers are thereby unlikely to widely adopt preventive behavior if these measures confl ict with their economic interests. Despite the ongoing government regulations regarding avian infl uenza in Hong Kong (6), a complete ban on live poultry is unrealistic because of the culturally entrenched demand for fresh poultry. Increasing knowledge and risk perceptions while simultaneously reducing occupational barriers to preventive behavior thereby continues to be the cornerstone of effective zoonotic infection control among animal workers. Implications of these fi ndings extend to other poultryborne pathogens, such as Campylobacter spp. and Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 17, No. 12, December 2011 Salmonella spp., which share common preventive measures. Close adherence to workplace measures will likely reduce outbreak risk for other poultry-borne diseases. Therefore, a framework for greater integration of risk management strategies and worker education of these poultry-borne infections tailored to the local context is worthwhile and cost-effective. In the spirit of the One Health Commission, which calls for an integrated, interdisciplinary approach to human-veterinary-environmental health challenges (15), the fi ght against global pandemics, such as those of avian infl uenza virus (H5N1), necessitates greater dialogue and collaborative leadership between governments and livestock industries. Development of realistic occupational safety measures is an ongoing challenge for national governments.

A Pilot Study of Host Genetic Variants Associated with Influenza-associated Deaths among Children and Young Adults

We compared the prevalence of 8 polymorphisms in the tumor necrosis factor and mannose-binding lectin genes among 105 children and young adults with fatal influenza with US population estimates and determined in subanalyses whether these polymorphisms were associated with sudden death and bacterial co-infection among persons with fatal influenza. No differences were observed in genotype prevalence or minor allele frequencies between persons with fatal influenza and the reference sample. Fatal cases with low-producing MBL2 genotypes had a 7-fold increased risk for invasive methicillin-resistant Staphylococcus aureus (MRSA) co-infection compared with fatal cases with high- and intermediate-producing MBL2 genotypes (odds ratio 7.1, 95% confidence interval 1.6–32.1). Limited analysis of 2 genes important to the innate immune response found no association between genetic variants and fatal influenza infection. Among children and young adults who died of influenza, low-producing MBL2 genotypes may have increased risk for MRSA co-infection.

I t is unknown why some apparently healthy persons become severely ill after infl uenza infection while others infected by the same strain remain asymptomatic or become only mildly ill. The presence of neutralizing antibody to a specifi c infl uenza strain is protective, and certain chronic medical conditions increase the risk for severe outcomes of infl uenza infections, but the risk factors for infl uenzaassociated deaths among previously healthy persons remain largely unknown (1) . Infectious disease mortality risk has a heritable component; children of parents who died of an infectious disease are ≈6× more likely to die of an infectious cause compared with the general population (2) . A recent large family study that used genealogic databases found an elevated risk for infl uenza death among relatives of persons who died of infl uenza (3) . By comparing the infl uenza mortality rate for relatives of persons who died of infl uenza with the infl uenza mortality rate for relatives of spouses of persons who died, the authors showed that the increased risk was not explained by shared exposure to infl uenza virus and thus may have a genetic component. However, to our knowledge, no published studies have examined the association between specifi c host genetic variants and severe infl uenza disease outcomes. To address the paucity of research on host genomics and infl uenza, the Centers for Disease Control and Prevention (CDC) convened a meeting of experts in 2007 to solicit opinions on how to explore the role of host genomics in public health activities for infl uenza conducted by the agency. A study of host genomic factors related to severe infl uenza outcomes in children was recommended as an activity that CDC was well positioned to pursue. This article reports the fi ndings of the study implemented in response to that recommendation. We conducted a hypothesis-generating pilot study to examine if host genetic variants were associated with fatal infl uenza virus infection by comparing prevalence of selected host genetic variants among children and A Pilot Study of Host Genetic Variants Associated with Infl uenzaassociated Deaths among Children and Young Adults 1 young adults who died of infl uenza with population-based prevalence estimates. We focused on 8 single-nucleotide polymorphisms (SNPs) in 2 candidate genes important in the innate immune response to infl uenza infection and for which national prevalence estimates were available: the gene for tumor necrosis factor superfamily, member 2 (offi cial symbol TNF) and the mannose-binding lectin gene (offi cial symbol MBL2). Because infl uenza-associated deaths in children, but not adults, are nationally reportable in the United States, most cases in this study were pediatric cases reported to CDC through the Infl uenza-associated Pediatric Mortality Surveillance system. This system requires state public health authorities to report to CDC any infl uenza-associated death among persons <18 years old that occurred within their jurisdiction. Information collected by this surveillance system constitutes the primary phenotypic information used in this study and includes underlying health status and chronic medical conditions, infl uenza vaccination status, clinical course and features, and results of microbiologic and virologic testing. Reporting to this surveillance system does not require submission of tissue samples; however, CDC routinely receives tissue samples for a subset of fatal pediatric infl uenza cases for diagnostic confi rmation. For some cases, medical records and autopsy reports provided additional information. A total of 442 infl uenza-associated deaths among children (<18 years old) and young adults (18-40 years old) residing in the United States were reported to CDC for the 1998-99 through 2007-08 infl uenza seasons; of these, 105 cases with laboratory-confi rmed infl uenza infection had suffi cient tissue specimens available for DNA extraction and constitute the analytic dataset for this study. Fatal infl uenza cases were considered laboratory confi rmed if a positive test result for infl uenza by viral culture, immunohistochemical analysis, or reverse transcription PCR (RT-PCR) had been documented. These represented 1) fatal pediatric cases reported to CDC during the 2003-04 infl uenza season when CDC conducted surveillance for infl uenza-associated pediatric deaths as part of an emergency response effort; 2) fatal pediatric cases identifi ed through national surveillance since 2004 when pediatric infl uenza-associated death was made nationally notifi able in the United States; or 3) fatal cases of infl uenza among young adults at any point in time or among children before 2003 whose case reports and specimens were received by the CDC Infectious Diseases Pathology Branch on a case-by-case basis. To obtain DNA for genotyping, a 10-μm section from blocks containing formalin-fi xed, paraffi n-embedded tissues was deparaffi nized with xylene and washed twice with absolute ethanol. After residual ethanol evaporated, tissues were digested overnight at 56°C in 200 μL Buffer PKD with 20 μL proteinase K (QIAGEN, Valencia, CA, USA). Extraction of the supernatant was completed with an EZ1 DNA Tissue Kit or a MagAttract DNA Mini M48 Kit (QIAGEN), with DNA eluted into a fi nal 100-μL volume. DNA quality was assessed with a human RNase P real-time PCR in 25-μL volumes by using Agilent Brilliant II QPCR Master Mix as described (4) . Validated TaqMan assays were used to genotype each SNP (protocols, primers, and probes available at http://snp500cancer.nci.nih.gov). Each 25-μL real-time PCR consisted of 12.5 μL of TaqMan Universal PCR Master Mix (Applied Biosystems, Foster City, CA, USA), 900 nmol of assay-specifi c primer, 200 nmol of assay-specifi c probe, and 5 μL of DNA. All controls (extraction blanks, no template controls, and positive controls for each genotype used at 5 ng per PCR; Coriell Institute for Medical Research, Camden, NJ, USA) and unknown samples were assayed in duplicate. Thermal cycling conditions consisted of 1 cycle at 50°C for 2 min, 1 cycle at 95°C for 10 min, and 50 cycles of 92°C for 30 s and 60°C for 1 min. Data were collected during the annealing plateau. For TNF, we examined 3 promoter SNPs: −308G>A (rs1800629), −238G>A (rs361525), and −555G>A (rs1800750) (5,6); we were unable to infer TNF haplotypes. For MBL2, we examined 5 SNPS, 3 in the coding region of exon 1 and 2 in the promoter region. The 3 structural SNPs in MBL2 that we examined encode variant alleles known as D (codon 52, rs5030737), B (codon 54, rs1800450), and C (codon 57, rs1800451); the wild-type is A (7, 8) . These variants are typically pooled and designated as the O allele. The MBL2 genotype A/A refers to wild-type homozygotes, A/O refers to heterozygotes, and O/O refers to homozygotes or compound heterozygotes. Promoter polymorphisms at positions −550 (H/L variant, rs11003125) and −221 (X/Y variant, rs7096206) encode variants that mediate MBL2 expression. Case-patients were classifi ed as low, intermediate, or high producers of MBL on the basis of their structural and promoter variants (referred to as a "truncated haplotype") (7). Case-patients homozygous or compound heterozygous for any of the 3 variant structural alleles and case-patients with a variant structural allele on 1 chromosome and the X variant on the other were categorized as low MBL producers. Case-patients homozygous for the wild-type structural allele were categorized as high MBL producers except for those also homozygous for the X variant, who were classifi ed as intermediate MBL producers on the basis of evidence that possession of the X/X promoter genotype signifi cantly down-regulates MBL production (9) . Case-patients with the YA/O genotype were classifi ed as intermediate MBL producers on the basis of analyses indicating that this genotype confers intermediate levels of functional MBL (9) . For some analyses, the intermediate and high producers were combined into 1 group and compared with MBL low-producers. The prevalence of genetic variants among cases was compared with population-based prevalence estimates for the same genetic variants for the 12-19-year age group available from the National Health and Nutrition Examination Survey (NHANES) III CDC-National Cancer Institute Collaborative Genomics Project databank (10) . NHANES is a nationally representative survey of the US population conducted by the CDC National Center for Health Statistics. During the second phase of NHANES III (1991-1994), leukocytes from participants were used to create a DNA bank maintained by CDC's National Center for Environmental Health that contains specimens from >7,000 participants, including ≈1,200 children. To our knowledge, the NHANES DNA bank is the only currently available source of nationally representative prevalence estimates for genetic variants among US residents. The 12-19-year age group is the youngest age group available in the NHANES DNA bank. Cases were stratifi ed by presence or absence of any chronic medical conditions in the patients known to increase the risk for infl uenza-associated complications (including moderate to severe developmental delay; hemoglobinopathy, immunosuppressive disorders, asthma or reactive airway disease, diabetes mellitus, history of febrile seizures, seizure disorder, cystic fi brosis, or cardiac, renal, chronic pulmonary, metabolic, or neuromuscular disorders) (11) . Case-patients without chronic medical conditions were classifi ed as "previously healthy." Casepatients who were admitted to an inpatient ward or intensive care unit were classifi ed as "hospitalized." Length of illness was defi ned as the duration of time between the reported date of illness onset and death. Case-patients with length of illness <3 days were classifi ed as having "sudden death." Bacterial co-infection was defi ned as at least 1 positive culture for a bacterial pathogen from a normally sterile site (e.g., blood, cerebrospinal fl uid). Minor allele frequencies between groups were compared with a test of binomial proportions. The null hypothesis was that there was no difference in minor allele frequency between the cases and the reference sample. A priori groups examined in subgroup analyses included previously healthy case-patients, case-patients <5 years old, case-patients with invasive bacterial co-infection, and case-patients with sudden death. Differences in length of illness were evaluated with the Kaplan-Meier estimator with differences tested with the log-rank statistic. Tests of signifi cance were based on a 2-sided test with α = 0.05. Tests of departure from Hardy-Weinberg equilibrium for the reference sample have been published (10) . Analyses were conducted in SAS version 9.2 (SAS Institute, Cary, NC, USA). This study was exempted from institutional review board review for approval of human subjects research. Data were obtained only from deceased case-patients, and reference sample data were used only in a de-identifi ed and aggregate manner. Of 442 cases of fatal infl uenza in children and young adults reported to CDC during the 1998-99 through 2007-08 infl uenza seasons, 105 (24%) cases had available autopsy specimens with suffi cient DNA for genotyping. Case-patient characteristics are summarized in Table 1 . Genotyped casepatients had a median age of 6.0 years (range 1 month-40 years) and 52% were female. Sixty-one percent of casepatients were white, and 17% were black. Seventy-four percent of cases occurred during 3 infl uenza seasons: 2003-04 (31%), 2006-07 (21%), and 2007-08 (22%). Eighty-one (77%) of 105 case-patients were infected with infl uenza A and 24 (23%) with infl uenza B. There were no signifi cant differences in the distribution of infl uenza types by season between cases and the national pattern of types found in the US viral surveillance system (data not shown). Compared with case-patients who were not genotyped, the 105 case-patients with DNA available for genotyping were slightly older (median age 6 years vs. 4 years; p<0.05), less likely to have had a preexisting medical condition (28% vs. 61%; p<0.001), and less likely to have been vaccinated for infl uenza during the season of death (7% vs. 16%; p<0.01). Case-patients genotyped were more likely to have experienced sudden death (31% vs. 22%; p<0.05) and to have died before reaching medical care (34% vs. 22%; p<0.001). It is not surprising that case-patients with sudden death were more likely to have undergone autopsy and, hence, to have had tissues available for DNA extraction. Genotyped case-patients were less likely to have had pneumonia evident on chest radiograph (22% vs. 46%; p<0.05) and about equally likely to have had invasive bacterial co-infection (21% vs. 23%; not signifi cant), but differences in these characteristics are diffi cult to interpret because genotyped case-patients were less likely to have received medical care for their illnesses (presumably because of a greater frequency of sudden death). Genotype and minor allele frequencies among casepatients are summarized in Table 2 . Minor allele frequencies comparing case-patients to the NHANES reference sample are shown in Figure 1 . No statistically signifi cant differences were observed in minor allele frequencies or genotype prevalence between the case-patients and the NHANES reference sample for the 3 TNF variants with all case-patients examined together or with black and white racial groups examined separately. No statistically signifi cant differences were observed in minor allele frequencies for the 5 MBL2 SNPs examined (Figure 1 ) or the prevalence of pooled MBL2 genotypes (Figure 2 ) between the case-patients and the NHANES reference sample with all case-patients examined together or with black and white racial groups examined separately. In a subgroup analysis, the minor allele frequency of rs5030737 was signifi cantly less common among case-patients <5 years old than in the reference sample (2% vs. 7.2%; p = 0.02). Among low producers of MBL, we observed an estimated odds ratio of 7. Table 3 ). Low-producing MBL2 genotypes were also associated with an approximate 3-fold increased risk for bacterial coinfection in general and with S. aureus infection overall, but these associations did not reach statistical signifi cance. Characteristics of case-patients with invasive MRSA coinfection are shown in Table 4 . We found no signifi cant differences in allele frequencies or genotype prevalence for variants in the TNF and MBL2 genes between fatal infl uenza cases in patients <40 years old and a nationally representative reference sample. However, among the case-patients who died, most of whom died in childhood, variants of MBL2 responsible for low production of MBL were associated with MRSA co-infection. This observation should be viewed cautiously as a hypothesis for further exploration, given the small number of case-patients with MRSA in our study (n = 8). This fi nding is consistent with results from previous studies that found associations between MBL insuffi ciency (defi ned by genotype) and respiratory infection in children (12) (13) (14) , severe and fatal sepsis (9, (15) (16) (17) , and systemic infl ammatory response syndrome in children (18) . TNF is a potent proinfl ammatory cytokine produced early in the innate immune response to infection that promotes a wide range of immunologic responses. Excessive systemic TNF is responsible for many symptoms of clinical infection and may lead to fatal complications. Studies have demonstrated a signifi cant genetic contribution to circulating TNF levels, with 50%-60% of variance in TNF levels genetically determined (19) (20) (21) . The most studied SNP is at position −308 (rs1800629), with the A allele associated with 20%-40% greater TNF production (22) (23) (24) and with susceptibility to and severity of numerous infectious diseases (20, 22, 25, 26) . Carriage of the A allele at the −238 position (rs361525) also has been associated with a variety of diseases (20, 22) . MBL, another key component of the innate immune system, is a soluble protein of the collectin family that binds to microbial surfaces and promotes phago-opsonization directly and indirectly by activating the lectin complement pathway. Low serum MBL levels are common and associated with an increased risk for a variety of infections and autoimmune diseases (15, (27) (28) (29) , including acute respiratory infection in young children (12) . MBL levels are strongly infl uenced by genetic factors, with >75% of variation in MBL levels explained by a small number of polymorphisms in the MBL2 gene (30) . Variant proteins are unstable and of lower oligomeric form, which decreases affi nity for microbial ligands and complement-activating ability. Each variant produces signifi cantly reduced serum MBL levels. MBL has been shown to strongly bind S. aureus (31) and susceptibility to fatal S. aureus infection due to MBL defi ciency has been convincingly demonstrated in murine models (32) . Phase I clinical trials of MBL replacement therapy indicate that this therapy is well tolerated and effective at improving MBL defi ciency in healthy persons (33) . Reports of MBL replacement therapy administered to severely ill persons (34) (35) (36) or to patients with S. aureus sepsis (37) suggest that therapy can improve clinical conditions, although results of these studies were mixed, and in some cases, clinical improvements were temporary. The clinical implications of MBL replacement therapy for infl uenza treatment or prevention are unknown. Among persons with fatal cases, we observed an increased risk for sudden death in carriers of the variant allele of TNF rs1800750. We are unaware of previous literature reporting a similar association; there is no obvious biologic mechanism to explain the fi nding. The TNF rs1800750 variant is in linkage disequilibrium with other TNF variants (http://pga.gs.washington.edu), some of which (including TNF rs361525) have been associated with increased TNF serum levels. Therefore, it is possible that the observed association may be due to linkage disequilibrium with unmeasured polymorphisms that are the causal variants, and more exhaustive analysis of TNF variants is worthy of future study. A strength of this study is its use of a cohort of case-patients particularly well-suited for investigation of potential host genetic risk factors-these case-patients died with active infl uenza infections, yet were predominantly children and young adults without severe preexisting medical conditions. In such a group, other factors associated with severe infl uenza are less likely to obscure possible genetic associations. An additional strength was access to postmortem lung tissue for immunohistochemistry and/or RT-PCR confi rmation of infl uenza infection. We recognize that this study has several limitations. Although the study cohort is, to our knowledge, the largest sample of fatal infl uenza cases in children and young adults, the analysis has limited statistical power to detect associations because of small sample sizes, especially when examining subsamples. We had access to limited information about racial and ethnic background of casepatients. Clinical data were obtained primarily from a US surveillance system and were not validated with medical chart review. Although we were able to infer truncated haplotypes for MBL2, haplotype information for TNF was unavailable. Despite these shortcomings, the possibility that specifi c variants of the MBL2 gene known to infl uence serum MBL levels appear to be associated with severe bacterial co-infection is an intriguing fi nding deserving of additional study, especially given the prevalence of co-infection among case-patients who died of pandemic (H1N1) 2009 virus infection (38) and observations that children co-infected with infl uenza and S. aureus may have higher case-fatality rates (39) . That we observed a stronger relationship between low-producing MBL genotypes and MRSA infection than between those genotypes and S. aureus infection in general is puzzling. We are unaware of an obvious physiologic explanation for why low MBL would predispose more strongly to infection with methicillin-resistant versus methicillin-sensitive S. aureus. One possibility is that MRSA is a marker for other strain characteristics. For example, such an association could arise if MRSA infections were predominantly the USA300 strain while other S. aureus infections were predominantly the USA100 strain. Unfortunately, we do not have data on S. aureus genetic strain types. We also found that of the 4 fatal infl uenza cases in which patients had both MRSA co-infection and low-producing MBL genotypes, 2 patients reportedly also had asthma. It is well-established that asthma increases the risk for serious complications of infl uenza, and although we know of no evidence suggesting that low-producing MBL genotypes are associated with increased risk for asthma (40) , this fi nding may be worth further exploration in future studies. Our fi ndings suggest several opportunities for additional infl uenza-related research. An obvious next step is examination of all functional variants of the MBL2 gene in conjunction with gene expression and functional assays in a larger group of severely ill infl uenza case-patients with suffi ciently detailed clinical data to defi ne important phenotypes (e.g., MRSA co-infection). Interest in association studies of rare variants, the availability of new sequencing technologies that dramatically decrease the cost of sequencing, and access to reference human sequence data suggest that investigating rare variants in candidate genes (including MBL2 and TNF) and their functional effects may be a promising avenue of research. Large-scale genotyping of a sample of case-patients to look for common variants by using methods such as genomewide association studies may be possible if a network of collaborators capable of pooling a suffi cient number of case-patients is developed. Recent initiatives such as the Genome-based Research and Population Health International Network (www.graphint. org/ver2) are aimed at encouraging such networks. Given the rapid acceleration in laboratory technologies, enhancement in bioinformatics methods and capacity, and trends toward collaborative research within large consortia, exploration of the role of host genomic factors in serious illness associated with infl uenza and other viral pathogens is increasingly feasible. We believe that host genomics is a promising area for future research regarding who is at risk for severe complications of acute infectious diseases, including infl uenza.

Clinical features and risk factors for severe and critical pregnant women with 2009 pandemic H1N1 influenza infection in China

BACKGROUND: 2009 pandemic H1N1 (pH1N1) influenza posed an increased risk of severe illness among pregnant women. Data on risk factors associated with death of pregnant women and neonates with pH1N1 infections are limited outside of developed countries. METHODS: Retrospective observational study in 394 severe or critical pregnant women admitted to a hospital with pH1N1 influenza from Sep. 1, 2009 to Dec. 31, 2009. rRT-PCR testing was used to confirm infection. In-hospital mortality was the primary endpoint of this study. Univariable logistic analysis and multivariate logistic regression analysis were used to investigate the potential factors on admission that might be associated with the maternal and neonatal mortality. RESULTS: 394 pregnant women were included, 286 were infected with pH1N1 in the third trimester. 351 had pneumonia, and 77 died. A PaO(2)/FiO(2 )≤ 200 (odds ratio (OR), 27.16; 95% confidence interval (CI), 2.64-279.70) and higher BMI (i.e. ≥ 30) on admission (OR, 1.26; 95% CI, 1.09 to 1.47) were independent risk factors for maternal death. Of 211 deliveries, 146 neonates survived. Premature delivery (OR, 4.17; 95% CI, 1.19-14.56) was associated neonatal mortality. Among 186 patients who received mechanical ventilation, 83 patients were treated with non-invasive ventilation (NIV) and 38 were successful with NIV. The death rate was lower among patients who initially received NIV than those who were initially intubated (24/83, 28.9% vs 43/87, 49.4%; p = 0.006). Septic shock was an independent risk factor for failure of NIV. CONCLUSIONS: Severe hypoxemia and higher BMI on admission were associated with adverse outcomes for pregnant women. Preterm delivery was a risk factor for neonatal death among pregnant women with pH1N1 influenza infection. NIV may be useful in selected pregnant women without septic shock.

Pregnant women are at an increased risk for contracting influenza and its complications associated with influenza [1] . Like previous epidemic and pandemic diseases, 2009 pandemic H1N1 (pH1N1) influenza posed an increased risk of severe illness among pregnant women [2] [3] [4] [5] [6] [7] [8] [9] . A report from the first month of the pH1N1 outbreak noted that the rate of hospitalization among pregnant women was approximately four times the rate in the general population in the USA [3] . As reported by the California Department of Public Health (CDPH), a total of 10% of the 1088 patients who were hospitalized or died from the 2009 pH1N1 influenza were pregnant [10] . According to the Ministry of Health (MOH) of the People's Republic of China, pregnant women accounted for 13.7% of deaths associated with 2009 pH1N1 influenza [11] . Pregnant women with influenza appear to have an increased risk of miscarriage, premature birth and stillbirth [2, 12, 13] . Reports from Victoria in Australia [14, 15] , New York [16] , and California [17] , demonstrate that 2009 pH1N1 infection was associated with substantial maternal and fetal morbidity and mortality. However, information is limited concerning the risk factors for maternal and neonatal death when pregnancy is complicated by severe or critical illness related to 2009 pH1N1 influenza. In this report, we described the characteristics of pH1N1 influenza in pregnant women and the risk factors for maternal and neonatal death. All patients who were admitted to hospitals with confirmed 2009 pH1N1 influenza from Sep. 1 to Dec. 31, 2009 from 27 Chinese provinces were screened if they fulfilled the diagnostic criteria for severe or critical cases. A confirmed case was a person whose pH1N1 virus infection was verified by real-time reverse-transcriptase polymerase chain reaction (rRT-PCR) with or without the presentation of other clinical symptoms. Patients were excluded if they had been treated as outpatients or in emergency rooms or duration of hospitalization < 24 h, or if they had incomplete records of clinical outcomes. Severe and critical cases were defined according to the H1N1 2009 Clinical guidelines (Third Edition, 2009) released by the MOH (Additional file 1: Table S1 ). Our research retrospectively collected the patient's clinical information and did not involve the patient's personal information and samples, so there was no informed consent. The case report form included demographic information, underlying conditions, gestational age, vaccination status, treatment, intensive care unit (ICU) admission, complications, and maternal and neonatal outcomes. Body mass index (BMI) was calculated using height and weight recorded in the case report form, patients with BMI ≥ 30 were categorized as obesity. Indications for applying noninvasive ventilation (NIV): pregnant women who complained shortness of breath or blood gas analysis confirmed hypoxemia PaO 2 to FiO 2 < 300. One nonpulmonary major organ dysfunction or unconsciousness was contraindications for NIV. Indications to change from NIV to invasive ventilation: A cautious trial of NIV was attempted and response to NIV was monitored after the first hour or two. If there was a deterioration of oxygenation, invasive ventilation was considered. Definition of successful NIV: PaO 2 to FiO 2 improved and respiratory rate decreased during one or two hour NIV therapy. The patients successfully weaned off NIV and survived. Definition of failed NIV: During the one or two NIV trial, a deterioration of oxygenation was observed and invasive ventilation was needed. Data collection and analysis were coordinated by the MOH. A standard data collection form was used for each study site. Site investigators were primarily infectious disease physicians closely involved in taking care of such patients at their centers. The data were entered in duplicate into a computerized database. Patient confidentiality was maintained by recording only patient date of birth and gender on the data collection form. The research ethics board at Beijing Chao-Yang Hospital and The First Affiliated Hospital, School of Medicine, Zhejiang University approved the study. We analyzed the reported demographic characteristics, underlying conditions, symptoms, treatments, complications, clinical course and maternal and neonatal outcomes. Means (standard deviations, SD) or medians (interquartiles, IQR) were calculated as summaries of continuous variables. For categorical variables, percentages of patients in each category were calculated. We compared clinical characteristics and clinical outcomes by using an ANOVA test, chi-square test, or Fisher's exact test or Wilcoxon rank-sum test as necessary. The primary outcome was in-hospital mortality. We performed univariable logistic analysis to investigate the potential factors on admission that might be associated with the maternal mortality. Factors with statistical significance (p < 0.05) in the univariate analyses were included in the multivariate logistic regression analysis. A p value of less than 0.05 was considered to indicate statistical significance. All analysis was carried out using SPSS for Windows (release 13.0). Clinical description of cohort 3570 severe or critical cases were screened and 394 cases involved pregnant women ( Figure 1 ). Demographic characteristics, underlying conditions, symptoms, and lab findings of the 394 pregnant women are illustrated in diseases were rare in this analysis. None of the patients had been immunized against seasonal influenza or 2009 pH1N1. The median APACHE II score was 7.0 (IQR, [4] [5] [6] [7] [8] [9] [10] [11] . At the time of admission, 351 patients (90.0%) had pneumonia with an abnormal chest radiography or chest computed tomography. The most common symptoms were cough (372; 94.7%) and dyspnoea (199; 50.6%). The median PaO 2 /FiO 2 on admission was 154.7 (IQR, 89.5-320.5) ( Table 1 ). Of the 394 hospitalized patients, 246 (63.7%) were admitted to an ICU at a median of 8 days from onset of illness (IQR 5 to 14; Table 2 ). Medication 378 (95.9%) patients received oseltamivir. The median time from onset of illness to oseltamivir therapy was 5 days (IQR 3 to 7), among them only 52 patients (14.0%) received oseltamivir within 48 h of onset of illness. 387 out of 394 patients received antibiotics. 244 received traditional Chinese medicine. Corticosteroid therapy was administered to 242 patients ( Table 2) . The most commonly reported complication in this study was acute respiratory disease syndrome (ARDS) (151; 53.4%) ( Table 2) . 211 (59.4%) women delivered at a median of 6 days (IQR 3 to 12) after pH1N1 symptom onset. 122 out of 211 women delivered prematurely (Additional file 2: Table S2 ). The most common delivery method was cesarean delivery (172 patients, 82.7%) ( Table 2 ). Among 143 live-birth deliveries for which the gestational age was known, 68 were premature (Additional file 2: Table S2 ). Among the 394 pregnant women in the study, 77 died (Table 2) , 56 out of the 77 patients who died were in their third trimester. The main cause of death was refractory hypoxemia (66 patients, 85.7%). Of 5 patients with secondary infection, three patients had Acinetobacter baumannii, one patient had Aspergillus spp, and one patient had both Acinetobacter baumannii and Aspergillus spp. 62.4% of women included in the study required intensive care and 47.2% required mechanical ventilation. 83 (Table 5 ). The first case of 2009 pH1N1 virus infection in China was documented on May 10, the virus has rapidly spread throughout the mainland. A total of 126,000 confirmed cases were reported by Mar 31, 2010, including 7414 patients severe and 800 patients died. Among all these severe cases, about 13.7% of patients were pregnant women [18] . In this large study of pregnant women who were hospitalized with severe 2009 pH1N1 influenza, the clinical characteristics were similar to those reported by others [3, 4, 17, 19] . 95.6% of patients were infected in the second or third trimester. In our study, the most common comorbidities were cardiovascular diseases (3.3%), diabetes mellitus (1.0%), respiratory diseases (2.8%), and obesity (18.5%). In our study, the prevalence of underlying diseases was much lower than reports from the United States (49.3%) [19] , 56% in Australia [14] , 34% in California [17] , 22 .8% in Brazil [20] , and 62% in France [4] . In those studies, the main cause of underlying disease was asthma. A study compared asthma prevalence of Chinese adolescents living in Canada and in China. The authors found that for girls, the range of asthma was 4.3% in Guangzhou to 9.8% in Canadian-born Chinese adolescents. These results suggest that the lower prevalence of pre-existing asthma in our samples reflects prevalence of the disease in the Chinese population [21] . The mortality rate for severe or critically infected pregnant women in our study was 20%, similar to what was reported in Canada, Mexico, and New Zealand [22] [23] [24] [25] , but higher than in France (8% death in ICUhospitalized pregnancy women) [4] . Risk analysis showed that a PaO 2 /FiO 2 ≤ 200 and higher BMI (i.e. ≥ 30) on admission were risk factors for maternal death. Pregnancy and ARDS are associated with increased oxygen consumption, which can result in hypoxemia in the mothers and the neonates. We reported that a higher BMI was associated with maternal mortality after adjusting for baseline clinical factors. Observations of a high prevalence of obesity in severe and fatal cases of 2009 pH1N1 infection have been reported in Chile, Canada, the United Kingdom and Mexico [10, 26, 27] . As observed in Australia, 42% of patients had a BMI of more than 30 and 22% of patients more than 35, while the corresponding proportions in the general Australian pregnant population was 24% and 10% respectively [28] . However, our research retrospectively collected the patient's clinical information recorded in CRFs. Proportion of obesity has been overestimated based on BMI in the 3 rd trimester of pregnancy. Data from previous pandemics and seasonal influenza epidemics suggested that the risk of complications associated with influenza might be higher in the second and third trimester of pregnancy than in the first trimester [2, 3, 17] . We also observed a higher proportion of maternal death occurring in the second and third trimester. During the 2009 H1N1 influenza pandemic, in the United States, the rate of premature birth (30.2%) was higher than the rate of premature births (13%) reported [29] , consistent with data demonstrating a higher rate of premature delivery during previous pandemics [2] . Among women in our study for whom data on pregnancy outcomes was available, the rate of premature birth was 57.8%. In a multivariable analysis, preterm delivery contributed to fetal mortality. Delivery in severe and critically infected women after 37 weeks' of gestation had improved neonatal outcomes compared to similar patients who delivered before 37 weeks of gestation. Evidence on the useful role of NIV in pregnant patients with ARDS secondary H1N1 viral infection was lacking. Dr. Amit Banga [30] reported a 28-year-old pregnant female with ARDS (PaO 2 /FiO 2 155) due to community-acquired severe pneumonia who successfully treated with NIV. In 2009, Dr. Michel Djibre and collegues [31] reported a 38-year-old pregnant woman at 31 weeks' gestation with PaO 2 /FiO 2 98 who was successfully treated with NIV. In our study, the success rate among pregnant women with H1N1 infection for NIV was 45.8%. A recent prospective multicenter survey also found that when NIV was used as first-line therapy for selected ALI/ARDS patients (those with 2 organ failures, hemodynamic instability, or encephalopathy were excluded), 54% avoided intubation and had excellent outcomes [32] . Apart from previous findings that major organ dysfunction and obtunded sensorium would obviously be unsuitable candidates for NIV, we found that pregnant women complicated by septic shock were less likely to be successfully treated by NIV. Our data also support that cautious selection of appropriate patients is important for successful application of NIV. Patients should be monitored closely for signs of NIV failure until stabilized. If there are signs of NIV failure, patients should be intubated promptly before a crisis develops. Our investigation has several limitations. Firstly, we only evaluated pregnant women admitted to a hospital who fulfilled the diagnostic criteria of severe or critical cases. Secondly, it was an observational study, and could therefore only demonstrate associations and could not infer cause. Thirdly, we lacked follow up visits for maternal and neonatal outcomes. Lastly, despite the use of a standardized data-collection form, not all information was collected for all patients. The clinical data reported herein is consistent with previous studies that demonstrate that pregnant women with influenza are at an increased risk of serious illness and death. Our novel findings included: 1) NIV was useful for some selected pregnant women with pH1N1 virus infection complicated by respiratory failure, but septic shock should be considered a contraindication; 2) a PaO 2 /FiO 2 ≤ 200 and higher BMI (i.e. ≥ 30) on admission were independent risk factors for maternal death; 3) Premature delivery was an independent risk factor for neonatal death. Additional file 1: The diagnosis criteria for severe and critical cases. Additional file 2: Maternal and neonatal outcomes by different delivery methods in different trimesters. Data are presented as no. (%)/total no.(%), if otherwise stated. Percentages are based on patients with complete information in the respective categories. * Two patients missed the detailed information in maternal outcomes. Neonatal outcomes were unknown in four cases. ** One patient missed the detailed information in maternal outcomes. Neonatal outcomes were unknown in two cases.

Membrane Fusion and Cell Entry of XMRV Are pH-Independent and Modulated by the Envelope Glycoprotein's Cytoplasmic Tail

Xenotropic murine leukemia virus-related virus (XMRV) is a gammaretrovirus that was originally identified from human prostate cancer patients and subsequently linked to chronic fatigue syndrome. Recent studies showed that XMRV is a recombinant mouse retrovirus; hence, its association with human diseases has become questionable. Here, we demonstrated that XMRV envelope (Env)-mediated pseudoviral infection is not blocked by lysosomotropic agents and cellular protease inhibitors, suggesting that XMRV entry is not pH-dependent. The full length XMRV Env was unable to induce syncytia formation and cell-cell fusion, even in cells overexpressing the viral receptor, XPR1. However, truncation of the C-terminal 21 or 33 amino acid residues in the cytoplasmic tail (CT) of XMRV Env induced substantial membrane fusion, not only in the permissive 293 cells but also in the nonpermissive CHO cells that lack a functional XPR1 receptor. The increased fusion activities of these truncations correlated with their enhanced SU shedding into culture media, suggesting conformational changes in the ectodomain of XMRV Env. Noticeably, further truncation of the CT of XMRV Env proximal to the membrane-spanning domain severely impaired the Env fusogenicity, as well as dramatically decreased the Env incorporations into MoMLV oncoretroviral and HIV-1 lentiviral vectors resulting in greatly reduced viral transductions. Collectively, our studies reveal that XMRV entry does not require a low pH or low pH-dependent host proteases, and that the cytoplasmic tail of XMRV Env critically modulates membrane fusion and cell entry. Our data also imply that additional cellular factors besides XPR1 are likely to be involved in XMRV entry.

Enveloped viruses must fuse with host cell membranes in order to gain entry and initiate infection. For retroviruses, this process is mediated by the envelope glycoprotein (Env) acquired from the viral producer cells. The Env is initially synthesized as a precursor in the endoplasmic reticulum (ER) and subsequently cleaved by cellular proteases in the trans-Golgi complex into the surface (SU) and transmembrane (TM) subunits [1] . The SU subunit contains a receptor binding domain (RBD) that is responsible for interactions with specific cellular receptors or coreceptors, and the TM subunit possesses a fusion peptide, two heptad repeats (HRs), a membranespanning domain (MSD), and a cytoplasmic tail (CT), all of which have been shown to control or regulate membrane fusion [2] . Upon proper triggering, the TM subunit undergoes a large scale conformational rearrangement, leading to the formation of a stable helix bundle (6-HB) that drives fusion between the viral and cellular membranes [3] . The retroviral Env-mediated fusion is controlled at multiple steps to prevent premature activation [2, 4] . First, the cleavage of retroviral Env precursor into SU and TM is a pre-requisite for fusion as it liberates the fusion peptide located at the amino terminus of TM so that it can insert into the target membrane upon triggering [3] . Second, post-translational modifications, such as glycosylation, are also critical for proper folding and receptor binding of Env thereby influencing membrane fusion and cell entry [5, 6, 7] . In addition, several retroviruses, such as murine leukemia virus (MLV), Mason-Pfizer monkey virus (M-PMV), equine infectious anemia virus (EIAV), etc, contain a ,16 aminoacid stretch in the CT of Env, known as R peptide, that intrinsically restricts membrane fusion [8, 9, 10] . In the latter case, the Env proteins containing the full length CT are not fusogenic in the virus-producer cells, but become fully fusogenic after viral protease cleavage of the R peptide upon budding from host cells [9, 11, 12] . The mechanism underlying the R peptide-mediated control of retroviral Env fusion is still not known. Whereas fusion of most retroviruses is triggered by receptor binding, increasing numbers of retroviruses have been shown to require a low pH, or receptor binding plus low pH, for membrane fusion [13, 14, 15, 16, 17, 18, 19, 20] . It is interesting that infection by ecotropic murine leukemia virus (E-MLV) has been shown to be blocked by inhibitors of cellular cathepsins [21] , suggesting host proteases are involved in the fusion activation of E-MLV and perhaps of other retroviruses. Similar mechanisms have been reported for other enveloped viruses [22, 23, 24, 25, 26] . Xenotropic murine leukemia virus-related virus (XMRV) is a gammaretrovirus that was originally identified from human prostate cancer patients and subsequently linked to chronic fatigue syndrome (CFS) [27, 28] . However, recent studies have shown that this virus is a recombinant mouse retrovirus that was likely generated during the passages of a human prostate tumor in nude mice [29, 30] . Moreover, numerous groups have failed to detect XMRV from human prostate cancer samples as well as CFS patients, making the claim of its association with these human diseases questionable [31, 32] . Regardless, it is still important to understand how the Env protein of XMRV mediates membrane fusion and cell entry from the virology perspective, especially in light of the emerging diverse mechanisms of retroviral Env-mediated fusion activation and cell entry [2] . The Env of XMRV shares significant sequence homology with that of other xenotropic and polytropic MLVs (X/P-MLV), especially in the SU subunit, and these viruses share the same xenotropic and polytropic retrovirus receptor 1 (XPR1) for entry [27, 33, 34, 35, 36] . XMRV has been shown to infect a wide range of cell lines derived from different species including humans, with the notable exception of hamster and mouse cells; overexpression of XPR1 in NIH 3T3 and CHO cells renders these cells susceptible to XMRV infection, indicating that XPR1 is the key cellular receptor for XMRV [37, 38, 39, 40, 41] . In this study, we aimed to understand the mechanisms of membrane fusion and cell entry mediated by the XMRV Env protein, particularly the possible role of its relatively long CT (compared to Mo-MLV) and of the viral receptor, XPR1, in modulating this process. Retroviruses have been historically believed to fuse directly at the plasma membrane of target cells for entry and infection [42] . However, recent studies have shown that some retroviruses, including avian sarcoma leukosis virus (ASLV), mouse mammary tumor virus (MMTV), Jaagsiekte sheep retrovirus (JSRV), enzootic nasal tumor virus (ENTV), foamy virus, EIAV, and ecotropic Moloney MLV (MoMLV) require a low pH or low pH-dependent proteases for cell entry [13, 14, 15, 16, 17, 18, 19, 20, 21] . Here, we produced MoMLV pseudotypes bearing XMRV Env, and investigated the cell entry of XMRV by using classical chemical inhibitors that block pH-dependent viral entry [4] . We first treated human HTX cells (a subclone of HT1080) with a lysosomotropic agent, NH4Cl, and observed that it did not inhibit but rather somewhat enhanced XMRV infection (p.0.05). As expected, the infection of pH-dependent vesicular stomatitis virus (VSV) pseudovirions was dramatically decreased (p,0.01, Fig. 1A ). We next treated cells with a proton-pump inhibitor, Bafilomycin A1 (BafA1), and found interestingly that XMRV infection was again increased (p.0.05), yet that VSV entry was almost completely blocked by BafA1 even at the 5 nM concentration (p,0.01, Fig. 1B ). We noted that entry of 10A1 MLV was also slightly enhanced by BafA1 (p.0.05), but the effect was not dosedependent (Fig. 1B) . Similar effects of NH4Cl and BafA1 on XMRV entry were also observed in 293 and a human prostate cancer cell line, DU145 (data not shown), together supporting the idea that XMRV entry does not require a low pH as do the typical pH-dependent viruses, such as VSV and influenza A [4] . The modest but reproducible enhancement of XMRV infection in the presence of NH4Cl and BafA1 could be explained by a block of viral particle degradation in the endosomes or lysosomes. To investigate this possibility and explore if XMRV entry requires cellular proteases, we performed pseudoviral infection in the presence or absence of leupeptin or cathepsin III inhibitor, both of which are broad spectra, lysosomal protease inhibitors. XMRV infection was enhanced by both protease inhibitors, albeit the increase was not statistically significant (p.0.05); however, infection of Ebola pseudovirions was dramatically impaired (p,0.01, Fig. 1C and 1D ). We noted that VSV infection was also slightly enhanced by these two protease inhibitors but the effect was not dose-dependent. The effect of these protease inhibitors on Ebola infection was consistent with the notion that Ebola GP-mediated membrane fusion with endosome requires cellular cathepsin B and L [22, 26] . Taken together, these results show that XMRV entry does not require a low pH or low pHdependent cellular proteases, and that endocytosis XMRV may occur for XMRV but this would likely result in virions inactivation through pH-dependent host proteases. In order to investigate the role of interactions between XMRV Env and its receptor XPR1 in modulating membrane fusion and cell entry of XMRV, we created a soluble form of XMRV SU fused to the human IgG Fc fragment ( Fig. 2A) . The fusion protein was produced by transient transfection of 293T cells and purified using protein A beads using a procedure we had previously described for the JSRV SU fusion protein [43] . As shown in Figure 2B , incubation of XMRV SU-human IgG fusion protein with the permissive HTX cells resulted in an apparent fluorescence shift relative to that of secondary antibody alone (which served as a negative control), and overexpression of XPR1 receptor in HTX cells substantially increased the XMRV SU binding to the cells, indicating that the binding was specific. Similar results were also obtained in the permissive human 293, DU145, A549, dog MDCK, and monkey Vero cells (data not shown). The specific binding of XMRV SU for XPR1 was further confirmed in CHO/XPR1 cells which were established by transduction using a retroviral vector expressing XPR1; but surprisingly, we reproducibly detected a fluorescent shift in the parental CHO cells (Fig. 2B) , which are known to be nonpermissive for XMRV infection [37, 41] (also see Table 1 below). We next assessed the effects of purified XMRV SU fusion protein on pseudoviral infection in HTX cells. Cells were pre- incubated with different amounts of XMRV SU fusion proteins for 1 h at 4uC, followed by switching the temperature to 37uC to initiate infection in the constant presence of the fusion protein for 6 h. As shown in Figure 2C , the XMRV SU fusion proteins substantially blocked the XMRV pseudoviral infection (p,0.05) in a dose-dependent manner, with the JSRV SU having no apparent effect (p.0.05). As would be expected, the JSRV SU fusion protein specifically blocked the JSRV pseudoviral infection but not that of XMRV (p,0.05) ( Fig. 2C and 2D ). The concentration of soluble XMRV SU required to block 50% of XMRV infection was ,10 ug/ml, which was relatively higher compared that of JSRV SU (,5 mg/ml, which is necessary to block 50% of JSRV infection) ( Fig. 2C and 2D ). Together, these results demonstrate that the soluble XMRV SU fusion protein interacts with the XPR1 receptor on the cell surface and functionally blocks the XMRV pseudoviral infection. Truncation of XMRV Env from the C-terminal cytoplasmic tail (CT) promotes SU shedding and syncytia-forming activity While identical in the N-terminal and central regions, including the conserved R peptide cleavage site between 624 and 625, the Cterminal CT of XMRV Env differs from that of MoMLV Env, with a relatively longer length (Fig. 3A ). Here we sought to determine the membrane fusion property of XMRV Env, particularly the effect of CT truncation on cell fusion. We first created a series of truncation mutants in the CT and examined the Env processing and expression by metabolic labeling. 293T cells were pulse-labeled with [ 35-S] Met-Cys for 1 h and chased for 4 h; the XMRV Env proteins in the cell lysates and their SU shed into the culture media were immunoprecipitated with anti-FLAG beads (FLAG is tagged at the N-terminus of SU). As shown in Figure 3B , all the Env constructs were properly processed and expressed in the transfected cells, except CT635 which consistently showed a decreased level of expression of the processed SU (,30% of wildtype) (note the SU subunits of CT624, CT613, CT609, CT608 and CT606 co-migrated with their full length precursors because of their reduced size of precursor, Fig. 3B , upper panel). Of note, CT624, CT613 and CT609 exhibited enhanced SU shedding into culture media as compared to that of wildtype and other mutants (Fig. 3B , lower panel). We also examined the SU surface expression of these Env constructs in 293T cells by flow cytometry using an anti-FLAG antibody, and observed that CT624 exhibited a wildtype level of expression whereas all the other truncation mutants had reduced SU on the cell surface (,50%) (Fig. 3C) . Altogether, these results demonstrate that truncation of the CT of XMRV Env affects SU shedding and surface expression. We next performed syncytia-forming assay in 293 cells and assessed the membrane fusion properties of XMRV Env and mutants. 293 cells were chosen because they are highly transfectable and also permissive to XMRV infection [41] . The full length XMRV Env was unable to induce syncytia formation, presumably due to the presence of an R peptide in the CT (Fig. 4A ). CT624 and CT613, in which the CT of XMRV Env was truncated at the putative R peptide cleavage site and further towards the N-terminus (Fig. 3A) , respectively, elicited apparent syncytia (typically ,30 syncytia per mg DNA, with .6 nuclei per syncytium) (Fig. 4A) . Interestingly, CT609, which contains the first arginine residue of the CT only, showed a much reduced fusion activity as compared to CT624 and CT613 (,5-10 syncytia per mg DNA, with smaller size) (Fig. 4A ). These results were somewhat different from what had been reported for MoMLV, where an identical mutant largely retained the fusogenicity of R peptide-minus mutant [11, 44] . Noticeably, the increased fusion activity of CT624 and CT613, and to lesser extent of CT609, correlated with the enhanced SU shedding of these mutants in culture media (Fig. 3B) . These results are similar to our previous findings made on JSRV Env, severe truncation of which led to pronounced SU shedding accompanied with greatly increased fusogenicity [19] . Interestingly, we observed that the tailless CT608 and CT606 mutants were virtually fusiondefective, possibly due to their truncation into the MSD and/or reduced surface expression ( Fig. 3A and 3C ). We further treated the individual Env-expressing cells with a low pH buffer (pH 5.0) for 1 min or 5 min (pictures not shown), but did not observe apparent effect on syncytia induction of any of these constructs, supporting the above conclusion that XMRV Env-mediated fusion and cellular entry is pH-independent. The finding that syncytia induction can be observed in cells expressing XMRV Env truncation mutants prompted us to further quantitatively measure their membrane fusion activities using a flow cytometry-based cell-cell fusion assay adapted from our previous studies on JSRV [18, 19, 45] . In this assay, the effector 293T/GFP cells were transfected with Env-encoding plasmids, and the target 293 cells were labeled with a red-fluorescent dye, CMTMR. Consistent with the syncytia formation data, the XMRV Env wildtype and CT635 were unable to induce cellcell fusion as evidenced by no fluorescent dye transfer (,1.8%, similar to the No-Env background), whereas truncation at the putative R-peptide cleavage site or further upstream towards the MSD, i.e., CT624 or CT613, induced apparent cell-cell fusion (4,5%) (p,0.05) ( Fig. 4B and 4C ). Again, CT609 showed a relatively low cell-cell fusion activity (,3%), and the tailless CT608 and CT606 mutants were incapable of inducing fusion with background signals (Fig. 4B and 4C ). We also examined the fusion activities of all constructs in 293 cells overexpressing the XPR1 receptor (293/XPR1), but observed only modest increases for CT624, CT613 and CT609 (,10-20%). The titers of XMRV wildtype and mutants in 293/XPR1 cells were approximately 5fold higher than those in the parental 293 cells, and the overexpression of XPR1 in 293/XPR1 cells was confirmed by flow cytometry using the soluble XMRV SU (data not shown). The differential fusion activities of CT624, CT613 and CT609 could be due to their different levels of Env expression on the cell surface or/and intrinsic fusogenicity. To distinguish these possibilities, we transfected effector 293T cells with different amounts of plasmid DNA encoding individual truncated Envs, and determined their cell-cell fusion activities and SU surface expression in parallel. As shown in Figure 5A , the fusion profiles of CT624 and CT613 were almost identical, as evidenced by their similar slopes (,0.095 and ,0.010, respectively, R 2 = 0.97-0.99). In contrast, CT609 exhibited a slightly decreased slope (,0.068, R 2 = 0.93), implying that its reduced fusogenicity relative to CT624 and CT613 cannot be fully attributable to its low level of surface expression. We further performed cell-cell fusion using different co-culture periods, i.e., 0, 2, 4, and 8 h, and again observed faster fusion kinetics for CT624 and CT613 (Fig. 5B ) as compared to CT609, further confirming that additional truncation of XMRV Env beyond the R peptide cleavage site does not increase fusion activity as we had seen for JSRV Env and that CT609 has an intrinsically relatively low fusogenicity. The expression of XMRV Env on the 293T cell surface was measured using anti-FLAG and flow cytometry. Fluorescence geometric means were normalized to XMRV Env (100%). Shown are the averages of 3 independent experiments 6 S.D. XMRV Env: XMRV Env tagged with a FLAG sequence only at the N-terminus. All truncations were also tagged similarly with an N-terminal FLAG. F-XMRV-F: an XMRV Env construct that is tagged by FLAG sequences on both N-and C-termini. Mock: untransfected 293T cells. doi:10.1371/journal.pone.0033734.g003 To examine the possibility that the lack or reduced fusion for CT609, CT608 and CT606 might be due to a block at hemifusion, we treated co-cultured target and effector cells with chlorpromazine (CPZ, 0.2-0.5 mM), a membrane permeable reagent that promotes the transition from hemifusion to full fusion [46] , but observed no apparent increase in fusion for any of these Env constructs (data not shown). These results suggest that the fusion suppression in these XMRV Env constructs unlikely takes place at the hemifusion step. Overall we conclude that, distinct from JSRV Env, severe truncation of the CT of XMRV Env towards the MSD does not further enhance but rather impairs the Env fusogenicity. The reason for the decreased fusogenicity of CT609, CT608 and CT606 remains unclear, but is likely related to reduced surface expression or/and altered Env conformation (see Discussion). We next determined the role of XPR1 in XMRV Env-induced membrane fusion by using nonpermissive CHO cells and CHO cells expressing human XPR1 (CHO/XPR1). The CHO/XPR1 cell line was established by transducing CHO cells with a LXSN retroviral vector encoding XPR1 [33] . Expression of XPR1 in the CHO/XPR1 cell line was demonstrated by the specific binding of soluble XMRV SU fusion protein to those cells as shown in Figure 2B , and was further confirmed by immunostaining using an anti-XPR1 antibody (Fig. 6A) . The titers of XMRV wildtype and CT truncation mutants in these two cell lines are shown in Table 1 . CHO cells were apparently not susceptible to XMRV Env pseudoviral infection (Table 1) , consistent with previous reports from other groups [38, 39, 40, 47] . Overexpression of XPR1 in CHO cells resulted in a titer of 10 4 IU/ml for the wildtype and somewhat reduced titers for the CT truncation mutants (Table 1) . Overall, these data support the notion that XPR1 is a critical cellular receptor for XMRV. The cell-cell fusion activities of XMRV Env and mutants in CHO and CHO/XPR1-expressing cells were then examined. For this purpose, we labeled CHO or CHO/XPR1 cells with CMTMR, and co-cultured them with the effector 293T/GFP cells expressing XMRV Env or truncation mutants plus GFP. We observed that, surprisingly, CT624 and CT613 reproducibly induced a detectable level of cell-cell fusion activity in the nonpermissive CHO cells (p,0.05) (Fig. 6B) , despite the fact that this cell line is nonpermissive for XMRV infection (Table 1) . Interestingly, overexpression of human XPR1 in CHO cells only slightly increased the fusion activities of XMRV Env CT mutants, CT624 and CT613 (Fig. 6B) , despite their significantly increased pseudoviral titers (Table 1) . These results, along with the data using 293/XPR1 cells described above, imply that XPR1 may not be the sole trigger for XMRV Env-mediated membrane fusion and cell entry. The CT of retrovirus Env plays various roles in the replication cycle, including entry and assembly; this has been mostly studied in HIV-1 [48] . Here, we wished to determine the ability of XMRV Env and CT truncation mutants to pseudotype the MoMLV retroviral and HIV-1 lentiviral vectors as well as its relationship to membrane fusion and cell entry. As shown in Table 2 , all the XMRV Env constructs (tagged with a FLAG sequence at the N- terminus) were able to pseudotype both types of vectors but with distinct efficiencies. The full length XMRV Env exhibited approximately 2610 4 infectious units per ml for both vectors ( Table 2 , and data not shown), similar to a recent report [49] . The titers of MoMLV retroviral pseudotypes harbouring CT635 or CT624 were slightly reduced as compared to that of the wildtype Env (,4-6 fold), whereas the other more severely truncated mutants exhibited a 2,3-log decrease in the infectious titer ( Table 2) . Similar patterns were also observed for the HIV-1 lentiviral pseudotypes (Table 2 ), but interestingly we found that CT635 consistently exhibited pronounced reductions in the lentiviral pseudoviral titers (,100 fold) as compared that of MoMLV retroviral pseudotypes (,6-fold). The generally reduced viral titers for the CT truncations cannot be fully explained by the enhanced SU shedding, at least for some of these mutants, but appeared to correlate with the differential levels of SU surface expression (Fig. 2C) . We also examined the incorporation efficiencies of these Envs into the MLV pseudovirions by Western blot using concentrated pseudoviral particles, and detected similar levels of SU for CT624, CT613, CT609 and the wildtype Env, as compared to CT635, CT608 and CT606 for which the SU incorporation efficiency was greatly reduced (Fig. 7) . We have attempted to detect the XMRV TM in viral producer cells and the viral particles using an antibody against the MoMLV TM but without success (a gift from Marc Johnson, data not shown). Nevertheless, the Env incorporation data based on the SU (Fig. 7 ) and the XMRV pseudotype titers shown in Table 2 correlated with the SU expression profiles shown in Figure 3C . We noticed that the titers of pseudoviral infection for CT624 and CT613 did not correlate with their enhanced Env fusogenicity based on the syncytia formation and cell-cell fusion assays, and this was particularly the case for CT613, which showed a strongly enhanced fusogenicity (Figs. 4 and 5) but a much reduced titer relative to the wildtype Env (Tables 1 and 2 ). Retroviruses use distinct mechanisms for membrane fusion and cell entry, the mechanisms of which are still poorly defined. In this report, we provided evidence that XMRV entry does not require a low pH or pH-dependent host proteases, but uses a mechanism that is similar to that of typical pH-independent viruses. Interestingly, we find that XMRV entry is enhanced by NH4Cl and BafA1, the two most commonly used agents that neutralize acidic endosomal environments, as well as by leupeptin and cathepsin inhibitor III, which broadly inhibit the lysosomal protease activities. Together, these observations suggest that endocytosis may occur in non-productive entry of XMRV, leading to viral particle degradation. Consistent with this notion, we did not observe specific block of XMRV entry by Dynasore or a dominant negative mutant of Dynamin (K44A) in 293T cells (data not shown). Previous studies have shown that different endocytic pathways mediate entry of some pH-independent retroviruses, including amphotropic and ecotropic MLV as well as HIV-1 [21, 50, 51] , however the exact mechanisms and the underlying significance remain largely unknown. It should be added that, while we have not observed any inhibitory effects of leupeptin and cathepsin III inhibitor on XMRV infection in HTX and 293 cells, we cannot rule out the possibility that cellular proteases may be involved in the XMRV entry of other cell types. In this sense, it is interesting to note that endocytosis and cathepsins were recently shown to affect the entry of several gammaretroviruses, including XMRV, in human TE671 and rat XC cells [52] . Additional studies are warranted to clarify this issue and further characterize the entry pathway of XMRV, perhaps with assistance of the recently developed single molecule labeling and confocal imaging technique. One important objective of this study was to understand the possible roles of the CT of XMRV Env in modulating membrane fusion and cell entry. We showed here that CT624 and CT613, which are truncated at or beyond the putative R peptide cleavage site of the XMRV Env (Fig. 3A) , induced apparent syncytia formation and cell-cell fusion in permissive 293 cells (Figs. 4 and 5), presumably due to the removal of the putative R peptide. Surprisingly, we observed apparent cell-cell fusion of CT624 and CT613 also in CHO cells (Fig. 6B) , which are known to be nonpermissive for XMRV infection, including these two truncation mutants (Table 1) . These results suggest two possibilities: first, CHO cells may express a low but functional level of XPR1 that permits cell-cell fusion of XMRV Env mutants and that the resistance of CHO cells to XMRV infection may be due to a block at the post-fusion steps. This possibility is supported in part by our observation that a soluble form of XMRV SU fusion protein reproducibly binds CHO cells relative to the negative control using secondary antibody alone (Fig. 2) . These binding results also argue against the possibility that potential N-linked glycosylation of XPR1 in CHO accounts for its resistance to XMRV infection, a situation that has been previously shown to be the case for several retroviruses [53, 54] . Second, the XPR1-mediated binding may The MLV titers were determined as described in Table 1 not be the sole trigger for XMRV Env-mediated membrane fusion. This scenario is in line with our finding that overexpression of XPR1 in 293 and CHO cells did not significantly increase the cell-cell fusion activities of XMRV Env truncation mutants despite their increased infection in these cells (Fig. 6B and Table 1 ; data not shown). We also considered the possibility that the XMRV Env truncation mutants may have acquired a receptor-independent, spontaneous cell-cell fusion or are pre-activated in 293T/ GFP cells due to their reduced kinetic barrier required for membrane fusion; however, the lack of infection in the CHO cells for the truncation mutants did not support this hypothesis ( Table 2) . Taken together, we favour the notion that, while XPR1 is a critical receptor for XMRV and is required for membrane fusion and cell entry, other cellular factors as yet to be identified are likely to be involved in cell entry and membrane fusion of XMRV. Consistent with this idea, it has been recently reported that XMRV does not infect BHK cells even when XPR1 is overexpressed in this cell line [47] , and that XMRV can infect A549 cells even though this cell line does not express a functional XPR1 receptor [55] . Hence, identification of additional cellular factors involved in XMRV entry would help to better understand the mechanisms of membrane fusion and cell entry mediated by XMRV Env. Previous studies from HIV and other simple retroviruses have suggested that the enhanced fusion activities of some retroviral Env truncations in the CT may be due to increased steady-state levels of Env expression on the cell surface [56, 57, 58, 59] . However, here we have found little evidence that suggests that this might be the case for XMRV (Fig. 3C ) -despite the highly conserved endocytosis motif, YXXh (Y = tyrosine, X = any amino acid, h = residue with hydrophobic side chain) present in the CT of XMRV Env (Fig. 3A) . Another commonly assumed mechanism is that truncation of the retroviral Env CT can somehow alter the conformation of Env ectodomain, resulting in a reduced association between SU and TM thereby promoting membrane fusion [60] . Indeed, we observed that all three truncation mutants with enhanced fusogenicities, i.e., CT624, CT613 and CT609, exhibited increased levels of SU shedding, which was in sharp contrast to that of wildtype Env and other mutants (CT635, CT608 and CT606) having minimal cell-cell fusion activity (Fig. 3B, Figs. 4 and 5) . Future studies will focus on how the CT of XMRV Env structurally modulates the Env fusion activation. Another surprising finding of this study is that CT609, which harbours the single arginine residue in the CT possesses a reduced fusogenicity relative to that of CT624 and CT613, which cannot be solely explained by its reduced surface expression (Figs. 4, 5, and 6). This observation is clearly different from what we had seen for JSRV Env [19] and is also somewhat different from some though not all of the previous studies on MoMLV [11, 44, 61] . Importantly, the tailless mutants, CT608 and CT606, are virtually fusion-defective (Fig. 4) , collectively leading us to propose that the N-terminal CT proximal to the MSD of XMRV Env is critical for Env-mediated membrane fusion. One possible mechanism is that residues in this region, including the highly conserved arginine present in many transmembrane proteins including the retroviral Envs, may interact with cell membrane and thus modulate lipid movement during the membrane fusion process [62, 63] . Despite enhanced fusogenicity of CT624 and CT613, we found that the infection efficiency of their pseudovirions were rather low compared to those of wildtype Env (Tables 1 and 2 ). One plausible explanation is that the incorporations of these truncations into the MoMLV and HIV-1 vectors might be reduced as compared to those of wildtype; however, based on our immunoblot analysis using an anti-FLAG antibody to detect the XMRV SU, we found no evidence to support this scenario (Fig. 7) . Alternatively, the reduced pseudoviral titers for the truncation mutants might result from their altered ability to bind to the viral receptor, XPR1. While we do not have direct evidence in support of this possibility, the apparent SU shedding induced by the CT truncations in the Env-expressing cells (Fig. 3B ) strongly suggests that conformational changes likely occurs in the ectodomain of the truncated Env, including their SU subunits. Indeed, prior studies from HIV and other retroviruses have demonstrated that CT truncation of retroviral Env can alter the Env receptor binding capability thus affect viral infection [60, 64] . In this regard, it would be interesting to explore the role of the CT of XMRV Env in the infectious virus system. The HTX (a subclone of HT1080), 293T, 293, 293T/GFP (293T cell line stably expressing GFP) and 293/GP-LAPSN (293 cells stably expressing MoMLV Gag-Pol and alkaline phosphatase or AP) cell lines have been previously described [19, 65] . The 293/ XPR1, HTX/XPR1 and CHO/XPR1 cell lines were generated by transducing the 293, HTX or CHO cells using a retroviral vector, LXSN, encoding the XPR1 receptor (LhXPR1SN, kind gift of Dusty Miller) [33] and bearing VSV-G. Infected cells were selected using G418 (Invitrogen, Carlsbad, CA) for ,10 days. All cell lines were cultured in DMEM (Invitrogen) supplemented with 10% fetal bovine serum (FBS) at 37uC at 10% CO 2 -air atmosphere at 100% relative humidity. The anti-FLAG monoclonal antibody, the EZview Red anti-FLAG affinity gel, and anti-mouse immunoglobulin G (IgG) coupled to phycoerythrin (PE) were purchased from Sigma (St. XMRV Env was initially engineered to contain a FLAG tag at both N-and C-termini by using pcDNA3.1-VP62 (gift of Robert Silverman) [37] as a template for PCR, and cloned in a pCIneo expression vector (Promega, Madison, WI), the resulting construct was referred to as pCIneo-F-Xenv-F. To create the N-terminal FLAG-tagged XMRV Env wildtype and CT truncations, the pCIneo-F-Xenv-F construct was used as a template, with the following lower primers being used for PCR amplification (Not I sites are underlined): XMRV Env, 59-ATCGGCGGCCGCT-CATTCACGTGATTCCACTTC-39; CT635, 59-TTCTGCG-GCCGCTCATGATTTGAGTTGGTGATA-39; CT624, 59-C-TGTGCGGCCGCTCACAGGGCCTGCACTACCGA-39; CT-613, 59-AATTGCGGCCGCTCAAAACTGGACCAAGCGGT-TG-39; CT609, 59-TACAGCGGCCGCTCAGCGGTTGAGA-ATACAGGGTCCGA-39; CT608, 59-AAACGCGGCCGCT-CAGTTGAGAATACAGGGTCCGA-39; CT606, 59-GACC-GCGGCCGCTCAAATACAGGGTCCGAAGA-39. The pCI-neo-10A1, pMD.G and pCIneo-Ebola GP expression vectors have been previously described [18] . The soluble XMRV SU construct was generated by overlapping PCR using pcDNA3.1-VP62 [37] and the previously described pCSI-JSU (for JSRV SU fusion protein) [43] as templates. The first fragment containing XMRV SU was amplified using the following primers: upstream primer (Not I underlined), 59-GCATGCGGCCGCATGGAAAGTCCAGCGTTCTC-39; downstream primer, 59-CCTAGGCCTGTCGACGCCTTTTCA-AACTGGCC-39. The second fragment containing human IgG Fc was generated using the following primers: upstream primer, 59-GGCCAGTTTGAAAAGCTGTCGACAGGCCTAGG -39; downstream primer, 59-TGTATCTTATCATGTCTGGATC-CCC-39. The XMRV SU fused to the human IgG Fc was generated using the two fragments as templates, and the upstream and downstream primer of the first and second fragment, respectively, and then the PCR product was cloned into the pCSI vector. The MoMLV retroviral pseudotypes encoding the alkaline phosphatase (AP) were produced by transfection of 293/GP-LAPSN cells with plasmid DNA encoding individual XMRV Env, CT truncations, or JSRV Env. The MoMLV retroviral pseudotypes encoding the green fluorescent protein (GFP) were generated by co-transfection of 293T cells with pCMV-gag-pol-MLV, pCMV-GFP-MLV (both vectors are kind gifts of François-Loïc Cosset) and plasmids encoding XMRV Env, XMRV Env CT truncations, Ebola GP (pCIneo-Ebola GP) [18] , VSV-G (pMD.G), or MLV 10A1 Env (pCIneo-10A1) [18] . The HIV-1 lentiviral pseudotypes encoding AP were produced by co-transfecting 293T cells with pCMV-HIVD8.2, pHR'CMVAP [66] and plasmid DNA encoding individual Envs. All pseudotypes were harvested 48 and 72 h post-transfection and cell debris were removed by centrifugation at 2,5006 g. MLV pseudovirions were purified by ultracentrifugation on a 20% sucrose cushion for 2 h at 185,0006 g and 4uC, and Western blot was performed to examine SU incorporation using an anti-FLAG antibody. All viral infections were carried out in the presence of 5 mg/ml polybrene (Sigma) and viral titers were determined by AP staining or flow cytometry analysis to measure GFP + cells 48-72 h post-infection. For infection in the presence of drugs or soluble XMRV SU or JSRV SU, cells were first pre-treated with the indicated concentrations of drugs at 37uC or the soluble proteins at 4uC for 1 h, and then incubated with retroviral pseudotypes for 6 h in the presence of drugs or fusion proteins before inactivation using citrate buffer (40 mM sodium citrate, 10 mM KCl, 135 mM NaCl, pH 3.15). The syncytium induction assay was performed as described previously with some modifications [6, 19] . 293 cells were cotransfected with plasmids encoding XMRV Env or CT truncation mutants plus a GFP-encoding plasmid in order to monitor the transfection efficiency and syncytia formation. Syncytia formation was typically observed and photographed 24 h post-transfection. Where applicable, cells were treated for 5 minutes at 37uC with pre-warmed pH 5.0 buffer (phosphate-buffered saline (PBS), 10 mM MES, 10 m M HEPES) or 0.2-0.5 mM CPZ for 1 min and incubated in normal growth media at 37uC for 1 h. The cell-cell fusion assay was performed as described previously [19, 45] . Briefly, effector 293T/GFP cells were transfected with plasmid DNA encoding XMRV Env or CT truncation mutants using Lipofectamine 2000 (Invitrogen). Twenty-four hours later, cells were washed with PBS and detached using PBS containing 5 mM EDTA. Target 293, 293/XPR1, CHO, or CHO/XPR1 cells were detached using PBS-5 mM EDTA and labeled with 3.5 mM CMTMR in serum-free media for 30 min at 37uC, washed, incubated for an additional 30 min at 37uC in fresh media and washed 3 times with media. Effector cells and target cells were cocultured on 24-well plates for the indicated time periods. Cell-cell fusion was measured by flow cytometry using FACSCalibur (BD Bioscience, Missauga, Canada). The surface expression of XMRV Env in the 293T/GFP cells was measured by flow cytometry using the anti-FLAG antibody and anti-mouse IgG coupled to PE. Production of XMRV SU fusion protein and its binding to cells Soluble XMRV SU and JSRV SU fusion proteins were produced as described previously [43] . 293T cells were transfected using the calcium-phosphate method with plasmids encoding the different SU. Twelve hours post-transfection, media were replaced with DMEM supplemented with 2% ultra-low IgG FBS (Invitrogen). The proteins in the media were purified using protein A beads (GE Healthcare, Uppsala, Sweden) and analyzed by SDS-PAGE and Sypro Ruby staining (Bio-Rad, Hercules, CA). XMRV SU binding assays were performed as described previously [6, 43] . Cells were incubated with 2-10 mg of soluble XMRV SU-IgG fusion protein for 3-4 h on ice, washed 3 times with PBS-2% FBS and incubated with anti-human IgG coupled to FITC for detection. Fluorescence was measured by flow cytometry using FACSCalibur (BD Bioscience, Missauga, Canada). Immunostaining CHO or CHO/XPR1 cells were fixed using 4% paraformaldehyde in PBS, permeabilized using 0.5% Triton X-100 and stained using anti-XPR1 and anti-rabbit IgG coupled to FITC. Before mounting the slides, cells were counterstained with the nuclear stain DAPI. Pictures were taken using a fluorescence microscope (Carl Zeiss, Goettingen, Germany) and images were processed using the ImageJ software (U.S., National Institutes of Health). Metabolic labeling was performed as previously described [19, 45] . Briefly, 293T cells were transfected using the calciumphosphate method with plasmid DNA encoding individual Env. Twenty-four hours later, cells were starved in cysteine and methionine-free DMEM for 30 minutes, pulse-labeled with 62.5 mCi 35 S-cysteine and -methionine for 1 h at 37uC, washed with fresh media and chased for 4 h at 37uC in complete growth medium. Media were then collected and cells washed and lysed (50 mM Tris pH 8.0, 150 mM NaCl, 0.4 mM EDTA, 1% Triton X-100, 0.1% NP-40, 10 mg/ml aprotinin (Sigma), 10 mg/ml leupeptin (Sigma) and 1 mM phenylmethylsulfonyl fluoride (Sigma). The XMRV Env proteins in media and in cell lysates were immunoprecipitated using anti-FLAG beads and resolved by SDS-PAGE. Dried gels were autoradiographed and band intensities of XMRV SU in the cultured media were quantified using the Quantity One software (Bio-Rad, Hercules, CA).

The Impact of Weather on Influenza and Pneumonia Mortality in New York City, 1975–2002: A Retrospective Study

The substantial winter influenza peak in temperate climates has lead to the hypothesis that cold and/or dry air is a causal factor in influenza variability. We examined the relationship between cold and/or dry air and daily influenza and pneumonia mortality in the cold season in the New York metropolitan area from 1975–2002. We conducted a retrospective study relating daily pneumonia and influenza mortality for New York City and surroundings from 1975–2002 to daily air temperature, dew point temperature (a measure of atmospheric humidity), and daily air mass type. We identified high mortality days and periods and employed temporal smoothers and lags to account for the latency period and the time between infection and death. Unpaired t-tests were used to compare high mortality events to non-events and nonparametric bootstrapped regression analysis was used to examine the characteristics of longer mortality episodes. We found a statistically significant (p = 0.003) association between periods of low dew point temperature and above normal pneumonia and influenza mortality 17 days later. The duration (r = −0.61) and severity (r = −0.56) of high mortality episodes was inversely correlated with morning dew point temperature prior to and during the episodes. Weeks in which moist polar air masses were common (air masses characterized by low dew point temperatures) were likewise followed by above normal mortality 17 days later (p = 0.019). This research supports the contention that cold, dry air may be related to influenza mortality and suggests that warning systems could provide enough lead time to be effective in mitigating the effects.

It is well known that intra-annual mortality exhibits a pronounced winter peak in locations with seasonal climates [1] , [2] . In the United States, mortality arising from respiratory disease is 50% higher in winter than in summer [1] . One contributor to this excess winter mortality is influenza [3] , the timing and severity of which can vary significantly from year to year. The association between the annual influenza peak and winter weather in temperate locations has lead to the hypothesis that weather variability could influence influenza mortality variations [4] , [5] . Recent research examining climatic influences on influenza transmission has galvanized interest in this topic. Airborne transmission of Influenza A/Panama virus between guinea pigs was more likely at low temperatures and relative humidities [4] . Because temperature is physically/mathematically linked to relative humidity and thus confounds interpretation, a mass-based humidity measure (such as specific humidity or vapor pressure) provides a stronger relationship to influenza transmission [5] . Given the cold-season decline in specific humidity in temperate climates, a retrospective study [6] showed an association between year-to-year humidity variations and the timing of the influenza seasonal onset in U.S. states. This differs from tropical locations which generally lack a seasonal influenza peak [7] . One theory is that tropical climates are dominated by direct contact transmission which, unlike airborne transmission, is not influenced by air temperature and humidity [7] . These recent studies [4] , [5] , [6] , [7] have motivated several reviews on climate and influenza seasonality [8] . [9] , [10] . These reviews, which approach the issue from atmospheric sciences, virological, and epidemiological perspectives, do not reach firm conclusions on the causes of influenza seasonality but suggest that those causes are complex and multifactorial and that the solution will require interdisciplinary cooperation. A variety of theories exist as to how weather and climate might exert some influence on influenza seasonality. Low temperatures enhance viral stability [4] , reduce mucosal blood flow [11] , and/or diminish mucociliary clearance [12] . Correlations exist between the number of upper respiratory infections and cold-air outbreaks [13] . It is commonly assumed that winter indoor crowding enhances influenza virus transmission, though direct evidence is lacking [8] . Low humidity conditions, which are often but not always accompanied by low temperatures, enhance survival times of viral aerosols [14] , [15] , [16] . If correct, these relationships suggest that indoor winter heating without humidification could enhance influenza transmission [8] , as indoor absolute humidity tends to be correlated with outdoor values and thus is typically lower in winter [6] . From a micro-physical perspective, there is evidence that both droplet size and transmission mode depend on ambient environmental factors [17] , [18] . Small droplets can remain airborne longer whereas large drops tend to precipitate, suggesting that ambient conditions would influence whether airborne-or contactmode transmission predominates. The virulence of the influenza virus changes constantly as the virus undergoes antigenic shift and antigenic drift [8] . Although virulence depends upon a variety of factors, one possible atmospheric influence is that cold, dry air allows the lipid envelope encasing the influenza virus to remain intact longer, increasing the likelihood of infection [19] . Theories proposing that factors other than weather/climate are responsible for influenza seasonality include cycles in viral interference [20] and intrinsic temporal viral dynamics that operate independent of external forcing factors [21] . Some research suggests that the seasonality is driven by the school calendar in which students reconvene after a summer break, but this timing is not consistent with the typical onset of influenza in early winter [22] . We examine the hypothesis that cold and/or dry weather enhances human pneumonia and influenza (P&I) mortality through a retrospective study of daily mortality in New York City and environs from 1975-2002. We hypothesize that periods with colder and/or less humid conditions exhibit excess P&I mortality for a period of time following those climatic conditions. Our research differs from recent work on this topic that examined the timing of the influenza season onset over large geographic areas [6] -our focus is on the influence of daily weather on influenza characteristics for a single, large metropolitan area. We selected New York City for this study for several reasons. Our study examines daily mortality, and statistical robustness is enhanced when the daily sample size is sufficiently large. Weather obviously has a high spatial variability, so it is important that the observed weather be representative of the environmental conditions likely experienced by the decedents. In addition, New York City's mid-latitude location provides a high degree of both interannual and intra-annual variability in weather and climate, so this variability provides a wider range of sample conditions. Thus, New York City has both a large enough population to provide a consistent daily mortality signal while the population density is high enough that the weather observed at a single station is sufficiently representative of conditions experienced throughout the metropolitan area. We conducted a retrospective cohort study of pneumonia and influenza (P&I) mortality of residents of the New York City metropolitan area. Daily frequencies of P&I mortality were tallied from National Center for Health Statistics archives for the New York City Consolidated Metropolitan Statistical Area (which, as defined in the year 2000, includes 30 counties in New York, New Jersey, Connecticut, and Pennsylvania). This period of record spans three revisions of the International Classification of Diseases (ICD) codes (Table 1 ). In the National Center for Health Statistics mortality files that we used for this research, all information that could allow an individual to be identified has been removed. This research utilized only mortality counts for a large metropolitan area. These de-identified counts are stored in governmental archives for the purposes of retrospective research; because all personal identifying information is redacted, consent is not required. Thus, this research is exempt from IRB review under the auspices of Title 45 Part 46 exemption category 4. Pneumonia or influenza must be listed as the primary cause of death to be included in this analysis. These diseases are commonly combined as an endpoint because of specific challenges associated with influenza. First, the number of deaths attributable to influenza is difficult to estimate directly because of a lack of virologically-confirmed infections. Second, many influenza-associated deaths occur from secondary complications when influenza viruses are no longer detectable by laboratory means [6] , [23] , [24] , [25] . Third, the use of P&I mortality in the study of influenza also reflects the problems associated with other measures of prevalence of influenza. Reporting of cases of influenza through routine channels is unsatisfactory because mild influenza may be under-diagnosed and the use of laboratory confirmation is skewed by the impact of variable testing patterns based on prevalence of disease [26] . Finally, multiple studies have shown that there exists a relationship between influenza morbidity and P&I mortality that can be mathematically described and used in epidemiological studies [3] , [27] , [28] , [29] , [30] , [31] . Daily deaths were aggregated into ten age groups (0-4, 5-14, 15-24, 25-34, 35-44, 45-54, 55-64, 65-74, 75-84 and .84) and standardized via direct standardization [32] based upon the age distribution of population of the United States in the year 2000. This procedure adjusts for temporal changes in age demographics over the period of record based on data for each county using U.S. Census archives, thereby allowing for consistent comparisons of mortality rates over time. After age standardization, the long-term mean for each day (within ICD period) was subtracted from that day's standardized P&I mortality to remove the inherent seasonal signal from the time series. Because our goal was to determine if cold and/or dry days or periods within a year were related to P&I mortality peaks, this level of granularity requires the examination of daily data with the seasonal signal removed. Examination of the P&I mortality time series exhibits obvious temporal discontinuities that are exactly coincident with the dates of ICD revision code changes. We removed this artifact by converting each day's P&I mortality to a z-score by dividing the mean departure by the standard deviation separately for each of the three relevant ICD periods ( Table 1 ). The time series of agestandardized mortality prior to and after deseasoning and z-score adjustment is shown in Figure 1a and Figure 1b , respectively. Because of the general lack of influenza in the summer months, June, July, and August were removed from the analysis as they had the potential to distort any relationships during the primary influenza season. Mortality data were smoothed using a 17-day leading moving average (e.g., mortality on January 1 is the mean from January 1-17). This smoother was selected after testing a variety of filter lengths and based upon prior research [6] . We examined both daily P&I mortality ''events'' and longer mortality ''episodes.'' ''Events'' are days with (smoothed) mortality at least one standard deviation above the long-term (smoothed) mean for that date. This z$1 criterion was chosen because the frequency distribution of smoothed mortality is positively skewed with the tail beginning at approximately one standard deviation. After smoothing, there is an obvious tendency for high mortality events to cluster into prolonged periods when the z$1 threshold is (Figure 2 ). We thus identified 12 P&I mortality episodes over the 28-year period of record (Table 2) . To summarize the outcome data treatment, after age standardization the daily mortality data were deseasoned to remove the large influence of season on respiratory infection and converted to z-scores to adjust for discontinuities related to ICD coding. These data were then smoothed using a 17-day leading smoother to account for the inherent lag between infection and mortality. Days or periods with z-scores.1 were identified as mortality ''events'' or ''episodes,'' respectively ( Figure S1 ). Hourly climate data from La Guardia Airport, New York, were retrieved from National Climatic Data Center archives. We utilize 1200 and 1900 Universal Time Coordinate (UTC) air temperature (T) and dew point temperature (T d ) to approximate the typical times of the warmest and coldest hours of the day (7 or 8 a.m. and 2 or 3 p.m. local time). The dew point temperature is the temperature at which water vapor begins to condense via cooling at constant pressure. We use dew point as a measure of the amount of moisture in the air because, unlike relative humidity, it is independent of air temperature [5] . In addition to dew point temperature, we employ an air mass classification, which has the advantage of incorporating a variety of weather variables into a single nominal variable. Specifically, we utilize the Spatial Synoptic Climatology (SSC) [33] which classifies each day's weather into one of six air mass types [dry moderate (DM), dry polar (DP), dry tropical (DT), moist moderate (MM), moist polar (MP), moist tropical (MT)]. A seventh transition (TR) category identifies days characterized by a significant change in weather-typically a frontal passage [34] . The SSC uses air temperature, dew point temperature, wind, surface air pressure, and total cloud cover observations taken four times per day as input to the classification. Thus, the SSC is a multivariate nominal classification of daily weather conditions. The approach has been utilized in a variety of human health applications [35] , [36] , [37] , [38] , [39] , [40] , [41] . The temperature and dew point time series were converted to zscores to remove seasonality and then smoothed using a centered 5-day moving average filter. This filter length was employed after examining various options because it represents a balance between Figure 1b smoothed with a 17-day centered moving average filter. A centered smoother is used here to more clearly present the peak times of the mortality episodes (see Table 2 ). doi:10.1371/journal.pone.0034091.g002 high frequency weather events and more long-term (monthly to seasonal) trends. The seasonality of the SSC air mass types was removed by comparing the presence (coded 1) or absence (coded 0) of each air mass type on each day of the year to the long-term average frequency. For example, if Moist Moderate air was present on average 30% of the time on January 1 over the period of record, then its occurrence on January 1, 2000 would result in a value of +0.7 for that day. These daily anomalies were then converted into continuous variables using a centered 7-day moving average filter for each SSC category. In summary, raw dew point observations were first de-seasoned by conversion to z-scores and then smoothed using a 5-day filter. Daily air mass frequencies were converted from a nominal to a continuous variable by first adjusting for the long-term frequency on each day and then smoothing those frequencies using a 7-day moving average ( Figure S2 ). Daily Analyses. A series of t-tests were employed to determine if temperature, dew point, and air mass frequency differed between high P&I mortality events (z$1) and non-events. To address the temporal autocorrelation in the weather variable time series and the resulting overestimate of the true degrees of freedom, the sample size was adjusted based upon the lag one temporal autocorrelation [42] (Wilks 2006) as follows: where N = number of observations N9 = adjusted degrees of freedom P = lag one temporal autocorrelation. N9 was then adjusted again based on the length of the smoother employed to determine the final effective sample size. For these and all other tests, a Type I error rate of 0.05 was employed and Levene's test for equality of variances was used to determine if pooling of the samples was required. The following tests were performed: 1) smoothed temperature, dew point, and air mass frequency, lagged 17-days, between mortality events (z$1) and nonevents (z,1) using an unpaired two-sample t-test ( Figure S3 ); 2) same as in 1 for unsmoothed temperature and dew point temperature (to determine if a strict 17-day lag exists); and 3) same as in 1 but using a one-sample t-test (to account for the possible influence of disparate sample sizes between groups). Episodic Analyses. For each high P&I mortality episode, we calculated the duration (in days), the summed total mortality over the entire episode, and the average daily episode mortality ( Table 2 ). These quantities served as dependent variables in a linear regression analysis vs. the independent (weather) variables (1200 and 1900 UTC T and T d and the deseasoned SSC frequencies). We used a 17-day lag in which the mean for each variable was calculated across the days in the episode and the preceding 17 days. This lag was selected based upon prior research that examined absolute humidity [6] and the results of other studies [28] , [43] . Given the relatively small number of episodes, we used bootstrapped regression analysis to generate a robust estimate of the regression coefficients. Based on the initial full sample, data sets of the same size were generated by randomly sampling variable pairs, with replacement, and estimating the regression parameters from that sample using ordinary least-squares. This procedure was repeated 10,000 times and the resulting suite of regression coefficients was examined to determine if the 2.5 percentile and 97.5 percentile observations were of the same sign. If so, the regression slope was deemed to be statistically significant [44] . In the daily analysis, 1200 UTC dew point temperature was significantly lower for events than for non-events (p = 0.003), a result that is consistent with the hypothesis that drier conditions are related to enhanced P&I mortality. However, 1900 UTC dew point was higher during events (p = 0.036), a result that contradicts the underlying hypothesis. Table 3 . Results of t-tests comparing weather variables between mortality event days to non-event days. When this test was repeated without smoothing the weather variables, only 1200 UTC dew point was significant (p = 0.028; Table 3 ). The lack of a relationship for 1900 UTC dew point (p = 0.181) suggests that the finding of high afternoon dew points during mortality events using smoothed weather data was not robust. Because the high number of non-event days vs. event days can bias the t-test, an additional test was performed comparing the 5day smoothed weather variables during events to the long-term mean. Here, 1200 UTC temperature and dew point were both significantly lower during high P&I mortality events (Table 3) . For the daily SSC analysis, lower frequencies of moist moderate (MM) air (p = 0.019) and higher frequencies of moist polar (MP) air (p,0.001) occurred 15-19 days before high mortality events (Table 3) . No relationship was found for dry polar (DP) air (p = 0.277). Temperature, dew point temperature, and air mass frequencies were examined 17 days prior to and throughout each of the 12 high P&I mortality episodes identified from 1975-2002. There is a statistically significant negative relationship between episode duration and mean 1200 UTC dew point (r = 20.61, p,0.05; Figure 3a ). The longest episode (57 days in 1981) was associated with a dry period during which the mean dew point was more than 0.8uC below normal for that time of year. Total episode mortality is likewise negatively correlated with 1200 UTC dew point (r = 20.56, p,0.05; Figure 3b ). Two of the three episodes with the lowest mortality also had dew points that were near normal, whereas the higher mortality episodes exhibited drier conditions. No significant relationships were found for afternoon variables, air temperature, or any of the SSC air mass types. Dew point temperature is commonly used by atmospheric scientists to measure humidity because it is relatively invariant to pressure and temperature changes and thus is a conservative quantity. In New York City from 1975-2002, periods with high P&I mortality were preceded 2-3 weeks by periods with low morning dew points. Furthermore, for the 12 high mortality episodes identified in that period, morning dew point was negatively correlated with both episode duration (r = 20.61) and total episode mortality (r = 20.56). This finding of a linkage between dry air and influenza mortality is consistent with the results of recent research [6] showing that absolute humidity (a correlate of T d ) influences the timing of the onset of the influenza season in various U.S. states (including New York state). Our results provide some limited evidence supporting laboratory studies linking dry air to higher airborne infection rates among guinea pigs [4] . The association between high frequencies of Moist Polar air masses prior to high mortality events is consistent with the dew point results. The average morning dew point in Moist Polar air in New York City is lower than for any air mass other than Dry Polar (Table 4 ). Although it may seem counter-intuitive that a ''moist'' air mass has a low dew point temperature, dew point must (physically) be less than or equal to air temperature, so cold air masses typically have low dew points, especially in winter. In general, Moist Polar air masses are relatively uncommon in the cold season in New York, occurring only 7.3% of the time (Table 4 ) and are associated with cold, overcast and often stormy conditions with moist air arriving from the Atlantic Ocean [33] . We also identified a significant relationship between low Moist Moderate air mass frequencies 3-4 weeks before mortality episodes. In an effort to understand this association, we calculated the correlation between Moist Moderate frequencies and the other air mass types during that period. Moist Moderate is negatively correlated with the two driest air masses-Dry Polar (r = 20.47) and Moist Polar (r = 20.31) (Table 4 ). Thus, the significant association between low Moist Moderate frequencies prior to P&I mortality events appears to be a proxy for the cool, low dew point conditions that are common when Moist Polar and Dry Polar air masses are present. On average, cold season dew points are 5.5uC higher in Moist Moderate air masses than in Moist Polar ( Table 4) . The lack of a direct Dry Polar relationship is surprising, as Dry Polar air masses exhibit the most extreme combination of cold air and low humidity. Dry Polar is far more common than Moist Polar in New York winters, however, so its high daily frequency during the influenza season limits the likelihood of identifying an underlying relationship. It might be more fruitful to examine an extreme cold, dry subset of Dry Polar air masses to identify the coldest and driest days. In New York City, high P&I mortality periods within a given year were preceded by multiple day periods with unusually low temperature and humidity. Over the 28-year period of this study, we identified 12 episodes of high P&I mortality and found that both the total mortality occurring during each episode and duration of each episode were inversely correlated with the average morning dew point temperature prior to and during the episodes. These results support the burgeoning hypothesis that unusually cold dry air enhances the airborne transmission of influenza virus. The exploratory nature of this analysis was necessitated by the lack of an underlying theory of influenza seasonality, sociobehavioral factors, and inherent variability in disease transmission and virulence. The time between infection and a resulting mortality event (i.e. ''latency'') varies between individuals depending on age, overall health, co-morbid conditions, and other factors. Thus, lags must be estimated to best fit the overall data structure. Similarly, the high frequency variability in the variables requires some smoothing to elucidate relationships, and the selection of appropriate smoothers is somewhat subjective. Nevertheless, our findings are consistent with several others. For example, there is evidence supporting a two-week lag between rising influenza virus and pneumonia mortality [28] . Other research showed fairly convincingly the existence of a 2-4 week lag between laboratory-confirmed cases of influenza and increased incidence of invasive pneumococcal disease [43] . The weather variables used in this study do not directly account for the ambient conditions experienced by the influenza victims while indoors, but cold and/or stormy weather could result in the decedents spending more time in heated indoor environs with low humidity, thereby enhancing infection opportunities [8] . For this study, P&I mortality was used to characterize the influenza time series in New York City. The limitation of this method is the potential for confounding as P&I mortality includes mortality from infections other than influenza. In addition, in nonpandemic years, P&I mortality is skewed by the extremes of age. This limitation is unlikely to be a major contributor in this study as 90% of influenza-related deaths involve persons over the age of 65 during seasonal epidemics [45] and our mortality data are ageadjusted to account for demographic changes in New York City over the period of the study. We chose to focus on New York City because the large population (and thus large daily P&I mortality rate) enhances statistical robustness, and New York weather is highly variable owing to its midlatitude, coastal location. These results should be confirmed using a similar methodology in other cities worldwide to determine if the humidity-influenza linkage is pervasive. It would be particularly interesting to determine how these relationships evolve in subtropical or tropical climates where the P&I mortality seasonality is more muted or nonexistent. It is likely that the underlying causes of influenza seasonality are multi-factorial, and we suspect that weather is but one of those factors. A predictive model for P&I mortality based on weather alone would likely be unsuccessful in accounting for most of the short-term influenza variability. Nevertheless, our results confirm recent emerging hypotheses of a relationship between cold, dry air and influenza transmission or virulence [6] , [8] . Identifying periods of low dew point temperatures a few days or even weeks in advance is well within the skill of existing weather forecast models. Given the lag between infection and mortality, it seems reasonable to propose that, during the prime influenza season, skillful forecasts of high P&I mortality periods could be made weeks to months in advance. Figure S1 Sample of mortality data from December 1, 1998 through March 10, 1999 . Daily mortality (z-scores) (red dashed lines) shows evidence of the beginning of a prolonged peak starting around day 43. When these data are smoothed using a 17day centered moving average filter (solid red line), the mortality peak becomes more evident. In our analysis, we instead employ a leading 17-day smoother (blue line), which effectively shifts the red line forward by 8 days. High mortality episodes are classified when the z-scores exceeds 1, so the 1999 episodes begins on day 39 and ends on day 77. (TIF) The decline in dew point on day 10 preceded the start of the mortality increase by approximately 17 days. During the subsequent period of declining dew points, mortality continued to rise. When dew point reached its minimum for this period on day 31, 17-day lagged mortality was one standard deviation above the mean. Although there is no consistent 1:1 lagged relationship between dew point temperature and mortality, this example illustrates the procedure and shows a general linkage between a low dew point period in mid-late December, 1998 and a subsequent high pneumonia and influenza mortality anomaly several weeks later. (TIF) Conceived and designed the experiments: RED CER KBE. Performed the experiments: CER. Analyzed the data: CER RED KBE. Wrote the paper: RED CER KBE.

Finding and removing highly connected individuals using suboptimal vaccines

BACKGROUND: Social networks are often highly skewed, meaning that the vast majority of the population has only few contacts whereas a small minority has a large number of contacts. These highly connected individuals may play an important role in case of an infectious disease outbreak. METHODS: We propose a novel strategy of finding and immunizing highly connected individuals and evaluate this strategy by computer simulations, using a stochastic, individual-and network-based simulation approach. A small random sample of the population is asked to list their acquaintances, and those who are mentioned most frequently are offered vaccination. This intervention is combined with case isolation and contact tracing. RESULTS: Asking only 10% of the population for 10 acquaintances each and vaccinating the most frequently named people strongly diminishes the magnitude of an outbreak which would otherwise have exhausted the available isolation units and gone out of control. It is extremely important to immunize all identified highly connected individuals. Omitting a few of them because of unsuccessful vaccination jeopardizes the overall success, unless non-immunized individuals are taken under surveillance. CONCLUSIONS: The strategy proposed in this paper is particularly successful because it attacks the very point from which the transmission network draws its strength: the highly connected individuals. Current preparedness and containment plans for smallpox and other infectious diseases may benefit from such knowledge.

Super-spreader events (cf. [1] ) crucially influence the course of infectious disease outbreaks, as has been shown for SARS, measles and smallpox. Targeting control efforts on individuals with highest potential to spread disease is more effective than mass control [1] . This is very important for diseases like smallpox for which herd immunity is decreasing and stockpiled vaccines are of low eligibility or uncertain immunogenicity [2] [3] [4] [5] [6] . Specific information on social networks and on their contact structures is still scarce, but common properties have been revealed for many networks [7] : it has been shown that the degree distribution of social networks are frequently highly skewed, i.e. they are bound together by just a few very highly connected individuals. The frequency of contacts in such networks is not Poisson distributed (as would be expected in networks which originate from a random mixing process), but follows a skewed and long-tailed distribution [8] : the vast majority of the population has rather few contacts whereas a small minority has a huge number of contacts. Highly skewed networks are ubiquitous in nature and it seems that this particular topology confers "dynamical robustness and reliability to perform a certain function in the presence of perturbations" [9] . Indeed, a highly skewed frequency distribution of the number of contacts per person has some astonishing effects on the transmission of infection diseases and on the effect of interventions [1] . In contrast to the results with an assumption of a homogeneous mixing population, the disease transmission is only marginally reduced if a percentage of individuals is immunized at random and, thus, "removed" from the transmission network. In the presence of highly connected individuals, even the celebrated basic reproduction number R 0 , originally defined as "the average number of secondary cases caused by a single index case in a completely nonimmune homogenously mixing population where no interventions are taken", no longer predicts whether an outbreak can occur or not [8, [10] [11] [12] [13] . These same highly-connected individuals which stabilize the transmission properties of a skewed contact network in the case of random "removal" of individuals, also make these networks vulnerable, if they can be identified and "removed". One approach to identify them, considering a theoretical infection process spreading on a skewed contact network, has been termed "acquaintance immunization" [14, 15] . Here, people are picked at random and asked to name one contact each who will then be vaccinated. People with many contacts are most likely to be mentioned by somebody and are very likely to be vaccinated. An even more efficient strategy has been found by assuming that individuals can guess information about their neighbors and their contacts [16] . Acquaintances have also been shown to be good social network sensors for early detection of outbreaks [17] . Inspired by Cohen et al. [14, 15] , the aim of the present paper is to explore the effect of such targeted immunization strategies on the course of an epidemic considering a real disease -smallpox-using pessimistic assumptions and suboptimal vaccines. We use computer simulations based on a stochastic transmission model where individuals are connected with each other in two superimposed networks, which allows us to distinguish between close contacts (comprising family members and close friends which can easily be traced) and casual contacts who will be more worrisome in case of an outbreak, because they are more difficult to detect and to be placed under surveillance. For the latter ones, we use a highly skewed network. Our baseline intervention scenario considers case isolation (with a limited capacity) and tracing of close contacts which we combine with a pre-emptive vaccination of highly connected individuals. In our simulations, we identify highly connected individuals by first "asking" a small random fraction of the population to supply the names of their casual contacts. The most frequently mentioned contacts are then offered to be vaccinated. Depending on the simulation scenario, only a fraction of the contacts can be named, or alternatively, only a fraction of the most frequently mentioned contacts can be immunized (this may be caused by combination of a low vaccination eligibility and an imperfect vaccine efficacy). In some of the considered scenarios, highly connected individuals are additionally placed under surveillance for an indefinite duration, so that they can be prevented from spreading the infection. We use a stochastic, individual-and network-based simulation approach. Individuals have discrete states which can be changed by events that are scheduled on a continuous time scale and executed using a discrete event simulation algorithm. Executed events can trigger future events which affect the same individual or -through the contact network-other individuals. The population consists of 100,000 fully susceptible individuals which form the nodes of a network graph. The contact network is a combination of two networks, representing two types of links: close contacts are represented by a two-dimensional toroidal square lattice with eight nearest neighbours as contacts, a so-called Moore neighbourhood; casual contacts are represented by a highly skewed network created with the Barabási-Albert algorithm that starts with a fully interconnected network of 12 nodes and then uses preferential attachment to add nodes [7] . Each individual of the population is characterized by two internal states: the infection state and the surveillance state. All individuals are 100% susceptible at the beginning of the simulation. The infection model which controls the natural history of the disease is an extended SEIR-model (see Table 1 and Additional file 1: Figure S1 ). The simulation starts with all individuals being susceptible except for 100 randomly selected individuals who are newly infected with the virus. As their infection progresses, the infected individuals develop prodromal fever and then proceed through an early rash, a middle rash and a late rash state. A fraction of individuals dies in the early rash state, all others acquire a lasting immunity after recovery. A detailed description of the infection process is given in the Appendix. The surveillance model (Additional file 2: Figure S2 ) controls individual states related to case detection, contact tracing, observation, and interventions like isolation or seclusion. At the start of each simulation, all individuals are unobserved. Two days after a yet unobserved new case develops the earliest signs of a rash, he or she is detected. Immediately after case detection, all close contacts and 10% of the casual contacts of the case are traced and put under observation which lasts for a maximum of 21 days (which is longer than the maximum incubation period). Observed individuals are detected immediately after developing prodromal fever. After detection, cases ought to be isolated immediately, but the number of isolation units is limited. To deal with this limited resource, all detected cases first enter a waiting queue; as soon as free isolation units become available, they are isolated (isolation is assumed to prevent all further contacts). While no free isolation units are available, detected cases are asked to seclude themselves which means that they will prevent all casual and 50% of their close contacts. Secluded cases are immediately isolated when free isolation units become available. Vaccination is implemented by first asking a fraction of the population to supply the names of their casual contacts. The most frequently mentioned contacts are then pre-emptively vaccinated, before the outbreak occurs. Technically, we first define the index cases and then define the 'vaccinated' individuals, different from those who are index cases. We do this to ensure a standard number of index cases (100), all of them unvaccinated. Depending on the simulation scenario, only a fraction of the contacts can be named, or alternatively, only a fraction of the most frequently mentioned contacts can be immunized (this may be caused by combination of a low vaccination eligibility and an imperfect vaccine efficacy). In some of the considered scenarios, vaccinated individuals are additionally placed under surveillance for an indefinite duration, so that they can be prevented from spreading the infection. In Figure 1 , we explore the theoretical limits of what can be achieved by "removing" highly connected individuals from the contact network: the vertical axis shows the maximum eigenvalue of the next generation matrix of the casual contact network which predicts the initial spread of the epidemic. The horizontal axis shows the fraction of the population which is removed from the transmission network (e.g. by immunization), whereby the individuals are sorted by their number of contacts, such that the most highly connected individuals will be removed first (red curve). Even if only a small fraction of the population is removed, Figure 1 shows a steep decline in the eigenvalue, which clearly indicates that removing the most highly connected individuals has a huge effect on the index cases' capacity to spread the infection, as has been shown in other publications (1). The blue curve in Figure 1 shows a much less dramatic effect which is gained if individuals are removed at random -irrespective of their number of contacts. The relevance of this finding can be seen in Figure 2 , which shows how the targeted immunization influences the simulation results of an outbreak which starts with 100 index cases in a population of 100,000 individuals. We start with a highly over-simplified situation in which we assume that we can ask everybody in the population to supply the names of all casual contacts, and later progress towards more realistic scenarios. The contacts are first sorted by frequency and then scheduled for vaccination, starting with the most frequently named person. As described in detail in the Methods section, detected cases are isolated or are asked to stay at home or to seclude themselves from contacts with other individuals (if all 500 isolation units are occupied); their contacts are traced and taken under surveillance. If only case isolation and contact tracing are used to fight against the outbreak, the large initial attack overwhelms the public health resources and the smallpox outbreak afflicts more than 50% of the population. When additionally targeting the vaccination on the most frequently named contacts, an immunization coverage of only 4% yields a median outbreak of less than 1,600 cases; a vaccination coverage of 6% can further reduce the median to less than 1,300 cases with a worst case scenario of under 1,700 cases. Asking everybody in the population is clearly an unrealistic task for any public health system. We will next explore how the results change, if only a small random fraction of the population is asked to name all casual contacts. Figure 3 shows how the median outbreak size changes if only a small random sample is asked. The horizontal axis again shows what percentage of the population is vaccinated, whereby immunization begins with the most frequently named contact. If we are willing to vaccinate 10% of the population, it suffices to ask a random sample of 4% of the population for their contacts in order to obtain a median outbreak size of less than 2,000 cases. For the highly optimistic goal of having a median outbreak size of less than 1,000 cases, we have to ask 10% of the population to supply the names of their contacts. In reality, hardly anybody may be able to recall all casual contacts. In the next refinement, we assume that people are asked to supply only a limited number of casual contacts. Figure 4 shows that an immunization of 10% of the population still yields very optimistic results, even if the 10% randomly sampled individuals only name 6 to 10 contacts each. In the previous scenarios, we have assumed a vaccine with 100% efficacy and we have assumed that every contact which was scheduled for vaccination was also eligible, but these assumptions cannot be made for the existing smallpox vaccines [2] . Figure 5 explores the influence of an imperfect vaccine on the median of the outbreak size. Successfully immunizing as much as 80% or 90% of the scheduled individuals may only be possible with one of the new generation vaccines [2] [3] [4] [5] [6] , and even this may already be regarded as over-optimistic by some, yet this assumption already increases the median outbreak size considerably ( Figure 5 ). The disproportionate effect of a few vaccination failures on the simulation result can be Figure 2 Influence of Perfect Vaccination of Highly Connected Individuals on a Smallpox Outbreak. Efficacy = 100%. Simulated outbreak sizes, caused by 100 index cases in a population of 100,000: the boxes show 25%, 50%, 75% quantiles, the whiskers show 10% and 90% quantiles and the triangles show minimum and maximum results (for 2% vaccination, the upper limit is not given, as one of the 100 simulations afflicted more than half of the population, as did all simulations for 0% and 1% vaccination). Targeted vaccination: (a) everybody is asked to name all casual contacts; (b) these contacts are sorted by frequency; (c) vaccination starts with the most frequently named person and progresses towards less frequently named ones. The horizontal axis shows what percentage of the population is vaccinated. explained by a closer look at the maximum eigenvalues depicted in Figure 1 : The removal of all highly connected individuals (red curve) is necessary to really incapacitate the transmission network; if only a random sample of 90% of the highly connected individuals are removed (yellow curve), it practically stays intact. In our final scenario, we combine immunization targeted to highly connected individuals (using an imperfect vaccine) with the surveillance of the vaccinated individuals: each contact which is scheduled for vaccination, will be taken under surveillance (if the vaccination success can be determined, it is sufficient to observe contacts with failed vaccination and contacts who were excluded from vaccination). Figure 6 assumes that a random sample of 90% of the individuals who are scheduled for vaccination can really be immunized. The blue curve shows the median outcome with, the red curve without additional surveillance of selected individuals. Our results show that a combination of a novel strategy of finding and immunizing highly connected individuals with case isolation and contact tracing can prevent a large smallpox outbreak, with the advantage of vaccinating only about 10% of the population (Figure 2 ). This low vaccination coverage considerably reduces the number of severe side effects and deaths due to vaccination [2, 3, 18] . In the case of smallpox vaccines this is of paramount importance, as reported by Casey et al. 2006 , "after the inoculation of 37,901 people in the United States, three deaths, two permanent disabilities, and ten life-threatening illnesses were attributed to vaccination during 2003" [6] . Successful control strategies proposed for an intentional release of smallpox, vary from targeting high-risk individuals to broader random vaccination campaigns and no one control method can be identified a priori as best [19] [20] [21] [22] . Uncertainty with respect to transmission before the onset of symptoms, residual herd immunity, demographics and mobility exist, so their differing conclusions can largely be attributed to underlying differences in model structures and parameter assignments. Our study is the first that uses a highly skewed contact structure for a smallpox outbreak. In comparison to previous modeling studies, and in order to show the strength of this strategy, we focus on a highly pessimistic scenario: (a) the outbreak starts with the simultaneous infection of 100 independent index cases on a fully susceptible population; (b) a high proportion of transmission takes place during the prodromal fever phase which does not yet reveal the nature of the infection and which does not trigger case detection except for suspects who are already under surveillance; (c) we consider a limited number of isolation units which are easily exhausted by a major outbreak; and (d) we consider suboptimal vaccines. We have made a big effort to design a realistic model and to use plausible parameter values, yet we made a number of inevitable simplifying assumptions on aspects that we consider out of the scope of the present paper. Worth of mention is that we use a pre-assigned and bidirectional contact network, to which we superimpose a stochastic transmission process (see Appendix). The contact network constitutes 'potential contacts' and the actual transmissions occur according to the transmission process. In reality, contact structures change over time so it is uncertain whether all pre-emptively identified highly connected individuals would play a role in a future outbreak. Specially in acts of bioterrorism, superspreading events (cf. [1] ) might be difficult to foresee unless awareness is increased. We assume, however, that highly connected individuals would have a potential for disease spread and constitute a good first guess. Our study is the first that addresses the problem of suboptimal vaccines in the case of smallpox. We show that a network-based vaccination strategy strongly depends on a very high vaccine efficacy and on a high eligibility and compliance of the people selected for vaccination. It can quickly lose its effect if some of the most highly connected individuals cannot be vaccinated or if vaccination fails protect them ( Figure 5 ). If only a few of them are left unprotected, they can fuel a super-spreading event (cf. [1] ) multiplying the infection in the population. Adopting the described containment strategy may require to overcome political, social and ethical hurdles. A vaccination campaign against smallpox will not be initiated anywhere before a strong suspicion of bioterror attack occurs or the appearance of smallpox cases inside or outside a country has been confirmed. As long as a smallpox bioterror attack is regarded to be highly unlikely, most people would not accept to receive a potentially harmful smallpox vaccination anyway. This perception may change drastically after smallpox have reappeared somewhere in the world. Individuals who have been identified to be under the highest risk of contracting the infection may gladly accept to be chosen to receive a protective vaccine. Yet, due to their health condition, not all of them may be eligible for vaccination. As it is very important to remove all potential highly infectious individuals (Figure 5) , highly connected individuals who cannot be vaccinated or whose immunization has failed must be taken under surveillance ( Figure 6 ). Considering the described containment strategy in preparedness plans would shift their focus to identifying potential highly infectious individuals and to preparing for pre-emptive vaccination of a pre-selected fraction of the population. It would also put emphasis on the availability of isolation units and of people trained in contact tracing and surveillance. If the public health system is overtaxed by an outbreak despite this targeted vaccination, containment strategies involving ring vaccination or large-scale vaccination should be implemented. Public health emergency plans have to provide instructions on (1) the observation of the dynamics of an outbreak and (2) which observations trigger the transition from a targeted containment strategy to a population-based one. Infections which share the same route of transmission may also have the same highly connected individuals. Knowing which individuals have many contacts (whereby a "contact" depends on the mode of transmission of an infectious agent) may help public health agencies to control whole groups of infectious diseases, including newly emerging ones, rather than individual diseases. In specific settings, highly connected individuals may be identified using other approaches: in hospitals and companies, an analysis of the team structure and its meeting schedules can hint at who has most contacts. This may also affect the focus of business continuity plans, written to prepare for pandemic influenza or similar events. Simulation studies show that highly connected individuals are reached rather early by newly introduced infections. Thus, the knowledge of such people can also be used to improve outbreak detection [17] . Current preparedness and containment plans for smallpox and other directly transmitted diseases like measles and pandemic influenza can benefit from considering the described pre-emptive strategy which effectively targets the fraction of the population fuelling disease spread and minimizes vaccine related side effects. We define eight expected numbers E s,n of secondary infections per index case which depend on the index case's infectious stage s {prodromal, earlyrash, middlerash, laterash} and on the network n {close, casual} in which the secondary cases are produced (Table S1 ). This distinction enables us to consider a more intense exposure of close contacts and to modify the cases' contagiousness in their different disease stages, as suggested by previous smallpox studies [19, 23] . We set the sum of these expected values to 5, the estimate at smallpox's eradication [24] . In a homogeneously mixing population, this would be called the basic reproduction number R 0 . We have decided to parameterize our model with these expected values rather than the maximum eigenvalue of the next generation matrix, not only because expected values are more intuitive, but also because estimates of R 0 are frequently based on secondary cases [14, 23] or on homogeneous mixing assumptions [25, 26] . As the maximum eigenvalue better predicts the initial spread of an epidemic, we use it in Figure 3 in the results. In the simulation model, we implement the following infection process: when an individual enters one of the four contagious states, it triggers future infection events for each of its contacts. This is done in the following way: (1) we chose the effective contact rate b s,n such that where D s is the duration of the specific contagious stage s and c n is the average number of contacts per person in network n (for close contacts, c n is 8, for casual ones, it is 24). Solving Equation 1 yields β s,n = − ln(1 − E s,n /c n ) D s (2) (2) we draw a random number r from an exponential distribution with mean 1/b s,n , (3) if r is smaller than the duration of the current contagious state D s , we schedule an infection event for the corresponding contact with delay r, otherwise the infection event is discarded. Infection events are stored in an event queue. At the scheduled time, it is checked whether the infection source is still contagious, the prospective victim is still susceptible and the contact between the two has not yet been impeded by interventions (e.g. isolation or seclusion) (see Table 1 ). The infection only takes place if all of these three conditions are fulfilled. Additional file 1: Figure S1 . Infection model. Additional file 2: Figure S2 . Surveillance model.

The impact of media coverage on the transmission dynamics of human influenza

BACKGROUND: There is an urgent need to understand how the provision of information influences individual risk perception and how this in turn shapes the evolution of epidemics. Individuals are influenced by information in complex and unpredictable ways. Emerging infectious diseases, such as the recent swine flu epidemic, may be particular hotspots for a media-fueled rush to vaccination; conversely, seasonal diseases may receive little media attention, despite their high mortality rate, due to their perceived lack of newness. METHODS: We formulate a deterministic transmission and vaccination model to investigate the effects of media coverage on the transmission dynamics of influenza. The population is subdivided into different classes according to their disease status. The compartmental model includes the effect of media coverage on reporting the number of infections as well as the number of individuals successfully vaccinated. RESULTS: A threshold parameter (the basic reproductive ratio) is analytically derived and used to discuss the local stability of the disease-free steady state. The impact of costs that can be incurred, which include vaccination, education, implementation and campaigns on media coverage, are also investigated using optimal control theory. A simplified version of the model with pulse vaccination shows that the media can trigger a vaccinating panic if the vaccine is imperfect and simplified messages result in the vaccinated mixing with the infectives without regard to disease risk. CONCLUSIONS: The effects of media on an outbreak are complex. Simplified understandings of disease epidemiology, propogated through media soundbites, may make the disease significantly worse.

Infectious diseases are responsible for a quarter of all deaths in the world annually, the vast majority occurring in low-and middle-income countries [1] . There are diseases such as SARS and flu that exhibit some distinct features such as rapid spatial spread and visible symptoms [2] . These features, associated with the increasing trend of globalization and the development of information technology, are expected to be shared by other emerging/ re-emerging infectious diseases. It is therefore important to refine classical mathematical models to reflect these features by adding the dimensions of massive news coverage that have great influence not only on the individual behaviours but also on the formation and implementation of public intervention and control policies [2] . People's response to the threat of disease is dependent on their perception of risk, which is influenced by public and private information disseminated widely by the media. While government agencies for disease control and prevention may attempt to contain the disease [3] , the general information disseminated to the public is often restricted to simply reporting the number of infections and deaths. Mass media are widely acknowleged as key tools in risk communication [4, 5] , but have been criticised for making risk a spectacle to capitalise on audience anxiety [6, 7] . The original interpretation of media effects in communication theory was a "hypodermic needle" or "magic bullet" theory of the mass media. Early communication theorists [8, 9] imagined that a particular media message would be directly injected into the minds of media spectators. This theory of media effects, in which the mass media has a direct and rapid influence on everyday understanding, has been substantially revised. Contemporary media studies analyses how media consumers might only partially accept a particular media message [10] , how the media is shaped by dominant cultural norms [11, 12] and how media consumers resist dominant media messages [13, 14] . It follows that media effects may sway people into panic (eg swine flu), especially with a disease where scientific evidence is thin or nonexistent. Conversely, media may have little effect on seasonal diseases (eg regular influenza). Media reporting plays a key role in the perception, management and even creation of crisis [6] . Since media reports are retrievable and because the messages are widely distributed, they gain authority as an intersubjective anchorage for personal recollection [4] . At times of crisis, non-state-controlled media thrive, while statecontrolled media are usually rewarded for creating an illusion of normalcy [6] . Media exposure and attention partially mediate the effects of variables such as demographics and personal experience on risk judgments [5] . The role of media coverage on disease outbreaks is thus crucial and should be given prominence in the study of disease dynamics. Klein et al., [15] noted that much more research is needed to understand how provision of information influences individual risk perception and how it shapes the evolution of epidemics; for example, individuals may overprotect, which can have additional consequences for the spread of disease. An example of such complex dynamics is the 1994 outbreak of plague in a state in India: after the announcement of the disease, many people fled the state of Surat in an effort to escape the disease, thus carrying the disease to other parts of the country [16] . Even though information on the number of cases and deaths can have an adverse effect, the number of those vaccinated has not been given prominence. A handful of mathematical models have described the impact of media coverage on the transmission dynamics of infectious diseases. Liu et al. [2] examined the potential for multiple outbreaks and sustained oscillations of emerging infectious diseases due to the psychological impact from reported numbers of infectious and hospitalized individuals. Liu & Cui [3] analysed a compartment model that described the spread and control of an infectious disease under the influence of media coverage. Li & Cui [17] incorporated constant and pulse vaccination in SIS epidemic models with media coverage. Cui et al., [18] showed that when the media impact is sufficiently strong, their modelwith incidence rate being of the exponential form capturing the alertness to the disease of each susceptible individual in the populationexhibits multiple positive equilibria (also see [2] ) which poses a challenge to the prediction and control of the outbreaks of infectious diseases. The aim of this study is to investigate the impact of media coverage on the spread and control of an influenza strain when a vaccine is available, and where the media reporting of both disease dynamics and vaccination is high. Vaccination is one of the most effective tools for reducing the burden of infectious diseases [19] . However, despite their public-health benefit, vaccination programs face obstacles. Individuals often refuse or avoid vaccinations they perceive to be risky. Recently, rumours that the polio vaccine could cause sterility and spread HIV have hampered polio eradication in Nigeria [20] , while misplaced fears of autism in the developed world have stoked vaccination fears [21] . Reporting the number of individuals who vaccinate may have a positive effect on the disease transmission by increasing the vaccination rate. Conversely, behavioural interventions can also have an enormous effect on the course of a disease [22, 23] Our model considers the same contact rate after a media alert, as proposed by Liu & Cui [3] , but there are fundamental differences in both models. They consider the classical SIR type model, while vaccination is included in ours to reflect transmission dynamics of human influenza. We divide the population (N) into four sub-populations, according to their disease status: susceptible (S), vaccinated (V ), infected (I), and recovered (R). Our model monitors the dynamics of influenza based on a single strain without effective cross-immunity against the strain. The susceptible population is increased by recruitment of individuals (either by birth or immigration), and by the loss of immunity, acquired through previous vaccination or natural infection. This population is reduced through vaccination (moving to class V ), infection (moving to class I) and by natural death or emigration. The population of vaccinated individuals is increased by vaccination of susceptible individuals. Since the vaccine does not confer immunity to all vaccine recipients, vaccinated individuals may become infected, but at a lower rate than unvaccinated. The vaccinated class is thus diminished by this infection (moving to class I) by waning of vaccine-based immunity (moving to class S) and by natural death. The population of infected individuals is increased by infection of susceptibles, including those who remain susceptible despite being vaccinated. It is diminished by natural death, death due to disease and by recovery from the disease (moving to class R). The recovered class is increased by individuals recovering from their infection and is decreased as individuals succumb to natural death. Media coverage is introduced into the model via a saturated incidence function. A schematic model flowchart is depicted in Figure 1 . The transmission model with media coverage is given by the following deterministic system of nonlinear ordinary differential equations: where Λ is the rate at which individuals are recruited into the population (recruitment of infectives is ignored for now); θ is the rate at which susceptible individuals receive the vaccine; µ is the the rate at which people leave the population, through natural death or emigration. We assume this rate to be the same for all sub-populations. b 1 is the rate at which susceptibles get infected; ω is the rate at which vaccine-based immunity wanes; g is the vaccine efficacy; a is the death rate due to the infection; and l is the recovery rate from infection. The terms b 2 I m I I + and b 3 I m I I + measure the effect of reduction of the contact rate when infectious and vaccinated individuals are reported in the media [2, 3, 18] . The half-saturation constant m I > 0 reflects the impact of media coverage on the contact transmission. The function g I I m I I ( ) = + is a continuous bounded function which takes into account disease saturation or psychological effects [24] . We note that recovered individuals cannot be vaccinated. Also, a vaccinated individual who gets infected and then recovers will return to the susceptible class with no vaccine protection. This is true even if ω is quite small but s and l are large. For example, if vaccination lasts three years, but recovery and loss of immunity takes 6 months, then we are assuming this person is subsequently unvaccinated. In the Michaelis-Menten functional response, the rate at which information is spread by the media rises as infectives increase, but eventually levels off at a plateau (or asymptote) at which the information (rate) remains constant (i.e. it has reached a maximum number of individuals due to information saturation) regardless of the increase in infections. Such dynamics can easily be observed in the spread of rumours, gossip and jokes (also known as randomized broadcast) [25, 26] . This constant coverage is extended by examining more complex effects which involve more than just reducing contacts down the line. The news in particular is extremely fickle so that what is news one day may be forgotten about next week; including the media effects in some more sophisticated way such as by an impulsive pulsing is also investigated. The limited power of the infection due to contact is accounted for by the saturation incidence. The first available information is the reported number of infected individuals when the disease is emerging. We assume that media coverage can slow but not prevent disease spread, so b 1 ≥ b 2 and b 1 ≥ b 3 . The above model is closely related to those in [27, 28] to analyze the transmission dynamics of human influenza, but there are some differences. In [27] , the authors consider the inflow of infective immigrants, while in [28] the model includes treatment. Neither of these are considered here. Our model is clearly a crude reflection of the complicated nonlinear phenomena of the transmission dynamics, and it does not incorporate the self-control property due to the change of avoidance patterns of individuals at different stages of the infectious process [2] . News coverage may have a significant impact on avoidance behaviours at both individual and society levels, which may reduce the effective contact between susceptible and infectious individuals; we include this via a saturation incidence functional response. Since the model monitors human populations, all the variables and parameters of the model are nonnegative. Based on biological considerations the system of equations (1)-(4) will be studied in the following region, which is positively invariant and attracting (thus, the model is mathematically and epidemiologically wellposed); it is therefore sufficient to consider solutions in Ω. Existence, uniqueness and continuation results for model system (1)-(4) hold in this region and all solutions of this system starting in Ω remain in Ω for all t ≥ 0. The disease-free equilibrium of the system is given by The endemic equilibrium of the system is given by Substituting the above into the second equation at equilibrium will yield the expression for Î after some rearrangement. The basic reproductive ratio, R v , is defined as the expected number of secondary infections caused by an infective individual upon entering a totally susceptible population [29] [30] [31] . This quantity is not only important in describing the infectious power of the disease, but can also can supply information for controlling the spread of the disease [32] . The linear stability of E v0 is governed by the basic reproductive ratio R v . Using the next-generation method [31] , we have The basic reproductive ratio is the spectral radius r(FV -1 ) which is Local stability of the disease-free equilibrium Proof. The Jacobian of the system evaluated at E v0 is given by For local stability of the disease-free equilibrium, we require that all the eigenvalues be negative. Three of the eigenvalues satisfy this condition while ς 2 < 0 implies that R v < 1 and, consequently, all the eigenvalues of the Jacobian matrix above have negative real part. Thus, the disease-free equilibrium is locally asymptotically stable. We adopt the method of Castillo-Chavez et al, [33] and we rewrite the set of model equations in the form with G(X G ,0) = 0. X G ℝ 3 denotes the number of uninfected classes and Z G ℝ denotes the number of infected classes. U X For the set of equations in (1)-(4), we set X G = (S, V, R) and Z G = (I). The conditions (H1) and (H2) below must be met for global stability. (H1) For * 0 is an M-matrix (the off-diagonal elements of A are nonnegative) and Ω is the region where the model makes biological sense. If the above two conditions are satisfied, then the following theorem holds. Theorem 2 (Castillo-Chavez et al, [33] ): The fixed point U X is a globally stable equilibrium of (2.28) provided that R v < 1 and that assumptions (H1) and (H2) are satisfied. Therefore, E v0 is globally asymptotically stable (GAS) since Ĝ(X G ,Z G ) > 0. The GAS of E v0 excludes any possibility of the phenomenon of backward bifurcation. We note that the GAS of the DFE E v0 when s = 0 is straightforward. Our objective in this section is to extend the initial model to include two intervention methods, called controls, represented as functions of time and assigned reasonable upper and lower bounds, each representing a possible method of influenza intervention. Using optimal control theory and numerical simulations, we determine the benefit of vaccination and media coverage when the latter has positive or negative effect on the former. We will integrate the essential components into one SIVR-type model to accommodate the dynamics of an influenza outbreak determined by population-specific parameters such as the effect of contact reduction when infectious and vaccinated individuals are reported in the media. Let u v and u m be the control variables for vaccination and media coverage respectively. Thus, model (1)-(4) now reads A balance of multiple intervention methods can differ between populations. A successful mitigation scheme is one which reduces influenza-related deaths with minimal cost. A control scheme is assumed to be optimal if it maximizes the objective functional The first two terms represent the benefit of the susceptible and vaccinated populations. The parameters B 1 and B 2 represent the weight constraints for the infected population and the control, respectively. They can also represent balancing coefficients transforming the integral into dollars expended over a finite time period of T days [34] . The goal is to maximize the populations of susceptible and vaccinated individuals, minimize the population of infectives, maximize the benefits of media coverage and vaccination, while minimizing the systemic costs of both media coverage and vaccination. The value u v (t) = u m (t) = 1 represents the maximal control due to vaccination and media coverage, respectively. The terms B u t v 2 2 ( ) and B u t m 2 2 ( ) represent the maximal cost of education, implementation and campaigns on both vaccination and media coverage. S(t) and V(t) account for the fitness of the susceptible and the vaccinated groups as a result of a reduction in the rate at which the vaccine wanes, and vaccination and treatment efforts are implemented [35] . We thus seek optimal controls u t v * ( ) and u t m * ( ) such that The basic framework of this problem is to characterize the optimal control. The existence of an optimal control can be obtained by using a result by Joshi [36] and Fister et al. [37] . Theorem 3Consider the control problem with the system of Equations (4.1)-(4.4). There exists an optimal control ( ) Proof. To prove this theorem on the existence of an optimal control, we use a result from Fleming and Rishel [38] (Theorem 4.1 pp. 68-69), where the following properties must be satisfied. 1. The set of controls and corresponding state variables is nonempty. 2. The control set U is closed and convex. 3. The right-hand side of the state system is bounded above by a linear function in the state and control. 4. The integrand of the functional is concave on U and is bounded above by An existence result in Lukes [39] (Theorem 9.2.1) for the system of equations (6)-(9) for bounded coefficients is used to give the first condition. The control set is closed and convex by definition. The right-hand side of the state system (Equations (4.1)-(4.4)) satisfies Condition 3 since the state solutions are a priori bounded. The integrand in the objective functional, , is concave on U. Furthermore, c 1 , c 2 > 0 and k > 1, so Therefore, the optimal control exists, since the lefthand side of (11) is bounded; consequently, the states are bounded. Since there exists an optimal control for maximizing the functional (10) subject to equations (6)-(9), we use Pontryagin's Maximum Principle to derive the necessary conditions for this optimal control. Pontryagin's Maximum Principle introduces adjoint functions that allow us to attach our state system (of differential equations), to our objective functional. After first showing existence of optimal controls, this principle can be used to obtain the differential equations for the adjoint variables, corresponding boundary conditions and the characterization of an optimal control u v * and u m * . This characterization gives a representation of an optimal control in terms of the state and adjoint functions. Also, this principle converts the problem of minimizing the objective functional subject to the state system into minimizing either the Lagrangian or the Hamiltonian with respect to the controls (bounded measurable functions) at each time t [40] . The Lagrangian is defined as where w 11 (t) ≥ 0, w 12 (t) ≥ 0 are penalty multipliers satisfying w 11 (t)(a 11u v (t)) + w 12 (t)(u v (t)b 11 ) at the optimal u v * , and w 21 (t) ≥ 0, w 22 (t) ≥ 0 are penalty multipliers satisfying w 21 (t)(a 22u m (t)) + w 22 (t)(u m (t)b 22 ) at the optimal u m * . Given optimal controls u v * and u m * , and solutions of the corresponding state system (6) with transversality conditions l i [t f ] = 0, for i = 1, 2, 3, 4. To determine the interior maximum of our Lagrangian, we take the partial derivatives of L with respect to u v and u m , respectively, and set it to zero. Thus, To determine an explicit expression for our controls u m * , u m * (without w 11 , w 12 , w 21 , w 22 ), a standard optimality technique is utilized. The following cases are considered to determine a specific characterization of the optimal control. The optimal system The optimality system consists of the state system coupled with the adjoint system, with the initial conditions, the transversality conditions and the characterization of the optimal control: [36] . The uniqueness of the optimal control follows from the uniqueness of the optimality system. The state system of differential equations and the adjoint system of differential equations together with the control characterization above form the optimality system solved numerically and depicted in Figures 2, 3 , 4, 5. The general model with pulse vaccination is given as In this model, vaccination occurs at fixed times, not continuously. This is closer to reality, since vaccination centres are only open at certain times, when people may get vaccinated in waves. Similarly, media stories tend to clump together, so that a big news story occurs on one day, which may trigger a short period of intense vaccination. We shall use a simplified version of this model to illustrate the possibility that media may have an adverse effect. Consider the following scenario. At the onset of the outbreak, the media -and hence the general population -is unaware of the disease and thus nobody gets the vaccine, allowing the disease to spread in its initial stages. At some point, there is a critical number of infected individuals, whereupon people are sufficiently aware of the infection to change their behaviour. We suppose that, initially, new infected people arrive at fixed times. We further assume that vaccinated people mix more than susceptibles. In this case, people who are vaccinated feel confident enough to mix with the infected, even though they may still have the possibility to contract the virus. This might be the case for health-care workers, for instance, who get vaccinated and then have to tend to the sick. Mathematically, we have a threshold for the critical number of infectives, I crit . For I <I crit , this model would look like For I >I crit , the model becomes However, to illustrate the adverse affect, we shall simplify the model even further. For a short timescale, we can assume recovery is permanent, so s = 0. Thus, we can ignore the R equation. For I <I crit , we assume that there is no mixing, but rather that new infectives arrive impulsively into the system at fixed times t k and in numbers I i , where I i ≪ I crit . (If the new infectives arrive at irregular times, then the broad results will be unchanged.) For I >I crit , fear of the disease keeps susceptibles from mixing with the infected, but the vaccinated will. Thus b 4 = b 6 = 0. Since I i ≪ I crit , we can assume that, for I >I crit , the effects of new infectives are negligible. The model then becomes for I <I crit and for I >I crit . Thus, the effects of the media are to trigger a vaccinating panic whenever the number of infectives is large enough. We kept the model with impulse vaccination as simple as possible since even this simplified version shows that media reports could have an adverse effect. Suppose new infectives appear regularly, so that t k+1t k = τ. (If not, the analysis generalizes quite easily.) For t k <t <t k+1 , we have It follows that the endemic equilibrium is stable if Î >I crit . Thus, even in an extremely simplified version of the model, the media may make things significantly worse than if no media effect were included. We kept this model deliberately simple, partly for mathematical tractability and partly to show that the media effects apply even in this idealised scenario. Note that, in reality, the fluctuations would apply in the upper region as well, making the actual value even larger. In the lower region, we ignored interaction between susceptibles and infectives (ie we assume b 4 = b 6 = 0). The effect of including these terms would be to slow the exponential decay between impulses (or possibly cause it to increase). This would only increase the effect seen here. In summary, a small series of outbreaks that would equilibrate at some maximum level m + >I crit will, as a result of the media, instead equilibrate at a much larger value I >m + >I crit . The driving factor here is if an imperfect vaccine causes overconfidence, so that people who have been vaccinated mix significantly more with infectives than susceptibles do. If this happens (as would be quite likely; most people who have been vaccinated feel invulnerable, even if the vaccine is not perfect, largely thanks to media oversimplifications), then the media effect is likely to be adverse. A simplified version of the model with pulse vaccination shows that the media can make things worse, if the vaccine is imperfect because the vaccinated mix over-confidently with the infectives. We now return to model (6)- (9) and illustrate some of the properties discussed in the previous sections. The parameter values that we use for numerical simulations are in Table 1 [41] . We consider an imperfect flu vaccine for which the waning rate is about 0.15. The relationship between b 2 and b 3 is not very obvious; consequently, we can either assume equality or that the former is slightly greater than the latter. Transmission dynamics of infectious diseases with and without media coverage have already been carried out in previous studies, but these models do not account for the vaccination coverage. Therefore, we illustrate some numerical results for the model with optimal control when media coverage has (i) a beneficial effect (see Figures 2 and 3) (ii) and adverse effect on the vaccination rate (see Figures 4 and 5 ). The optimality system is solved using an iterative method with a fourth order Runge-Kutta scheme. Starting with a guess for the adjoint variables, the state equations are solved forward in time. Then the state values obtained are used to solve the adjoint equations backward in time; the iterations continue until convergence. Simulations are carried out to determine how maximizing media coverage enhances vaccination. The effects of costs that can be incurred, which include education, implementation and campaigns on media coverage, are also studied to evaluate how these costs can affect the transmission of human influenza. We increase the value of B2 (the cost weight) in Figure 2 to assess how the populations of susceptibles, infectives, vaccinated and recovered individuals are altered. In Figure 3 , we investigate how increasing minimization of infectives through increasing the weight B1 affects the control of human influenza transmission. We do the same in Figures 4 and 5 , respectively, to see how, if media coverage has an adverse effect, the various populations behave. In Figure 4 , we vary the cost weight, while in Figure 5 , we vary the weight of minimizing infectives. We note from Figure 2 that, during the initial days, there is a very sharp drop in the population of infected individuals, while other populations show increases. Increasing minimization of infectives, while keeping costs low, can lead to the disease being controllable. The slight rise and fall, after the initial 20 days, in the population of infectives could be attributed to complacency on the part of some individuals (or may be due to oscillations in the system independent of external factors). We find that people tend to relax after the initial shock of the disease threat. However, we note that this is not for long, and this could be attributed to the fact that vaccination levels continue to rise, so as people continue to receive vaccination, infection is controlled. Thus, if costs are kept minimal, and more people are able to access media and vaccination, then infection can be controlled. Both vaccination and media coverage continue at optimum levels as a result of the low costs and minimization of infectives. From Figure 3 , as costs are increased, few people have access to media and vaccination; as a result, low numbers get vaccinated against the disease. In the long run, the infection levels rise. The degree of media coverage and vaccination also decrease as a result of the exorbitant costs. With the little available media coverage and the few vaccinated individuals, we find that, due to information filtration, there is a jump in the vaccination levels, though these only last briefly; as the degree of media coverage and vaccination decrease, so do the vaccination levels. From Figure 4 , even though costs are kept at minimal levels, the negative reports concerning vaccination result in a drastic reduction in the vaccination levels. After some time, we note a slight increase in the vaccination levels; however, these numbers remain very low. This could be due to the fact that, as infection rises, a few will risk getting vaccinated in the hope of being cured. Thus, media coverage can have adverse effects if people's perception towards the vaccine is negatively influenced by the media. In Figure 5 , both media coverage and vaccination are eventually withdrawn. Very low numbers get vaccinated. It is only when infection escalates that vaccination levels also increase as some might find it better to try to prevent the infection, despite the negativity towards vaccination in the media. Figure 6 illustrates other potential adverse effects that media may have, if the effect is to trigger a vaccinating panic where vaccinated individuals are not fully protected and mix with infected individuals but susceptible individuals do not. In this case, the number of infected individuals may increase sharply as a result of the media. Figure 7 illustrates the long-term Media simplifications can lead to overconfidence in the idea of a vaccine as a cure-all. The result is not just a vaccinating panic and a blow-out epidemic, but a net increase in the endemic equilibrium. Thus, media coverage of an emerging epidemic can fan the flames of fear and also implicitly reinforce an imperfect solution as the only answer. We have formulated and investigated a simple deterministic vaccination model describing the effects of media coverage on the transmission dynamics of influenza. The media effect due to reporting the number of infections as well as the number of individuals successfully vaccinated is introduced into the compartmental model via a saturated incidence-type function. The With media effects Figure 6 The vaccination panic threshold The effect of the vaccination panic threshold using the simplified model (14)- (20) . Without media triggering a vaccinating panic, the number of infected individuals remains low (solid purple curve). However, if the media triggers a vaccinating panic, then the number of infected individuals rises sharply (dashed green curve). Inset: Comparison of the two outcomes around the vaccination threshold. impact of costs that can be incurred, which include vaccination, education, implementation and campaigns on media coverage, are also investigated using optimal control theory applied via the Pontryagin's maximum principle. A simplified version of the model with pulse vaccination shows that the media can have an adverse effect if the vaccine is imperfect and the vaccinated mix over-confidently with the infectives. Numerical simulations are carried out to support the analytical results. We note, however, that our caricature model is not complete; a more comprehensive study will require interdisciplinary research across traditional boundaries of social, natural, medical sciences and mathematics [2] . Nevertheless, our work provides some insights into the effects of media reporting on the transmission dynamics of infectious diseases for which a vaccine exists. The present study is in no way exhaustive and can be extended in various ways: for example, to investigate the case in which there is media coverage but people ignore it (in which case the vaccination rate is unchanged despite the control). Thus, the effects of media on an outbreak of influenza with a partially effective vaccine may be complicated. While the media may encourage more people to get vaccinated, they may also trigger a vaccinating panic or promote overconfidence in the ability of a vaccine to fully protect against the disease. This may have potentially disastrous consequences in the face of a new pandemic.

Detection of Mycobacterium ulcerans by the Loop Mediated Isothermal Amplification Method

BACKGROUND: Buruli ulcer (BU) caused by Mycobacterium ulcerans (M. ulcerans) has emerged as an important public health problem in several rural communities in sub-Saharan Africa. Early diagnosis and prompt treatment are important in preventing disfiguring complications associated with late stages of the disease progression. Presently there is no simple and rapid test that is appropriate for early diagnosis and use in the low-resource settings where M. ulcerans is most prevalent. METHODOLOGY: We compared conventional and pocket warmer loop mediated isothermal amplification (LAMP) methods (using a heat block and a pocket warmer respectively as heat source for amplification reaction) for the detection of M. ulcerans in clinical specimens. The effect of purified and crude DNA preparations on the detection rate of the LAMP assays were also investigated and compared with that of IS2404 PCR, a reference assay for the detection of M. ulcerans. Thirty clinical specimens from suspected BU cases were examined by LAMP and IS2404 PCR. PRINCIPAL FINDINGS: The lower detection limit of both LAMP methods at 60°C was 300 copies of IS2404 and 30 copies of IS2404 for the conventional LAMP at 65°C. When purified DNA extracts were used, both the conventional LAMP and IS2404 PCR concordantly detected 21 positive cases, while the pocket warmer LAMP detected 19 cases. Nine of 30 samples were positive by both the LAMP assays as well as IS2404 PCR when crude extracts of clinical specimens were used. CONCLUSION/SIGNIFICANCE: The LAMP method can be used as a simple and rapid test for the detection of M. ulcerans in clinical specimens. However, obtaining purified DNA, as well as generating isothermal conditions, remains a major challenge for the use of the LAMP method under field conditions. With further improvement in DNA extraction and amplification conditions, the pwLAMP could be used as a point of care diagnostic test for BU

Buruli ulcer (BU) caused by Mycobacterium ulcerans (M. ulcerans) is a necrotizing skin disease endemic mostly in rural wetland of tropical countries of Africa, America, Asia and Australia. The disease also occurs in non-tropical areas of Australia, China and Japan. Globally, BU has been reported in over 30 countries [1] [2] [3] . The burden of BU is however most severe in sub Saharan Africa where the true incidence of the disease is difficult to determine as a result of poor surveillance measures and case confirmation [2] . Available data however reveals an increase in BU incidence over the last several years in the west African countries of Ivory Coast, Ghana and Benin. In these countries BU has replaced leprosy as the second most prevalent mycobacterial disease [1] , [3] [4] [5] . BU begins as painless nodule, papule, plaque or edema that evolves into characteristic ulcers with undermined edges. If untreated, extensive ulceration (that can cover 15% of the body), scarring and contractures may cause serious functional disabilities in patients [5] [6] [7] . Unfortunately most patients seek treatment late and present with large ulcers [8] [9] [10] . Previously treatment of such lesions involved surgical removal of all the affected tissue and part of the surrounding tissues, eventually followed by skin grafting [9] [10] [11] [12] . In 2004 antimycobacterial treatment alone (if necessary in combination with surgery) was introduced and has since been considered as the treatment of choice for BU [6] , [13] [14] [15] [16] . Laboratory confirmation of clinically suspected BU cases has therefore become crucial for the clinical management of BU [17] . Four laboratory tests are recommended for the diagnosis of BU. These include microscopic examination, culture, IS2404 PCR and histopathological analysis. Microscopic examination detects 29%-78% of clinically suspected BU cases and is currently the only rapid and affordable test available for BU diagnosis in many endemic areas. The detection rate of culture is between 34%-79% and takes an average of 9-12 weeks to yield positive results. Culture therefore cannot be used for rapid laboratory confirmation of BU. Histopathological analysis is reported to detect 30% additional cases than other confirmatory tests, however this technique is restricted to external reference laboratories and are unavailable in peripheral health centres or district or regional hospitals. IS2404 PCR has close to 96% sensitivity and is considered the method of choice for laboratory confirmation of BU [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] . The WHO recommends that at least 50% of cases must be confirmed by IS2404 PCR before commencement of antibiotic therapy [22] [23] [24] [25] . However technical difficulties (eg, cold chain requirement, stable power supply and qualified laboratory staff) limit the use of this diagnostic test in BU endemic areas. A dry reagent PCR consisting of lyophilized PCR mix which is reconstituted with water for testing DNA was developed to simplify BU diagnosis by PCR [26] but this method also requires the use of a thermocycler, electrophoresis and gel imaging equipment and therefore similarly makes the use of this diagnostic test for BU diagnosis in endemic areas unlikely. The Loop Mediated Isothermal Amplification (LAMP) is a novel nucleic acid amplification method for molecular detection and identification [27] . The principle of LAMP is autocycling strand displacement DNA synthesis in the presence of Bst DNA polymerase with high strand displacement activity under isothermal conditions between 60-65uC within 60 minutes [28] . The assay is highly specific due to the recognition of target DNA by 4 to 6 independent sequences and the amplification efficiency of LAMP is equivalent to that of PCR based methods ( [27] , [29] , [30] ). The LAMP reaction enables easy identification of positive tests due to the accumulation of high amounts of amplification products in the reaction tubes. Further improvement in visual identification can be realized through the addition of intercalating dyes such as SYBR green or hydroxynapthtol blue to reaction tubes [31] . This therefore precludes the need for post amplification analysis and hence reduces cost and labour. LAMP has also been shown to be less affected by a number of inhibitors of conventional PCR [32] . Additionally the closed tube format of this assay reduces problem of carry over contamination which is likely in less controlled environments [33] . With all of these characteristics LAMP of DNA has emerged as a powerful tool to facilitate point of care diagnostic test [31] . In order to develop a field applicable technique that offers high detection sensitivity and specificity for the diagnosis of BU, we explored the use of the pocket warmer LAMP (pwLAMP) technique, a DNA amplification method using isothermal conditions (60uC) provided by a disposable pocket warmer [34] . Ethical approval for analysing patients' specimens was obtained from the ethical review board of the Noguchi Memorial Institute for Medical Research. Specimens used were anonymously taken from an already existing collection of patients' specimens processed for diagnosis of BU from Agogo Presbyterian Hospital in Ghana. Thirty clinical specimens consisting of 20 swabs and 10 fine needle aspirates taken respectively from ulcers and pre-ulcerative lesions of suspected BU patients were used in this study. The fine needle aspirate specimens were kept in 1 ml phosphate buffered saline (PBS) and swabs were stored dry in sterile tubes. Each swab was transferred into a tube containing 2 ml milli-Q purified water (Millipore Corporation, Billerica, MA) and gently vortexed for 5 sec and then removed. Portions 250 ml of the sample suspensions were transferred to separate new sterile eppendorf tubes containing 250 ml of lysis buffer (1.6 M GuHCl, 60 mM Tris pH 7.4, 1% Triton X-100, 60 mM EDTA, Tween-20 10%), 50 ml proteinase-K and 250 ml glass beads. The mixtures were incubated horizontally in a shaker (200 rpm) at 60uC overnight. To capture the DNA, 40 ml of diatomaceous earth solution (10 g diatomaceous earth obtained from Sigma Aldrich Chemi GmbH in 50 ml of H 2 O containing 500 ml of 37% (wt/vol) HCl) was added to the suspensions and incubated at 37uC with shaking (200 rpm) for 60 min. The mixtures were centrifuged at 14,000 rpm for 10 sec and the resulting pellets were washed twice with 900 ml of 70% ethanol (2-8uC) followed by 900 ml of acetone. The pellets were dried at 50uC for 20 min and resuspended in 100 ml milli Q purified water and centrifuged at 14,000 rpm for 10 sec. The purified DNA was used as templates for both IS2404 PCR and LAMP assays to detect M. ulcerans. To investigate the performance of the LAMP assay on crude DNA preparations, we obtained 2 types of DNA extracts for each clinical specimen. One crude extract consisted of 250 ml suspensions of the specimen boiled for 10 min followed by centrifugation at 14,000 rpm for 5 min (boiled extract). The other crude extract used was a 250 ml suspension of the unboiled specimen. Ten M. ulcerans strains grown on LJ slants were harvested and DNA was extracted as previously described [35] . Serial dilutions of purified M. ulcerans DNA containing 300,000, 30,000, 300, 30 and 3 copies of IS2404 element per 5 ml were prepared. The number of copies of the insertion sequence element was determined based on the genome size of 5,806 kb and presence of an average number of 207 copies of IS2404. This was used to determine the detection limit of the LAMP assays. In order to develop a simple and rapid test that can be used to diagnose Buruli ulcer under field conditions, we modified the conventional LAMP assay by using a disposable pocket warmer as a heating device for generating a constant temperature for the test reaction and employed the use of crude sample preparations consisting of boiled and unboiled extracts of the clinical specimen instead of using purified DNA as the diagnostic specimen. Thirty clinical specimens from suspected Buruli ulcer patients were investigated by the modified LAMP (or pocket warmer LAMP) and the conventional LAMP, as well as IS2404 PCR, a reference method for the detection of Mycobacterium ulcerans. There was no significant difference in the detection rate (63-70%) in all of the methods when purified samples were used for the tests. On the other hand the use of crude specimen preparation resulted in a drop in detection rate (30-40%) . This study demonstrates that the LAMP test can be used for rapid detection of M. ulcerans when purified DNA preparations are used. With further improvements in the sample reaction, as well as in specimen purification, the pocket warmer LAMP may provide a simple and rapid diagnostic test for Buruli ulcer. LAMP for Detection of Mycobacterium ulcerans www.plosntds.org PCR for IS2404 PCR targeting IS2404 was performed as described previously [21] . The first and second round PCRs used primers pGp1: 59-AGGGCAGCGCGGTGATACGG-39and pGp2: 59-CAGTG-GATTGGTGCCGATCGAG-39 and pGp3: 59-GGCGCAGAT-CAACTTCGCGGT-39 and pGp4: 59-CTGCGTGGTGCTTT-ACGCGC-39, respectively. For the First round, the 30 ml reaction volume contained 3 ml DNA, 25 pmol/ml of each primer (pGp1 and pGp2), 3 ml of 106 PCR buffer (containing 1.5 mM magnesium chloride), 6.0 ml Qsolution, 0.2 mM deoxynucleotide triphosphates (dNTPs) and 1.0 U HotStar Taq polymerase (QIAGEN). For the second run, 1 ml of the first run product was added to 24 ml reaction volume containing, 25 pmol/ml of each primer (pGp3 and pGp4), 2. Pocket warmer LAMP (pwLAMP) was performed using a loopamp DNA amplification kit (Eiken Chemical) described previously [32] . Each 25 ml reaction mixture contained 1.6 mM each of FIP and BIP, 0.2 mM each of F3 and B3, 0.8 mM each of LF and LB, 26 reaction mixture (12.5 ml), 1 ml of Bst DNA polymerase, 1 ml of fluorescence detection reagent (Eiken Chemical), 3.5 ml distilled water and 1 ml sample. Reaction tubes were incubated at 60uC for 60 min in the heat block (GeneAmp 9700, Applied Biosystems, Foster City, CA) while with the pwLAMP, the tubes were sandwiched in a twofold pocket warmer (Hokaron Haru-type, Lotte Health Products, Tokyo, Japan) surrounded by a paper towel, and put in a Styrofoam box for 120 min (60 min reaction incubation). The reaction was terminated at 85uC for 5 min and the results were read by eye in ambient light and also using UV illumination. A chi-squared test was performed to reveal the statistical difference using SPSS (version 16.0; SPSS Inc., Chicago, IL) software. LAMP reaction requires a constant temperature of about 60u-65uC for 60 min for amplification of DNA [27] [28] [29] [30] [31] [32] [33] [34] . In a previous study a pocket warmer reached 58uC in 30 min and stayed around 60uC for more than 60 min in a Styrofoam box [34] . The 3 pocket warmers (of a pack of 30 hand warmers) tested in this study achieved a temperature of 60uC after 60 min and maintained this temperature for about 90 min. The pocket warmer thus provided a suitable temperature (60uC) and time range (60 min) for amplification. Both pwLAMP and the conventional LAMP assays were able to detect to the limit of 300 copies of the target sequence after 60 min of amplification. This limit improved to 30 copies when the conventional LAMP was carried out at 65uC (the pocket warmer was not able to attain this temperature and was therefore not investigated). The sensitivity and specificity of the LAMP assays for the detection of M. ulcerans is shown in tables 1 and 2. Under ambient illumination, positive specimens in the LAMP assay produced greenish colouration (Figure 2 ). When purified DNA extracts were used, 21 (16 swabs, 5 fine needle aspirates) (70%) of 30 clinical specimens were positive by IS2404 PCR as well as by the conventional LAMP. None of the PCR positive specimens were negative by conventional LAMP. However 19 samples of purified DNA extracts were positive with the pwLAMP, but the 90.5% sensitivity (19/21) of the pwLAMP compared to the results by IS2404 PCR was not statistically significant (p = 0.58, Chi-square test). All negative specimens in IS2404 PCR were negative in both LAMP assays, indicating specificities of both LAMP assays to the reference method were 100%. Twelve unboiled (9 swabs and 3 fine needle aspirates) and 9 boiled (6 swabs and 3 fine needle aspirates) extracts were positive by all 3 detection assays with sensitivities of 57.1% (unboiled, 12/21) and 42.9% (boiled, 9/21) compared to results using purified DNA extracts respectively for both LAMP and IS2404 PCR assays. The positivity of swabs was found to be in the range of 30% to 80% compared to 30% to 50% for fine needle aspirates. When the positivities in crude DNA specimens were compared with those in purified DNA, the differences were statistically significant by chi-square test (unboiled vs purified DNA, p = 0.0195, and boiled vs purified DNA, p = 0.0019). None of the IS2404 PCR negatives was positive in the LAMP assays irrespective of the DNA extracts type used. These data suggest that sensitivity of LAMP and PCR assays for the detection of M. ulcerans in clinical specimens is enhanced when purified DNA extracts are used. BU is a neglected tropical disease that mostly affects the poor in resource limited communities in sub-Saharan Africa [1] [2] [3] . IS2404PCR, the method of choice for confirmation of BU diagnosis cannot be operational in BU endemic areas [16] , [17] , [24] [25] . The development of rapid and reliable point of care diagnostic assays is of high priority to BU management and prevention. This study explored the potential use of the LAMP method for field diagnosis of BU. Some important limitations to the use of this assay in the field include specimen purification and difficulty in maintaining isothermal condition for the reaction. A study suggested that omission of DNA extraction or the use of crude DNA extracts have no effect on the LAMP test [32] . Hatano et al used a disposable pocket warmer to provide isothermal condition for LAMP reaction [34] . Based on this knowledge, we applied the pwLAMP to crude and purified DNA extracts in order to determine whether this method will be suitable for use as a point of care diagnostic test for BU. Although the pocket warmers used in this study reached 60uC after 1 hr (instead of 30 min in previous studies [34] ), both devices achieved the requisite temperature and holding time for executing LAMP reaction, a major advantage in the use of amplification based assay for the detection of an infectious agent under field condition. The pwLAMP did not cross-react with other mycobacteria ( Figure 1) . Moreover, the experiment on clinical specimens demonstrated that the pwLAMP had a 100% specificity in clinical specimens of BU. The pwLAMP was found to have comparable sensitivity as the conventional LAMP at 60uC as both assays were able to detect 300 copies of IS2404 element (equivalent of 1.5 genomes of M. ulcerans). The detection limit of the conventional LAMP at 65uC improved to 30 copies of IS2404 and this level of sensitivity may probably be achieved with a pocket warmer that can attain a temperature of 65uC and maintain a holding time of 60 min. When applied to purified DNA extracts of clinical specimens, the pwLAMP, conventional LAMP and IS2404 PCR yielded concordant results (Tables 1 and 2) . However, 2 of the samples that were positive by IS2404 PCR/ conventional LAMP were negative by the pw LAMP. None of the IS2404 PCR negative samples were positive in both types of LAMP assays. On the other hand we observed a drop in detection rate from 63-70% to 30-40% when crude extracts of clinical specimens were used (Tables 1 and 2) . This indicates that the use of crude DNA extracts as template may not be appropriate for the detection of M. ulcerans by the LAMP method. This observation contradicts a previous study that suggested omission of DNA extraction had no effect on sensitivity of the LAMP assay [32] . For the crude preparations however, it is noteworthy that the detection rate of the LAMP assay was significantly higher for the unboiled extracts than for the boiled extracts. Explanations for these results were not explored. The observation that the LAMP assay was not inhibited especially for the unboiled specimens is quite consistent with previous work that have shown LAMP to be tolerant to culture medium and to certain biological substances including phosphate buffered saline, serum, plasma, urine and vitreous [32] . In conclusion, the study demonstrates that the LAMP assay yields comparable results as IS2404 PCR when it is performed at 60u-65uC for 60 min on purified DNA extracts and further supports the use of the pocket warmer as a device for providing isothermal amplification condition for the LAMP assay. This therefore is a potential boost to the application of pwLAMP in resource poor settings. Challenges of obtaining pure DNA extracts of clinical specimen as well as the use of a pocket warmer capable of maintaining 65uC for 1 hr, however needs to be addressed in order to improve the performance of the pwLAMP assay. Further development and testing in larger numbers of specimens is therefore necessary to access the potential use of pwLAMP as a simple and rapid point of care diagnostic test for BU.

A Human PrM Antibody That Recognizes a Novel Cryptic Epitope on Dengue E Glycoprotein

Dengue virus (DENV) is a major mosquito-borne pathogen infecting up to 100 million people each year; so far no effective treatment or vaccines are available. Recently, highly cross-reactive and infection-enhancing pre-membrane (prM)-specific antibodies were found to dominate the anti-DENV immune response in humans, raising concern over vaccine candidates that contain native dengue prM sequences. In this study, we have isolated a broadly cross-reactive prM-specific antibody, D29, during a screen with a non-immunized human Fab-phage library against the four serotypes of DENV. The antibody is capable of restoring the infectivity of virtually non-infectious immature DENV (imDENV) in FcγR-bearing K562 cells. Remarkably, D29 also cross-reacted with a cryptic epitope on the envelope (E) protein located to the DI/DII junction as evidenced by site-directed mutagenesis. This cryptic epitope, while inaccessible to antibody binding in a native virus particle, may become exposed if E is not properly folded. These findings suggest that generation of anti-prM antibodies that enhance DENV infection may not be completely avoided even with immunization strategies employing E protein alone or subunits of E proteins.

Dengue virus (DENV) is a flavivirus with four related but antigenically distinct serotypes (DENV1-4). It infects approximately 50-100 million people each year, of which 500,000 people exhibit the life-threatening form of severe dengue -dengue haemorrhagic fever (DHF) and dengue shock syndrome (DSS) [1] . The current lack of treatment or licensed vaccine means dengue poses a serious public health threat [2] . Infection by one serotype of DENV confers lifelong immunity against the homologous serotype, but only limited crossprotection to the remaining three serotypes [3, 4] . The presence of cross-reactive, non-neutralizing antibodies generated during a primary infection has been suggested to enhance the pathogenicity of subsequent infections via the process of antibody-dependent enhancement (ADE) [5] . A successful and safe vaccine candidate must therefore elicit a protective long-lasting immune response to all four serotypes [4, [6] [7] [8] . Recent immunological studies have shown the human anti-DENV immune response to be dominated by prM-specific antibodies in both primary and secondary infections [9, 10] . These prM-specific antibodies are highly cross-reactive and non-neutralizing. When complexed with immature DENV (imDENV), it has the ability to render normally non-infectious imDENV highly infectious [11, 12] . This has caused concern over current vaccine candidates that contain native dengue prM [9] -which is a component of most current vaccine strategies whether naturally attenuated, recombinantly attenuated, yellow fever-dengue-virus chimeras, chemically inacti-vated virus, DNA vaccine or recombinant subunit protein vaccines [13, 14] . Vaccine candidates that do not contain prM proteins, such as soluble recombinant Envelope (E) protein or E domain subunit vaccines may thus become increasingly important. To gain a deeper understanding of the early DENV-specific immune response in humans, the four serotypes of DENV were sequentially screened with a non-immunized human Fab phage display library. Broadly cross-reactive prM-specific antibodies dominated the screen and the Fab with highest affinity, D29 Fab-IgG, was converted into full-length human IgG1 format for thorough characterization. This antibody (D29 Fab-IgG) was found to have high affinity for a conformational epitope on prM and, like other prM antibodies [11, 12] , was capable of restoring the infectivity of virtually non-infectious immature DENV (imDENV) in FccR-bearing K562 cells. The antibody also cross-reacted with E protein -fine mapping and site-directed mutagenesis studies localized the epitope to the DI/ DII junction of E, which would be inaccessible in a native virus particle. This suggests the possibility that immunization strategies employing E protein alone or subunits of E proteins may not be able to completely avoid generation of anti-prM antibodies that enhance DENV infection. Aedes albopictus C6/36 cells (ATCC) were maintained in Leibovitz L-15 media (GIBCO, Invitrogen) supplemented with 8% fetal bovine serum (FBS), at 28 o C, 5% CO 2 . BHK-21 cells (ATCC) and human erythroleukemic K562 cells (ATCC) were maintained in RPMI 1640 GlutaMAX medium (RPMI) (GIBCO, Invitrogen) containing 10% FBS, and incubated at 37 o C, 5% CO 2 . Vero cells (ATCC) were grown in 199 medium (M199) (GIBCO, Invitrogen) supplemented with 8% FBS, 1% sodium pyruvate and 1% NEAA. HEK293 T cells (ATCC) was cultured in Dulbecco's modified Eagle's medium (DMEM) (GIBCO, Invitrogen) supplemented with 10% FBS at 37 o C in 5% CO 2 . Mouse monoclonal antibodies 3H5 (m3H5), 4G2 (m4G2) and 2H2 (m2H2) are specific to EDIII of DENV2, EDII of flaviviruses and prM of DENV1-4 respectively. Chimeric humanized 3H5 (h3H5) and 4G2 (h4G2) were constructed by cloning the Fab portion (variable light and heavy chains) into an expression vector containing the human IgG1 framework for expression in HEK293 T cells [15] . Conjugation of HRP to antibodies-D29, -m3H5 and -m2H2, was performed using the Lightning-Link HRP Conjugation Kit (Innova Biosciences). Dengue 1 (DENV1) strain Hawaii, Dengue 2 (DENV2) strains New Guinea C (NGC) and ST, Dengue 3 (DENV3) strain H87 and Dengue 4 (DENV4) strain H241 were passaged in Vero cells for all assays unless otherwise stated. Virus titers were determined by plaque assay on BHK cells. Immature DENV (imDENV) was produced as described previously [16] . Briefly, the media of infected Vero cells was replaced 2 dpi with M199 NH 4 Cl medium (M199 with 10% FBS, 1% sodium pyruvate, 1% non essential amino acids, 1% PS, 2.5 mM L-Glutathione and 20 mM NH 4 Cl). Culture supernatants were harvested after 5 days and imDENV was purified by PEG precipitation. Virus titer was determined as genome containing particles (GCP)/ml by quantifying real-time PCR (qPCR) described previously [17] . To ascertain the binding specificity of D29 Fab-IgG, direct binding ELISA against dengue virus was performed using standard ELISA methods. Briefly, 2610 5 pfu/well purified DENV1-4 was coated on Maxisorb plate followed by blocking using 5% SM. Following incubation with 1 mg/ml dengue-specific antibodies, plates were probed with HRP-conjugated anti-human IgG-Fc secondary antibody or anti-mouse IgG-Fc secondary antibody (Pierce). Detection steps in subsequent ELISAs followed this method unless otherwise stated. To determine the binding affinity of D29 Fab-IgG direct binding ELISA was performed with serially diluted D29 Fab-IgG. Results were fitted to a one-site binding hyperbola using Prism 5.03 (GraphPad, San Diego, USA). For Peptide phage ELISA, 2 mg of D29 and non-related control antibodies were coated on Maxisorb plate and incubated with phage, followed by detection with HRP-conjugated anti-M13 secondary antibody. DENV-infected Vero cells or C6/36 cells were fixed and permeablilized with 80% acetone at 220 o C for 10 min. The presence of virus was detected by D29 Fab-IgG, h3H5 and h4G2 at 1 mg/ml, followed by staining with FITC-conjugated antihuman IgG (ZyMax, Invitrogen). Cells were mounted with MOWIOL (Sigma) and images were captured using a fluorescent microscope (Olympus). DENV-infected Vero cells were harvested 5 days post-infection and lysed with cold lysis buffer (1.5% Triton X-100, 0.2 mM Phenylmethylsulfonyl Fluoride (PMSF), 1 mg/ml pepstatin A and 75 mM KCl in PBS) for 40 min on ice. An equal volume of 56 loading dye was added to the cell lysate before 12% SDS-PAGE and transferred onto nitrocellulose membranes. Blocked membranes were probed with dengue-specific antibodies at RT for 1 hr and detected with HRP-conjugated anti-human IgG-Fc or antimouse IgG-Fc secondary antibody and developed with chemiluminescence (Supersignal West Pico Chemiluminescent Substrate, Pierce). To establish the binding target of D29 Fab-IgG, competition ELISA against epitope-characterized monoclonal antibodies was carried out. Plates were prepared as for direct binding ELISA. After blocking, serially diluted m3H5, m4G2 or m2H2 was added to the plates and incubated for 1 hr at RT, followed by addition of D29 Fab-IgG at 2.5 mg/ml and incubation for 1 hr at RT. Bound antigen-antibody complexes were detected by HRP-conjugated anti-human IgG-Fc secondary antibody (Pierce), and developed as described above. For competition Western, Fab-IgG D29 was incubated with m3H5, m4G2 or m2H2 for 1 hr at RT before applying to blocked membranes for 30 min at RT. HRP-conjugated anti-Human IgG-Fc secondary antibody was applied to detect bound D29 for 45 min at RT after 4 washes with PBST. For radioactive immunoprecipitation, DENV2-infected BHK-21 cells were harvested 48 hr later and prepared as described previously [18] . For immunoprecipitation with SDS to disrupt protein complexes, lysate from infected cells was incubated with 1.25% SDS for 30 min at 4 o C. The SDS concentration was diluted to 0.2% before immunoprecipitation and precipitated proteins were analysed by silver stained SDS-PAGE (Silver stain Plus Kit, Bio-Rad) and Western blot. The phage-displayed-12 random dodecapeptide (Ph.D-12) library (New England Biolabs) was panned against D29-Fab-IgG to identify the antibody epitope as per instruction manual. The first round of panning was carried out with 100 mg/ml D29 Fab-IgG immobilized on Maxisorb Immunotube (Nunc). The amplified phage was enriched by three further rounds of panning in solution using Protein A or G sepharose in alternate rounds of panning. One round of negative selection with Protein A and G sepharose was carried out to minimize non-specific binders. The isolated peptide phage sequences were mapped against 3D protein structure using the Peptiope server with crystal structures of prM-E heterodimer at neutral pH (Protein Data Bank (PDB) accession code 3C6E). Graphic visualization was carried out using MSI WebLab ViewLite (Accelrys). To test whether isolated peptide phage inhibits binding of D29 to its natural epitope, inhibition ELISA was carried out. 10-fold serially diluted peptide phage (starting at 10 12 pfu/ml) or 2610 5 pfu DENV2 was incubated with D29 Fab-IgG for 1 hr before applying to DENV2-coated plate for 5 min at RT. Bound antibody was detected with HRP-conjugated anti-human IgG-Fc secondary antibody for 1 hr at RT. To test the inhibition capacity of peptide phage by Western blot, pre-incubation of 0.1 mg/ml of D29 Fab-IgG with 4610 11 pfu peptide phage; 8610 6 pfu DENV2 or 5%SM for 1 hr was carried out before applying to membrane transblotted with DENV2 lysate for 30 min. HRP-conjugated anti-Human IgG-Fc secondary antibody was applied to detect bound D29 Fab-IgG for 45 min at RT and the membrane was processed as described previously to visualize the reactive bands. The gene portion containing prM-E, as described in Puttikhunt, et al [24] , was amplified from DENV2 (NGC) cDNA using the primers D2-FW (59-AATTAATACGACCGTCTCCCATGAA-TAGAAGACGCAGATCTGCAGGC-39) and D2-RV (59-CGC-CCGTTTGATCTCGAGCTACTAGGCCTGCACCATGACT-CCC-39) and cloned into pCMV-myc-ER vector via the restriction sites Nco I and Xho I to create plasmid construct pCMV-prM-E. Using this as template, selected amino acids residues of the predicted epitopes, P3 and P9, were mutated by site-directed mutagenesis using Quikchange Multi Site-directed Mutagenesis kit (Stratagene, Agilent Technologies). PCR reactions and subsequent cloning steps were carried out according to manufacturer's instructions. Substitution of amino acids in all mutant constructs was confirmed by sequencing. Mutants were expressed in HEK293-T cells and harvested 48 hr posttransfection for immunoblot analysis and immunoprecipitation experiments. To determine the ability of D29 Fab-IgG to neutralize Dengue infection in vitro, PRNT was carried out on BHK cells as described previously [19] . The ADE assay was performed as described previously [20] . Briefly, pre-formed immune complexes were prepared by incubating serially diluted antibody with DENV2 or imDENV2 at a multiplicity of 25 GCP per cell (MOG 25) in RPMI MM at 37 o C for 1 hr before applying to 2610 4 K562 cells in 96-well U-bottom plate (Nunc). Supernatant was harvested 48 hr postinfection and virus titer was quantified by plaque assay. Following isolation from the human non-immune Fab-phage library by panning against the 4 serotypes of DENV sequentially (Fig. S1 ), D29 Fab-IgG was converted into full length human IgG1 (Materials and Methods S1). To ensure the antibody was specific for DENV viral proteins, ELISA and IFA were performed with all 4 serotypes of DENV grown in C6/36 and Vero cell lines. D29 Fab-IgG reacted with infected cells from DENV1-4 without crossreaction to the non-infected cell controls, indicating that the determinant for D29 Fab binding is indeed viral and not host (Fig. 1A , B, C). Using direct ELISA, D29 Fab-IgG was found to have high affinity for all 4 serotypes of DENV ( Fig. 1D hypermutation. This suggests that antibodies like D29 can be generated in the early immune response against DENV. To broadly characterize the epitope of D29 Fab-IgG on DENV, we used competition ELISA with well characterized DENV antibodies including mouse-derived 3H5 (m3H5) which is specific to DENV2 EDIII; m4G2, a cross-reactive antibody that recognizes EDII of all flaviviruses and m2H2, a cross-DENV reactive anti-prM antibody [21, 22] . Making use of humanized versus parental murine antibodies for 3H5 and 4G2, and HRPconjugated m2H2 versus non-conjugated for m2H2, each of the reference antibodies was shown to self-compete ( Fig. 2A) . No significant competition was observed between D29 Fab-IgG and m3H5, m4G2 or non-DENV specific control antibody. However, binding of D29 Fab-IgG to DENV2 decreased significantly in the presence of m2H2 in a dose-dependent manner, indicating that D29 Fab-IgG recognized an epitope on prM similar to that of m2H2 ( Fig. 2A) . To confirm the target protein of D29 Fab-IgG, Western blot analysis was carried out with DENV-infected Vero cell lysates and the recombinant ectodomain of DENV2 E protein (rE) under reducing and non-reducing conditions. Under non-reducing conditions, the control antibodies recognized their respective target proteins from the cell lysates: h4G2 detected the ,55 kDa E protein of all 4 serotypes and the ,50 kDa rE protein, and m2H2 detected the ,20 kDa prM of all 4 serotypes (Fig. 2B) . Remarkably, D29 Fab-IgG detected both E and prM of all 4 serotypes; however, the binding was abolished in the presence of reducing agent, indicating that the D29 epitope is conformational. All control antibodies except h3H5 lost binding to their target proteins under reducing condition (Fig. 2C) , which is expected since h3H5 recognizes a linear epitope on EDIII [22, 23] , whereas both h4G2 and m2H2 are conformationally sensitive [10, 24] . In agreement with the competition ELISA data, h3H5 and h4G2 did not compete with D29 Fab-IgG for epitopes in the Western blot format, whereas incubation with m2H2 abolished D29 binding to prM but not to E protein (Fig. 2D) . Given the ability of D29 Fab-IgG to bind both prM and E on western blots, immunoprecipitation was carried out with DENV2infected BHK cell lysate to elucidate if D29 Fab-IgG binds both proteins in solution. An initial radioactive immunoprecipitation showed D29 Fab-IgG binding to both E (,55 kDa) and prM (,20 kDa), as both proteins are precipitated whereas h4G2 only precipitated E protein (Fig. 3A) . However, as prM intrinsically interacts with E to form heterodimers, it is possible that antibody binding to one of the components in the heterodimer would pull down the other. To address this, dissociation of the heterodimer with 1.25% SDS was performed prior to the assay. Immunoprecipitated proteins were resolved by reducing SDS-PAGE and visualized by silver staining. Identity of the precipitated proteins was verified by non-reducing Western blot analysis with h3H5 and m2H2, respectively (Fig. 3B ). The control antibodies were able to precipitate their corresponding binding partners. Despite the ability to recognize both prM and E on western blots, D29 Fab-IgG only precipitated prM. Taken collectively, our data indicate that D29 binds both prM and E but the epitope on E is not accessible on the native protein. As the DTT-sensitive m2H2 had previously been mapped to its epitope using linear peptides [21] , we attempted a similar approach to locate the binding epitope of D29 Fab-IgG. An inhibition ELISA was carried out using synthetic ,20 mer linear peptides covering DENV2 prM [25] as well as a library of short 15-mer overlapping linear peptides (Mimotope) corresponding to aa1-166 of prM and aa1-495 of E (Materials and Methods S1). However, none of the linear peptides were able to significantly inhibit the binding of D29 Fab-IgG to DENV2 (Fig. S3) , suggesting that the epitope of D29 Fab-IgG1 is discontinuous. Next, D29 Fab-IgG was screened using a random dodecapeptide (Ph.D-12)-phage library as this approach has previously been used to map antibody-binding motifs [26, 27] . Peptide-phage clones were randomly selected from the second, third and fourth round of panning, and their binding specificity was confirmed by direct binding ELISA against D29 Fab-IgG. Reactive clones were sequenced and checked against a list of common target-unrelated peptides (TUPs) [28] to eliminate peptides that react with constant regions of antibodies, protein A/G sepharose or plastic surfaces. No TUPs were detected and the resulting 20 unique clones fall into three main consensus groups: 26% contained the motif W(TL/SV)K(L/X)PXW; 14% contained the motif AKTMP and 33% contained the motif KXPXW ( Table 1) . Mapping of peptide sequences using the Peptiope server against the 3D crystal structure of the prM-E heterodimer at neutral pH (PDB 3C6E) identified three main clusters of predicted epitope location. Since Pepitope assumes all input peptides mimic surface residues; all buried residues are eliminated from the search. The Figure 4 . Localization of D29 Fab-IgG predicted epitopes in 3D crystal structures of DENV2 prM-E heterodimer (PDB 3C6E) and the binding specificity of peptide-phages. (A) Clusters of conformational epitopes predicted by Pepitope server are displayed on the prM-E heterodimer crystal structure at neutral pH with their solvent-accessible surfaces highlighted. The prM is pink, EDI is red, EDII is yellow, EDIII is blue, FP is cyan. (B) The binding specificity of peptide-phages (P1-P9) to D29 Fab-IgG was tested in a direct ELISA format with 10 mg/ml of D29 Fab-IgG, h3H5, m2H2 and non-DENV specific human antibody (Hu) immobilized on a Maxisorb plate. Ability of peptide-phages to inhibit D29 Fab-IgG binding to DENV2 was investigated by (C) ELISA and (D) Western blot analysis. For ELISA, peptide-phages (10 12 pfu/ml) were incubated with D29 Fab-IgG for 1 hr at RT before application to immobilized DENV2 for 5 min at RT. Bound D29 Fab-IgG was detected with HRP-conjugated anti-Human IgG-Fc. The percentage of inhibition of D29 Fab-IgG binding by the peptide-phages shown is the average of three experiments. Error bars represent the standard errors of the mean (***p-value ,0.005). For Western blot analysis, 0.5 mg/ml of D29 Fab-IgG was incubated with 8610 6 pfu of purified DENV2, 5% SM or 4610 11 pfu of peptide-phage clones for 1 hr at RT before applying to membrane transblotted with DENV2 viral lysate for 30 min at RT. doi:10.1371/journal.pone.0033451.g004 peptides with the KXPXW motif could mostly be mapped to the highest scoring Cluster 1 located at the prM/EDII interface. A few exceptions mapped to Cluster 3, most probably due to poorer sequence match and a lower alignment score. The remaining two motifs were mapped to the second highest scoring Cluster 2 that spans EDI and EDII (Fig. 4A) . To verify the specificity of the identified motifs, nine representative clones were selected (Table 1) and assessed for their ability to bind D29 Fab-IgG by direct ELISA. No crossreaction by these peptide-phage clones with control antibodies was observed, confirming that the binding to D29 Fab-IgG was specific (Fig. 4B) . The peptide-phage clones were then tested for their ability to inhibit D29 Fab-IgG's binding to immobilized DENV2. Although most of the selected peptide-phage clones only weakly inhibited D29 Fab-IgG's binding to DENV2, clone P3 (Cluster 1) and P9 (Cluster 2) significantly competed by 30% and 70%, respectively (Fig. 4C ). Significant inhibition of binding to both prM and E was also observed on Western blot for P9, however no noticeable inhibition was observed for P3 (Fig. 4D) . Alignment of P3 and P9 peptide sequences to their respective clusters revealed 9 and 11 aa match, respectively (Fig. 5 ). This may explain the difference in the ability of the two peptide phage clones to inhibit binding of D29 Fab-IgG. To confirm the residue assignation of the predicted epitopes for D29 on prM and E, clusters of 2-3 residue mutants were made on a prM-E construct that Puttikhunt and co-workers had previously shown to fold correctly when expressed in HEK293 [24] For the P3 (prM-E) epitope, two mutants, M1 (P56L-A, P58Q-G, P59N-G) and M2 (E246K-A, E247K-A) were generated (Fig. 6A ). Their reactivity with D29 Fab-IgG was tested by Western blot analysis, using h4G2 and m2H2 as control antibodies. All antibodies reacted with their respective protein targets for the non-mutated prM-E protein; however, binding of m2H2 and D29 Fab-IgG to the prM protein was almost completely abolished by mutation at the prM residues in M1 (Fig. 6B) . Mutation at the E residues within the P3 sequence (M2) on the other hand, did not have significant impact on the reactivity with D29 Fab-IgG (Fig. 6B) . Detection of both M1 and M2 by h4G2 verified the expression of both mutants. To verify the P9 (E) epitope, eight mutants were generated: E57R-G, E129V-A, E131Q-G) and M5/6 (E129V-A, E131Q-G, E133E-G, E134N-G) (Fig. 6A ). Both control antibodies were able detect their respective protein targets for all mutants, apart from Figure 5 . Localization of P3 and P9 peptide sequence on 3D crystal structure of prM-E heterodimer (PDB 3C6E). P3 (Purple) and P9 (green) peptide sequences were aligned with the predicted clusters (Cluster 1 -Navy blue; Cluster 2 -Black) on the prM-E crystal structure. The path of peptide phage sequences was highlighted with the participating residues from the cluster and the peptide phage labeled. Matched residues were purple (P3) or green (P9) and numbered accordingly; mis-matched residues were grey. The respective proteins and domains were highlighted as above. doi:10.1371/journal.pone.0033451.g005 M3/4 and M4/5 which appeared to be non-viable (Fig. 6B) . The binding of h4G2 to E of M4 also appeared to be affected by the mutation introduced (Fig. 6B) ; likely to be caused by a change of local conformation, rather than a drop of expression level since prM-binding by m2H2 was normal. Notably, the binding of D29 Fab-IgG to the E protein of the M3, M5 and M6 mutants were severely affected; in particular, D29 Fab-IgG failed to detect E of M4 and M5/6, whereas binding of prM for all these mutants was not affected at all (Fig. 6B ). Together these results confirm the epitopes identified through peptide-phage display. Previous studies with anti-prM antibodies have shown that while the antibodies are generally non-neutralizing [9, 25] , they are capable of rendering non-infectious immature DENV (imDENV) particles infectious through an FccR-mediated process [9, 12] . Indeed, D29 Fab-IgG failed to neutralize any of the 4 serotypes of DENV at the highest concentration tested (400 mg/ml) in a PRNT assay (Table S1 ). Next, the ability of D29 Fab-IgG to enhance infection of DENV2 and imDENV2 was determined by carrying out ADE assays with FccR-bearing K562 cells. The specific infectivity of the viruses was established by qPCR and plaque assay to be 8.5610 5 :1 for imDENV2, which was significantly higher than that of the standard-grown DENV2 at 240:1 (Fig. 7A) . Enhancement of DENV2 infection was observed for all antibodies, with the exception of the non-DENV specific antibody. All the DENV-specific antibodies caused an approximately 20-fold increase in infection compared with IgG control (Fig 7B) . As D29 Fab-IgG is non-neutralizing, it was capable of enhancing infection of prM-containing virus in the preparation for the entire range of concentrations tested. Infection of virtually non-infectious imDENV was also significantly enhanced by D29 Fab-IgG to a level similar to m2H2, causing an 80-100 fold increase in infection relative to control IgG (Fig. 7C) . The enhancement of imDENV infection by h3H5 was similar to that observed for DENV2, since its epitope was not affected by the maturity state hence it enhanced both equally well; h4G2 failed to enhance imDENV2 infection in K562 cells possibly be due to the occlusion of its epitope by the pr peptide [29, 30] , and indicated by the immunoprecipitation protein profile (Fig. 3B ). To gain a deeper understanding of the early immune response against DENV, we have isolated and characterised a highly crossreactive antibody fragment -D29, from a non-immune human Fab-phage library that shows near germline sequence. Upon conversion into full-length IgG, D29 showed cross-reactivity against all four serotypes of DENV and could be competed with the prM-specific mouse antibody 2H2. Intriguingly, D29 Fab-IgG was found to recognize both E and prM during Western blot analysis in a conformational dependant manner. Cross-reactivity between E and prM has been observed in antibodies isolated from mice as well as human patients [10, 11, 21] . However, the binding epitopes of these antibodies remain unknown and the observation is usually explained by general cross-reactivity of anti-prM antibodies. Initial immunoprecipitation experiments with D29 Fab-IgG precipitated both prM and E proteins. However, upon dissociation of the prM interaction with E by the addition of SDS, only prM was precipitated by D29 Fab-IgG, pinpointing prM as the native binding target. Similarly, h3H5 only precipitated E upon addition of SDS, whereas the protein profile precipitated by h4G2 was not affected possibly due to its epitope at the fusion loop of E being occluded by the presence of prM [29, 30] , and only 'free' E proteins were pulled down in both the presence and absence of SDS. The epitope of D29 was mapped using a random-peptide phage display library, an approach which has been widely used for the identification of both linear and conformational epitopes [26, [31] [32] [33] [34] . Two peptide phage clones, P3 (Cluster 1) and P9 (Cluster 2), displayed significant inhibition of D29 Fab-IgG binding to immobilized DENV2 by ELISA; P9 was also able to significantly block the binding of D29 Fab-IgG in a nonreducing Western blot analysis. This may be due to a better sequence match of P9 to its predicted cluster compared with P3 (Fig. 5) , hence, a higher affinity of the peptide to D29 Fab-IgG. To ascertain the authenticity of the predicted epitopes, the predicted residues were confirmed by site-directed mutagenesis of the P3 and P9 peptide sequence. The residues P56L, P58Q and P59N within the P3 epitope were found to be the principle residues involved in the interaction between prM and D29 Fab-IgG; they were also critical for the binding of m2H2, corroborating with results of competition assays. The binding for E protein was not really affected by mutation of aa residues within the P3 epitope; contrasting to the significant impact caused by mutation of 1-2 residues within the P9 epitope. The combined mutation of E129V, E131Q, E133E, and E134N completely abolished binding of D29 Fab-IgG to E protein, establishing the critical residues within the P9 epiotpe. Taken together, these results suggest that D29 Fab-IgG recognises two epitopes on DENV -a solvent-accessible epitope on prM as the principle binding epitope, and a cryptic epitope on E which mimics the prM epitope but is not available on a native, functionally-folded E protein. The P3 and P9 epitopes are highly conserved across the four DENV serotypes, as well as two other flaviviruses, Japanese encephalitis virus and West nile virus; but with increasing divergence in Yellow fever virus and Tick-borne encephalitis virus (Fig. S4) . Most current dengue vaccines consist of both prM and E proteins [12, 13, 35] ; studies performed with tick-borne encephalitis virus indicate that proper folding of E requires the chaperone function of prM [36] . However, as demonstrated in this study and recent reports, prM-specific antibodies are able to restore and enhance the infectivity of imDENV and partially mature DENV [9, 11, 12] ; it may be worthwhile to consider alternative vaccine approaches that minimize anti-prM responses during vaccine design [9] . The findings from our study also suggest that any partial denaturation or unfolding of an E protein vaccine preparation may result in the exposure of a cryptic P9-like epitope, which has the potential to induce D29-like antibodies with threatening ADE capability. A recent study investigating the acute and early convalescent B cell response in dengue patients has found dengue virus to have a significant B cell activation capacity, causing a transient but high appearance of plasmablast and plasma cell, coinciding with that of dengue-specific IgG antibodies at day 4-7 after onset of fever [37] . The similarity of D29 Fab-IgG to germline sequence suggests that such antibodies are present in the immune repertoire of naïve individuals and given the inherent high affinity of D29, such antibodies may be preferentially selected during clonal expansion and maintained as memory. Indeed, antibodies that cross-react between prM and E has been isolated from memory B cells in primary and secondary dengue patients [9, 11] . The original immunogens for these cross-reactive antibodies are difficult to determine since they were generated in the course of natural infections, but it would be interesting to identify the epitope and germline of these antibodies and compare the sequences with D29 Fab-IgG. and prM, 10 mg/ml of synthetic peptides corresponding to prM (A) and E (B) was incubated with D29 Fab-IgG. Peptides corresponding to E were tested individually but presented as groups of 5. For all experiments, 2610 6 pfu/ml of DENV2 or imDENV2 was included as control antigen. M1-32: 15 mer peptides corresponding to prM. Pr1-5: .20 mer custom-made peptides covering parts of prM. E1-73: 15 mer peptides corresponding to E. (**p-value,0.005) (TIF)

Lack of Association between CLEC5A Gene Single-Nucleotide Polymorphisms and Kawasaki Disease in Taiwanese Children

Background. Kawasaki disease is characterized by systemic vasculitis of unknown etiology. Previous genetic studies have identified certain candidate genes associated with susceptibility to KD and coronary artery lesions. Host innate immune response factors are involved in modulating the disease outcome. The aim of this study was to investigate CLEC5A (C-type lectin domain family 5) genetic polymorphisms with regards to the susceptibility and outcome of KD. Methods. A total of 1045 subjects (381 KD patients and 664 controls) were enrolled to identify 4 tagging single-nucleotide polymorphisms (tSNPs) of CLEC5A (rs1285968, rs11770855, rs1285935, rs1285933) by using the TaqMan Allelic Discrimination Assay. The Hardy-Weinberg equilibrium was assessed in cases and controls, and genetic effects were evaluated by the chi-square test. Results. No significant associations were noted between the genotypes and allele frequency of the 4 CLEC5A tSNPs between controls and patients. In the patients, polymorphisms of CLEC5A showed no significant association with coronary artery lesion formation and intravenous immunoglobulin treatment response. Conclusions. This study showed for the first time that polymorphisms of CLEC5A are not associated with susceptibility to KD, coronary artery lesion formation, and intravenous immunoglobulin treatment response in a Taiwanese population.

Kawasaki disease (KD) is characterized by acute, febrile, and systemic vasculitis and was first described by Kawasaki et al. in 1974 [1] . In developed countries, KD is the leading cause of acquired heart diseases in children [2, 3] . KD occurs worldwide and particularly in Japan, Korea, and Taiwan and mainly affects children less than 5 years of age [4] [5] [6] . The most serious complication of KD is the occurrence of coronary artery lesions (CALs) [7, 8] . The prevalence of KD in children younger than 5 years is the highest in Japan, followed by Korea and Taiwan, and lowest in Europe. Previous studies have either failed to identify causative pathogens for KD or reported discrepant results [9] [10] [11] . Therefore, it is possible that a genetic background plays an important role in the pathogenesis of KD. CLEC5A (C-type lectin domain family 5, member A; also known as myeloid DAP12-associating lectin (MDL-1)) Journal of Biomedicine and Biotechnology contains a C-type lectin-like fold similar to the naturalkiller T-cell C-type lectin domains and is associated with a 12-kDa DNAX-activating protein (DAP12) on myeloid cells [12] [13] [14] . Signaling via this complex constitutes a significant activation pathway in myeloid cells and plays an important role in immune defense. Recently, it has been demonstrated that CLEC5A acts as a signaling receptor for proinflammatory cytokine release, and that blockade of CLEC5A-mediated signaling attenuates the production of proinflammatory cytokines by macrophages infected with dengue virus [14] . In contrast, it has been demonstrated that MDL-1 stimulation induces a significant amount of RANTES and macrophage-derived chemokine (MDC) production in cooperation with signaling through TLR in mouse myeloid cells [15] . Furthermore, there is ample evidence that activation of peripheral blood monocytes/macrophages [16] [17] [18] , proinflammatory cytokines [16] , and the RANTES gene play a central role during acute KD [19, 20] . A persistent or increased expression of chemokine genes in the convalescent phase in patients is associated with coronary artery lesions [17, 19] . In addition, infiltration by the cells is notable in affected tissues in autopsy cases and in skin biopsy specimens from KD patients [21] . However, no CLEC5A genetic association with KD has previously been reported. To gain further understanding of the genetic role of CLEC5A in the pathogenesis of KD, the aim of our study was to determine if any CLEC5A SNPs are associated with susceptibility to KD, CAL formation, or IVIG treatment response in Taiwanese children. All study cases were children enrolled from Chang Gung Memorial Hospital, Kaohsiung Medical Center, between 2001 and 2009, who fulfilled the diagnostic criteria for KD. All patients were treated with IVIG (2 g/kg) and aspirin as per our previous studies [7, 8, 18] . This study was approved by the Institutional Review Board of Chang Gung Memorial Hospital. Blood samples were collected after informed consent was obtained from parents or guardians. CAL formation was defined as the internal diameter of the coronary artery measuring at least 3 mm (4 mm if the subject was over the age of 5 years) or the internal diameter of a segment at least 1.5 times that of an adjacent segment, as observed in echocardiography [8, 22] . IVIG responsiveness was defined as defervescence within 48 h after the completion of IVIG treatment and no recurrence of fever (temperature > 38 • C) for at least 7 days after IVIG with marked improvement or normalization of inflammatory signs [7, 8] . Blood cells were subjected to DNA extraction by treating them first with 0.5% SDS lysis buffer and then protease K (1 mg/mL) for digestion of nuclear protein for 4 h at 60 • C. Total DNA was harvested by using a Gentra extraction kit followed by 70% alcohol precipitation as described in our previous report [23] . Susceptibility of KD. A total of 381 KD patients and 664 controls were included in this study ( Table 1 ). The distribution of CLEC5A genotypes was in accordance with the Hardy-Weinberg equilibrium for both cases and controls (Table 2) . However, none of the tSNPs was significantly associated with the genotype or allele frequency of the controls or KD patients under 3 genetic models (dominant, recessive, or allelic models) ( Table 2) . In this study, 37 patients (9.9%) had CAL formation and 49 patients (13.1%) had resistance to the initial IVIG treatment (Table 1) . However, no tSNPs were significantly associated with genotype or allele frequency in the KD patients with or without CAL formation (Table 3) . Additionally, the CLEC5A polymorphisms tested in this study failed to show any significant associations with genotype or allele frequency in the KD patients who showed a response to IVIG treatment (Table 4 ). Table 1 ), CAL formation (Supplemental Table 2 ) and IVIG treatment response (Supplemental Table 3 ) in the KD patients. However, none was significantly associated with the phenotype. The C-type lectin-like super domain (CTLD) family has diverse functions, and in particular, is important in innate immunity including nature killer (NK) function or pathogen recognition [25] . CLEC5A belongs to the Group V "NK cell receptors" family, and MDL-1 expression is upregulated in activated myeloid cells [26] and acts as a signaling receptor for proinflammatory cytokine and chemokine release [14] . Even though a number of reports have demonstrated that KD involves activation of a wide array of immunological elements such as T cells and macrophages [16] [17] [18] 27] , with the subsequent release of several cytokines [28] , only a few reports have addressed the role of lectin in the pathogenesis of KD. Several genetic associations with susceptibility to KD and CAL formation have been reported, but the results are inconsistent [29] [30] [31] [32] . Previous genetic association studies have indicated that the intronic SNP (rs28493229) of ITPKC, 1,4,5-trisphosphate 3-kinase C, reduces gene expression by altering splicing efficiency, and the C allele contributes to immune hyperreactivity in KD patients [29] . Recently, it has been demonstrated that rs28493229 is associated with susceptibility to KD and CAL formation [24, 29] . ITPKC is able to regulate the immune system via calcium-dependent NFAT pathways [29] . Similarly, previous studies have indicated that C-type lectin receptors (CLRs) are critical in the activation of the Syk-mediated NFAT signaling pathway [33] . In addition, CLEC5A has been shown to play a key role in host defense and to be involved in dengue virusmediated disease [14] . This finding suggests CLEC5A may be a potential target protein that involves calcium-dependent immune regulation and contributes to the development of coronary artery lesions. However, we did not find evidence to support a genetic role of CLEC5A in the pathogenesis of KD. Since we picked tagging SNPs from the HapMap database, only the tagging SNPs with a minor allele frequency of more than 10% were selected. Although our tSNP could capture majority of the underlying genetic variances with MAF > 10% across the CLEC5A gene, the rare causal genetic polymorphisms in CLEC5A may not have been detected in this study. Therefore, we cannot rule out or exclude rare causal genetic polymorphisms in CLEC5A. In addition, there are, at least, seventeen groups of CLRs in vertebrates. Indeed, it has been reported that mannose-binding lectin gene polymorphisms are associated with susceptibility to KD [34] and CAL formation [35] . Thus, large-scale DNA sequencing to CLR family is needed to better understand KD. In conclusion, this study showed for the first time that tSNPs of CLEC5A are not associated with susceptibility to KD, CAL formation, and IVIG treatment response in a Taiwanese population.

A Non-VH1-69 Heterosubtypic Neutralizing Human Monoclonal Antibody Protects Mice against H1N1 and H5N1 Viruses

Influenza viruses are among the most important human pathogens and are responsible for annual epidemics and sporadic, potentially devastating pandemics. The humoral immune response plays an important role in the defense against these viruses, providing protection mainly by producing antibodies directed against the hemagglutinin (HA) glycoprotein. However, their high genetic variability allows the virus to evade the host immune response and the potential protection offered by seasonal vaccines. The emergence of resistance to antiviral drugs in recent years further limits the options available for the control of influenza. The development of alternative strategies for influenza prophylaxis and therapy is therefore urgently needed. In this study, we describe a human monoclonal antibody (PN-SIA49) that recognizes a highly conserved epitope located on the stem region of the HA and able to neutralize a broad spectrum of influenza viruses belonging to different subtypes (H1, H2 and H5). Furthermore, we describe its protective activity in mice after lethal challenge with H1N1 and H5N1 viruses suggesting a potential application in the treatment of influenza virus infections.

Seasonal influenza causes up to 500,000 deaths worldwide each year [1] . Infants, immunocompromised individuals and the elderly are particularly susceptible, with 90% of deaths occurring in the latter group [2] . Influenza viruses can also cause pandemics that, although rare, are recurrent events historically associated with high levels of morbidity and mortality [3, 4, 5, 6] . Preventive vaccination has historically been the most efficient measure of influenza control, but this approach presents important limitations due to the accumulation of antigenic mutations in the virus, known as antigenic drift. Vaccines typically elicit a potent neutralizing antibody response limited to the specific viral strains included in the preparation and to closely related viruses [2] . For this reason, seasonal vaccines need to be annually reformulated based upon the forecasting of viral strains that will circulate in the coming influenza season. Furthermore, influenza vaccines have suboptimal immunogenicity and efficacy in the groups at highest risk of severe disease [7] . Moreover in the case of a pandemic, the use of vaccine is limited by the time required for its development and deployment [8] . The current therapeutic regimen for influenza A viruses is limited to two classes of drugs: the adamantanes (amantadine and rimantadine) and the neuraminidase inhibitors (oseltamivir and zanamivir). However, the natural and/or acquired resistance to these drugs has been reported [9, 10] . Resistance to adamantanes is prevalent among seasonal and avian influenza A viruses significantly reducing their usefulness [11, 12] . The sudden and widespread emergence of resistance to oseltamivir among prepandemic H1N1 viruses has raised further concerns over the current therapeutic options [13, 14] . Oseltamivir resistance was reported in patients infected with the pandemic H1N1 viruses and highly virulent H5N1 viruses [15, 16] . The resistance to zanamivir is rare [17] , but its use is limited to patients who can actively inhale it, which often excludes young children, impaired older adults or patients with underlying airway disease [14] , that is the groups of patients most vulnerable to serious influenza infection complications. Alternative strategies are needed to combat the constant threats posed by influenza. One of such strategies may come from passive immunoprophylaxis with monoclonal antibodies (mAbs) recognizing broadly conserved influenza epitopes and endowed with broad-range neutralizing activity [18] . The most important protective antigen on the surface of influenza virus is HA, whose structure can be divided in two distinct regions: the globular head, responsible for the binding to the sialic acid, and the stem region that contains the fusion peptide and the membrane anchor domain. On the globular head, constituted by the HA1 subunit, lie several epitopes targeted by neutralizing antibodies [18] . However, mAbs recognizing this region are of restricted application due to the antigenic drift that this region encounters [18] . In contrast, the stem region of HA, formed mostly by the HA2 subunit, is relatively conserved among different influenza A subtypes [19] and indeed could represent an universal target for the development of cross-neutralizing monoclonal antibodies. Several human heterosubtypic neutralizing mAbs, directed against HA stem region and with protective features in animal models, have been recently described [20, 21, 22, 23, 24, 25] . All these mAbs recognize epitopes located in the most conserved region of influenza viruses HA and neutralize influenza viruses by blocking fusion of the viral and the host endosomal membranes. Many of these anti-influenza heterosubtypic neutralizing mAbs utilize the VH1-69 germline gene and bind to a hydrophobic region on the HA stem using their complementary determining region 2 (CDR2). In this study, we describe a human monoclonal antibody named PN-SIA49 recognizing a highly conserved epitope on the stem region of HA and featuring one of the strongest in-vitro neutralizing activity described so far against a broad spectrum of viruses belonging to different influenza subtypes (H1, H2 and H5). Furthermore, we describe its protective activity in mice after lethal challenges with H1N1 and H5N1 viruses suggesting its potential as a broad-spectrum monoclonal antibody for treatment of influenza virus infection. It was previously described that the Fab fragment of PN-SIA49 binds to the stem region of HA and neutralizes all tested H1N1 isolates [26, 27] . The heavy chain variable region of PN-SIA49 uses the VH3-23 gene and is paired with a VL1-39 light chain. The nucleotide sequence homology of PN-SIA49 with the germline sequence is 93.11% for the VH gene and 92.30% for the VL gene, demonstrating its origin from a somatic mutation process. In this study, in order to better characterize its neutralizing activity, PN-SIA49 was produced as whole IgG1 molecule using the BD BaculoGold System. The whole IgG molecule was tested in fluorescence inhibition assay, infectious foci formation reduction assay and plaque reduction assay against human, swine and avian influenza A viruses belonging to phylogenetic group 1 (H1N1, H2N2, H5N1 and H9N2) and group 2 (H3N2 and H7N2). The results obtained showed that IgG PN-SIA49 has a stronger neutralizing activity compared to Fab PN-SIA49 (Table 1) [26, 27] . Indeed, IgG PN-SIA49 neutralized all tested viruses belonging to group 1, except the virus strain belonging to H9N2 subtype, with a half maximal inhibitory concentration (IC50) ranging between 0.1-1.9 mg/ml (Table 1 and Figure S1 ). On the contrary, PN-SIA49 showed no neutralizing activity against the viruses belonging to group 2 ( Figure S2 ). These data suggest that the epitope recognized by PN-SIA49 is conserved among group 1 viruses. Therapeutic efficacy of PN-SIA49 against H1N1 and H5N1 challenge in a mouse model To determine whether the in-vitro neutralizing activity displayed by PN-SIA49 would be predictive of its protective efficacy in-vivo, BALB/c mice were inoculated intranasally with 3 fifty percent lethal dose (LD50) of A/Wilson Smith/33(H1N1) (WS33) or A/Vietnam/1203/2004 (H5N1) (VN04) virus, and were treated 24 hours later with 10, 1 or 0.1 mg/kg of PN-SIA49. An anti-HCV/E2 mAb (e137) [28] was used as control at 10 mg/ kg. PN-SIA49 protected mice from lethal challenge with WS33 in a dose-dependent manner providing 100% protection (6/6) against death in animals that received 10 mg/kg of the antibody and 83.3% protection (5/6) in animals that received 1 mg/kg ( Figure 1C ). Consistent with the in-vitro neutralizing activity, PN-SIA49 at 10 mg/kg afforded 66.6% protection against lethal H5N1 virus challenge ( Figure 1D ) whereas control mice rapidly succumbed to infection by day 8 post-challenge (p.c.). Surviving mice remained healthy and showed minimal body weight loss (maximum weight loss: 7.2% in WS33 group, 14.4% in VN04 group) over the 2-week observation period. At the conclusion of the experiment, The mean body weight loss was 2.2% in the treated mice challenged with VN04 virus ( Figure 1B ) while mice infected with WS33 regained their full body weight ( Figure 1A ). To gain further insights into the kinetics of viral neutralization in vivo, 4 mice from each group were euthanized on day 4 p.c. and viral titer was determined in whole lung tissues. PN-SIA49 significantly reduced virus titer in the lungs of mice infected with WS33 by approximately 2,000 fold at 10 mg/kg and about 700 fold at 1 mg/kg ( Figure 1E ). A 10-fold reduction in pulmonary virus titer was noted in mice challenged with VN04 virus that received PN-SIA49 at 10 mg/kg ( Figure 1F ). Taken together, these results indicate that the survival was associated with an important reduction of the virus burden in the lungs of mice treated with PN-SIA49 and that the reduction is concordant with its in-vivo activity. In order to better define the HA region recognized by PN-SIA49, several approaches were used. Firstly, the hemagglutination inhibition (HI) activity for PN-SIA49 was evaluated and the resulting HI titre was 2.5 mg/ml and 0.039 mg/ml for VN04 and WS33 virus, respectively. Secondly, to evaluate if PN-SIA49 recognizes an epitope on the HA stem region, a competition assay on A/Puerto Rico/8/1934-HA (A/PR/8/34-HA) human epithelial kidney HEK293T transfected cells was performed between PN-SIA49 and the commercially available mouse monoclonal antibody C179 (Takara Bio inc., Otsu, Shiga, Japan), which binds to an epitope on the HA stem region [29] . The results obtained showed that PN-SIA49 completely inhibited C179 binding to the A/PR/8/34-HA ( Figure S3 ). Further evidence that PN-SIA49 binds to an epitope on the HA stem region is given by lack of protease susceptibility of the HA at low pH in the presence of PN-SIA49. Exposure to low pH followed by trypsin digestion results in degradation of HA. In contrast, when the HA is pre-treated with PN-SIA49, most HA is retained in a protease resistant, pre-fusion form ( Figure 2 ). Taken together, these preliminary data suggest that the epitope recognized by PN-SIA49 is localized in the HA stem region, but in close proximity of the HA globular head. Based on the PN-SIA49-C179 competition assay results, a large panel of HA mutants carrying an alanine substitution were generated [29, 30, 31, 32] . The binding of PN-SIA49 to these mutants was then evaluated by FACS analysis. Data obtained revealed that the binding of PN-SIA49 was decreased by His25Ala, Asn336Ala, Pro338Ala mutants on HA1 and Me-t360Ala, Asp362Ala, Gly363Ala, Trp364Ala, Thr384Ala, Va-l395Ala, Asn396Ala, Glu400Ala mutants located on HA2 (sequence numbering refers to A/PR/8/34, GenBank accession number ABO21709). Importantly, these residues are extremely conserved among viruses belonging to H1N1 subtype spanning from 1918 and 2009 and also highly conserved in H2N2 and H5N1 subtypes ( Figure 3 ). All the other mutations did not have any effect on the antibody PN-SIA49 binding to the HA ( Figure 3 ). To exclude the possiblity that the reduced PN-SIA49 binding to HA was due to reduced expression of HA on the cell surface, we performed a FACS analysis in which wild-type HA and mutants HA were stained with a mouse anti-influenza A HA (H1 subtype) monoclonal antibody. This antibody was directed against a linear epitope in order to evaluate the expression level for each HA on cell surface. As shown in figure S4 , the HA mutants are expressed on the cell surface at a similar levels to that of wild type HA. Consistent with the great phylogenetic distance, the amino acidic difference in positions 25, 360 and 362 between the tested viral strains belonging to H1N1, H2N2, H5N1 subtypes and the viral strains belonging to H9N2 subtype may at least partially explain the lack of neutralizing activity against the H9N2 strain. Based on the results obtained from the alanine scanning study, an in-silico analysis on the HA crystal structure of A/PR/8/34 (PDB ID code 1RU7) was carried out. The analysis confirmed that the residues identified lie on the stem region of HA, that they belong to the HA1 and HA2 subunits and that they are exposed on the surface of the HA molecule ( Figure 4 ). In this study, we characterize a human mAb, IgG PN-SIA49, directed against influenza viruses and previously described as Fab fragment [26, 27] . IgG PN-SIA49 neutralizes a broad spectrum of viruses belonging to different influenza subtypes (H1, H2 and H5) and is characterized by the lowest mean IC50 values described to date for a heterosubtypic mAb [20, 21, 22, 23, 24, 25] against the influenza subtypes tested in this study (Table S1 ). Furthermore, IgG PN-SIA49 is endowed with a stronger neutralizing activity compared to its monovalent molecule (Table 1) . Indeed, it is well documented in the literature that bivalency of IgG molecule may be an essential feature for the biological activity of a mAb, mostly due to an increase in antibody avidity [33, 34, 35, 36] . The increased avidity may also play an important role in the prophylactic and therapeutic application of an antibody, allowing the administration of smaller amount compared to the Fab fragment. Additionally, the whole IgG molecule has a longer half-life than its fragments, with a consequent prolonged effect of the molecule. For all these reasons, we used IgG PN-SIA49 to treat mice infected with H1N1 and H5N1 viruses and data obtained show that PN-SIA49 protects mice from H1N1 and H5N1 virus lethal challenge suggesting its potential application in treatment of influenza virus infections. Most of the anti-influenza heterosubtypic neutralizing mAbs described thus far, derived from the VH1-69 germline, bind to a conserved epitope in the HA stem region present only on group 1 influenza A viruses. The binding is mainly through highly hydrophobic amino acidic residues in the heavy chain CDR2 [20, 21, 22, 23, 24, 25] . Instead, the heavy chain variable region of PN-SIA49 results from VH3-23 gene rearrangement with the D3-3 and JH6 gene segments demonstrating that, despite the preferential usage of VH1-69 in the heterosubtypic response to influenza HA [20] , an in-vivo heterosubtypic protection may be conferred also by non VH1-69 derived Abs. PN-SIA49 recognizes a novel broadly neutralizing epitope on HA stem region, which is highly conserved among group 1 subtypes that have been confirmed in humans. In particular, this region is broadly shared among H1N1 isolates spanning from 1918 to 2009, H2N2 subtype responsible for the 1957 pandemic, and highly pathogenic and potentially pandemic H5N1 influenza virus. Finally, PN-SIA49 is protective in mice when given after lethal challenge with either H1N1 or H5N1 virus. This suggests its great potential as broad-spectrum monoclonal antibody for in vivo treatment of influenza virus infections. Taken together, these data underline the importance of PN-SIA49 for the development of anti-influenza strategies based on passive immunization. Further studies will be necessary to define the most effective prophylactic and therapeutic administration protocols. Finally, these data and a more detailed definition of the epitope recognized by PN-SIA49 could also be useful to develop new vaccination strategies able to elicit a humoral immune response directed against key regions of the influenza HA protein. This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of . Amino acidic sequence conservation in hemagglutinin groups and subtypes at the region bound by PN-SIA49. Circles below residues indicate PN-SIA49 percentage binding to each HA alanine mutant compared to binding to the wild-type HA: red 25% binding; yellow 50-75% binding, blue 100% binding. Sequence numbering is based on H1N1 A/Puerto Rico/8/1934 coding region (GenBank accession number ABO21709). Subtypes that can be neutralized by PN-SIA49 are indicated with a green '+' on the left, while the ones that can not be neutralized are indicate with a red '2'. a Recombinant HA from H1N1 A/South Carolina/1/1918 pandemic strain was previously shown to be bound by PN-SIA49 [26, 27] . b H1N1 A/New Caledonia/20/1999 was previously shown to be neutralized by PN-SIA49 as Fab fragment [26, 27] . doi:10.1371/journal.pone.0034415.g003 We previously described the molecular cloning of a human monoclonal antibody Fab fragment named PN-SIA49 [26, 37] . An anti-influenza A antibody directed against H1N1 subtype HA, named RB62, and an anti-HCV E2 glycoprotein antibody, named e137 [28] , produced and purified with an identical procedure were used as controls in all experiments. The following human H1N1 and H3N2 reference strains were acquired from the American Type and Culture Collection , 100 mg/ml of streptomycin (Gibco Invitrogen, Carlsbad, CA, USA) and 2 mg/ml TPCK-trypsin (Roche Applied Science). The A/swine/Parma/1/97 isolate was analogously grown on Newborn Swine Kidney (NSK) cells, kindly provided by the Zooprophylactic Institute of Brescia, Italy. At 80% confluence, cells in MEM supplemented with 2 mg/ml serum-free TPCK-trypsin, were infected with each strain at a MOI of 0.001. After 1 hour of infection, cells were washed with phosphate buffered saline (PBS); MEM supplemented with 2 mg/ml TPCK-trypsin was then added and cells were incubated at 37uC in 5% CO2 atmosphere. Cells were observed daily to monitor the cytopathic effect and, usually after 72-96 hours, the supernatant was collected, centrifuged at 1000 rcf for 10 minutes to eliminate cells debris and filtered with 0.22 mm filters (Millipore, Billerica, MA, USA). The supernatant was then aliquoted and stored at 280uC as cell-free virus. well). Serial dilutions, 10 mg/ml-0.03 mg/ml, of IgG PN-SIA49 were preincubated for 1 hour at 37uC with 100 median tissue culture infective doses (TCID50) of virus. Following incubation, 100 ml of the mix antibody-virus were added to the cells and incubated for another hour at 37uC in 5% CO2. At the end of this incubation, cells were washed with PBS and 100 ml of MEM TPCK-Trypsin (2 mg/ml) were added in each well. Cells were incubated for 7 hours at 37uC in 5% CO2 and then washed with PBS, fixed and permeabilized with ice-cold ethanol. Cells were incubated with anti-influenza A mouse antibody (Argene, Shirley, NY, USA) for 30 minutes at 37uC in a humid chamber. The cells were then washed with PBS and incubated for 30 minutes at 37uC in a dark humid chamber with a FITC-conjugated secondary antibody (Argene, Shirley, NY, USA). Nuclei staining was obtained with Hoechst 33342 (Sigma Aldrich). An infection control without antibody was included, as well as a negative control with the anti-HCV/E2 antibody e137. Each neutralization assay was performed in triplicate and repeated in two different sessions. The neutralization activity for each antibody concentration was expressed as the percentage reduction of fluorescent nuclei compared with the nuclei count in the infection control. Nuclei counting was performed by using the GE Healthcare's IN Cell Analyzer 1000, an automated epifluorescence based microscope system. The neutralization curves were then fit by non-linear regression with the GraphPad Prism software, allowing IC50 calculation. The A/PR/8/34 (H1N1) and A/Milan/UHSR1/2009 (H1N1) viruses were also tested in plaque reduction assay as previously described [26] . Briefly, neutralizing assays were carried out in 6 wells plates using MDCK cells (56105 cells/well). Two dilutions, 1-0.1 mg/ml, of IgG PN-SIA49 were preincubated for 1 hour at 37uC with 100 TCID50 of virus. Following this incubation, 1 ml of each virus-antibody mix was added on MDCK monolayer and the plate was incubated 1 hour at 37uC in 5% CO2. Then, the medium was removed and the monolayer washed twice with PBS. Two ml of MEM-agarose 0.8% supplemented with penicillin (50 mg/ml) (Gibco Invitrogen, Carlsbad, CA, USA), streptomycin (100 mg/ml) (Gibco Invitrogen, Carlsbad, CA, USA), L-glutamine (2 mM) (Gibco Invitrogen, Carlsbad, CA, USA) and trypsin (2 mg/ml) (Roche Applied Sciences) were added to each well and the plates were incubated 48 hours at 37uC in 5% CO2. After this incubation, the agarose medium was removed from each well and 1 ml of 70% methanol-crystal violet 1% (w/v) was added to each well at room temperature. Finally, the wells were washed with tap water and dried. An infection control without antibody was added as well as a negative control with anti-HCV/E2 e137 mAb. The neutralization was determined counting the PFU reduction in presence of antibodies compared to the infection control. Infectious foci formation reduction assay. The following H1N1, H2N2, H3N2, H5N1, H7N2 and H9N2 viruses were . Each viral isolate was titrated to establish working dilution that produces 15-30 foci forming units per well in 96 tissue culture plates. Neutralizing assays were carried out in 96 wells plate using MDCK/SIAT-1 cells. Serial dilutions, 30 mg/ml-0.37 mg/ml, of IgG PN-SIA49 were preincubated for 1 hour at 37uC with the subset of viruses. Following this incubation, 100 ml of the antibody-virus mix was added to the cells and incubated for another hour at 37uC in 5% CO2. At the end of this incubation, the cells were washed twice in PBS and 100 ml of virus growth media containing 2 mg/ml of TPCK treated trypsin was added. Cells were incubated for 12-16 hours at 37uC in 5% CO2 and then washed with PBS, fixed and permeabilized with ice cold methanol/acetic acid (95:5) for 30 min at 220uC. Cells were incubated with anti-NP antibodies (Millipore, Billerica, MA, USA) for 30 minutes at 37u. The cells were then washed and incubated for 30 minutes at 37uC with a mouse HRP-conjugated secondary antibody. True Blue chromogenic substrate (KPL) was used to count the number of foci. Female BALB/c mice were purchased at 6 to 8 weeks of age from Charles River Co. (Wilmington, MA). All mice were maintained in specific pathogen-free barrier facilities. All animal experiments and procedures conformed to protocols approved by the Centers for Diseases Control and Prevention (CDC), Atlanta, GA, USA. For each virus, four groups of 10 mice were inoculated intranasally with 3 LD50 of A/Wilson Smith/33 or A/ Vietnam/1203/2004 virus in a 50 ml volume. At 24 h after inoculation, graded doses (10, 1, 0.1 mg/Kg) of PN-SIA49 or the control antibody (e137, 10 mg/Kg) were administrated to mice by intraperitoneal injection in a final volume of 0.2 ml. A subset of six mice in each group were weighed on the day of virus challenge and then observed and weighed every 2 days for 2 weeks after inoculation. Mice that lost more than 25% of their initial body weight were euthanized. A subset of four animals treated with mAbs were euthanized on day 4 after inoculation, and whole lungs were homogenized in 1 ml of sterile PBS. Virus titers in lung tissue homogenates were determined by plaque titration in MDCK cell monolayer cultures. Hemagglutination (HI) assay. HI tests using mAb PN-SIA49 or e137 control antibody against live WS33 and VN04 viruses were performed according to standard protocols [38] . Briefly, serial dilutions of purified mAbs in PBS were performed from initial concentration of 5 mg/mL. Positive and negative control ferret sera were diluted initially 1:10 in receptor-destroying enzyme from Vibrio cholerae (Denka Seiken, Tokyo). Serial dilutions of control sera or mAbs were pre-incubated with 4 HA units of virus per well. For WS33 virus, turkey red blood cells (RBCs) were added to a final concentration of 0.5%, whereas horse RBCs were used at a 1% suspension for VN04 virus. Normal ferret serum gave a value of less than 10. Specific HI activity of mAbs was calculated as the lowest concentration of mAb that displayed HI activity. Protease susceptibility assay. Each reaction contained 3 ml of the anti-influenza vaccine season 2011-2012 (InflexalV-Crucell), which contains 30 ng of A/California/7/09 (H1N1) HA or 3 ml of the anti-influenza vaccine season 2011-2012 combined with 2 fold molar excess of PN-SIA49. Titron X-100 was added to prevent aggregation of the post-fusion HA. The pH was lowered in all samples except controls using citric acid 0.1 M pH 3. Reactions were mixed, briefly centrifuged and incubated at 37uC for one hour. After incubation, reactions were equilibrated to room temperature and the pH was neutralized by addition of 1 M Tris, pH 9. The actual pH reached was determined in parallel using larger buffer volumes without protein. Trypsin was added to all samples except controls at a final ratio of 1:25 by mass and samples were digested overnight at 37uC. Non-reducing SDS buffer was added to each reaction. Samples were boiled for ,2 minutes and loaded on a nonreducing 4-15% polyacrilamide pre-casted gel (Biorad, Italy). After running, samples were transferred on a PVDF membrane (PerkinElmer, Belgium) for 2 hours at 350 mA. The membrane was then blocked with 5% not fat milk in PBS-Tween20 0,1% (PBST) for 1 h at room temperature and then washed three times with PBST. PN-SIA28, a human anti-HA monoclonal antibody recognizing HA0 [39] , was used as primary antibody at 1 mg/ml in 5% not fat milk-PBST. The membrane was incubated for 1 hour at room temperature and then was washed three times with PBST. Secondary anti-human antibody was added and incubated for 1 h at room temperature. After incubation the membrane was washed, the substrate solution (SuperSignalH West Pico Chemiluminescent Substrate, PIERCE) was added and incubated for 2 min. To determine the pH required to convert the HA to the postfusion form, pH titrations using the assay describing above was performed. Samples were exposed to a range of pH conditions (pH 4.5, 4.9, 5.3, 5.7, 6.1, 6.5, 7 and 8), neutralized and processed as described above. A/Puerto Rico/8/1934 (H1N1) hemagglutinin (A/PR/8/34-HA) was amplified as previously described [26, 37] using the following PCR primers: APR834_fw:CACCATGAAGGCAAACC-TACTGGTCCTGTTATGTG. APR834_rev:TCAGATGCATATTCTGCACTGCAAA-GATCCATTAGA. The PCR products were cloned into the pcDNA 3.1D/V5-His-TOPO vector (Invitrogen, Carlsbad, CA, USA). Subsequently, HA alanine mutants were generated using Gene Tailor Site-Directed Mutagenesis System (Invitrogen, Carlsbad, CA, USA). A total of 20 A/PR/8/34-HA mutants were generated (His25Ala, His45Ala, Thr315Ala, Asn336Ala, Ile337Ala, Pro338Ala on the HA1 subunit and Trp357, Thr358, Gly359, Met360, Ile361, Asp362, Gly363, Trp364, Thr384, Ile388, Thr392, Val395, Asn396, Glu400 on the HA2 subunit. Sequence numbering refers to A/PR/8/34, GenBank accession number ABO21709). The binding activity of PN-SIA49 was assayed using full-length wild type and mutants HAs. Human epithelial kidney HEK293T cells (ATCC CRL-1573) were transfected in 6 wells plate (Corning, Corning, NY, USA) (16106 cells/well) with 4 mg of pcDNA 3.1D/V5-His-TOPO vector containing the HA nucleotide sequences described above. After centrifugation and fixation with 4% paraformaldehyde for 15 minutes at RT, the transfected cells were incubated for 30 minutes at room temperature with PN-SIA49 or conformational controls for H1N1 (RB62) at 10 mg/ml. Additionally, the isotype control, e137 (10 mg/ml) was introduced as well as untransfected cells and a mouse anti-influenza A HA (H1 subtype) monoclonal antibody (GeneTex Inc., Irvine, CA, USA) directed against a linear epitope to evaluate the transfection efficiency and the expression level for each HA. The cells were then washed with PBS and incubated for 30 minutes at room temperature with FITC-conjugated anti-human (Sigma Aldrich) or anti-mouse (Argene, Shirley, NY, USA) antibody. Afterwards, the cells were washed with PBS and analyzed by FACS. The FACS data were analyzed using the software Weasel w 2.5 (Waler+Eliza Hall, Institute of Medical Research, Parkville Victoria, Australia). The binding of PN-SIA49 to the different HA-mutants was then expressed as a binding percentage compared to wild-type. The data showing the PN-SIA 49 binding decrease between H1N1 wild type HA and H1N1 HA mutants, were obtained normalizing each PN-SIA 49 binding value to corresponding anti-H1 expression control values. For the competition assay, serial dilutions of PN-SIA49 were used in combination with a fixed concentration (1 mg/ml) of mouse monoclonal antibody C179 (Takara Bio inc., Otsu, Shiga, Japan) which binds to an epitope on the HA stem region [29] . For sequences analysis the following software packages were used: SeqScape (Applied Biosystems), ClustalX (Toby Gibson), Bio Edit (Tom Hall, Ibis Therapeutics) and Treeview (GubuSoft). For molecular visualization and rendering UCSF Chimera package from the Resource for Biocomputing Visualization and Informatics at University of California, RasMol (Roger Sayle), Jmol (Jmol: an open-source Java viewer for chemical structures in 3D. http:// www.jmol.org/), Cn3D (United States National Library of Medicine, NLM) were used. Finally for data analysis and graphical editing GraphPad Prism was used. Figure S1 Neutralization assays against group 1 influenza viruses. Dose-response curve fit nonlinear regression is reported for IgG PN-SIA49 against neutralized group 1 influenza viruses. (A) Results from fluorescence inhibition assays, (B) plaque reduction assays and (C) infectious foci formation reduction assays. Data from at least two different experiments for each virus are reported. Each point was performed in triplicate. (PDF) Figure S2 Influenza hemagglutinin unrooted phylogenetic tree of all the viral strains tested in neutralization assays with PN-SIA49. Viral isolates belonging to group 1 and group 2 are divided into two different boxes. Subtypes that can be neutralized by PN-SIA49 are indicated with a green '+', while the ones that cannot be neutralized are indicate with a red '2'. As reported in the text, PN-SIA49 is able to neutralize all of the group 1 viruses tested in this study except for the H9N2 strain. No neutralizing activity was detected against the H3N2 viruses tested. * The recombinant HA from A/South Carolina/1/1918 (H1N1) pandemic strain was previously shown to be bound by PN-SIA49 [26, 27] . # H1N1 A/New Caledonia/20/1999 was previously shown to be neutralized by PN-SIA28 as Fab fragment [26, 27] . (PDF) Figure S3 C179/PN-SIA49 competition assay. Graphic representation of cell staining and flow cytometric analysis of HEK293T cells transfected with the pcDNA 3.1D/V5-His-TOPO vector containing the HA-A/PR/8/34 were performed. Serial dilutions of PN-SIA49 were used in combination with a fixed concentration (1 mg/ml) of C179 (blue line). A monoclonal antibody directed against the HA globular head was used as competition negative control (pink line). (PDF) Figure S4 HA mutants that determine a decrease of PN-SIA49 binding to HA are expressed at the same level of wild type HA on cell surface. FACS curves showing the binding of anti-H1N1 HA antibody (directed against a linear epitope) to untransfected cells, HA wild-type and HA-mutants. White and red curves represent, for each graph, respectively the binding of anti-HA expression control to untransfected cells and wild type H1N1-HA. The different colour curves represent the different mutants. (PDF)

Web-based Investigation of Multistate Salmonellosis Outbreak

We investigated a large outbreak of Salmonella enterica serotype Javiana among attendees of the 2002 U.S. Transplant Games, including 1,500 organ transplant recipients. Web-based survey methods identified pre-diced tomatoes as the source of this outbreak, which highlights the utility of such investigative tools to cope with the changing epidemiology of foodborne diseases.

T he epidemiology of foodborne illnesses is influenced by a variety of factors, some of which have changed dramatically in recent years. The increased availability of preprocessed foods and the improved survival of persons with immune defects have affected the sources and nature of foodborne illness (1) (2) (3) (4) . Increased mobility of Americans through interstate travel has complicated the identification and investigation of outbreaks. New technologies for outbreak investigation have the potential to greatly assist public health officials in successfully managing these changing factors. We describe an outbreak of Salmonella enterica serotype Javiana infections affecting a large group of geographically dispersed organ transplant recipients. The prompt and successful investigation of this outbreak was facilitated by the use of Web-based surveys. On To identify additional cases, state health departments were asked to report any S. Javiana isolates with a PFGE pattern indistinguishable from the outbreak strain. To develop hypotheses about potential sources of infection, we conducted in-depth telephone interviews with several persons who were identified with culture-confirmed illness. On the basis of interview results, we conducted a Webbased cohort study among Transplant Games attendees to identify risk factors for infection by using eQuest, a software package developed by the Centers for Disease Control and Prevention (CDC) that allows rapid development of Web-based surveys (5) . Using email addresses provided to us by the Transplant Games organizers, we electronically distributed a message on July 20 to attendees (including athletes and spectators), requesting that they complete an outbreak survey. We included information about salmonellosis and its treatment and provided a link in the email to the secure Web site containing the outbreak survey. In each survey respondent's household, we collected information for a single person visiting Orlando, regardless of whether he or she had been ill. A case was defined as fever or diarrhea with onset between June 25 and July 7 in a person who visited Orlando. Submitted answers were automatically stored in a secure electronic database and linked only to a random survey number. To identify the specific food item responsible for illness, we performed a Web-based case-control study. On July 31, we distributed a survey containing detailed questions about specific food items available in theme park A to persons who had responded to the first survey. Casepatients were questioned about food items eaten in the 3 days before illness onset. Controls were defined as well survey respondents, and all were questioned about the middle 3 days of the Transplant Games (June 26-28). Plant X, the processing plant that supplied tomatoes to theme park A, was inspected on August 13. Molecular subtyping of confirmed S. Javiana isolates was performed at state public health laboratories (6) . Diced Roma tomatoes from unopened boxes that had been stored frozen at theme park A were cultured at the Florida State Public Health Laboratory. Statistical analyses were conducted by using SAS software version 8.2 (SAS Institute Inc., Cary, NC, USA) to calculate odds ratios (OR) and 95% confidence intervals (CI). Multivariable logistic regression analyses were conducted for variables that were significantly associated with illness. Through laboratory surveillance, 21 additional S. Javiana infections with indistinguishable PFGE patterns were identified in 10 states, for a total of 23 identified culture-confirmed cases. Dates of illness onset were from June 24 to July 8. Of 22 patients for whom travel information was collected, 19 reported visiting theme park A in the last week of June; 16 visited theme park A but did not report any contact with the Transplant Games, which suggests a true outbreak of S. Javiana infections among visitors to the theme park. An electronic link to the Web-based cohort study survey was distributed on July 20, 2002, to 1,100 Transplant Games attendees. Among these 1,100, we received survey responses from 369 persons (34%) in 42 states; 80% responded within 48 hours. Of the 369, a total of 82 (22%) reported illness and 41 (53%) were female. The median age of ill respondents was 47 years (range 4-71 years); 48 (59%) were transplant recipients. Dates of symptom onset were June 26-July 7 (Figure) . Predominant symptoms included diarrhea (93%), abdominal pain (79%), and fever (51%). Three respondents (4%) had been hospitalized. No deaths were reported to the organizers of the Transplant Games or CDC. Among ill respondents, 75 (91%) reported eating food items at specific food courts in theme park A. The Web-based case-control study was distributed on July 31 to the 369 persons who responded to the first survey. By August 2, a total of 222 persons (60%) responded. Of 217 valid responses, 41(19%) were ill persons who met the case definition; the remaining 176 were healthy controls. Ill persons were significantly more likely to report eating dishes containing diced Roma tomatoes than were well persons (44% of ill vs. 15% of well, OR = 4.3, 95% CI 2.1-9.1). Other food items that were significantly associated with illness on univariate analysis were dishes containing shredded iceberg lettuce (OR = 3.7, 95% CI 1. 8-7.4) , pre-shredded cheddar cheese (OR = 2.9, 95% CI 1.5-5.9), fresh ground beef (OR = 3.0, 95% CI 1.4-6.4), and pre-sliced beefsteak tomatoes (OR = 4.6, 95% CI 1.1-19.4) ( Table) . In multivariable logistic regression modeling, only diced Roma tomatoes remained independently associated with illness at the 0.05 significance level. Diced Roma tomatoes were supplied to theme park A from plant X, where whole Roma tomatoes were mechanically diced and washed in a manually chlorinated recycled water tank. Levels of chlorine in the tank were variable (≈1.5-3.5 ppm free chlorine), providing potential opportunity for the amplification of any existent microbial contamination. Review of invoices showed that diced Roma tomatoes used in food courts patronized by persons with outbreak-related illness were processed at plant X from June 20 through July 3. No diced tomatoes from the implicated lots were available for testing. Microbiologic evaluation of an unopened box of plant X diced Roma tomatoes processed on July 12 indicated the presence of fecal coliforms (150-1,000 CFU/g). The nature of this outbreak highlights several changing features of foodborne disease epidemiology, including the enhanced mobility of persons through air travel, an increasing reliance on pre-processed foods, and an expanding immunocompromised population at risk. Through a Web-based investigation, we were able to rapidly identify the source of this outbreak and inform an immunocompromised population of its potential risk for illness. Our approach allowed us to contact and question several hundred geographically dispersed persons in a matter of days. Survey respondents' answers were automatically stored in a secure electronic database, eliminating the need for data entry. With the development of questionnaire templates, a public health official could select sets of questions and precoded answers from pull-down menus and modify them to design an outbreak-specific, Web-based questionnaire that is automatically linked to an electronic database (5) . These methods make it increasingly possible to design and post a Web-based investigative tool for wide distribution within hours. Our investigation has several limitations. Use of a Web-based investigation tool limited responses to only those Transplant Games attendees with known email addresses and Internet access. The initial response rate to our survey was only 34%; households with ill persons may have been more likely to respond to our Web-based survey. However, most (>75%) respondents to both surveys were from households in which no one had experienced illness, which provided us with a sufficient number of responses from both well and ill persons to identify the source of the outbreak. Hospitalized and severely ill persons may have been too sick to respond to the survey or may have been unable to access the Internet, which limited our ability to calculate accurate hospitalization or attack rates among persons attending the Transplant Games. Although isolation of S. Javiana from plant X diced tomatoes would have strengthened our findings, the tomatoes tested were processed well after the outbreak period, when levels of contamination may have differed considerably. As use of the Internet becomes more widespread for participation in regional, national, and international conferences, groups, and listservs, electronic mail cohorts are becoming more commonplace. The development of Webbased public health investigative tools can facilitate future investigations of outbreaks affecting geographically dispersed persons who may be part of an electronic mail cohort. A Web-based approach to data collection can also play a critical role in rapidly sharing data in outbreaks involving multiple jurisdictions (7) . The growing use of Web-based technologies in public health investigations will have to be balanced with the need to protect the privacy of personal information in the online environment (8, 9) . The careful application of emerging technologies and conventional epidemiologic techniques can help public health officials effectively cope with the multitude of changing factors that shape public health in the United States.

Dengue Fever, Hawaii, 2001–2002

Autochthonous dengue infections were last reported in Hawaii in 1944. In September 2001, the Hawaii Department of Health was notified of an unusual febrile illness in a resident with no travel history; dengue fever was confirmed. During the investigation, 1,644 persons with locally acquired denguelike illness were evaluated, and 122 (7%) laboratory-positive dengue infections were identified; dengue virus serotype 1 was isolated from 15 patients. No cases of dengue hemorrhagic fever or shock syndrome were reported. In 3 instances autochthonous infections were linked to a person who reported denguelike illness after travel to French Polynesia. Phylogenetic analyses showed the Hawaiian isolates were closely associated with contemporaneous isolates from Tahiti. Aedes albopictus was present in all communities surveyed on Oahu, Maui, Molokai, and Kauai; no Ae. aegypti were found. This outbreak underscores the importance of maintaining surveillance and control of potential disease vectors even in the absence of an imminent disease threat.

D engue viruses cause a wide range of illness, including dengue fever (DF), dengue hemorrhagic fever (DHF), and dengue shock syndrome (DSS). Four dengue serotypes, known as DENV-1, -2, -3, and -4, can cause severe and fatal disease. Dengue typically occurs in tropical and subtropical areas in the world and is transmitted by Aedes mosquitoes; Aedes aegypti is the principal vector worldwide (1) . DF and DHF are the most important arboviral diseases of humans; ≈50-100 million dengue infections and several hundred thousand cases of DHF occur annually (2) . The first large-scale dengue fever epidemic in Hawaii occurred in the late 1840s; a second outbreak occurred at the turn of the century, with an estimated 30,000 cases (1, 3) . During those periods Ae. aegypti was widespread in Hawaii (4) . Epidemic dengue occurred again on Oahu in 1943 to 1944, when 1,498 infections were reported, mostly in urban areas of Honolulu (5) . Ae. albopictus had been introduced into Hawaii at the beginning of the century, and by 1940 it was the dominant day-biting Stegomyia mosquito species in the islands (4, 5) . After the Second World War, no confirmed autochthonous dengue infections were reported in Hawaii. Nevertheless, dengue illnesses were occasionally identified among travelers to Hawaii who had been infected overseas. The annual number of imported cases was low, with 20 infections recorded during the 10-year period from 1991 through 2000 (P. Effler, unpub. data). On September 12, 2001 , the Hawaii State Department of Health (HDOH) received a call from a physician in Hana, Maui, who had seen a patient with febrile illness and rash 1 week earlier. The physician indicated that several of the patient's family members had become symptomatic; none had a history of recent foreign travel. On investigation by HDOH staff, dengue fever was suspected, and clinical specimens were collected and forwarded to the Centers for Disease Control and Prevention (CDC) for diagnosis. On September 21, CDC confirmed recent dengue infection in the index patient. We report the results of an investigation into the first outbreak of dengue fever in Hawaii in 56 years. From September 23 to 28, 2001 , HDOH contacted all licensed physicians in the state by email or facsimile to request that any patient with a denguelike illness (DLI) be tested for dengue, regardless of travel history. DLI was defined as fever or chills plus 2 or more of the following symptoms (6): myalgia, headache, arthralgia, eye or retroorbital pain, rash, or hemorrhagic manifestation (e.g., petechiae, hematuria, hematemesis, menorrhagia, melena). On September 24, 2001 , active surveillance was established at 51 clinical settings across the state. All acute-care hospitals and major clinics were contacted daily to determine the number of clinically compatible illnesses seen in the previous 24 hours and to arrange for laboratory evaluation of suspected cases. Although HDOH recommended dengue testing only for patients meeting DLI criteria, it was performed whenever requested by a physician. HDOH staff interviewed persons with suspected dengue infection to obtain symptom and travel histories. Visits to residences and work sites were conducted. Patients' household contacts or co-workers with a history of illness were urged to be tested for dengue. All clinical laboratories in Hawaii were asked to report any requests for dengue diagnostic testing and to forward aliquots of serum samples obtained for dengue testing to the HDOH State Laboratories Division. Laboratory analyses to detect anti-dengue immunoglobulin (Ig) M and IgG and to isolate and identify the virus were performed by methods previously described (7) (8) (9) (10) (11) (12) . RNA was extracted by using QIAmp Viral RNA Mini kits (Qiagen GmbH, Hilden, Germany). Sequencing was performed by using the Taq DyeDeoxy Terminator Cycle Sequencing kits (Applied Biosystems, Foster City, CA, USA). Sequencing products were cleaned by using agarose gel electrophoresis and silica gel adsorption (Qiagen PCR purification columns) and analyzed on an ABI PRISM 377 DNA sequencer (Applied Biosystems). Sequences were assembled and aligned with Lasergene software (DNAStar, Madison, WI, USA), and phylogenetic trees were generated with PHYLIP v. 3.5c (University of Washington, Seattle, WA, USA). Laboratory-positive recent dengue infection was defined as a person who had 1) dengue virus isolated from serum, 2) a positive dengue IgM antibody test result, or 3) a positive IgG antibody test result in a person initially tested for dengue >60 days after onset of DLI and who was epidemiologically linked to another person with recent dengue infection identified by virus isolation or positive IgM serologic test result. Persons were classified as negative for dengue infection if they had at least 1 specimen collected 6-60 days after illness onset that was IgM negative or a first specimen collected >60 days after illness onset that was IgG negative. Persons were classified as indeterminate for dengue infection if all specimens were collected <6 days after illness onset and were negative for virus isolation and for anti-dengue IgM. Imported dengue was defined as illness in a person with laboratory evidence of recent dengue infection and a history of international travel within 14 days of illness onset. During the outbreak investigation, a CDC entomology team conducted spot checks of potential breeding sites in 29 communities (at least 20 sites per community) on all islands except Hawaii and Lanai. From March to May 2002, HDOH vector-control staff placed ovitraps at 295 sites throughout the state; local vector-control staff relied on prior experience to select sites with known populations of day-biting mosquitoes. In both surveys, larvae were collected from breeding sites and identified to species. In the second survey, eggs were reared to the fourth larvae or adult stage before speciation. Adult mosquitoes attracted to humans were also captured and identified at many of these sites; in the outbreak areas; landing counts were obtained by recording the number of mosquitoes landing on a stationary person during a 5-minute period. Univariate analyses were conducted by using EpiInfo Version 6.4c (CDC, Atlanta, GA, USA). A difference in proportions was considered significant if the chi-square p value was <0.05. From September 12, 2001 , to April 30, 2002, a total of 1,644 persons in Hawaii without a history of recent foreign travel were tested for possible dengue infection. Of these, 122 (7%) had laboratory evidence of a recent dengue infection: 15 (12%) were positive by virus isolation; 99 (81%) had anti-dengue IgM; and 8 (7%) had a history of DLI, anti-dengue IgG, and an epidemiologic link to a patient with recent infection (Table 1) . Testing was indeterminate for 422 (26%) persons, and the remaining 1,100 (67%) did not have dengue infection. The median age was 41 years (range 1-77), 35 years (range 0-89), and 29 years (range 0-81) for persons who were laboratory positive, negative, and indeterminate for dengue infection, respectively. Autochthonous dengue infections were identified on 3 of 6 islands (Table 1) . Exposures on Maui, Oahu, and Kauai accounted for 76%, 21%, and 3% of all recent dengue infections, respectively. Eighty (66%) of the laboratory-positive infections were from persons who stayed in the Hana area of Maui, an area with <2% of the island resident population (Figure 1 ). On Oahu, 20 (77%) of the infections occurred among residents of 2 nearly adjacent communities on the windward side with a combined population of 25,709 (<3% of the island's total). The heavily affected areas of Maui and Oahu both have thick vegetation and heavy precipitation (average annual rainfall >177 cm/year, 4 times the annual rainfall in Honolulu). The outbreak spanned >8 months, with a peak incidence in late September 2001. (Figure 2 ) The first suspected dengue illness was reported with an onset date September 5, 2001; subsequent investigations identified an additional 31 laboratory-positive patients with illness onset before that date, and the earliest was May 27, 2001 . Of laboratory-positive cases, 89% met the clinical criteria for DLI (Table 2) . Patients with recent dengue infection reported a greater number of symptoms than those who did not have dengue. One or more hemorrhagic manifestations were reported in 42 (34%) persons with dengue infection. Myalgia, chills, arthralgia, and rash were significantly more common among patients with laboratory-positive dengue infection than in persons with negative or indeterminate results. No cases of DHF or DSS, as defined by the World Health Organization, were reported, and no deaths occurred (14) . Three patients with laboratory-positive dengue infection were hospitalized for their illness. Eighty-one (66%) of the recent infections were initially reported by physicians treating acutely ill patients, while the remaining 41 (34%) were identified through HDOH field investigations. Thirteen household clusters accounted for 53 (43%) of the 122 patients. One-hundred and fifteen (95%) of the 122 persons with laboratory-positive infection were residents of the state of Hawaii. All 7 visitors with dengue stayed at rental properties in the Hana area of Maui. Another 70 nonresidents with possible dengue infections who visited Hawaii during the outbreak were reported to HDOH; 30 of these nonresidents were serologically tested, and results for all were negative. From January 1, 2001, to April 30, 2002, a total of 43 cases of imported dengue infection were reported to HDOH ( Figure 3) . Oahu had the greatest number of imported infections (31 infections), followed by Maui (6 infections), Hawaii (4 infections), and Kauai (2 infections). Eighteen (42%) of the imported dengue infections were from the Society Islands, 13 (30%) were from American or Western Samoa, 7 (16%) were from the Philippines, and 1 each was from Cambodia, Easter Island, Indonesia, Thailand, and Vietnam. Imported dengue peaked in July and August 2001; exposures in the Society Islands accounted for the largest proportion of cases during this time (n = 9, 47%). All 15 dengue virus isolates obtained from patients with exposure in Hawaii were DENV-1. Phylogenetic analysis of envelope glycoprotein sequences showed that the Hawaiian isolates belonged to a group composed primarily of Pacific Island isolates from recent years (Figure 4 ). High bootstrap values showed the Hawaiian isolates were associated more closely with contemporaneous Tahiti and subsequent Easter Island isolates than with a 2001 isolate from American Samoa. In entomologic surveys conducted during the outbreak, Ae. albopictus was present in all 29 communities surveyed on Oahu, Maui, Molokai, and Kauai, but no Ae. aegypti were found at any site. In drier areas, on the leeward sides of the islands, container indices were high (>50%), but landing rates were generally low. However, in Nahiku, a small community in densely vegetated woodland near Hana, Maui, that was heavily affected during the outbreak, adult Ae. albopictus populations were high, with landing rates of 70 to 90 mosquitoes per person in 5 minutes. In the surveys conducted at 300 sites in 2002, Ae. albopictus larvae were ubiquitous on all islands, including Lanai and Hawaii, but Ae. aegypti was only found in 3 communities in the southern part of the island of Hawaii. This report describes the first outbreak of dengue fever in Hawaii since the mid-1940s. Understanding the factors that contributed to the reemergence of dengue after such a prolonged absence and to the cessation of transmission will help public health authorities develop future prevention and control strategies. At the time of the 2001 Hawaii outbreak, a large DENV-1 epidemic was occurring in the Society Islands, 4,400 km south of Hawaii. More than 33,000 dengue illnesses were recorded in the Society Islands from February to November 2001, and of the 1,400 persons hospitalized, DHF was diagnosed in 45%, and 20% had symptoms of DHF or DSS. Ae. aegypti was identified as the vector (15, 16) . Virologic and epidemiologic data strongly suggest that the Hawaii dengue outbreak was directly linked to the one in French Polynesia. Travelers are a potential source for dengue outbreaks; many epidemic introductions are thought to result from the arrival of a single viremic person into an Ae. aegypti-or Ae. albopictus-infested area (17) . DENV may have been introduced to Maui when a group of >30 persons from Hana visited Tahiti during April-May 2001. One of the travelers (patient A) became ill shortly after returning to Hana and later tested positive for anti-DENV IgM and IgG. Patient A was a close associate of the first known autochthonous case-patient in the Hawaii outbreak, whose illness onset occurred ≈2-3 weeks later. Although patient A may have been the source for the Hana outbreak on Maui, available information suggests that additional separate virus introductions led to independent foci of autochthonous cases on the other 2 affected islands. In Kauai, only 1 of 4 dengue case-patients had any known exposure to persons from Maui. Moreover, the first identified case-patient in Kauai shared accommodations with a person in whom a febrile illness developed shortly after the patient returned from Tahiti. On Oahu, none of the 26 confirmed infections could be epidemiologically linked to exposures on Kauai or Maui. Furthermore, during an investigation of an autochthonous cluster on Oahu, the likely index patient was as an IgM-positive family member who had a DLI 4 days after returning from a trip to Tahiti. Ae. albopictus was the vector responsible for the 2001 Hawaii outbreak. Both entomologic surveys support that Ae. albopictus is ubiquitous, often common on all the islands, whereas Ae. aegypti is restricted to a few small foci on the relatively sparsely inhabited island of Hawaii. Several factors may explain why the outbreak in Hawaii followed a much different course than the concurrent epidemic caused by an apparently similar DENV-1 strain elsewhere in the Pacific. First, differences in mosquito species, behavior, and ecology are critical to understanding why the Hawaii outbreak was less severe than that described in the Society Islands, where Ae. aegypti was the principal mosquito vector. Ae. aegypti females are highly anthropophilic and often feed on several persons before obtaining enough blood to complete a gonotrophic cycle. This tendency towards multiple feeding may contribute to the explosive nature of dengue outbreaks in areas where Ae. aegypti is present. Compared with Ae aegypti, Ae. albopictus is considered to be an inefficient epidemic dengue vector because it is less anthropophilic and not as well adapted to urban domestic environments (18) . Ae. albopictus will readily feed on humans, but usually only on a single person, and it also feeds on other animals, which decreases the probability of human contact (19, 20) . Lifestyle factors may also help explain why Hawaii's dengue outbreak was limited (21) . Residences in many affected areas often had dense, uncultivated vegetation near housing and, not infrequently, an abundance of items that could serve as suitable Aedes breeding sites: tires, buckets, and discarded vehicles. Furthermore, dwellings in these areas often lacked window screens and doors. The combination of ample mosquito breeding sites and relatively unrestricted access to residents in some sections of windward Oahu and Hana, Maui, probably enhanced opportunities for mosquito-human contact beyond levels that existed in Hawaii's major population centers. Public health measures may also have helped mitigate the spread of Hawaii's outbreak. This response consisted of 4 simultaneous, integrated initiatives: 1) enhanced surveillance to detect new foci of transmission; 2) rapid education of healthcare providers to improve the diagnosis and treatment of dengue; 3) health promotion activities directed toward the general public, including visitors; and 4) vector-control efforts, which included a combination of source reduction activities, limited use of larvicides, and area spraying (Appendix). Worth noting is that most of the illnesses in the Hawaii outbreak were mild, given that an apparently similar DENV-1 strain caused a major epidemic of DHF and DSS in French Polynesia. One possible explanation for the difference in illness severity observed between these locations is that the number of cases in Hawaii was too small to manifest the extremes of the clinical spectrum. A second explanation is that a history of dengue infection, i.e., antibody-dependent enhancement, may have been important in French Polynesia (22) . A third explanation is that the Hawaiian virus had changed genetically and became less virulent or lost its epidemic potential. This loss of epidemic potential occurred in the 1970s when both DENV-1 and DENV-2 were reintroduced into the Pacific after an absence of 25 years (23) . Despite close similarities in the envelope protein sequences of the 2001 Tahiti and Hawaii viruses, important differences may exist in other areas of the genome that could influence these properties. Recent studies in Sri Lanka and Puerto Rico suggest that the genetic changes associated with epidemic potential occur in the nonstructural virus genes and not the envelope gene commonly usually used to show genetic relatedness between dengue viruses (24, 25) . Full-length genomic sequencing of DENV-1 viruses is pending. The Hawaii experience demonstrates the potential of Ae. albopictus, under suitable conditions, to transmit small outbreaks of dengue within the United States. During the last 15-20 years, this mosquito has expanded its geographic range within the United States and now is found in at least 24 states on the mainland (26, 27) . From 1986 to 2000, a total 516 laboratory-confirmed and 2,128 suspected dengue infections were imported into the United States (28) (29) (30) (31) (32) (33) . The true incidence of imported dengue infection is probably higher, since dengue may often go undiagnosed in areas where the virus is not endemic (4, 20, 23, 34, 35) . Given the high volume of travel between the US mainland and dengue-endemic areas of the world (an estimated 14 million passengers to and from the Caribbean, Central and South America, and Oceania in 2001), we recommend that health officials keep local clinicians informed of dengue activity in these regions and that clinicians consider the possibility of autochthonous transmission when evaluating febrile rash illnesses, particularly when local vector surveillance indicates high populations of Ae. aegypti or Ae. albopictus mosquitoes (36,37). This investigation has several limitations. First, despite extraordinary efforts to obtain specimens, ≈25% of all persons initially evaluated for dengue did not submit a convalescent-phase specimen (>5 days after illness onset) required for definitive case classification. During followup attempts to obtain convalescent-phase sera, we often learned that patients or their physicians had decided that dengue was unlikely and no further testing was necessary; however, some dengue infections may have been missed. Secondly, because persons acquire dengue from mosquitoes that feed during the daytime, infection might have occurred at a location other than where the patient lived. We mapped the distribution of residences, however, because this information is not subject to recall bias. Thirdly, when investigating newly reported cases, we did not routinely elicit the number of household members and obtain serum samples from them in a standardized manner. Therefore, we cannot calculate the proportion of close contacts who were infected. The Hawaii dengue experience is another example of how readily pathogens can cross great expanses of ocean to cause outbreaks in new territory (1, (38) (39) (40) . Important lessons learned from this episode include the need to closely monitor and respond to disease developments in the global community and the need to maintain surveillance and control of potential disease vectors even in the absence of an imminent disease threat. samples, 4) creating a patient-tracking system, and 5) notifying all state epidemiologists though Epi-X to identify any possible dengue cases exported from Hawaii. Provider education included 1) issuing medical alerts to physicians, 2) conducting grand rounds and other lectures on dengue at local medical centers, and 3) distributing CDC video tapes on dengue diagnosis and treatment to physicians. Health promotion efforts included 1) issuing frequent press releases, including daily case counts and messages about eliminating mosquito breeding sites around the home; 2) giving multiple news interviews by HDOH staff with radio, television, and print media; producing public service announcements by HDOH for radio and television; 3) conducting joint town meetings by HDOH and Department of Education health educators; 4) distributing >600,000 dengue brochures through high-volume stores and other venues; 5) developing a dengue education Web site, which provided the public and officials with information on the latest developments; 6) distributing educational brochures to Maui rental car agencies and hotels; and 7) establishing checkpoints along the Hana Highway staffed by public health nurses and others who distributed educational materials and mosquito repellent. Vector control efforts included 1) inspecting private and public properties for mosquitoes, larvae, and potential breeding sites; 2) conducting door-to-door source reduction campaigns by HDOH staff and community volunteers in Hana and windward Oahu; and 3) treating >2,500 residences statewide with insecticides or larvicides.

Community responses to communication campaigns for influenza A (H1N1): a focus group study

BACKGROUND: This research was a part of a contestable rapid response initiative launched by the Health Research Council of New Zealand and the Ministry of Health in response to the 2009 influenza A pandemic. The aim was to provide health authorities in New Zealand with evidence-based practical information to guide the development and delivery of effective health messages for H1N1 and other health campaigns. This study contributed to the initiative by providing qualitative data about community responses to key health messages in the 2009 and 2010 H1N1 campaigns, the impact of messages on behavioural change and the differential impact on vulnerable groups in New Zealand. METHODS: Qualitative data were collected on community responses to key health messages in the 2009 and 2010 Ministry of Health H1N1 campaigns, the impact of messages on behaviour and the differential impact on vulnerable groups. Eight focus groups were held in the winter of 2010 with 80 participants from groups identified by the Ministry of Health as vulnerable to the H1N1 virus, such as people with chronic health conditions, pregnant women, children, Pacific Peoples and Māori. Because this study was part of a rapid response initiative, focus groups were selected as the most efficient means of data collection in the time available. For Māori, focus group discussion (hui) is a culturally appropriate methodology. RESULTS: Thematic analysis of data identified four major themes: personal and community risk, building community strategies, responsibility and information sources. People wanted messages about specific actions that they could take to protect themselves and their families and to mitigate any consequences. They wanted transparent and factual communication where both good and bad news is conveyed by people who they could trust. CONCLUSIONS: The responses from all groups endorsed the need for community based risk management including information dissemination. Engaging with communities will be essential to facilitate preparedness and build community resilience to future pandemic events. This research provides an illustration of the complexities of how people understand and respond to health messages related to the H1N1 pandemic. The importance of the differences identified in the analysis is not the differences per se but highlight problems with a "one size fits all" pandemic warning strategy.

During an influenza pandemic the lack of experience of dealing with such hazards increases public reliance on information from government, public health agencies, employers, the community and the media. A challenge for government and health agencies is sustaining public awareness and alertness over a protracted period [1, 2] . Responding to the challenges posed by managing the risk from relatively unknown health hazards requires recognition of the fact that promoting sustained action involves not only getting the information right but also ensuring that it is communicated in ways that accommodate diversity in community characteristics, needs and expectations. The lack of experience also calls for research to provide the evidence base health authorities need to respond effectively in circumstances (e.g. rapid onset of disease outbreak) that precludes learning in situ. There is evidence that greater levels of perceived susceptibility to and perceived severity of the disease and greater belief in the effectiveness of recommended preventative and avoidant measures are important predictors of behavior [3] [4] [5] . It is important to accommodate the fact that disbelief in the effectiveness of measures can result in people failing to act and developing distrust of sources of information [6] . Previous New Zealand research identified a general belief that local agencies will manage a pandemic well and that New Zealand is a relatively safe place to be in the event of a pandemic. However, there was also a lack of trust in information providers as well doubt that the health system would cope [4] . Trust is a crucial component of effective risk communication when people face uncertainty and levels of trust are correlated with people's perceptions of the integrity of information. Lack of trust can be exacerbated by scepticism about the veracity of health risk warnings and the view that media are sensationalist and untrustworthy [7, 8] . Being uncertain about an outbreak, and thinking it, and its consequences had been exaggerated have been associated with a lower likelihood of behavioural change [5] . Increasing trust is a function of the degree to which agencies engage with and empower communities [6] . Lack of trust in authorities may also affect how people process and interpret health messages and advice, increase concerns and interfere with the way that the risk messages are interpreted and acted on [9] . Transparency and honest communication where both good and bad news is conveyed can empower the public to make their own decisions [1] . The public are more likely to take appropriate action and accept the recommendations if they have been involved in the decision-making process, and the quality of the relationship between authorities and the community has a direct effect on the uptake of risk messages, and trust in the message providers [2] . The primary objective of this study was to provide health authorities with evidence-based practical information to guide the development and delivery of key health messages for H1N1 and other health campaigns. The study focused on community responses to key health messages in the 2009 and 2010 H1N1 campaigns. The study was part of a rapid response initiative; therefore focus groups were selected as the most efficient means of data collection in the time available. Eight semi-structured focus groups were recruited between May and July 2010 (the New Zealand winter season) comprising 7 to 13 participants each and lasting approximately 1 hour. Separate focus groups were conducted for each of the target groups with a total of 80 participants representative of five target populations groups identified in consultation with Ministry of Health staff: Māori, Pacific Peoples, children (or parents of children), general population, and vulnerable people with chronic conditions (defined as those who are eligible for subsidised vaccinations, such as pregnant women, those with diabetes, using asthma inhalers, with heart disease or kidney problems. Participant characteristics are summarised in Table 1 . Purposive sampling methods were used to ensure the sample met the criteria specified by the Ministry of Health. The period of data collection was selected to coincide with the period in which people are most aware of flu and the need for preventive and protective actions. Participants were recruited using posters in a community library and through contacts within key agencies (e.g. university research centres, Pacific health services, district health boards, and the Ministry of the Social Development). Staff of the Research Centre for Māori Health and Development, Massey University were responsible for the Māori components of the research to ensure a culturally appropriate methodology given the time restraints of the study. In contrast to New Zealanders of European descent, Māori possess different cultural characteristics. For example, based on cultural dimensions [10] Māori score higher on dimensions such as individualism-collectivism, power distance and uncertainty avoidance compared with their counterparts. Because these differences have significant implications for social influences on behaviour and for the nature of the relationships that exist between community members and health authorities, it was essential to ensure that data were collected in culturally sensitive and competent ways. Hence Māori were treated differently from the perspective of the data collection approach adopted. This is consistent with effective cross cultural research methods [11] . All focus group sessions (except Māori groups) were recorded and independently, professionally transcribed. Transcriptions were mainly verbatim, with verbal padding and hesitations omitted. Apart from the facilitators, specific individuals were not identified in the transcripts or any subsequent reports. The extracts used to illustrate the content of each theme are identified by codes which correspond to the focus group transcripts from which they were taken ( Table 1) . The analysis of the focus group data (excluding Māori groups) was undertaken by a single researcher who was neither present at the focus groups nor had read any preliminary findings. This work was verified by the focus group facilitators to ensure that any "contextual richness" had not been missed in the data. Thematic analysis was used to identify themes and concepts across the entire data set (6 transcripts) to "identify repeated patterns of meaning" [12] . The process involved working through the six phases of thematic analyses as identified by Braun and Clarke [12] . This study was approved by the Massey University Human Ethics Committee: Southern A (ref 10/32, 21 May 2010) . Written informed consent was obtained from all participants. The final coding scheme consisted of 17 themes which were grouped into four main categories: risk, building community understanding, responsiveness and information preferences. Transcript extracts were selected on the basis of their relevance to the theme under discussion and all identifying information has been removed. People's perception of risk is a product of the perceived likelihood of a pandemic and its perceived consequences. The analysis identified that people made judgements about both likelihood and consequences. Understanding these inputs helps tailor risk communication strategies. Previous research identified a general belief that New Zealand, as a result of factors such as geographic isolation, reduce the likelihood of a pandemic and thus increase a belief that New Zealand is a relatively safe place to be in the event of a pandemic [4] . Perceived risk was also attenuated by a belief that border control agencies will prevent pandemic flu entering New Zealand, with this possibly resulting in people transferring responsibility for preparing from themselves to border control agencies [6] . "It's pretty much a one-border control place. That's why that's better. We're lucky, because we've got only that one contact point coming in." (G3) Other participants were aware that our perceived geographical isolation does not in fact translate to a decrease in risk. In fact H1N1 arrived in New Zealand early in the global pandemic of 2009 with students returning from a school trip to Mexico and the USA. "Well we think that we're different because we're far away. But actually, if you think of how people travelled here, it's the biggest factor for it always, because everyone who comes here comes in an aeroplane, pretty much. And they come from everywhere." (G3) Discussion reflected variable levels of awareness of and reaction to pandemics. Many regarded the 2009 H1N1 pandemic as an overreaction that was not taken very seriously by many people. The following extract illustrates the concept of "normalisation bias" in which people extrapolate from the current experiences (H1N1) to define what a future pandemic would look like and therefore underestimate their future risk which makes them less respective to current health risk messages [6] . "I thought it was scaremongering, personally. I thought the way it was handled was quite interesting, and I think it also made a lot of people very paranoid about things that sometimes people don't need to be paranoid about." (G3) For some participants it was the use of term "pandemic" that gave rise to the sense of overreaction by suggesting something more serious than what actually occurred. "It became reasonably clear reasonably quickly last time that hundreds and thousands and millions weren't dying. Even when they kept on sort of saying things were happening, and then you saw the numbers, it just didn't add up." (G1) There is evidence that higher levels of general anxiety and perceptions of high risk severity are related to a greater likelihood of carrying out preventive and avoidant protective behaviours [3, 5] . However, it is important to note that this relationship is only present when people with high risk perceptions also know what to do to manage their risk. If people do not know what to do, or question the veracity of the advice offered, they are more likely to respond by denying the risk or transferring it to others [13] . Whilst most of the participants in the general population groups regarded swine flu as "bit of a joke, really" others reported feelings of anxiety, fear or panic. In particular, reports of flu-related deaths gave rise to more serious concern. For those who felt that geographical remoteness provided some form of protection, it was not until cases were reported within New Zealand that the threat became more real. This shift in thinking is important as the perceived relevancy and immediacy of risk affects the decision to act or not to act on information [4, 8] . "Because you kind of go, well, we're just little old New Zealand, where nothing happens. You know, we'll be right. And it wasn't until you actually heard that people in New Zealand had brought it back from overseas. And that's when you really does go "Ooh! Alright." (G3) Concern about news reports from overseas also related to the reliability of the information in terms of both its trustworthiness and relevance to "us". 'The other thing, people are a bit cynical about the whole thing. Because there are these people dying in Mexico, you think, well, there are people sick in hospital. You know, I've got no idea what the health system's like there." (G2) Such comments reflect a belief "it won't happen to me because -it happens to others". Through the process of "othering", individuals focus on differences in others, effectively creating a separation between "us" and "them" [14] . By projecting the risk of infection and death onto "them" the sense of powerlessness and vulnerability is reduced for "us" [15] . Cynicism about the veracity of media reports was not limited to news from overseas. Indeed there was a strong feeling that the New Zealand media had a central role in the "overhyping" of the 2009 and 2010 pandemic risk. Media are credited with amplifying the risk perceptions in such a way that risk communicated may not be an accurate reflection of the true risk [7] . "I think the news media have a lot to blame, because they want to make news. They sensationalise stuff. That's not helpful." (G2) Public distrust in journalists and the sensationalising of health related stories can also be a hindrance to taking the risk seriously and of undertaking precautionary measures [5] . A belief that risk has been exaggerated is associated with an increased sense of helplessness and frustration and a reduction in the likelihood that people will prepare in the short-term [5, 7, 16] . In an attempt to understand potential risks people situate risk-knowledge in historical and local contexts [17] . This can be seen in the focus group discussions in which participants were more concerned about other risks, including illnesses such as heart disease, cancer, meningitis, and respiratory disease. Much of the discussion in the general population groups focused on their reassuring themselves that the pandemic was indeed an overreaction and a "media beat-up" as they believe that this had happened before with other events. In this condition, people preferentially seek information that reinforces this belief and this may reduce the likelihood of their attending to public information. Maintaining a sceptical and cynical perspective is a means of "distancing" themselves from the threat. "Like the bird flu that was... that, I think, killed some... a few people in Canada. And people said it was going to be the next big flu, and it wasn't." (G2) There is evidence that individual perceptions of risk are important determinants in undertaking preventive, and/ or protective behaviours for events such as pandemics as long as people know how to manage their risk [3, 5, 12, 21] . Some participants appeared to have a sense of "personal invulnerability" based on their history of rarely having experienced flu's or colds. The comments were also indicative of the phenomenon of unrealistic optimism. This means that while people accept the existence of a risk to the community in general, they see themselves as less vulnerable or more capable than others. This bias results in their transferring risk from themselves to others and seeing risk communications as applying to others rather than to themselves [6] . This bias represents a significant constraint on the effectiveness of risk communication. However, encouraging active discussion of pandemic issues in community groups can reduce its influence. "I've been fortunate enough to... I don't know that I've even actually ever had the flu, even ordinary flu, in my life... So I seem to have a bit of a natural immunity to it, luckily enough, so... yeah." (G2) In contrast there were others who expressed a heightened sense of risk due to their personal circumstances and health history. Assessing their personal risk in this way supports the view that people's understanding of risk is developed not only through cultural and sub-cultural membership, but also through personal experience [17] . This highlights the importance of risk communication encouraging people to personalize information. While all groups were aware of the concept of "high risk" groups the relationship between knowledge of "high risk factors" and individual perceptions of risk was less clear. Although some participants identified themselves as "high risk" they appeared to be uncertain about what this meant for them. Others used "othering" to minimise their own perception of risk by differentiating themselves from those who they identified as "high risk". This social construction of boundaries of "self" and "other" and their relationship to boundaries of "safety" and "danger" are particularly relevant to understanding notions of health and disease [18] . However, the self/other divide can be used to facilitate preparing [13] . By considering whether something can be done to assist those more vulnerable, people are more likely to also consider what they can do for themselves. Recall of key health messages was varied, however most participants were aware of hygiene self-efficacy measures such as hand washing, sanitiser use, covering of coughs and sneezes, and staying at home. This was particularly strong in one of the Pacific Peoples groups and for many appeared to be translated into action. "Mainly they tell you to wash your hands....Cover your mouth when you cough.......And don't share hankies, they say, yeah." (P2) There was some recall of the 2009 posters and the messages which they contained. Although some who recalled the posters expressed concern that they were inaccurate and overused. There was generally less recall of the 2009 television advertisements, and even less recall of the messages they were conveying. Overall, participants did not feel that they were better prepared or had changed their behaviour as a result of the information which they recalled from 2009. Where lessons had been learnt from the previous year, these were mostly related to improved personal hygiene measures. Awareness of the 2010 campaign was scant and what was recalled tended to be from commercial advertising, for example, from private commercial companies promoting flu vaccinations and household hygiene products. Participants reported receiving pandemic information from a variety of media sources including newspapers, TV, radio and the internet. However the primary source of information for participants was their workplace and/ or community. This differs from previous research in which Google was listed as a primary source of information [8] and television was the preferred means of receiving information during a pandemic [19] . The general population groups tended to report workplace as a key source of information, including workplace intranets. In contrast there was general agreement in the Pacific Peoples groups that their primary sources of information were the community. This included social networks and family, regular forums and meetings, church groups and health centres. This supports the view that when faced with uncertainty, people turn to others to reduce their uncertainty and guide their preparation; often these are family and friends, but also health agencies with whom they have a direct relationship [2] . This highlights the importance of both recognising the existence of diverse communities that facilitate developing understanding and engaging with these communities to ensure that information can be tailored to meet the needs, goals and expectations of each group. "And even in the church. Because one thing in the church is, our people fear God, and they always go church no matter what. And the pastor is one of the key people that talks into the community." (P2) The 2010 campaign had largely gone unnoticed by Māori tribal elders (kaumātua). They did not believe that information was readily accessible, no one had seen articles in the local press regarding H1N1 and pamphlets and posters "were not freely available". This view was mirrored in the young Māori mothers (Tamariki Ora) focus group. The kaumātua group felt that information is best disseminated to places of work, school and family and to Māori health providers to ensure coverage of the Māori population. It appeared that the messages in the media did not make an impact with Tamariki Ora mothers. Previous research has shown that community participation and trust in emergency management agencies played significant roles in increasing community preparedness, willingness to take responsibility for own safety, risk acceptance and satisfaction with communication [2] . Public are more likely to take appropriate action and accept the recommended actions if they have been engaged in all aspects of the risk management and decision-making processes through mechanisms such as focus groups or forums in ways that empower people to take action [6, 20] . Kaumātua were of the view that the 2007/08 campaign had been successful because the District Health Board had come out into the community and involved them in planning or providing information, but "it had not been effectively followed up on". A number of other comments also highlighted the importance of engaging with communities and disseminating information through community mechanisms. "Civil Defence needs to be proactive, and actually make sure that they've got the right community people, and the right community organisations on the board." (G3) With respect to workplace pandemic or disaster response plans, most participants had little or no knowledge of these. Some were aware of general plans or the existence of emergency supplies at work. There was a general belief in one group that emergency preparedness was seen as an "individual responsibility" by their employers. The apparent lack of clearly-articulated workplace response strategies is consistent with previous research [20] . As with previous research [2, 20, 21] few participants had stocked up on emergency supplies or prepared for the pandemic any other way. Some individuals had stocked up on food and essential supplies and/or had a family disaster plan. "I've certainly thought about how would we get food, or how much food did we have in the house, if we were to get quarantined." (G1) While some participants reported that they had already prepared emergency boxes, others were yet to act on their intentions to do so; it was on the list of things to do. Participants were aware of advice to stockup on pharmaceutical products but a number found the array of over the counter products available very confusing. Perceived or actual economic impact influences psychological and behavioural responses [22] . Many participants expressed concern that the cost of emergency kits could be a barrier for low income families and that some sort of financial assistance should be available, particularly as they may believe that the costs of acting will only incur a benefit in the event of a pandemic and this may not occur until some future date. "And the thing is that it would affect... I mean, imagine someone on a really base-level income... isn't going to have a preparedness kit. So they're the ones that are going to suffer, through in some ways, no fault of their own." (G2) Attitudes towards having the flu vaccination varied greatly as did uptake. Discussion concerned H1N1, H5N1, pandemic flu and seasonal influenza. There were those who routinely have them and those who "don't do flu shots". Uptake rates ranged from none in the Tamariki Ora mothers group to over 90% of those present in the kaumātua group. For some participants the decision whether or not to have the vaccination was related to their perception of risk. "Yes. I normally do have it. But I insisted on having it early, even though the GP was only giving it to people who were in vulnerable groups, because of the fact that I was travelling ..." (G1) There were others who, having identified themselves as being in an "at risk" group, still chose not to be vaccinated. For these and others it appeared to be a balancing act between their perceived risk of influenza against the perceived risks associated with the vaccine itself. The "costs", psychological and health "benefits", and expected outcome are all important issues that influence people when making a decision about whether to act on advice about a pandemic [2] . "I'm kind of stuck on that one. I've certainly thought more about it this year, as to whether I should just take the risk; but I'm not going to. I'm still not having one." (G1) Several of the Tamariki Ora mothers were apprehensive and confused about the flu injection and wanted more information about "immunisation for their children". "Will the current flu injection help us to stay immune for several years?" (M2) Whilst some participants elected not to have the vaccination even if it was free, for others cost was a significant issue. There was also a degree of cynicism about the vaccination and flu treatment amongst those that reported not having had one. In particular cynicism was expressed with respect to Tamiflu ® . "The whole Tamiflu... again, it makes me cynical, you know. Somebody was making a heck of a lot of money out of that, you know?" (G2) The reluctance to be vaccinated and the cynicism illustrated by these extracts is consistent with research showing that decisions to engage in preventive and avoidant behaviours is influenced by attitudes towards public health interventions [9] including having confidence in the efficacy of the behaviour [3] . It is worth noting that the latter beliefs influence the level of trust in health agencies and that specifically advising people (about all preparedness measures and not just antivirals) about why specific preparations are required increases the likelihood of adoption and helps maintain trust in health agency sources of information [12] . Participants had heard the social isolation/distancing message. "If you're sick, go home. And if you're sick and you're at home, stay there....That's right. I'm quite vocal about that anyway. I hate seeing people sick around me." (G1 with agreement from several other participants) Although participants recognised that isolation is an important response strategy, the economic pressures to go to work instead of staying at home was a major concern. "That's a big push, yeah, that's a big reason why people still go out, even though they know they have a cough. ...Yeah, that pushes me (into employment). ......You send your kids to school sick. .........Yeah, because you haven't the time, yeah." (P2) This is consistent with previous research showing that perceived or actual economic impact is a major factor in decisions around avoiding the flu [9, 19, 21, 22] . These issues can be compounded by not preparing, being unaware of workplace policies and plans (e.g., policies about wage payments if people are advised to stay home). Kaumātua believed should a person develop flu-like symptoms they should isolate themselves from other members of the community, however they were concerned that when individuals are isolated they may not be contacted by members of the community or whānau. Concerns were also expressed by this group about observing specific cultural practices and greeting protocols, reflecting the role that cultural and sub-cultural membership has in people's understanding of risk [17] . Participants wanted guidelines about who should stay home and when, and they wanted backing from their employers with respect to this issue. Participants felt that they had been given contradictory information about when to stay home and when to go to work or school which left them feeling uncertain about what to do. Consistency of advice is a significant important factor in communications from key agencies [3] . Information preferences "Knowing the difference" between swine flu and other flu emerged as a significant issue. This reflects previous reports of a strong desire from the public for symptom details about influenza [19] and the finding that public information about signs and symptoms are beneficial to public understanding of a pandemic [8] . Participants across all groups wanted to know about the specific swine flu symptoms which they could use to identify it and protect themselves from possible infection. There is a view that surveillance combined with good scientific information and operational research is crucial in limiting the spread of H1N1 [23] . A number of participants were concerned about testing and monitoring. They felt that the Ministry should not have "cut off monitoring quite so early" and that the failure to test for the H1N1 after health services became inundated with patients undermined the advice that the Ministry of Health was giving. In addition to the desire for symptom details many participants wanted concrete facts, such as how many people were diagnosed with or dying from swine flu. Some participants wanted to know percentages; others preferred the information to be given as numbers because they found percentages confusing. For many participants it was not necessarily numbers they wanted but information that helped them judge the "seriousness" of the pandemic and their level of personal risk. There was general agreement in a Pacific Peoples group that simplicity in the framing of messages was important. This aligns with the contention people can absorb only a small amount of information at a time and have difficulty understanding some kinds of information. Risk communication should take this into account and identify the most critical facts [24] . It is apparent from previous research that trust in authorities and satisfaction with communications received are associated with compliance of preventive, avoidant, and management behaviours [2, 3, 22] . People want the truth, even if it is worrisome, so honesty is crucial [24] even if that meant being told "we don't know, at this stage". "Give it to us the way it is. Because I'm sure adults are capable of dealing with that information, and then, you know, making their own choices later of how they deal with the information, but to actually be given that information, without any drama, and yet not being, you know, pushed under the rug somewhere." (P3) While some participants expressed confidence in the organisations providing information, others felt that they were not being given all the facts and that this affected their ability to make informed decisions. Trust in the information given is important because it affects the perceived credibility of risk assessments from authorities which in turn can influence response behaviours [3] . It is important to note that trust can be easily lost if people believe that agencies are not acting in their interests or do not provide information that meets their needs and once lost, trust is difficult to regain [6] . "Well, I'm always dubious about the... particularly the death rates that... that was not real... from what I understand, a lot of people that died had pre-existing conditions." (G2) Transparency and honest communication where both good and bad news is conveyed empowers the public to make their own decision [1] and that openness of government communication and acknowledging uncertainty is important for fostering trust [3] . Consistency of advice also appears to be an important factor in communications from key agencies [3] . Participants provided a number of examples where conflicting information or advice led to a feeling of confusion and frustration and loss of trust in key sources. "...at the beginning of the swine flu, the communication breakdown between the hospital and the local GP services. Because people were coming to the GPs, and they were referring them to the hospital, and then they were... (saying "No, we only take emergencies"), there's a lot of confusion." (P1) Dissatisfaction with health providers was not limited to poor inter-agency communication. Just getting an appointment with a GP was difficult for some and others had concerns about how they were treated and the advice which they were given. The issue of financial impact is also relevant to seeking medical assistance with a number of participants reporting that the high cost of a doctor's appointment was definitely a deterrent to seeking treatment. The issue of trust is illustrated by this extract from a woman who believes she was misdiagnosed in the emergency department putting vulnerable family members at risk. "I don't think she knew what I had. I just think she wanted to go tick, "Goodbye, here's the antibiotic." ... and I believed her. And that's how naive I was. I should have gone again, but I thought, "Oh no, she's got the ticket. She's got the certificate that says 'doctor'." I trusted her. ... when I came here further, I had the swine flu, I just wanted to get out and go back up there and find that little lady, and taking her and her certificate, and bang her up the side of the head and say, "You could have taken out my granddaughter, my daughter, and my father, because you were in too much of a hurry to go tick, tick, tick." (P2) Focus group participants expressed a desire for practical information to guide their responses to the influenza threat. Such views are consistent with previous reports of research participants preferring risk messages that empower with information about actions that could reduce risk and/or mitigate consequences [4, 5, 12] . People want to know how to protect themselves and their families during an influenza pandemic and the ability to act provides a sense of relief that "they could do something" [8] . It has been reported that participants who received little or no information about protective actions they could take, expressed helplessness and frustration [8] , Language preference has been shown to be an important factor in satisfaction with risk communications [4] . Participants in the Pacific Peoples groups stated strongly that messages should be communicated in an appropriate language. "Sometimes our older folk, they don't understand English. They have to be in original languages. The diversity of the Pacific -I mean, for some of our older people, because it's their cultural background, and it's so hard for them to understand." (P2) The acceptance of public health messages can be affected by factors such as socio-cultural behaviours, gender roles, generational differences, religious beliefs and language preferences [9] . The following extract also supports the argument that emotions can also cloud people's decision making, so communicators must treat audiences respectfully [24] . In order to accommodate these cultural and demographic factors it is important to work through communities. "Sometimes I don't think it's just not only their language....it's the way you use your language. You speak too fast, your English words are beyond me. And they're not dumb, the people. .... Sometimes it's just the way you... your tone. If you talk to them like they're dumb, well they'll just...They'll shut up." (P2) The following comment from a kaumātua also supports the argument that key health information and advice must take into account factors such as socio-cultural behaviours, spiritual beliefs and language preferences. In the event of an outbreak or pandemic, it would be difficult for kaumātua not to observe the protocols that are very much part of their traditional practices. If there was a outbreak we would have been ok, we were concerned at the lack of knowledge of tikanga Māori and it is not usual to have to stay away from marae or to stay at home, the thought of mass burials in a pandemic was culturally irresponsible." (M1) Discussions around preferences for the timing and frequency of key health messages were contradictory. On one hand participants, especially those in one Pacific Peoples group expressed a desire to be given early information through frequent messages. Other participants however, felt that they were being "bombarded" with too much information too soon. Participants wanted to be warned about potential risk well in advance to allow time for them to prepare. This supports the view that occasional media reports are insufficient to adequately inform individuals about pandemic preparedness, and interventions are needed before a pandemic occurs to improve public awareness, promote effective coping responses and help in the successful implementation of plans [25] . In contrast others thought it would be better if information was given much closer to the actual event. This is consistent with previous research that found participants preferred "just in time" delivery of information to avoid having to think about pandemic influenza unless they had to or unless a pandemic was imminent [8] . "Just-in-time" messaging that included technical terms, risks, health benefits and protective actions has been shown to help align public perception with realistic assessments of pandemic threat [24] . Both of these perspectives reflect problems with people's risk beliefs. Those who desire advance warnings may be unaware that a pandemic could be in New Zealand very quickly and possibly before its existence is formally identified. They may not have the time they expect to prepare. Those who adopt a "just in time" approach may overestimate their capacity to prepare in a short time frame (e. g., food supplies in supermarkets being rapidly depleted and not restocked). "I don't know we should say anything unless there's a clear danger, because otherwise you just get used to it. I mean you ignore it." (G1) The extract above illustrates the serious problem that arises from warning fatigue. Too many early warnings can result in cynicism, disengagement and a decline in trust [26] . Warning fatigue is more likely to result from potential threats with a 'long-lead-time'like pandemics, where there are many warnings in the absence of the actual threat. People "get sick of hearing it" and are likely to "switch off" and ignore future warnings. "But I think part of it is, that we've had so many health scares in the past decade. Like SARS, and different types of flu, and it's all this big media hype thing....and then it just kind of disappears." (G3) Focus group discussion about other information campaigns highlighted differences between what participants considered to be effective communications and from who they wanted to hear important messages. Some advertisements were seen as very effective because they were "straight-up, to the point" and included people to whom the participants could relate. "Because they're using the Pacific People, and the different languages. Like after the Cook Island one, they have a Cook Island lady saying, ("Don't have hesitation")." (P2) The perceived success of other campaigns was associated with the person fronting that campaign as much as the quality of the presentation. "Going back to those first John Kirwan (sporting celebrity) ads, they were just really nicely shot, with someone that... most New Zealand men would respect John Kirwan. He was the ideal person." (G1) Such comments confirm previous research that shows that people are more likely to act when information comes from within their own community; community leaders are highly credible sources of information and people would prefer them to be trained in issues of pandemic risk management [27] . The front person was an important factor in presenting important messages. There was some interesting debate about who could "be trusted" and who was "believable". Some participants felt that important messages should come from medical professionals or official agencies such as the Ministry of Health, District Health Board or WHO. This view highlights the importance of credibility which is regarded in the literature as a critical element in effective risk management and communication [28, 29] . Communicators who are also scientists and perceived as being impartial and knowledgeable are more likely to be regarded as credible [28] . "Well, the Ministry of Health, ideally, should front it, because you know, they're the national organisation." (G3) Others argued that the front person should be a role model, or someone recognisable to the public at large. "...I think it might stick in your mind a lot more if you'd got someone that looks familiar to you telling you the information. And I think that we do actually trust someone that maybe is a household name, more than someone that they have no idea who they are". (P1) As some participants pointed out, it may be that a variety of communicators may be best. Indeed there is support in the literature for employing a multidisciplinary approach to risk communication with input from a range of experts [1, 24] . Any conclusions drawn from this study should be considered tentative as the findings cannot be generalised to the population at large. It is not known whether the individuals who chose to participate differed from those who were eligible but chose not to participate. Whilst this study intentionally involved participants with diverse cultural and ethnic backgrounds, and included individuals from vulnerable groups, the sample does not permit conclusions regarding the effect of socio-demographic factors such as age or gender. Further research is needed to explore the complexities involved in the way in which the framing of risk messages impacts on people's perception of risk and subsequent preparedness and response behaviours. The results of this study highlight the problem with a "one size fits all" pandemic warning strategy that risks antagonising and distancing communities and thereby reducing trust in agencies and the likelihood that advice will be followed. Agencies must acknowledge that the public are diverse and need to be involved in the development and management of pandemic response initiatives appropriate for different communities and sensitive to existing cultural and/or spiritual practices. Pacific Peoples and Māori focus groups identified community institutions with established mechanisms which could provide useful vehicles for the dissemination of information and engaging with the community. With a community engagement perspective the role of the health agencies would be that of consultant to the community or a change agent rather than trying to disseminate directly to the public in a top-down approach. For all segments of the population, the effectiveness of risk communication can be increased by using community engagement and empowerment principles that help tailor information to the needs and expectations of diverse groups. Pandemic management strategies need to be seen in relation to the wider context of overall societal emergency preparedness. Preparedness for specific events, such as pandemics, might be better situated within more general risk campaigns rather than as stand alone approaches. Such a collaborative approach would also help reduce the sense that people have of being bombarded with information from multiple sources with frequently conflicting messages. Information alone is insufficient to motivate people to prepare. The way in which information is presented or conveyed is an important factor in determining an individual's response. People wanted messages about specific actions that they could take to protect themselves and their families and to mitigate any consequences. They wanted transparent and honest communication where both good and bad news is conveyed. There was a desire across all groups for clear and specific information, such as infection and/or death rates and defining symptoms. This reflects a failure to distinguish between the pandemic and its consequences and highlights the importance of doing so for risk communication.

Quantifying Type-Specific Reproduction Numbers for Nosocomial Pathogens: Evidence for Heightened Transmission of an Asian Sequence Type 239 MRSA Clone

An important determinant of a pathogen's success is the rate at which it is transmitted from infected to susceptible hosts. Although there are anecdotal reports that methicillin-resistant Staphylococcus aureus (MRSA) clones vary in their transmissibility in hospital settings, attempts to quantify such variation are lacking for common subtypes, as are methods for addressing this question using routinely-collected MRSA screening data in endemic settings. Here we present a method to quantify the time-varying transmissibility of different subtypes of common bacterial nosocomial pathogens using routine surveillance data. The method adapts approaches for estimating reproduction numbers based on the probabilistic reconstruction of epidemic trees, but uses relative hazards rather than serial intervals to assign probabilities to different sources for observed transmission events. The method is applied to data collected as part of a retrospective observational study of a concurrent MRSA outbreak in the United Kingdom with dominant endemic MRSA clones (ST22 and ST36) and an Asian ST239 MRSA strain (ST239-TW) in two linked adult intensive care units, and compared with an approach based on a fully parametric transmission model. The results provide support for the hypothesis that the clones responded differently to an infection control measure based on the use of topical antiseptics, which was more effective at reducing transmission of endemic clones. They also suggest that in one of the two ICUs patients colonized or infected with the ST239-TW MRSA clone had consistently higher risks of transmitting MRSA to patients free of MRSA. These findings represent some of the first quantitative evidence of enhanced transmissibility of a pandemic MRSA lineage, and highlight the potential value of tailoring hospital infection control measures to specific pathogen subtypes.

Methicillin-resistant Staphylococcus aureus (MRSA) is responsible for a high burden of morbidity and mortality worldwide [1] [2] [3] [4] . While community-associated MRSA is becoming increasingly important globally [5, 6] , in many countries, including the United Kingdom, MRSA remains predominantly a nosocomial pathogen [7, 8] . The dominant sequence type (ST) in Asia is ST239, and recent analysis of whole-genome sequence data has shown that this ST has distinct lineages in Asia, Europe and South America which probably share a European ancestor [9] [10] [11] . Little is known about what has enabled this ST to be so successful, or whether its propensity to transmit between hosts differs from other MRSA types in certain settings. A recent concurrent outbreak due to an ST239 MRSA strain (ST239-TW, subsequently referred to as TW) and the two dominant endemic UK MRSA types (ST22 and ST36, which we refer to as non-TW) in two linked adult intensive care units in a London teaching hospital provided a rare opportunity to compare the transmissibility of different MRSA types in the same clinical setting [12] . The transmissibility of a potentially emerging pathogen (the rate at which it spreads from an infected host to exposed susceptible hosts) is an important factor in determining its success and, in the case of an established pathogen, for estimating how effective interventions must be to bring an epidemic under control [13, 14] . Quantifying the degree to which strains of a nosocomial pathogen differ in their transmissibility in a particular setting could lead to a better understanding of why major clonal replacements occur. Measuring how such transmissibility changes in response to interventions would allow us to quantify the value of specific control measures, which may vary according to the strain [15] . This could lead to better resource use by allowing us to choose control measures appropriate for the specific strain. Such an analysis has greatest relevance for predominantly clonal organisms, such as S. aureus, where distinct lineages cocirculate over extended periods of time [10] . A fundamental measure of the overall transmission potential of a pathogen in a given setting is the basic reproduction number, R 0 . This is defined as the mean number of secondary cases generated by a typical case in a fully susceptible population [16, 17] . If the transmissibility of each infected host remains constant throughout its infectious period, and if each infected host has an equal chance of infecting each susceptible host, then R 0 is simply the product of the mean rate at which an infected host generates secondary infections and the mean infectious period (provided the two are not correlated). The self-sustaining chain reaction that constitutes a major epidemic is possible only if R 0 is greater than one. If it is less than one, although there may be some self-limiting chains of secondary transmission following the introduction of an index case (and quite large clusters become possible as R 0 approaches one), this will not lead to a sustained increase in cases and, in a large population, only a small proportion of susceptible hosts will be infected [18] . An important related number is the net (or effective) reproduction number, R t . This is defined as the average number of secondary cases generated by a case infected at time t, accounting for incomplete host susceptibility to infection and control measures in place. If R t is greater than one at time t, the epidemic will (on average) be growing. If R t is less than one, it will be declining [16, 19] . These reproduction numbers are central to a mechanistic understanding of infectious disease epidemiology, and a number of methods for estimating them from different types of surveillance data have been devised [16, [19] [20] [21] [22] [23] . However, epidemics that predominantly affect hospitalized patients require some special considerations. First, unlike the community setting, the population of those exposed to infection changes rapidly over time as patients are admitted and discharged. Second, most common nosocomial pathogens are bacteria which can be carried asymptomatically over long periods, during which time colonized hosts may have several hospital admissions. This can give rise to distinctive dynamics: in addition to the usual explosive outbreaks, we also see epidemic patterns characterized by a sequence of self-limiting clusters of transmission which, over time, become more frequent and eventually coalesce into an exponentially growing epidemic [24, 25] . The concept of the single admission reproduction number, R a , can help in the understanding of these features of hospital epidemics [25, 26] . R a is defined as the mean number of secondary cases caused by a typical infectious patient during a single admission to a particular hospital or ward otherwise free of the pathogen. Necessarily, R a is less than or equal to R 0 . However, if R a v1 and R 0 w1 then every outbreak will be locally controlled in the short term, but, with repeated challenges to the hospital, longterm control failure will be inevitable. This results from the persistence of carriage following discharge which, over time, leads to a gradual increase in numbers colonized on admission. To account for changing numbers of susceptibles, we can also define a net single admission reproduction number, R a,t . This is analogous to R t and represents the average number of secondary cases generated during a single hospital/ward admission where not everyone is necessarily susceptible. Direct ascertainment of R a and R a,t would be possible if we could reliably assess who infected whom during a hospital outbreak. In practice, even with detailed surveillance and molecular typing data, there is almost always considerable uncertainty about the true transmission tree. Instead, computationally-intensive approaches based on fitting mechanistic mathematical models to data which account for uncertainty in transmission routes and screening data represent the state-of-the art for analysing nosocomial transmission dynamics [27] [28] [29] [30] [31] [32] . However, such approaches require detailed data on both susceptible and colonized or infected patients, and an assumption that temporal changes in the transmissibility can be described parametrically by some standard functional form (most commonly, piecewise constant). As currently implemented they do not allow direct estimates of the number of transmission events associated with each patient. The aims of this paper are twofold: to describe a new approach (method 1) for estimating R a,t using hospital surveillance data; and to use it to analyse MRSA data from concurrent outbreaks with different MRSA types (TW and non-TW) in two linked adult intensive care units (ICUs). The method is simple to use and enables us to track how R a,t changes over time without the assumption that changes in transmissibility follow a fixed functional form, and without requiring data on susceptible patients. The method extends techniques for the probabilistic reconstruction of epidemic trees developed for analyzing foot and mouth disease and SARS data [21, [33] [34] [35] . We contrast results using this approach with that from a fully parametric mechanistic model (method 2), which represents an adaptation of previously described parametric models for nosocomial infection to a multistrain system [31, 32] . This second approach allows R a to be estimated. It requires more detailed data and stronger assumptions, but allows us to explicitly test hypotheses about how transmissibility is affected by interventions, and how it varies between different wards and subtypes of MRSA. While betweenclone and between-ward differences in single admission effective reproduction numbers, R a,t , calculated using method 1 may be caused by differences in transmissibility, number of susceptibles, and lengths of stays, with method 2 we assume all MRSA positive patients have the same length of stay distribution and explicitly adjust for different numbers of susceptible patients when calculating R a . Different strains of hospital pathogens may differ in their ability to spread between patients and respond differently to control measures. Attempts to quantify such betweenstrain variation are lacking in high prevalence settings. We analysed data from concurrent outbreaks with different MRSA strains in two adult intensive care units. MRSA is usually carried by patients asymptomatically, and most of our data came from routine screening swabs used to detect such carriage. We divided strains into two groups: common United Kingdom strains and strains from a type often found in Southeast Asia. We developed a new method to estimate how transmission changes over time and compared results with those from an adaptation of a previously described approach. An advantage of the new method is that it makes weaker assumptions about the process generating the data. The methods gave broadly similar results: the introduction of daily antiseptic bodywashes for all patients was the only intervention associated with a substantial fall in transmission, but this intervention was less effective for the Asian strain. This work should be useful for assessing the between-strain variation in the transmission of other hospital pathogens, and for assessing the impact of interventions on patientto-patient transmission. Under baseline assumptions, on ICU1 there were 282 MRSA importation events (episodes where patients were assumed to be MRSA positive when admitted to the ICU) and 132 acquisition events. These comprised of 12 importations and 23 acquisitions with TW MRSA and 270 importations and 109 acquisitions with non-TW MRSA. On ICU2 there were 285 importations (25 with TW) and 166 acquisitions (43 with TW) (figure 1). Importations with non-TW to the respective ICUs decreased from 0.20 and 0.19 per day in phase 1 to 0.11 and 0.13 per day in phase 4. In contrast, importations with TW MRSA peaked in phase 2 in both ICUs (at 0.03 and 0.12 per day) and were at or below 0.01 per day in phases 1 and 4. Amongst patients who were MRSA positive on admission the median length of stay was 12 days (inter quartile range [IQR]4, 18) for TW-positive patients and 6 days (IQR 3, 13) for non-TW patients (p~0:01, Wilcoxon rank sum test with continuity correction). For patients who acquired MRSA the corresponding numbers were 26.5 (13.25, 42.5) for TW patients and 19 (12, 29.25) for non-TW patients (p~0:02). There was no evidence that length of stay differed by ward (p~0:79), or by study phase (p~0:44, Kruskal-Wallis rank sum test). Over the study period (January 2002 to April 2006) there were three interventions (referred to as A, B and C) and these define four study phases. Estimated net single admission case reproduction numbers (expected number of secondary cases per case during a single ward admission) associated with each MRSA-positive patient episode are shown in figure 2 (bottom panel) together with histograms of case reproduction numbers for each ward and study phase (top panel). These highlight wide between-patient variability which decreases in the second half of phase 4 when transmission is reduced and the TW clone is eliminated. While most patients have a very low expected number of secondary cases, 22 out of 103 patients (21%) with TW MRSA are expected to transmit to at least one other patient. Corresponding numbers for non-TW MRSA are 40 out of 762 (4%). This proportion was consistently higher for TW MRSA in all four study phases: 25, 11, 18 and 31% versus 7, 0, 9 and 2% for non-TW MRSA. Aggregating these reproduction numbers into four-week intervals highlights the temporal trends, differences between wards and impact of interventions (figure 3). In ICU1 these suggest similar patterns of transmission for the different MRSA types for the period prior to intervention C (a surface antiseptic protocol). In contrast, there were marked differences between MRSA types in ICU2 and throughout the study period the four-week averaged reproduction numbers for the TW clone usually exceeded those for non-TW clones when both types were present. These differences are also seen when reproduction numbers are averaged over study phases (table 1); the TW clone had a higher reproduction number than the non-TW MRSA in each phase in ICU2 but not in ICU1. Reproduction numbers for TW MRSA were also more volatile than those for non-TW MRSA in ICU2. There was evidence from both units to suggest differences between the MRSA types in their response to infection control interventions: while the net reproduction number of non-TW MRSA fell to a low level following intervention C in both ICUs, this was not the case for the TW clone which continued to transmit for several months at pre-intervention levels. Eventually, the TW outbreak came to an end after all patients with TW MRSA were treated empirically with systemic antibiotics (linezolid) from 1st September 2004 [12, 15] . After this intervention, although patients with TW MRSA continued to be imported into the ICUs, only three isolated apparent transmission events occurred (figure 1). When reproduction numbers for the two ICUs combined were estimated (allowing for cross transmission between ICUs) the results suggested the reproduction number of the TW clone was consistently higher than that for the non-TW MRSA and varied little throughout the study period (table 1) . Reproduction numbers for the non-TW clones, in contrast, fell in phase 2 and 4. These results were not Results from method 2 showed broad agreement with these findings, but in contrast to method 1 made a priori assumptions about the timing of the changes in transmissibility (tables 2-3, figure S1 ). On both wards, averaging over all phases, there was about a 1 in 400 chance of a given susceptible patient acquiring MRSA from a particular MRSA-positive patient on a particular day (table 2) . When estimates of daily transmission probabilities from a single MRSA positive patient were constrained to take the same values for the TW and non-TW clones but were allowed to vary by ward and study phase, we found clinically significant variation between the four study phases in both ICUs (table 2). In particular, while estimates were similar in phases 1 to 3, there was a marked reduction in phase 4. There was no strong evidence that these joint estimates (for all MRSA clones) varied by ICU in any of the study phases (table 2) . These findings were robust to the assumptions made about acquisition events and times (supplementary table S3). Extending this analysis to allow transmission probabilities to vary with the MRSA type enabled differences between MRSA clones and wards to be quantified (table 3) and allowed hypothesis tests about whether the daily transmission probabilities differed between strains (thus allowing us to test whether the observed differences in transmissibility found using method 1 could be entirely explained by the longer length of stay of the TW patients). In ICU1 no consistent differences were seen. In ICU2 the TW clone had a higher daily transmission probability to susceptible patients in each of the four study phases under the baseline assumption of complete bacterial interference, though confidence intervals were wide and showed considerable overlap. Differences between TW and non-TW MRSA transmission probabilities reached statistical significance at the 5% level for both wards combined and for ICU1, but not for ICU2 alone, and in only one of the four phases (phase 4) using combined data from both ICUs. This phase corresponded to the introduction of the surface antiseptic bodywash protocol, which was associated with a more than halving of the transmission probability from a patient with non-TW MRSA compared to earlier phases. The fall in the transmission probability for TW MRSA in phase 4 was smaller, and likely to be confounded by the use of linezolid for TW carriers in this phase. Large differences in transmission probabilities for the two MRSA types were also seen in phase 2 (corresponding to the introduction of hand hygiene promotion), but in this case confidence intervals were wider reflecting the short duration of this phase. In other phases differences between TW and non-TW estimates were much smaller. The magnitude of the differences depended on which patients were assumed to be susceptible. Under the baseline assumption that patients colonized with one strain were not susceptible to acquiring another (complete bacterial interference) the differences were larger than in the sensitivity analysis where no bacterial interference was assumed (table 3) . This can be explained by the higher prevalence of non-TW MRSA clones; under the assumption of no bacterial interference all non-TW MRSA positive patients would be considered susceptible to infection or colonisation with TW MRSA and vice versa. Changing from complete interference to no interference therefore results in a greater increase in the number of susceptibles available for the TW clones to infect than it does for the non-TW clones. To accommodate these changes, a larger reduction in the daily transmission probability for TW clones is required. Overall, combining data from both wards, the TW clone was estimated to have a daily transmission probability that was between 63 and 100% higher than the non-TW clones in phase 4, and between 53 and 94% higher in phase 2 (the lower numbers corresponding to the no bacterial interference assumption) though the differences only reached significance at the 5% level in phase 4 and only under baseline interference assumptions. Transmission probabilities were broadly similar in the two other study phases. Estimates of the single-admission reproduction number (R a ) from the model without background transmission (and assuming TW and non-TW patients have the same length of stay distribution) are reported in the supplementary material (figure S1). Results were robust to the assumptions made about the number and timing of MRSA acquisition events (supplementary table S4), but fitting a more complex model allowing patient-to-patient transmission and transmission from background sources suggested that the relative importance of patient-to-patient and background transmission could not be reliably identified in such hyperendemic settings without additional data (supplementary table S5 ). Common bacterial nosocomial pathogens have distinct dynamics from typical community pathogens and call for different analytical approaches. Important features of hospital epidemics with such organisms include: i) a host population that changes rapidly over time in comparison with the timescale of epidemic dynamics; ii) a high proportion of infected (or colonized) hosts who are already infected when they enter the population (the hospital or ward); iii) a dominant role for asymptomatic infection so infected hosts can usually only be identified using screening swabs, leading to large uncertainty in the timing of transmission events; iv) a lack of a well-defined serial interval or generation time (since asymptomatic carriage can persist for months or years, but transmission is only intermittently observed during hospital admissions). The probabilistic tree reconstruction approach described above (method 1) overcame these limitations by using a hazards-based approach applied to patient screening data to assign probabilities to potential source patients for observed Phase-specific estimates for the daily probability of a susceptible patient acquiring MRSA from an MRSA positive patient in the same ward, for ICU 1 and ICU 2 (without distinguishing between TW and non-TW strains). In the Combined row, the estimates are constrained to be the same in both wards, and the All phases column constrains the estimates to be the same in the four phases. acquisition events. Using hazards in this way to reconstruct epidemic trees and estimate reproduction numbers appears to have first been suggested by Kenah et al [36] . Results using this method were supplemented with a maximum likelihood approach (method 2) where the timing of cross-infection events was assumed to be known but which allowed estimation of the daily transmission probability, enabling us to study effects related to study phase and MRSA type while controlling for differences in length of stay. These methods were applied to data from two adjacent general ICUs in which admission and weekly MRSA screens and culture results from clinical samples identified patients admitted with and acquiring MRSA over a four year period. During that time there was sustained transmission with endemic MRSA and a newly introduced TW variant. Both analytical methods supported the hypothesis that intervention C (the surface antiseptic protocol) was associated with a sustained reduction in MRSA transmission, and both indicated a reduced effect for the TW clone. Both methods gave point estimates that indicated elevated transmission of TW MRSA compared with endemic strains in all four study phases in ICU2 but not ICU1. There were, however, some differences: the wardlevel reproduction numbers (method 1) tended to indicate greater increased transmission for the TW compared to non-TW MRSA than was seen using method 2. This reflects the fact that the two methods are quantifying different things: method 1 estimates secondary cases per case, which depends both on transmissibility and the length of ICU stay while carrying MRSA; method 2, in contrast, estimates only the daily transmission probability from one MRSA carrier to one susceptible patient. This will not be affected by length of stay. Indeed, there was some evidence that patients colonised with TW MRSA (particularly those colonised on ICU admission), had a longer length of stay than those colonised with non-TW MRSA. This may reflect the link between MRSA infection and excess length of stay in this cohort [37] , and the increased virulence of the TW strain which was over four times more likely to cause blood stream infection in colonised patients compared to non-TW MRSA strains in the same ICUs [12] . Even in the absence of an increased rate of transmission to other patients, increased length of stay would lead to a higher singleadmission reproduction number. It is possible that such differences in length of stay reflect underlying differences in the characteristics of patients most vulnerable to acquiring the different MRSA types. For example, because the TW outbreak was centred on the two ICUs, patients carrying TW on ICU admission might be more likely than patients carrying non-TW MRSA to have had recent ICU admissions. The TW clones showed a far broader range of antibiotic-resistance than endemic MRSA clones and have previously been shown to preferentially colonise vascular catheters but not carriage sites compared with endemic strains [12] . Taken together, these observations suggest that the TW MRSA could represent a phenotype particularly adapted to transmission in Table 3 . Estimates of the daily transmission probability (q) from one exposed to one susceptible patient. Estimates of the daily transmission probability (q) from one exposed to one susceptible patient. settings, such as ICUs, with high levels of antibiotic usage and patient catheterisation, perhaps at the expense of persistence outside these areas. There is some evidence that such adaptation results from both increased persistence in the ICU (perhaps by targeting long-stay patients, and causing infections that increase length of stay) and from an increased daily transmission probability (particularly in the presence of widespread antiseptic use). Caveats, of course, apply: differences in lengths of stays between TW and non-TW colonized/infected patients could be confounded by exposure history (the recent arrival of the TW clone rather than its biological properties may account for the different patient characteristics). Differences in daily transmission probabilities could also be subject to such confounding and could also have arisen by chance (in all phases -even phase 4, where the effect size was largest -confidence intervals were wide). The mechanisms underlying variations in transmissibility of different MRSA (and S. aureus) strains are poorly understood. Reasons for the differences in the two ICUs are also unclear. Chance variation cannot be ruled out, as the formal investigation of transmission potential of different MRSA types was, in part, motivated by perceived differences in transmissibility (using the same data), and the usual limitations of post hoc analyses therefore apply. Also, although the analyses accounts for demographic stochasticity, there may also be important sources of environmental stochasticity which are not accounted for. It seems unlikely that the difference in TW transmission in the two ICUs can be explained by colonized staff: a universal staff screening programme failed to detect the TW clone during the outbreak [12] . Differences in infection control practice also seem unlikely but cannot be ruled out: the two wards share the same infection control policies and staff pool, with medical and nursing staff rotating between units at 3-6 monthly intervals, though only physiotherapy, radiology and pharmacy staff worked across both units at the same time. It is possible that the built environment influences MRSA transmission. ICU2 was last refurbished in 1969, retaining a mixture of original materials including wood, and has much less open space, only eight sinks, and one side room, whereas ICU1 was refurbished in 1999 to an open plan configuration with better space utilization, 19 sinks and three side rooms. The reduced availability of sinks, side rooms and space to circulate may have adversely affected the ability to carry out infection control practice or cleaning, although it is unclear why this should only affect TW MRSA, which was not detected on environmental screening during the outbreak [12] . Despite anecdotal reports that some lineages of S. aureus strains have an enhanced epidemic potential in hospital settings [38] , objective assessments of between-strain variation in transmissibility are largely lacking. Such variation is nonetheless to be expected given the large degree of phenotypic variation in different S. aureus and MRSA clones, and the dominance of a small number of MRSA lineages [39] . One of the few instances where the nosocomial transmission potential of different subtypes of the same nosocomial pathogen have been quantified comes from a comparison of the onward transmission from patients admitted to hospitals in the Netherlands carrying MRSA [40] . In this case, because MRSA introductions were infrequent (as MRSA prevalence in hospitals in the Netherlands is below 1%) and contact tracing extensive, the secondary cases could be assigned to distinct clusters of transmission following identified introductions. This allowed the authors to use methods based on a branching process model to estimate the single admission reproduction number, R A [41] . It was found that newly admitted ST398 MRSA strains (which are commonly associated with livestock production) had a greatly reduced propensity to spread compared with other MRSA sequence types, with an R A value (95% CI) of only 0.16 (0.04-0.40), about one sixth of the corresponding value for non-ST398 MRSA. The authors concluded that less stringent control measures were likely to be sufficient to control ST398 MRSA clones than those needed for non-ST398 MRSA types. Such methods would not have been applicable for our data, and the first method used here to quantify the transmissibility of different strains (method 1) instead built on recent approaches to estimate reproduction numbers by probabilistically reconstructing epidemic trees. Such tree-reconstructions have used simple rulebased methods, for example assigning sources from a candidate list based on proximity data [33] , more formal semi-parametric methods using partial likelihoods and assuming a known serial interval distribution [21, 34] , and, most recently, semi-parametric hazard-based approaches [36] . Hazard-based approaches have some advantages over the first two methods: they avoid some of the arbitrary assumptions of the rule-based approaches, do not require knowledge of the serial interval distribution, and can avoid biases that arise from the fact that the serial interval distribution changes over the course of an epidemic. Advantages over approaches based on fitting a full transmission model include fewer assumptions, in particular with regard to the functional form of changes in the transmission potential over time. In this respect, tree reconstruction approaches have some similarities with other semi-parametric approaches that make use of survival analytical methods, such as the approach adopted by Wolkewitz et al., who derived non-parametric estimates of a time-varying transmission rate changed over time using a Martingale-based method [42] . An important difference in the current approach is that we are specifically interested in estimating how the distribution of the number of secondary cases resulting from each case changes over time. The method 1 approach described here also makes relatively low demands for data (with no information required for patients who do not become colonized or infected), has a low computational burden, and can be easily adapted to cope with cocirculating subtypes as in the application here. This approach is appropriate when the daily probability of a patient acquiring MRSA is small, as in this case reconstructed epidemic trees will be approximately independent of this probability. This approximation is likely to be reasonable for all but the most explosive outbreaks. For example, using the exact formula we found that changing this probability from a baseline of 0.005 to 0.001 and 0.025 changed the estimated mean reproduction numbers for each phase and MRSA type by less than 3%. Two assumptions underlying the analytical approaches used here are i) that new MRSA acquisitions can be explained by patient-to-patient spread within the units (which is likely to be mediated by contacts with transiently colonized healthcare workers) and ii) that risk of transmission increases in line with colonization pressure (the number of patients with MRSA on the ward). While these assumptions are supported by observational and quasi experimental studies [43, 44] , it would be desirable to more rigorously challenge them. Unfortunately, unpublished simulation studies and analysis here with a more complex model allowing different transmission routes (table S5) both suggest that the ability to identify the relative importance of background and patient-to-patient transmission may be limited in hyper-endemic settings in the absence of more discriminatory typing data. The inability of our typing methods to reliably distinguish between non-TW MRSA types, or to identify genetic variants of the TW clone therefore represent important limitations of this work. High resolution genotyping data would enable more definitive assessments of who infects whom, and therefore allow us to quantify the risks of transmission of different MRSA subtypes in different wards at different times with greater certainty. Figure 1 confirms that not all acquisition events can be explained by transmission from a known MRSA positive patient from the same ward. The combined-ICU analysis, allowing for between-ward transmission, is able to account for some MRSA acquisitions where no known source was present on the same ward, and this explains why combined ICU estimates of the reproduction number are sometimes outside the range of individual ICU estimates (table 1), but unknown MRSA sources are also likely to be present in the patient population [32] . A full model-based analysis using data augmentation (which estimates model parameters and latent parameters that represent ''unobserved'' -or augmented -data, typically using Markov chain Monte Carlo methods for fitting) could account for such unknown sources. Such an approach retains some important advantages for analysing typical surveillance data. These include the ability to account for imperfect swab sensitivity and for uncertainty in the number and timing of acquisition events, circumventing the need to make arbitrary assumptions about which patients were colonized on admission to a ward. In the present context such an analysis would allow us to explicitly account for the change in the screening protocol in November 2004. Since this involved screening more body sites, it is likely to have increased screening sensitivity and led to increased detection of MRSA, potentially biasing the estimated effect of intervention C. To date, however, no published work has adapted such approaches to cope with multiple co-circulating subtypes. The method 2 used here can be thought of as a simplified version of such an approach (in that it is based on a fully-specified mechanistic transmission model) but it avoids the complexities of data augmentation by assuming the epidemic process is perfectly observed. An important area for future work will be to extend data augmentation methods to cope with carriage of multiple types. Such approaches have been developed for the sequential carriage of community pathogen subtypes [45] . Addressing issues of co-colonisation with different subtypes may be particularly important for some nosocomial pathogens, and neglecting such effects is a potential source of bias. In our analysis here we considered two possibilities -complete bacterial interference (where one strain completely inhibits the acquisition of another), and no bacterial interference. The reality may lie somewhere between these two extremes. Such an analysis will be complicated by the fact that routinely-used laboratory methods are not well-suited to detecting the simultaneous carriage of multiple types [46] , and sensitivity for detecting a second type will not, in general, be the same as sensitivity for detecting a single type. Ethical approval for this research was granted by the NHS National Research Ethics Service, South East Research Ethics Committee. All data were analyzed anonymously. Anonymised data from two 15-bed adult general intensive care units (ICU) within a 1050-bed teaching hospital in London, United Kingdom, were collected between 1st January 2002 and 30th April 2006 as described elsewhere [12, 15] . Dates of admission and discharge and MRSA culture results from screen and clinical samples were analysed for all 4,570 consecutive patient admissions to both ICUs. Infection control policies were in place including specifying hand hygiene between patient contacts and use of contact precautions for known MRSA colonized patients throughout. On this background three main new MRSA control interventions were introduced: intervention A (introduced on 15th July 2003) was an education campaign to promote hand hygiene and barrier nursing; intervention B (introduced on 15th October 2003) was isolation of known MRSA colonized patients in side rooms or in patient and nursing cohort pairs; intervention C (introduced on 26th April 2004) was a surface antiseptic protocol which included daily chlorhexidine bodywashes for known MRSA positive patients, and daily triclosan bodywashes for other patients. The three interventions defined four study phases for analysis: phase 1 from 1st January 2002 to 14th July 2003; phase 2 from 15th July to 14th October 2003; phase 3 from 15th October 2003 to 25th April 2004; and phase 4 from 26th April 2004 to 30th April 2006. Patients were swabbed for MRSA carriage on admission and every Monday morning. Swabs were taken from nose, axillae and perineum until 1st November 2004, when additional rectal and throat samples were included (a change associated with an approximate 30% increase in the proportion of patients identified as carriers on admission to ICU) [47] . Clinical samples were collected when infection was suspected. S. aureus colonies were identified using a combination of catalase positivity, Staphaurex (Remel Europe Ltd., Dartford, England) and/or salt mannite positivity with confirmation by a tube coagulase test. Methicillin resistance was determined by disc testing. Screen samples were identified using a selective mannitol broth technique [47] . TW MRSA was defined initially by its distinctive and extensive antimicrobial resistance pattern, sequence typing and microarray analysis [12] . More extensive typing of available admission and acquisition isolates has shown all antimicrobial resistance patterns defined TW isolates to belong to CC8/239 and non-TW isolates to be w90% ST22 and ST36 [15] . When TW and non-TW MRSA isolates were recovered from the same patient, only the first type recovered was considered. This was, however, rare: two patients had both types recovered from pooled screening sites; nine had both types from sputum; and seven had both types from wounds. Thirteen patients had both types recovered from different sites. Further details of patient characteristics, interventions, swabbing sites and microbiological procedures have been described elsewhere [12, 15] . We analyse the data using two separate approaches which we refer to as method 1 and method 2. In both analyses we define a new MRSA acquisition to have occurred if a patient has a negative admission screening swab, a subsequent MRSA positive screen or clinical sample while in the ICU and more than 48 hours after being admitted to the ward, and no prior MRSA positive isolate in the 90 days preceding ICU admission. Patients with any MRSA positive samples taken within 48 hours of admission are assumed to be positive on admission (MRSA importations). A patient who is believed to be neither colonized nor infected on a given day is assumed to be susceptible to becoming colonized or infected by either MRSA type (see supplementary material for further details). In the first approach (method 1), which probabilistically reconstructs the epidemic tree, we assume that the acquisition occurred one or more days before the first positive screening swab. In the second approach (method 2) we assume a new acquisition to have occurred on day t{1 if a patient has his or her first MRSA positive swab on day t, following a negative MRSA admission screening swab during the same ward admission. We also assume i) that once MRSA positive, a patient remains so until ward discharge (hence no information from swab results after the first positive is used), and ii) MRSA-positive patients only become potential sources for transmission to other patients after their first positive swab, unless they are assumed to be positive on admission, in which case they are potential sources from their date of admission. For patients readmitted to one of the wards following ward discharge, we apply the same criteria that we use for first time admission to determine admission colonisation status. We use a time unit of one day, and take dates of admission and discharge to represent the first and last whole days of a patient admission. Notation. We introduce the following notation: let q ijt represent the daily probability of a single susceptible patient in ward i acquiring MRSA of type j from a single patient on the same ward at day t who is colonized or infected with MRSA type j. We also define the daily avoidance probability of acquiring MRSA, q' ijt~1 {q ijt . Denote by S wjt , C wjt , and A wjt the number of patients on ward w on day t who are, respectively, susceptible to MRSA type j, known to be colonized or infected with type j, and found to be colonized or infected with type j for the first time on day t (having had a prior negative admission screening swab). Here we take C wjt to be the number of patients on ward w on day t who have had at least one previous positive swab with MRSA type j on or before day t. We take A wjt to be the number of patients on ward w who have their first positive swab with MRSA of type j on day t. In our default method 2 analysis we take S wjt as the number of remaining patients (i.e. those with no prior MRSA positive swabs during the current episode) excluding those patients who are discharged on day t since we assume that acquisitions on the day of discharge would be not be detected. The tree reconstruction approach (method 1), in contrast, does not require knowledge of S wjt to estimate reproduction numbers. In the application we consider here there are two wards and we define two subtypes (TW and non-TW), so both w and j can take values 1 or 2. We also define N w to be the total number of patient episodes on ward w over the study period for which there was at least one MRSA positive swab and M w to be the total number of new MRSA acquisitions on ward w over the study period, i.e. M w~P t P j A wjt . Method 1: Reconstruction of the epidemic tree. The treereconstruction approach calculates the probability that each observed new MRSA acquisition was acquired from each of the other MRSA positive patients in one of the two ICUs. In this approach we condition on the probabilities q ijt and assume that all new acquisitions were acquired from a known patient source. A scaling factor, s, specifies the reduction in the daily risk of transmission from an MRSA positive patient in one ward to an MRSA negative patient in a different ward. We explore the sensitivity of the results to both the q ijt and the s values. Let p kl represent the conditional probability that patient k acquired MRSA from patient l given that patient k acquired MRSA from one of the other P w N w {1 MRSA positive patients. We calculate the elements p kl of the P w M w | P w N w matrix P kl as follows. Define u ijt to be the probability that a susceptible patient on ward i at time t escapes cross-infection from one of the P w C wjt MRSA type j positive patients in the ICUs on that day. Therefore u ijt~Pw 1{s (1{ fwg (i)) q ijt À Á Cwjt where fwg (i) is an indicator function that equals one when w~i and zero otherwise. Now consider a single patient k on ward i who is free of MRSA on admission and whose first and last days on the ward are t k f and t k l . The probability that this patient is free of MRSA type j at the end of day s (sƒt k l ) is where the h wijt terms represent the hazards of transmission of MRSA type j from patients in ward w at time t to a patient in ward i and lijs is an indicator function that takes the value 1 if patient l is present on ward i and MRSA type j positive on day s and 0 otherwise. These hazards can be expressed in terms of the probabilities q ijt as h wijt~{ C wjt ln(1{q ijt s (1{ fwg (i))), which is approximated by C wjt q ijt s (1{ fwg (i)) when q ijt is small, as will usually be the case. The conditional probabilities, p kl , which represent the probability that patient k was infected by patient l given that patient k was infected by one other patient, are then given by the following expression, which is approximately independent of q ijt when q ijt is small: Here m indexes all the patients who could potentially have infected patient l. The net single admission reproduction number for patient episode l (i.e. the expected number of secondary cases resulting from this episode) is then given by R l~X m p ml and corresponding reproduction numbers for a given time period are obtained by averaging over these patient reproduction numbers for all patient episodes starting in the given period. Associated confidence intervals are derived by simulation, by repeatedly drawing the source of infection for each of the P w M w new infections from a multinomial distribution with probability vectors given by the rows of P kl . All confidence intervals reported for reproduction numbers are based on the quantiles of 1000 such simulations. By default, when analysing both wards together, we assume minimal cross-infection between the wards (s~0:0001), though we also consider the opposite extreme (s~1), representing complete ward mixing. We report results here where q ijt is fixed at 0.005, though also describe results of sensitivity analyses with values of 0.001 and 0.025. Method 2: A likelihood-based approach. The second approach estimates the probabilities q ijt using maximum likelihood estimation (MLE). With two MRSA types, we assume that patients susceptible to type 1 are also susceptible to type 2, so S i1t~Si2t~Sit . This implicitly assumes complete bacterial interference (i.e. that colonisation with one type of MRSA prevents acquisition of another type one or more days after the first acquisition [48] ). As a sensitivity analysis we also considered the other extreme: complete lack of bacterial interference, so that a patient colonized with one MRSA type had the same daily risk of acquiring a different subtype as an uncolonized patient. With the additional assumption that new MRSA acquisitions occur the day before they are detected, the log likelihood of the new MRSA acquisition data in ward i on day t is given by: where c is a constant and Here P i1t (P i2t ) is given by the product of the probability of not acquiring MRSA type 2(1) on day t{1 and the probability of acquiring type 1 (2) . P 0 it is the product of the probabilities of not acquiring either MRSA type. For completeness we could add a term to represent acquisition of both types on the same day. In practice we had no such observations. The overall log-likelihood is then given by the sum of LL it terms over i (the two wards) and is maximized using unconstrained optimization with the Nelder-Mead algorithm [49] as implemented in the function optim in R version 2.11.1 (www.project-r.org). Approximate 95% confidence intervals (CIs) were derived by inverting the square matrix of second-order partial derivatives of the loglikelihood function, i.e. the Hessian. In practice, with method 2 rather than estimating separate q ijt terms for every day t (which would over-parameterize the model), we apply constraints. We consider constraints where i) q ijt terms from the same ward and same study phase are required to take the same value; ii) q ijt are the same across all time periods within each ward; iii) q ijt terms from the same study phase are constrained to take the same value and do not vary across study wards; and iv) q ijt terms are the same for different MRSA types, q i1t~qi2t . These constraints imply a series of nested models, and we apply likelihood ratio tests to determine whether there is evidence to reject the hypotheses that these equality constraints represent. To obtain estimates of R a requires consideration of the length of stay distribution. To simplify matters we ignore any potential additional length of ICU stay caused by infection, but account for the facts that the longer a susceptible patient stays the greater the risk of acquiring MRSA, and the longer an MRSA-positive patient stays the greater the expected number of secondary transmission events they will cause. In simple models it is often assumed that there is constant hazard of hospital discharge and the length of stay distribution is exponential. In such cases the risk of a patient acquiring MRSA is unrelated to the subsequent length of stay and R a is trivially calculated as the product of mean length of stay, the mean number of exposed patients on the ward, and the daily probability of transmission from a single source to a single exposed patient. In practice, the hazard of ICU discharge is likely to be dependent on the day of stay, and typically decreases with increasing day of stay. In this case, the patients at greatest risk of acquiring MRSA (i.e. those who have stayed longest on the ward) will also have longer expected future stays and will therefore tend to cause more secondary infections. To account for this we partition patients into groups defined by the number of ICU days a randomly-selected patient on the ward would stay after becoming MRSA positive (assuming no additional stay due to MRSA). To do this we use the empirical length of stay distribution and calculate the probability, p i , that a randomly selected patient on the ICU is a member of group i. This is given by p i~P ? k~1 l kzi{1 = P ? k~1 kl k , where l k is the probability that a newly admitted patient stays for k days. Assuming an average of n exposed patients per day on the ward, and that each patient has a daily probability, q, of acquiring MRSA from a single MRSA positive patient on the ward, an MRSA patient in group j will, on average, transmit MRSA to k ij~pi |q|j|n patients in group i (ignoring saturation effects, which will be negligible for sufficiently small q). These k ij values are the elements of the next generation matrix. The dominant eigenvalue of this matrix gives the reproduction number, R a [20] . In contrast to method 1, which excludes susceptible patients from the analysis and enables estimates only of the net (or effective) reproduction number, this method accounts for susceptibles in the model and therefore allows us to estimate the single-admission reproduction number (the transmission potential of an MRSA positive patient in an otherwise fully susceptibility ward). This will be greater than or equal to the net single-admission reproduction number, and determines the threshold epidemic behaviour [25] . A further difference is that this method allows the transmission potential of an MRSA positive patient to change over time, according to the current study phase. In contrast, method 1 makes no explicit assumptions about the timing of changes in the transmission potential, and net reproduction numbers for a particular time period relate to the transmission potential of patients admitted to the ward during that time period even though the actual transmission events may occur at a later time. Both methods 1 and 2 assume that MRSA acquisition events occur as a result of patient-to-patient transmission from known carriers and exclude observations where there are no potential source patients. An additional sensitivity analysis therefore extended method 2 by allowing for both patient-to-patient transmission and background transmission (for example, from colonized staff or persistent environmental contamination). If the daily probability of a susceptible patient on ward i at time t acquiring strain j from such background sources is q' ijt , then the P i1t term above is replaced by (1{q' i2t )(1{q i2t ) C i2(t{1) 1{(1{q' i1t )(1{q i1t ) C i1(t{1) À Á in the new model, with similar changes for other terms. When applying this model, q' i1t was assumed to vary by ward, MRSA type and study phase, but to remain constant within in a phase. Figure S1 Single admission reproduction numbers (R a ) estimated using method 2. Estimates (95% CIs) of the wardlevel reproduction number, R a , according to study phase, MRSA type and ward obtained using method 2 and assuming complete bacterial interference and no interaction between ICU 1 and ICU 2. (PDF) Protocol S1 Protocol for defining MRSA importation and acquisition events. Table S1 TW and non-TW MRSA importation and acquisition events under different assumptions. The baseline assumption classifies all episodes where MRSA was recovered from an isolate taken within 48 hours of admission as importations. The SA1 assumption uses a 24 hour cutoff instead. See protocol S1 in supporting material for full details of baseline and SA1 assumptions. (PDF) Table S2 Estimated ward-level reproduction numbers (s.e.) for TW and non-TW MRSA clones under alternative assumptions. Phase-specific estimates of ward-level reproduction numbers for TW MRSA and Non-TW MRSA derived using Method 1 under baseline assumptions with perfect ward coupling (applies to combined ICU estimates only) and under SA1 assumptions (see protocol S1 in supporting material for details of baseline and SA1 assumptions). (PDF) Table S3 q estimates for TW and non-TW combined under SA1 and SA2 assumptions. Sensitivity analysis for phase-specific estimates for q, the daily probability of a susceptible patient acquiring MRSA from an MRSA positive patient in the same ward, for ICU 1 and ICU 2 (without distinguishing between TW and non-TW strains). In the Combined row, the estimates are constrained to be the same in both wards, and the All phases column constrains the estimates to be the same in the four phases. See protocol S1 in supporting material for details of the SA1 and SA2 assumptions used in the sensitivity analyses. 1 P-values test the null hypothesis that transmission does not vary between study phase (likelihood ratio test, df = 3). 2 P-values test the null hypothesis that transmission in the current phase does not differ between wards (likelihood ratio test, df = 1). (PDF) Table S4 Estimates of the daily transmission probability (q) from one exposed to one susceptible patient under SA1 and SA2 assumptions. Estimates of the daily transmission probability (q) from one exposed to one susceptible patient under assumptions SA1 and SA2. See protocol S1 in supporting material for details of the SA1 and SA2 assumptions used in thesse sensitivity analyses. 1. P-values test the null hypothesis that transmission varies between study phases but not MRSA types against the alternative that it varies between study phases and MRSA types (likelihood ratio test, df = 4). 2 P-values test the null hypothesis that transmission in the study phase does not differ between TW and Non-TW MRSA using combined data from both wards (likelihood ratio test, df = 1). (PDF) Table S5 Estimates of the daily transmission probability (q) from one exposed to one susceptible patient and from background transmission sources. 'Patient to patient' estimates corresponds to the daily transmission probability (q) from one exposed to one susceptible patient. Background estimates corresponds to the daily probability of acquisition from background sources (such as environmental contamination). This probability is assumed to remain constant within each phase for each of the two MRSA types. In some cases confidence intervals could not be estimated for numerical reasons, while in others the very wide confidence intervals indicate that parameters are weakly identifiable. For both ICUs we compared the model with background and patient-to-patient transmission (with both phase and MRSA-type specific parameters) with nested models with only background transmission (but still with both phase and MRSAtype specific parameters) using a likelihood ratio test based on the chi-squared distribution with eight degrees of freedom. The results gave strong evidence to prefer the more complex model in the case of ICU2 (p = 0.008), but no evidence to prefer it in the case of ICU 1 (p = 0.70). (PDF)

Discovery and Genomic Characterization of a Novel Bat Sapovirus with Unusual Genomic Features and Phylogenetic Position

Sapovirus is a genus of caliciviruses that are known to cause enteric disease in humans and animals. There is considerable genetic diversity among the sapoviruses, which are classified into different genogroups based on phylogenetic analysis of the full-length capsid protein sequence. While several mammalian species, including humans, pigs, minks, and dogs, have been identified as animal hosts for sapoviruses, there were no reports of sapoviruses in bats in spite of their biological diversity. In this report, we present the results of a targeted surveillance study in different bat species in Hong Kong. Five of the 321 specimens from the bat species, Hipposideros pomona, were found to be positive for sapoviruses by RT-PCR. Complete or nearly full-length genome sequences of approximately 7.7 kb in length were obtained for three strains, which showed similar organization of the genome compared to other sapoviruses. Interestingly, they possess many genomic features atypical of most sapoviruses, like high G+C content and minimal CpG suppression. Phylogenetic analysis of the viral proteins suggested that the bat sapovirus descended from an ancestral sapovirus lineage and is most closely related to the porcine sapoviruses. Codon usage analysis showed that the bat sapovirus genome has greater codon usage bias relative to other sapovirus genomes. In summary, we report the discovery and genomic characterization of the first bat calicivirus, which appears to have evolved under different conditions after early divergence from other sapovirus lineages.

The caliciviruses are a family of small non-enveloped viruses, and can be classified into five genera: Vesivirus, Lagovirus, Norovirus, Sapovirus and Nebovirus. They possess a non-segmented, polyadenylated, positive-sense ssRNA genome of about 7.5 to 8.5 kb in length, enclosed in an icosahedral capsid of 27 to 40 nm in diameter. Among them, noroviruses and sapoviruses (SaVs) are well known to cause enteric disease in a range of mammals, including humans, while vesiviruses and lagoviruses cause systemic diseases in specific animal hosts. Nebovirus is the most recently established genus in the family Caliciviridae [1] , and its members are associated with enteric diseases in cattle [2, 3] . A putative sixth genus, Recovirus, has been proposed for a novel calicivirus detected in stool specimens from rhesus monkeys [4, 5] . Another new genus, Valovirus, has been proposed for a novel group of swine caliciviruses known as the St-Valérien-like viruses [6] . In addition, there exist other unclassified caliciviruses, such as the recently described chicken calicivirus [7] . The genus Sapovirus currently contains only one recognized species, the Sapporo virus, which was discovered in 1977 in Sapporo, Japan [8] . The SaV genome is approximately 7.1 to 7.5 kb in length, and may have two or three ORFs. ORF1 encodes a polyprotein that undergoes proteolytic cleavage to form the non-structural proteins and the major capsid protein VP1. ORF2 encodes the minor structural protein VP2. ORF3 encodes a small basic protein of unknown function [9, 10] . Interestingly, it is located in an overlapping reading frame within ORF1, and is present only in SaVs from selected genogroups. At present, SaVs are classified formally into 5 genogroups based on phylogenetic analysis of the full-length VP1 sequence, though additional genogroups have been proposed to accommodate some novel SaVs discovered in recent years. Further classification of SaVs into genotypes has also been undertaken, though taxonomic assignment at the genotype level appears to be less well-defined than at the genogroup level [11] . As mentioned above, both noroviruses and SaVs generally cause mild to asymptomatic enteric infections in human and animal hosts [12] . Human SaV infections are reported to be similar to or milder than human norovirus infections, but SaV infections have a shorter duration of viral shedding and are less associated with projectile vomiting [13] [14] [15] [16] . Incidence of SaV-associated gastroenteritis infections remains less than norovirusassociated infections for both sporadic and outbreak settings, though various studies have reported increasing rates of SaV infections around the world [17] [18] [19] [20] [21] . The genetic diversity of SaVs is comparable to that of noroviruses, and the diversity of reported animal hosts is also similar. Noroviruses have been discovered in specimens from humans, pigs, cattle, dogs, sea lions, African lion, and mice [22] [23] [24] [25] . In comparison, SaVs have been found in specimens from humans, pigs, dogs, minks and California sea lions [23, [25] [26] [27] . Bats (order Chiroptera of class Mammalia) constitute a significant portion of biological diversity in many ecosystems and have a wide geographical distribution [28] . We have previously discovered novel viruses in several local bat species [29] [30] [31] [32] [33] [34] [35] , and there were many similar discoveries of novel bat viruses by researchers in other parts of the world. In particular, important human viral pathogens like the SARS virus, Nipah virus and Ebola virus were found to have originated from bats and contributed to substantial human morbidity and mortality in recent outbreaks. Taken together, these discoveries hint that these small mammals are important reservoirs of diverse and undiscovered animal viruses, with significant risk of zoonotic transmission to humans [36] . In the present study, we investigated the presence of unknown calicivirus diversity in bats by targeted RT-PCR screening. Novel SaV sequences were amplified from several faecal samples of the bat species Hipposideros pomona, and genome sequences were obtained for three strains of the bat SaV. Sequence analysis indicated that the novel virus possesses several genomic features atypical of SaVs, and phylogenetic analysis revealed that it descended from a lineage that had diverged early from other SaV. A total of 728 anal swabs from different bat species in Hong Kong were obtained. No obvious signs of enteric disease, like anorexia and diarrhoea, were observed in the bats during the brief period of captivity needed for sampling. RT-PCR using broadly reactive degenerate primers for a 185 nt fragment in the 3D-like RNA-dependent RNA polymerase (RdRp) region of the calicivirus ORF1 gene was positive in two specimens. Repeated screening using more sensitive specific primers revealed three additional positive specimens. Further information on the species and RT-PCR screening results are presented in Table 1, Table S1 and Figure S1 . Sequence similarity search using BLASTN against the NCBI nonredundant nucleotide database did not reveal significant similarity to known SaV sequences. Another search using BLASTX against the NCBI non-redundant protein database produced hits to SaV sequences, with the most closely related sequence being the RdRp sequence of porcine SaV (GenBank accession number ACT98315) at 43% aa identity. A phylogenetic tree was constructed from the nucleotide alignment based on the length of the partial RdRp sequence obtained from bat SaV/TLC72 (GenBank accession number JQ267527) ( Figure S2 ). Complete or nearly full-length genome sequences (with incomplete 59 ends) were obtained for three positive samples using the sequencing strategy as described in the Methods section. For two of the samples that were positive only for RT-PCR screening with specific primers, only sequences for short segments of the viral genome were obtained. Additional viral genome sequencing on these samples was unsuccessful due to limited clinical materials available and possibly low viral titres. The complete genome of bat SaV strain TLC58 (Genbank accession number JN899075) is 7696 nt in length and has a genomic G+C content of 60.2 mol%. Both the length and G+C content of the bat SaV genome are significantly higher than that of other known SaVs (Table 2) . Each genome is predicted to contain 3 overlapping ORFs, comparable to the genome organization of SaVs in GI, GIV and GV ( Figure 1 ). The 59-UTR and the 39-UTR are 9 nt and 225 nt in length, respectively. The length of the 39-UTR is considerably longer than other SaVs ( Table 2 ). The two other nearly full-length bat SaV genomes were found to be highly similar to that of the complete bat SaV/TLC58 genome in nucleotide sequence and genome organization, and were not analysed separately. The complete ORF1 is 6855 nt long, and encodes a large precursor polyprotein with an estimated molecular mass of 246.8 kDa. The polyprotein contains characteristic amino acid motifs conserved in caliciviruses: 2C-like NTPase at residue 482 (GPPGIGKT), VPg at residues 958 (KGKTK) and 972 (SEYEE), 3C-like protease at residue 1183 (GDCG), 3D-like RNAdependent RNA polymerase at residues 1520 (GLPSG) and 1568 (YGDD), and VP1 at residue 1867 (PPG). It undergoes proteolytic processing to produce the nonstructural viral proteins and the major capsid protein VP1. Based on comparison with the ORF1 cleavage map of SaV/Mc10 [37, 38] , a human SaV GII strain, we predicted the cleavage site that generates the major capsid protein to be located between residues 1740 (E) and 1741 (G). An in-frame AUG start codon is located in a favourable context for translation initiation (GUGUUUGUGAUGGA) just upstream to the cleavage site, which has also been reported in other caliciviruses [6, 39] . This sequence is noted to be similar to the 59 UTR of the genome, and it was postulated that the site might permit internal translation initiation from subgenomic RNA [39] . The sequence identities of the bat SaV/TLC58 with other SaVs in the complete ORF1 protein sequence vary between 36.0% and 37.4% (Table 3) . While comparison with caliciviruses of other genera, the ORF1 sequence identities are overall lower (15.6%-22.8%) than those between different SaVs (Table S2) . For individual alignment of protease -polymerase, sequence identities with other SaVs (45.3%-48.4%) are overall higher than those with other genera (22.7%-32.1%) ( Table 3 and Table S2 ). The VP1 is predicted to be 546 aa long, and has a molecular mass of 56.6 kDa. It shares 36.1% to 39.2% amino acid identities with VP1 of other SaVs (Table 3) . Likewise analysis with caliciviruses of other genera reveals lower similarities with 14.9% to 23.4% sequence identities (Table S2 ). The complete ORF2 is 615 nt long, with an overlapping region of 8 nt with the 39 terminus of ORF1. Its reading frame is +1 relative to that of ORF1, unlike most other SaVs (Table 2) . ORF2 encodes the minor structural protein VP2, which has an estimated molecular mass of 21 kDa. The mechanism of translation initiation in ORF2 of SaVs has not been fully elucidated. In the present case, a translational upstream ribosome binding site (TURBS) motif (CAUGGGACC; underline indicates region complementary to 18 S ribosomal rRNA sequence) could be identified at 24 nt upstream of the ORF2 start codon. Sequence identities for VP2 with other SaVs vary from 15.5% to 19.9% (Table 3) . By comparison with caliciviruses of other genera, sequence identities are generally lower than those between SaVs. (4.8%-12.3%) (Table S2 ). Phylogenetic trees were constructed using the predicted amino acid sequences of the ORF1 precursor polyprotein (Figure 2 ), VP1 and VP2 ( Figure 3 ). The LG+G+F model was found to be the bestfit substitution model in all cases. Phylogenetic analysis was not performed for the putative ORF3 product as no homologous sequences were available. Sequence analysis with the Recombination Analysis Tool did not reveal any potential recombination breakpoints in the bat SaV sequences. There are subtle but important differences in the phylogenetic position of the bat SaV in the three phylogenetic trees. In the tree based on the full-length amino acid sequences of the ORF1 polyprotein, the bat SaVs are clustered tightly with the porcine SaVs in a monophyletic clade constituting the SaVs. However, in the VP1 tree, the bat SaVs are positioned just outside the clade of other SaVs. In the VP2 tree, the bat SaVs are located approximately equidistant from the GII noroviruses and porcine SaVs. The phylogenetic positions of the bat SaVs are supported by high Shimodaira-Hasegawa-like approximate likelihood ratio test (SH-like aLRT) branch support values as calculated by PhyML. Although the phylogenetic positions of the novel bat virus are slightly divergent in the three trees, they generally show the bat SaV as being most closely related to the SaVs. In our opinion, there is insufficient ground for proposing a new genus for the novel virus under the current framework of taxonomic classification. The ORF1 polyprotein and VP1 capsid protein sequences of the novel bat virus showed obvious phylogenetic clustering with other SaV sequences. It should also be noted that the VP2 protein sequences are shorter and more divergent, and therefore are considered to be less useful in the phylogenetic classification of caliciviruses [4] . Lastly, the genome organization of the bat SaV is highly similar to that of the SaVs as shown above. Hence, together with relatively high sequence identities with other SaVs rather than with calicivirues in other genera (Table S2) , we propose that the novel bat virus be classified as a new member of the genus Sapovirus in the family Caliciviridae. Codon usage and compositional bias analysis As genomic nucleotide composition is strongly associated with codon usage bias in viruses, we examined the codon usage in the genomes of the novel bat SaV and other SaVs given their different nucleotide composition. The bat SaV genome was found to have significantly greater codon usage bias than the other SaV genomes, as measured by their effective number of codons (N c ) ( Figure 4 ). Adjusting N c for background nucleotide composition (N c 9) did not significantly affect the observed difference in codon usage bias. Next, we examined CpG dinucleotide bias in the SaV genomes, as studies on other animal RNA viruses suggest that CpG suppression is a major factor in their genome evolution [40, 41] . Odds ratio of CpG and GpC dinucleotides (r CG and r GC ) and the CpG/GpC ratio were calculated to assess the degree of CpG suppression. Results confirm the presence of significant CpG suppression (r CG #0.78) in examined SaV genomes, with the only exception being the bat SaV genome (Table 4 ). r GC values are similar across examined SaV genomes, suggesting that the difference in CpG suppression is specific. All SaV genomes are found to have a slightly negative GC skew, and there is no major difference between the degree of GC skew in bat SaV and the other SaVs (Table 4 ). This suggests that the degree of cytosine deamination is not a major factor in the altered GC composition and CpG suppression in the bat SaV genome. Although the taxonomic classification of caliciviruses has improved with the availability of full-length gene sequences and robust phylogenetic methods [42] , the increase in genetic diversity introduced by novel caliciviruses would necessitate further taxonomic revisions within the family. The International Committee on Taxonomy of Viruses has adopted a systematic polythetic approach towards virus taxonomy, but classification at or below the genus level may be complicated by the specific biology of diverse viruses. As a case in point, the proposed assignment of the novel bat virus to the genus Sapovirus might be opposed on the basis of an increased genomic G+C content, the different reading frame of ORF2, and the increased length of the 39 UTR. On the other hand, the polythetic criteria for inclusion in the genus are not fully clear, and phylogenetic distances between viral gene sequences have assumed overriding importance in previous and current classifications. It should be noted that even phylogenetic analysis may be confounded by other factors such as the cleavage pattern of ORF1 polyprotein, which has not been determined experimentally for many caliciviruses. Among the various notable genomic features and properties in the novel bat SaV, we were most intrigued by its remarkably high G+C genomic content. Most caliciviruses have a genomic G+C content of 44.2-57.4 mol%. Among them, the genomic G+C content of the SaVs lie within the relatively narrow range of 49.0-53.6 mol% in spite of their genetic diversity. Hence, the presently observed G+C genomic content of 60.2% is significantly higher than that for other SaVs or caliciviruses, and indeed would rank amongst mammalian RNA viruses with the highest G+C genomic content [43] . Relatively little is known about the evolution of genome composition in caliciviruses. A number of factors have been postulated to exert selectional pressure on the G+C content of viral genomes, including host body temperature, immune pressure, codon and nucleotide usage patterns [44] [45] [46] [47] . Our results suggested that the increased G+C content is associated with a decrease in CpG suppression, but does not have a direct correlation with codon usage bias. We are unaware of any Table 3 . Comparison of genome identities and amino acid identities between the predicted polyproteins of bat SaV and the selected SaV. previous findings indicating that genomes of bat viruses are under less CpG suppression, thus the observed reduction in CpG suppresion is unlikely to result from host-related factors. The greater codon usage bias in the bat SaV genome is another interesting genome feature, which could be associated with altered dinucleotide frequencies. The association could be tested by Markov modelling of the dinucleotide and codon frequencies in the SaV genomes, although the small genome sizes and the presently small number of complete genomes would limit the usefulness of this approach [48] . The novel SaV described presently is the first known member of the Caliciviridae in bats. The approach to its discovery is based on the established strategy of targeted genetic screening informed by conserved sequences of related viruses. Although this ''homologybased'' strategy has been successful in the discovery of numerous viruses, the advent of affordable high-density microarrays and high-thoughput sequencing has given rise to virus discovery through metagenomics. Indeed, the first canine SaVs were discovered recently by metagenome sequencing of canine diarrhoea samples on a high-throughput pyrosequencer [23] . Important advantages of the new method include detection of novel viruses not closely related to known viruses, and the capacity to detect multiple divergent viruses in cases of co-infection. However, metagenomics sequencing can suffer from possible bias during sample preparation [49] , and it is unlikely to detect very low titres of viruses in a specimen, such as the three bat faecal samples that were positive upon repeat screening with specific PCR primers in the present study. While we anticipate the increasing utilization of the metagenomics approach, existing methods such as viral culture, electron microscopy and targeted nucleic acid amplification would continue to serve important roles in virus discovery. As Hong Kong is a highly urbanized city, the local roosting sites of bats are mainly man-made structures, such as water tunnels and abandoned mines. Hipposideros pomona is very common and widespread throughout Hong Kong countryside areas. It is a small-sized leaf-nosed bat with body weight ranged from 6-8 g. It possesses a small nose leaf which is simple, small, and lacking of lateral leaflets (Figures S3 and S4 ). This species may aggregate in small chambers or enclosures where the air flow is relatively limited. The 5 SaV-infected specimens were all captured in a place called Tai Lam -Shek Kong located next to a major country park of Hong Kong, and this roosting site shares similar ecological characteristic with other sampled roosting sites. Due to the extremely high human population density in Hong Kong, direct contact between humans and bats is relatively frequent. Fortunately, no local case of bat zoonosis has ever been reported [36] . The relatively large genetic distance between the present bat SaV and other mammalian SaVs suggests that the zoonotic risk posed by this virus is likely to be low, though this should be confirmed with further in vitro and in vivo studies. There are two main limitations in the current study. First and foremost, clinical information on the sampled bats is limited to the brief period of captivity needed for sample collection, which is unlikely to reflect the disease association of the virus accurately. In other words, the scope of the study is limited to surveillance of viral diversity and possible discovery of new viruses. Secondly, the number of samples for the novel virus is quite small, despite the use of specific PCR primers for screening and the relatively large number of samples collected. Thus, we were unable to draw conclusions on the seasonality of its detection or its host specificity. To address these limitations, long-term follow-up studies would be required to identify sufficient positive samples with associated clinical data. Increasing the scale of surveillance would also help, though there are practical geographical and logistic constraints in our locality. In conclusion, we identified a novel bat SaV with several genomic features and properties that set it apart from other members of the genus Sapovirus. Phylogenetic analysis suggests that its ancestral lineage had diverged early from the other SaVs and evolved under different conditions. Further discovery and characterization of additional strains would enhance our understanding of the evolutionary history of the SaVs and other caliciviruses. The study was approved by the Department of Agriculture, Fisheries and Conservation, HKSAR; and Committee on the Use of Live Animals in Teaching and Research, The University of Hong Kong. Bats from 14 different locations in rural areas of Hong Kong, including water tunnels, closed mines, sea caves and forested areas, were captured over a 36-month period. Anal swabs were collected by an experienced veterinary surgeon, and kept in viral transport medium at 4uC before processing. Viral RNA was extracted from the anal swabs using a QIAamp Viral RNA mini kit (Qiagen). The RNA was eluted into 50 ml RNase-free water and was used as the template for RT-PCR. Screening was performed by amplifying a 185 nt fragment in the RdRp region of the ORF1 gene of caliciviruses. Conserved degenerate primers (59-GAYTAYTCNMRRTGGGAYTC-39 and 59-GGCATNCCNGAKGGNAYNCC -39) were designed from the multiple sequence alignment of the available calicivirus gene sequences in NCBI GenBank. First-strand cDNA synthesis was performed using SuperScript III kit (Invitrogen) according to manufacturer's instructions. The PCR mixture (25 ml) contained cDNA, PCR buffer (10 mM Tris/HCl pH 8.3, 50 mM KCl, 2 mM MgCl 2 and 0.01% gelatin), 200 mM of each dNTP and 1.0 U AmpliTaq Gold polymerase (Applied Biosystems). PCR cycling conditions were as follows: hot start at 94uC for 7 min, followed by 50 cycles of 94uC for 1 min, 50uC for 1 min and 72uC for 1 min with a final extension at 72uC for 10 min in an automated thermal cycler (Applied Biosystems). Standard precau-tions were taken to avoid PCR contamination and no false-positive signal was observed in the negative controls. The PCR products were gel-purified using a QIAquick gel extraction kit (Qiagen). Both strands of the PCR products were sequenced twice with an ABI Prism 3730xl DNA Analyser (Applied Biosystems), using the two PCR primers. Additional RT-PCR screening was performed on the same samples using specific primers designed from the RdRp nucleotide sequences of bat SaVs obtained from previous rounds of RT-PCR and sequencing, as RT-PCR screening with specific primers usually offers higher sensitivity than a comparable screening with consensus degenerate primers. Sequences of the specific primers are as follows: forward primer 59-CACAATGCAGCCAGCCA-39 and reverse primer 59-GGTGCGCGTGGTGAACAC-39. PCR cycling conditions were as follows: hot start at 94uC for 7 min, followed by 50 cycles of 94uC for 1 min, 52uC for 1 min and 72uC for 1 min with a final extension at 72uC for 10 min in an automated thermal cycler (Applied Biosystems). Standard precautions were taken to avoid PCR contamination and no falsepositive signal was observed in the negative controls. PCR product purification and sequencing were performed as above. Purified PCR products were cloned into a pCR2.1-TOPO vector (Invitrogen) according to manufacturer's instructions. The vector was then used to transform the competent Escherichia coli strain DH5a by electroporation. Positive transformants were identified by blue-white screening, and eight colonies were selected for DNA sequencing of the construct using the M13 forward and reverse primers according to the manufacturer's instructions. Sequencing reactions were performed as described above. Viral genome sequences were obtained using strategies we had previously used for other RNA viruses [50] [51] [52] . RNA extraction and cDNA generation were performed as described above. PCR primers were designed by targeting conserved regions, which were identified from the multiple alignment of genomes of related SaVs, as primer-binding sites. Additional primers for subsequent rounds of PCR were designed based on the results of earlier rounds of genome sequencing. The complete set of primer sequences is available from the authors upon request. The 59 and 39 ends of the viral genomes were sequenced following amplification of the segments by rapid amplification of cDNA ends, which was performed using the SMARTer RACE cDNA Amplification kit (Clontech) according to the manufacturer's instructions. ORFs were located using the ORF Finder tool at NCBI (http:// www.ncbi.nlm.nih.gov/projects/gorf/). Annotation of the predicted proteins was performed by BLAST sequence similarity search against annotations in the NCBI RefSeq database. Multiple sequence alignments were constructed using MUSCLE version 3.8.31 [53] , and phylogenetic informative regions were extracted using BMGE [54] . Maximum-likelihood phylogenetic trees were constructed using PhyML version 3 [55] , under the best-fit protein evolution model as selected by ProtTest 3 [56] . Branch support values were estimated by calculation of SH-like aLRT values [57] . Recombination detection was performed by analysing the translated sequences of ORF1 and ORF2 separately using the Recombination Analysis Tool [58] . The full-length ORF1 and ORF2 coding sequences were extracted from selected SaV genomes and concatenated for codon usage analysis (see Table 4 for the list of included genome sequences). Codon usage and summary statistic of codon usage bias (N c and N c 9) were calculated using the INCA package version 2.1 [59] , where N c is the effective number of codons in the coding regions of the genome [60] , and N c 9 is the effective number of codons adjusted for background nucleotide composition [61] . For CpG dinucleotide bias analysis, odds ratio of CpG and GpC dinucleotides and the CpG/GpC ratio were calculated as described in previous studies [40, 41] . Odds ratio of #0.78 indicates significant suppression of the dinucleotide, same as the interpretation criteria of previous studies. Symmetrized nucleotide frequencies and dinucleotide odds ratio were not considered in the present study, as SaV genomes consist of positive-sense ssRNA only. To investigate the possible effects of cytosine deamination, genomic GC skew, which is the ratio (G-C)/(G+C), was calculated for the SaV genomes. The strength of the GC skew had been suggested to correlate with the degree of cytosine deamination [41, 44, 62] .

Angiotensin Converting Enzyme (ACE) and ACE2 Bind Integrins and ACE2 Regulates Integrin Signalling

The angiotensin converting enzymes (ACEs) are the key catalytic components of the renin-angiotensin system, mediating precise regulation of blood pressure by counterbalancing the effects of each other. Inhibition of ACE has been shown to improve pathology in cardiovascular disease, whilst ACE2 is cardioprotective in the failing heart. However, the mechanisms by which ACE2 mediates its cardioprotective functions have yet to be fully elucidated. Here we demonstrate that both ACE and ACE2 bind integrin subunits, in an RGD-independent manner, and that they can act as cell adhesion substrates. We show that cellular expression of ACE2 enhanced cell adhesion. Furthermore, we present evidence that soluble ACE2 (sACE2) is capable of suppressing integrin signalling mediated by FAK. In addition, sACE2 increases the expression of Akt, thereby lowering the proportion of the signalling molecule phosphorylated Akt. These results suggest that ACE2 plays a role in cell-cell interactions, possibly acting to fine-tune integrin signalling. Hence the expression and cleavage of ACE2 at the plasma membrane may influence cell-extracellular matrix interactions and the signalling that mediates cell survival and proliferation. As such, ectodomain shedding of ACE2 may play a role in the process of pathological cardiac remodelling.

Heart failure is characterised as a decline in cardiac contractility, which is associated with structural changes, collectively termed cardiac remodelling. Cardiac myofibroblasts are key mediators of cardiac remodelling via their proliferation, invasion and secretion of extracellular matrix proteins. Angiotensin II (Ang II) stimulates cardiac myofibroblast transdifferentiation leading to fibrosis. Ang II also stimulates proliferation [1] , NADPH oxidase activation [2] (and thereby reactive oxygen species production), the production of proinflammatory cytokines [3] and the activation of matrix metalloproteinases (MMPs) [4] . As a result, Ang II is a major contributor to the pathology of cardiovascular diseases. Ang II is generated from the biologically inert peptide, Ang I, by the catalytic action of angiotensin converting enzyme (ACE), a key proteolytic step in the renin angiotensin system (RAS). Aberrant functioning of the RAS is a feature of a variety of cardiovascular, renal and other pathologies and ACE inhibitors and Ang II receptor 1 (AT1R) antagonists are widely used in the clinic. Accordingly, ACE inhibition has been shown to prevent cardiac remodelling after myocardial infarction (MI) and preserves cardiac function [5, 6] . A combination of ACE inhibitors and AT1R blockers has been shown to be more effective than either alone [7] . A decade ago a new member of this system was identified, termed angiotensin converting enzyme 2 (ACE2) [8, 9] . ACE2 acts to hydrolyse Ang II into the vasodilator Ang-(1-7), thereby contributing to reductions in blood pressure. Current models of the RAS are based on the concept that the two enzymes counterbalance each other. The balance between the two angiotensin converting enzymes has been highlighted by ACE2 deletion murine models, which have a significantly higher mortality rate post-MI than wild-type mice. Mortality was associated with enhanced adverse ventricular remodelling following MI [10] , a state which was reversed by the use of an AT1R blocker and as such the pathology of ACE2 deletion was attributed to the increased levels of Ang II [10] . A mounting body of evidence is forming in support of a cardioprotective role for ACE2, through the metabolism of Ang II [10, 11] , but also through the direct action of Ang-(1-7) via its own receptor, Mas [12] . Like Ang II the actions of Ang-(1-7) extend beyond vasopressor control, and for the most part appear to counteract the effects of Ang II and therefore mediate cardioprotection [13] . Ang-(1-7) reduces interstitial fibrosis [14] , myocyte hypertrophy [15] and inhibits myocyte cell growth [16] . The reduction in myocyte hypertrophy resulting from expression of Ang-(1-7) was associated with a decrease in pro-inflammatory cytokines (TNF-a and IL-6) and also a reduction in exogenous ACE transcript [17] . Both ACE and ACE2 are increased in the failing heart [18] [19] [20] . Over-expression of ACE2 and inhibition of ACE exert a protective influence on the heart post-myocardial infarction (MI) and prevent the pathological remodelling [21] . These data together suggest that the regulated activity of angiotensin converting enzymes may play a role in cardiac homeostasis. In addition to its catalytic actions, ACE2 is the cellular receptor for the SARS virus; more recently, other regulatory actions of ACE2 through protein-protein interactions have been identified [22] . ACE2 acts as a chaperone protein for the neutral amino acid transporter, B 0 AT, [23] and binding of calmodulin to the cytoplasmic tail of ACE2 regulates its retention on the cell surface [24] . The reported observation that ACE2 binds integrin b1 (ITGB1) in the failing human heart [25] adds another dimension to the role of the ACE family in cardiac homeostasis. Integrins are a family of ab heterodimeric cell surface receptors, which link extracellular matrix proteins with the intracellular cytoskeleton. Integrins have an important role in the regulation of gene expression, cell proliferation, differentiation, migration and apoptosis. Activated myofibroblasts develop specialised focal adhesions, containing high levels of a5, b1 and b3 integrins [26] . ITGB1 serves as a mechanotransducer and expression of this integrin increases in the heart after MI [27] . ITGB1 is highly implicated in left ventricular remodelling and MI models have shown it is essential for the adaptive remodelling response [28] leading to the suggestion that the functional activity of ACE2 in cardiac remodelling may, at least in part, be mediated through ITGB1. Both the angiotensin converting enzymes and ITGB1 traverse the plasma membrane; the angiotensin converting enzymes also exist in soluble forms in the plasma when shed from the cell membrane. Recent studies of heart failure patients have linked elevated levels of soluble (shed) ACE2 (sACE2) to increased myocardial dysfunction and thus have indirectly identified a protective role for the cell surface-associated form [29] . Hence, the ectodomain of ACE2 may have a role in cardiac remodelling independent of its catalytic activity. Here we show that both angiotensin converting enzymes bind ITGB1 and integrin a5 (ITGA5) in vitro. The data presented provide evidence that the enzymes provide a substrate for cellular adhesion and that the interaction between ACE2 and an integrin increases cellular adhesion when both are present on the cell membrane. We demonstrate that the ectodomain of ACE2 regulates integrin induced cell signalling via modulation of the phosphorylation of focal adhesion kinase (FAK) and Akt expression levels. A typical integrin binding motif is the tripeptide sequence RGD. Bioinformatic analysis was used to compare the protein sequences of ACE and ACE2; a highly conserved integrin binding domain was identified in the ectodomain of ACE2 but not ACE ( Figure 1A ). There are two isoforms of ACE: somatic ACE, that contains two homologous catalytic ectodomains (the N and Cdomains) and the testicular ACE (tACE), composed solely of the C-domain. The RGD sequence in ACE2 is replaced by the sequence RSW in the ACE N-domain and RSM in the C-domain. To assess the ability of ACE2 to bind an integrin, immunoprecipitation was performed on HEK cells over-expressing ACE2 (HEK-ACE2 cells). Anti-ITGB1 antibody was used to pull down any interacting proteins and western blotting performed using an anti-ACE2 antibody ( Figure 1B ). An interaction was found to occur in the HEK-ACE2 cells, represented by a single band at 120 kDa corresponding to the fully glycosylated ACE2 protein ( Figure 1B ). This interaction was specific, since ACE2 was not detected when immunoprecipitation was performed with isotype control IgG antibody. Immunoprecipitation was repeated in Huh7 cells, which endogenously express ACE2. Crosslinking was performed in order to fix any interaction of less than 9 Å in length. An interaction between ACE2 and ITGB1 was readily detected in these cells ( Figure 1C ). In addition an interaction between ACE2 and the common binding partner of ITGB1 in cardiac tissue, ITGA5, was probed by subjecting cell lysates to immunoprecipitation with antiintegrin a5 (ITGA5) antibody. ITGA5 was also found to bind to ACE2 in Huh7 cells ( Figure 1C ). The subcellular location of both ACE2 and ITGB1 was visualised in HEK-ACE2 cells by immunofluorescence microscopy ( Figure 1D ). Antibodies to ACE2 (green) and ITGB1 (red) located both proteins to the plasma membrane. Co-location of the two proteins was observed in some areas (yellow), highlighted by arrows. Both isoforms of ACE lack the RGD motif present in the extracellular domain of ACE2 and we therefore hypothesised that ACE would not bind integrins. Cells over-expressing tACE were used to examine any potential interaction between ACE and ITGB1. Immunoprecipitation revealed that ACE does bind ITGB1 ( Figure 2A ) and that ACE also binds ITGA5 similarly ( Figure 2B ). In order to clarify any role of the RGD motif in the binding of ACE2 to integrins, cross-linked immunoprecipitation was repeated in Huh7 cells, in the presence and absence of an RGD peptide. The interaction between ACE2 and ITGB1 was not blocked by the presence of RGD peptide ( Figure 2C ). Hence, the ability of ACE2 to bind the RGD-independent integrin subunit ITGA2 was additionally investigated. ACE2 bound ITGA2 at a comparable level to ITGA5 and ITGB1 ( Figure 2D ). We utilised molecular modelling to ascertain the location of the RGD motif in the structure of ACE2 to investigate the apparent functional redundancy of this motif in integrin binding. Examination of the ACE2 structure in silico [30] revealed that the motif was on the protein surface ( Figure 3A ). However, the space filling model revealed that the aspartate residue of the RGD motif faces into the active site cleft ( Figure 3B and 3C) and is, therefore, inaccessible for protein-protein interactions. The RSM sequence in tACE superimposes in exactly the same position as the RGD sequence in ACE2 and is therefore also inaccessible for proteinprotein interactions ( Figure 3D ). The similarity between the two proteins is illustrated by the close proximity of the outline trace and the degree of overlap. Variations in amino acid sequences are illustrated by the slight offset in the a-helical loops. To examine the functional significance of an interaction between the angiotensin converting enzymes and integrins, adhesion assays were used to explore the possibility that they may act as a cell adhesion substrate. We designed an in vitro technique representative of an in vivo cellular environment to study the effect of ACE2 expression on cellular attachment. A cell to cell adhesion assay was developed in order to examine the ability of membrane bound ACE or ACE2 to act as a ligand for cell adhesion. Cells over-expressing ACE or ACE2 or their mock transfected controls were used as cell adhesion substrates and Huh7 cells were labelled with BCECF fluorescent dye and allowed to adhere. Huh7 cell adhesion to the substrate cells was confirmed by immunofluorescence microscopy ( Figure 4A ). Calibrations confirmed that the relative fluorescence measured was proportional to the number of cells seeded ( Figure 4B ). A significant difference was seen between the adhesion of Huh7 cells to ACE2-expressing cells compared with non ACE2expressing cells ( Figure 4C ); the expression of ACE2 on the cell surface increased cell adhesion by approximately 25%. To further examine any role of RGD-mediated cell adhesion, cells were incubated with an RGD peptide prior to plating. Pre-incubation with RGD peptide significantly reduced the adhesion of Huh7 to HEK control cells ( Figure 4C ); cellular expression of ACE2 abolished the decrease mediated by pre-incubation with an RGD peptide ( Figure 4C ). Conversely, cellular expression of ACE conferred no enhanced adhesion properties compared to control cells ( Figure 4D ). In this model the presence of an RGD peptide reduced cellular adhesion by approximately 35% independent of tACE expression ( Figure 4D ). To examine the physiological importance of the adhesion properties of ACE2 and ACE, adhesion assays were performed using primary human cardiac myofibroblast (CF) cells, important mediators of cardiac remodelling. All cells adhered strongly to fibronectin (data not shown), an important component of the extracellular matrix, adhesion to which is integrin-mediated [31] . A significant difference in cell binding was seen in the presence and absence of ACE2 ( Figure 5A ) in patient samples. The average fold increase in cell adhesion in the presence of ACE2 was 3.9 fold and p = 0.0035. Differences in adhesion to both fibronectin and ACE2 were seen in all patient cells ( Figure 5A ); this is likely due to an inherent variation in integrin and/or ACE2 expression levels since these were primary cells [32] . As previous experiments had shown ACE2 and ACE both bind integrins, investigations were performed to determine if ACE could also act as a cell substrate. Cells adhered comparably to both ACE and ACE2 ( Figure 5B ) and the presence of an RGD peptide had no effect on the adhesion to ACE or ACE2 (data not shown). In light of the association of ACE2 with integrins, experiments were performed to determine if ACE2 could elicit integrin signalling. Focal adhesion kinase (FAK) is stimulated early in any integrin signalling cascade. Given that the extracellular domain of ACE2 binds integrin, cells were stimulated with the ectodomain of ACE2. The levels of phosphorylated FAK (pFAK) in the presence and absence of sACE2 were quantified by ELISA. At 0.1 mg/ml sACE2 significantly reduced levels of pFAK in Huh7 cells and in primary myofibroblasts ( Figure 6A and B). No further decrease was observed when cells were incubated with 1 mg/ml ACE2. Downstream translation of this signal to Akt was investigated by quantifying the levels of phosphorylated Akt. Western blot analysis revealed that levels of phosphorylated Akt in Huh7 cells increased in response to stimulation with sACE2 and Ang II for 30 min (data not shown). However, this increase was accounted for by an up-regulation in Akt protein expression ( Figure 6 ). The catalytic product of ACE2, Ang-(1-7), alone did not elicit the same effects as ACE2 ( Figure 6 ). Further analysis revealed that this signal was not transmitted to the downstream effector of Akt signalling, NF-kB. Cells were transfected with NF-kB reporter vector. Treatment of cells with Figure 1 . ACE2 binds an integrin. A) Evolutionary alignment of a part of the ACE2 amino acid sequence. ACE2 protein sequence conservation surrounding and including the proposed integrin binding site. Amino acids are shown as their one letter codes. The predicted integrin binding site is highlighted in red, whilst the homologous region in ACE is coloured blue. This alignment of the ACE2 protein sequence was taken from the Uniprot database. B) Immunoprecipitation of ITGB1 with ACE2 in HEK-ACE2. HEK cell lysates incubated with ITGB1 antibody and eluted using protein G. Immunoblotting for ACE2 was performed with anti-ACE2 antibody. C) Immunoprecipitation of ITGB1 and ITGA5 with ACE2 in Huh7 cell monolayers. Cell monolayers were cross-linked using DTBP before lysis and immunoprecipitation as before. D) Immunocytochemical detection of ACE2 and ITGB1 location in HEK-ACE2 cells. Cell imaging shows ACE2 (green) and ITGB1 (red) are located together (yellow) on the cell membrane of HEK-ACE2 cells. doi:10.1371/journal.pone.0034747.g001 sACE2 or ACE for between 2 and 24 h resulted in no significant change in luminescence compared to control (Figure 7 ). Furthermore, neither Ang II nor Ang-(1-7) had any significant effect (data not shown). The present study reveals that both ACE and ACE2 bind integrins in an RGD-independent manner. We have shown that ACE2, in particular, increases cellular adhesion and, moreover, affects integrin signalling. Shedding of the ACE2 protein may relieve repression of integrin signalling, exerted by the presence of ACE2 on the cell membrane. Using primary cardiac myofibroblasts we have demonstrated that the actions of ACE and ACE2 exerted on cell models are physiologically relevant to the diseased human heart. Both ACE and ACE2 are increased in the failing heart [18] [19] [20] . ACE2 expression has consistently been seen to be up-regulated in the peri-infarct area after MI [10, 20, 33] and in end stage heart failure [34] indicative of a role in injury. Knockout of the ace2 gene increases MMP2 and MMP9 levels in the peri-infarct region of mice, resulting in disruption of the extracellular matrix and enhanced adverse remodelling [10] . ACE2 over-expression has been shown to inhibit collagen production in response to hypoxic injury [35] . Moreover, activation of remodelling pathways has been demonstrated in ace2 knockout animals in the absence of an increase in Ang II [36] . Cardiac remodelling is a key pathological process in the development of heart failure. Integrins play a key role in this process by mediating cell-ECM interactions and cellular signalling. ITGB1 has also been implicated in myocardial dysfunction [28] . An association between ITGB1 and ACE2 has previously been reported in the failing heart and attributed to the presence of an RGD motif in ACE2 [25] . We have established that both ACE and ACE2 binds ITGB1 and also its common cardiac binding partner, ITGA5 [37] , as well as the RGD-independent and liver rich ITGA2 [38] . However, in contrast to the study of Lin et al. [25] our data clearly suggest that these interactions occur independently of an RGD motif. 3-dimensional modelling has demonstrated that the RGD motif present in the ectodomain of ACE2 is inaccessible, which explains its redundancy in integrin binding. The aspartate residue is positioned facing the active site cleft of the protein and as such is not available to bind into the integrin binding pocket [39] . Given the structural homology between the two proteins, it is not surprising that ACE is also capable of binding integrins as the interaction appears to be independent of the RGD sequence, which is lacking in ACE. Prothrombin, like ACE2, contains a partially buried RGD motif, however in prothrombin this sequence is exposed upon activation. The ability of prothrombin to bind integrin avb3 is key to its biological activity in the fibrotic cascade [40] . As in ACE2, the carboxylate group of the Asp of the RGD motif is directed towards the specificity pocket of the enzyme [41] . In its native state prothrombin does not exert strong adhesive properties. However, proteolytic maturation exposes the RGD motif and thereby enhances the adhesive properties of thrombin [42] . This mechanism is unlikely to occur in ACE2 which, when present on the cell membrane, is in its mature form (unlike thrombin it does not occur in an inactive proenzyme form during biosynthesis). What is more, any rearrangement of the active site of ACE2 is likely to inhibit its catalytic activity; sACE2 is catalytically active. Although RGD motifs are the most common mechanism of integrin binding, there are other cell surface proteins which bind integrins despite lacking an RGD motif. ADAM 9, for example, binds through a hypervariable loop stabilised by disulphide bridges, which protrudes from the surface of the protein structure [43, 44] . The cellular adhesion molecule ICAM-1 is hypothesised to bind via immunoglobulin type domains [45] , whereas other proteins have been predicted to bind through a (D/E)ECD motif [46] . Neither ACE nor ACE2 contain an ECD motif. We have shown that the integrin binding to ACE2, but not ACE, is essential for the role of ACE2 as a cellular anchor when expressed on the cell surface. What is more, the presence of ACE2 on the cell surface partially removed the requirement of RGDinteractions for cellular adhesion. Fibroblast motility is a key process in the development of scar tissue and thus cellular adhesion is an important homeostatic mechanism. In order to study the role of ACE and ACE2 in the remodelling response, primary cardiac myofibroblasts were used as a disease model. These cells play a key role in the maintenance of cardiac architecture under conditions of injury, by forming scar tissue through their ability to proliferate and adhere [47] . We demonstrate that both ACE and ACE2 exert comparable effects over myofibroblast adhesion, but that again this effect is not mediated through an RGD motif. Recent clinical investigations have highlighted that elevated levels of sACE2 in patient plasma correlated with increased myocardial dysfunction [29] and vascular compliactions in type 1 diabetes [48] . We have shown that both ACE and ACE2 interact with the cell surface in adhesion assays and, furthermore, ACE2, in particular, enhances cell adhesion and may modulate integrin signalling. Cellular retention of ACE2 is therefore required for its role as a cellular anchor and, as such, the cleavage of ACE2 by ADAM17 [49] may be a pathological step in the development of heart failure. The Figure 2 . Both angiotensin converting enzymes bind integrins independent of an RGD motif. A) Immunoprecipitation of ITGB1 with tACE in HEK-tACE cells. HEK cell lysates were incubated with ITGB1 antibody and eluted using protein G. Immunoblotting for ACE was performed with anti-ACE mC5 antibody. This interaction was similar to that between ACE2 and ITGB1 in ACE2 over-expressing cells. B) Immunoprecipitation of ITGB1 and ITGA5 with tACE in SHSY5Y cells over-expressing the testicular form of ACE. Immunoprecipitation was performed as before. C) Immunoprecipitation of ACE2 with ITGB1 in the presence of a RGD peptide. Huh7 cells were pre-incubated and crosslinked using DTBP, in the presence or absence of RGD peptide before lysis and immunoprecipitation as before. D) Immunoprecipitation of ITGA2, ITGA5, ITGB1 with ACE2 in Huh7 cell monolayers. Cell monolayers were cross-linked using DTBP before lysis and immunoprecipitation as before. doi:10.1371/journal.pone.0034747.g002 retention of ACE2 on the cell membrane is known to be regulated by cell signalling [24] and viral infection [50] . Shed ACE2 could activate integrins by binding to them and transducing activating signals or, additionally, by interacting with non-integrin sites given the multiple protein-protein interactions with which the ACE2 protein is involved. FAK is a critical signalling component associated with areas of substratum adhesion; signalling via FAK is mediated through autophosphorylation of Tyr 397 [51] . We demonstrate that sACE2, at levels comparable to those reported in human plasma, significantly reduces FAK phosphorylation levels. Furthermore, we show that treatment with sACE2 increases the levels of Akt expression, a pro-survival, pro-proliferative protein. Changes in the level of phosphorylated Akt were also seen in response to sACE2; however, these were accounted for by the increase in the amount of total Akt. As such, signalling by Akt was not transmitted to its downstream effector NF-kB. sACE2 has no known function. We do not dismiss the possibility that this circulating ACE2 may bind integrins, when released from the plasma membrane, and elicit autocrine or paracrine signalling. In fact, one of the shed forms of the amyloid precursor protein, sAPPa, binds to ITGB1 and signals to enhance axon outgrowth [52] . What is more, cellular expression of full length APP inhibits axonal outgrowth and an excess of the shed form overcomes this inhibition [52] . We therefore propose some of the anti-proliferative actions of ACE2 may in part be mediated through a non-catalytic interaction with integrins, rather than by metabolism of Ang II per se. These data suggest that ACE2, through its integrin binding abilities, may have regulatory roles in cellular attachment and support a novel mechanism of integrin activation upon ACE2 shedding. All routinely used reagents were purchased from Sigma unless otherwise stated. Cell culture reagents were purchased from Lonza (Slough, UK). Cell Extraction Buffer and Elisa (pFAK) kit were purchased from Invitrogen (Paisley, UK) along with lipofectamine and BCECF reagent (2979-bis-(2-carboxyethyl)-5-(and6)-carboxyfluorescein). Roche (Welwyn, UK) supplied protease inhibitor tablets. Fluorescent ACE2 substrate Mca-APK-Dpn was supplied by Enzo (Exeter, UK) and MTS reagent by Promega (Southampton, UK). Secondary antibodies, the chemiluminescence system used and Protein G Sepharose 4 fast flow were supplied by GE healthcare (Chalfont St. Giles, UK). DTBP (Dimethyl 3,39dithiobispropionimidate) was purchased from Pierce (Cramlington, UK). The ACE2 inhibitor 416F2 [53] was a generous gift from Prof V. Dive (CEA, Gif sur Yvette, France). Polyclonal ACE2 antibody raised in goat was purchased from R&D systems (Abingdon, UK). Integrin antibodies raised in mice against ITGB1 and ITGA5 antibodies were bought from Santa Cruz Biotech (Heidelberg, Germany). Polyclonal ADAM17 antibody raised in rabbit was purchased from Calbiochem (Nottingham, UK), while Akt antibody raised in rabbit was purchased from Cell Signaling (Hertfordshire, UK). Purified ACE enzyme was a kind gift from Prof N. Hooper (The University of Leeds, UK) and Prof S. Danilov (University of Illinois at Chicago, USA) generously provided the ACE monoclonal antibody [54, 55] . HEK (human embryonic kidney) and Huh7 (hepatocellular carcinoma-derived) [24] cells were cultured in Dulbecco's modified Eagle's medium (DMEM), supplemented with 10% (v/ v) foetal bovine serum, 2 mM essential amino acids, 1% (v/v) nonessential amino acids. HEK cells stably transfected with full length ACE2, designated HEK-ACE2 [49] , and those overexpressing the testicular form of ACE (tACE), HEK-tACE [56] , were cultured in the same conditions with the addition of G418 (0.5 mg/ml) to the medium. SHSY5Y cells were cultured in DMEM-F12 media [57] , and SHSY5Y cells overexpressing tACE, SHSY5Y-tACE cells, were a kind gift from Dr. C. Rushworth, and were cultured in DMEM-F12 supplemented with G418 (0.5 mg/ml). Cardiac myofibroblasts were obtained by enzymatic digestion of biopsies of human right atrial appendage. Patients were undergoing elective coronary artery bypass surgery and had normal ventricular function (ejection fraction normal ($50% by cardiac ultrasound and/or LV Angiography). Local (Leeds West) Research Ethical Committee (LREC) approval is in place for this study; written consent was given, reference number 01/040. Informed, written patient consent is obtained. The investigation conformed to the principles outlined in the Declaration of Helsinki, 1997. Primary cultures of cardiac fibroblasts were harvested, characterized as myofibroblasts co-expression of smooth muscle a-actin and vimentin and cultured as described previously [58] . Experiments were performed on cells from different patients at passages 2-5 [59] . Cell images were taken on a Nikon Eclipse TS100 microscope using a Nikon COOLPIX 4500 4.0 megapixel camera. Recombinant ACE2 purification HEK cells stably expressing a FLAG-tagged ACE2 ectodomain (sACE2) were created. HEK cells were transfected with plasmid DNA (pCl-neo containing nucleotides 104-2323 of ACE2 cDNA with the FLAG peptide conjugated to the C-terminus). Successfully transfected cells were selected by passage in media containing G418 (1 mg/ml). sACE2 was collected from the conditioned media of these cells and purified by affinity chromatography using an anti-FLAG M2-agarose column and eluted into tubes containing 25 ml 1M Tris pH 8.0 by addition of 0.1 M glycine, pH 3.5. Eluted fractions were analysed for ACE2 activity by fluorometric assay [24] and purity was checked by silver stain, using SilverXpress (Invitrogen) as per manufacturer's instructions, and then immunoblotted for ACE2. Cells were treated at 80% confluency and all pharmacological reagents were diluted in OptiMEM. For pFAK quantifications and phospho-Akt quantifications, incubations with sACE2 were carried out with 100 ng/ml or 1 mg/ml sACE2 for the time indicated (pFAK, 20 min). After treatment, cells were placed on ice and lysed in either ice cold RIPA buffer containing protease inhibitor and phosphoSTOP (phosphor-Akt), or Cell Extraction Buffer (pFAK). Lysates were then analysed by pFAK ELISA as per manufacturer's instructions, or by western blot for Akt levels. Huh7 cells were transfected with pGL4.32 [luc2P/NF-kB-RE/ Hygro] (NF-kB reporter) vector (500 ng) using Lipofectamine 2000 in OptiMEM. Renilla CMV (1 ng) was co-transfected. Cell medium was changed 4 h post-transfection and stimulated after 24 h with either sACE2/ACE, 100 ng/ml, IL-1b, 100 ng/ml, diluted in DMEM containing 1% FCS. After treatment, luciferase levels were analysed using Dual-Luciferase Reporter Assay System following manufacturer's instructions. Lysates were routinely prepared by solubilisation of cells in RIPA buffer (0.1 M Tris-HCl, pH 7.4, 0.15 M NaCl, 1% (v/v) Triton X-100, 0.1% (v/v) Nonidet P-40). Protein concentrations were determined by bicinchoninic acid (BCA) protein assay [60] . Bovine serum albumin was used as a standard with a 50:1 ratio of 4% (w/v) CuSO 4 .5H 2 O. All cells were washed twice and scraped into ice cold PBS, where crosslinking were preformed Huh7 cells were cross-linked with dimethyl 3,39 dithiopropionimidate (DTBP, 5 mM) for 30 min on ice prior to scraping. Cells were pelleted before resuspending in ice-cold RIPA lysis buffer (0.4% (v/v) with proteinase inhibitor cocktail). Lysates were passed through a 22G needle 5 times and re-cleared by centrifugation at 116006g for 2 min. For immunoprecipitation, protein-G-Sepharose was pre-cleared by rotation in 5% (w/v) BSA in TBS for 1 h at 4uC, prior to washing 3 times with protein binding buffer (50 mM Tris-HCl pH 7, 50 mM NaCl, 1 mM ZnSO 4 ). Samples were incubated with monoclonal anti-b1 integrin antibody (4uC, overnight), pre-cleared protein-G-Sepharose was then added and samples rotated (2 h, 4uC). Bound samples were eluted by heating to 85uC with 1xSDS-PAGE sample buffer. Sepharose beads were pelleted and the eluted supernatant heated to 95uC with b-mercaptoethanol. Where cells had been crosslinked, crosslinking was denatured by heating with DTT before elution (30 min, 37uC). Proteins were separated by SDS-PAGE and then transferred onto PVDF membranes using 5% (v/v) transfer buffer, 20% (v/v) Figure 5 . The angiotensin converting enzymes are cell adhesion substrates. Microtitre plates were coated with or without sACE2 and blocked with BSA (1%). Wells were washed; CmF cells plated and allowed to adhere (2.5 h at 37uC). After incubation unadhered cells were washed off and cell adhesion was quantified using MTS reagent and reference to a calibration curve. A) Tabulated cell adhesion results for different patients. B) Graphical representation of (A). C) Cell adhesion is comparable between ACE and ACE2. Plates were coated with sACE2, ACE or PBS and adhesion assays performed as before. doi:10.1371/journal.pone.0034747.g005 methanol. The membrane was saturated with blocking solution (TBS 0.1% (v/v) Tween 20, 2% (w/v) BSA, 5% (w/v) dried milk) (1 h, room temp). Membranes were incubated with primary antibody (4uC, overnight). After washing with TBST four times at 10 min intervals the membranes were incubated with secondary antibody for 1 h at room temperature and then washed as before. Bound antibody was detected using the enhanced chemiluminescence system following the manufacturer's instructions. Densitometric analysis was performed using AIDA software. Immunostaining HEK-ACE2 cells were plated onto coverslips and fixed with 4% (w/v) paraformaldehyde (10 min), washed twice with PBS and incubated with blocking buffer (5% BSA in PBS) for 30 min at room temp. Blocking buffer was removed and cells were placed in primary antibodies (ACE2 and ITGB1) for 2 h. Antibody binding was visualised using anti-goat Alexa Fluor 488 and anti-mouse Alexa Fluor 594 (Molecular Probes) for 2 h. Coverslips were mounted using Vectashield (Vector Laboratories Ltd.). Cells were imaged using a Delta Vision microscope and SoftWoRx software. Adhesion assays were carried out in 96 well plates. Wells were coated overnight (4uC) with protein (10 mg/ml) or PBS and washed in PBS. Wells were then blocked with 1% BSA in serum free media (1 h, 37uC.) Blocking solution was removed and wells washed twice with PBS. Cells were plated at a density of 10,000 or 20,000 cells per well in serum-free medium and allowed to attach by incubation at 37uC for 2.5 h. Non-adherent cells were removed by rinsing wells twice with PBS. Adhered cells were quantified using MTS reagent and measuring absorbance at 492 nm. Figure 6 . sACE2 inhibits integrin signalling. A &B) Treatment with sACE2 decreases the level of pFAK. Huh7 cells (A) or CF cells (B) were starved for 12 hours, then treated in the presence or absence of sACE2 at concentration indicated for 20 min. Significance determined by one-way ANOVA (Huh7) or Students' t-test (CF) where *p#0.05. Following treatment with sACE the level of pFAK within the cells was assayed using an ELISA kit as per manufacturer's instructions. C, D&E) Treatment with sACE2 increases cellular Akt expression. Cells were treated as above with either: sACE2, ACE, or Ang-(1-7), at the concentration and for the length of time indicated. Levels of Akt (C) and b-actin (B) were visualised by immunoblot and quantified by densitometric analysis (E). doi:10.1371/journal.pone.0034747.g006 Cell to cell adhesion assay HEK, HEK-ACE2 or HEK-tACE cells were seeded into a microtitre plate and starved in serum free medium (16 h). Huh7 cells were labelled with BCECF in OptiMEM (37uC, 30 min); labelled cells were washed three times with PBS. Huh7 cells were resuspended in serum free medium and plated onto the HEK, HEK-ACE2 or HEK-tACE cells for 2 h, non-adherent cells were removed by washing in PBS, PBS was added to each well and the fluorescence was read, (excitation 440 nm/emission 535 nm). Cell adhesion was examined using an inverted microscope (TE-2000E, Nikon) illuminated with a halogen lamp filtered through a GFP bandpass filter (450-480 nm excitation wavelength). Protein structures were taken from the PDB.org, file 1r42 and manipulated with Discovery Studio 2.0 (DS2.0, Accelrys Inc.). Turquoise, ACE2 extracellular domain, residues 1-615; red, tACE; yellow, RGD motif; Mln-4760, green; zinc, lilac; active site residues, pink; all other colours in the ribbon structure are sections of the C-terminal domain disordered in solution. In the spacing filling model: Overall surface, light pink; hydrogen bond acceptors, red; hydrogen bond donors, blue. Results are expressed as mean +/2 standard error of the mean (SEM). Significance was assessed by Student's t test or one-way ANOVA and p#0.05 was considered significant.

Altered Thymic Function during Interferon Therapy in HCV-Infected Patients

Interferon alpha (IFNα) therapy, despite good efficacy in curing HCV infection, leads to major side effects, in particular inducement of a strong peripheral T-cell lymphocytopenia. We here analyze the early consequences of IFNα therapy on both thymic function and peripheral T-cell homeostasis in patients in the acute or chronic phase of HCV-infection as well as in HIV/HCV co-infected patients. The evolution of T-cell subsets and T-cell homeostasis were estimated by flow cytometry while thymic function was measured through quantification of T-cell receptor excision circles (TREC) and estimation of intrathymic precursor T-cell proliferation during the first four months following the initiation of IFNα therapy. Beginning with the first month of therapy, a profound lymphocytopenia was observed for all T-cell subsets, including naïve T-cells and recent thymic emigrants (RTE), associated with inhibition of intrathymic precursor T-cell proliferation. Interleukin (IL)-7 plasma concentration rapidly dropped while lymphocytopenia progressed. This was neither a consequence of higher consumption of the cytokine nor due to its neutralization by soluble CD127. Decrease in IL-7 plasma concentration under IFNα therapy correlated with the decline in HCV viral load, thymic activity and RTE concentration in blood. These data demonstrate that IFNα-based therapy rapidly impacts on thymopoiesis and, consequently, perturbs T-cell homeostasis. Such a side effect might be detrimental for the continuation of IFNα therapy and may lead to an increased level of infectious risk, in particular in HIV/HCV co-infected patients. Altogether, this study suggests the therapeutic potential of IL-7 in the maintenance of peripheral T-cell homeostasis in IFNα-treated patients.

The hepatitis C virus (HCV) causes persistent infection in approximately two thirds of cases leading to chronic liver disease, liver failure, and, eventually, hepatocellular carcinoma in a substantial proportion of infected individuals. The most common therapy for chronic hepatitis C consists of pegylated interferon-a (IFNa) and ribavirin administration which results in viral clearance in 43-46% (genotype 1) to 80%, (genotype 3) of treated patients [1] . Interferon will continue to be a major component of new direct acting antivirals for treatment of HCV [2] . IFNa is produced in large amounts during the acute phase of many viral infections [3, 4, 5, 6] . Direct activation of interferonstimulated genes enhances naïve T-cell survival through increased Bcl-2 and reduced Bax activation [7] and contributes to clonal expansion of antigen-specific T-cells [8] . Recent data suggest that early therapeutic intervention with pegylated IFNa rescues polyfunctional memory T-cells expressing high levels of the IL-7 receptor alpha chain (CD127) and Bcl-2, allowing a higher rate of sustained viral response [9] . However, despite good efficacy, IFNa-based therapies lead to sustained anemia, thrombocytopenia, neutropenia and lymphocytopenia [10, 11, 12, 13, 14] . Moreover, pegylated IFNa therapy enhances the risk of infection in older HCV-infected patients and HIV-infected individuals, independently from neutropenia [15, 16, 17] . The mechanisms of action of IFNa include inhibition of different hematopoietic growth factors [18, 19] , possibly affecting lymphoid differentiation at an early stage [20] , and modifications in cell homing [12, 21, 22] . The mechanisms involved in IFNa therapy-associated leukocyte depletion remain poorly understood. Others and we have documented a strong reduction in the ability of HIV-infected patients to sustain thymic production as a direct consequence of a drop in intrathymic precursor T-cell proliferation [23, 24, 25] . Similar thymic impact was also seen during early SIV-infection in the rhesus macaque model [26] . The capacity of the thymus to produce recent thymic emigrants (RTEs) is, in large part, dependent on thymocyte proliferation [27] . Indeed, extensive thymocyte proliferation occurs between T-cell receptor beta (TCRB) and alpha (TCRA) chain rearrangements. The extent of this proliferation directly correlates with thymic export [28] . The extent of cell proliferation in the thymus can be measured in patients through estimating, in peripheral blood cells, the ratio (sj/bTREC ratio) between the frequency of signal joint T-cell receptor excision circles (sjTREC), produced during the excision of the TCRd locus prior to TCRa chain rearrangement, and that of DbJbTREC T-cell receptor excision circles (TRECs) produced during TCRBD to TCRBJ rearrangement [29] . These by-products of TCR rearrangement processes are generated by the circularization of the chromosomal DNA excised during TCR rearrangements respectively occurring at the DN3 (DbJbTREC) and DP (sjTREC) stages of differentiation. Due to the fact that TRECs do not replicate during mitosis, increased proliferation between DN3 and DP leads to the reduction of DbJbTREC frequency in RTEs as compared to sjTREC frequency and consequently to an increase of the sj/bTREC ratio [23] . The correlation between initial plasma IFNa levels and the speed of thymic dysfunction observed during HIV primary infection suggested that IFNa, produced as part of the innate immune response to infection, participates in the impairment of thymopoiesis. However, no direct evidence of the relationship between IFNa production and thymic dysfunction was provided by these studies. In contrast, Arizcorreta and colleagues showed that IFNa and ribavirin therapy induces a substantial reduction of circulating sjTRECs, in HIV/HCV co-infected patients, accompanied by sustained naïve CD4 + T-cell defect, suggesting thymic dysfunction [10] . Similarly, in the SIV-infected rhesus macaque model, we showed that IFNa therapy induced a strong decrease of circulating RTE numbers as defined either by sjTREC frequency and numbers or by CD31 hi expression on naïve T-cells [30] . Interestingly, in these animals, recombinant interleukin (IL)-7 therapy more than abrogated the deleterious effects of IFNa therapy [30] . IL-7 is a key cytokine implicated at various levels of thymocytes differentiation. It allows cell survival during the rearrangement processes, and is implicated in the extensive thymocyte proliferation, in particular in intermediate single positive (ISP) and early DP cells [31, 32, 33, 34] . This proliferation participates in the development of naïve T-cell diversity [35] . While up regulated by IFNa [36, 37] , the cyclin-dependent kinase inhibitor P27/Kip1 is negatively regulated by IL-7 [38] , allowing ISP and early DP thymocytes to proliferate. Moreover, IFNa also inhibit IL-7 dependent proliferation through down modulation of the common c chain, that participates, together with CD127 to the IL-7 receptor [39] . We here investigated the early impact of IFNa therapy on thymic function and naïve T-cell homeostasis in both HCV-infected and HIV/HCV co-infected patients who started IFNa therapy. We first evaluated the evolution of naïve T-cell subsets in three groups of HCV infected individuals: 1) Acute HCV infection (n = 8), defined as ,6 months post estimated date of infection; 2) chronic HCV infection (n = 8), defined as .6 months post estimated date of infection; and 3) HIV/HCV co-infected individuals (n = 10). In all groups, patients were enrolled at the beginning of IFNa therapy and were followed for a total of 4 months. While, for any group of patient's, naïve CD4+ and CD8+ T-cell counts were not significantly different from healthy individuals (figure 1A), as early as one month following treatment initiation, naïve CD4+ T-cell counts were significantly reduced in chronically HCV-infected patients (39%, 58%, 46% and 35% decrease at M1, M2, M3 and M4 respectively; p#0.025; Figure 1B , top central panel). A similar trend was also observed in the CD8 compartment (40%, 39%, 33% and 33% decrease; Figure 1B , bottom central panel). A comparable effect was also observed in most co-infected patients (mean cell count declines were 19%, 32%, 52% and 43% at M1, M2, M3 and M4 in the CD4+ T-cell compartment and 9%, 21%, 41% and 42% in CD8+ T-cell subset; p#0.05 by M2-M3; Figure 1B , right panels). In contrast, naïve T-cell counts were only barely affected in acutely-HCV infected patients under IFNa therapy ( Figure 1B , left panels). Similarly, central memory CD4+ T-cells (CD45 RA-CCR7+; TCM) demonstrated 38% and 28% decrease in HCV and HIV/HCV patients respectively (59% and 60% in CD8+ TCM) while effector memory (CD45RA2 CCR7-; TEM) CD4+ T-cell counts declined by 45% and 10% in the same groups (61% and 65% in CD8+ TEM) ( Figure S1 ). Within CD4+ naïve T-cells, RTEs can be identified by their higher expression of the platelet endothelial cell adhesion molecule-1 (PCAM-1 or CD31) [40] . While the number of RTEs was similar in HCV-infected patients at study entry and healthy individuals ( These data demonstrate that, as early as one month following treatment initiation, IFNa induces stronger alterations of naïve Tcell subsets, and more specifically in the RTE compartment than in any other T-cell subset, suggesting a specific effect on thymopoiesis. We thus analyzed the evolution of intrathymic precursor T-cell proliferation, peripheral T-cell cycling, IL-7 plasma concentration and IL-7 receptor alpha chain (CD127) expression, different factors affecting naive T-cell homeostasis. Despite differences between the 3 groups at study entry, RTE cycling rate, as estimated through measurement of Ki-67 expression, did not change significantly during the follow-up period ( Figure 3A ). These data demonstrate that the observed changes in sjTREC frequencies were not a consequence of variations of RTE proliferation during IFNa therapy but more probably due to reduced thymic production. We thus estimated thymic output through quantification of the sj/bTREC ratio in all groups of patients ( Figure 3B ). The sj/ bTREC ratio estimates the extent of thymocyte proliferation between TCRB rearrangement and the excision of the T-cell receptor delta (TCRD) locus [23] . This parameter directly reflects the extent of thymic production and, contrarily to sjTREC values, is independent from peripheral RTE proliferation or survival capacity [28] . The sj/bTREC ratio was already low in HIVinfected patients (p,0.005 as compared to healthy control donors; Figure 3B bottom left panel) and did not evolve further under IFNa therapy in co-infected patients ( Figure 3B , bottom right panel). In contrast, acutely HCV-infected patients demonstrated higher than normal sj/bTREC ratio at baseline (p,0.05 as compared to aged matched healthy controls), showed a significant reduction in sj/bTREC ratio at M1 (p = 0.014) and M2 (p = 0.001; Figure 3B , top panel). Finally, a similar decline in the sj/bTREC ratio was observed during IFNa therapy in chronically HCV-infected patients (p,0.02 at M1, M2 and M3; Figure 3B , central panel). Precursor T-cell proliferation in the thymus is, at least in part, dependent upon IL-7. We thus quantified plasma IL-7 concentration in all groups of patients. At study entry, HCV-and HIV/ HCV-infected patients presented with elevated plasma IL-7 (median = 10.3 pg/mL, range (6.7-12.9) in acutely HCV-infected patients; 8.3 pg/mL (6.3-10.5) in chronic HCV-infected patients and 7.15 pg/mL (4.3-13.5) in co-infected subjects), as compared to that observed in healthy control individuals (p,0.001 for any patients' group; Figure 4A) . Surprisingly, while lymphocytopenia established, IL-7 plasma concentrations significantly decreased in both groups of HCV-infected patients (30, 54, 18 and 29% decrease at M1 to M4 in acute infection, p,0.05; 25, 46, 26 and 16% decrease at M1 to M4 in chronic infection, p,0.05; Figure 4B left and central panels). In contrast, IL-7 plasma levels did not significantly evolve in co-infected individuals during the first month of IFNa therapy ( Figure 4B right panel) . Only patients with the highest IL-7 plasma levels showed a reduction in the concentration of this cytokine. Decreased plasma IL-7 concentrations could be a consequence of reduced IL-7 production, increased consumption by T-cells or sequestration by soluble IL-7 receptor (sCD127). In both HCVinfected and HIV/HCV co-infected patients, neither sCD127 Considering the variations in all the parameters we used to evaluate thymic function, we then sought to evaluate the impact of changes in IL-7 plasma levels on de novo production from the thymus and on the number of both sjTREC and circulating CD4+ RTEs. In a majority of patients, IL-7 plasma level, sj/bTREC ratio, sjTREC/ml and blood RTE concentration fluctuated in parallel ( Figure S2 ). Variation of IL-7 plasma concentration (DIL-7) during the first month of therapy correlated with variations in naïve T-cell counts (CD4+ + CD8+; DNaïve T-cell counts) and RTE CD4+ T-cell counts (DRTE T-cell counts) in both HCV (r = 0.521, p = 0.039 and r = 0.595, p = 0.025; Figure 5A and 5B, left panels) and, to a lesser extent, HIV/HCV co-infected patients (r = 0.636, p = 0.048 and r = 0.539, p = 0.108; Figure 5A and 5B, right panels). Moreover, in HCV-infected patients, DIL-7 also correlated with variations in intrathymic precursor T-cell proliferation (Dsj/bTREC ratio; r = 0.601, p = 0.020; Figure 5C ). Variations in plasma IL-7 levels also correlated with changes in the proportions (D%Ki-67+ in CD4+RTEs; r = 0.806, p = 0.0002; Figure 5D , left panel) and numbers (DKi-67+RTEs; r = 0.706, p = 0.002; Figure 5E , left panel) of cycling RTEs in acute and chronic HCV infected patients and with D%Ki-67+RTE counts in co-infected patients (r = 0.709, p = 0.022; Figure 5E , right panel). Overall, IL-7 concentration was associated with reduced thymopoiesis and RTE proliferation, lower consequently leading to limited circulating RTE and naïve T-cell counts. These data strongly suggest that changes in IL-7 plasma levels during IFNa therapy directly impact the homeostasis of RTEs. We herein demonstrated that IFNa-based therapy leads to major lymphocytopenia in naïve T-cell compartments, in particular in the RTE subset. Several mechanisms could be implicated in the establishment of such a lymphocytopenia [41] . Among these, enhanced apoptosis [42, 43] , cell sequestration in lymphoid or non-lymphoid organs [12, 21, 22] and regulation of peripheral T-cell homeostasis [20] . In our study, no major change in cell survival (Bcl-2 expression) or T-cell activation (CD25 and CD69 expression) was observed during the follow-up period (data not shown). Moreover, we did not observe any significant modification in Ki-67 expression in any T-cell subset during the first month of therapy (data not shown and Figure 3 ). Finally, IFNa-induced T-cell homing, although rapid and massive, is only a transient process [22] suggesting that this mechanism marginally contributes to the observed long lasting lymphocytopenia. Interestingly, both sjTREC quantification (sjTREC/mL) and intrathymic precursor T-cell proliferation (sj/bTREC ratio) were affected very early on after initiation of therapy ( Figures 2B and 3B ). While sjTREC frequency and concentration in peripheral blood can be affected by modifications of parameters that impact on peripheral T-cell homeostasis (cycling, survival/apoptosis, homing), the sj/bTREC ratio is a marker of the intrathymic proliferation history of RTEs. Indeed, this parameter is generated by cell proliferation that occurs between TCRb chain rearrangement and the excision of TCRd locus. Further cell cycling after TCRa chain rearrangement does not modify the sj/bTREC ratio as both type of TRECs are similarly diluted upon cell proliferation. Accordingly, while exported to the periphery, the sj/bTREC ratio of mature T-cells cannot be modified. Therefore, while the observed decrease in sjTREC concentration (figure 2) can be a consequence of modifications of circulating T-cell homeostasis, the decline of the sj/bTREC ratio observed during the first months of IFNa therapy (figure 3) defines changes in thymocyte proliferation, thus in thymic output [28] . Acutely infected patients demonstrated a higher sj/bTREC ratio at baseline than patients in the chronic phase. However, this group was younger (Median = 31.5 (26-47)) versus Median = 53.5 (37-61)) than the chronic group (p,0.01; data not shown) and demonstrated normal sj/bTREC ratio for their age. Similar evolution of thymic function and circulating T-cell subsets were observed in both groups of HCV-infected patients, irrespective of the development stage of HCV pathology. The lack of effect of IFNa therapy in HIV/HCV co-infected patients might be due to the fact that, as expected for chronically HIV-infected individuals, these patients already had a low thymic function at study entry. The impairment of thymopoiesis in HCV-infected patients under IFNa therapy is reminiscent of that observed during the acute phase of HIV-1 infection [23] which suggested that long term production of IFNa, as part of the anti-HIV innate immune response, may play a role in the observed thymic defect. The correlation between decline in IL-7 plasma levels under IFNa therapy and both thymic dysfunction and reduced T-cell counts, in particular in the naïve and RTE compartments ( Figures 5A and 5B) , confirms this hypothesis. Finally, in a recent study, we showed that IFNa treatment leads to decreased sjTREC frequency as well as reduced naïve T-cell and RTE counts in SIV-infected rhesus macaques [30] . Such an effect was accompanied by a 30-40% decrease in IL-7 plasma levels in these animals and could be counteracted by injection of recombinant simian IL-7 [30] . One could expect that such an effect of type I IFNs is not restricted to HIV-infection as many viral infections induce IFNa responses and cause transient lymphocytopenia in the infected hosts [3, 4, 5, 6] . Moreover, the IFNa-induced reduction of thymic function and its probable consequences on naïve T-cell diversity may contribute to the higher infectious risk associated with IFNa therapy, in particular observed in older patients [15, 16, 44] . There are multiple sources for circulating IL-7 during viral infections including lymphoid organs, epithelial cells and recently the liver was identified as a major source of IL-7. Moreover, increased plasma IL-7 levels can also be observed during viral infection in non-lymphopenic individuals ( [33] and unpublished data), suggesting a role in the development of immune responses. Indeed, this cytokine participates to T-cell homing in various lymphoid and non-lymphoid tissues through stimulation of local chemokine productions [45] . Increased IL-7 plasma levels in lymphopenic individuals is likely due to reduced consumption [46] yet augmented production to counteract lymphopenia cannot be excluded [33] . The recent identification of the liver as an IL-7 producing tissue upon TLR stimulation [47] makes it tempting to speculate that HCV-infection can also, through TLR activation, stimulate IL-7 production by the liver. Indeed, non-lymphopenic HCV-infected patients demonstrate similar IL-7 plasma levels than lymphopenic HIV-infected individuals [33, 48] suggesting that most of the IL-7 production in untreated HCV-infected patients was not linked to circulating T-cell counts. The reduction of IL-7 plasma levels while lymphocytopenia establishes under IFNa therapy, the absence of a correlation between IL-7 plasma levels and CD127 expression and the concomitance of decreases in IL-7 plasma levels and HCV viral load under therapy suggest that viremia might be driving IL-7 production before initiation of therapy. Our data suggest that, before initiation of IFNa therapy, actively replicating HCV leads to the overproduction of IL-7. Subsequent reduction of IL-7 production upon initiation of therapy probably reflects the elimination of IL-7 producing HCV-infected hepatocytes. This sudden reduction of IL-7 plasma levels may lead to diminished thymopoiesis. The fact that IL-7 plasma levels did not reach normal levels when HCV became undetectable may suggest that, after the initial decline that follows the drop in viremia, IL-7 plasma levels were regulated, as in HIVinfected patients [33] and in IFNa-treated SIV-infected rhesus macaques [30] , as a consequence of lymphocytopenia through either reduced consumption or increased production in lymphoid organs [49] . Future studies with a longer follow-up period, in particular after the end of IFNa therapy and recovery from lymphocytopenia are required to further elucidate this point. We herein demonstrated that a substantial reduction in thymic export was observed in HCV-infected patients, during the first months of IFNa therapy. This effect directly paralleled IFNainduced lymphocytopenia and decreased IL-7 plasma levels, initially high in HCV-infected patients. These data suggest that IL-7 production by the liver, a consequence of active HCV replication, was reduced while patients controlled HCV viremia. Restricted IL-7 plasma levels might, in association with the antiproliferative effect of IFNa, limit T-cell production in the thymus. Our study highlights the therapeutic potential of IL-7 as a complement to the standard IFNa based treatment to help HCVinfected patients to sustain normal circulating T-cell counts, and restore the diversity of the peripheral T-cell repertoire through its central thymopoietic effect. Restoring the breadth and intensity of T-cell control over the HCV virus might be immediately beneficial for the HIV/HCV co-infected population and offer new promising avenues for chronic HCV in the context of massive drop of HCV viral load after short term treatment with new antiviral compounds that will continue to be administered in combination with IFNa [50] . Sixteen HCV-infected patients (C-1 to C-16) and ten HIV/ HCV co-infected patients (I-1 to I-10) naïve to IFNa therapy were enrolled in this study. A summary of the virological and immunological status of patients at baseline is shown in table 1. All the HIV/HCV co-infected patients but one were under HAART with undetectable viremia (,40 HIV copies/mL). Chronically infected patients (C-9 to C-16 and I-1 to I-10) initiated pegylated IFNa/ribavirin treatment (IFNa-2a: Pegasys, 180 mg weekly, Ribavirin: Copegus, 800 mg to 1000 mg daily) and were followed over a 4 months period. Patients included in the acute phase of HCV infection (C-1 to C-8) were treated with pegylated IFNa (IFNa-2a: Pegasys, Roche, 180 mg weekly) [51, 52] . Blood samples were taken monthly on EDTA. Two milliliters of total blood were 2-fold diluted in FCS/20%DMSO frozen at 280uC and conserved in liquid nitrogen. These total blood samples were subsequently used for flow cytometry analyses. Plasma was separated from the remaining eight milliliters and mononuclear cells were purified on Ficoll Hypaque (Eurobio, Courtaboeuf, France) and frozen for further analyses. Patients from the HCV mono-infection group were followed at the Centre de Recherche du CHUM, Hôpital Saint Luc, Montreal, QC, Canada and its collaborators as previously described [9, 53] . Patients from the HIV-HCV groups were followed at the Hôpital Henri Mondor, Creteil, France. Clinical protocols conformed to Figure 5 . Variations in IL-7 plasma levels correlate with evolution of RTE production. Correlations. between variations in IL-7 plasma levels (DIL-7) and either variations in (A) total (CD4 + + CD8 + ) naïve T-cell counts (Dnaïve T-cell counts), (B) RTE defined as CD31 hi naïve CD4 + T-cells (DRTE CD4 counts), (C) the sj/bTREC ratio (Dsj/bTREC ratio), (D) the frequency of Ki-67 + cells in the RTE CD4 + T-cell subset (D%Ki-67 + in CD4 + RTEs) or (E) the number of circulating Ki-67 + CD4 + RTEs (DKi-67 + RTE counts) between study entry and month 1 of therapy were calculated for acutely (black symbols) and chronically (white symbols) HCV-infected patients (left panels) and HIV/HCV co-infected patients (right panels). Correlation coefficients (Spearman's r) and the associated probabilities (p) are shown. doi:10.1371/journal.pone.0034326.g005 ethical guidelines of the authors' institutions and the US Department of Health and Human Services' human experimentation guidelines. This study was approved by both the Ethical committee of Centre Hospitalier de l'Université de Montreal (CHUM) and the ethical committee of Hôpital Henri Mondor, Créteil, France. Samples were obtained with the written subjects' informed consent. Immunophenotyping and flow cytometry analysis FACS analyses were performed on cryopreserved samples. After thawing blood cells were incubated for 15 minutes at 4uC with conjugated monoclonal antibodies (mAbs). For intracellular labeling, cells were permeabilized with the Cytofix/Cytoperm Kit (Becton Dickinson) before incubation with specific mAbs according to the manufacturer's instructions. Samples were then washed, fixed in 2% paraformaldehyde phosphate-buffered saline (PBS/PFA 2%) and acquired using a Cyan cytofluorometer (Dako) and analyzed with FlowJo 8.7 software. The monoclonal antibodies used in this study were: CD3-pacific blue (PB) (clone UCHT-1; Dako, Trappes, France), CD4-peridin chlorophyll protein-cyanine 5.5 (PerCP-Cy5.5) (clone L200; BD, Le-Pont-de-Claix, France), CD45RA-phycoerythrin (PE) (clone HI100; BD), CCR7-allophycocyanin (APC) (clone 150503; R&D Systems Europe, Lille, France); CD8-phycoerythrin-cyanine 7 (PE-Cy7) (RPA-T8; BD), CD31-biotin (clone WM59; AbDSerotec, Düsseldorf, Germany); Ki-67-fluorescein isothiocyanate (FITC) (clone MIB-1; Dako), Bcl-2-FITC (clone 124; Dako) and strepatavidin-PE-Texas-RED (BD). IL-7 was quantified in the plasma using the IL-7 Quantikine HS kit according to the manufacturer's instructions (R&D Systems Europe). Plasma soluble-CD127 quantification Soluble plasma IL-7 receptor (sCD127) quantification was performed as previously described [54] . Parallel quantification of the sjTREC and the 13 DJbTRECs, together with CD3c gene (used as a housekeeping gene) was performed for each sample using LightCyclerTM technology (Roche Diagnostics) with a technique adapted from [29] . Intrathymic precursor T-cell proliferation was evaluated through calculation of the sj/bTREC ratio as described [23] . HCV RNA quantification was performed using an in-house quantitative real-time reverse transcription-PCR assay as previously described [9] , COBAS Amplicor HCV Monitor test TM , Version 2.0 (sensitivity 600 IU/ml)), qualitative COBAS Ampli-Prep/COBAS Amplicor HCV test TM , version 2.0 (sensitivity 50 IU/ml) or Abbott RealTime HCV assay TM (sensitivity 12 IU/ ml). Statistical analyses (Spearmans rank correlations and Wilcoxon matched -paired signed-rank tests) were performed using the Stata/IC 10.0 (Stata corporation, College Station, Tx U.S.A.). Due to the exploratory nature of the study there was no correction for multiple comparisons, and calculated p values are reported herein.

Baculovirus-based Vaccine Displaying Respiratory Syncytial Virus Glycoprotein Induces Protective Immunity against RSV Infection without Vaccine-Enhanced Disease

BACKGROUND: Respiratory syncytial virus (RSV) is a major cause of severe lower respiratory tract diseases in infancy and early childhood. Despite its importance as a pathogen, there is no licensed vaccine against RSV yet. The attachment glycoprotein (G) of RSV is a potentially important target for protective antiviral immune responses. Recombinant baculovirus has been recently emerged as a new vaccine vector, since it has intrinsic immunostimulatory properties and good bio-safety profile. METHODS: We have constructed a recombinant baculovirus-based RSV vaccine, Bac-RSV/G, displaying G glycoprotein, and evaluated immunogenicity and protective efficacy by intranasal immunization of BALB/c mice with Bac-RSV/G. RESULTS: Bac-RSV/G efficiently provides protective immunity against RSV challenge. Strong serum IgG and mucosal IgA responses were induced by intranasal immunization with Bac-RSV/G. In addition to humoral immunity, G-specific Th17- as well as Th1-type T-cell responses were detected in the lungs of Bac-RSV/G-immune mice upon RSV challenge. Neither lung eosinophilia nor vaccine-induced weight loss was observed upon Bac-RSV/G immunization and subsequent RSV infection. CONCLUSION: Our data demonstrate that intranasal administration of baculovirus-based Bac-RSV/G vaccine is efficient for the induction of protection against RSV and represents a promising prophylactic vaccination regimen.

Respiratory syncytial virus (RSV) is the most important viral pathogen of causing serious bronchiolitis and pneumonia in infants and young children worldwide. RSV is also receiving increasing recognition as an important cause of lower respiratory tract illness in immunocompromised patients, the young children, and the elderly (1) (2) (3) . Despite the importance of RSV as a respiratory pathogen, there is no licensed vaccine currently available against RSV infection. In the 1960s, the immunization of children with formalin inactivated-RSV (FI-RSV) vaccine not only failed to protect, but also yielded enhanced pulmonary disease in vaccinated infants following RSV infection (4, 5) . Studies with BALB/c mice have become a useful model for RSV pathogenesis, since FI-RSV-enhanced disease is also observed in vaccinated BALB/c mice. It is likely that the augmented lung disease and the development of pulmonary eosinophilia are attributed to an excessive Th2 type immune response (6) . The RSV G protein is one of the major protective antigens and good inducer of strong serum and mucosal neutralizing antibody responses. It also has single immunodominant I-E d epitope spanning RSV G amino acid 183 to 198 and largely induces a specific subset of CD4 T cells (7, 8) . Previously, we have reported that RSV G protein fragment (spanning amino acid residues 131-230) delivered by recombinant adenoviral vector successfully elicited long-term protective immunity against RSV infection in mice (9) . Baculoviruses are enveloped viruses possessing a rod-shaped nucleocapsids in which double stranded circular DNA genome of 88-135Kbp is packaged (10) . Baculoviruses are generally utilized as a vector for the high level production of recombinant proteins but they could be also employed in gene transfer to mammalian cells (11) (12) (13) . Several research groups have demonstrated that vaccination with recombinant baculovirus can induce high-level humoral and cell-mediated immunity against various antigens, suggesting that baculovirus could be used as a vaccine carrier (14, 15) . In addition, it has been reported that immunization with a recombinant baculovirus expressing the hemagglutinin gene of influenza virus elicited a strong innate immune response and protected mice against influenza virus challenge (16) . Moreover, it is known that there is no baculovirus-specific immunity in mammals that might interfere the action of baculovirus-based vaccine (17) (18) (19) . In the present study, a recombinant baculovirus displaying the RSV G protein (Bac-RSV/G) was constructed. Our results clearly show that strong humoral and cellular immune responses were induced by intranasal administration of Bac-RSV/G in a mouse model. Importantly, complete protection against RSV challenge without vaccine-induced illness was observed in Bac-RSV/G-immune mice, suggesting that vaccination of Bac-RSV/G could be utilized as a prophylactic vaccination regimen against RSV infection. Baculoviruses were propagated in Spodoptera frugiperda 9 (Sf9) insect cells using SF-900 serum-free medium (Invitrogen) at 27 o C. RSV A2 strain was propagated in HEp-2 cells (ATCC, Manassas, VA) in Dulbecco's modified Eagle's medium (Life Technologies, Gaithersburg, MD) supplemented with 3% heat-inactivated fetal calf serum, 2 mM glutamine, 20 mM HEPES, nonessential amino acid, penicillin, and streptomycin and titrated for infectivity by plaque assay as described elsewhere (20) . The coding sequence of RSV G protein from RSV A2 strain was amplified from cDNA by PCR and cloned into the EcoR I and Xho I sites of pFastBac-1 vector (Fig. 1A) . The recombi-nant baculovirus was subsequently generated by using the Bac-to-Bac Ⓡ system (Invitrogen) following the manufacturer's instructions. The recombinant baculoviruses were purified from supernatants of infected Sf9 insect cells with 25% (w/v) sucrose in 5 mM NaCl, 10 mM EDTA in a SW28 rotor (Beckman, USA) at 24,000 rpm for 75 min at 4 o C. The supernatant was decanted, and the pellet was resuspended in phosphate-buffered saline (PBS) and centrifuged for 4 h at 24,000 rpm, 4 o C. The viral pellet was resuspended in PBS and titrated by plaque assays on Sf9 cells. Female BALB/c mice were purchased from Charles River Laboratories Inc. (Yokohama, Japan). Mice kept under specific-pathogen-free conditions. For immunization, 6-to 8-weekold mice were inoculated with baculoviruses via the intranasal (i.n.) route. For i.n. immunizations, mice were lightly anesthetized by ether/chloroform inhalation, and 2×10 8 PFU of Bac-RSV/G or Bac-control in a volume of 70μl was applied to the left nostril. Three to four weeks after second immunization, the mice were challenged i.n. with 1×10 6 PFU of live RSV A2 strain. All animal studies were performed according to the guidelines of our Institutional Animal Care and Use Committee (Approval No. 2010-9-4). Blood was obtained from the retro-orbital plexus with a heparinized capillary tube, collected in an Eppendorf tube, and centrifuged, and serum was stored at −20 o C. RSV G protein-specific antibody titers in immunized mice were measured by a direct enzyme-linked immunosorbent assay (ELISA). Briefly, 96-well plates were coated overnight with 100μl/well of 0.5μg/ml of purified G protein fragment diluted in PBS and then blocked with PBS containing 1% skim milk and 0.05% Tween-20 for 2 h. Sera were then added in serial dilutions and incubated for 2 h. The plates were washed five times with PBS containing 0.05% Tween 20 and incubated for 30 min with various dilutions of horseradish peroxidase-conjugated affinity-purified rabbit anti-mouse total IgG, secondary antibody (Zymed Laboratories, San Francisco, CA). The plates were washed five times and developed with 3,3',5,5'-tetramethylbenzidine, and the reaction was stopped with 1 M H 3 PO 4 and analyzed at 450 nm with a Thermo ELISA plate reader. The wells receiving no serum were used to calculate cut-off values. The lungs were perfused with 5 ml of PBS containing 10 U/ml heparin (Sigma-Aldrich, St. Louis, MO) through the right ventricle using a syringe fitted with 25-gauge needle. The lungs were then removed and placed in RPMI medium supplemented with glutamine, gentamicin, penicillin G, and 10% FBS (HyClone, Logan, UT). The tissue was then processed through a steel screen to obtain a single-cell suspension, and particulate matter was removed by passage through a 70-μm Falcon cell strainer (BD Labware, Franklin Lakes, NJ). Freshly explanted BAL fluid or lung cells were purified by density gradient centrifugation and stained in PBS-3% FBS-0.09% NaN 3 using fluorochrome-conjugated antibodies. The antibodies used were anti-CD4 (clone RM4-5), anti-CD44 (clone IM7) or anti-CD43 (clone 1B11). Both antibodies were purchased from BD PharMingen (San Diego, CA). After staining, cells were fixed in PBS-2% (wt/vol) paraformaldehyde, and events were acquired using a FACSCalibur flow cytometer (BD Biosciences, San Diego, CA). To enumerate the number of cytokine-producing cells, intracellular cytokine staining was performed as described elsewhere (21) . In brief, 2×10 6 freshly explanted lung lymphocytes were cultured in a culture tube. Cells were left untreated or stimulated with 10μM G (183-195) peptide (WAICKRIPNKKPG) and then incubated for 5 h at 37 o C in 5% CO2. Brefeldin A (5μg/ml) (Sigma-Aldrich) was added for the duration of the culture period to facilitate intracellular cytokine accumulation. Cells were then stained for surface markers, washed, fixed, permeabilized with fluorescence-activated cell sorter buffer containing 0.5% saponin (Sigma-Aldrich, Seoul, Korea), and stained for cytokines. The antibodies used were anti-IFN-γ (clone XMG1.2) or anti-IL-17A (TC11-18H10.1). Dead cells were excluded on the basis of forward and side light scatter patterns. Data were collected using CELLQuest software (BD Biosciences) and analyzed with CELLQuest and WinMDI version 2.9 software (Scripps Research Institute, La Jolla, CA). Lung supernatants were also collected for analysis with the FlowCytomix (eBioscience), according to the protocol. Kits containing antibody beads (IL-4, IL-5, IL-6, IL-10, IL-13) were used to measure cytokine levels in each of the samples. Four or five days after RSV challenge, subsets of mice were euthanized and the lungs were removed into Eagle's modified essential medium. The tissues were then processed through a steel screen to obtain a single-cell suspension, and particulate matter was removed by passage through a 70-μm cell strainer (BD Labware, Franklin Lakes, NJ). The supernatants were collected, and RSV titers in the supernatants were measured by standard plaque assay on subconfluent HEp-2 monolayers. The data are expressed as the PFU per gram of lung tissue. Comparison of differences was conducted by using an un-IMMUNE NETWORK http://www.ksimm.or.kr Volume 12 Number 1 February 2012 paired, two-tailed Student t test. The difference was considered statistically significant when the p value was ≤0.05. Generation of recombinant baculovirus displaying RSV G The recombinant baculovirus expressing RSV G coding sequences under the control of the polyhedrin promoter, Bac-RSV/G, was generated using Bac-to-Bac system with a pFastBac-1 plasmid as shown in Fig. 1A . To examine whether Bac-RSV/G express and display RSV G in the virus particle, western blotting analysis was performed with sucrose gradient-purified baculoviruses. Using the G-specific monoclonal antibody, specific bands of approximately 70 kDa to 90 kDa were detected in the purified Bac-RSV/G particles, but not in Bac-control virus particles (Fig. 1B) . As a positive control, purified RSV particles produced in HEp-2 cells were used and slightly higher molecular weight bands were detected (Fig. 1B) . The size difference might be due to different patterns of glycosylation between insect cells and human cells. To investigate the immunogenicity of Bac-RSV/G, the groups of mice were immunized i.n. with Bac-RSV/G, Bac-control, or PBS. RSV G-specific antibody levels in sera from the immune mice at 2 weeks after priming and 2 weeks after boost-ing were determined by ELISA. The specific antibody responses were barely detectable in all groups of mice at 14 days after priming (data not shown). Following booster immunization, however, the mean titers of serum antibodies increased significantly only in the group of mice immunized with Bac-RSV/G ( Fig. 2A) . Secretory IgA has been shown to directly mediate local immunity against aerial pathogens, implying an important role for antibody in protection in the upper respiratory tract (22) . Thus, for an effective RSV vaccine development, the induction of secretory IgA on respiratory mucosal surface is critical. To examine whether Bac-RSV/G vaccination elicits IgA response in the respiratory tract, bronchoalveolar lavage (BAL) was performed at day 5 after RSV challenge and levels of IgA were determined by RSV-specific ELISA. Each group of mice immunized with Bac-control or PBS did not exhibit any specific IgA response, whereas Bac-RSV/G-immune mice showed significantly enhanced level of specific IgA response (Fig. 2B) . These results suggest that intranasal immunization of Bac-RSV/G effectively induce both RSV G-specific serum IgG and respiratory IgA. Since the RSV G contains I-E d -restricted CD4 T-cell epitope, we examined whether Bac-RSV/G immunization induces specific CD4 T cells or not. To this end, Bac-RSV/G, Bac-control, or PBS immune mice were challenged with RSV. CD4 T cell responses were evaluated at 5 days after challenge by measuring the production of IFN-γ in lung lymphocytes that had been stimulated with I-E d -restricted G (183-195) epitope peptide ex vivo. As shown in Fig. 3 , G-specific CD4＋IFN-γ＋ cells were detected in the lungs of Bac-RSV/G-immunized group (∼2.5% of gated CD4 T cells on average), while few specific cells (＜0.2% of CD4 T cells) were observed in the lungs of Bac-control or PBS-immunized group. We next investigated whether immunization of Bac-RSV/G induces other CD4＋ T cell subset such as Th2 and Th17. Bac-RSV/G-vaccinated mice showed increased frequency of G-specific IL-17-producing CD4 T cells in the lungs after RSV challenge (∼3.7% of gated CD4 T cells on average; Fig. 4 ). However, the levels of Th2-type cytokines such as IL-4, IL-5, IL-10, and IL-13 measured by multiplex antibody-based bead assays were not significantly different in the lungs from all groups of mice (Fig. 5) . Together, these results indicated that mixed G-specific Th1/Th17-cell responses were induced by Bac-RSV/G vaccination. To determine whether Bac-RSV/G has protective efficacy against RSV infection, mice were challenged with live RSV A2 virus at four weeks after booster immunization. While there was active RSV replication in the lungs of the Bac-control or PBS immune mice, immunization of Bac-RSV/G prevented any detectable RSV replication in the lungs during the peak of infection (Fig. 6) . Interestingly, immunization of Bac-control resulted in partial decrease of viral replication at the peak. This might be due to nonspecific innate immunity induced by baculovirus inoculation. It was previously reported that intranasal administration of wild-type baculovirus induces strong innate immune responses and provides protection against lethal challenge of influenza virus (16) . Fig. 3 . BAL was performed five days after RSV challenge, and BAL cells were stained with antibodies to CD45, Siglec-F, and CD11c, and eosinophils were quantitated among CD45 It was reported that immunization with vaccinia virus expressing the entire RSV G glycoprotein results in pulmonary eosinophilia following challenge with live RSV (6, 23, 24) . These studies indicated that RSV G expressed vaccines have a possibility for the development of pulmonary eosinophilia. To determine whether the intranasal immunization of Bac-RSV/G potentiates eosinophilia, the levels of eosinophils in the BAL fluids of the immune mice were examined by flow cytometry 5 days after RSV challenge using antibodies to Siglec-F, CD45, and CD11c as described previously (25) . Bac-RSV/G-immunized mice have a higher frequency of CD11c-SiglecF＋ in the BAL as compared with Bac-control or PBS group (Fig. 7A , p＜0.05), but the level of eosinophil influx in Bac-RSV/G immune mice was relatively weak (∼2% of the total CD45＋ BAL cells). These results suggest that intranasal Bac-RSV/G immunization barely increase the risk of vaccine-induced eosinophilia. To evaluate other RSV-induced pathology by Bac-RSV/G immunization, we monitored weight loss in all immune-mice after RSV challenge. Following infection with live RSV, there was no significant weight loss in Bac-RSV/G-immune mice (Fig. 7B ) and disease score (data not shown). Taken together, these results suggest that intranasal Bac-RSV/G vaccination give rise to protective immunity in the absence of subsequent vaccine-enhanced disease. To develop a safe and effective RSV vaccine, many strategies and platforms have been applied in pre-clinical and clinical phases (26) . Many RSV vaccine candidates specifying target antigens employed two envelop proteins, G attachment protein and F fusion protein, because these antigens are known to induce protective immunity against live RSV infection. However, in the BALB/c mouse model, immunization of G protein expressed from recombinant vaccinia virus elicited Th2-biased responses, which was responsible for the vaccine-enhanced diseases (27) . Thus, G protein has been falsely regarded as a bad target antigen for a long time, although it could induce strong neutralizing antibody responses upon immunization. Recently, it has been suggested that G protein itself is not the cause of vaccine-enhanced diseases and the proper balance between RSV-specific Th1 and Th2 responses is rather an important factor controlling both safety and efficacy of G-targeted RSV vaccine. Using appropriate platforms and/or adjuvants, this balance could be achieved with G-targeted vaccines. For example, we have recently demonstrated that mucosal immunization of recombinant adenovirus vac-cine expressing the core domain of G successfully induced protective immunity without vaccine-induced diseases (9) . Baculovirus has recently been emerged as a new tool for vaccine vector development since it has many advantages as a vaccine platform (28) (29) (30) . First, baculovirus, which has been used to overexpress recombinant proteins in insect cells, is neither replication-competent nor pathogenic in mammalian cells, making it safer vaccine vector than other mammalian viral vectors. In addition, baculovirus-based vaccines produced in insect cells are not contaminated by LPS. Second, baculovirus-based vaccine possesses several advantages upon production, such as easy manipulation, relatively simple scale-up, and high titers (11, 31) . Thirdly, there is no pre-existing vector immunity against baculovirus in mammals, which enables baculovirus to escape vector neutralization by pre-existing immunity during in vivo delivery (19) . Lastly, baculovirus has the ability to stimulate strong innate immunity, exhibiting strong adjuvanticity itself. It has been previously shown that inoculation of wild-type baculovirus alone can stimulate the secretion of inflammatory cytokines from innate immune cells and confer protection from lethal virus infection in mice (16) . Consistent with this report, intranasal inoculation of control baculovirus also provided partial protection against live RSV challenge in our study (Fig. 6) . The baculovirus-mediated stimulation of innate immunity might be associated with the induction of type I interferons, mediated by both TLR9-dependent (32) and TLR9-independent pathways (33) . Thus, it becomes evident that baculovirus can act as a natural adjuvant by stimulating innate immunity and be used as a safe and effective vaccine carrier. Our data demonstrate that intranasal immunization of Bac-RSV/G vaccine induces complete protection from RSV infection, which is associated with enhanced Th1 and Th17 responses in the absence of significant Th2 responses. Although the role of Th1 or CTL responses to RSV clearance is well delineated (34) , the role of Th17-mediated responses have not been fully elucidated in host defense against RSV. Th17 cells, which are characterized by massive production of IL-17, are thought to contribute to autoimmune diseases but also play a crucial role in host defense against extracellular bacteria (35, 36) . In addition, Th17-associated effector molecules seem to be necessary and sufficient for protection against several viral pathogens. For example, recent studies have reported that Tc17 cells are potently cytolytic against vaccinia virus-infected cells (37, 38) . In these studies, both IL-17-producing CD4＋ and CD8＋ T cells are necessary to confer pro-tection against viral infections, exhibiting cytotoxic activity for clearance of vaccinia virus-infected cells. Mice immunized with a recombinant vaccinia virus expressing the G protein of RSV (vvG) exhibit pulmonary eosinophilia induced by Th2 type cytokines after challenge RSV infection (6) . Th2 type cytokines are thought to promote eosinophil infiltration to the lungs of RSV-infected mice and may contribute to immunopathology associated with vaccine-enhanced diseases. Interestingly, we showed that the levels of Th2-type cytokines, such as IL-4, IL-5, IL-10, or IL-13, were very low in the lungs of Bac-RSV/G-immune mice following RSV challenge, not significantly different to those in the PBS control group. Several evidences suggest that IL-17 negatively regulates Th2 cytokine production (39, 40) . Thus, the induction of Th1/Th17 responses by Bac-RSV/G vaccination may have attenuated the development of Th2 responses. Furthermore, IL-17 may play certain roles in immune defense in the lungs, including augmentation of mucosal immunity by delivery of IgA and IgM into the airway lumen (41) . Though the exact mechanism by which the Th17 responses are induced by Bac-RSV/G vaccination is not clear, the G-specific Th17 response may be associated with clearance of RSV from the lungs. Further studies will be needed to elucidate the exact role of Th17 immunity in RSV infection. In conclusion, our present study demonstrates that baculovirus displaying RSV G protein, Bac-RSV/G, induces antigen-specific humoral and cellular immunity, and provides protection against RSV challenge without vaccine-induced immunopathology. Thus, our study provides strong evidence that Bac-RSV/G vaccine could be further developed as a mucosal RSV vaccine.

Divergent lineage of a novel hantavirus in the banana pipistrelle (Neoromicia nanus) in Côte d'Ivoire

Recently identified hantaviruses harbored by shrews and moles (order Soricomorpha) suggest that other mammals having shared ancestry may serve as reservoirs. To investigate this possibility, archival tissues from 213 insectivorous bats (order Chiroptera) were analyzed for hantavirus RNA by RT-PCR. Following numerous failed attempts, hantavirus RNA was detected in ethanol-fixed liver tissue from two banana pipistrelles (Neoromicia nanus), captured near Mouyassué village in Côte d'Ivoire, West Africa, in June 2011. Phylogenetic analysis of partial L-segment sequences using maximum-likelihood and Bayesian methods revealed that the newfound hantavirus, designated Mouyassué virus (MOUV), was highly divergent and basal to all other rodent- and soricomorph-borne hantaviruses, except for Nova virus in the European common mole (Talpa europaea). Full genome sequencing of MOUV and further surveys of other bat species for hantaviruses, now underway, will provide critical insights into the evolution and diversification of hantaviruses.

Discovery of phylogenetically divergent hantaviruses in shrews and moles (order Soricomorpha, family Soricidae and Talpidae) [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] raises the possibility that rodents (order Rodentia, family Muridae and Cricetidae) may not be the principal or primordial reservoirs. Moreover, newfound hantaviruses harbored by soricomorphs of multiple species, distributed in widely separated geographic regions across four continents, suggest that their host diversity may be far more expansive than previously assumed. Specifically, other mammals having shared ancestry or ecosystems with soricomorphs may serve as reservoirs and may be important in the evolutionary history and diversification of hantaviruses. In particular, bats (order Chiroptera) may be potential reservoirs by virtue of their rich diversity and vast geographical range, as well as their demonstrated ability to host myriad medically important, disease-causing viruses [14] [15] [16] [17] [18] . Surprisingly little attention, however, has been paid to this possibility. As in our previous investigations on the spatial and temporal distribution of hantaviruses in soricomorphs [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] , we relied on the availability of archival tissues. Using the PureLink Micro-to-Midi total RNA purification kit (Invitrogen, San Diego, CA), total RNA was extracted from 168 frozen and 45 ethanol-fixed liver and other visceral tissues of 213 insectivorous bats (representing 13 genera), collected during May 1981 to June 2011 in Asia, Africa and the Americas (Table 1) . cDNA was then prepared with the SuperScript III First-Strand Synthesis System (Invitrogen) using random hexamers, and PCR was performed as described previously, using an extensive panel of oligonucleotide primers, designed on conserved genomic sequences of rodentand soricomorph-borne hantaviruses [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] 19, 20] . Each reaction mixture contained 250 μ dNTP, 2 mM MgCl 2 , 1 U AmpliTaq polymerase (Roche, Basel, Switzerland) and 0.25 μ oligonucleotide primers. Initial denaturation at 94°C for 5 min was followed by two cycles each of denaturation at 94°C for 40 s, two-degree step-down annealing from 48°C to 38°C for 40 s, and elongation at 72°C for 1 min or 1 min 20 s, then 32 cycles of denaturation at 94°C for 40 s, annealing at 42°C for 40 s, and elongation at 72°C for 1 min, in a GeneAmp PCR 9700 thermal cycler (Perkin-Elmer, Waltham, MA). Amplicons were purified using the QIAQuick Gel Extraction Kit (Qiagen, Hilden, Germany), and DNA sequencing was performed using an ABI Prism 377XL Genetic Analyzer (Applied Biosystems, Foster City, CA). After innumerable failed attempts, hantavirus RNA was detected by RT-PCR in ethanol-fixed liver tissues from two of 12 banana pipistrelles (Neoromicia nanus Peters 1852), captured during June 2011 near Mouyassué village (5°22'07"N, 3°05'37"W) in Aboisso District, 130 km from Abidjan, in the extreme southeastern region of Côte d'Ivoire in West Africa ( Figure 1 ). The taxonomic identity of the hantavirus-infected vesper bats was confirmed by phylogenetic analysis of the cytochrome b gene of mtDNA (GenBank JQ287717), amplified by PCR as previously described [8, 9] . Despite similarly exhaustive efforts, hantavirus RNA was not detected in any of the other bat species tested (Table 1) , including frozen liver tissue of six tiny pipistrelles (Pipistrellus nanulus), collected in Parc National du Mont Péko, 700 km northwest of Mouyassué, in February 1992, and ethanol-fixed liver tissue of three tiny pipistrelles, collected in December 2009 in Azagny, where a hantavirus was previously found in the West African pygmy shrew (Crocidura obscurior) [8] . A 423-nucleotide region of the RNA-dependent RNA polymerase-encoding L segment, amplified using a hemi-nested primer set (outer: 5'-GAAAGGG-CATTNMGATGGGCNTCA GG-3', 5'-AACCADT-CWGTYCCRTCATC-3'; inner: 5'-GNAAAYTNATGT-ATGTNAGT GC-3', 5'-AACCADTCWGTYCCRT-CATC-3'), was aligned and compared with hantavirus sequences available in GenBank, using ClustalW (DNASTAR, Inc., Madison, WI) [21] and transAlign [22] . The newfound hantavirus, designated Mouyassué virus (MOUV), exhibited low nucleotide and amino acid sequence similarity of less than 69% to all representative soricomorph-and rodent-associated hantaviruses, except for the 76.3% sequence similarity with Nova virus (NVAV), previously reported in the European common mole (Talpa europaea) [12] . Interestingly, MOUV sequences were identical in the two banana pipistrelles (KB576 and KB577), a male-female pair captured simultaneously and presumed to be a mating couple, suggesting horizontal virus transmission or common-source infection. MOUV formed a uniquely divergent lineage, distant from all other hantaviruses identified to date, except for NVAV (Figure 2 ), in phylogenetic trees based on L-segment sequences, generated by the maximum-likelihood and Bayesian methods, implemented in PAUP* (Phylogenetic Analysis Using Parsimony, 4.0b10) [23] , RAxML Blackbox webserver [24] and MrBayes 3.1 [25] , under the best-fit GTR+I+Γ model of evolution established using jModeltest 0.1.1 [26] . Topologies were well supported by bootstrap analysis of 100 iterations, and posterior node probabilities based on two runs each of 2 million generations sampled every 100 generations with burn-in of 25%. Despite the overall success of our brute-force RT-PCR approach at identifying previously unrecognized hantaviruses in frozen tissues [2, 3, [5] [6] [7] [10] [11] [12] [13] and tissues preserved in RNAlater ® RNA Stabilization Reagent [4, 8] , designing universal primers for the amplification of soricomorph-borne hantaviruses has presented continuing challenges. Thus, while it is likely that many more hantaviruses await discovery, overcoming technical barriers is essential to facilitating their detection. Viewed in this context, the failure to detect hantavirus RNA in all but one bat species was not altogether unexpected and may be attributed simply to suboptimal primer design and imperfect cycling conditions. Also, low RNA yields and poor RNA preservation in tissues fixed in ethanol under field conditions may have thwarted our efforts at obtaining more of the MOUV genome. That said, the successful amplification of hantavirus RNA from ethanol-fixed tissues is highly instructive and augments the pool of archival tissues for future exploratory studies of hantaviruses in bats, and possibly other insectivorous small mammals that share ancestral lineages with soricomorphs, such as hedgehogs (order Erinaceomorpha, family Erinaceidae). Dating to the seminal discovery of Hantaan virus in lung tissue of the striped field mouse (Apodemus agrarius) [27] , lung has been the preferred tissue in studies aimed at finding new hantaviruses [28] [29] [30] . However, lung is not the only tissue in which hantaviruses can be detected [27, 31] . In our search of genetically distinct hantaviruses in long-stored archival tissues from shrews and moles, lung tissue was frequently unavailable. Instead, liver tissue was more often accessible and proved to be quite suitable [4, 5, 12, 13] . Similarly, liver tissues were more often available in the present study. As in reservoir rodents and soricomorphs, hantavirus RNA is likely to be present in many tissues of persistently infected bats. Real-time quantitative RT-PCR analysis of lung, liver and other viscera will clarify the tissue distribution of MOUV in newly captured banana pipistrelles from Mouyassué. Having their fossil origins in the Eocene epoch, approximately 50 million years before present, bats occur on every continent except Antarctica and are among the most speciose orders of mammals, with more than 1,100 extant species [32] . The banana pipistrelle, which is distributed widely in forests and savannas across sub-Saharan Africa ( Figure 1C, inset) , is one of 13 species in the genus Neoromicia of the family Vespertilionidae and subfamily Vespertilioninae. Like other vesper bats, the banana pipistrelle is insectivorous. Unlike large fruit bats, such as the straw-colored fruit bat (Eidolon helvum) and hammer-headed bat (Hypsignathus monstrosus), which are sold as bush meat, the banana pipistrelle, weighing approximately 3 g, is not consumed as food. However, because banana pipistrelles occasionally roost within houses or reside near human habitation, rare human encounters raise the possibility of hantavirus exposure. Previously, serological evidence of hantavirus infection was reported in the common serotine (Eptesicus serotinus) and greater horseshoe bat (Rhinolophus ferrumequinum) captured in Korea [33] , but genetic analysis of hantaviral isolates from these insectivorous bat species proved to be indistinguishable from prototype Hantaan virus [34] , suggesting laboratory contamination. In the present study, the strikingly divergent lineage of MOUV precluded any possibility of contamination and lends support to our earlier conjecture that the ancient origins of hantaviruses may have involved insect-borne viruses [7, 10] , with subsequent adaptation to and host switching between early soricomorph and chiropteran ancestral hosts in the mammalian superorder Laurasiatheria. However, since the biological and evolutionary implications of bats as reservoirs of hantaviruses are considerable, studies are underway to establish that the banana pipistrelle is the natural host of MOUV. Moreover, high-throughput sequencing technology is being applied to obtain the full genome of MOUV and to ascertain the geographic range and genetic diversity of hantaviruses harbored by bats.

Protective Immunity to Listeria Monocytogenes Infection Mediated by Recombinant Listeria innocua Harboring the VGC Locus

In this study we propose a novel bacterial vaccine strategy where non-pathogenic bacteria are complemented with traits desirable for the induction of protective immunity. To illustrate the proof of principle of this novel vaccination strategy, we use the model organism of intracellular immunity Listeria. We introduced a, low copy number BAC-plasmid harbouring the virulence gene cluster (vgc) of L. monocytogenes (Lm) into the non-pathogenic L. innocua (L.inn) strain and examined for its ability to induce protective cellular immunity. The resulting strain (L.inn::vgc) was attenuated for virulence in vivo and showed a strongly reduced host detrimental inflammatory response compared to Lm. Like Lm, L.inn::vgc induced the production of Type I Interferon's and protection was mediated by Listeria-specific CD8(+) T cells. Rational vaccine design whereby avirulent strains are equipped with the capabilities to induce protection but lack detrimental inflammatory effects offer great promise towards future studies using non-pathogenic bacteria as vectors for vaccination.

Current state of the art vaccine technology focuses on three distinct strategies 1) the creation of live attenuated pathogens based on the deletion of virulence factors [1] 2) the use of subunit vaccines [2] which contain one or more semi-pure antigens that are critical in inducing an immune response and 3) the use of metabolically active but non-viable bacteria [3] . For the first strategy, it must be considered that in today's medicine vaccines will often be administered to immunocompromised individuals and that the use of live vaccines in such subpopulations poses serious difficulties [4, 5] . The greatest disadvantage of subunit vaccines is their general requirement for strong adjuvants, as these adjuvants often induce detrimental tissue reactions. Lastly, the concept of so-called killed but metabolically active (KBMA) bacteria involves bacteria which are unable to form colonies on growth media but still have an intact protein synthesis and secretion machinery. Such mutants are reportedly capable of inducing CD4 + and CD8 + T cell responses and protection [3] . However this requires multiple injections. Lm is a facultative intracellular microorganism and many of the bacterial determinants necessary for pathogenesis, including intracellular growth and spread of Lm, have been identified and are clustered on a 10-kb region of the chromosome termed the virulence gene cluster (vgc) which encodes the prfA, plcA, hly, mpl, actA and plcB genes organized in three transcriptional units [6] . Being a facultative intracellular bacterium makes Lm particularly attractive as a potential live vaccine vector for the induction of cell-mediated immunity to foreign antigens [7, 8] . However, despite its capability to induce effective CD8 + T-cell responses the safety of recombinant Lm remains an important issue, as infections with Lm can cause severe and life-threatening infections [9] . Moreover, infection with Lm is mainly accompanied by undesired CD4 + T-cell mediated delayed type hypersensitivity (DTH) responses and granulomatous inflammation [10, 11] . Therefore the use of Lm in a clinical setting is associated with major risks limiting its potential as an effective vaccine vector. An alternative strategy would entail the transfer of a core set of virulence genes from pathogenic Lm to create a strain that is attenuated for virulence but is capable of inducing an effective immune response. To explore this approach we have transferred the vgc locus of Lm into a non-pathogenic species of Listeria such as L. innocua (L.inn) as a carrier strain. Here we show that a single immunization with this recombinant strain (L.inn::vgc) fulfills the desired requirements for a successful bacterial vaccine vector. These include low virulence in association with induction of protective antigen-specific CD8 + T-cell responses and reduction of CD4 + T cell-mediated inflammation. In vivo survival of the recombinant L.inn::vgc strain The ability of Listeria to survive in vivo at the early stage of infection is crucial for the induction of cell-mediated immunity [12, 13] . We examined the ability of L.inn::vgc to survive in the spleen and liver in infected mice and compared it to that of the wild type Lm. BALB/c mice were infected intravenously (i.v.) with sub-lethal doses of wild type Lm EGD-e (10 3 ), L.inn::vgc (10 7 ), or the wild type L.inn strain (10 7 ). Time points correlating with the critical phases of host immune response to listerial infection were selected and used to compare bacterial growth and induction of immune effectors in wild-type Lm, L.inn and L.inn::vgc strains. Day 3 of a Listeria infection refers to the end of the pre-immune phase before the expansion of specific T cells in the mouse model of listeriosis [14] . The presence of viable bacteria on this day has been shown to be critical for the successful induction of T cell-mediated immunity [15] . Therefore on day 3, bacterial load as well as spleen morphology was analyzed. Day 9 corresponds to the primary immune effector phase. At this time point, DTH to soluble antigen was measured in vivo as an indicator of DTH reaction and CD4+ T cell activity. Moreover, the numbers of antigen specific IFN-c producing CD8+ cytotoxic T cells were analyzed. Day 60 postinfection as well as day 5 post-challenge were chosen to analyze the memory immune effector phase [16] . To this end the number of memory effector T-cells was determined quantitatively. In vivo survival and growth kinetics of bacteria were followed by determining the number of bacteria in spleens and livers of infected mice. As expected, regardless of the dose of infection, the wild type L.inn strain was progressively cleared from both organs ( Fig. 1A) whereas the L.inn::vgc strain successfully survived in both spleen and liver during the first two days after infection as indicated by the bacterial numbers that increased in both spleen and liver till day 2 and gradually decreased over days 3 and 4 postinfection. On the other hand, the bacterial numbers of the wild type Lm, increased from day 1 till day 4 post-infection in both spleen and liver. Stimulation of Type I interferon's by the L.inn::vgc strain A striking phenomenon for cytosolic resident microbes is the ability to induce expression of Type I interferons. In contrast to the wild type Lm, its isogenic mutant lacking listeriolysin remains trapped in vacuoles and does not induce Type I interferon's [17] . We have recently documented that the L.inn::vgc can successfully survive inside phagocytic cells, thereby egressing from the phagolysosome [18] . In order to confirm if cytosolic persistence of the recombinant L.inn::vgc strain is efficient enough to stimulate production of such cytokines, we examined the transcriptional responses of IFN-a2 and IFN-b1 in bone marrow-derived macrophages following infection with Lm, L.inn as well as the recombinant L.inn::vgc strain. L.inn::vgc and the wild type Lm showed significantly higher transcriptional induction of both IFN-a2 and IFN-b1 than wild type L.inn at 2 hours post-infection (Fig. 1B) . This effect was more pronounced at a later time point (8 hours) post-infection reflecting the efficient intracellular survival pattern of the L.inn::vgc strain. The recombinant L.inn::vgc strain exhibits a lowered inflammatory response At the early stages of infection, wild type Lm is engulfed by professional phagocytes like macrophages, dendritic cells, or neutrophils. These cells produce a variety of proinflammatory cytokines which recruite or activate other inflammatory immune cells. The levels of IL-1ß, IL-6, IL-12, and TNF-alpha in mice sera were measured over the first 4 days after infection with Lm (10 3 ), L.inn (10 7 ), and the L.inn::vgc strain (10 7 ). Like L. inn, , L.inn::vgc was not able to produce significant amounts of these cytokines over the first 4 days post-infection in spite of high infection doses (10 7 ) while primary infection with Lm led to high proinflammatory cytokine production (Fig. 2) . Both granuloma formation and delayed-type-hypersensitivity footpad responses have previously been shown to be CD4 + T cell dependent inflammatory responses following infection with Lm. Morphological changes were examined in the spleens on day 3 after i.v. infection. Although the numbers of bacteria in spleens at day 3 post-infection for both Lm and L.inn::vgc were approximately the same (Fig. 1A) , distinct differences in the morphological appearance between spleens isolated from mice infected with Lm and those isolated from mice infected with L.inn::vgc were observed (Fig. 3A ). Splenomegaly associated with extensive granuloma formation was observed in spleens of Lm infected mice, as a result of intensive leukocyte infiltration which was visualized in stained spleen sections (Fig. 3B) , whereas splenomegaly in the absence of granuloma formation was observed in spleens of L.inn::vgc infected mice. Infection with the wild type L.inn did not result in any morphological changes in spleens. These observations were confirmed by antigen-elicited skin responses showing corresponding results (Fig. 3C ). Mice were injected into the left hind footpads with 50 ml of somatic soluble Lm EGD-e antigen (60 ng/ml) at day 9 post-infection. Twentyfour hours later, thickness of the left and right footpads of individual mice were measured. Footpads of mice pre-immunized with L.inn::vgc showed reduced thickness than those of mice preimmunized with the wild type Lm. The wild type L.inn strain did not induce a DTH response in the footpads of these mice. Moreover, antigen-induced CD4+ T cell-derived IFN-gamma production of spleen cells was measured as an indication for a proinflammatory T cell response. Spleen cells were isolated at day 9 post-infection and stimulated in vitro with the released soluble antigen of L. monocytogenes EGD-e (100 ng). Spleen cells from mice immunized with L.inn::vgc produced significantly lower levels of IFN-gamma when compared to spleen cells from mice immunized with wild type Lm. The wild type L.inn strain failed to prime T cells for the production of IFN-gamma (Fig. S1 , supplementary information). Induction of T cell-mediated immunity by the recombinant L.inn::vgc strain A number of cell types are involved in host defense against Listeria. Antigen-specific T lymphocytes mediate recovery from primary listerial infections and protective immunity to subsequent infections [13, 19] . Both CD4 + (helper, MHC class II restricted) and CD8 + (cytotoxic, MHC class I restricted) T cell subpopulations have been implicated [20] . Experimental evidence indicates, however, that CD8 + T cells play the predominant role in mediating protective immunity [21] [22] [23] [24] . The ability of the recombinant L.inn::vgc to induce T-cell mediated immunity as a prerequisite for protective immunity was analyzed. Groups of BALB/c mice were infected with Lm (10 3 ), L.inn (10 7 ), or L.inn::vgc (10 7 ). Two months later, all mice were challenged with a lethal i.v. dose (10 5 ), corresponding to 206LD50, of the wild type Lm, and survival was monitored. As controls, a group of untreated BALB/c mice that received a similar lethal dose of the wild type Lm were included. A single pre-immunization with the L.inn::vgc strain led to a significant protection against subsequent lethal infection with Lm. As expected, all mice that were pre-immunized with sub-lethal doses of Lm were also protected against a lethal listerial infection and survived whereas all non-immunized mice as well as those preimmunized with L.inn died within 4 days after challenge (Fig. 4A) . Entry of Listeria into the cytosol is a critical event for CD8 + T cell recognition and induction of immunity [22] . In order to establish the correlation between the protection of mice preinfected with Lm or the L.inn::vgc strain upon lethal challenge and the induction of CD8 + T cells in response to infection, the generation of antigen-specific MHC class I restricted CD8 + T cells were quantitatively examined. The numbers of antigen-specific MHC class I restricted effector CD8 + T cells induced in mice spleens 9 days after primary infection and 5 days after challenge with the wild type Lm (2610 3 ) was determined through evaluation of the number of IFN-c producing CD8+ T cells induced showing reactivity against the dominant H-2K d restricted LLO 91-99 epitope [25] in an in vitro ELISPOT assay. As shown in Fig. 4B , infection with wild type Lm as well as the L.inn::vgc strain induced significant numbers of LLO 91-99 specific CD8 + T-cells. After recall infection the numbers of LLO 91-99 specific CD8 + T-cells showed a significant increase. On the other hand, infection with L.inn failed to induce a significant number of CD8 + T-cells either after primary infection or after challenge. To address the contribution of effector memory CD8+ T cells in mediating long-lasting immunity after re-infection with the wild type Lm, the expression level of the cell surface adhesion molecule CD62L was quantified. Expression of CD62L is down regulated on the surface of cytotoxic CD8 + T-cells when developed to protective memory T cells [21] . Two months after the primary infection, the number of CD8 + CD62L lo lymphocytes was approximately identical in all groups of primarily infected mice. This number increased dramatically upon re-infection with the wild type Lm (2610 3 ) in mice pre-immunized with Lm as well as with L.inn::vgc while pre-immunization with L.inn was not able to induce CD62L down-regulation seen in the other groups (Fig. 5 ). In addition we have monitored the expression of CD44 on CD8 + T-cells. CD44 is expressed at high levels on memory but not in naïve T-cells [26] . In mice that were primarily infected with Lm and L.inn::vgc, the expression of CD44 was upregulated on CD8 + T-cells 5 days post-challenge infection with the wild type Lm (2610 3 ) while primary infection with L.inn did not lead to a significant change in CD44 expression pattern (Fig. S3 ). In this study, we define a unique vaccine strategy which is based on a rationally designed pathogen by complementation of a non- pathogenic strain with selected genes necessary to induce a vigorous immune response. To illustrate the proof of principle of this strategy we used the Listeria model by taking a non-pathogenic L.inn strain and complementing it with genes from pathogenic Lm which were previously shown to be a sine qua non requirement for intracellular growth and survival [18] . This novel vaccine strategy resulted in generation of a recombinant strain (L.inn::vgc) that possesses properties needed to induce the marked protective immunogenic properties of the wild type Lm but is attenuated in virulence as well as in its capacity to induce host detrimental cellmediated inflammation. The recombinant L.inn::vgc strain showed a significant in vivo survival rate in the first 3 days post-infection (Fig. 1A) . This observation is in accordance with our recent finding that the L.inn::vgc strain is able to survive in phagocytic host cells [18] . Moreover, it has the capability to induce identical Type I interferon at levels similar to wild type Lm. As previously shown by McAffrey et al., 2003, induction of type 1 IFN is a surrogate marker indicating access of Listeria into the cytosol of antigen presenting cells [17] . We have shown that the L.inn::vgc strain could use the complemented virulence factors to escape into the cytosol and subsequently be presented to CD8 T lymphocytes. In this context Zwaferink et al. 2008 have shown a role for IFNb in macrophage cell death. Treatment of macrophages with this cytokine could enhance host-cell membrane permeabilization by listeriolysin consequently leading to cell apoptosis [27] . Especially encouraging was the observation that, although L.inn::vgc strain was injected at a high dose of 10 7 cfu, mice could still efficiently control the infection and showed very low blood levels of pro-inflammatory cytokines thus reducing the detrimental inflammatory responses caused by the wild type Lm. Morphological and histological analysis of spleen after infection with Lm have shown the induction of splenomegaly and granuloma as a result of monocytic infiltrations of the white pulp which were most pronounced on day 3 post-infection while infection with L.inn::vgc only resulted in a splenomegaly without any significant morphological changes detectable (Fig. 3A,3B) . The intensity of the morphological and histological alterations in spleens paralleled the level of Listeria-induced DTH responses, as the in vivo induction of DTH after L.inn::vgc infection was also significantly lower than DTH induction following Lm infection (Fig. 3 and Fig. S1 ). We therefore show that the recombinant L.inn::vgc strain shows a significantly reduced proinflammatory and CD4 + mediated inflammatory response compared to Lm, thereby addressing one major concern regarding the use of live Lm as a vaccine. However the crucial question remained as to whether the L.inn::vgc strain elicits significant adaptive immune responses resembling those of the wild-type strain. Since Lm is located in both phagosomes and cytosol of professional antigen-presenting cells during infection, epitopes derived from Listeria proteins are presented by the MHC pathway thereby priming both effector CD4 + and CD8 + T cells [28, 29] resulting in full elimination of Listeria from the host. Previous experimental studies [30, 31] have revealed that persistence and number of viable microorganisms are important parameters for establishing efficient T cell-mediated immunity. Moreover, it has been shown that the presence of live bacteria in mice organs over the first 48 hours after immunization is critical for the induction of effector CD8 + T cell mechanisms [32] . Indeed we were able to show that all animals immunized with Lm or L.inn::vgc were protected against 206LD 50 of virulent Listeria (Fig. 4A) . Although the L.inn::vgc strain elicits lowered CD4 + mediated inflammatory responses as compared to infection with Lm, it is capable of mounting a successful anti-listerial protective response, indicating that the observed in vivo survival pattern of the L.inn::vgc strain was sufficient to induce protection. The entry of effector T cells into a memory stage, however, is accompanied by the ability to rapidly expand their population during recall responses and to down regulate expression of cell surface markers such as CD62L and CCR7 [33] . It was previously reported that primary infection with the wild type Lm induces down regulation of CD62L on the surface of effector CD8 + T cells which reaches its lowest levels at day 8 post-infection [34] . However, over the following weeks, expression of CD62L is up regulated. During recall infection, CD62L is then rapidly down regulated on the surface of memory CD8 + T cells [32, 35] . In order to correlate protection against challenge with Listeria with antigen specific CD8 + T cells, we examined the induction of LLO 91-99 specific CD8 + T cells in response to primary infection with the different Listeria strains. Infection with L.inn::vgc induced a significant population of cytotoxic CD8 + T lymphocytes (Fig. 4B) which, upon challenge with the wild type Lm, showed a CD62L expression pattern similar to that presented in mice primarily infected with the wild type Lm (Fig. 5) . The identity of the memory T-cells induced in response to L.inn::vgc infection was confirmed by testing the CD44 expression on the CD8 + T-cells following recall infection with Lm where high expression of CD44 was observed on CD8 + T-cells derived from mice primarily infected with the recombinant L.inn::vgc strain but not with the wild type L.inn (Fig. S3) . The inability of the Lm strain lacking listeriolysin O (LLO) [36] as well as the L.inn strain expressing only LLO [37] to induce a protective T cell response reflects the requirement of the entire virulence gene cluster in conferring a long lasting immunity. We therefore show that a non-pathogenic L.inn strain complemented with the entire vgc is capable of inducing a vigorous anti-listerial response. Even though we have demonstrated a vigorous immune response following i.v. infection the immune response to Listeria can vary considerably depending on the route of administration. Using the intraperitoneal route of infection, we obtained a similar result i.e. protection following pre-infection with the Lm and L. inn::vgc strains but not with mice pre-immunized with the L. inn strain (Fig. S2) . Thus despite a different route of infection Lm::vgc is able to induce protection in-vivo. The mouse is not a suitable and reproducible model for evaluating oral immunization protocols because of the specificity of the listerial InlA molecule [38] . Therefore experiments examining mucosal immunity will have to be carried out in the guinea pig model of listerial infection. Recently, highly attenuated mutants of Lm have been developed as candidates for vaccine vectors [3, 39] , however, a single immunization with these strains was not sufficient for the induction of protective cellular immunity. Here a transcomplemented strain of a non-pathogenic L. inn strain expressing genes of the vgc cluster provides robust protection with a single dose of 10 7 cfu bacteria without causing any signs of overt illness. The LD50 of the wild type Lm is around 5000 cfu. As shown in figure 1A , the L.inn::vgc strain does not grow in-vivo beyond day 3 post-infection and is subsequently eliminated. These properties, imparting protective responses and rapid elimination from the host are considered to be desirable properties for successful vaccine vectors. Our results, namely, the in vivo survival pattern, the induction of interferon's and antigen specific CD8 + T cells, the lack of overt detrimental inflammatory reactions and most importantly the induction of protection against challenge with Listeria, allow the conclusion that the L.inn::vgc strain is potentially capable of inducing protection and that further development of this strain as a suitable live bacterial vaccine vector in clinical settings are warranted. Mice experiments were done according to the requirements of Justus-Liebig University Giessen Animal Ethics Committees with ethics approval number: 63/2007. Animals were sacrificed using CO2 asphyxiation and the appropriate organs aseptically harvested. Six to eight week-old female BALB/c mice, purchased from Harlan Winkelmann (Borchen, Germany), were kept at our breeding facilities in specific-pathogen-free conditions and used in all experiments. Bacterial strains used in this study are wild type Listeria monocytogenes EGDe serotype 1/2a (Lm) [40] , wild type L. innocua strain (serotype 6a NCTC 11288) [41] transformed with either the recently characterized gram+ve/gram-ve shuttle pUvBBAC+vgc1 vector and referred to as (L.inn::vgc strain) or the pUvBBAC vector without the inserted vgc and referred to as L.inn [18] Bacteria were grown in brain-heart infusion (BHI) (Difco, Augsburg; Germany) broth in presence or absence of 5 mg/ml erythromycin. For each experiment, erythromycin was used as a selective antibiotic for growth of L.inn::vgc and the wild type L.inn harbouring the pUvBBAC vector. Wild type L. monocytogenes was grown in absence of erythromycin. In all experiments, fresh cultures of bacteria, prepared from an overnight culture, were used. Briefly, bacteria were grown in Brain Heart Infusion (BHI) at 37uC, harvested in the exponential growth phase and washed twice with PBS. The pellet was resuspended in PBS and the bacterial concentration was calibrated by optical absorption. Further dilutions were prepared in PBS to obtain required numbers of bacteria for infection. The protocols for animal handling were previously approved by our institutional Animal Ethics Committee (protocol number 63/ 2007). Bone marrow-derived macrophages were isolated from 4 to 6 week old C57Bl/6 female mice and grown and differentiated for 7 days in L929 conditioned medium to an approximate concentration of 2,5610 5 cells/well in 6-well plates. On the day of infection the medium was exchanged against MDEM medium with 1% FCS and the cells were infected with 5610 6 cfu per well with the wild type Lm and L.inn strains as well as the recombinant L.inn::vgc strain for 2 h and 8 h. The cells were lysed and their total RNA was isolated. For every bacterial strain and negative control the cells of at least two wells of a six well tissue culture plaque were lysed and total RNA was isolated. Prior to lysis culture medium was aspirated and cells were lysed using RLT lysis buffer (Qiagen, Germany). Total RNA was isolated using the RNeasy Mini Kit and the RNase free DNase I set (Qiagen) following the manufacturers protocol. The RNA was recovered in RNase free water, heat denatured for 10 min. at 65uC; quantified with the NanoDropH ND-1000 UV-Vis Spectrophotometer (NanoDrop Technologies, USA) and a quality profile with the Agilent 2100 bioanalyzer (Agilent Technologies, Germany) was made. First-strand cDNA was synthesized with 500 ng of purified RNA using SuperScriptII (Invitrogen) and a mixture of T21 and random nonamer primers (Metabion) following the instructions for the reverse transcription reaction recommended for the Quanti-Tect SYBR Green PCR Kit (Qiagen). Real-time quantitative PCR was performed on an ABI Prism 7700 real time cycler. The relative expression of the targets IFNa2 (Interferon alpha 2) and IFNb1 (Interferon beta) were normalized to that of two reference genes: SDHA (Succinate dehydrogenase alpha subunit) and PPIA (peptidylprolyl isomerase A). Finally a mean of the fold change of the target versus each of the reference genes was taken as the final value. Somatic soluble antigen was prepared by culturing Lm in tryptic soy broth for 18 h, washing it in PBS, and subsequently subjecting it to ultrasonication.1 g (wet weight) of bacterial cells were suspended in 10 ml of PBS and sonicated five times for 1 min (87.5%, output, degree 7 on a sonifier model S-125; Branson Sonic Power, USA) on ice. The sonicated suspension was centrifuged at 39 000 U for 50 min, and the supernatant was filter sterilized (pore size,0.45 mm) and stored at 220uC at a dilution of 1:100 in PBS [42] . Primary in vivo infection with Lm (10 3 ), the wild type L.inn (10 7 ), or the L.inn::vgc (10 7 ) strain was performed by an intravenous injection of viable bacteria in a volume of 0.2 ml PBS. Bacterial growth in spleens and livers was determined by plating 10-fold serial dilutions of the organ homogenates on BHI agar plates. The detection limit of this procedure was 10 2 colony forming units (CFU) per organ. Colonies were counted after 24 h of incubation at 37uC. Cytokine production was assayed from the collected sera of infected mice using a multiplex cytokine assay kit and Luminex technology (Bio-Rad). Balb/C mice were infected with Lm (10 3 ), the wild type L.inn (10 7 ), or the L.inn::vgc (10 7 ) strain. Sera were aseptically isolated on days 1, 2, 3, and 4 post-infection. Four cytokines were tested: TNFa, IL-1b, IL-6, and IL-12(p70) and cytokine levels were presented as absolute concentrations in pg/ ml. Spleens were aseptically isolated from mice previously infected with the different Listeria strains as mentioned above and examined for morphological alterations. Tissues were fixed in 10% neutral buffered formalin, embedded in paraffin, sectioned, and 5 mm sections were stained with hematoxylin and eosin (HE). Pathoplogical foci in spleen sections were then microscopically examined (Keyence). Spleens were aseptically removed from mice at day 9 postinfection with the wild type Lm, the wild type L.inn, or the L.inn::vgc strain. Spleen cells were isolated and antigen (LLO 91-99 ) specific IFN-c producing CD8+T cells were determined in the spleens of mice after i.v. infection with the same bacterial strains mentioned above by using an ELISPOT system as previously described [23, 35] . Quantification of IFN-gamma in cell culture supernatants IFN-gamma was measured in the supernatants of splenocytes by using a mouse IFN-gamma ELISA kit, BD optEIA TM (BD Biosciences Pharmigen) according to the manufacturer instructions. The assay was performed in duplicates, and data represent means 6 standard error. For flow cytometry analysis, approximately 1610 6 splenocytes, isolated from infected mice (Lm, L.inn, and L.inn::vgc strains) were stained with FITC labelled anti-CD8 and biotinylated anti-CD62L or anti-CD44 (pharMingen, Becton Dickinson). PEconjugated streptavidin was used to detect the binding of anti-CD62L or anti-CD44 on the cell surface. Flow cytometry was performed using a FACS Calibur flow cytometer and further analyzed with CELL Quest software (Becton Dickinson, CA). All mice, pre-immunized with wild type Lm, the wild type L.inn strain and the L.inn::vgc strain were challenged 2 months later with a 206LD 50 (10 5 ) lethal dose of wild type Lm. A group of non preimmunized Balb/c mice were included as controls. Survival of mice was monitored for several days and expressed as percentage of animals surviving challenge with Lm. Data are representative of at least three independent experiments. Significance of the represented data was calculated using ANOVA (analysis of variance). Data are expressed as mean 6 standard errors (S.E.). Figure S1 Listeria-induced IFN-gamma production by spleen cells 9 days after infection (i.v.). Mice were infected with 10 3 CFU of Lm, 10 7 CFU wild type L.inn, or with 10 7 CFU of L.inn::vgc strain. On day 9 after infection, mice were killed and spleens removed. Single cell suspensions were stimulated in vitro with secreted soluble Listeria antigen to produce IFN-gamma. After 48 hours, culture supernatants were tested for presence of IFNgamma by ELISA. *P,0.05 (EGD-e vs L.inn::vgc). (TIF) Figure S2 Intraperitoneal infection with the L.inn::vgc strain induces protective immunity. Mice were infected intraperitoneally with Lm, L.inn and the L.inn::vgc strain as described in figure 4A . After 2 months all mice were challenged i.v. with a lethal dose (206LD 50 ) of the wild type Lm. As a control, a group of uninfected normal mice was included. Survival was monitored up to 8 days after challenge. (TIF) Figure S3 Quantification of CD44 expression on CD8 + splenocytes following primary and recall infection with Lm, L.inn and the L.inn::vgc strain. Flow cytometry was performed on spleen cells, isolated from mice on day 60 after the primary infection or day 5 after the challenge. Cells were stained with FITC-labelled anti-Lyt-2 and PE-labelled anti-CD44. Numbers shown are gated CD8 + CD44 hi T cells and analyzed with CELLQuest software. (TIF)

Transcriptomics of In Vitro Immune-Stimulated Hemocytes from the Manila Clam Ruditapes philippinarum Using High-Throughput Sequencing

BACKGROUND: The Manila clam (Ruditapes philippinarum) is a worldwide cultured bivalve species with important commercial value. Diseases affecting this species can result in large economic losses. Because knowledge of the molecular mechanisms of the immune response in bivalves, especially clams, is scarce and fragmentary, we sequenced RNA from immune-stimulated R. philippinarum hemocytes by 454-pyrosequencing to identify genes involved in their immune defense against infectious diseases. METHODOLOGY AND PRINCIPAL FINDINGS: High-throughput deep sequencing of R. philippinarum using 454 pyrosequencing technology yielded 974,976 high-quality reads with an average read length of 250 bp. The reads were assembled into 51,265 contigs and the 44.7% of the translated nucleotide sequences into protein were annotated successfully. The 35 most frequently found contigs included a large number of immune-related genes, and a more detailed analysis showed the presence of putative members of several immune pathways and processes like the apoptosis, the toll like signaling pathway and the complement cascade. We have found sequences from molecules never described in bivalves before, especially in the complement pathway where almost all the components are present. CONCLUSIONS: This study represents the first transcriptome analysis using 454-pyrosequencing conducted on R. philippinarum focused on its immune system. Our results will provide a rich source of data to discover and identify new genes, which will serve as a basis for microarray construction and the study of gene expression as well as for the identification of genetic markers. The discovery of new immune sequences was very productive and resulted in a large variety of contigs that may play a role in the defense mechanisms of Ruditapes philippinarum.

The Manila clam (Ruditapes philippinarum) is a cultured bivalve species with important commercial value in Europe and Asia, and its culture has expanded in recent years. Nevertheless, diseases produced by a wide range of microorganisms, from viruses to metazoan parasites, can result in large economical losses. Among clam diseases, the majority of pathologies are associated with the Vibrio and Perkinsus genera [1] [2] [3] . Although molluscs lack a specific immune system, the innate response involving circulating hemocytes and a large variety of molecular effectors seems to be an efficient defense method to respond to external aggressions by detecting the molecular signatures of infection [4] [5] [6] [7] [8] ; however, not many immune pathways have been identified in these animals. Although knowledge of bivalve immune-related genes has increased in the last few years, the available information is still scarce and fragmentary. Most of the data concern mussels and Eastern and Pacific oysters [9] [10] [11] [12] [13] [14] , and very limited information is available on the expressed immune genes of R. philippinarum. Recently, the expression of 13 immune-related genes of Ruditapes philippinarum and Ruditapes decussatus were characterized in response to a Vibrio alginolyticus challenge [15] . Also, a recent 454 pyrosequencing study was carried out by Milan et al. [16] , who sequenced two normalized cDNA libraries representing a mixture of adult tissues and larvae from R. philippinarum. Even more recently Ghiselli et al. [17] , have de novo assembled the R. philippinarum gonad transcriptome with the Illumina technology. Moreover, a few transcripts encoded by genes putatively involved in the clam immune response against Perkinsus olseni have been reported by cDNA library sequencing [18] . Currently (19/12/ 2011) , there are 5,662 ESTs belonging to R. philippinarum in the GenBank database. The European Marine Genomics Network has increased the number of ESTs for marine mollusc species particularly for ecologically and commercially important groups that are less studied, such as mussels and clams [19] . Unfortunately, most of the available resources are not annotated or well described, limiting the identification of important genes and genetic markers for future aquaculture applications. The use of 454-pyrosequencing is a fast and efficient approach for gene discovery and enrichment of transcriptomes in non-model organisms [20] . This relatively low-cost technology facilitates the rapid production of a large volume of data, which is its main advantage over conventional sequencing methods [21] . In the present work, we undertook an important effort to significantly increase the number of R. philippinarum ESTs in the public databases. Specially, the aim of this work was to discover new immune-related genes using pyrosequencing on the 454 GS FLX (Roche-454 Life Sciences) platform with the Titanium reagents. To achieve this goal, we sequenced the transcriptome of R. philippinarum hemocytes previously stimulated with different pathogen-associated molecular patterns (PAMPs) to obtain the greatest number of immune-related transcripts as possible. The raw data are accessible in the NCBI Short Read Archive (Accession number: SRA046855.1). The R. philippinarum normalized cDNA library was sequenced with 454 GS FLX technology as shown in Figure 1 . Sequencing and assembly statistics are summarized in Table 1 . Briefly, a total of 975,190 raw nucleotide reads averaging 284.1 bp in length were obtained. Of these, 974,976 exceeded our minimum quality standards and were used in the MIRA assembly. A total of 842,917 quality reads were assembled into 51,265 contigs, corresponding to 29.9 megabases (Mb). The length of the contigs varied from 40 to 5565 bp, with an average length of 582.4 bp and an average coverage of 5.7 reads. Singletons were discarded, resulting in 37,093 contigs formed by at least 2 ESTs, and 26,675 of these contigs were longer than 500 bp. Clustering the contigs resulted in 1,689 clusters with more than one contig. The distribution of contig length and the number of ESTs per contig, as well as the contig distribution by cluster are all shown in Figure 2 . Even though the knowledge of expressed genes in bivalves has increased in the last few years, it is still limited. Indeed, only 41,598 nucleotide sequences, 362,149 ESTs, 24,139 proteins and 704 genes from the class Bivalvia have been deposited in the GenBank public database (19/12/11) , and the top entries are for the Mytilus and Crassostrea genera. For Ruditapes philippinarum, these numbers are reduced to 5,662 ESTs, 612 proteins and 12 genes. This evidences the lack of information which prompted the recent efforts to increase the number of annotated sequences of bivalves in the databases. For non-model species, functional and comparative genomics is possible after obtaining good EST databases. These studies seem to be the best resource for deciphering the putative function of novel genes, which would otherwise remain ''unknown''. NCBI Swissprot, NCBI Metazoan Refseq, the NCBI nonredundant and the UniprotKB/Trembl protein databases were chosen to annotate the contigs that were at least 100 bp long (49, 847) . The percentage of contigs annotated with a cut off evalue of 10e-3 was 44.7%. Contig sequences and annotations are included in Table S1 . Of these contigs, 3.26% matched sequences from bivalve species and the remaining matched to non-Bivalvia mollusc classes (4.13%), other animals (81.38%), plants (2.58%), fungi (1.78%), protozoa (1.50%), bacteria (4.95%), archaea (0.20%), viruses (0.21%) and undefined sequences (0.01%). As shown in Figure 3A , the species with the most sequence matches was Homo sapiens with 3,106 occurrences. The first mollusc in the top 35 list was Lymnaea stagnalis at position 11. The first bivalve, Meretrix lusoria, appeared at position 17. R. philippinarum was at position 25 with 124 occurrences. Notably, a high percentage of the sequences had homology with chordates, arthropods and gastropods ( Figure 3B and C), and only 343 contigs matched with sequences from the Veneroida order ( Figure 3D ). These values can be explained by the higher representation of those groups in the databases as compared to bivalves and the quality of the annotation in the databases, which has been reported in another bivalve transcriptomic study [22] . The data shown highlight, once again, the necessity of enriching the databases with bivalve sequences. A detailed classification of predicted protein function is shown for the top 35 BLASTx hits ( Figure 4A ). The list is headed by actin with 903 occurrences, followed by ferritin, an angiopoietin-like protein and lysozyme. An abundance of proteins directly involved in the immune response was predicted for this 454 run; ferritin, lysozyme, C1q domain containing protein, galectin-3 and hemagglutinin/amebocyte aggregation factor precursor are immune-related proteins present on the top 35 list. Ferritin has an important role in the immune response. It captures circulating iron to overcome an infection and also functions as a proinflammatory cytokine via the iron-independent nuclear factor kappa B (NF-kB) pathway [23] . Lysozyme is a key protein in the innate immune responses of invertebrates against Gram-negative bacterial infections and could also have antifungal properties. In addition, it provides nutrition through its digestive properties as it is a hydrolytic protein that can break the glycosidic union of the peptidoglycans of the bacteria cell wall [24] . The C1q domain containing proteins are a family of proteins that form part of the complement system. The C1q superfamily members have been found to be involved in pathogen recognition, inflammation, apoptosis, autoimmunity and cell differentiation. In fact, C1q can be produced in response to infection and it can promote cell survival through the NF-kB pathway [25] . Galectin-3 is a central regulator of acute and chronic inflammatory responses through its effects on cell activation, cell migration, and the regulation of apoptosis in immune cells [26] . The hemagglutinin/amebocyte aggregation factor is a single chain polypeptide involved in blood coagulation and adhesion processes such as self-nonself recognition, agglutination and aggregation processes. The hemagglutinin/ amebocyte aggregation factor and lectins play important roles in defense, specifically in the recognition and destruction of invading microorganisms [27] . Other proteins that are not specifically related to the immune response but could play a role in defense mechanisms include the following: angiopoietin-like proteins, apolipoprotein D and the integral membrane protein 2B. In other animals, angiopoietin-like proteins (ANGPTL) potently regulate angiogenesis, but a subset also function in energy metabolism. Specifically, ANGPTL2, the most represented ANGPTL, promotes vascular inflammation rather than angiogenesis in skin and adipose tissues. Inflammation occurs via the a5b1 integrin/Rac1/NF-kB pathway, which is evidenced by an increase in leukocyte infiltration, blood vessel permeability and the expression of inflammatory cytokines (tumor necrosis factor-a, interleukin-6 and interleukin-1b) [28] . Apolipoprotein D (apoD) has been associated with inflammation. Pathological and stressful situations involving inflammation or growth arrest have the capacity to increase its expression. This effect seems to be triggered by LPS, interleukin-1, interleukin-6 and glucocorticoids and is likely mediated by the NF-kB pathway, as there are several conserved NF-kB binding sites in the apoD promoter (APRE-3 and AP-1 binding sites are also present). The highest affinity ligand for apoD is arachidonic acid, which apoD traps when it is released from the cellular membrane after inflammatory stimuli and, thus, prevents its subsequent conversion in pro-inflammatory eicosanoids. Within the cell, apoD could modulate signal transduction pathways and nuclear processes such as transcription activation, cell cycling and apoptosis. In summary, apoD induction is specific to ongoing cellular stress and could be part of the protective components of mild inflammation [29] [30] [31] . Finally, the short form of the integral membrane protein 2B (ITM2Bs) can induce apoptosis via a caspase-dependent mitochondrial pathway [32] . To avoid redundancy, the longest contig of each cluster was used for Gene Ontology terms assignment. A total of 23.05% of the representative clusters matched with at least one GO term. Concerning cellular components ( Figure 4B ), the highest percentage of GO terms were in the groups of cell and cell part with 25.9% in each; organelle and organelle part represented 19.67% and 11.38%, respectively. Within the molecular function classification ( Figure 4C ), the most represented group was binding with 49.25% of the terms, which was followed by catalytic activity (29.12%) and structural molecular activity (4.60%). With regard to biological process ( Figure 4D ), cellular and metabolic processes were the highest represented groups with 16.78% and 12.43% of the terms, respectively, which was followed by biological regulation (10.18%). Similarities between the R. philippinarum transcriptome and another four bivalve species sequences were analyzed by comparative genomics (Crassostrea gigas of the family Ostreidae, Bathymodiolus azoricus and Mytilus galloprovincialis of the family Mytilidae and Laternula elliptica of the family Laternulidae). This analysis could identify specific transcripts that are conserved in these five species. A Venn diagram was constructed using unique sequences from these databases according to the gene identifier (gi id number) of each sequence in its respective database: 207,764 from C. gigas, 76,055 from B. azoricus, 121,318 from M. galloprovincialis and 1,034,379 from L. elliptica. C. gigas was chosen because is the most represented bivalve species in the public databases. The other three species are bivalves that have been studied in transcriptomic assays. Figure 5 shows that of the total 29,679 clusters, 72% were found exclusively in the R. philippinarum group, while only 7.59% shared significant similarity with all five species. The number of coincidences among other groups was very low (4.14% to 0.31% of sequences), suggesting that 21,454 new sequences were discovered within the bivalve group. The percentage of new sequences is very high compared to previous transcriptomic studies [33] [34] , in which the fraction of new transcripts was approximately 45%. One possible explanation for this discrepancy is the low number of nucleotide and EST sequences currently available in public databases for R. philippinarum, but these transcripts could also be regions in which homology is not reached, such as 59 and 39 untranslated regions or genes with a high mutation rate. On the other hand, a comparison between our 454 results and the Milan et al. [16] transcriptome using a BLASTn approach is summarized in Table 2 Immune-related sequences R. philippinarum hemocytes were subjected to immune stimulation using several different PAMPs to enrich the EST collection with immune-related sequences. The objective was to obtain a more complete view of clam responses to pathogens. A keyword list and GO immune-related terms were used to find proteins putatively involved in the immune system. After this selection step, we found that more than 10% of the proteins predicted from the contig sequences had a possible immune function. Some sequences were found to be clustered in common, well-recognized immune pathways, such as the complement, apoptosis and toll-like receptors pathways, indicating conserved ancient mechanisms in bivalves ( Figures 6, 7, 8 ). The complement system is composed of over 30 plasma proteins that collaborate to distinguish and eliminate pathogens. C3 is the central component in this system. In vertebrates, it is proteolytically activated by a C3 convertase through both the classic, lectininduced and alternative routes [35] . Although the complement pathway has not been extensively described in bivalves, there is evidence that supports the presence of this defense mechanism. ESTs with homology to the C1q domain have been detected in the American oyster, C. virginica [36] , the tropical clam Codakia orbicularis [37] , the Zhikong scallop Chlamys farreri [38] and the mussel M. galloprovincialis [39] [40] . More recently, a novel C1q adiponectin-like, a C3 and a factor B-like proteins have been identified in the carpet shell clam R. decussatus [41] [42] . These data support the putative presence of the complement system in bivalves. Our pyrosequencing results, using the BLASTx similarity approach, showed that the complement pathway in R. philippinarum was almost complete as compared to the KEGG reference pathway ( Figure 6 ). Only the complement components C1r, C1s, C6, C7 and C8 were not detected. i. Lectins. Lectins are a family of carbohydrate-recognition proteins that play crucial self-and non-self-recognition roles in innate immunity and can be found in soluble or membraneassociated forms. They may initiate effector mechanisms against pathogens, such as agglutination, immobilization and complement -mediated opsonization and lysis [43] . Several types of lectins have been cloned or purified from the Manila clam, R. philippinarum [44] [45] [46] , and their function and expression were also studied [18, 47] . Also, a Manila clam tandemrepeat galectin, which is induced upon infection with Perkinsus olseni, has been characterized [46] . Lectin sequences have been found in the stimulated hemocytes studied in our work: 23 of the contigs are homologous to C-type lectins (calcium-dependent carbohydrate-binding lectins that have characteristic carbohydrate-recognition domains), 115 are homologous to galectins (characterized by a conserved sequence motif in their carbohydrate recognition domain and a specific affinity for bgalactosides), 4 contigs have homology with ficolin A and B (a group of oligomeric lectins with subunits consisting of both collagen-like and fibrinogen-like domains) and 34 contigs have homology with other groups of lectins such as lactose-, mannoseor sialic acid-binding lectins. ii. b-glucan recognition proteins. b-glucan recognition proteins are involved in the recognition of invading fungal organisms. They bind specifically to b-1,3-glucan stimulating short-term immune responses. Although these receptors have been partially sequenced in several bivalves, there is only one complete description of them in the scallop Chlamys farreri [48] . Two contigs with homology to the beta-1,3-glucan-binding protein were found in our study. iii. Peptidoglycan recognition proteins. Peptidoglycan recognition proteins (PGRPs) specifically bind peptidoglycans, which is a major component of the bacterial cell wall. This family of proteins influences host-pathogen interactions through their pro-and anti-inflammatory properties that are independent of their hydrolytic and antibacterial activities. In bivalves, they were first identified in the scallops C. farreri and A. irradians [49, 50] and the Pacific oyster C. gigas, and from the latter four different types of PGRPs were identified [51] . Peptidoglycan-recognition proteins and a peptidoglycan-binding domain containing protein have been found for the first time in R. philippinarum in our results and were present 4 and 1 times, respectively. iv. Toll-like receptors. Toll-like receptors (TLRs) are an ancient family of pattern recognition receptors that play key roles in detecting non-self substances and activating the immune system. The unique bivalve TLR was identified and characterized in the Zhikong Scallop, C. farreri [52] . TLR 2, 6 and 13 were present among the pyrosequencing results. TLR2 and TLR6 form a heterodimer, which senses and recognizes various components from bacteria, mycoplasma, fungi and viruses [53] . TLR13 is a novel and poorly characterized member of the Toll-like receptor family. Although the exact role of TLR13 is currently unknown, phylogenic analysis indicates that TLR13 is a member of the TLR11 subfamily [54] suggesting that it could recognize urinary pathogenic E. coli [55] . It has been demonstrated that TLR13 colocalizes and interacts with UNC93B1, a molecule located in the endoplasmic reticulum, which strongly suggests that TLR13 might be found inside cells and might play a role in recognizing viral infections [56] . Figure 7 summarizes the TLR signaling pathway with the corresponding molecules found in the R. philippinarum transcriptome. Pathogen proteases are important virulence factors that facilitate infection, diminish the activity of lysozymes and quench the agglutination capacity of hemocytes. Because protease inhibitors play important roles in invertebrate immunity by protecting hosts through the direct inactivation of pathogen proteases, many bivalves have developed protease inhibitors to regulate the activities of pathogen proteases [1] . Some genes encoding protease inhibitors were identified in C. gigas [57] , A. irradians [58] , C. farreri [59] and C. virginica; in the latter a novel family of serine protease inhibitors was also characterized [60] [61] [62] . A total of 23 contigs with homology to Serine, Cystein, Kunitzand Kazal-type protease inhibitors and metalloprotease inhibitors were found among our results. Lysozyme was one of the most represented groups of immune genes in this transcriptome study with 208 contigs present. It is an antibacterial molecule present in numerous animals including bivalves. Although lysozyme activity was first reported in molluscs over 30 years ago, complete sequences were published only recently including those of R. philippinarum [24] . Antimicrobial peptides (AMPs) are small, gene-encoded, cationic peptides that constitute important innate immune effectors from organisms spanning most of the phylogenetic spectrum. AMPs alter the permeability of the pathogen membrane and cause cellular lysis [63] . In bivalves, they were first purified from mussel hemocyte granules [64, 65] . In mussels, the AMP myticin C was found to have a high polymorphic variability as well as chemotactic and immunoregulatory roles [66, 67] . In clams, two AMPs with similarity to mussel myticin and mytilin [68] and a big defensin [69] are known. We were able to detect 36 contigs with homology to different defensins: defensin-1 (American oyster defensin), defensin MGD-1 (Mediterranean mussel defensin) and the big defensin previously mentioned. Four contigs were similar to an unpublished defensin sequence from Venerupis ( = Ruditapes) philippinarum. The primary role of heat shock proteins (HSPs) is to function as molecular chaperones. Their up-regulation also represents an important mechanism in the stress response [70] , and their activity is closely linked to the innate immune system. HSPs mediate the mitochondrial apoptosis pathway and affect the regulation of NF-kB [71] . HSPs are well studied in bivalves. For R. philippinarum, several assays have been developed to better understand the HSPs profile in response to heavy metals and pathogen stresses [72] [73] [74] . The most important and well-studied groups of HSPs were present in our R. philippinarum transcriptome (HSP27, HSP40/ DnaJ, HSP70 and HSP90), but other, less common HSPs were also represented (HSP10, HSP22, HSP83 and some members from the HSP90 family). Recently, several genes related to the inflammatory response against LPS stimulation have been detected in bivalves. Such is the case of the LPS-induced TNF-a factor (LITAF), which is a novel transcription factor that critically regulates the expression of TNFa and various inflammatory cytokines in response to LPS stimulation. It has been described in three bivalve species: Pinctada fucata [75] , C. gigas [76] and C. farreri [77] . Other TNF-related genes have been identified in the Zhikong scallop, such as a TNFR homologue [78] and a tumor necrosis factor receptor-associated factor 6 (TRAF6), which is a key signaling adaptor molecule common to the TNFR superfamily and to the IL-1R/TLR family [79] . Figure 7 shows that several components of the TLR signaling pathway that are present in our transcriptomic sequences (MyD88, IRAK4, TRAF-3 and -6, TRAM, BTK, RAC-1, PI3K, AKT, BTK and TANK). A total of 1,918 contigs, 8.43% of those annotated, had homology with the main groups of putatively pathogenic organisms such as viruses (47 hits), bacteria (1,126 hits), protozoa (341 hits) and fungi (404 hits). Figure 9 displays the taxonomic classification of these sequences and Table 3 summarizes a list of the known bivalve pathogens found in our results. Bacteria constitute the main group found among the sequences not belonging to the clam. As filter-feeding animals, bivalves can concentrate a large amount of bacteria and it could be one of their sources of food [24] . Because Vibrio spp. are ubiquitous in aquatic ecosystems, it was expected that the Vibrionales order, with 141 hits, would be the most predominant. Several species of the Vibrio genus are among the main causes of disease in bivalves specifically causing bacillary necrosis in larval stages [80] . Is noticeable that sequences belonging to the causative agent of Brown Ring Disease in adults of Manila Clam, Vibrio tapetis, have not been found. Perkinsus marinus, with 2 matches, is the only bivalve pathogen found within the protozoa (Alveolata) group. Perkinsosis is produced by species from the genus Perkinsus. Both P. marinus and P. olseni have been associated with mortalities in populations of various groups of molluscs around the world and are catalogued as notifiable pathogens by the OIE. Viruses were the least represented among pathogens. The Baculoviridae family was the most predominant with 21 matches, but the corresponding sequences were inhibitors of apoptosis (IAPs) [81] that could also be part of the clam's transcriptome. Five viral families were found in our transcriptome study: Iridoviridae, Herpesviridae, Malacoherpesviridae, Picornaviridae and Retroviridae. A well-known bivalve pathogen was also identified, the ostreid herpesvirus 1, which has been previously been found to infect clams [82] . Fungi had 404 matches in our results. It is known that bivalves are sensitive to fungal diseases, which can degrade the shell or affect the larval bivalve stages [83, 84] . This study represents the first R. philippinarum transcriptome analysis focused on its immune system using a 454-pyrosequencing approach and complements the recent pyrosequencing assay carried out by Milan et al. [16] . The discovery of new immune sequences was effective, resulting in an enormous variety of contigs corresponding to molecules that could play a role in the defense mechanisms. More than 10% of our results had relationship with immunity. This new resource is now gathered in the NCBI Short Read Archive with the accession number: SRA046855.1. Our results will provide a rich source of data to discover and identify new genes, which will serve as a basis for microarray construction and gene expression studies as well as for the identification of genetic markers for various applications including the selection of families in the aquaculture sector. We have found sequences from molecules never described in bivalves before like C2, C4, C5, C9, AIF, Bax, AKT, TLR6 and TLR13, among others. As a part of this work, three immune pathways in R. philippinarum have been characterized, the apoptosis, the toll like signaling pathway and the complement cascade, which could help us to better understand the resistance mechanisms of this economically important aquaculture clam species. Animal sampling and in vitro stimulation of hemocytes R. philippinarum clams were obtained from a commercial shellfish farm (Vigo, Galicia, Spain). Clams were maintained in open circuit filtered sea water tanks at 15uC with aeration and were fed A total of 100 clams were notched in the shell in the area adjacent to the anterior adductor muscle. A sample of 500 ul of hemolymph was withdrawn from the adductor muscle of each clam with an insulin syringe, pooled and then distributed in 6-well plates, 7 ml per well, in a total of 7 wells, one for each treatment. Hemocytes were allowed to settle to the base of the wells for 30 min at 15uC in the darkness. Then, the hemocytes were stimulated with 50 mg/ml of Polyinosinic:polycytidylic acid (Poly I:C), Peptidoglycans, ß-Glucan, Vibrio anguillarum DNA (CpG), Lipopolysaccharide (LPS), Lipoteichoic acid (LTA) or 1610 6 UFC/ml of heat-inactivated Vibrio anguillarum (one stimulus per well) for 3 h at 15uC. All stimuli were purchased from SIGMA. Pyrosequencing. After stimulation, hemolymph was centrifuged at 1700 g at 4uC for 5 minutes, the pellet was resuspended in 1 ml of Trizol (Invitrogen) and RNA was extracted following the manufacturer's protocol. After RNA extraction, samples were treated with Turbo DNase free (Ambion) to eliminate DNA. Next, the concentration and purity of the RNA samples were measured using a NanoDrop ND1000 spectrophotometer. The RNA quality was assessed in a Bioanalyzer 2010 (Agilent Technologies). From each sample, 1 mg of RNA was pooled and used for the production of normalized cDNA for 454 sequencing in the Unitat de Genòmica (SCT-UB, Barcelona, Spain). Full-length-enriched double stranded cDNA was synthesized from 1,5 mg of pooled total RNA using MINT cDNA synthesis kit (Evrogen, Moscow, Russia) according to manufacturer's protocol, and was subsequently purified using the QIAquick PCR Purification Kit (Qiagen USA, Valencia, CA). The amplified cDNA was normalized using Trimmer kit (Evrogen, Moscow, Russia) to minimize differences in representation of transcripts. The method involves denaturation-reassociation of cDNA, followed by a digestion with a Duplex-Specific Nuclease (DSN) enzyme [85, 86] . The enzymatic degradation occurs primarily on the highly abundant cDNA fraction. The single-stranded cDNA fraction was then amplified twice by sequential PCR reactions according to the manufacturer's protocol. Normalized cDNA was purified using the QIAquick PCR Purification Kit (Qiagen USA, Valencia, CA). To generate the 454 library, 500 ng of normalized cDNA were used. cDNA was fractionated into small, 300-to 800-basepair fragments and the specific A and B adaptors were ligated to both the 39 and 59 ends of the fragments. The A and B adaptors were domain; PKC: Protein kinase C; PTEN: Phosphatidylinositol-3,4,5-trisphosphate 3-phosphatase and dual-specificity protein phosphatase PTEN; RAIDD: Caspase and RIP adapter with death domain; TNF R1: Tumor necrosis factor receptor 1; TNF-a: Tumor necrosis factor alpha; TRADD: TNF receptor type 1-associated DEATH domain protein; TRAF2: TNF receptor-associated factor 2; TRAIL: TNF-related apoptosis-inducing ligand; TRAIL decoy: Decoy TRAIL receptor without death domain; TRAIL-R: TRAIL receptor. doi:10.1371/journal.pone.0035009.g008 used for purification, amplification, and sequencing steps. One sequencing run was performed on the GS-FLX using Titanium chemistry. 454 Sequencing is based on sequencing-by-synthesis, addition of one nucleotide, or more, complementary to the template strand results in a chemiluminescent signal recorded by the CCD camera within the instrument. The signal strength is proportional to the number of nucleotides incorporated in a single nucleotide flow. All reagents and protocols used were from Roche 454 Life Sciences, USA. Pyrosequencing raw data, comprised of 975,190 reads, were processed with the Roche quality control pipeline using the default settings. Seqclean (http://compbio.dfci.harvard.edu/tgi/software/) software was used to screen for and remove normalization adaptor sequences, homopolymers and reads shorter than 40 bp prior to assembly. A total of 974,973 quality reads were subjected to MIRA, version 3.2.0 [87] , to assemble the transcriptome. By default, MIRA takes into account only contigs with at least 2 reads. The other reads go into debris, which might include singletons, repeats, low complexity sequences and sequences shorter than 40 bp. NCBI Blastclust was used to group similar contigs into clusters (groups of transcripts from the same gene). Two sequences were grouped if at least 60% of the positions had at least 95% identity. The 51,265 contigs were grouped into a total of 29,679 clusters. An iterative blast workflow was used to annotate the R. philippinarum contigs with at least 100 bp (49,847 contigs out of 51,265). Then, BLASTx [88] with a cut off value of 10e-3, was used to compare the R. philippinarum contigs with the NCBI Swissprot, the NCBI Metazoan Refseq, the NCBI nr and the UniprotKB/Trembl protein databases. After annotation, Blast2GO software [89] was used to assign Gene Ontology terms [90] to the largest contig of a representative cluster (minimum of 100 bp). This strategy was used to avoid redundant results. Default values in Blast2GO were used to perform the analysis and ontology level 2 was selected to construct the level pie charts. To make a comparison between R. philippinarum and other bivalve species, the nucleotide sequences and ESTs from C. gigas, M. galloprovincialis, L. elliptica and B. azoricus were obtained from GenBank and from dedicated databases, when available. [93] . Unique sequences from these databases (based on gi number) were used from each of the databases. These sequences were compared by BLASTn against the longest contig from each of 29,679 R. philippinarum clusters with a cut off e-value of 10e-05. Hits to R. philippinarum sequences were represented in a Venn diagram. The comparison between our 454 results, the longest contig from each of 29,679 clusters, and the Milan et al. [16] transcriptome, contigs downloaded from RuphiBase (http:// compgen.bio.unipd.it/ruphibase/query/), was made by BLASTn with a cut off e-value of 10e-05. Another analysis was carried out to compare just the longest contig from each of 2,005 clusters identified as immune-related and the Milan et al. contigs as well. The results were summarized in a table ( Table 2 ). The percentage of coverage is the average % of query coverage by the best blast hit and the percentage of hits is the % of query with at least one hit in database, in parenthesis were added the total number of hits. Identification of immune-related genes All the contig annotations were revised based on an immunity and inflammation-related keyword list (i.e. apoptosis, bactericidal, C3, lectin, SOCS…) developed in our laboratory to select the candidate sequences putatively involved in immune response. The presence or absence of these words in the BLASTx hit descriptions was checked to identify putative immune-related contigs. The remaining non-selected contigs were revised using the GO terms at level 2, 3 and 4 assigned to each sequence after the annotation step that had a direct relationship with immunity. Selected contigs were checked again to eliminate non-immune ones and distributed into functional categories. Immune-related genes were grouped in three reference immune pathways (complement cascade, TLR signaling pathway and apoptosis) to describe each route indicated by our pyrosequencing results. To identify and classify the groups of organisms that had high similarity with our clam sequences, the Uniprot Taxonomy [94] was used except for the protozoa group. Because protozoa are a highly complex group, a specific taxonomy [95] was followed. Briefly, after the BLASTx annotation step all the hit descriptions included the species name (i.e. Homo sapiens) or a code (i.e. HUMAN) meaning that protein has been previously identified as belonging to that species. With such information sequences were classified in taxonomical groups and represented in pie charts. Table S1 List of contigs (e-value,10-3) of Ruditapes philippinarum including sequence, length, description (Hit description), accession number of description (Hit ACC), e-value obtained and database used for annotation (Blast).

Distribution and Risk Factors of 2009 Pandemic Influenza A (H1N1) in Mainland China

Data from all reported cases of 2009 pandemic influenza A (H1N1) were obtained from the China Information System for Disease Control and Prevention. The spatiotemporal distribution patterns of cases were characterized through spatial analysis. The impact of travel-related risk factors on invasion of the disease was analyzed using survival analysis, and climatic factors related to local transmission were identified using multilevel Poisson regression, both at the county level. The results showed that the epidemic spanned a large geographic area, with the most affected areas being in western China. Significant differences in incidence were found among age groups, with incidences peaking in school-age children. Overall, the epidemic spread from southeast to northwest. Proximity to airports and being intersected by national highways or freeways but not railways were variables associated with the presence of the disease in a county. Lower temperature and lower relative humidity were the climatic factors facilitating local transmission after correction for the effects of school summer vacation and public holidays, as well as population density and the density of medical facilities. These findings indicate that interventions focused on domestic travel, population density, and climatic factors could play a role in mitigating the public health impact of future influenza pandemics.

In early April 2009, human cases of infection with 2009 pandemic influenza A (H1N1) virus were first identified in the United States and Mexico (1) . The virus then spread rapidly to other regions of the world. As of January 24, 2010, laboratory-confirmed cases of pandemic influenza (H1N1-2009) were being reported in more than 209 countries or regions worldwide, with 14,711 deaths among confirmed cases (2) . Pandemic influenza was introduced to mainland China on May 9, 2009 (3, 4) , and then spread across the whole country. By the end of 2009, more than 120,000 confirmed cases were reported to the Chinese Center for Disease Control and Prevention (CCDC), including 648 deaths (5) . Information on reported cases was released daily by the Ministry of Health of the People's Republic of China in the early stages of the pandemic and then twice weekly later on. Analyzing the information gathered and unearthing underlying risk factors provide an opportunity to identify epidemic characteristics and transmission patterns of the pandemic in China, thereby producing useful information for prevention and control measures during future epidemics. In this study, we aimed to characterize the temporal and spatial distribution of pandemic influenza in mainland China, to understand the diffusion pattern of the disease within the country, and to identify risk factors for invasion and local transmission of this disease. We used a database that included all cases of pandemic influenza (H1N1-2009) reported to the China Information System for Disease Control and Prevention (CISDCP) from May 9, 2009 , when the first confirmed case in China was reported, to December 31, 2009 (6) . The CISDCP covers all provincial, prefectural, and county centers for disease control and prevention, 95.3% of the provincial, prefectural, and county hospitals (9,084 in total), and 84.0% of township clinics (38,175 in total) across mainland China. After the World Health Organization issued an alert about the novel influenza virus (H1N1-2009), pandemic influenza was classified as a class B notifiable infectious disease on April 30, 2009 , by the Ministry of Health but was managed according to the criteria for class A notifiable infectious diseases. According to the Law for Prevention and Control of Infectious Diseases in China, information regarding each patient, once identified, must be reported to the CCDC within 2 hours through the Web-based CISDCP system. A suspected case was defined as a person with influenza-like symptoms who had had close contact with a confirmed case within the past 7 days, had a history of travel to affected areas within the past 7 days, or tested positive for influenza A virus, excluding other known subtypes of influenza A. A laboratory-confirmed case was defined according to World Health Organization criteria-that is, a person with influenza-like symptoms and laboratory-confirmed pandemic influenza A virus infection by one or more of the following tests: reverse-transcriptase polymerase chain reaction, real-time reverse-transcriptase polymerase chain reaction, viral culture, or a 4-fold rise in specific antibodies to pandemic influenza A virus (7) . Influenzalike symptoms were defined according to World Health Organization criteria: sudden onset of fever greater than 38°C, cough or sore throat, and absence of other diagnoses (8) . All laboratory-confirmed cases from 2009 were included in our database, including information about age, sex, occupation, residence address, work address, onset date and location, hospital admission date and address, and clinical outcome. Furthermore, census information was obtained from the National Bureau of Statistics of China (9) . Epidemic curves were created by plotting the daily number of newly confirmed cases and the number of deaths. Mainland China is divided into 2,925 counties, which are political subdivisions of provinces, usually containing several townships. The incidences for different sex and age groups were calculated using 2009 census data. As was done previously (10, 11) , each case was linked to a digital map of China (1:100,000) according to its onset location. The incidence for each county was calculated and standardized using direct standardization for age and sex according to the overall composition of the 2009 Chinese population, using 5-year age-group categories. To explore the spatial and temporal diffusion trend of pandemic influenza in mainland China, a map of pandemic influenza spread was developed using trend surface analysis, which is a spatial smoothing method that uses polynomials with geographic coordinates, as defined by the central point of each county (12) (13) (14) . The time delay of the first confirmed case for each county was defined as the duration of time (in days) since May 9, 2009, the date of the first confirmed case in mainland China. A trend surface on these durations was created in ArcGIS 9.2 (ESRI Inc., Redlands, California) using a second-order trend surface model with a local polynomial method to explore the diffusion patterns of pandemic influenza over time. In addition, the epidemic curves were also plotted for 5 selected cities/provinces in different regions of mainland China: Guangdong (southern China), Shanghai (eastern China), Shaanxi (central China), Tibet (western China), and Beijing (northern China). To assess the association between the invasion of pandemic influenza and different means of domestic travel, we performed a survival analysis (Cox analysis) of the time delay of the first confirmed case for each affected county, considering unaffected counties as right-censored. The time delay of the first confirmed case was defined as the duration (in days) since May 9, 2009, the date of the first confirmed case in mainland China. Domestic travel was expressed using 4 variables: distance (from the midpoint of each county) to the nearest civil airport and whether or not a county was intersected by national highways, freeways, or railways. This information was obtained from the National Bureau of Statistics of China (9, 15) . Spatial analyses were used to extract data on these variables in ArcGIS 9.2 (ESRI Inc.). Since population density is also linked to human activities and may facilitate influenza transmission (16, 17) and since density of medical facilities could be linked to patient reporting, we adjusted for the effects of these variables. In this study, the population density for each county was obtained from the National Bureau of Statistics of China (9, 15) , and the densities of medical facilities were based on the CISDCP, including all reporting sectors for notifiable infectious diseases, comprising provincial, prefectural, and county centers for disease control and prevention and provincial, prefectural, and county hospitals and township clinics. Hazard ratios and their 95% confidence intervals and P values were estimated using maximum likelihood methods. Hazard ratios for the continuous variables were calculated for the following units: distance to the nearest airport (in 50-km increments), population density (in 1,000 persons per km 2 ), and density of medical facilities (in number of reporting sectors for pandemic influenza per 10,000 persons). To explore the effect of climatic factors on local transmission within counties, we performed multilevel Poisson regression. Climatic data (temperature, relative humidity, and precipitation) during May-December 2009 were obtained from the National Meteorological Bureau of China (18) . Owing to probable time lags, the climatic variables were processed by calculating the average value for the current day and a lag of 1-3 days, which is the observed incubation period of pandemic influenza (19) . Poisson regression deals with the daily number of laboratoryconfirmed cases per county. The inclusion of the population size for each county as an offset makes it an analysis of incidence. To account for possible confounding, we included school summer vacation and public holidays, the proportion of the school-age population (ages 6-19 years), population density, and the density of medical facilities as correction factors in the analysis. The percentage change in incidence in response to the change of the variable by a given amount (10°C for temperature, 10% for relative humidity, 1 mm for precipitation, 10% for school-age population, 1,000 persons per km 2 for population density, and number of facilities per 10,000 persons) was used to reflect the impact of each variable. The 95% confidence intervals and corresponding P values were estimated after correcting for overdispersion because of the nature of infectious diseases with spatial clustering patterns (20, 21) . For temperature, we also included a quadratic term in the analysis. For all analyses, univariate analysis was performed first to examine the effect of each variable separately. Multivariate analysis was then performed by including all variables with P values less than 0.20 in univariate analysis and exclusion of variables with P values greater than 0.10, using a standard backward likelihood ratio method. For all continuous variables, we also presented categorical results in 3-5 categories to allow inspection of the data and determine whether or not the assumption regarding continuous variables (quadratic for temperature) was justified. Statistical analyses were performed using the Stata package (StataCorp LP, College Station, Texas) (20) . Readers interested in further research can contact the corresponding author to obtain the full data set used in this study. A total of 121,805 cases of pandemic influenza (H1N1-2009), distributed in all 31 provinces in mainland China, were reported from May 9, 2009, to December 31, 2009. There was much variation in the numbers of confirmed cases in different provinces, ranging from 881 to 12,748, with a median of 2,958 cases, and in the incidence of confirmed cases in different provinces, ranging from 3.94 per 100,000 population to 71.72 per 100,000, with a median of 8.41 per 100,000. From the time profile, we found that the number of confirmed cases increased rapidly beginning at the end of August, when a new term began for school students, and peaked by the end of November. The first death caused by pandemic influenza was reported on October 4, 2009. The number of deaths eventually rose to 648 by the end of the year, and peaked in early December (Figure 1 ). The age-and sex-standardized incidence map shows that the epidemic spanned a large geographic area, and the most affected areas were in western China (see Web Figure 1 , which appears on the Journal's website (http://aje.oxfordjournals.org/)). Significant differences in incidence were found among age groups, with incidences peaking in school-age groups (Web Figure 1 ). Boys showed a higher incidence than girls (ages <20 years). Web Figure 2 shows the trend of the spatial spread of pandemic influenza over time in mainland China and indicates that the epidemic areas during the first 120 days after May 9, 2009, were limited to the circumferences of cities with international airports, such as Beijing, Shanghai, Guangzhou, Shenzhen, Chengdu, and Changchun. Thereafter, it spread to the rest of mainland China, roughly from southeast to northwest. The largest-scale spread took place 150-180 days after the first case (Web Figure 2) . Figure 2 shows the large variation in the temporal patterns of pandemic influenza for the 5 selected cities/provinces, but there was a marked drop in incidence during the first week of October for all locations. Survival analysis of the duration of time to the first confirmed case in each county indicated that all 4 factors related to domestic travel or human mobility were significantly associated with the invasion of pandemic influenza in the Cox univariate analysis (Table 1) . Population density and the density of medical facilities also showed a significant association. The significant effect of being intersected by railways disappeared, and the density of medical facilities showed borderline significance after correction for other factors in multivariate analysis, whereas being intersected by national highways and freeways and proximity to airports and higher population den-sity remained as significant factors, all showing a positive association (Table 1) . Table 2 shows that all climatic factors (except precipitation), school summer vacation and public holidays, proportion of the school-age population, population density, and the density of medical facilities were significantly associated with the extent of local transmission in univariate multilevel Poisson regression. School summer vacation and public holidays showed a significant negative association with the incidence of pandemic influenza. The significant effect of the proportion of schoolage children disappeared after correction for other factors; thus, temperature, relative humidity, school summer vacation and public holidays, population density, and the density of medical facilities remained as significant factors in multivariate analysis. Temperature showed a peak pattern, with the highest incidences for the range from 0°C to 10°C, which was also reflected in the statistically significant quadratic term. Our study provides a complete overview of the spatial and temporal characteristics of the pandemic influenza (H1N1-2009) epidemic in mainland China in 2009. The epidemic spanned a large geographic area and presented spatial and temporal heterogeneity in different regions of mainland China. Our analyses of the invasion of pandemic influenza indicated that domestic travel by air and by national highways and freeways and population density contributed to the spread of the epidemic. Lower temperatures and lower relative humidity were climatic factors that facilitated local transmission after correction for the effects of school summer vacation and public holidays, as well as population density and the density of medical facilities. The density of medical facilities could have influenced pandemic influenza patient reporting, and this effect seemed more important for the reporting of local transmission (a highly significant positive association) than for reporting of the invasion (borderline significance). This indicates that the CISDCP can be further improved. In the initial phase of the epidemic, the Chinese government took measures to prevent and control the spread of the novel influenza virus, declaring it a notifiable infectious disease in order to strengthen national surveillance and find newly confirmed cases quickly. In addition, quarantine measures were implemented at the international airports (e.g., Beijing, Shanghai, Guangzhou, and Fuzhou) to identify and isolate prob-able cases and close contacts in order to decrease the risk of local transmission caused by imported cases. It is possible that these measures were effective in achieving a slower pace of spread in the early stages of the epidemic, but the current evidence is inconclusive (22) . However, following the development of the global pandemic and the beginning of the new school term, a rapid spread of the epidemic occurred in mainland China, and local outbreaks increased at the end of August 2009. A reduction in incidence was observed during the first week of October, when there was an 8-day public holiday during the National Days from October 1 to October 8. This drop in incidence was largely due to a lower tendency for patients to visit medical facilities at that time, together with the fact that many hospitals had reduced the open hours of their outpatient clinics during the holidays. This is clearly visible in Figure 1 , where we see the drop beginning on September 28, 3 days (i.e., the average duration between onset and seeing a physician) before the start of the holiday period. Thus, the holiday period led to a reduction in the number of people being diagnosed with pandemic influenza. Apparently, most undiagnosed patients recovered in the following days, because there was no marked compensation visible in the days after the holiday period. Additionally, there Abbreviations: CI, confidence interval; NS, not significant. a Results were adjusted for school summer vacation and public holidays, population density, and the density of medical facilities. b For all continuous variables, categorical results are also reported to allow inspection of the data and assessment of whether or not the assumption regarding continuous variables was justified. may have been some reduced transmission because of school closure, as was observed in Japan, where transmission was substantially reduced during school closure (23) . In addition, the temporal death curve could reflect the massive rise in confirmed cases with approximately 1 week's delay following the peak of confirmed cases at the end of November. The direction of the spread of pandemic influenza was from the southeast to the northwest, indicating how the virus benefited from entering the country through international airports in the coastal areas and spreading further along routes of long-distance domestic travel. With the fast-growing public transportation infrastructure and increasing socioeconomic activities, travel has become an important issue in the prevention of emerging airborne infectious diseases such as influenza, especially during the introduction period. Obviously, the presence of airports and high densities of transportation routes coincide with more developed areas (i.e., those with a higher population density and more medical facilities); however, after we corrected for these factors, proximity to airports and the presence of national highways or freeways remained significantly associated with the spread of the infection. Geoinformatics plays an important role in the study and control of infectious disease outbreaks, and it includes techniques such as geographic mapping and location-based alert services (16, (24) (25) (26) . As was recognized previously, the international spread of pandemic influenza and severe acute respiratory syndrome was largely related to air travel (27) (28) (29) . Our study confirms that air travel and transportation routes accelerated the spread of pandemic influenza between counties in mainland China. Air travel and travel by national highways and freeways especially appeared to play a role, whereas railways were less important. In mainland China, trains are mainly used for occasional long-distance travel, whereas highways are more often used for daily or weekly commuting, especially because bus schedules are more flexible. Our previous study on the geographic spread of the severe acute respiratory syndrome epidemic in China also demonstrated that domestic travel along national highways played a more important role than travel by railway (30) . Data from the National Bureau of Statistics of China show that passenger traffic by highways in 2009 was 18.2 times that of railway travel (9) . Transportation by highway remains an important mode of travel between Chinese cities and provinces and therefore is a potential target for controlling any future emerging airborne infections. We also showed that lower temperature and lower relative humidity create a higher risk of local transmission of pandemic influenza. However, much lower temperature (e.g., <0°C) did not facilitate local transmission, as indicated by the daily incidence over categorical temperature in Table 2 , and required inclusion of a quadratic variable in the model. The observation of an influence of temperature and relative humidity on pandemic influenza is in accordance with animal experiments on seasonal influenza virus (31) . In addition, recent studies have suggested that absolute humidity could also play an important role in the transmission of pandemic influenza and seasonal variations in influenza epidemics in temperate regions (32) (33) (34) . As expected, population density further facilitated both invasion and local transmission, whereas holiday periods reduced spread (16, 17, 35, 36) . We also used population size as a correction factor instead of population density, which led to similar results (not shown). In conclusion, this is the first complete documentation of pandemic influenza in mainland China, to our knowledge. The findings indicate that interventions focused on domestic travel, population density, and climatic factors could play a major role in mitigating the public health impact of future influenza pandemics.

Replication, Neurotropism, and Pathogenicity of Avian Paramyxovirus Serotypes 1–9 in Chickens and Ducks

Avian paramyxovirus (APMV) serotypes 1–9 have been isolated from many different avian species. APMV-1 (Newcastle disease virus) is the only well-characterized serotype, because of the high morbidity, mortality, and economic loss caused by highly virulent strains. Very little is known about the pathogenesis, replication, virulence, and tropism of the other APMV serotypes. Here, this was evaluated for prototypes strains of APMV serotypes 2–9 in cell culture and in chickens and ducks. In cell culture, only APMV-1, -3 and -5 induced syncytium formation. In chicken DF1 cells, APMV-3 replicated with an efficiency approaching that of APMV-1, while APMV-2 and -5 replicated to lower, intermediate titers and the others were much lower. Mean death time (MDT) assay in chicken eggs and intracerebral pathogenicity index (ICPI) test in 1-day-old SPF chicks demonstrated that APMV types 2–9 were avirulent. Evaluation of replication in primary neuronal cells in vitro as well as in the brains of 1-day-old chicks showed that, among types 2–9, only APMV-3 was neurotropic, although this virus was not neurovirulent. Following intranasal infection of 1-day-old and 2-week-old chickens, replication of APMV types 2–9 was mostly restricted to the respiratory tract, although APMV-3 was neuroinvasive and neurotropic (but not neurovirulent) and also was found in the spleen. Experimental intranasal infection of 3-week-old mallard ducks with the APMVs did not produce any clinical signs (even for APMV-1) and exhibited restricted viral replication of the APMVs (including APMV-1) to the upper respiratory tract regardless of their isolation source, indicating avirulence of APMV types 1–9 in mallard ducks. The link between the presence of a furin cleavage site in the F protein, syncytium formation, systemic spread, and virulence that has been well-established with APMV-1 pathotypes was not evident with the other APMV serotypes.

The family Paramyxoviridae consists of enveloped viruses with a nonsegmented, single-stranded, negative-sense RNA genome [1] . These viruses have been isolated from a great variety of mammalian and avian species around the world. Many members of the family cause important human and animal diseases, while the disease potential of many other members is not known. The family is divided into two subfamilies: Paramyxovirinae and Pneumovirinae. The subfamily Paramyxovirinae comprises five genera: Rubulavirus, Respirovirus, Morbillivirus, Henipavirus, and Avulavirus. Subfamily Pneumovirinae is divided into two genera: Pneumovirus and Metapneumovirus. All paramyxoviruses isolated from avian species are classified into the genus Avulavirus, except avian metapneumoviruses, which are classified in the genus Metapneumovirus. Avian paramyxoviruses (APMVs) have been divided into nine different serotypes (APMV 1 to 9) based on Hemagglutination Inhibition (HI) and Neuraminidase Inhibition (NI) assays [2] . APMV-1 comprises all strains of Newcastle disease virus (NDV) and has been well characterized because of its economic importance in poultry industry. As an initial step towards characterizing other APMV serotypes, complete genome sequences of one or more representative strains of APMV serotypes 2 to 9 have been determined [3] [4] [5] [6] [7] [8] [9] [10] . The genomes of all paramyxoviruses range between 15 and 19 kb in length and contain 6-10 genes that encode up to 12 different proteins [11] . For members of subfamily Paramyxovirinae, efficient RNA replication requires that the nucleotide (nt) length of the genome is an even multiple of six, known as 'rule of six', reflecting the precise packaging of the polynucleotide in the nucleocapsid [12] . The genomes of APMVs are very similar in organization. All the members contain six genes in the order of 39-N-P-M-F-HN-L-59 with the exception of APMV-6, which contains an additional small hydrophobic protein (SH) gene in its genome. APMVs encode a nucleocapsid protein (N), a phosphoprotein (P), a matrix protein (M), a fusion protein (F), a hemagglutininneuraminidase protein (HN), and a large polymerase protein (L). Two additional proteins called, V and W, are produced by RNA editing of the P gene. The F and HN proteins form spike-like projections on the outer surface of the viral envelope and are the neutralizing and protective antigens of NDV. Significant sequence divergence in these two proteins exists among APMV serotypes [6] . The HN protein possesses receptor-binding and neuraminidase activity; whereas, the F protein is directly involved in membrane fusion which is necessary for the entry of the virus. Homotypic interactions between the HN and F proteins are hypothesized to control initiation of the fusion process for most paramyxoviruses [13, 14] . The M protein forms the inner layer of the envelope and plays a key role in assembly by interacting with the HN and F proteins as well as ribonucleocapsid [15, 16] . APMVs have been isolated from many different avian hosts [17] . APMV-1 is the only well-characterized serotype, because of the high morbidity, mortality, and economic loss caused by highly virulent strains. NDV isolates vary greatly in their pathogenicity for chickens, ranging from no apparent disease to severe respiratory and neurological disease causing 100% mortality [18] . NDV strains are categorized into three main pathotypes: lentogenic (avirulent), mesogenic (moderately virulent), and velogenic (virulent), based on their pathogenicity in chickens [19] . In contrast, the disease potential of APMV-2 to -9 is not well known because many of these viruses were isolated from birds dying in quarantine, hunter killed, trapped wild birds, apparently healthy poultry or exotic birds [20] . APMV-2 and -3 have been reported to cause significant disease in poultry, whereas the pathogenic potential of APMV-4 to -9 is generally unknown. In general, APMV-2 strains have been isolated from chickens, turkeys and wild birds across the globe and have been found to cause mild respiratory disease, decreases in egg production, and infertility [21] [22] [23] . APMV-3 strains have been isolated from wild and domestic birds and their infections have been associated with encephalitis and high mortality in caged birds [24] . APMV-5 strains have only been isolated from budgerigars (Melopsittacus undulatus) and cause depression, dyspnoea, diarrhea, torticollis, and acute fatal enteritis in immature budgerigars, leading to very high mortality [25] . Infections from APMV-4, -8, and -9 appear to be restricted to ducks and geese. APMV-6 and -7 infections in turkeys cause drops in egg production and induce respiratory disease. There are no reports of isolation of APMV-5, -8 and -9 from poultry [2] . But recent serosurveillance of commercial poultry farms in the U.S. indicated the possible prevalence of all APMV serotypes excluding APMV-5 in chickens [26] . The pathogenicity of APMV-2 and APMV-3 has been studied in experimentally infected chickens [27, 28] . However, replication, pathogenicity, and neurovirulence of APMV serotypes 2 through 9 have not been comprehensively studied. Therefore, we characterized in vitro replication of APMVs (growth kinetics and cytopatic effect in chicken fibroblast cells) and their in vivo replication and tropisms by infecting prototype strains of each serotype in two different ages of chickens (1-day-old and 2-week-old chickens) and 3-week-old ducks. Specifically, neurotropism of APMVs was evaluated in primary chicken neuronal cells and brain tissue of 1-day-old chicks. The chicken embryo fibroblast cell line (DF1, ATCC, Manassas, VA, USA) was grown in Dulbecco's minimal essential medium (DMEM) with 10% fetal bovine serum (FBS) and maintained in DMEM with 5% FBS. The African green monkey kidney Vero cell line (ATCC, Manassas, VA, USA) was grown in Eagle's minimum essential medium (EMEM) containing 10% FBS and maintained in EMEM with 5% FBS. Primary chicken neuronal cells were grown in Neurobasal medium with B-27 supplement (Invitrogen). The viruses used in this study were nine prototype strains of APMV serotypes 1 to 9: APMV-1 (NDV), lentogenic strain LaSota/46 and mesogenic strain Beaudette C (BC); APMV-2, APMV-2/Chicken/California/Yucaipa/56; APMV-3, APMV-3/ PKT/Netherland/449/75; APMV-4, APMV-4/duck/Hon-gKong/D3/75; APMV-5, APMV-5/budgerigar/Kunitachi/74; APMV-6, APMV-6/duck/HongKong/18/199/77; APMV-7, APMV-7/dove/Tennessee/4/75; APMV-8, APMV-8/goose/ Delaware/1053/76; and APMV-9, APMV-9/duck/New York/ 22/1978. All of the viruses were grown in 9-day-old embryonated specific-pathogen-free (SPF) chicken eggs inoculated by the allantoic route, except for APMV-5, which was grown in Vero cells. The ability of viruses to produce plaques was tested on Vero and DF1 cells under 0.8% methylcellulose overlay. Exogenous protease was supplemented into the cells for replication of APMV-1 LaSota, APMV-3 and -9 (10% allantoic fluid) and APMV-8 (1 mg/ml of acetyl trypsin) (4, 9, 10) . The monolayers were fixed with methanol and plaques were visualized by immunoperoxidase staining using virus specific antiserum raised against N protein. Virus titers for in vitro and in vivo replication were quantified by immunoperoxidase staining with N-specific antibodies on DF1 cells (APMV-1, -2, -3, -4, -6, and -9) or Vero cells (APMV-5, -7, and -8) [3-10]. The multicycle growth kinetics of the viruses was evaluated in DF1 cells. Duplicate wells of six-well plates were infected with each APMV at an MOI of 0.01 PFU/cell. After 1 h of adsorption, the cells were washed and then covered with DMEM containing 2% FBS at 37uC in 5% CO 2 . APMV-1 LaSota, APMV-3, -8, and -9 were supplemented with protease, as described above. Supernatants were collected and replaced with an equal volume of fresh medium at 12-h intervals until 72 h post-infection (hpi). Virus titers in the collected supernatants were quantified in DF1 cells or Vero cells by limiting dilution and immunostaining with N-specific antibodies [29] [30] . Virus titers were expressed as 50% tissue culture infectious dose (TCID 50 /ml) by the end-point method of Reed and Muench [31] . The pathogenicity of APMVs was determined by the mean death time (MDT) test in 9-day-old SPF embryonated chicken eggs and by the intracerebral pathogenicity index (ICPI) test in 1-day-old SPF chicks [19] . Briefly, for the MDT, a series of 10fold dilutions of infected allantoic fluid (0.1 ml) was inoculated into the allantoic cavities of five 9-day-old eggs per dilution and incubated at 37uC. The eggs were examined once every 8 h for 7 days, and the time of embryo death was recorded. The MDT was determined as mean time (h) for the minimum lethal dose of virus to kill all the inoculated embryos. The criteria for classifying the virulence of NDV isolates are: ,60 h, virulent strains; 60 to 90 h, intermediate virulent strains; and .90 h, avirulent strains. For the ICPI test, 0.05 ml of a 1:10 dilution of fresh infective allantoic fluid for each virus was inoculated into group of 10 1-day-old SPF chicks via the intracerebral route. The birds were observed for clinical symptoms and mortality once every 8 h for a period of 10 days. At each observation, the birds were scored as follows: 0 if normal; 1 if sick; and 2 if dead. The ICPI is the mean of the score per bird per observation over the 8-day period. Highly virulent velogenic viruses give values approaching 2, and avirulent or lentogenic strains give values at or close to 0. All experiments involving experimental animals were approved by the committee of IACUC, University of Maryland (protocol number R-09-81) and conducted following the guidelines. To evaluate neurotropism of APMVs in vitro, primary chicken neuronal cells were prepared from 9-day-old chicken embryos for virus infection. Briefly, the section of hippocampus was dissected from embryos, digested with trypsin, passed through a cell strainer (40 mm nylon), seeded onto poly-L-lysine-coated plates, and then infected with each virus [32] . Spread of viruses in neuronal cells was determined by confocal microscopy analysis. Briefly, at 48 hpi, the cells were fixed, permeabilized, stained with polyclonal antibodies against the respective N protein and a neuronal marker (anti-neuron specific beta III tubulin antibody, abcamH, Cambridge, MA) followed by anti-Alexa Fluor 488 and 594 antibodies, and then analyzed by confocal microscopy. In addition, virus replication in neuronal cells was determined by collecting supernatants at 12-h intervals until 72 hpi. Virus titers in the collected supernatants were quantified in DF1 or Vero cells by limiting dilution and immunostaining with N-specific antibodies as described above. To evaluate the ability of APMVs to replicate in chicken brains, ten 1-day-old SPF chicks were inoculated with 0.05 ml of a 1:10 dilution of fresh infective allantoic fluid for each virus via the intracerebral route. Two birds were sacrificed daily until 5 days post-infection (dpi). Brain tissue samples were collected from the sacrificed birds and processed for immunohistochemistry as described later. Brain tissue samples also were homogenized and virus titers were determined by limiting dilution and immunoperoxidase assay in DF1 cells or Vero cells using polyclonal antibodies against the respective N protein. 2.5. Replication of APMVs in 1-day-old Chicks, 2-week-old Chickens, and 3-week-old Ducks Following Intranasal Inoculation Three 1-day-old chicks per group were inoculated with 100 ml of each virus (256 HA units/bird) via the intranasal route. On day 3 post-infection, tissue samples (lung, trachea, spleen, and brain) were collected for virus titration by limiting dilution and immunoperoxidase assay as described above. To evaluate the tropism of APMVs in older birds, groups of ten 2week-old SPF chickens were inoculated with 200 ml of each virus (256 HA units/bird) by the intranasal route. Three birds from each group were sacrificed at 4 dpi and tissues samples (lung, trachea, spleen, and brain) were collected and processed for immunohistochemistry as described later. Tissue samples also were homogenized for virus titration. Virus titers in DF1 or Vero cells were determined by limiting dilution as described above. To confirm the replication of viruses in infected birds, the tissue samples were also inoculated into 9-day-old embryonated chicken eggs. On 3 dpi, virus growth was determined by HA assay. The remaining birds were observed daily for 10 days for any clinical signs and then sacrificed for virus titration of various tissues as described above. To evaluate virus replication in different hosts, we further determined replication of APMVs in 3-week-old mallard ducks. Six birds each were infected with 500 ml of individual APMVs (256 HA units/bird) via the combined intranasal and intratracheal routes and sacrificed at 4 dpi for collection of tissue samples (lung, trachea, spleen, and brain). Homogenates were prepared for virus titration in cell cultures and replication in eggs as described above. The remaining birds were observed daily for 10 days for any clinical signs. From the experiments described above, brain tissue harvested 3 dpi from 1-day-old chicks infected by intracerebral route and various tissue samples harvested 4 dpi from 2-week-old chickens infected by the intranasal route were fixed in phosphate-buffered formalin (10%). Fixed tissues were embedded in paraffin and sectioned (Histoserv, Inc., Germantown, MD). Sections from mock-infected birds were used as controls. The tissues were deparaffinized, rehydrated, and subsequently, immunostained to detect viral N protein using the following protocol. Briefly, the sections were blocked with 1% BSA in PBS for 1 h at room temperature, incubated with a polyclonal antibody (1:200 dilution) against the respective N protein (29, 30) followed by horseradish peroxidase-conjugated goat anti-rabbit antibodies for 30 min, and then stained with AEC (3-amino-9-ethylcarbazole) substratechromogen. Syncytium formation is a hallmark of the cytopathic effect (CPE) caused by many paramyxoviruses, including APMV-1, in cell culture [1] . To investigate syncytium formation by APMV serotypes 2-9 and to compare their CPE, Vero cells were infected with mesogenic APMV-1 strain BC or with representatives of the other APMV serotypes at a multiplicity of infection (MOI) of 0.1 PFU/cell, incubated for 48 h, and visualized directly by photomicroscopy (data not shown) and following immunostaining with rabbit antiserum to the respective N protein (Fig. 1 ). Exogenous protease was supplemented into the culture medium for replication of APMV-3, 8, and 9, based on our previous studies [4, 9, 10] . APMV-1 strain BC was known to be independent of protease supplementation. APMV-3 and -5 produced distinctive CPE with syncytium formation similar to those of APMV-1. In contrast, the rest of APMVs produced single cell infections leading to cell rounding and detachment of infected cells but a lack of evident syncytia. Similar results were observed in chicken embryo fibroblast DF1 cells (data not shown). We further evaluated the multicycle replication of the APMVs in DF1 cells (Fig. 2 ). There was a great difference in the kinetics and magnitude of replication among different APMVs. APMV-1 strain BC and APMV-1 lentogenic strain LaSota grew better than any other APMVs and reached the highest titers (10 8 TCID 50 /ml) at 32 and 60 hpi, respectively. APMV-3 replicated relatively well, reaching 10 7 TCID 50 /ml at 72 hpi. The highest titers of APMV-2 and -5 also were detected at 72 hpi, but their titers were 4.0 log 10 lower than those of APMV-1. In general, replication of other APMV viruses was limited in DF1 cells (,10 2 TCID 50 /ml). Inefficient virus replication in vitro (i.e., APMV-4, 6, 7, 8, and -9) was correlated with the pattern of a single-cell infection as observed with CPE in infected cells (Fig. 1 ). The pathogenicity of APMVs was evaluated with the standard pathogenicity tests used for APMV-1, namely MDT in embryonated chicken eggs and ICPI in 1-day-old chicks ( Table 1 ). The pathogenicity of the two strains of APMV-1 has been well characterized: strain BC is neurovirulent (MDT 58 h and ICPI 1.55), whereas strain LaSota is avirulent (MDT 112 h and ICPI 0.00) and is used as a vaccine strain. For the most part, APMV 2-9 were even more avirulent based on these assays than strain LaSota, regardless of their isolation sources. The one exception was APMV-3, which had a MDT value (117 h) that is similar to that of the LaSota strain, but had an ICPI value (0.53) that was higher than that of the LaSota strain (0.00), but nonetheless remained in the avirulent range. The MDT and ICPI values of other APMVs were .144 h and 0.00, respectively, consistent with being avirulent. Chicks infected with APMV serotypes 2-9 had no apparent clinical signs during the 8-day period of the ICPI test. To evaluate neurotropism of APMVs, virus replication was evaluated in primary chicken neuronal cells in vitro. We first evaluated infection (MOI of 0.1 PFU) in neuronal cells by confocal microscopy analysis of cells that had been immunostained at 72 hpi with N protein against each respective APMV (Fig. 3A) . Expression of the N protein of APMV-1 strain BC was clearly detected in dendrites and axons. In contrast, APMV-1 strain LaSota failed to replicate in neuronal cells even after supplementation with allantoic fluid as a source of protease. Among serotypes 2-9, APMV-3 was able to replicate well without the addition of allantoic fluid, whereas the presence of APMV-7 and -8 antigens in infected cells was sporadically detected. Replication of the other APMVs was not detected up to 3 dpi. We also examined replication in the neuronal cell cultures by collecting supernatants from the cultures at 12 h intervals and assaying for infectious virus by titration. This confirmed the ability of APMV-3 to replicate in neuronal cells, although its titer was lower than that of neurovirulent APMV-1 strain BC (Fig. 3B) . Titers of BC and APMV-3 in neuronal cells increased gradually and reached 5.5 and 3 log TCID 50 /ml, respectively, at 72 hpi. In contrast, other APMVs, including APMV-1 strain LaSota, were not detected in the supernatants of neuronal cells, indicating their inability to replicate in these cultures of primary chicken neuronal cells. We further determined the ability of the APMV serotypes to replicate in chicken brains by inoculating 1-day-old chicks via the intracerebral route. The infected chickens were sacrificed daily and brain tissue was collected for virus titration (Fig. 4) . In chicks infected with neurovirulent APMV-1 strain BC, the virus reached the highest titer (.6.0 log 10 TCID 50 /g) in the brain at 2 dpi, resulting in the death of all of the infected chicks on 3 dpi. The non-neurovirulent LaSota strain was detected at low titer on day 1, but was not detected on subsequent days and did not cause disease or death during the 5 days of observation. APMV-3 strain reached a titer of 5 log 10 TCID 50 /g on 3 dpi, but did not cause disease or death of any of the infected chicks during the 5 days of observation. We detected a low level of replication of APMV-8 (,2 log 10 TCID 50 /g), with no observed disease or death. Replication of the other APMVs was not detectable on any day, and there was no detectable disease or death. Thus, APMV-3 was identified as the only neurotropic virus among APMV serotypes 2-9. However, despite the ability of APMV-3 to replicate to moderate titer in the brain, it did not cause discernable disease or death and thus was non-neurovirulent. The neurotropism of the APMVs in 1-day old chicks also was evaluated by immunohistochemistry analysis of brain tissue harvested on day 3 (Fig. 5) . Viral antigens were detected in the brain tissues that had been infected with APMV-1 strain BC and APMV-3, confirming the replication of the two viruses in chicken brain. Our result showed a more extensive distribution of viral antigens in brain tissue infected with BC compared to APMV-3. In contrast, presence of viral antigens was not detected in the brain tissues that had been infected with the other APMVs (Fig. 5) . We next determined the replication and tissue tropism of APMVs in 1-day-old and 2-week-old chickens following intranasal inoculation. Chickens were sacrificed on 3 dpi for 1-day-old chicks and on 4 dpi for 2-week-old chickens, and the following tissues were harvested for quantitative virology: trachea, lungs, spleen, and brain (Fig. 6) . In 1-day-old chicks, APMV-1 strain BC replicated to high titers (.5.0 log 10 TCID 50 /g) in each of the sampled tissues in each of the birds, whereas LaSota replicated in trachea, lung and spleen, but not the brain, and its titers were less than those of BC (Fig. 6A) . APMV-3 replicated to moderate titers (.4.0 log 10 TCID 50 /g) in all of the collected samples, including the brain. However, replication of other APMVs was restricted to the trachea, and their titers were low, ranging from 1.5 to 3 log 10 TCID 50 /g (data for APMV-2 are shown in Fig. 6A ; the other serotypes are not shown). Chicks infected with APMV-1 strain BC began to show clinical signs on 1 dpi and distinctive neurological signs on 2 dpi, whereas chicks infected with other APMVs had no apparent clinical signs during the 3-day post-infection. In 2-week-old chickens, virus replication was more restricted and viral titers were lower compared to 1-day-old chickens (Fig. 6B) . APMV-1 strain BC was able to replicate in all of the collected tissues, but the virus titers in the brains of older birds (2.6 log 10 TCID 50 /g) were much lower than that of 1-day-old chickens (6.5 log 10 TCID 50 /g). APMV-1 strain LaSota replicated in the trachea, lungs, and spleen, but not the brain, as was observed in the 1-day-old chickens. Interestingly, APMV-3 replication was detected only in the trachea and the brain, indicating that it is neutrotropic in these older chickens despite this restricted replication. APMV-2 was also able to replicate in trachea of 2week-old chickens. However, replication of the other APMVs was not detected in any of the harvested tissue samples. Since APMV 2-9 can replicate much better in embryonated chicken eggs than in cell culture [20] , the collected tissue homogenates from this same experiment also were inoculated into 9-day-old embryonated chicken eggs to confirm the replication of viruses in various tissues ( Table 2 ; APMV-5 was not assayed because it does not replicate in the allantoic cavity of chicken eggs). This approach with increased sensitivity enabled us to detect replication of all of the tested APMVs in the trachea of most of the birds. In addition, APMV-2, 3, 6, 7, and 9 were detected in the lungs of at least one of the three chickens in each group, and APMV-3 was detected in the spleen of two chickens. Among the tested APMVs, APMV-4 showed the least replication in chickens both in vitro and in vivo ( Fig. 2 and Table 2 ), suggesting that it has a strong host range restriction in chickens. On 10 dpi, none of the APMVs were detected in any of the sampled tissues from any of the birds (not shown). Clinical signs of illnesses in any of the infected groups were not found up to 10 dpi. Histopathological examinations of tissue samples collected on 4 dpi revealed similar microscopic findings in all APMVs. Specifically, the trachea showed mild lymphocytic tracheitis, mild to moderate multifocal mucosal attenuation, and loss of tracheal alveolar mucous glands (Fig. 7A) . Lung sections exhibit moderate, multifocal, lymphocytic to lymphohistiocytic bronchitis with mild to moderate perivascular and peribronchial interstitial inflammation and focal perivascular cuffing with varying severity (Fig. 7B) . Minimal lymphoid depletion was detected in the spleen, and microscopic lesions were not found in any of the brain tissues (data not shown). The presence of virus antigens in various tissues of infected chickens was further evaluated by immunohistochemistry analysis. Deparaffinized sections of the virus-infected and uninfected control tissue were immunostained using polyclonal antibodies against the N protein of the respective APMV. The presence of antigens for most APMVs was detected in the epithelial lining of trachea (Fig. 7C) . However, no antigen was detected in any tissues harvested from chickens infected with APMV-5. In the lungs, the viral antigens were mostly localized in the epithelium surrounding the medium and small bronchi (Fig. 7D) . Lung tissues showed extensive presence of antigens for APMV-1 strain BC followed by APMV-1 strain LaSota, APMV-3, and APMV-2. However, presence of antigens for other serotypes was not detected in lung tissues. Our results suggested that APMV 2-9 were avirulent, and their replication was mostly restricted to trachea and lungs of infected chickens. Since several of the specific APMV strains under evaluation (i.e., APMV-4, -6, and -9) were isolated from other avian species including the mallard duck (see Table 1 ), their replication and pathogenicity were further evaluated in groups of 3-week-old ducks following intranasal inoculation. No clinical signs of illnesses in any of the infected groups were found up to 10 dpi. Three birds from each group were sacrificed on 4 dpi and the tissues were harvested for virus titration. None of the APMVs were detected in any of the tested tissues from any of the ducks by virus titration in DF1 cells (data not shown). However, APMVs were detected in some of the samples of trachea, lung, and spleens by inoculation into 9-day-old embryonated chicken eggs, although none of the APMVs (including APMV-1) were detected in any of the brain samples (Table 3 ; APMV-5 was not evaluated). APMV-1 strains BC and LaSota were the only APMVs that were detected in all three ducks in their respective groups, although replication was mostly restricted to the trachea. APMV-2 was found only in the trachea of a single duck, in contrast to its more efficient replication in chickens. APMV-3 was not detected in any samples from any duck, in contrast to its efficient replication and neurotropism in chickens. APMV-8 and -9 also were not detected in any duck. The other serotypes (APMV-4, -6, and -7) were each detected in the trachea of two ducks from its group, and in the lungs of one or two ducks from its group. APMV-4 replicated slightly better in ducks than in chickens. APMV-7 was the only serotype detected in the spleen (of a single duck), whereas it had not been detected in the spleen of infected chickens. Thus, the pattern of infection in chickens and ducks was substantially different and could not be predicted by the isolation history. APMVs are frequently isolated from a wide variety of avian species and are grouped into nine serotypes based on antigenic analysis [33] , with a likely tenth serotype recently described in penguins [34] that was not evaluated in the present study. APMV-1 (NDV) is the most extensively characterized member of the APMV serotypes. APMV-2 to -9 have been isolated from both wild and domestic birds, but their disease potential in wild or domestic birds was largely unknown. APMV-1 also has been shown to infect a number of non-avian species [35] and presently is being evaluated as a potential human vaccine vector for human pathogens [36] . There is a possibility that APMV-2 to -9 could also be used as human vaccine vectors for human pathogens, providing multiple vectors with minimal cross-restriction from vector-specific immunity. Our previous studies demonstrated that APMV-2 to -9 can replicate in mammalian hosts, specifically in the hamster and mouse models [29, 30] . However, their replication and pathogenicity in avian host has not been comprehensively Chicks were inoculated with each virus by the intracerebral route, and brain tissue was harvested for immunohistopathology on 3 dpi. The tissues were fixed in phosphate-buffered formalin, and sections were prepared and stained using an antibody against the respective N protein (stained red). doi:10.1371/journal.pone.0034927.g005 evaluated and compared. Therefore, in this study, we evaluated in vitro and in vivo replication and pathogenicity of APMVs in chickens and their in vivo replication and pathogenicity in ducks and characterized differences in their tropisms, including neurotropism in chickens. Virulent APMV-1 strains contain a multibasic cleavage site (with the general consensus sequence of RRQKRQF) with a polybasic furin motif (RX[R/K]RQ) that is readily cleaved in cell culture by intracellular furin or furin-like protease. Avirulent APMV-1 strains lack this polybasic site and depend on extracellular protease for cleavage, which in vitro can be supplied by exogenous protease and in vivo is supplied by secretory trypsinlike protease found in the lumen of the respiratory and enteric tracts [20] . Our study showed that APMV-1 strain BC readily forms syncytia in cell culture and replicates more efficiently in vitro than any of the other APMVs, followed by the APMV-1 LaSota strain. Characterization of the CPE of APMV types 2-9 demonstrated that only APMV-3 and -5 induced syncytium formation in infected cells. The cleavage site of APMV-3 (RPRGRQL) lacks the furin motif, and efficient growth and syncytium formation depended on supplementation with exogenous protease, resembling an avirulent APMV-1 strain. The cleavage site of APMV-5 (KRKKRQF) is the only example from APMV2-9 to have a furin motif [6] , and syncytium formation and growth were independent of added protease, resembling a virulent APMV-1 strain. Production of syncytia by APMV-3 and APMV-5 correlated with efficient multicycle replication in cell culture in the case of APMV-3 and a moderate level of replication in cell culture in the case of APMV-5. Although APMV-5 contains the greatest number of basic residues for any of the APMV except for some strains of APMV-1, its replication in DF1 cells was much less efficient compared to APMV-1 and -3. Interestingly, the level of replication of APMV-5 in cell culture was very similar to that of APMV-2, which lacks a furin motif (KPASRQF), does not form Figure 6 . Replication of APMVs in 1-day-and 2-week-old chickens. Groups of (A) 1-day-or (B) 2-week-old chickens were inoculated with each virus (256 HA units) by the intranasal route. Three birds from each group were sacrificed on 3 dpi (1-day-old chicks) or 4 dpi (2-week-old chickens), and virus titers in the collected tissues samples (brain, trachea, lung, and spleen) were determined by limiting dilution in DF1 cells and immunostaining with polyclonal antibodies raised against the respective N protein. doi:10.1371/journal.pone.0034927.g006 syncytia, and whose replication and lack of ability to form syncytia are independent of exogenous protease. Thus, for those two viruses (APMV-2 and -5), the growth phenotype in cell culture did not correlate well with the presence or absence of a furin motif in the cleavage sequence. The other APMVs (types 4, 6, 7, 8, and 9) have one or two basic amino acids in their cleavage site sequences and exhibited single cell infection and inefficient replication in vitro. However, only in the case of serotypes 8 and 9 did the addition of exogenous protease increase the efficiency of replication, and the overall level of replication remained very low and syncytia were not observed. Thus, APMV types 2, 4, 6, 7, 8, and 9 have cleavage sites that seemed similar to those of avirulent APMV-1 strains, but the inclusion of protease had little effect on growth in cell culture. The pathogenicity of APMVs was first characterized by standard pathogenicity tests, namely the ICPI and MDT assays. We included two well-characterized strains of APMV-1, mesogenic BC and lentogenic LaSota, and compared their pathogenicity with that of APMV serotypes 2-9. This showed that APMV serotypes 2-9 were avirulent by both assays. Of types 2-9, only APMV-3 was associated with embryo death in the MTD assay and disease in the ICPI assay, although these effects were modest and APMV-3 was categorized as avirulent by the standards of these assays. By the MDT assay, the other types (types 2, 4, 5, 6, 7, 8, and 9) lacked detectable virulence and thus were less virulent than the well-known avirulent vaccine strain APMV-1 LaSota, and by the ICPI assay these APMVs were equivalent to LaSota in lacking detectable virulence. Thus, possession of a furin cleavage site and the ability to form syncytia by APMV-5 did not correlate with virulence in vivo, nor did the ability of APMV-2 to replicate to moderate titers in cell culture without the addition of trypsin. We investigated the neurotropism of APMVs in chicken neuronal cells in vitro and in the brains of 1-day-old chicks infected via the intracerebral route. The neurovirulent APMV-1 strain BC was observed to replicate in chicken neuronal cells both on the basis of antigen expression and the production of infectious virus. In contrast, the non-neurovirulent LaSota strain did not detectably express viral antigens in these cells, and infectious virus was detected in the culture only at a low level on a single day early in infection. Among serotypes 2-9, expression of viral antigen in the neuronal cells was detected only with APMV-3, -7, and -8, and production of infectious virus was detected only with APMV-3. Thus, only APMV-1 strain BC and APMV-3 appeared to replicate productively in these neuronal cell cultures. Following intracerebral inoculation of 1-day-old chicks, only APMV-1 strain BC, APMV-3, and (to a much lesser extent) APMV-8 could be detected in brain tissue samples. With APMV-1, virus neurotropism depends on the presence of a furin cleavage site, since secretory proteases are unavailable in the brain [37] . The lack of neurotropism of the APMV-1 strain LaSota is consistent with this idea. However, APMV-3 appears to be an exception, since it lacks a furin site and is dependent on added protease in cell culture and thus presumably dependent on secreted protease in vivo, yet it replicated efficiently in the chick brains and neutronal cells in vitro. Interestingly, however, while APMV-3 was clearly neurotropic, it caused no disease or death and thus was not neurovirulent. These findings indicate that the simple paradigm that neurotropism depends on the presence of a furin cleavage site does not hold for APMV-3. Furthermore, it shows that neurotropism by an APMV does not necessarily confer neurovirulence. Our previous study with replacement of the APMV-2 F protein with that of APMV-1 BC suggested an important role for the BC F protein in virus neurotropism, neuroinvasiveness, and neurovirulence in 1-day-old chicks [38] . Based on the generally held model for APMV-1 pathogenesis, it would have been reasonable to suggest that the presence of the furin site in the BC F protein was the major determinant of the observed neurotropism, neuroinvasiveness, and neurovirulence. However, the present study challenges the predominant role for the cleavage site, and suggests that other features of the F protein may be involved in the observed neurotropism, neuroinvasiveness, and neurovirulence. Evaluating the difference in the roles of the F proteins in virus neurotropism versus neurovirulence may provide information on these activities and may help design safer vaccines for neurotropic pathogens. We also evaluated replication, tropism, and pathogenesis following intranasal infection in 1-day-old chicks and 2-week-old chickens. In 1-day-old chickens, only the APMV-1 strain BC and APMV-3 were able to spread to the brain and replicate, demonstrating neuroinvasiveness and neurotropism. These two viruses, plus APMV-1 strain LaSota, also were detected in the spleen of 1-day-old chicks. Conversely, replication of the other APMVs was mostly detected the in trachea of chicks, and the virus titers were moderate. In 2-week-old birds, at 4 dpi, infection of APMV-1 strain BC was found in all of the tested tissues, including the brain, and LaSota was detected in all of the tested tissues except the brain. APMV-3 also was detected in all of the tested tissues including the brain. Replication of the other APMVs was restricted to the trachea and lungs. In general, virus titers in the 2-week-old chickens were lower than in the 1-day-old chicks. By 10 dpi, no virus could be detected in any of the tissues in any of the infected chickens for any serotype by virus isolation, suggesting that the virus was cleared from all tissues and disease was resolved, indicating the self-limited nature of the infections. Histopathology analysis of APMV-infected 2-week-old chickens showed that infection by each of the APMVs, including APMV-5, produced tracheitis and mild pathology that was mainly restricted to the respiratory tract. Previous study with serologic assays in chickens infected with APMVs demonstrated a good humoral response on 14 dpi [39] . Thus, the birds indeed appeared to be infected even thought the recovery of infectious virus was often sporadic and low. A single APMV type, namely APMV-5, was not detected by virus isolation or by immunohistochemistry in any of the chickens in this study. However, the serologic assay of chickens that had been inoculated with APMV-5 also showed the development of virus specific antibodies detected by virus plaque reduction neutralization assay [39] , suggesting that infection had occurred. This suggests a low level of virus replication in chickens. Thus, although APMV-5 bears a furin cleavage site and can cause 100% mortality in budgerigars [25] , it was completely avirulent and replicated inefficiently in chickens. This is suggestive of a host range difference that was not greatly ameliorated by the presence of a multi-basic cleavage site. Experimental infection of mallard ducks with APMVs indicated that all APMV serotypes, including duck isolates, are avirulent in ducks. Differences were observed in the replication of various APMVs between chickens and ducks. For example, replication of APMV-1 strain BC was low in ducks compared to its efficient replication chickens, although APMV-1 has shown to infect waterfowls, such as geese and Muscovy and Pekin ducks [40] [41] [42] . Several APMV serotypes (APMV-3, -8, and -9) did not replicate in ducks, even though APMV-3 replicated well in chickens and the APMV-9 strain had been isolated from ducks. In chickens, APMV-4 poorly replicated in both 1-day-old and 2-week-old chickens in our study. In addition, the virus has been shown to induce low HI titers in chickens compared to other APMVs, which was taken as evidence of a low level of replication of this virus in chickens [26, 39] . However, we detected replication of APMV-4 in trachea and lungs of infected ducks, indicating its preference to ducks from which this strain had been isolated. moderate multifocal mucosal attenuation, and loss of tracheal alveolar mucous glands. Lung sections (B) exhibit moderate, multifocal, lymphocytic to lymphohistiocytic bronchitis with mild to moderate perivascular and peribronchial interstitial inflammation and focal perivascular cuffing with varying severity. The presence of antigens for most APMVs was detected in the epithelial lining of trachea (C) and in the epithelium surrounding the medium and small bronchi of the lungs (D). doi:10.1371/journal.pone.0034927.g007 Table 3 . Replication of APMVs in 3-week-old ducks. Three birds from each group were sacrificed on day 4, and tissues samples (brain, trachea, lung, and spleen) were collected and homogenized. To confirm the virus replication, each sample (100 ml) in the collected samples was inoculated into three eggs, and allantoic fluids were collected on 3 dpi. Virus replication was determined by hemagglutination assay. Note that APMV-5 was not analyzed because it does not replicate in the allantoic cavity of chicken eggs. doi:10.1371/journal.pone.0034927.t003 In summary, our findings indicate that APMV serotypes 2-9 were avirulent in both chickens and ducks as well as in standard international assays. Each of the APMV serotypes replicated to low-to-moderate titers in the trachea of chickens, with APMV-3 having the highest titers. Among APMV types 2-9, only APMV3 replicated systemically and was neuroinvasive and neurotropic, although not neurovirulent. The sequence of the F protein cleavage site was not a reliable predictor of pathogenicity of APMVs in chickens, indicating incongruity with the well-known APMV-1 paradigm. For example, the systemic replication and neurotropism of APMV-3 is inconsistent with its lack of a furin cleavage site and dependence on exogenous trypsin for replication in cell culture. As another example, several of the APMV serotypes, including types 2, 4, 6, and 7, lacked a furin site, but replication of these viruses in cell culture was not improved by the addition of exogenous trypsin. As yet another example, APMV-5 was the only one of serotypes 2-9 to have a furin site, and yet it replicated to only moderate levels in cell culture, and replication in chickens could be detected only by seroconversion. Thus, the paradigm from studies with APMV-1 that virulence and systemic spread correlates with the presence of a furin site generally was not observed with APMV serotypes 2-9. Reverse genetics for different pathotypes of APMV-1 and for prototype strains of APMV-2, 3, and -7 have been developed and used for studying determinants of virus pathogenesis [38, [43] [44] [45] . Further characterization of virus replication and pathogenicity using reverse genetics will enhance our understanding of overall APMV pathogenesis.

Patterns and influencing factor of synonymous codon usage in porcine circovirus

BACKGROUND: Analysis of codon usage can reveal much about the molecular evolution of the viruses. Nevertheless, little information about synonymous codon usage pattern of porcine circovirus (PCV) genome in the process of its evolution is available. In this study, to give a new understanding on the evolutionary characteristics of PCV and the effects of natural selection from its host on the codon usage pattern of the virus, Patterns and the key determinants of codon usage in PCV were examined. METHODS: We carried out comprehensive analysis on codon usage pattern in the PCV genome, by calculating relative synonymous codon usage (RSCU), effective number of codons (ENC), dinucleotides and nucleic acid content of the PCV genome. RESULTS: PCV genomes have relatively much lower content of GC and codon preference, this result shows that nucleotide constraints have a major impact on its synonymous codon usage. The results of the correspondence analysis indicate codon usage patterns of PCV of various genotypes, various subgenotypes changed greatly, and significant differences in codon usage patterns of Each virus of Circoviridae.There is much comparability between PCV and its host in their synonymous codon usage, suggesting that the natural selection pressure from the host factor also affect the codon usage patterns of PCV. In particular, PCV genotype II is in synonymous codon usage more similar to pig than to PCV genotype I, which may be one of the most important molecular mechanisms of PCV genotype II to cause disease. The calculations results of the relative abundance of dinucleotides indicate that the composition of dinucleotides also plays a key role in the variation found in synonymous codon usage in PCV. Furthermore, geographic factors, the general average hydrophobicity and the aromaticity may be related to the formation of codon usage patterns of PCV. CONCLUSION: The results of these studies suggest that synonymous codon usage pattern of PCV genome are the result of interaction between mutation pressure and natural selection from its host. The information from this study may not only have theoretical value in understanding the characteristics of synonymous codon usage in PCV genomes, but also have significant value for the molecular evolution of PCV.

Genetic information is transmitted from mRNA to protein in a mode of triplet codon. Each amino acid matches with at least one codon, at most six codons. The codons encoding the same amino acid is called synonymous codon. During biosynthesis of protein, usage probability of those synonymous codons is different. Some species or some genes are usually prone to use one or several particular synonymous codons. These codons are called preferable codons, which is called as codon bias. Usage bias of codons from various species has been studied, and it is found that during protein biosynthesis synonymous codons encoding amino acid is not used randomly [1] [2] [3] . Many studies have indicated that obvious bias exists between different genes from different species or the same species [4] [5] [6] . Usage bias of codons is influenced mainly by mutation bias, translation selection, secondary protein structure, replication and selective transcription, hydrophobia and hydrophilia of protein, and external environment [7] [8] [9] [10] [11] [12] [13] . PCV belongs to genus of porcine circovirus, family of porcine circovirus. It has two genotype, PCV genotype I and PCV genotype II, and it is the smallest virus which has been discovered so far [14] . Among the different genotype, PCV genotype II infection and its related diseases have become one big problem across the globe for pig feeding, which threatens greatly to normal development of the industry of pig feeding. The PCV genome is a single-stranded negative circular DNA, and very small; full length of the PCV genotype I is only 1,759 bp, and PCV genotype II, 1,767 bp or 1,768 bp. The genome contains 11 open reading frames (ORF), among which, ORF1 encodes replication-associated proteins (Rep and Rep'); ORF2, structural proteins (viral capsid proteins, Cap); ORF3, toxicity-associated proteins, which can cause apoptosis [15, 16] . By analyzing the whole sequence of PCV genome, it is found that ORF2 has smaller selective pressure than ORF1, and more mutation. Nucleotide sequences among various strains in the same genotype are very conservative, their homology is over 90%, while similarity between nucleotide sequences from various strains respectively from the two genotypes is less than 80% [17, 18] . However, so far, studies have not related to usage of PCV codons. Explanation of codon usage pattern of PCV has significance on PCV evolution, gene prediction, gene classification, design of high expressed genes and viral vectors, and understanding of interaction between PCV and its host cells. Therefore, in this study, we first performed comprehensive analysis on codon usage pattern of PCV genome and the related factors affecting on codon usage. This study will play a major role in explanation of evolution process of PCV genome and further studies. The characteristics of synonymous codon usage in PCV In order to investigate usage pattern of the PCV codons, we calculated various RSCU values of various codons from 28 different strains from different genotype. It can be seen from the three-dimension mesh plot of the analysis results of correspondence between 59 synonymous codons in the PCV genome (See Figure 1) , range of Z axis (f'1) is between 15 to 2.5, which indicates that synonymous codons usage in the PCV genome is not balanced, that is to say, among all the 59 synonymous codons, a part of codon is rarely used, while others have higher usage frequency. Additionally, it can be seen from Table 1 that among 18 preferable codons, 12 ones have the end base of G or U, only 6 have the end base of A or C, and so those codons with the end base of G or U are prone to use in PCV genome. Nevertheless, compared with other vertebrate DNA viruses, PCV genome has lower GC%, from 48.35% to 49.12%, with an average content of 48.61% and SD value of 0.19 (Table 2 ). And hence, the phenomenon, that in PCV genome GC content is lower while the condons with the end base of G is used in a way of bias, suggests that content of G or C as the end base of codons has effect on usage pattern of synonymous codons. Apart from this, we can also see from Table 2 , ENC values between PCV genomes has less fluctuation, with a range from 55.32 to 58.67, and an average value of 56.80 and SD value of 0.85, which indicates that codon bias of the PCV genome is stable. Natural selection and mutation pressure has been considered to be two key factors which have effect on codon usage patterns of organisms [19] . In order to explore whether determinative factors for codon usage mutation in PCV is mutation pressure or natural selection, we compared correlation between A 3 %, U 3 %, G 3 %, C 3 %, GC 3 % and U 3 %, G 3 %, C 3 %, GC 3 % with correlation analysis ( Table 3 ). The analysis results show, except GC %, GC3% has marked correlation with A%, U%, G% and C%. This indicates that GC 3 % can reflect interaction between natural selection and mutation pressure to some extent. In addition, the correlation between f'1,f'2 value and A %, U%, G%,C%, GC%, A 3 %, U 3 %, G 3 %, C 3 %, GC 3 % was analyzed ( Table 4 ). The results showed that significant correlations exist between synonymous codon usage pattern and nucleotide composition in PCV. This result further verifies the conclusion that during the shaping of synonymous codon usage pattern of PCV, Composition constraints play and very important role. In order to compare synonymous codon usage patterns between different PCV genomes from different genotype, we carried out analysis on codon usage of different PCV genotype with correspondence analysis (CA). In correspondence analysis, the first dimension variable f'1 and the second dimension variable f'2 can reflect 39.87 and 23.83 percent of total mutation respectively. We can see from Figure 2 , except for two strains of PCV genotype I in deviation from the cluster, other all the strains of PCV genotype II lie in the same cluster and overlap partly with each other. Obviously, PCV genotype I and PCV genotype II lay in two independent areas, which demonstrate that codon usage between the different PCV genotype is of great significance. Meanwhile, PCV genotype II-A and PCV genotype II-B almost lie in the same area, but they have tiny difference (Figure 2 ), which suggests that different sub-genotype from the same PCV genotype have difference in the aspect of codon usage. From ENC values and corresponding relation distribution diagram of GC 3 % ( Figure 3 ) we can see, most points are near or under the theoretical curve, which suggests that apart from mutation pressure which influences on codon usage pattern in PCV, the usage pattern is also influenced by other factors. As parasitic organisms, virus's codon usage pattern would be subject to its host to some extent [6] . In this study, patterns of codon usage are compared in PCV and its natural host, and then found that there are high similarities between them. In detail, the high frequently used codons in the swine were also the non-preferred codons of PCV, such as CUG, GUG, CAG, AAU, GAC, GAA, UGU, AGA and UUU. Further more, all preferentially used codons of the genome of PCV and swine were all G-ended or U-ended codons (Table 1 ). These results suggest that the selection pressure from the host affects codon usage pattern of PCV. It is worthwhile to note that PCV genotype II have high similarities with swine than PCV genotype I ( Figure 4 ). In details, the values of RSCU in PCV genotype II and swine codon such as, AGA for Arg, GUG for Val, AGC for Ser, ACC for Thr were clearly different from that of PCV genotype I. It may be important one of Molecular mechanisms of infection and pathogenesis of PCV genotype II. Researches in recent years indicated that dinucleotide biases can affect codon bias [20] . To study the possible effect of dinucleotide composition on codon usage of the PCV genome, the relative abundances of the 16 dinucleotides in genome of the 28 PCV strains were calculated ( Table 5 ). The result show that the occurrences of dinucleotides are not randomly distributed and no dinucleotides were present at the expected frequencies. The relative abundance of CpG showed a strong deviation from the "normal range" (mean ± SD = 0.622 ± 0.029) and was markedly under represented. Correspondingly, the synonymous codon containing the CpG were inhibited because of CpG was present at the expected frequencies. In detail, low RSCU values were present in 7 of all 8 codon containing CpG and they are all not preferentially used codons (such as CCG, UCG, ACG, CGC, CGG, CGA, CGU) ( Table 1) . These observations indicate that the composition of dinucleotides also plays a key role in the variation found in synonymous codon usage among PCV. The correspondence analysis has been performed in order to compare the synonymous codon usage pattern between the viruses of Circoviridae. From which we could detect one major trend in the f'1 which accounted for 20.27% of the total variation, and another major trend in the f'1 for 15.46%of the total variation. A plot of the f'1 and the f'1 of each virus of Circoviridae was shown in Figure 5 . We can see from the plot that there were considerable differences for codon usage patterns among PCV, DCV, GCV, CoCV, CAV and BFDV. Concretely, PCV belong to the different genotype tends to come together. Moreover, DCV, GCV, CoCV and BFDV tend to come together and CAV was alone in separate area. It's clear that although different virus tends to come together, differences of synonymous codon usage pattern still exist between each virus. Therefore, the synonymous codon usage pattern of each virus of Circoviridae varies by the species of virus. To investigate whether there is a correlation between the codon usage of PCV and geographic factor, 28 virus genes of PCV were divided into eleven groups according to obtained area, and correspondence analysis was also used. As can be seen from the plot, coordinate of virus isolates from different country is separated, and these relatively isolated spots tend to cluster into several groups according to the genotype ( Figure 6 ). All above During protein biosynthesis synonymous codon encoding amino acids are not used randomly, and some species or some gene always prefers to use of one or several particular synonymous codons, which is called as codon usage bias. Precious studies reveal that different genes from different species or the same one have obvious codon usage bias [21, 22] . Codon usage bias is influenced mainly by mutation bias [23, 24] , translation selection [25, 26] , secondary protein structure [20, 27] , replication and transcription selection [28] , secondary mRNA structure [29] , gene length [30] , tRNA abundance [31] , gene function and external environment [32] . However, most of these studies focus on some higher organism and many microorganisms with large genome and more genes, and there are few studies on virus with small genome and few genes or comparison between virus and host. Relatively, there are more reports on codon usage in genomes from viruses with great harm to mankind, such as SARS, human immunodeficiency virus, influenza virus A and hepatitis virus. PCV is a primary pathogen of postweaning multisystemic wasting syndrome (PMWS), which has threatened the development of pig feeding industry seriously because in recent year's occurrence of this disease has increased so as to bring about great economic loss in the world industry of pig feeding. Further studies on codon usage pattern in PCV have great significance on mutation pattern and molecular evolution of PCV. However, reports on codon usage pattern in PCV are rare, and this study is the first report. By comparison with reported DNA viruses such as Duck plague virus, Duck enteritis virus, Iridovirus, Herpesvirus [33] [34] [35] [36] , synonymous codon usage bias in the Table 4 The correlation analysis between the first two axes in CA and the nucleotide contents of PCV a b NS in superscript represent non-significant ***P-value < 0.001. **P-value < 0.01. *0.01 < P-value < 0.05 Figure 2 Correspondence analysis plot of relative synonymous codon usage in PCV. The first dimension variable f' 1 and the second variable f' 2 can reflect 39.87% and 23.83% of total mutation respectively. ). This suggests that low codon bias may result from increase in itself replication efficiency in PCV in order to adapt to replication system of its hosts. In this study, relation between main indices (f'1 and f'2)for the correspondence analysis on PCV usage cofon usage and its nucleotide composition (See Table 2 ) indicates, mutation pressure has a significant role in PCV codon usage. Other factors which can influence on PCV codon usage are also analysed and the initial results show that mutation pressure is the main factor to influence on PCV codon usage variation. There were reports that natural selection can influence on synonymous codon usage pattern in viruses and the same conclusions are also obtained from this study. Three evidences support this conclusion. The first evidence is that PCV genome is GC3% -poor (average value = 47.08, SD = 2.88), but most of preferentially used codons are G/ T-ended codons. Meanwhile, average of A 3 % is higher than that of T 3 %, but among the codons which PCV prefers to using, there are only three preferable codons with the end base of A 3 % while six those with the end base of T 3 %. The second evidence is that the high similarities exist between PCV and its natural host. The third evidence is that CpG and the synonymous codon including it were inhibited. The three above evidences both state that natural selection is involved in formation of PCV synonymous codon usage pattern. At present, according to pathogenicity, antigenicity and nucleotide sequence difference, PCV is divided into two genotypes, PCV genotype I and PCV genotype II, of which PCV genotype II includes various subtypes. From significance of PCV codon usage between different genotypes in Figure 1 , we can see that PCV codon bias may have association with genotypes. In addition to this, the results in this study also reveal that geological factor may almost have relation with codon usage in PCV. In some reports, gene length has certain correlation with codon usage [30] . Similarly, in some viruses, gene length has no effect on codon usage [22] . With correlation analysis we surveyed codon usage bias and gene length in PCV, and it is found that in these viral genes, codon usage bias has no notable correlation with gene length (Spearman, r = 0.075, p > 0.1). The results indicate that PCV gene length has no effect on synonymous codon usage. Other factors, including GRAVY and aromaticity may also significantly influence codon usage of PCV Taken together, the codon usage patterns of PCV possibly result from interactions between natural selection and mutation pressure. These results not only provide an insight into the variation of codon usage pattern among the genomes of PCV, but also may help in understanding the processes governing the evolution of PCV. The information of 28 PCV genomes, including the genotype, length value, the isolated area and GenBank accession numbers of these strains was listed in the Table 6 . In order to compare the differences between PCV and its host, twenty swine gene were gained and detailed information of these genes is listed in Table 7 . In addition, to compare the codon usage patterns among different viruses, twenty-five viral genomes of Circoviridae were taken into account ( Table 8 ). All of the sequences were downloaded from NCBI (http:// www.ncbi.nlm.nih.gov/Genbank/). Each general nucleotide composition (T%,A%,C% and G%) and each nucleotide composition in the third site of codon (T 3 %,A 3 %, C 3 % and G 3 %) in PCV coding sequence were calculated by biosoftware DNAStar7.0 for windows. In order to eliminate the influence of amino acid composition on codon usage and directly reflect the usage Figure 4 Compare the codon the codon usage pattern among PCV genotype I, PCV genotype II and swine. characteristics of codon, the study evaluates synonymous codon usage bias through statistical estimation on relative synonymous codon usage frequency (RSCU) [37] . RSCU value refers to the ratio between the usage frequency of one codon in gene sample and expected frequency in the synonymous codon family. If the synonymous codon usage of one amino acid has no preferences, that is, codon usage frequency is close to expected frequency, the RSCU values of codons are equal to 1; if a codon RSCU value is greater than 1, the codon use frequency is higher than expected frequency, whereas it is less than expected value. The definition on a single gene codon bias is mainly based on effective number of codons (ENC) [38] . ENC values can reflect the preference degree of synonymous codon non-equilibrium use in codon family. The range of ENC values is from 20 (each amino acid only uses one codon) to 61 (all synonymous codons are equivalently used). ENC value is closer to 20, the degree of being used non-randomly is higher, and the bias is stronger. It is generally believed that the genes are provided with significant codon bias when ENC ≤ 35. The values of RSCU and ENC were obtained by codonW program. A comparison of actual and expected dinucleotide frequencies of the 16 dinucleotides in coding region of PCV genomes was also undertaken using SPSS 17.0. Correspondence analysis is mainly used for detecting the changes of codon RSCU values in genes [39] . It is an effective multivariate statistical method of studying the internal relation between the variables and samples, and it is successfully applied to the study of codon. In correspondence analysis, all genes in samples are distributed in a 59-dimensional (59 justice codons, in addition to the stop codon, Met, and Trp) vector space, each gene is described with 59 (f' 1 , f' 2, ..., f' 59 ) variables, the results can be applied for finding out the major factors affecting codon usage bias in genes [40, 41] . This was done using the CodonW program. Correlation analysis of PCV was used to identify the relationship between nucleotide composition and synonymous codon usage pattern [42] . All statistical processes were carried out by with statistical software SPSS17.0 for windows.

Development of a Humanized Antibody with High Therapeutic Potential against Dengue Virus Type 2

BACKGROUND: Dengue virus (DENV) is a significant public health threat in tropical and subtropical regions of the world. A therapeutic antibody against the viral envelope (E) protein represents a promising immunotherapy for disease control. METHODOLOGY/PRINCIPAL FINDINGS: We generated seventeen novel mouse monoclonal antibodies (mAbs) with high reactivity against E protein of dengue virus type 2 (DENV-2). The mAbs were further dissected using recombinant E protein domain I-II (E-DI-II) and III (E-DIII) of DENV-2. Using plaque reduction neutralization test (PRNT) and mouse protection assay with lethal doses of DENV-2, we identified four serotype-specific mAbs that had high neutralizing activity against DENV-2 infection. Of the four, E-DIII targeting mAb DB32-6 was the strongest neutralizing mAb against diverse DENV-2 strains. Using phage display and virus-like particles (VLPs) we found that residue K310 in the E-DIII A-strand was key to mAb DB32-6 binding E-DIII. We successfully converted DB32-6 to a humanized version that retained potency for the neutralization of DENV-2 and did not enhance the viral infection. The DB32-6 showed therapeutic efficacy against mortality induced by different strains of DENV-2 in two mouse models even in post-exposure trials. CONCLUSIONS/SIGNIFICANCE: We used novel epitope mapping strategies, by combining phage display with VLPs, to identify the important A-strand epitopes with strong neutralizing activity. This study introduced potential therapeutic antibodies that might be capable of providing broad protection against diverse DENV-2 infections without enhancing activity in humans.

Dengue is the most important arthropod-borne viral disease in humans and an increasing public health concern in tropical and subtropical regions of the world. Approximately 50-100 million cases of dengue fever (DF) and 500,000 cases of dengue hemorrhagic fever (DHF) occur every year, and 2.5 billion people are at risk of dengue infection globally [1, 2] . Dengue infection may lead to fever, headache and joint pain in milder cases but may also lead to the more severe life-threatening DHF/dengue shock syndrome (DSS) has plasma leakage, thrombocytopenia, and hemorrhagic manifestations, possibly leading to shock [3, 4] . Dengue virus (DENV) is positive-sense single-stranded RNA virus of approximately 11 kb genome of the genus Flavivirus, a family Flaviviridae. It has four genetically and antigenically related viral serotypes: DENV-1, -2, -3 and -4. Flaviviruses encode a single polyprotein processed by host and viral protease to produce three structural proteins, including capsid (C) protein, precursor membrane/membrane (prM/M) and envelope (E) protein, and seven nonstructural proteins: NS1, NS2A, NS2B, NS3, NS4A, NS4B and NS5 [5] . The E protein, a 53 kDa glycoprotein important for attachment, entry, and viral envelope fusion, can bind to cellular receptors and induce neutralizing antibodies [6, 7] . The DENV consists of an icosahedral ectodomain, containing 180 copies of the E protein [8] . E protein monomer contains three structural and functional domains [9, 10] . E protein domain I (E-DI) is a central b-barrel structure. E protein domain II (E-DII) is organized into two long finger-like structures and contains the flaviviruses conserved fusion loop. E protein domain III (E-DIII) has an immunoglobulin-like fold and may mediate interactions between the virus and the receptors on the host cell [11] . Studies of the biological characteristics and epitope specificities of mouse monoclonal antibodies (mAbs) have elucidated the antigenic structure of flavivirus E proteins [12] [13] [14] [15] . Serotype-specific mAbs with neutralizing activity against DENV-2 have been found to be located on the lateral ridge of E-DIII and the subcomplex-specific mAbs recognized A-strand of E-DIII [14, 16, 17] . Antibody-mediated neutralization has been found to alter the arrangement of viral surface glycoproteins that prevent cells from viral attachment [16] . Binding of an antibody to the viral surface can interfere with virus internalization or membrane fusion [6] . Primary DENV infection is believed to provide lifelong immunity against re-infection with the same serotype [18, 19] . However, humoral immune responses to DENV infection are complex [20] [21] [22] , and may exacerbate the disease during heterologous virus infection [18, 19] . Antibody-dependent enhancement (ADE) in dengue pathogenesis results from the increase in the efficiency of virus infection in the presence of nonneutralizing or sub-neutralizing concentrations of anti-E or anti-prM immunoglobulins [21, 23] . The attachment of antibody-virus complex to such Fcc receptor-bearing cells as monocytes and macrophages can lead to an increased virus replication [18, 24, 25] . A better understanding of the neutralizing epitopes may facilitate the generation of new antibody-based therapeutics against DENV infection. In this study, we generated several mAbs against DENV-2. We found that serotype-specific anti-E-DIII mAbs played an important role in the neutralization of virus infectivity. Studies of the neutralizing epitopes found the strongest mAbs to be DB32-6 and DB25-2, both DENV-2 serotype-specific antibodies. These two mAbs recognized the A-strand of E-DIII at residues K310 and E311, respectively. Humanized DB32-6 mAb efficiently neutralized DENV-2 infection in a therapeutic mouse model and its variant version prevented enhancing activity. BHK-21 cells were grown at 37uC with 5% CO 2 in Minimal Essential Medium (MEM, Gibco) supplemented with 10% heatinactivated fetal bovine serum (FBS, Gibco) and 100 U/ml penicillin, 100 mg/ml streptomycin, 0.25 mg/ml amphotericin B (Antibiotic-Antimycotic, Gibco). Aedes albopictus C6/36 cells were grown at 28uC in 1:1 Mitsuhashi and Maramorosch (MM) insect medium (Sigma-Aldrich)/Dulbecco's modified Eagle's medium (DMEM, Gibco) containing 10% FBS and 100 U/ml penicillin, 100 mg/ml streptomycin, 0.25 mg/ml amphotericin B (Antibiotic-Antimycotic, Gibco). The four DENVs (DENV-1 Hawaii, DENV-2 16681, DENV-3 H87 and DENV-4 H241) were provided by Dr. Duane J. Gubler from the Centers for Disease Control and Prevention, Fort Collins, U.S.A. The various DENV-2 strains including New Guinea-C (NGC), NGC-N (mouse-adapted neurovirulent), PL046 and Malaysia 07587 were used in this study [26, 27] . These viruses were passaged in C6/36 cells. Anti-DENV-2 mAbs were generated according to previously described procedures [28, 29] . Female 4-to 6-week-old BALB/c mice were immunized with 10 7 plaque-forming units (pfu) of DENV-2 (16681). The DENV-2 was purified from viral culture supernatant using 4G2 (an anti-E protein mAb)-coupled protein G-Sepharose 4 Fast Flow gel. After four inoculations with the same concentration of antigens, the splenocytes from the immunized mouse spleen were harvested and then fused with mouse myeloma NS-1 cells. Fused cells were cultured in DMEM supplemented with 15% FBS, HAT medium and hybridoma cloning factor (Roche) in 96-well tissue culture plates. Two weeks after fusion, culture supernatants were screened by ELISA. Selected clones were subcloned by limiting dilutions. Hybridoma clones were isotyped using a commercially isotyping kit (Southern Biotech) by ELISA. Ascites fluids were produced in pristine-primed BALB/c mice. mAbs were affinity-purified by standard protein G-Sepharose 4 Fast Flow (GE Healthcare Bio-Sciences) according to manufacturer's directions. Screening of mAbs against DENV1-4 by ELISA C6/36 cells at 80% confluency in 96-well plates were infected with DENV-1 to -4 to produce viral antigens. These cells were then harvested 5-7 days after infection. One mg/ml mAbs was added to the plates and incubated at room temperature (RT) for 1 h. After washing with PBS, horseradish peroxidase (HRP)-conjugated antimouse IgG (Jackson ImmunoResearch Laboratories) was incubated at RT for 1 h. Finally, plates were incubated with peroxidase substrate o-phenylenediamine dihydrochloride (OPD; Sigma-Aldrich). Reaction was stopped with 3N HCl and optical density was measured using a microplate reader set at 490 nm. C6/36 cells were harvested after viral infection. Lysates or expression proteins were collected. Cell extracts were mixed with sample buffer (Bio-Rad Laboratories). Protein samples were separated by SDS-PAGE and transferred to nitrocellulose membrane (Hybond-C Super). Nonspecific antibody-binding sites were blocked with 5% skimmed milk in PBS, and membranes were incubated with primary antibody. Blot was then treated with horseradish peroxidase-conjugated goat anti-mouse immunoglobulin (Jackson ImmunoResearch Laboratories) and then developed with enhanced chemiluminescence reagents (ECL, Thermo Fisher Scientific). Immunofluorescence assay (IFA) BHK-21 cells at 80% confluency were infected at a multiplicity of infection (MOI) of 0.5 with DENV-2 (16681). After 2 days infection, the cells were fixed with 1:1 methanol/acetone for 10 min at 220uC. Cells were blocked using PBS supplemented with 1% BSA for 1 h at RT. Primary anti-DENV antibodies or control antibodies (normal mouse IgG, Jackson ImmunoResearch Laboratories) were diluted (1:250) in block solution for 1 h at RT. Secondary antibody, FITC-conjugated goat anti-mouse IgG Dengue virus (DENV) infection remains a serious health threat despite the availability of supportive care in modern medicine. Monoclonal antibodies (mAbs) of DENV would be powerful research tools for antiviral development, diagnosis and pathological investigations. Here we described generation and characterization of seventeen mAbs with high reactivity for E protein of DENV. Four of these mAbs showed high neutralizing activity against DENV-2 infection in mice. The monoclonal antibody mAb DB32-6 showed the strongest neutralizing activity against diverse DENV-2 and protected DENV-2-infected mice against mortality in therapeutic models. We identified neutralizing epitopes of DENV located at residues K310 and E311 of viral envelope protein domain III (E-DIII) through the combination of biological and molecular strategies. Comparing the strong neutralizing activity of mAbs targeting A-strand with mAbs targeting lateral ridge, we found that epitopes located in A-strand induced stronger neutralizing activity than those located on the lateral ridge. DB32-6 humanized version was successfully developed. Humanized DB32-6 variant retained neutralizing activity and prevented DENV infection. Understanding the epitope-based antibody-mediated neutralization is crucial to controlling dengue infection. Additionally, this study also introduces a novel humanized mAb as a candidate for therapy of dengue patients. Humanized mAb against Dengue Virus www.plosntds.org (Jackson ImmunoResearch Laboratories) was diluted to 1:250 and supplemented with DAPI (Invitrogen) diluted 1:2,000 for 1 h at RT. The binding activity of antibodies to the DENV-2-infected or transfected cells were observed and photographed through a fluorescence microscope. The expression constructs of E-DI-II and E-DIII were cloned into the pET21a vector (Merck). The E-DI-II, comprising amino acids 1-295 of the E protein, was tagged to flag and hexahistidine at the C terminus for affinity purification. The E-DIII, comprising amino acids 295-400 of the E protein, was tagged to flag and hexahistidine, too. The plasmids were expressed in Escherichia coli strain BL21 (DE3). The recombinant proteins E-DI-II and E-DIII were analyzed using 12% SDS-PAGE by Western blot analysis. The DNA fragments corresponding to E-DI-II and E-DIII were also cloned into a mammalian expression vector, pcDNA3.1 (Invitrogen). The expression constructs of DENV-2 C, prM, prM-E, E, NS1, NS2A, NS2B, NS2B-3, NS3, NS4A, NS4B and NS5 were obtained from Dr. Y.-L. Lin [30] . Transient expression of DENV-2 proteins in BHK-21 cells was transfected by PolyJet (SignaGen Laboratories) according to manufacturer's recommendations and then to test specificity of mAbs. In vitro neutralization assay (i) For the plaque reduction neutralization test (PRNT), eight 3fold serial dilutions of mAbs (from 200 mg/ml to 0.1 mg/ml) were mixed with an equal volume of 200 pfu of DENV-2 (16681) and incubated at 4uC for 1 h. The final concentration of mAbs at the PRNT ranged from 100 to 0.05 mg/ml. Antibody-virus mixtures (100 ml) were added to BHK-21 cells at 80%-90% confluency in 12-well plates. After absorption of virus for 2 h, BHK-21 cells were washed and 2 ml of 1% (w/v) carboxyl methyl cellulose (Sigma-Aldrich) in MEM plus 2% (v/v) FBS was layered onto the infected cells. After incubation at 37uC for 5 to 7 days, the viral plaque that had formed on the cell monolayer was fixed by 1 ml 3.7% formaldehyde (Sigma-Aldrich) at RT for 1 h. The cells were then stained with 1% crystal violet. Percentage of plaque reduction was calculated as: %Inhibition = 1002[(plaque number incubated with mAb/plaque number without mAb)6100]. (ii) For flow cytometry, serial dilutions of DB32-6 mAb were incubated with DENV-2 (16681, NGC, PL046 and Malaysia 07587) at MOI of 0.5 at 4uC 1 h before adding BHK-21 cells. After 2 h absorption, the monolayers were washed and incubated with MEM (Gibco) plus 2% (v/v) FBS at 37uC for 2 days. The cells infected with DENV-2 were washed and fixed with 3.7% formaldehyde at 4uC for 10 min. They were then permeabilized in PBS supplemented with 1% FBS, 0.1% saponin (Sigma) at 4uC for 10 min. For staining, cells were incubated with 4G2 at a concentration of 1 mg/ml at 4uC for 30 min. After two washes, R-Phycoerythrin (PE)conjugated AffiniPure F(ab9) 2 fragment goat anti-mouse IgG (H+L) (Jackson ImmunoResearch Laboratorie) diluted 1:250 was then added at 4uC for 30 min followed by two washes and analyzed by flow cytometry. % Infection = (the intensity of cells incubated with mAb/without mAb)6100. This study was carried out following strict guidelines from the care and use manual of National Laboratory Animal Center. The protocol was approved by the Committee on the Ethics of Animal Experiments of Academia Sinica. (Permit Number: MMi-ZOOWH2009102). The mice were killed with 50% CO 2 containing 50% O 2 . All efforts were made to minimize suffering. (i) Breeder mice of the ICR strain were purchased from the Laboratory Animal Center National Taiwan University College of Medicine. Purified mAbs at doses of 1, 10 and 100 mg/ml were incubated with 1610 4 pfu (25-fold LD 50 ) of DENV-2 (16681) at 4uC for 30 mins. Two-day-old suckling mouse brain was inoculated with 20 ml of the reaction mixture by intracranial (i.c.) injection. Survival rate and signs of illness, including paralysis, were observed daily for 21 days following challenge. In post-exposure therapeutic experiments, mice were passively injected with 5 mg of mAb via i.c. route after 1 day of infection. (ii) Stat1-deficient mice (Stat1 2/2 ) [31] were bred in the specific-pathogen-free animal facility at the Institute of Biomedical Sciences, Academia Sinica. Mice were challenged intraperitoneally with 1610 5 pfu (300-fold LD 50 ) of DENV-2 (NGC-N) in 300 ml of PBS and simultaneously injected intracranially (i.c.) with 30 ml of PBS. In prophylaxis experiments, antibodies (100 mg per mouse, intraperitoneally) were administered 1 day before infection and administered on day 0, 1, 3, 5 and 7 after infection. In postexposure therapeutic experiments, antibodies (100 mg per mouse, intraperitoneally) were administered on day 1, 3, 5 and 7 after infection. The phage display biopanning procedures were performed according to previous reports [28, 32] . Briefly, an ELISA plate was coated with mAbs at 100 mg/ml. Samples of 100 ml diluted mAb were then added to wells and incubated at 4uC for 6 h. After washing and blocking, the phage-displayed peptide library (New England BioLabs, Inc.) was diluted to 4610 10 pfu of phage and incubated for 50 mins at RT. After washing, bound phage was eluted with 100 ml 0.2 M glycine/HCl (pH 2.2) and neutralized with 15 ml 1 M Tris/HCl (pH 9.1). The eluted phage was amplified in ER2738 for subsequent rounds of selection. The phage was titrated onto LB medium plates containing IPTG and X-Gal. The biopanning protocol for the second and third rounds was identical to the first round except for the addition of 2610 11 pfu of amplified phage for biopanning. An ELISA plate was coated with 50 ml mAbs 50 mg/ml. After washing and blocking, amplified phage diluted 5-fold was added to coated plate and incubated at RT for 1 h. After washing, 1:5000 diluted HRP-conjugated anti-M13 antibody (GE Healthcare) was added at RT for 1 h. OPD developed and was terminated with HCl. Optical density was measured at 490 nm. We used the recombinant expression plasmid pCBD2-2J-2-9-1 [33] to generate VLP mutants. Various VLP mutants were generated by site-directed mutagenesis derived from pCBD2-2J-2-9-1 as a template. PCR was performed using pfu ultra DNA polymerase (MERCK) and all mutant constructs were confirmed by sequencing. BHK-21 cells at 80%-90% confluency in 48-well plates were transfected with plasmids of various VLPs. After two days transfection, the cells were washed with PBS supplemented with 1% FBS, fixed with 3.7% formaldehyde at 4uC for 10 min, and then permeabilized in PBS supplemented with 1% FBS, 0.1% saponin (Sigma-Aldrich) at 4uC for 10 min. For staining, cells were incubated with mAbs at 4uC for 30 min, DB32-6, DB25-2, 3H5 and mix mAbs (4G2, DB2-3, DB13-19, DB21-6 and DB42-3) at a concentration of 0.1, 1, 1 and 1 mg/ml, respectively. After being washed twice, R-Phycoerythrin (PE)-conjugated AffiniPure F(ab9) 2 fragment goat anti-mouse IgG (H+L) (Jackson ImmunoResearch Laboratories) diluted to 1:250 was then added at 4uC for 30 min and analyzed by flow cytometry. Relative recognition was performed according to previously described procedures and calculated as [intensity of mutant VLP/intensity of WT VLP] (recognized by a mAb)6[intensity of WT VLP/intensity of mutant VLP] (recognized by mixed mAbs) [34] . Total RNA was extracted from hybridoma cells using the TRIzol reagent (Invitrogen) and mRNA was isolated with the NucleoTrap mRNA Mini Kit (Macherey-Nagel GmbH & Co. KG.). Purified mRNA was reverse transcribed using oligo (dT) as a primer in a ThermoScript RT-PCR system (Invitrogen). The variable heavy-and light-chain domains (V H and V L ) were amplified from the cDNA product by PCR with a variety of primer sets [35, 36] . The PCR products were cloned using the TA kit (Promega) and the V H and V L sequences were determined by DNA sequencing. Software Vector NTI was used for sequence analysis. From these sequences, the framework regions (FRs) and complementarity-determining regions (CDRs) were analyzed by comparing them with those found in the Kabat database and the ImMunoGeneTics database [37] . Two human genes, GenBank accession DI084180 and DI075739, were 94.7% and 92.2% identical to DB32-6 V H and V L , respectively. Humanized DB32-6 V H consisted of the modified FR1 to FR4 from the accession DI084180 gene, and the CDR1 to CDR3 of the DB32-6 V H , respectively, while humanized DB32-6 V L consisted of the modified FRs from the accession DI075739 gene and the CDRs of the DB32-6 V L . Both were synthesized (GENEART) and amplified by PCR using pfu Turbo DNA polymerase (EMD Bioscience). The resulting V H was cloned into modified expression vector pcDNA3.1 (Invitrogen) with a signal peptide and human IgG1 constant region, while the V L was cloned into modified expression vector pSecTag (Invitrogen). We generated a variant of humanized DB32-6 (hDB32-6 variant) in which leucine residues at positions 1.2 and 1.3 of C H 2 domain were substituted with alanine residues [38] . The V H and V L plasmids were cotransfected into CHO-K1 cells and selected by G 418 and puromycin for 2-3 weeks. Transformed cells were limit diluted in 96-well plates. After two weeks, stable clones produced humanized antibodies in the McCoy's 5A medium (Sigma-Aldrich), as identified by ELISA. Humanized antibodies were produced by CELLine AD 1000 (INTEGRA Biosciences) according to manufacturer's directions. Murine and humanized DB32-6 mAbs affinity analysis for E-DIII of DENV-2 was performed by surface plasmon resonance (BIAcore X, Biacore, Inc). Purified E-DIII (50 mg/ml) was immobilized on a CM5 sensor chip (Biacore, Inc) and injected at a flow rate of 10 ml/min. The mAbs were diluted to 4, 2, 1, 0.5, 0.25 and 0 nM in HBS-EP buffer (Biacore, Inc). mAbs were injected at a flow rate of 30 ml/min for 3 min and then allowed to dissociate over 1.5 min. Regeneration of the surface was achieved with an injection of 10 mM glycine HCl/0.2 M NaCl (pH 3.0) before each mAb injection. The data were analyzed by the BIAevaluation software with a global fit 1:1 binding model. Serial dilutions of mAbs were mixed with DENV-2 (16681) at MOI of 1 at 4uC for 1 h. The 100 ml mixture were incubated with 5610 4 K562 cells [39] in 96-well plates at 37uC for 2 h. After infection, the cells were washed and incubated with RPMI (Gibco) plus 2% (v/v) FBS at 37uC for 2 days. The cells were washed with PBS supplemented with 1% FBS, fixed with 3.7% formaldehyde, and permeabilized in PBS supplemented with 1% FBS, 0.1% saponin (Sigma) at 4uC for 10 min. For staining, cells were incubated with DB42-3 at a concentration of 3 mg/ml at 4uC for 30 min. After two times washes, R-Phycoerythrin (PE)-conjugated AffiniPure F(ab9) 2 fragment goat anti-mouse IgG (H+L) (Jackson ImmunoResearch Laboratories, West Grove, PA) diluted 1:250 was then added at 4uC for 30 min follow by two times wash steps and analyzed by flow cytometry. Survival rate was expressed using Kaplan-Meier survival curve, and log rank test was used to determine the significant differences. For body weight change experiments, paired t-test was used to determine the significant differences, * P,0.05, ** P,0.01. Seventeen mAbs with high reactivity against E protein of DENV-2 were generated after immunization of mice with DENV-2 strain 16681. We identified 17 mAbs belonging to the IgG isotype that reacted with DENV-2-infected cells but not with mock-infected cells using immunofluorescence assay (IFA) ( Figure S1 ) and ELISA ( Figure 1A ). 4G2 was a pan-flavivirus mAb that could recognize the fusion loop of E-DI-II, and 3H5 (ATCC HB46) was a DENV-2 serotype-specific mAb that could recognize the lateral ridge of E-DIII [12, 17, 40] . Both 4G2 and 3H5 were used as positive controls (Figure 1 ). The specificities of the mAbs recognized as the four DENVs were further confirmed by ELISA and Western blotting (Figures 1A-1B and Table 1 ). Based on our Western blot analysis using a nonreducing condition, 14 of the mAbs recognized E protein (53 kDa) ( Figure 1B ). Three mAbs could not be identified by Western blotting. In order to identify the target proteins of these mAbs, we prepared BHK-21 cells transfected with plasmids expressing DENV-2 C, prM, prM-E, E, NS1, NS2A, NS2B, NS2B-3, NS3, NS4A, NS4B and NS5 ( Figure S2 ). Results indicated that three mAbs (DB21-6, DB22-4 and DB36-2) recognized E protein ( Figure 1C ). The identification and characterization of the 17 mAbs are summarized in Table 1 . To characterize the antigenic structure of the DENV E protein and to study the relationship between epitopes and their neutralizing potency, we constructed and expressed the recombinant E-DI-II and E-DIII from DENV-2 in E. coli and mammalian expression systems. Western blot analysis and IFA showed that, of the 17 mAbs recognizing E protein, 10 mAbs (DB2-3, DB9-1, DB13-19, DB21-6, DB22-4, DB23-3, DB27-3, DB33-3, DB39-2 and DB42-3) targeted to E-DI-II and 2 mAbs (DB25-2 and DB32-6) recognized E-DIII ( Figures 1D-1E and Table 1 ). However, 5 mAbs could not be identified by these two assays. We evaluated the ability of mAbs to inhibit DENV-2 infection in BHK-21 cells using a plaque reduction neutralization test (PRNT). Ten mAbs had neutralizing activity with 50% PRNT (PRNT 50 ) concentrations ranging from 0.14 mg/ml to 33 mg/ml (Table 1) . DB32-6 was found to be a DENV-2 serotype-specific mAb against E-DIII ( Figures 1A, 1B and 1E) and was the most efficient at neutralizing DENV-2 infection at a PRNT 50 concentration of 0.14 mg/ml (Figure 2A ). In addition, it could completely inhibit the infection at a lower concentration of 1.2 mg/ml (Figure 2A) . The mAb DB25-2 was found to be a DENV-2 serotype-specific mAb against E-DIII ( Figures 1A, 1B Humanized mAb against Dengue Virus www.plosntds.org and 1E) and to neutralize DENV-2 at a PRNT 50 titer of 1.2 mg/ ml (Figure 2A) . These findings indicate that serotype-specific mAb DB32-6 against E-DIII was the most potent in neutralizing DENV infection. Some serotype-specific mAbs, such as DB2-3 and DB23-3 against E-DI-II and DB25-2 against E-DIII showed strong neutralizing activity. Many complex reactive mAbs showed moderate-to-poor neutralizing activity (Table 1) . mAbs prevent DENV-2-induced lethality in suckling mice and Stat1 2/2 mice Two different mouse models were used to assess whether DB32-6 could efficiently protect mice against DENV-2 challenge. Protection assay of neutralizing mAbs was performed with ICR strain 2-day-old suckling mice [41] . Mice were inoculated intracerebrally with 20 ml of DENV-2-mAb mixture containing 1610 4 pfu (25-fold LD 50 ) of DENV-2 with neutralizing mAbs at concentrations of 1, 10 or 100 mg/ml. Generally, the nonneutralizing antibody normal mouse IgG (NMIgG) treated group showed paralysis, ruffling, and slowing of activity around 6 to 9 days. This was followed by severe sickness leading to anorexia, asthenia and death within 9 to 17 days ( Figures 2B and 2C) . In contrast, mAbs DB32-6 at a concentration of 10 mg/ml protected 93% of the mice from the lethal challenge of DENV-2 ( Figure 2B ). mAbs 3H5, DB23-3, DB2-3 and DB25-2 had survival rates of Humanized mAb against Dengue Virus www.plosntds.org 75%, 76%, 72% and 71%, respectively. DB42-3 and DB13-19 had survival rates of 46% and 28%, respectively ( Figure 2B ). The neutralizing mAbs showed a significant delay of the onset of paralysis and death relative to the NMIgG. To evaluate the therapeutic potential of the highly protective mAb DB32-6, we administered 100 mg/ml or 1 mg/ml to infected suckling mice. The survival rates for DB32-6 at 100 mg/ml or 1 mg/ml were 100% and 89%, respectively ( Figure 2C ). In comparison, 3H5 showed 82% and 40% survival rates at 100 mg/ml or 1 mg/ml, respectively. Stat1 2/2 mice, which lack a transcription factor involved in interferons (IFNs) signaling were sensitive to lethality induced by DENV-2 infection [27, 31] . To test the potential therapeutic effects of the strongest neutralizing mAb DB32-6, we challenged Stat1 2/2 mice at a strict condition with 1610 5 pfu (300-fold LD 50 ) of DENV-2 (NGC-N). After 21 days observation, mice showed ruffled fur, mild paralysis and lost approximate 20% of their initial body weight at day 7 after infection (P,0.01), and then they all died within 7-18 days of infection ( Figures 2D and 2E ). In the prophylaxis experiments, antibodies (100 mg per mouse, intraperitoneally) were administered 1 day before infection and at day 0, 1, 3, 5 and 7 after infection. The DB32-6 prophylactically treated group showed 100% protection ( Figure 2D left) . Even in the postexposure therapeutic experiments, the DB32-6 treated mice had a survival rate of 50% (Figure 2E left) . The mAb DB32-6 had excellent neutralizing activity against different DENV-2 strains (16681 and NGC-N) in two mouse models. To further evaluate whether the strongest mAb DB32-6 could broadly neutralize the diverse DENV-2 strains, we infected BHK-21 cells with four different DENV-2 Southeast Asian genotype strains, 16681, NGC, PL046 and Malaysia 07587. Remarkably, mAb DB32-6 exhibited effective neutralization against various DENV-2 strains ( Figure S3 ). Epitopes recognized by neutralizing antibodies have been identified in all three domains of the E protein [42] [43] [44] . To find out more about the epitopes of these neutralizing antibodies, we used phage display [29, 45] to identify the neutralizing epitopes. After three rounds of phage display biopanning, the phage titers were increased to 85-fold (DB32-6) and 331-fold (DB25-2) compared to the phage display biopanning results from the first round ( Figure 3A ). Individual phage clones from the third round of biopanning were randomly selected. ELISA was performed to determine whether the mAbs could specifically recognize selected phage clones. Of 20 selected phage clones, 17 and 18 clones had significant enhancement of binding activity to DB32-6 and DB25-2, respectively ( Figure 3B ). The selected phage clones PC32-6 and PC25-14 were specific and dose dependently bound to DB32-6 and DB25-2, respectively. They did not react with control NMIgG ( Figure 3C) . The 17 immunopositive phage clones that were highly reactive with DB32-6 were amplified and phage DNA was isolated for DNA sequencing. All of the phage clones displayed 12 amino acid (aa) residues ( Figure 3D left) . Phage-displayed peptide sequences selected by DB32-6 had the consensus motifs of histidine (H)-lysine (K)-glutamic acid (E)-tryptophan (W)/tyrosine (Y)-histidine (H) (Figure 3D left) . Similarly, 17 immunopositive phage clones selected by DB32-6 using phage library displayed 7 amino acid residues, which contained the consensus motif H-K-E-W/Y-H ( Figure 3D left) . Interestingly, all phage-displayed peptides selected by DB32-6 and DB25-2 contained lysine (K) and glutamic acid (E), respectively ( Figure 3D ). To further confirm the neutralizing epitopes, we developed various E protein epitope-specific variants VLPs and screened lossof-binding VLP mutants for identification of critical recognition residues. Using this strategy, we found that DB32-6 lost its VLP binding activity when the residue K310 in the A-strand of E-DIII was changed to alanine (K310A) or glutamine (K310Q) (Figure 4A left) . Similarly, DB25-2 lost its VLP binding activity when E311 was changed to arginine (E311R) in the A-strand of E-DIII ( Figure 4A right). Both the critical recognition residues K310 and E311 were located in the A-strand of E-DIII (Figures 4B and 4C) . We found that mAb 3H5 recognized residues K305, E383 and P384 ( Figure S4 ), as previously reported [17, 20] . Notably, even the adjacent residues (K310 and E311) induced antibodies with different levels of neutralizing activity. By comparing the amino acid sequences of E proteins from representing genotypes of DENV-2 (Table S1), we found residues K310 and E311 in E-DIII of the different genotypes (Southeast Asian, West African and American) ( Figure S5 ). Our data further showed epitopes in the A-strand of E-DIII were important for inducing neutralizing antibodies. Murine mAbs have been shown to have limited clinical use because of their short serum half-life, inability to trigger human effector functions and the production of human anti-murine antibodies (HAMA) response [46] . mAbs have been humanized by grafting their CDRs onto the V H and V L FRs of human Ig molecules [47] . DB32-6 was the most potent mAb against DENV-2 and showed potential as a therapeutic antibody. To develop humanized mAbs, we sequenced V H and V L segment of the neutralizing mAbs from hybridoma cell lines. The CDRs of DB32-6 were grafted onto human IgG1 backbone to create humanized DB32-6 (hDB32-6) ( Figure 5A ). The hDB32-6 was expressed in CHO-K1 cells and purified from culture supernatants. Both hDB32-6 and mDB32-6 were able to against DENV-2 ( Figure 5B ). The hDB32-6 maintained the specificity of murine DB32-6 (mDB32-6). Furthermore, we established stable clones of hDB32-6. After selection, mAbs hDB32-6-30, hDB32-6-48 and hDB32-6-51 were found to have highly binding activity ( Figure 5C ). Comparing to these mAbs, we found hDB32-6-48 to have the highest production rate in cells. mAb hDB32-6-48 was dose-dependent against DENV-2 and E-DIII ( Figure 5D ). The affinity was analyzed by surface plasmon resonance. The mDB32-6 and hDB32-6-48 bound to E-DIII of DENV-2 with a similar affinity (0.12 nM and 0.18 nM, respectively) ( Figure 5E ). The results revealed that hDB32-6 maintained the same binding affinity to the E protein as mDB32-6. mAb hDB32-6 protected mice from DENV-2-induced mortality We established a suckling mice model to determine the protective activity of mDB32-6 and hDB32-6. To evaluate Humanized mAb against Dengue Virus www.plosntds.org therapeutic effect of mAbs, we administered 5 mg of mAb at day one after 1610 4 pfu (25-fold LD 50 ) of DENV-2 (16681) infection. Through 21 days of observation, groups treated with mDB32-6, hDB32-6-48 and 3H5 mAbs were found to have survival rates of 96%, 94% and 56%, respectively ( Figure 6 ). However, none of the mice in control antibody normal human IgG (NHIgG)-treated group survived (Figures 6) . These results demonstrate that both mDB32-6 and hDB32-6 have excellent neutralizing activity against DENV-2. mAb hDB32-6 variant eliminate ADE phenomenon When developing the antibody-based therapy, ADE phenomenon is a major cause for concern in dengue pathogenesis because it might enhance DENV infection. Modification of Fc structure in an antibody can prevent Fcc receptors binding and lead to eliminate ADE [38, 39, 48] . We generated a variant of humanized DB32-6 (hDB32-6 variant) to prevent Fcc receptors binding while maintaining DENV neutralizing capability without enhancing infection (Figure 7) . The hDB32-6 variant retained the same neutralizing activity as unmodified mAb mDB32-6 at high concentrations (100 mg/ml and 10 mg/ml) but was completely devoid of enhancing activity at low concentrations (1 mg/ml and 0.1 mg/ml) (Figure 7) . The hDB32-6 variant eliminated the ADE phenomenon and holds great potential for being developed into therapeutic antibodies for the prevention and treatment of DENV-2 infection. mAbs of DENV have served as powerful research tools for antiviral development and pathological investigations. Here, we newly generated and characterized 17 mAbs with high reactivity against E protein of DENV-2. Several mAbs had potent neutralizing activity. The neutralizing epitopes were identified using a combination of strategies, including phage display, computational structure analysis [49] , and high-throughput epitope mapping of VLPs. From these results, the A-strand of E-DIII was found to be important in neutralizing DENV-2 than the Humanized mAb against Dengue Virus www.plosntds.org lateral ridge of E-DIII. mAb DB32-6 which had the strongest neutralizing activity against various strains of DENV-2 was humanized and modified to abrogate the ADE phenomenon. The mAb DB32-6 was demonstrated to increase the survival rate in two mouse models even after DENV-2 infection. Based on previous epitope mapping results, several epitopes have been shown to elicit strong neutralizing antibodies against individual flaviviruses that situated in E-DIII [14, 50] . Investigation of neutralizing epitopes on the E proteins may provide the framework for a detailed understanding of both specific mechanisms of the viral infection as well as the identification of the specific DENV domain that attaches to a cellular receptor. Phage display is useful in the identification of B-cell epitopes, including linear [32, 51] and conformational epitopes [29, 45] . However, these epitopes need to further elucidation using other methods. Combining different strategies provided a fast and reliable evidence for identifying epitopes (Figures 3 and 4) . To date, few mAbs possess better neutralizing activity than 3H5, which has been shown to bind to residues K305, E383 and P384 at the lateral ridge of E-DIII [17, 20] . DB32-6 had higher neutralizing activity than 3H5. Neutralizing epitope of DB32-6 was mapped on K310 residue in A-strand of E-DIII (Figure 4 ). Neutralizing epitope of another mAb DB25-2 was mapped on E311 residue in A-strand of E-DIII, too ( Figure 4) . These serotype-specific neutralizing epitopes located in the A-strand of E-DIII induced stronger neutralizing activity than those located on the lateral ridge of E-DIII. We aligned different DENV-2 genotypes and found that the K310 and E311 were frequently observed in DENV-2 ( Figure S5 ). The K310 may be important to DENV-2. Thus by binding DB32-6 to K310, it lead to dramatic neutralized DENV-2. To determine whether DB32-6 can neutralize diverse genotypes of DENV-2 is a critical step in evaluating the potential of therapeutic development in the future. Previous studies have shown that the strongly neutralizing mAb, subcomplex-specific 1A1D-2 and cross-reactive 9F12 recognized residues at K305, K307 and K310 in A-strand [15, 17] . Our mAb DB32-6 is a serotype-specific neutralizing mAb that recognized residue K310 but not residues K305 or K307. Although K310 is considered as a subcomplex-specific epitope, DB32-6 is a serotypespecific mAb. There may be other regions that affect the binding of DB32-6 to DENV-2. We found that by mutating residue I312, DB32-6's binding activity was reduced by 50% (data not shown). Residue I312 may be a minor epitope of DB32-6. Moreover, 1A1D-2 is a temperature dependent mAb due to its needs for dynamic motion on the virion surface to neutralize virus [16] . Different from 1A1D-2, DB32-6 is temperature independent. When DB32-6 was incubated with DENV at 4uC, it still exhibited significant neutralizing activity (Figures 2 and 6 ). As expected, when incubating the DENV and DB32-6 at 37uC, DB32-6 showed better efficacy than it did at 4uC (data not shown). The residue K310 on the surface of DENV-2 may be accessible to DB32-6 binding. Additionally, DB32-6 had high binding affinity (0.12-0.18 nM) to DENV-2. Based on the above finding, the residue K310 induce serotype-specific mAbs and is crucial in the neutralization of virus infectivity. Antibodies to E-DI-II tend to be more cross-reactive and less potent in neutralization of dengue infection [39] . However, there are fewer antibody concentrations capable of recognizing E-DIII than there are that recognize E-DI-II in dengue patients [20, 39] . Wahala et al. studied the human immune sera of DENV infection and found the E-DIII binding antibodies to play a minor role in DENV neutralization, similar to West Nile virus-infected human [52, 53] . The mAbs that bind to E-DIII expresses potent neutralizing activity, but only a few of them exist in serum of the patients infected with DENV or WNV. Combining the information from both mice and human mAbs studies of DENV infection is critical to understanding the complex mechanism behind the humoral immunity following natural DENV infection. According to one previous study, the immunoglobulin populations recognizing residues K310, E311 and P364 in dengue fever patients were much larger in IgM than in IgG [20] . The strong neutralizing IgG made up a small proportion of the antibody in dengue patients. de Alwis et al. has conducted an in-depth analysis of the human mAbs derived from memory B-cells of patients infected with primary DENV infections [54] . After the epitope mapping of anti-DENV-2 human mAbs, the strong neutralizing mAb 10.16 was mapped to K305, K310 and E311 in the A-strand. Together, the finding above suggest that the highly protective epitopes K310 and E311 in mouse play a role in humans as well. We also identified several E-DI-II specific mAbs with high to no neutralizing activity. Serotype-specific mAbs (DB2-3 and DB23-3) with potent neutralizing activity were found to recognize E-DI-II of DENV-2 ( Figure 2 and Table 1 ). Some studies have identified highly neutralizing and protective antibodies against JEV and Figure 5 . Construction and characterization of humanized DB32-6 mAb. (A) Amino acid sequences of humanized DB32-6 (hDB32-6). FR, framework region; CDR, complementarity determining region. Red residues represent the different amino acids from murine DB32-6 (mDB32-6). (B) mDB32-6 and hDB32-6 mAbs recognized DENV-2-infected BHK-21 cells by IFA. Cells were counterstained with DAPI (blue) and observed at 4006 magnification. (C) Binding activity of hDB32-6 mAbs. Three stable clones of hDB32-6 (hDB32-6-30, hDB32-6-48 and hDB32-6-51) recognized DENV-2infected C6/36 cells and recombinant E-DIII of DENV-2 by ELISA. (D) Various concentrations of mDB32-6 and hDB32-6-48 mAbs were reactive to DENV-2 and recombinant E-DIII of DENV-2 but not to mock control. NMIgG and NHIgG were used as negative controls. (E) Binding affinities of mDB32-6 and hDB32-6-48 to E-DIII of DENV-2. mAbs affinity analysis was performed by surface plasmon resonance (SPR). Binding affinity was tested at the mAb concentrations ranging 0 to 4 nM. Binding curves and kinetic parameters are shown. doi:10.1371/journal.pntd.0001636.g005 Humanized mAb against Dengue Virus www.plosntds.org DENV located in E-DI [55, 56] Currently, we are in the process of identifying the neutralizing epitopes of DB2-3 and DB23-3. mAbs that broadly cross-react with other flaviviruses are in E-DII near the fusion loop, which is immunodominant antigenic [20, 34, 42] . Binding an antibody to DENV can change the rearrangement of the E protein, which may neutralize or enhance viral infection [16, 57] . The high or no neutralizing activity of our mAbs can be used help identify neutralizing or immunopathogenic epitopes in the E protein. Studies that explore the mAbs mediated neutralization mechanism and mAbs dependent enhancement are currently underway. The mouse models for dengue infection developed to date do not represented the entirety of the pathogenesis of human dengue infection [58] . Developing of mouse models to studying its pathogenesis is important but challenging. We used two models, suckling mice protection assay and Stat1-deficient (Stat1 2/2 ) mouse model with different DENV-2 strains through intracerebral or intraperitoneal inoculation to evaluate the neutralizing activity of DB32-6 mAb (Figures 2 and 6 ). Our findings suggested that mAb DB32-6 might effectively block virus entry. However, disease manifestation of suckling mice is not relevant to dengue disease in humans since DENV infections in humans rarely involve the nervous system [58] . The Stat1-deficient mice are genetically mutated and not immunocompetent, hence they are not representative of the wild types' immune response to DENV. However, their survival rates might reflect the therapeutic potential of these mAbs. The results from these mouse models showed that the therapeutic potential of this newly generated mAb DB32-6 is worth further investigation. In the absence of an effective dengue vaccine, neutralizing antibodies can be used as a passive immunotherapeutic strategy for treating dengue. Previous studies of humanized antibodies against DENV were derived from two chimpanzee Fab fragments: humanized IgG1 1A5 cross-neutralizing DENV-1 and DENV2 and humanized IgG1 5H2 specific against DENV-4 [42, 48, 56] . Our newly generated hDB32-6 was derived from murine mAb. However, when developing antibody-based therapy, ADE phenomenon is a major concern. Modification of Fc structure in an antibody can prevent Fcc receptors binding and inhibit ADE (Figure 7 ) [39, 48] . Our studies show that the serotype-specific mAbs targeting the A-strand of E-DIII could serve as a dramatic neutralization determinant. Through testing in different mouse models, we have successfully generated a mAb hDB32-6 variant with high therapeutic potential against diverse DENV-2 strains without inducing ADE. Such an antibody-based therapy may help control severe dengue in the future. Figure S4 Identification of mAb 3H5 neutralizing epitopes by VLP mutants. BHK-21 cells expressed various DENV-2 VLP mutants. After fixation and permeabilization, mAbs were incubated with the cells. Binding activity was assessed Figure 6 . mAbs, mDB32-6 and hDB32-6-48, protected against DENV-2-induced mice mortality. Two-day-old suckling mice (ICR strain) were injected intracranially (i.c.) with 1610 4 pfu of DENV-2 (16681). After 1 day of infection, 5 mg of mAb were passively injected into mice through an i.c. route. Log rank test was used to determine significant differences in survival rate, and the mAbs with neutralizing activity were compared to the control group treated with NHIgG, P,0.001. doi:10.1371/journal.pntd.0001636.g006 Figure 7 . Antibody-mediated enhancement of DENV-2 infection by mAbs. Serial dilutions of NMIgG, 4G2, mDB32-6 and hDB32-6 variant were incubated with DENV-2 (16681) at MOI of 1 at 4uC for 1 h before they were added to K562 cells. After 2 days infection, cells were fixed, permeabilized, and stained with mAb DB42-3, and the percentage of cells infected with DENV-2 was detected by flow cytometry. doi:10.1371/journal.pntd.0001636.g007 Humanized mAb against Dengue Virus www.plosntds.org by flow cytometry. The fluorescence intensities were quantified to determine the relative recognition, calculated as [intensity of mutant VLP/intensity of WT VLP] (recognized by a mAb)6[intensity of WT VLP/intensity of mutant VLP] (recognized by mixed mAbs). Data shown are one representative experiment out of three independent experiments. (DOC) Figure S5 Sequence alignment of different DENV-2 genotypes and highlights of the neutralizing epitopes in E-DIII. The sequence of E-DIII from DENV-2 (strain 16681, Southeast Asian genotype) is aligned with other DENV-2 genotypes including NGC (Southeast Asian), PL046 (Southeast Asian), PM33974 (West African) and IQT2913 (American). Black blocks show residues of genotypic variation. The serotype-specific neutralizing epitopes located in E-DIII are K310 (green) and E311 (purple) which are recognized by DB32-6 and DB25-2, respectively. (DOC)

CD8(+) T Cells in Leishmania Infections: Friends or Foes?

Host protection against several intracellular pathogens requires the induction of CD8(+) T cell responses. CD8(+) T cells are potent effector cells that can produce high amounts of pro-inflammatory cytokines and kill infected target cells efficiently. However, a protective role for CD8(+) T cells during Leishmania infections is still controversial and largely depends on the infection model. In this review, we discuss the role of CD8(+) T cells during various types of Leishmania infections, following vaccination, and as potential immunotherapeutic targets.

CD8 + T cells play a major role in protective immunity to a wide variety of pathogens, including viruses, bacteria, and protozoan parasites. However, the protective role of CD8 + T cells during Leishmania infections has been controversial, mainly because of the discrepancy among infections with different Leishmania species. Different Leishmania species have different tropisms and their diversity is reflected in the various clinical manifestations they induce. Hence, it is not surprising that the contribution of CD8 + T cells to the immune response against the parasite depends on the clinical form and the species that is causing it. Here, we discuss the literature on the contribution of CD8 + T cells to the immune response against Leishmania, taking into account the various clinical forms and experimental models. CD8 + T cells recognize peptides that are presented in the context of major histocompatibility complex (MHC) class I molecules via the T cell receptor (TCR). Although peptides presented via MHCI mainly derive from endogenous antigens, various exogenous cellassociated antigens have also been shown to be uploaded onto the MHCI pathway, by a process referred to as cross-presentation. Leishmania antigens were also shown to be cross-presented (Bertholet et al., 2006) . During in vivo infections, cross-presentation of Leishmania antigens may result from several internalization pathways, such as direct infection, receptor-mediated uptake (Woelbing et al., 2006) , or internalization of apoptotic vesicles (Winau et al., 2006) . Thus far, two different processing pathways have been proposed. An early work demonstrated that a surface antigen of L. amazonensis was processed in a proteasome-dependent manner within the cytosol (Kima et al., 1997) . In contrast, a more recent study showed that cross-presentation of secreted leishmanial antigens is confined to an intraphagosomal processing pathway that is TAP-and proteasome-independent (Bertholet et al., 2006) . After activation, antigen-specific CD8 + T cells differentiate into effector cells and acquire the capacity to kill target cells, and produce several cytokines and chemokines (Kaech et al., 2002; Harty and Badovinac, 2008) . Among the various CD8 + T cell subsets, Tc1 were shown to play a major role in the fight against several protozoan parasites (Jordan and Hunter, 2010) . The hallmark of this subset is the production of IFN-γ and TNF, and cytotoxic capacity (Woodland and Dutton, 2003) . The precise mechanism underlying cytotoxic T lymphocyte (CTL) killing of microbes is still poorly understood. CTLs can exert cytotoxicity through various mechanisms: via exocytosis of lytic granula containing perforin, granzyme A/B, and/or granulysin; through the interaction between FasL and Fas expressed on targets cells; via TNF; or via TRAIL (Trapani and Smyth, 2002) . A study has also shown that reactivated memory CD8 + T cells efficiently killed Listeria monocytogenes via a mechanism mediated by CCL3 and involving the induction of radical oxygen intermediates (Narni-Mancinelli et al., 2007) . Direct killing of extracellular pathogen by CTLs has also been described. For example, CTLs can mediate killing of Mycobacterium tuberculosis through the release of anti-bacterial products (Stenger et al., 1998; Canaday et al., 2001) . However, direct killing of Schistosoma mansoni (Ellner et al., 1982) and Entamoeba histolytica (Salata et al., 1987) is thought to be contact-dependent. CTLs have also been reported to directly kill extracellular Toxoplasma gondii (Khan et al., 1990) . Interestingly, killing in this case appeared to be antigen-specific. To date there is no evidence that CD8 + T cells can mediate protection against Leishmania parasites through their cytotoxic activity. However, since CTLs have been www.frontiersin.org observed in various mouse models and also in human patients, a possible protective role for Leishmania-specific cytotoxic T cells should not be excluded. In addition to killing and releasing cytokines and chemokines, recent studies have ascribed a novel regulatory role for CD8 + T cells (Sun et al., 2009; Palmer et al., 2010; Trandem et al., 2011) . Regulatory CD8 + T cells represent a transient state of effector CD8 + T cells (Trandem et al., 2011) , which is possibly induced by potent TCR stimulation, which promotes the production of the immunosuppressive cytokine IL-10 (Zhang and Bevan, 2011) . Not only do these cells produce IL-10, but they are also excellent killers and produce normal to higher amounts of IFN-γ and TNF (Sun et al., 2009; Palmer et al., 2010; Trandem et al., 2011) . The main function of these cells is thought to lie in the prevention of immunopathology during infection without affecting the kinetics of pathogen clearance. IL-10-producing CD8 + T cells have also been observed in human patients infected with L. guyanensis (Bourreau et al., 2007) and in patients suffering from post-kalaazar dermal leishmaniasis (PKDL; Ganguly et al., 2008) . The role of regulatory CD8 + T cells in the immune response against parasitic infections is still unknown. The role of CD8 + T cells in the immunity to L. major has always been controversial. Early studies in BALB/c mice reported that CD8 + T cells were the main mediators of protection following CD4 + T cell depletion in mice infected with L. major (Titus et al., 1987; Hill et al., 1989; Muller et al., 1991) . Interestingly, depletion of CD4 + T cells was rendering susceptible BALB/c mice resistant to L. major infection. The results obtained using the CD4 + T cell depletion model suggested that CD8 + T cells could potentially control L. major infection in mice. A few years later, experiments in β2-microgobulin-deficient mice contradicted these findings and revealed that CD8 + T cells were not essential in mediating protection in L. major-infected BALB/c mice (Wang et al., 1993) . Moreover, a study in CD8 + T cell-deficient mice demonstrated that Cd8 −/− mice were able to control L. major infection for at least 1 year, suggesting that CD8 + T cells were not required for long-lasting immunity (Huber et al., 1998) . The contribution of CD8 + T cells in the control of primary L. major infection became less important also because of the strong evidence that Th1 cells were the primary cells involved in mediating protection against cutaneous leishmaniasis (Reiner and Locksley, 1995; Louis et al., 1998; Sacks and Noben-Trauth, 2002) . Several studies have demonstrated that Th1 cells producing IFN-γ were essential in controlling L. major infection, and that failure to develop a Th1 response resulted in susceptibility to the diseases. Hence, a consensus was reached in that if a mouse generates Th2 responses, this will lead to susceptibility; in contrast, Th1 responses were successfully controlling infection without the help of CD8 + T cells. This paradigm was later challenged when new findings arose from a more natural model of infection, where 100 metacyclic promastigotes were inoculated intradermally in the ears of C57BL/6 mice. In this model, Cd8 −/− and CD8 + T cell-depleted mice fail to control L. major infection, and CD8 + T cells were thought to be necessary for supporting Th1 responses (Belkaid et al., 2002) . The discrepancy between the findings in the low-and the highparasite dose model was clarified by another work that compared the requirements of CD8 + T cells in both systems (Uzonna et al., 2004) . Interestingly, in the low infection model CD8 + T cells producing IFN-γ were essential for modulating CD4 + T cell responses toward a Th1 response. In contrast, C57BL/6 mice inoculated with a high L. major dose did not require CD8 + T cell help to generate protective Th1 responses. The CD8 + T cell requirement for optimal IFN-γ production by Th1 cells was also proposed in a high-dose L. major infection model in BALB/c mice (Herath et al., 2003) . Moreover, CD8 + T cell-derived IFN-γ was reported to contribute to the induction of nitric oxide production in macrophages during experimental cutaneous leishmaniasis (Stefani et al., 1994) . Although the role of CD8 + T cells-derived IFN-γ has been clarified, little is known about the involvement of cytotoxic CD8 + T cells in cutaneous leishmaniasis. In a low-dose model of L. major infection, CD8 + T cell responses were shown not only to be protective, but also to mediate pathology (Belkaid et al., 2002) . Hence, it is possible that CTLs may be involved in the ulceration of skin lesions through tissue disruption. This suggests that perhaps two types of CD8 + T cell effectors are generated during L. major infection: antigen-specific CD8 + T cells that produce IFN-γ but lack cytotoxic activity; and CTLs that are potent killers but produce little to no IFN-γ and promote pathology. Although the role of CD8 + T cells during primary immune responses is controversial, these cells appear to play a prominent role in protecting mice from a secondary challenge (Muller et al., 1993 . Indeed, antigen-specific CD8 + T cells were expanding up to 50-fold in the spleen and lymph nodes of reinfected BALB/c mice . This expansion correlated with a substantial production of IFN-γ, which is thought to be essential for controlling Leishmania infections. These observations have major implications for vaccine design. In summary, during experimental cutaneous leishmaniasis, CD8 + T cells are necessary to support protective Th1 responses through IFN-γ production, but they are also involved in the development of immunopathology. Further investigations are needed to better identify various subtypes of CD8 + T cells that arise during cutaneous leishmaniasis. In contrast to the cutaneous models, CD8 + T cells have always been thought to play a major role in experimental visceral leishmaniasis (VL). Over 20 years ago, Stern et al. (1988) demonstrated for the first time that CD8 + T cells significantly contribute to the formation of granulomas in the liver of L. donovani-infected mice. Indeed, CD8 + T cell depletion resulted in impaired granuloma formation and exacerbation of liver disease. In agreement with these results, Kaye et al. (1992) also reported a delayed onset and a decrease of the liver granulomatous response in non-obese, diabetic mice expressing transgenic I-E molecules, suggesting that antigen-specific CD8 + T cells are required for proper granuloma formation. CD8 + T cells appear to participate in controlling parasite growth in the spleen as well, since CD8 + T cell depletion during chronic VL significantly increased splenic parasite burden (Stäger, unpublished) . This observation was underscored by the fact that adoptive transfer of antigen-specific CD8 + T cells during Frontiers in Immunology | Microbial Immunology chronic L. donovani infection resulted in 90% reduction in the splenic parasite burden (Polley et al., 2006) . Moreover, therapeutic vaccination aimed at reactivating CD8 + T cells during chronic VL ensued in the control of parasite growth in the spleen (Joshi et al., 2009) . Interestingly, CD8 + T cells do not only participate in primary responses to L. donovani, but are also the major mediators of resistance upon reinfection (Stern et al., 1988) . Indeed, protection was abrogated following CD8 + T cell but not CD4 + T cell depletion. A prominent function for CD8 + T cells was also described in L. infantum-infected mice. Using an intradermal infection model, Ahmed et al. (2003) demonstrated that CD8 + T cells contribute to parasite clearance in the skin of L. infantum-infected mice. Another study also showed that CD8 + T cells purified from L. infantum-infected mice expressed IFN-γ and TNF, and displayed considerable cytotoxic activity against cells expressing Leishmania antigens (Tsagozis et al., 2003) . Interestingly, killing of infected target cells was mediated by both the perforin and Fas/FasL pathways (Tsagozis et al., 2003) . The Fas/FasL pathway has also been implicated in the defense against L. donovani (Alexander et al., 2001) . Indeed, gld and lpr mice, which lack a functional Fas/FasL pathway, were shown to be more susceptible to L. donovani. Additionally to the classical cytotoxic pathways, a novel counterregulatory function for a subset of cytotoxic CD8 + T cells has recently been proposed in the L. donovani infection model (Martin et al., 2010) . In this study, CD3 + CD8 + CD40 + T cells are shown to suppress regulatory T cells via CD40/CD40L interaction during the early stages of infection in BALB/c mice. CD40 signals through Ras, PI3K, and protein kinase C, leading to the induction of granzyme and perforin, and ultimately to the killing of Tregs. CD8 + T cells may not be merely participating in the primary immune response by secreting IFN-γ and possibly killing infected target cells and/or Tregs, but they could also be involved in the recruitment of inflammatory cells and in the maintenance of granulomas. Indeed, a study using L. infantum demonstrated that CD8 + T cells expressed RANTES and MIP-1α (Tsagozis et al., 2003) , two chemokines that are involved in the recruitment of T cells at the inflammatory site and in the formation and maintenance of granulomas (Mackay, 2001) . The authors proposed that CD8 + T cells may thus be involved in granuloma formation. This hypothesis is in agreement with the depletion data (Stern et al., 1988) , showing that depletion of CD8 + T cells results in impaired granuloma formation and ultimately in disease exacerbation. Despite the documented evidence that CD8 + T cells strongly participate in the immune response to L. donovani and L. infantum, our recent findings suggest that L. donovani induces defective antigen-specific CD8 + T cell responses (Joshi et al., 2009 ). Interestingly, mice infected with L. donovani generate CD8 + T cell responses with limited clonal expansion. The extension of the clonal expansion is thought to be correlated with the effectiveness in eliminating pathogens. It was calculated that a naïve CD8 + T cell may go through 19 cell divisions in the first week after pathogen inoculation (Badovinac et al., 2007) . Massive clonal expansions have not only been observed during viral infections, but also following the injection of irradiated Plasmodium berghei sporozoites (Sano et al., 2001) . During L. donovani infection, CD8 + T cells underwent at least 8-9 rounds of division, but failed to accumulate in the spleen (Joshi et al., 2009) . Moreover, only 10% of CD8 + T cells during clonal expansion expressed markers typically associated with end-differentiated effector cells, such as KLRG1, unpublished) . The cause of this limited expansion is yet unknown and may depend on several factors. One of the possible explanations is limited antigen availability that may result from poor antigen-processing and presentation. Processing of Leishmania antigens is thought to be confined to a TAP-independent, intraphagosomal pathway that is less efficient and requires higher amounts of antigen than the endoplasmic reticulum-based, TAP-dependent cross-presentation pathway (Bertholet et al., 2006) . Furthermore, the major surface protease of Leishmania, gp63, was shown to cleave epitopes within the parasitophorous vacuole, further reducing antigen availability (Garcia et al., 1997) . Hence, Leishmania antigens may be poorly presented and this poor presentation may not be enough to induce and sustain a massive clonal expansion of antigen-specific CD8 + T cells. Nonetheless, antigen may be suddenly available in large amounts later on during L. donovani infection, since CD8 + T cells undergo a second round of activation, become dysfunctional, and ultimately die by "exhaustion" (Joshi et al., 2009) ; high antigen levels have been described as a cause of CD8 + T cell "exhaustion" during chronic viral infections (Mueller and Ahmed, 2009 ). Further research is needed to clarify the mechanisms involved in CD8 + T cell exhaustion during chronic VL. In conclusion, CD8 + T cells are required to control parasite growth during experimental VL and reactivation of these responses results in a dramatic reduction in parasite burden. Therefore, immune interventions that target CD8 + T cell responses may have great therapeutic potential against VL. The role of CD8 + T cells in human leishmaniasis patients is still unclear and seems to depend on the various species of parasites and the disease they cause. Few studies have been conducted with human VL patients. However, most of the studies ascribe a protective role for CD8 + T cells, in agreement with results obtained from experimental models. Indeed, the control of L. infantum infection was shown not only to be associated with IFN-γ-producing CD4 + T cells, but also with CD8 + T cells (Mary et al., 1999) . Interestingly, during active VL, CD8 + T cells are less responsive to stimulation and a greater percentage stains positive for Annexin V compared to healthy controls (Clarencio et al., 2009 ). These observations correlate very well with what we observed in mice experimentally infected with L. donovani, where CD8 + T cells became increasingly dysfunctional during chronic infection and died by exhaustion (Joshi et al., 2009) . Whether human CD8 + T cells also display signs of exhaustion during active VL still remains to be tested. Another study investigating CD8 + T cell responses in patients infected with L. chagasi has revealed that the frequency of CD18 + CD45RO + CD8 + T cells is significantly decreased in the spleen of patients with active VL (Clarencio et al., 2009 ). In contrast, CD18 + CD8 + T cells seem to be retained in the bone marrow of VL patients. CD18, or integrin β-2, is the β subunit of LFA-1 (CD11a), CD11b, CD11c, and CD11d. In humans, lack of CD18 www.frontiersin.org causes leukocyte adhesion deficiency, a disorder characterized by lack of leukocyte extravasation from blood into the tissue (Bunting et al., 2002) . With exception of the fact that CD18 + cells appear in the granulomas of dogs with asymptomatic VL (Sanchez et al., 2004) , very little is known about the function of CD18 + CD8 + T cells during VL and whether cells lacking CD18 expression have similar migratory capacity and effector functions to their CD18 + counterparts. Not only is the frequency of CD18 + CD8 + T cells reduced in L. chagasi patients, but also, the level of circulating memory T cells is significantly decreased during active VL (Hailu et al., 2005; Clarencio et al., 2009 ). This observation is in agreement with our findings in the experimental model of VL, where the majority of the CD8 + T cells displayed an effector phenotype during chronic infection (Joshi et al., 2009) . Although CD8 + T cells positively correlate with cure of VL patients, one report suggested that these cells may contribute to the immunopathogenesis of PKDL (Ganguly et al., 2008) . Indeed patients suffering from PKDL showed a significant increase in the percentage of CD8 + T cells producing IL-10, which disappeared after cure (Ganguly et al., 2008) . IL-10-secreting CD8 + T cells are thought to play a regulatory role in different viral infection models. This CD8 + T cell subset was shown to display great cytotoxicity and produce granzyme B, IFN-γ, and TNF (Sun et al., 2009; Palmer et al., 2010; Trandem et al., 2011) . IL-10 + CD8 T cells seem to represent a transient and reversible state of CD8 + effector T cell differentiation. Its primary function is to balance pathogen clearance with bystander tissue damage (Zhang and Bevan, 2011) . Interestingly, in viral model, this subset disappears after the infection is cleared. Hence, it is possible that the IL-10-producing CD8 + T cells in PKDL patients are actually killing parasites and protecting patients from tissue damage, rather than suppressing protective responses. Further studies are needed in order to define the nature of these cells. CD8 + T cells also actively participate in the immune response to cutaneous infections in human. As observed in the low-dose model in mice (Belkaid et al., 2002; Uzonna et al., 2004) , L. major also induces Th1 and CD8 + T cells in human patients and both responses are associated with disease resolution (Nateghi Rostami et al., 2010) . CD8 + T cells were not only observed in large numbers in the lesions of L. major patients during the acute phase, but also during the healing process (Da-Cruz et al., 1994 Gaafar et al., 1999) . The exact role of CD8 + T cells in L. major infections in humans is not yet known. A major correlate of protection appears to be the high amounts of IFN-γ produced by CD8 + T cells after restimulation (Nateghi Rostami et al., 2010) . In vitro studies have also demonstrated that Leishmania-specific CTLs are generated upon co-culturing human naïve T cells with antigens from L. amazonensis promastigotes and IL-12 (Russo et al., 1999) , or with L. major parasites (Da . Moreover, increased granzyme B activity was also found in patients with an active infection and was associated with a good prognosis (Bousoffara et al., 2004) . In this study, in vitro cytotoxicity by peripheral blood lymphocytes on L. major-infected macrophages appeared to be mediated by granzyme B, suggesting that CTL activity may be involved in controlling parasite growth. It is possible, though, that the cytotoxic activity not only contributes to disease clearance, but also to the development of skin ulceration, as observed in L. major-infected mice (Belkaid et al., 2002) . A strong CD8 + T cell expansion has also been observed in L. mexicana patients during the healing process (Salaiza-Suazo et al., 1999) . Interestingly, lesions of patients with localized cutaneous leishmaniasis (LCL) harbor a higher number of CD8 + T cells compared to patients with diffuse cutaneous leishmaniasis (DCL; Hernandez-Ruiz et al., 2010) . As already observed in VL patients, CD8 + T cells in DCL patients, unlike LCL patients, show a reduced capacity to respond to antigen-specific stimulation during active infection. In fact, these cells displayed low cytotoxicity and only produced little IFN-γ upon stimulation, therefore showing typical signs of functional exhaustion (Hernandez-Ruiz et al., 2010) . Strikingly, effector functions could be restored in vitro after stimulation with TLR2 agonists, highlighting the potential therapeutic benefit of the revival of CD8 + T cell functions in DCL patients. In contrast to the cutaneous and visceral forms of leishmaniasis -where CD8 + T cells seem to correlate with cure and contribute to the immune response -in mucocutaneous infections (ML) CD8 + T cells seem to be implicated in the pathogenesis of the disease. Indeed, high numbers of cytotoxic CD8 + T cells were observed in ML patients (Barral-Netto et al., 1995; Brodskyn et al., 1997) . Moreover, the recruitment of granzyme A + CD8 + T cells is associated with lesion progression (Faria et al., 2009) , suggesting that CTLs may contribute to immunopathology. The development of ML is not only associated with the presence of CTL, but also with a high frequency of activated CD4 + T cells, an extreme IFNγ and TNF production, and a reduced control of inflammation due to low expression of the IL-10 receptor (Gaze et al., 2006; Faria et al., 2009) . Furthermore, IL-17-secreting CD4 + and CD8 + T cells were also found in ML patients (Boaventura et al., 2010) . Consequently, neutrophils, which are typically recruited during a TH17-mediated inflammatory response, were also detected in necrotic and perinecrotic areas (Boaventura et al., 2010) . This suggests that neutrophils, together with CTLs, may be involved in tissue injury and in the development of immunopathology. Taken together, the literature shows that CD8 + T cells actively participate in the fight against most Leishmania infections in humans and their presence correlates with cure. In contrast, CD8 + T cells in ML patients contribute to disease exacerbation. Vaccination of humans with heat-killed Leishmania or recombinant parasite proteins has so far failed to induce long-term immunity and only recovery from natural or experimental infection has provided proper protection. Several trials have analyzed the protective effect of autoclaved L. major plus Bacillus-Calmette-Guérin (BCG) versus BCG alone assessing the cumulative incidence of cutaneous leishmaniasis caused by L. tropica (Sharifi et al., 1998) or L. major (Momeni et al., 1999) , or of VL (Khalil et al., 2000) caused by L. donovani. Although no trial showed a significant effect on disease incidence, the vaccination induced skin test conversion and provided limited protection. Additionally, a study showed that immunization of Colombian soldiers with three doses of L. amazonensis alone was non-protective (Velez et al., 2005) . In the human disease, there is evidence that mixed T helper cytokine profiles are present, while healing and protection against reinfection are associated with dominant Th1 and CD8 + T cells. These findings suggest that it is the cytokine balance that activates or suppresses activation of macrophages harboring Leishmania parasites. This, in turn, determines the outcome of the infection. Thus treatments or antigen/adjuvant formulations that can alter the type of T helper response may change the course of disease (Da-Cruz et al., 2002; Rogers and Titus, 2004; Mohajery, 2007) . For this purpose, different vaccination strategies have been examined in animal models including leishmanization (Modabber, 1990) , killed parasite (Grimaldi, 1995) , live attenuated parasite (Titus et al., 1995) , and subsequently, subunit vaccines composed of recombinant or native proteins from different stages of the parasite's life cycle, and DNA vaccines (Skeiky et al., 1998; Webb et al., 1998; Stager et al., 2000; Bottrel et al., 2001; Campos-Neto et al., 2001; Rafati et al., 2001; Coler et al., 2002) . The latter two strategies encompass candidates such as gp63, gp46, LACK, CPB, CPA, Kmp11, LmsTI1, TSA, LeIF, HASPB1, and LPG, and have shown promising results in murine models. Nonetheless, only Leish111f (a recombinant fusion protein of LmsTI1, TSA, and LeIF) progressed through phase I and II clinical trials (Llanos-Cuentas et al., 2010; Chakravarty et al., 2011) . Nowadays, it is clear that CD8 + T cells play an important role in the mechanisms for cure of and resistance to Leishmania infection, either by production of IFN-γ and activation of macrophages, or by direct killing of parasitized macrophages, or a combination of both effects. CD8 + T cells have been associated with protection against Leishmania reinfection in murine models; however, the induction of these T cell subsets in humans seems to be also related to the healing process. Today, there are several reports about different leishmanial antigens eliciting CTL responses such as P8, gp46 , HASPB1 (Stager et al., 2000) , Kmp11 (Basu et al., 2007) , CPB (Rafati et al., 2002) , nucleosomal histones (Iborra et al., 2004) , LmaCIN (Farajnia et al., 2005) , LmsTI1, and TSA (Coler et al., 2002) . The essential point to be considered in vaccine design for a heterogeneous population, such as that of humans, is the HLA polymorphism. Effective vaccination against a complex parasitic infection such as Leishmania would require a multivalent vaccine composed of several antigens to enhance the possibility of covering a good number of MHC types. This is possible either through recombinant fusion proteins encompassing the whole antigen or through vaccines composed of peptides from different antigens (Campos-Neto et al., 2001; Rafati et al., 2001; Mendez et al., 2002) . The latter strategy, called polytope vaccine, is finding its way in vaccinology because of its extraordinary properties, especially the ability to direct the immune response toward the induction of CTLs (Sbai et al., 2001; Schirmbeck et al., 2003; Robinson and Amara, 2005) . As CTL responses play a pivotal role in defense against viruses and tumor cells, polytope vaccines have found their way in these fields but there is still no report on leishmaniasis even it has been shown that CTLs could be very important in protection and long-lasting resistance to infection. Recently, we took advantage of the potential of immunoinformatics tools to screen for L. major epitopes that could be presented in HLA A2, which is the most prevalent HLA supertype in the Iranian population. In vitro stimulation to recall memory CD8 + T cells from Leishmania-infected individuals and intracellular cytokine assays for IFN-γ-producing cells confirmed that HLA A2 positive individuals that recovered from an L. major infection successfully generated CD8 + T cell responses against peptides derived from LmsTI1 and LPG-3 (Seyed et al., 2011) . Furthermore, Walden and co-workers have mapped the T cell epitopes from kinetoplastid membrane protein of L. major (Kmp11) via classical mapping for different human HLA class I alleles (Basu et al., 2007) . Gazzinelli and co-workers have studied CD8 + T cell responses against the Leishmania A2 antigen and mapped the CD8 + T cell epitopes in BALB/c mice (Resende et al., 2008) . Laouini and co-workers (Guerfali et al., 2009 ) and Dumonteil and co-workers (Dumonteil, 2009 ) started genomewide screenings for novel epitopes. Using a combination of T cell epitope prediction tools, they successfully validated such epitopes in both BALB/c and C57BL/6 mice. Although understudied, CD8 + T cells appear to play an important role in the immune response to most Leishmania infections. Pilot studies in the murine model of VL have also demonstrated that adoptive transfer of antigen-specific CD8 + T cells (Polley et al., 2006) or reactivation of CD8 + T cell responses through a therapeutic vaccine (Joshi et al., 2009) results in the control of parasite growth. A better understanding of the mode of activation, the specificity, and effector functions of the various CD8 + T cell subsets generated during Leishmania infections could ameliorate the design of vaccines and of novel therapeutic interventions. for the early control of parasite burden in the liver of Leishmania donovani-infected mice. Eur. J. Immunol. 31, 1199-1210. Badovinac, V. P., Haring, J. S., and Harty, J. T. (2007) . Initial T cell receptor transgenic cell precursor frequency dictates critical aspects of the CD8(+) T cell response to infection. Immunity 26, 827-841. Barral-Netto, M., Barral, A., Brodskyn, C., Carvalho, E. M., and Reed, S. G. (1995) . Cytotoxicity in human mucosal and cutaneous leishmaniasis. Parasite Immunol. 17, 21-28. Basu, R., Roy, S., and Walden, P. (2007). HLA class I-restricted T cell epitopes of the kinetoplastid membrane protein-11 presented by Leishmania donovani-infected human macrophages. J. Infect. Dis. 195, 1373 -1380 S., Lira, R., Caler, E., Bertholet, S., Udey, M. C., and . CD8+ T cells are required for primary immunity in C57BL/6 mice following low-dose, intradermal challenge with Leishmania major. J. Immunol. 168, 3992-4000. www.frontiersin.org

Macroevolutionary Immunology: A Role for Immunity in the Diversification of Animal life

An emerging picture of the nature of immune systems across animal phyla reveals both conservatism of some features and the appearance among and within phyla of novel, lineage-specific defense solutions. The latter collectively represent a major and underappreciated form of animal diversity. Factors influencing this macroevolutionary (above the species level) pattern of novelty are considered and include adoption of different life styles, life histories, and body plans; a general advantage of being distinctive with respect to immune defenses; and the responses required to cope with parasites, many of which afflict hosts in a lineage-specific manner. This large-scale pattern of novelty implies that immunological phenomena can affect microevolutionary processes (at the population level within species) that can eventually lead to macroevolutionary events such as speciation, radiations, or extinctions. Immunologically based phenomena play a role in favoring intraspecific diversification, specialization and host specificity of parasites, and mechanisms are discussed whereby this could lead to parasite speciation. Host switching – the acquisition of new host species by parasites – is a major mechanism that drives parasite diversity and is frequently involved in disease emergence. It is also one that can be favored by reductions in immune competence of new hosts. Mechanisms involving immune phenomena favoring intraspecific diversification and speciation of host species are also discussed. A macroevolutionary perspective on immunology is invaluable in today’s world, including the need to study a broader range of species with distinctive immune systems. Many of these species are faced with extinction, another macroevolutionary process influenced by immune phenomena.

Recent years have witnessed a dramatic increase in our understanding of the diversity of immune systems across animal phyla (Flajnik and Kasahara, 2010; Messier-Solek et al., 2010; Rast and Litman, 2010; this volume) . Availability of genome sequences from a broad variety of animals coupled with an increased appreciation for the diversity of their defenses has given the study of immunity a much stronger evolutionary foundation, one that has been further enriched by studies of plant immunity and responses of bacteria and archaea to threats to their genomes (Horvath and Barrangou, 2010) . The increasing depth and breadth of immunological studies is also bringing to light a greater awareness of the impact that immunity has had on all forms of life, especially parasites. Here "parasite" is used inclusively, referring to infectious agents ranging from viruses to bacteria to protists to multicellular helminths. The features uniting parasites are that they infect hosts, provoke some degree of fitness-diminishing harm, prompt the deployment of immune responses, and undertake immune evasive actions. "Immune systems" are referred to as those molecules, cells, tissues, and organs that protect hosts from parasites (see caveats below). This discussion excludes a broad range of behavioral defenses like preening (Bush and Clayton, 2006) or avoidance (e.g., Mooring et al., 2003; Garnick et al., 2010) . Here I attempt to draw together ideas that begin to put immunological phenomena into a broader macroevolutionary context. Macroevolution is the study of patterns, and the evolutionary processes that have generated them, at or above the species level (Stanley, 1998; Levinton, 2001) . It is the study of how and why life has diversified, and attempts to document how and why lineages of organisms have come into being and either given rise to additional lineages or gone extinct. The process of speciation is germane to macroevolutionary studies because it is the process responsible for increasing the diversity of life forms. Extinction and its causes are also an essential part of such studies. The attributes of immune systems across the spectrum of animal diversity provide a new way to view and reinterpret the diversity of animals. Immune systems exhibit unforeseen novelty and thus offer new insights into major selective forces influencing animal life. Also, phenomena that are fundamentally immunological provide fertile ground for investigating the impact of immunity as a driver of biodiversity. The role of immune systems in macroevolutionary processes is one that deserves recognition and more study. In considering what is to follow, several caveats should be borne in mind: (1) We are just beginning to view molecular components of immune systems from a broad sampling of animal phyla. Detailed analyses are still few for how these presumptive immune components actually function in defense, and how critical their roles might be in protecting the organisms in question. (2) Also poorly known are the specific parasite threats faced by the more obscure groups of animals serving as hosts. (3) Many of the examples of immunological novelty presented below emphasize differences at the phylum level. Some phyla such as the Arthropoda, Mollusca, and Chordata are immense in species numbers and undoubtedly collectively employ as yet many undiscovered immune capabilities. Also, some of the smaller animal phyla are essentially unexplored with respect to their immune systems. Once understood, these additional examples will only add to the overall diversity of immune responses. (4) It is not always easy to circumscribe "the immune system" or an "immune response." This is particularly so in cases where potent defenses for parasites result from selection for variant alleles for genes like hemoglobin B or apolipoprotein L-1 that otherwise might not be considered a core part of the immune system (Anstee, 2010; Barreiro and Quintana-Murci, 2010; Genovese et al., 2010; Wheeler, 2010) . Discoveries relating to the innate immune systems of plants, flies, and mammals have tended to accentuate the similarities among them, implying a grand conservatism even across kingdoms with respect to basic immune system design and function. Indeed, there are intriguing similarities between the membraneassociated and intracytoplasmic receptors of plants and animals suggestive that some basic solutions to recognition and response to parasites have been conserved since at least the time animal and plant lineages diverged. However, particularly given that some of these similarities are a likely result of convergent evolution rather than indicative of a common origin (Ausubel, 2005) , conserved immune features are not the emphasis here. Rather, this overview accentuates the emergence of immunological novelty among and within animal phyla (Figure 1 ; Table 1 ). The most basal animal group is the phylum Porifera, the sponges (Srivastava et al., 2010) . Sponges lack the complex tissue and organ structure found in other animal phyla, and lack cells specialized for protection from parasites. Although sponge immunobiology is in its infancy, one of the best-known sponges, Suberites domuncula, possesses membrane-spanning molecules that contain an intracellular Toll-interleukin 1 receptor (TIR) domain, though it lacks an external leucine-rich repeat pattern recognition receptor more typical of TLRs. On the basis of having a MyD88 homolog, S. domuncula has at least the rudiments of an NF-κB signaling pathway. Sponges also have molecules for attacking bacterial membranes, presumptive antiviral responses (Schroder et al., 2008) , and diversified scavenger receptor cysteine-rich molecules of unknown function (Wiens et al., 2007) . Among basal animals, it is members of the phylum Cnidaria (jellyfish, Hydra, anemones, and corals) that have proven most surprising with respect to the large size and content of their genomes, including their immune systems. Cnidarians have distinct tissues but lack organs and are generally considered to be diploblastic, meaning they have recognizable ecto-and endoderm, but lack well-developed mesoderm tissue. Like sponges, they lack recognizable specialized immune cells. However, the starlet sea anemone Nematostella vectensis has at least one TLR, an NF-κB signaling pathway, a homolog of a complement 3-like molecule, the likely presence of functioning intracellular NOD-like receptors (NLRs), perforin-like molecules, diverse C-type lectins (Wood-Charlson and Weis, 2009) , and even a recognizable homolog of the recombination activating gene, RAG1 Augustin et al., 2010) . Cnidarians often live in colonies and have to contend with encroaching competitors, including conspecifics. For this they have well-developed mechanisms to recognize self and non-self. One of the responsible molecules has been identified, and is surface expressed, polymorphic and possesses three external immunoglobulin superfamily (IgSF) domains (Nicotra et al., 2009) . The remaining animals, the Bilateria, are bilaterally symmetrical and triploblastic, with well-developed tissues and organs. They often have specialized immune cells. Most fall into two major lineages, the protostomes and deuterostomes. Among the protostomes, representatives of the molting clade (Ecdysozoa) have been most extensively studied in an immune context, as this clade includes nematodes and arthropods, both containing well-studied model organisms. Caenorhabditis elegans and other nematodes have reduced genomes and from an immunological perspective are surprising for what they do not have. Although C. elegans has one bonafide TLR that plays a role in defense against some bacteria (Tenor and Aballay, 2008) , it lacks Myd88, NF-κB and several other components of the canonical Toll pathway. NLRs are also lacking. Nonetheless, C. elegans can mount inducible, parasite specific responses. It has several novel signaling pathways for defense (Irazoqui et al., 2010) and produces many distinctive antimicrobial peptides (AMPs) for protection from bacteria (Roeder et al., 2010) . C-type lectins may serve as recognition molecules in C. elegans. The preoccupation with production of AMPs by gut cells reflects their diet of bacteria, which could include potential parasites (Roeder et al., 2010) . Another prominent model of ecdysozoan immunity is Drosophila (Lemaitre and Hoffmann, 2007) , but increasingly other insect are studied as well, such as mosquitoes because of their role in transmitting human parasites (Bartholomay et al., 2010) . Insects have dedicated immune cells such as plasmatocytes and lamellocytes that circulate through their open circulatory system and phagocytose or encapsulate foreign objects. The Drosophila immune system also shows evidence of gene loss: it lacks the C3-like complement component and NLR homologs found in cnidarians. Their TLRs are different from those of vertebrates in that they do not engage microbial ligands directly. Of their nine TLR genes, only one or two function in immunity, activating NF-κB signaling pathways in the fat body to produce AMPs (Lemaitre FIGURE 1 | An overview of some of the novel features associated with immune responses of representatives of major animal lineages (see text for details). TLR, Toll-like receptor; AMP, antimicrobial peptide; Dscam, Down syndrome cell adhesion protein; VCBPs, variable region-containing chitin-binding proteins; NLRs, intracellular NOD-like receptors; LRR, leucine-rich repeat; IgSF, immunoglobulin superfamily; Ab, antibodies; TCR, T cell receptor; MHC, major histocompatibility complex. and Hoffmann, 2007). Insects are by no means immunologically bereft though. They have a number of other effective defense components not seen in many other organisms. They have elaborate cascades of CLIP-serine proteases that mediate and coordinate phagocytosis, nodule formation, encapsulation, and AMP formation, and they can deposit layers of melanin around foreign objects (Kanost et al., 2004) . They engage multimeric fibrinogen-related proteins (FREPs) in parasite recognition (Dong and Dimopoulos, 2009 ) and employ Down syndrome cell adhesion molecule (Dscam), a member of the IgSF, in antigen recognition. Tens of thousands of Dscam isoforms can potentially be generated by alternative splicing (Schmucker and Chen, 2009 ) and parasite challenge-specific Dscam splice form repertoires can be produced (Dong et al., 2006) . Insect studies provide additional examples of immunological novelty, at the ordinal, family, or even genus level. One example is provided by Drosophila and Anopheles, both in the same order (Diptera), but representing very different life styles and having been separate lineages for 250 million years. Gene families involved in immunity have evolved rapidly and divergently in the two dipterans. For example, with respect to thioester containing proteins (TEPs), Anopheles has 10 genes and Drosophila only four, with only one orthologous pair between the two. Anopheles has 58 fibrinogen-like immunogenes whereas Drosophila has only 14, with only two shared orthologous pairs (Dong and Dimopoulos, 2009) . At the family level, a comparison of three different mosquito genera (Aedes, Anopheles, and Culex, all in the Culicidae) has revealed prominent genus specific expansion of some immune gene families (Bartholomay et al., 2010) . Comparative studies of Drosophila species are particularly revealing, showing that novel immune genes and immune gene families have originated relatively recently, suggestive of a role of parasites in driving adaptive evolution in flies (Sackton et al., 2007) . Furthermore, for particular immune proteins, the amino acids under positive selection vary between Drosophila species groups, suggesting different fly species experience different parasite pressures (Morales-Hojas et al., 2009) . Insects with very different life styles, such as the social honey bees (Evans et al., 2006) and ants (Smith et al., 2011) , and symbiont-dependent aphids (Pennisi, 2009 ) likewise have immune systems that are surprisingly divergent from the Drosophila immune system. The other major lineage of protostomes, the Lophotrochozoa, includes prominent groups such as the flatworms, annelids, and mollusks. In the polychete annelid Capitella capitata, TLRs have undergone an expansion to over 100 genes, most of which are similar, suggestive of recent duplication. Another annelid, the leech Helobdella robusta, has only 16 TLR homologs which are not only highly divergent from one another but also are not orthologous to any of the polychete sequences (Davidson et al., 2008) . In the freshwater snail Biomphalaria glabrata, FREPs are encoded by an expanded gene family, and are implicated in defense against gastropod parasites such as digenetic trematodes (Hanington et al., 2010a,b) . In B. glabrata, FREPs are particularly noteworthy for being comprised of juxtaposed IgSF and fibrinogen domains, and for the fact they are somatically diversified during the production of hemocytes by the snails (Zhang et al., 2004) . Expanded families of C-type lectins are present in other mollusks and the bivalve www.frontiersin.org Mytilus edulis is capable of generating diversified forms of the AMP myticin C both within and among individuals (Costa et al., 2009 ). Among invertebrate deuterostomes, the sea urchin has proven surprising in featuring dramatic echinoderm-specific expansion of several recognition molecules (Hibino et al., 2006) . Sea urchins possess >220 TLR genes (vertebrates usually have 21-25), >200 NLRs (mammals have 20-35), >200 SRCR genes (humans have 16; Messier-Solek et al., 2010) , and a novel Sp 185/333 gene family. The latter gene family produces a repertoire of defense proteins more diverse than the sequence diversity encoded in the genes, indicative of the presence of another mechanism to generate diversity (Ghosh et al., 2010) . Sea urchins also possess an NF-κB pathway, lectin and alternative complement pathways and homologs of RAG1 and RAG2, but do not produce immunoglobulins (Ig), T cell receptors (TCRs), or have a major histocompatibility complex (MHC; Hibino et al., 2006) . With respect to our own phylum, the Chordata, the cephalochordate Branchiostoma (better known as Amphioxus or the lancelet), is novel in having expanded families of TLRs, NLRs, and SRCRs (Huang et al., 2008) , over 1,200 C-type lectins, and an extraordinary diversity in adaptors/facilitators and signaling/effector domains functioning downstream from their NLRs (Huang et al., 2008; Messier-Solek et al., 2010) . Amphioxus also possesses distinctive variable region-containing chitin-binding proteins (VCBPs; Dishaw et al., 2008; Cannon and Litman, 2009) which are further distinguished by high levels of polymorphism, resulting in yet another distinct "hyper-diversified," multigene immune receptors family (Dishaw et al., 2010) . Cephalochordates have a functioning complement system operating via the alternative and lectin pathways, including with a distinctive expanded number of C1q-like genes (Huang et al., 2008; Messier-Solek et al., 2010) . A RAG1 gene is present, and possibly a RAG2 gene as well (Dong et al., 2005) . The urochordates, or tunicates, the sister group to the vertebrates, in the same immunological vein as nematodes and flies, are surprising for what they do not have. None of the genes playing a pivotal role in adaptive immunity in the jawed vertebrates are present. MHC, TCRs, Ig, RAG, and activation-induced cytidine deaminase (AID) genes are all lacking. V-like and C1-like domains are present and VCBPs have been identified (Cannon et al., 2004) , and they do have complement components, three TLRs, an expanded family of C-type lectins and FREPs. However, urochordates lack obvious expansions of any gene family highly relevant to vertebrate immunity (Azumi et al., 2003) . Based on what we know thus far, genome reduction is the hallmark of urochordate immunobiology. Even closer to home are the agnathans or jawless vertebrates, lampreys and hagfish, the sister group to the jawed vertebrates or gnathostomes. We now know they lack RAG1 and RAG2 and do not produce TCRs or Ig, however, they have a remarkable ability to make highly diverse variable lymphocyte receptors (VLRs) that consist of somatically re-arranged modules containing leucine-rich repeats (Pancer et al., 2004) . It is striking that agnathans and gnathostomes have adopted divergent solutions to the same problem of generation of recognition capability, both of which involve rearrangements of germ-line encoded genes, but in entirely different ways with different starting sets of molecules. The basic gnathostome immune system, the one most familiar to immunologists, features a close collaboration between innate and adaptive arms. As noted above, relative to some of the invertebrate deuterostomes such as echinoderms or cephalochordates, gnathostome innate immune components are modest in numbers, typically possessing 10-25 TLRs and 20-35 NLRs (Messier-Solek et al., 2010) . The gnathostome adaptive immune system features somatic diversification of both TCRs and Ig, requiring for this process RAG1 and RAG2, the former likely derived and modified from a transib-like transposon (Fugmann, 2010) . The gnathostome immune system works in conjunction with a unique antigen processing and presentation system, the MHC, to limit self-reactivity. It is notable for its specificity, its emphasis on expansion of relevant clones of lymphocytes, and for its memory and capacity to produce heightened secondary responses long after primary stimulation . The ongoing discovery of new types of immune cells (Neill et al., 2010; Saenz et al., 2010) and novel receptors (Parra et al., 2007) strongly suggests there are more fundamental insights to come with respect to gnathostome immunology. Furthermore, and a point relevant for the general discussion here, there is considerable variability among gnathostomes in how their immune systems function (Flajnik and Kasahara, 2010) . To conclude this overview, it is indeed remarkable that organisms as diverse as cnidarians and humans have some immune architecture such as TLRs (and associated pathways) and NLRs in common. However, it is argued from the examples provided above that at least as compelling are the differences among and within phyla, even among species in a genus. Surprises abound, such as in the unexpectedly complete set of immune genes found in basal cnidarians and the immune genome reductions exhibited by nematodes, arthropods, and urochordates. Even the more familiar examples of conservatism such as TLRs and NLRs in arthropods and vertebrates may have been derived independently (Hughes, 1998; Hughes and Piontkivska, 2008; Zhang et al., 2010) . Large lineage-specific gene expansions such as noted for echinoderms, and domain reshuffling such as for invertebrate NLRs (Zhang et al., 2010) have occurred, creating remarkable heterogeneities among and within phyla. Layered on top of this are other forms of innovation such as elaboration of novel signaling pathways and production of associated AMPs in nematodes, distinct antigen recognition and melanin-deposition systems in arthropods, and the emergence of several distinct mechanisms for generating diverse antigen receptors in mollusks, arthropods, echinoderms, cephalochordates, agnathans, and gnathostomes. From this it is concluded that immune systems across and within phyla have a remarkable propensity to generate novelty and distinctiveness. As we learn about the immune systems of more animals, this diversity is bound to increase. It is hardly surprising that immune systems are so variable. Animals have been extant and diversifying for up to 800 million years (Erwin et al., 2011) . They have adopted a diversity of life styles: sessile, colonial predators; inhabitants of extreme environments dependent on chemosynthetic symbionts; animals that are at times net producers of energy due to their photosynthetic symbionts; pelagic species that live in enormous schools; inhabitants of arid terrestrial environments; social species living in large colonies; and endoparasites that are so modified morphologically as to belie their origins, to name just a few. These different life styles will impose very different exposures to potential parasites. Similarly, life histories vary radically from wind-dispersed organisms like tardigrades or rotifers that live in ephemeral habitats or that have life spans measured in hours, to sessile filter feeders like marine bivalves that routinely live for over 100 years in the same location. The role of life history traits such as survival rates and reproductive output are predicted to strongly influence the extent and kinds of particular immune responses that might be expected both among and within particular host species (Lee, 2006) . Another factor likely to have influenced immune capability is the nature and extent of commitment to mutualistic symbionts. Some animals have established mutualistic associations with what are essentially monocultures of specialized bacteria (Nyholm and Nishiguchi, 2008; Pais et al., 2008) . Others, like ourselves, are dependent on a diversity of both archaeal and bacterial mutualists to which our immune system has made extensive accommodation. Third party symbionts have long played a role in educating, augmenting, and modulating animal immune systems (Turnbaugh et al., 2007) . The outcome of host-parasite interactions is often influenced by third party symbionts which probably play a far great role in host defense in many animal groups than customarily realized (Loker, 1994; Welchman et al., 2009; Gross et al., 2009; Eberl, 2010) . The adoption of body plans differing in complexity and mass has also influenced immune system structure and function. It has been argued that evolution of the vertebrate jaw and an accompanying predatory life style introduced parasites into the gut and required a more elaborate adaptive immune system that now typifies gnathostomes (Matsunaga and Rahman, 1998) . It has also been argued that the complexities of adaptive immunity could not have evolved in animals with limited numbers of cells or with small size or simplified body architecture (Hauton and Smith, 2007) . With respect to body mass, for vertebrates, it has been suggested that the number of B and likely T cells in a clone scale with body mass as does the B cell repertoire (Wiegel and Perelson, 2004) . The general point is that the adoption of different habitats, life histories, symbionts, and bodies of differing body mass and complexity are all factors that will influence immune system design and mode of action. In addition to the above considerations, all organisms have to contend with another category of symbionts, namely parasites. Because viruses, bacteria, and protists were present before animals arose, all animals from their inception would have had to contend with these parasites. Several modes of transmission of such parasites among early animals of disparate lineages were available, including: intimate proximity of many different kinds of animals (such as on a coral reef), predation, presence of vectors imbibing blood or plant juices containing parasites, and even one parasite serving as a vector for another, as for example a trematode vectoring a bacterium into new animal hosts. In such a situation, where frequent transfer of parasites was possible among hosts from even disparate phyla, if all emerging animal lineages had the same defenses, it would be possible for an effective parasite that had overcome the defenses of one host group to simply spread into another host phylum. Consequently, having an immune system with distinctive means of antigen recognition and/or novel effector mechanisms would have been a distinct advantage when inevitably confronted with parasites that had evolved in other host groups (Figure 2) . Being immunologically different increases the odds that parasites from other inhabitants of the same coral reef are not as easily acquired. The notion that a parasite can track and exploit a common host genotype creating an advantage of rareness has been predicted and observed in specific host-parasite systems (Trachtenberg et al., 2003; Wolinska and Spaak, 2009 ), further suggestive of a similar dynamic favoring distinctiveness or "avoidance of commonness" among members of different host lineages. Once animals began to diversify, a major trend was for some animals to parasitize others. Some animal parasites became wholly or largely committed to particular lineages of animal hosts, in which they subsequently diversified. Such lineage-specific parasites (some examples in Table 2 ) are another general factor expected to drive immunological novelty. These parasites often establish prolonged, intimate, and extensive infections in their chosen hosts that have profound fitness consequences such as castration or death (Lafferty and Kuris, 2009a) . Furthermore, given the phylogenetic diversity represented among these parasites, it is not surprising they would evolve novel methods of infectivity. For example, ichneumonid and braconid hymenopteran parasitoids have acquired mutualistic polydnaviruses that function to suppress the immune responsiveness of their hosts and facilitate parasitoid infection (Webb et al., 2009) . In contrast, without the aid of viral symbionts, larvae of digenetic trematodes secrete both anti-oxidants and immunosuppressive factors that down-regulate snail host immune components for a period sufficient to enable them to complete their lengthy period of larval development (Hanington et al., 2010a) . Therefore, the immune response devised by a particular host lineage afflicted with its own phylogenetically distinct, host specific, and harmful parasites would likely be divergent from the responses mounted by a different host group experiencing its own lineage-specific parasites (Figure 2) . This is not to imply that only animal parasites have developed lineage-specific associations with hosts, but merely serves to show that specialized parasites can help us understand the origins of immunological novelties. The macroevolutionary patterns noted above with respect to novelty in defense strategies among animal lineages could not occur if there were not microevolutionary processes ultimately involved in www.frontiersin.org FIGURE 2 | One scenario for early in the divergence of animals is that different lineages (A-D) have fundamentally similar immune systems such that they all are colonized by the same parasites. In the case shown at the top, an immunological innovation occurred in lineage A, that allowed it to resist these parasites. This may have permitted a subsequent radiation in "parasite-free space" in this host lineage. At the bottom, lineage D has acquired a lineage-specific parasite different from those previously experienced. This requires an immunological accommodation that causes the immune system of lineage D1 to diverge. Both the host lineage and the lineage-specific parasites along with them may subsequently diverge. generating them. These microevolutionary processes occur below the species level, within and among populations of either host or parasite species, and might culminate in speciation of either participant. Speciation may be accompanied by colonization of new habitats, and further divergence to create major new lineages. Starting with the process of parasite diversification, the sections below discuss how the involvement of immunology in microevolutionary processes could lead to events that can help explain the macroevolutionary patterns discussed above. The following quotes outline some of the key ideas for how immunity can play a role in generation of parasite diversity. "For a pathogen, the selective pressures arising from the host immune system are a major influence on the evolution of mechanisms of infectivity and of immune-recognition avoidance" (Acevedo-Whitehouse and Cunningham, 2006) . "In parasitism an essential factor in survival is immune escape, which allows a parasite to resist host attack. Immune escape is a mechanism for reducing gene flow at the level of the compatibility filter because its result is assortative survival (Combes and Théron, 2000) as opposed to assortative mating." (Combes, 2001, p. 154) . "The stepwise coevolutionary process results in extreme specialization and complex defense mechanisms. . .specialization is likely to increase the rate of speciation that may occur in both host and parasite" (Price, 1980) "Host specificity thus is an ideal prerequisite for rapid speciation" (Mayr, 1963) . Parasites are often cited as examples of specialists because they have limited ranges of host species, often with restricted ranges of habitats within their chosen hosts. For example, the lineage-specific parasites mentioned above, often show considerable specificity to particular species or genera within their adopted host lineages. Most animal parasites are host specific (Poulin and Keeney, 2008; Agosta et al., 2010) , but this by no means is to suggest that generalists do not occur: parasites like Schistosoma japonicum, Toxoplasma gondii, Borrelia burgdorferi, or the rabies virus routinely infect a remarkably broad range of host species. Using molecular techniques to identify parasites, species formerly considered to be generalists have in some cases been shown to be complexes of cryptic, host specific species (Poulin and Keeney, 2008) . As eloquently documented by Combes (2001) , both encounter and compatibility filters operate to restrict the spectrum of host usage. Encounter filters pertain in situations where host and parasite live in different geographic localities, have different ecological circumstances, or where host or parasite behavioral tendencies preclude contact. The compatibility filter refers to barriers imposed by the host that prevent infection once contact has occurred. The compatibility filter includes both physiological and biochemical suitability of the host to support the parasite, and the active defense provided by the immune system. Encounter filters are undeniably important in restricting parasite host range. Many examples of emerging infections (Goss et al., 2009; Gray et al., 2009; Pfeffer and Dobler, 2010) owe their emergence to a change in the encounter filter such that a new combination of parasite (often including its vector) and host are juxtaposed (Daszak et al., 2000; Parrish et al., 2008; Weissenbock et al., 2010) . Emerging diseases are an indication that parasite infectivity is not always dependent on a long accommodation to a particular host species: lack of contact may have prevented prior infections. Similarly, experimental infections of new hosts with parasites essentially bypass the encounter filter, and are sometimes successful (Poulin and Keeney, 2008) , affirming the reality and importance of the encounter filter. With respect to the compatibility filter, a role for unsuitability should not be discounted and could be manifested in several ways, such as a lack of receptors needed for efficient viral entry into cells (Parrish et al., 2008) , lack of appropriate structures for parasite attachment (Tompkins and Clayton, 1999) , or by a general failure to provide the biochemical environment needed for the parasite to survive (Sullivan and Richards, 1981) . Although unsuitability is probably underappreciated with respect to preventing infections, relying on the possibility of being unsuitable is not a cogent defense strategy. The importance of active immunity to the compatibility filter is illustrated by several lines of evidence. (1) As illustrated by HIV, when the immune system is compromised, the door is opened to opportunists that themselves can become life-threatening (Holmes et al., 2003) . (2) Genetic defects in the immune system, such as with TLRs, are associated with increased susceptibility to several different pathogens (Qureshi and Medzhitov, 2003) . (3) Experimental exposures of hosts to parasites they have not previously encountered often fail (Bowen, 1976; Bozeman et al., 1981; Vidal-Martinez et al., 1994; Philips and Clarkson, 1998; Sapp and Loker, 2000; Duke, 2004; de Vienne et al., 2009; Giraud et al., 2010) or the parasite replicates poorly or is inefficiently transmitted in a new host (Komar et al., 2003; Parrish et al., 2008) . Table 3 provides examples of parasites in novel hosts that are engaged and killed by immune responses. (4) Host defense genes are under strong selection and are conspicuous for evolving quickly (Sackton et al., 2007; Viljakainen et al., 2009; Barreiro and Quintana-Murci, 2010; Schulte et al., 2010) . (5) The extraordinary diversity of strategies undertaken by parasites to evade, manipulate, or suppress the immune system is testament to the impact of immunity on their success (Schmid-Hempel, 2008) . These evasive strategies have been shown in some cases to be specific with respect to the particular hosts involved (Table 4) , providing a mechanistic basis for the connection between immunity and parasite specialization. Given that the consequences to a parasite for engaging the compatibility filter of an atypical host could be disastrous and result in its death, strong selection to avoid colonization of such hosts would be expected in some cases. Similar considerations may also apply to the host as well, as mounting an immune response can be costly and detrimental (Graham et al., 2011) . This means that some avoidance behaviors attributed to the encounter filter may actually be a consequence of the operation of a strong host immune response (Kuris et al., 2007; Keesing et al., 2009) . To conclude this section, specialization and attendant host specificity is a central, emergent property of parasitism and has multiple underlying determinants, involving both encounter and compatibility filters. The ubiquity of host defenses and the evidence that they often eliminate novel parasites argue that host immune systems play a critical role in limiting parasite host www.frontiersin.org Table 3 | Examples of colonizing parasites, or parasites placed in novel hosts, that are killed or limited by immune responses. Infection of the crab Pachygrapsus marmoratus with the rhizocephalan barnacle Sacculina carcini results in melanization of larvae in thoracic ganglia (Kuris et al., 2007) Antibody/factor that activates complement in serum of the non-host Raja radiata kills the tapeworm Acanthobothrium quadripartitum whereas larvae survive in serum of the normal host, Raja naevus (McVicar and Fletcher, 1970) Destruction of cercariae of avian schistosomes in the skin of mammals associated with a mixed Th1/Th2 lymphocyte cytokine response followed by more polarized Th2 response upon repeated exposures (Horak and Kolarova, 2005) Encapsulation of hymenopteran parasitoids by hemocytes of non-permissive insect hosts (Schmidt et al., 2001) Lysis of the trypanosome Trypanosoma brucei brucei by apolipoprotein L-1 in serum of humans who are refractory to this subspecies (Wheeler, 2010) . Disruption of the Erk-STAT1 signaling pathway allows cross species transmission of the normally rabbit-specific myxoma virus to mice (Wang et al., 2004) Animal handlers who were exposed to a new coronavirus developed antibodies to the new virus and did not develop clinical infections (Guan et al., 2003) Species specific forms of APOBEC3G and other innate, intracellular defense components, can prevent cross species transfer of lentiviruses (Mangeat et al., 2004; VandeWoude et al., 2010) A staphylococcal complement inhibitor that specifically blocks human C4b2a and C3bBb, interfering with additional C3b deposition through classical, lectin or alternative pathways (Rooijakkers et al., 2005) . Sung et al. (2008) found several genes conserved in all Staphylococcus aureus isolates from humans were variable or missing in one or more animal isolates, including fnbA, fnbB, and coa. Human and murine chlamydial infections depend on different virulence factor genes that coevolved to counter host species specific IFN-γ-mediated effector responses mounted by the particular host species (Nelson et al., 2005) . Orf virus is evolutionarily adapted to sheep as its primary host (Seet et al., 2003) . Different strains of influenza A virus likely have NS1 genes adapted to antagonize the IFNα/β antiviral system of their specific host species (Garcia-Sastre, In a review of the interactions between monogenean parasites and their fish hosts, Buchmann and Lindenstrøm (2002) concluded that "immune evasion mechanisms are probably a main factor in host specificity." Rosengard et al. (2002) noted that the smallpox inhibitor of complement enzymes (SPICE) is nearly 100-fold more potent than the vaccinia homolog in inactivating human C3b and sixfold more potent at inactivating C4b, providing evidence for how variola proteins are particularly adept at overcoming human immunity relative to vaccinia. The host specificity of three species of Bacillus (B. cereus, B. thuringiensis, and B. anthracis) is determined by the presence of virulence plasmids that determine the type of particular virulence factors produced (Gohar et al., 2005). ranges and thereby at least in part dictate the specialization so characteristic of parasitic organisms. Also in support of this claim is that some patterns of parasite host specificity can be attributed to the operation of specific immune evasion strategies, and that such strategies are pervasive among parasites. As exemplified by the parasites indicated in Table 2 , in addition to being specialized to exploit particular host groups, they are remarkably diverse in species, as are the lineages of many parasites. One of the potential consequences of specialization, including in an immunological context, is diversification in species, of both parasite and host lineages. The mechanisms involved in promoting speciation remain a matter of active investigation and for the discussion below, the purpose is to indicate that immunological phenomena may play a role in this process that deserves further attention. One prominent mechanism of parasite speciation is switching to a new host species, and the role of accommodation to the immune system of new hosts to permit such switches is discussed in a separate section below. A second mechanism is co-speciation. For some parasite groups closely tied to their hosts and with limited options for colonization of new hosts, such as sucking lice on burrowing mammals (Light and Hafner, 2008) , speciation may occur if the hosts upon which they are found themselves speciate, often following a physical separation of populations of the host species. In such cases, persistence of new daughter parasite species should be favored by the fact that the parental species had already achieved successful accommodation to the parental host species. Although the actual role of specific immune phenomena in influencing the persistence of incipient parasite species in co-speciating systems is not known, an important underlying role for a preexisting immunological accommodation between parental host and parasite species that also favors persistence of the new parasite species should not be discounted. Another important way in which specialization dictated by immunological phenomena can increase the probability of formation of new parasite species is by promoting intraspecific diversification. The interactions between a particular host and parasite species can be expected to be variable across space (Wood et al., 2007) . Parasite abundance will vary across local scales, possibly because of the variable presence of other hosts needed to complete its life cycle (Byers et al., 2008) . Other parasite species impacting the same host may be present or absent, such that the host experiences different overall parasite pressure in different locations within its range. Furthermore, the host itself will be variable across its range owing to its responses to other local circumstances. All of these factors conspire to create heterogeneities with respect to how the host potentially mounts immune responses to the parasite (Kraaijeveld and Godfray, 1999; Thomas et al., 2000; Lindstrom et al., 2004; Kalbe and Kurtz, 2006; Blais et al., 2007; Bryan-Walker et al., 2007; Scharsack et al., 2007; Matthews et al., 2010) . Variability within parasite species with respect to infectivity to their hosts is a pervasive phenomenon (Carius et al., 2001; Schulenburg and Ewbank, 2004; Seppala et al., 2007; Vorburger et al., 2009) and this is likely driven in part by variations in immune evasive measures taken by parasites (Hammerschmidt and Kurtz, 2005; Cornet et al., 2009; Vorburger et al., 2009 ). These dynamics are compatible with general theoretical predictions that parasite variation is driven by immunity, and hosts themselves are variable with respect to immunity due to pressure posed by parasitism (Frank, 2002) . Immune responses are drivers for parasite diversification (Summers et al., 2003; Lazzaro and Little, 2009; McKeever, 2009 ). An overall increase in intraspecific genetic variability, with that variation partitioned into regionally differentiated parasite populations accommodated to local host populations provide rich opportunities for further divergence. One possibility for further divergence is that local adaptation to host immunity could potentially lead to "assortative survival" (Combes and Théron, 2000; Combes, 2001, p. 154) , meaning that the only options for mating (parasites frequently seek mates and undergo sexual reproduction within their hosts) occur between individuals able to survive in hosts with similar immune capability that are vulnerable to the same parasite immune evasive capacity (Giraud et al., 2010) . This would further accentuate local differentiation of parasites, potentially leading to ecological speciation of the parasite, particularly if subsequent gene flow is prevented by failure of immune adapted parasites to thrive in hosts (from other localities) with different immune capacities. It must be noted that fluctuations in local patterns of abundance of hosts and parasites may diminish the strength of local adaptation and promote gene flow such that speciation is precluded (Lazzaro and Little, 2009) , and that in general, evidence that parasite speciation is effected by underlying immune mechanisms is sparse. However, given the need for parasites to accommodate to a host's internal environment and that a host species is likely to be confronted with varying parasite pressure, it seems host immune responses will favor diversification in parasite lineages. To add an additional dimension to the concept that spatially variable relationships favor parasite diversification, it has recently been argued a general underlying mechanism favoring biological diversification is the existence of localized parasite-coevolutionary races that select hosts to prefer immunologically similar conspecifics and to avoid out-group individuals, thereby minimizing the risk of exotic disease acquisition (Fincher and Thornhill, 2008) . By promoting strong intraspecific diversification within host species based on avoidance of contagion, this mechanism has also been predicted to lead to parasite diversification (Fincher and Thornhill, 2008) . To conclude this section, all of these observations fit into the more generalized geographic mosaic theory of coevolved relationships (Thompson, 2005) : in this particular case, local adaptations based on immunological accommodation of host and parasite can lead to diversification of parasites and potentially speciation. An important way diversity in parasite lineages is generated, one that has increasingly come to light from molecular phylogenetic studies and the study of emerging diseases, is via switching to new hosts ( Table 5) . Although successful host switching cannot be a ubiquitous process, otherwise we might expect to find only a few species of generalist parasites instead of a predominance of host specific parasites, clearly it has been an important factor historically and examples continue to be regularly documented. A priori, it seems logical that most successful switches would be to hosts not phylogenetically distant from the original host species. Such close range switches are likely favored by a degree of phenotypic plasticity and preadaptation (exaptation) of the parasite and its use of phylogenetically conserved resources in the new host species such that new attributes are not needed to overcome a new host's immune system (de Vienne et al., 2009; Agosta et al., 2010) . For example, in a study of host switches in bats involving the fast-evolving RNA virus causing rabies, the success of cross species transfers diminished as the phylogenetic distances among the hosts involved increased (Streicker et al., 2010) . It is also possible for parasite switches to occur when the original and new host species are not closely related (Brant and Loker, 2005) . This has been observed for emerging human parasites for which ungulates and carnivores were more likely originating host species than primates, and it was concluded that an already broad host range as opposed to the phylogenetic relatedness of the new and old host species was the more important factor dictating success in interspecific parasite jumps (Woolhouse et al., 2005) . In any case, a host switch can lead to a speciation event if the parasite in the new host becomes isolated from the founding stock, or can have even more profound effects if the switch is into a new host lineage and leads to the founding of a diverse new parasite lineage (Agosta, 2006; Janz et al., 2006; Hoberg and Brooks, 2008; Martinsen et al., 2008; Refrégier et al., 2008; Winkler et al., 2009; Giraud et al., 2010; Nyman, 2010) . The isolation of the switching parasite from the founding stock is reinforced because even a single individual may be able to establish a new population and because differing ecological circumstances of the new host may preclude mixing of parasite progeny with the source population: the new parasites may never get back into the original host and thus mate only with other parasites in the new host. Assortative survival and mating would again be factors favoring isolation of the founding parasites. Host switching is also relevant to the idea stated above that if a host lineage acquires a new parasite, it may then have its immune system substantially altered. Particularly if the parasite is successful and radiates, then the immune system of the new host lineage may be forced to diverge to adjust to the new challenge. www.frontiersin.org Table 5 | Examples of parasite groups exhibiting hosts switches likely to have played a major role in diversification of that group. With respect to Plasmodium and related genera of blood parasites, major clades are associated with shifts into different families of dipteran vectors, and the Plasmodium species of birds and squamate reptiles show evidence of repeated switching back and forth (Martinsen et al., 2008) . Major lineages within the blood fluke genus Schistosoma are defined by acquisition of different genera of even families of snail intermediate hosts, by host switching (Morgan et al., 2003) . Long-range host shifts involving acquisition of both new snail and vertebrate hosts appear to have occurred during the history of schistosomes (Brant and Loker, 2005) . Zietara and Lumme (2002) note that as many as 20,000 species of the monogenean genus Gyrodactylus may exist, and note that in a study of one subgenus (Limnonephrotus) that several host switch events were statistically confirmed, including into new host families, supporting the idea that host switching is a means to drive innovation and adaptive radiation in these ectoparasites. It appears that host switching has been common in trypanorhynch tapeworms, one of the most diverse and abundant groups of metazoan parasites of elasmobranchs (Olson et al., 2010) Coronaviruses have likely undergone several host switches, between mouse and rat, chicken and turkey, birds and mammals, and between humans and other mammals (Rest and Mindell, 2003) . Braconid wasps of the subfamily Euphorinae have undergone extensive host switching among phylogenetically distantly related insect host groups, often followed by adaptive radiations of the parasitoids within a particular host lineage (Shaw, 1988) . "Infection of a novel host is the most frequent cause of fungal emerging disease" (Stukenbrock and McDonald, 2008; Giraud et al., 2010) Several examples from the literature of emerging infectious disease indicate that switches are often favored by changes in the encounter and not the compatibility filter (Woolhouse et al., 2005) . Ecological circumstances have exposed humans to a parasite they previously did not encounter. Such examples of host switches, particularly if the new host is distantly related to the original host, would seem to argue against the points made in the preceding sections regarding the importance of parasite accommodation to the idiosyncrasies of their host's immune system. If immunity is important in restricting parasite host ranges, how can such switches occur? First, these conspicuous successes need to be weighed against all the encounters between novel parasite and host combinations that fail and therefore go unnoticed, which is likely a far more frequent outcome (de Vienne et al., 2009; Tunaz and Stanley, 2009; Giraud et al., 2010 ; see also Table 3 ). In cross species transfers of rabies into bats, the vast majority are dead ends: they did not establish sustained infections (Streicker et al., 2010) . Although some of the failures could be explicable because of less frequent contact among more distantly related bats (the encounter filter), increasingly divergent defense systems leading to higher levels of innate resistance were also invoked as an explanation (Daszak, 2010; Streicker et al., 2010) . The role of immune systems in preventing such infections would be easy to underestimate because the result is a failed experiment that in all probability we never even knew had happened. In a similar vein, a survey of fieldtrapped insects in turkey revealed that 98% exhibited some kind of melanotic hemocyte nodule (Tunaz and Stanley, 2009 ). Such host reactions provide a convenient historical record of previous parasite encounters (Kuris et al., 2007) . It was concluded that insects are regularly challenged by infections from which they recover. The action of innate immunity in routinely preventing acquisition of new parasites is probably considerable and easy to underestimate. Secondly, host switches would be favored if the new parasite, as exemplified by HIV, directly attacks the host's immune system and compromises it, or if the new host is immunocompromised by some other means. Diminished levels of immune competence can occur for several reasons, including ones likely to have been in operation throughout animal evolution. One possible means is that the host's indigenous parasites might use immunosuppression to favor initiation and persistence of their own long-term infections ( Table 6 ) and thereby facilitate colonization of that host by other parasites (Krasnov et al., 2005) . An intriguing possibility is that the successful colonization of a host species by one or more immunosuppressive parasites might then favor colonization by opportunist parasites, resulting in an unusually diverse parasite fauna supported by that host. The large number of species of larval digenetic trematodes known to be supported by some snail species (Loker et al., 1981; Lafferty and Kuris, 2009b) might exemplify this possibility. High host density, stressful thermal (Bruno et al., 2007) or oxygen regimes (Aeby and Santavy, 2006) , and even mating (Rolff and Siva-Jothy, 2002) are some natural situations that can also lower immune competence. To this list can be added a number of human-imposed immune stressors including crowded aquaculture conditions (Flemming et al., 2007) , use of harmful chemicals on fields or roads (Rohr et al., 2008; Karraker and Ruthig, 2009) , altered diets (Sahu et al., 2008) , elevated or altered environmental nutrient conditions (McKenzie and Townsend, 2007; Wedekind et al., 2010) , and deliberate implementation of immunosuppressive therapies. Not only can these situations lower host immune competence, they may also increase parasite virulence and thereby alter probabilities of successful colonization in a new host species (Wedekind et al., 2010) . A switch into an immunocompromised individual of a new host species is likely to be temporary and not lead to speciation unless the new parasite can better adapt to its new host, at the same time with minimal gene flow occurring with conspecifics from its ancestral host. Availability of populations of similarly immunocompromised new hosts that allow continued transmission and adaptation of the new parasite host could favor divergence from the founding parasite and speciation. Varroa mites (Varroa destructor ) in honey bees (Apis mellifera) suppress the activity of several immune-related genes (encoding both antimicrobial peptides and enzymes) and favor higher infection titers with the deformed wing virus (Yang and Cox-Foster, 2005) . Drosophila simulans infected with Wolbachia have reduced ability to encapsulate eggs of the parasitoid Leptopilina heterotoma (Fytrou et al., 2006) . The malaria parasite Plasmodium gallinaceum suppresses the encapsulation response of the mosquito Aedes aegypti (Boëte et al., 2004) . Two acanthocephalan parasites Pomphorhynchus laevis and Polymorphus minutus both have the effect of decreasing the standing level of immune defense (as measured by reduced phenoloxidase enzyme activity) in their local gammarid hosts, Gammarus pulex, but not in their more recently introduced host Gammarus roeseli (Rigaud and Moret, 2003) . Hymenopteran parasitoids induce immunosuppression in their host insects in part by the injection of polydnaviruses which target and inhibit both cellular and humoral components of the host response (e.g., Labropoulou et al., 2008) and the parasitized hosts become increasingly susceptible to opportunistic infections by viruses (Rivkin et al., 2006) , bacteria (Shelby et al., 1998) , and other parasitoids (Guzo and Stoltz, 1985) . As noted by Lie (1982) , interference by trematode larvae with gastropod defense responses appears to be a common consequence of infection (Hanington et al., 2010a) , and the presence of one trematode infection can facilitate the colonization of an infected snail by trematodes that would not ordinarily be able to infect that species of host (Walker, 1979; Southgate et al., 1989) . HIV in people was associated with parasites that rarely if ever had been implicated in causing human disease including microsporidia, cryptosporidia, JC virus, and Mycobacterium avium (Kovacs and Masur, 2008) . Studies of parasite communities suggest that taxonomic distinctiveness of ectoparasites and endoparasite richness are positively correlated across species of rodent hosts, indicative of immune responses to some parasites depleting energy reserves and facilitating colonization by others (Krasnov et al., 2005) . How might immunological phenomena influence the degree of diversity shown among the hosts contending with parasites? As highlighted below, parasite pressure clearly favors immunological diversification at microevolutionary scales. Whether this diversification contributes in a meaningful way to host speciation remains controversial, but has attracted considerable attention and is gaining some support, as discussed below. As early as Haldane (1949) , and as more recently underscored (Frank, 2002; Lazzaro and Little, 2009) , it has been recognized that parasites drive polymorphisms in host immune competence, particularly in variable environments. This can occur by balancing (Wegner, 2008) or disruptive selection (Duffy et al., 2008; Matthews et al., 2010) . High levels of polymorphism are found in several genes of both the innate and adaptive immune systems (Hill, 2001; Trowsdale and Parham, 2004; Acevedo-Whitehouse and Cunningham, 2006) . MHC genes show a predominance of non-synonymous over synonymous mutations in their peptidebinding regions, and have extensive allelic diversity, indicative of strong role of selection, generally considered to be mediated by parasites. For example, parasite pressure is believed to have promoted maintenance of high MHC diversity in sticklebacks (Wegner et al., 2003) and Atlantic salmon (Dionne et al., 2007) . In humans, regional differences in HLA class I diversity has been associated with intracellular parasite richness (Prugnolle et al., 2005) . Across species comparisons of rodents have associated helminth species richness with increased MHC class II polymorphisms (de Bellocq et al., 2008) . MHC genes are also known for their role in mediating mate choice through olfactory systems in humans (McClintock et al., 2002) , rodents (Sommer, 2005) , and fishes (Landry et al., 2001; Milinski et al., 2005) . The evidence supporting the idea that variability in immune response driven by parasites can be a factor favoring speciation of host lineages is mostly correlational, but is supported by a growing body of literature. One general factor favoring this is the development of strong local immunological accommodations of particular host populations to the distinctive parasites they encounter, such that across a broader host range different host populations differ significantly with respect to the nature of their responses (Wheatley, 1980; Blais et al., 2007; Scharsack et al., 2007; Matthews et al., 2010) . For example, malaria might be encountered in some but not all parts of the host range where appropriate mosquitoes were present, or different suites of parasites might be encountered in different foraging habitats such as rivers or lakes (Eizaguirre et al., 2011) . Differences among populations of the same host species with respect to their immune defenses have been noted in Drosophila against parasitoids (Kraaijeveld and Godfray, 1999) , Darwin's finches (Lindstrom et al., 2004) , sticklebacks coping with eye flukes (Kalbe and Kurtz, 2006) , and marine amphipods infected with trematodes (Thomas et al., 2000; Bryan-Walker et al., 2007) . Differentiation resulting from spatially variable antagonistic interactions with parasites would in this case provide the substrate for further diversification of host lineages (Thompson, 2005 (Thompson, , 2009 . The impact of this local adaptation could be augmented by assortative mating mediated by sexual selection to favor further divergence (van Doorn et al., 2009 ). According to this line of thought, those locally adapted hosts that best resist parasites are able to elaborate ornaments that signal superior resistance to parasites, such that local mates preferentially breed with them. If these hosts were transplanted to other host populations with their own distinct parasite challenges, they would not be as resistant, their sexual ornamentation would suffer, and they would not be selected for mating. Thus a combination of natural and sexual selection could favor divergence of new host species. Parasites can exert strong selection on traits known to affect mate choice (Hamilton and Zuk, 1982; Poulin and Thomas, 1999; MacColl, 2009) and in some cases the genes involved also have immune functions, such as genes of the MHC (Milinski et al., 2005; Milinski, 2006; Eizaguirre et al., 2009 Eizaguirre et al., , 2011 . MHC genes www.frontiersin.org have been considered as possible "magic traits" in sticklebacks, influencing both defense and mate choice (Matthews et al., 2010) . MHC divergence in a closely related and sympatric pair of cichlid species from Lake Malawi has been proposed to be a result of local host-parasite-coevolutionary associations, and to have influenced odor-mediated mate choice, and ultimately to have favored speciation (Blais et al., 2007) . Diverging cichlids of the genus Pundamilia in Lake Victoria provide another example of how local adaptation and assortative mating (both potentially influenced by immune traits) may work together to promote host speciation. In this system, females have a preference for conspicuously colored males. These bright colors seem to be reliable indicators of male fitness, including resistance to parasites. Conditions of light penetration favor blue males in shallow depths and red males in deeper waters. The parasites encountered at different depths also vary in density and composition such that habitat-specific defenses could occur. Females from the depths prefer brighter red males whereas those from shallow water prefer brighter blue males, potentially leading to divergence, with visual cues playing a key role. This example points out the immunology may work in concert with a number of other forces such as predator avoidance or dietary preferences which all conspire to reinforce divergence of the two incipient species by visual means (Maan and Seehausen, 2010) . Mating among hosts between different locations would break down these differences, but might be disfavored if the progeny had reduced resistance to any local set of parasites. "Infectious speciation" as exemplified by interactions between Drosophila species and inherited, endosymbiotic Wolbachia bacteria provides another possible mechanism for the involvement of immune processes in host speciation. For a group of six related species in the D. paulistorum complex, each species has its own distinct host specific obligatory and mutualistic Wolbachia with which it has achieved accommodation. This accommodation likely involves suppression of immune pathways involving apoptosis of infected cells. In hybrids, the Wolbachia over-replicate and cause embryonic inviability and male sterility, suggesting the unique host accommodation has been lost. In addition to such postmating isolation, it has also been shown that females can detect and will reject males harboring the wrong symbiont, thus further reinforcing isolation . Among animal hosts, hybrids are often more susceptible to parasites than parental species (see overviews provided by Fritz, 1999; Wolinska et al., 2008) , potentially favoring isolation of the parental host species, but other outcomes have also been noted and the responsiveness of hybrids to infection recorded 1 year might differ from those reported the next. This implies that the interactions between hybrids and the parasites they experience exhibit complex temporal dynamics and that the parasites themselves have undergone complicated changes as a result of their hosts' hybridization history that have not been sufficiently investigated (Detwiler and Criscione, 2010) . Certainly some studies suggest that isolation of incipient parental species might be reinforced by a breakdown of co-adapted immune gene complexes among their hybrids. Also, in some cases, the act of hybridization contributes to the formation of new species by allopolyploidy, as has been postulated for anurans of the genus Xenopus. Hybrids in this case often have increased resistance for parasites, potentially providing a selective advantage to favor the persistence of new species of recent hybrid origin (Jackson and Tinsley, 2003) . Another hypothesized mechanism favoring diversification of host lineages is that localized interactions with particular parasites allows immunological accommodation to them, resulting in strong preference for interactions with individuals with similar immune accommodation and philopatry (limited host dispersal). This is coupled with avoidance of out-group individuals that might lead to introduction of exotic parasites (Fincher and Thornhill, 2008) . Diverse parasite populations are thus hypothesized to drive diverse host populations, and ultimately speciation, leading to a general correlation between host and parasite biodiversity (Fincher and Thornhill, 2008) . Extinction too is a macroevolutionary process, including the major pulse of extinction events currently underway. Habitat destruction, human overpopulation, industrialization, threats from introduced species, emerging diseases, and global climate change, have lead to predictions that up to 50% of all species will be lost in the next 50 years (Pimm and Raven, 2000; Koh et al., 2004; Thomas et al., 2004) . The study of immunology is relevant to extinction in at least three broad contexts outlined below. First, as noted above, the interactions between parasite and host often lead the parasite to specialization and host specificity which may in part be dictated by interactions with the host immune system. It has long been argued that specialization leads to a greater probability of extinction because if the host on which the specialized parasite is dependent undergoes severe population fluctuations or itself goes extinct, the parasite will soon follow: a co-extinction event has occurred. Co-extinctions involving pairs of mutualists or host-parasite units may be the most common forms of loss of biodiversity (Koh et al., 2004; Dunn et al., 2009) . By comparison, a generalist parasite able to exploit alternative hosts would have a greater likelihood of surviving under similar circumstances. The evolutionary trajectories taken by parasites have been much debated, and specialized parasites have been shown to give rise to large lineages of specialized, or even to more generalized descendents (Johnson et al., 2009) . However, host specificity remains a salient feature affecting the odds of extinction (Koh et al., 2004; Poulin and Keeney, 2008) . Second, an inescapable feature of the modern world is the frequency with which invasions of exotic species occur (Torchin et al., 2002) . Invasive species might be either hosts or parasites (possibly including their vectors), and all can present dire and often unpredictable consequences, including extinction, for indigenous organisms (Wyatt et al., 2008) . A role for immunology in influencing invasions can occur in at least three ways: (a) Introduced host species often leave their native parasites behind and although they are likely to be colonized by parasites in their new environment, in some cases this colonization is slow to occur, such that they experience relatively low parasite burdens for a long time. Insofar as immune responses are costly to mount and the harmful effects of parasites are avoided, the invading species may enjoy a distinct advantage over its competitors in its relative freedom from parasites, particularly if they can adopt relatively low cost defense measures against their maladapted parasites ( Lee and Klasing, 2004) . The relative inability of parasites to colonize the new intruder is a testament to the specialization often required to achieve infectivity, as noted above. (b) If an introduced host is accompanied by some of its indigenous parasite fauna, then related hosts in the newly colonized area may then have to contend with a parasite to which they are not accustomed, a process that may take decades to achieve Taraschewski, 2006) . This is particularly likely to cause problems if the colonized area is an isolated habitat like an island where hosts have simply not encountered comparable parasites previously. The devastating impact of the introduction of mosquitoes followed by avian malaria and avian pox on the endemic honeycreepers of the Hawaiian islands, likely causing both extinctions and slowing recovery of still extant species, is an iconic example (Atkinson and Samuel, 2010) . Islands often favor speciation yet the subsequent colonization of mainland habitats by island species is likely to be limited and unsuccessful due to active transmission of parasites there that a migrant is unable to handle immunologically. (c) In some cases, invasion of a parasite can occur even without the benefit of a simultaneous introduction of its indigenous host. An example is the eel swimbladder nematode Anguillicola crassus which has been introduced from the Orient into Europe where it provokes intense tissue reactions from European eel species (Taraschewski, 2006) . By contrast, Oriental eels mount immune responses that prevent a high and robust population of worms from building up. All of these examples centered on introductions have macroevolutionary implications as they might lead to expansion and radiation of hosts and their parasites into new habitats, or may directly cause extirpations of indigenous hosts (and possibly their co-adapted parasites as well). Immunological phenomena provide proximal causal explanations. Once again, as with the discussion above regarding host switching, there may be significance to what we do not see as well: many parasite introductions fail because the colonist is unable to breach indigenous host defenses and host introductions fail because they are ill-prepared for indigenous parasites. For example, the introduction of the American rainbow trout Oncorhynchus mykiss into Europe failed because they encountered the parasite Myxobolus cerebralis which is normally transmitted by the indigenous brown trout, Salmo trutta: rainbows succumbed to whirling disease in Europe . When this parasite was introduced into North America, brown trout (themselves also introduced) were already well-adapted to it but indigenous rainbows and other salmonids were not and have suffered outbreaks of whirling disease as a consequence. A third broad context in which immunology becomes relevant to extinction, and one that is much in evidence today, is the role played by diminution of diversity in key immune loci such as the MHC. This can result in vulnerability of endangered host species to general parasite attack and thus extinction (Radwan et al., 2010) . Loci other than the MHC may also play major roles in dictating susceptibility to particular groups of parasites (Behnke et al., 2003) , and polymorphism in non-MHC genes are relevant in resistance to both tuberculosis (Ottenhoff et al., 2005) and malaria (Hill, 2001) . MHC genes are estimated to account for only about half of the genetic variability for resistance traits (Acevedo-Whitehouse and Cunningham, 2006) . Regardless of the immune locus involved, the role of random drift in diminishing allelic diversity in bottlenecked or fragmented host populations would seem to increase the risk of successful parasite attack and greatly increase extinction probabilities. There is a need to determine if other factors like genome-wide inbreeding depression are more important in causing extinction, to understand why some species depauperate in MHC diversity seemingly continue to thrive (Hedrick, 2003; Radwan et al., 2010) , for increased transfer of information from the study of the main immunological models to endangered species, and to study other polymorphic genes involved in effective defense (Acevedo-Whitehouse and Cunningham, 2006) . A picture of the structure and function of immune systems across animal phyla is slowly emerging, but so many organisms remain unstudied we lack perspective on how representative our current picture is. Do all tunicates, for example, reveal evidence of pronounced genome reduction with respect to immune function, or is this a characteristic of just the few species studied to any extent? What kind of pressure from parasites or otherwise has driven the expansion of all the innate immunity genes evident in organisms like sea urchins, and how do these animals regulate and orchestrate effective responses given the multiplicity of defense molecules they possess? In their 100+ years life span, how frequently must marine bivalves mount immune responses and how do they avoid the problem of fast-evolving parasites from "locking on" and overrunning them? In the meantime, it is clear that the approaches taken to achieve immune defense are diverse, frequently innovative and often capable of generating diverse repertoires of defense molecules that blur the distinctions commonly made to distinguish adaptive from innate immunity. The diversity in immune systems among and within phyla is in and of itself a major macroevolutionary pattern that should become a more central part of how we characterize animal diversity, including in textbooks. We need to know what the full pattern actually is, and the pattern begs an explanation for the processes involved in generating it. Study of model parasite-host systems shows that the strength of selection imposed on particular immune genes is strong, and can result in some of the fastest evolutionary rates known for metazoans. In other words, parasite-host immune interactions strongly influence anagenesis, the evolutionary changes occurring within parasite or host lineages. If we adopt a broader perspective with a longer time frame, it seems likely such intense interactions will be seen to have an impact on cladogenesis, the origin of lineages, as well. It seems the overall impact of infection and immunity should attract as much attention as predation, competition, or other biotic interactions in shaping the overall diversity of animal life. Needed are more empirical studies over longer time www.frontiersin.org frames to provide more specific mechanistic insights as to how host immune responses drive diversification of parasites and how this can lead to speciation events, potentially in both parasite and host lineages. Further revelation of specific genes, often not considered part of the immune system per se, and how they facilitate defense against particular parasites and might favor evolutionary divergence among parasites are needed. Finally, placing immunology in a macroevolutionary perspective can hopefully provide insights for understanding today's world in which a host of rapid changes greatly increase the odds for extinction of many animals. This work was supported by NIH grants AI024340, and by 1P20RR18754, the latter from the Institute Development Award (IDeA) Program of the National Center for Research Resources, NIH. immunodeficiency virus type 1related opportunistic infections in sub-Saharan Africa. Clin. Infect. Dis. 36, 652-662. Horak, P., and Kolarova, L. (2005) . Molluscan and vertebrate immune responses to bird schistosomes. Parasite Immunol. 27,

Recent Advances in the Diagnosis and Treatment of Influenza Pneumonia

A potentially fatal complication of influenza infection is the development of pneumonia, caused either directly by the influenza virus, or by secondary bacterial infection. Pneumonia related to the 2009 influenza A pandemic was found to be underestimated by commonly used pneumonia severity scores in many cases, and to be rapidly progressive, leading to respiratory failure. Confirmation of etiology by laboratory testing is warranted in such cases. Rapid antigen and immunofluorescence testing are useful screening tests, but have limited sensitivity. Confirmation of pandemic H1N1 influenza A infection can only be made by real-time reverse-transcriptase polymerase chain reaction (rRT-PCR) or viral culture. The most effective preventive measure is annual influenza vaccination in selected individuals. Decisions to administer antiviral medications for influenza treatment or chemoprophylaxis should be based upon clinical and epidemiological factors, and should not be delayed by confirmatory laboratory testing results. Neuraminidase inhibitors (NI) are the agents of choice.

Influenza is acute respiratory illness caused by influenza A or B viruses with seasonal circulation during the winter months. However, outbreaks of novel recombinant strains that take place in animals have produced, along the years, many worldwide outbreaks with serious public health issues. Although normally a self-limited process in the general population, high risk groups for complications and death have been identified (Table 1) . During an outbreak, in otherwise healthy subjects the diagnosis of influenza infection can be made confidently based on clinical manifestations alone. However, in certain situations, such as sporadic cases, in patients at an increased risk for complications, or in hospitalized patients with severe pulmonary compromise, confirmation of etiology by laboratory testing is required to guide treatment and for surveillance purposes [1••, 2••]. A potentially fatal complication of influenza infection is the involvement of the lower respiratory tract caused directly by the influenza virus, and the development of secondary bacterial pneumonia. Novel recombinant influenza A strains carry the risk for more severe disease and have the potential to cause widespread illness and a large number of deaths, regardless of age or previous health status. Examples of this are the H5N1 "avian" influenza A outbreak since 2004, and the more recent H1N1 "swine" influenza A pandemic in 2009, both of which have prompted the development of quick and reliable laboratory test in an effort to optimize their management and reduce morbidity and mortality. In this article, we review the latest available data for the diagnosis of influenza lower respiratory tract infection, as well as for treatment and prevention strategies. Signs and symptoms of upper and/or lower respiratory tract infection, along with systemic involvement in the form of fever, myalgia, and headache, are usually the main presenting features of the disease. In the context of an outbreak, otherwise healthy subjects presenting with a self-limited acute febrile respiratory illness usually require no further diagnostic procedures. In two retrospective studies that examined which clinical signs and symptoms are most predictive of influenza infection in patients with influenza-like illness, cough and fever were the only symptoms significantly associated with a positive PCR test for influenza [3, 4] . In another study, no isolated symptom or sign was able to accurately predict influenza infection, though the absence of fever, cough and nasal congestion significantly decreased its likelihood [5] . In general, patients diagnosed with pandemic H1N1 influenza A virus had similar signs and symptoms compared to those with seasonal influenza. However, these patients had gastrointestinal manifestations more frequently [6, 7] , were more likely to have pneumonia [8] , and also had higher rates of extrapulmonary complications, intensive care unit admission, and death [9] . Pneumonia is the most frequent and severe complication of influenza, most commonly presenting in high risk patients (Table 1) . Primary influenza pneumonia represents direct lung involvement by influenza virus, and should be suspected in non-resolving influenza infections. Typically, primary influenza pneumonia presents in chest x-rays with bilateral reticular or reticulonodular opacities. Less frequently, focal areas of consolidation can be seen, particularly in the lower lobes. High-resolution computed tomography may show ground glass opacities with or without multifocal peribronchovascular and subpleural consolidation [10] . The cytopathic effect of the influenza virus on the tracheobronchial epithelium may predispose to secondary bacterial pneumonia [11, 12] . Secondary bacterial pneumonia must be suspected whenever there is an exacerbation of fever and respiratory symptoms after initial improvement in a patient diagnosed with acute influenza. Leukocytosis, instead of a normal or low white blood cell count, and lobar consolidation on chest imaging, instead of the diffuse pattern that is typical of viral pneumonia, are also suggestive [13] . In an observational study of 543 hospitalized patients with H1N1 influenza A infection in Spain, 43 % of the 243 patients in which chest radiographs were performed had pneumonia, 83 % of the 210 patients who had microbiologic confirmation had primary influenza pneumonia, and the remaining 17 % had concomitant secondary bacterial pneumonia. Bilateral pneumonia occurred in 48.3 % of patients; Streptococcus pneumoniae being the most frequent pathogen [14] . Several reports have identified methicillinresistant Staphylococcus aureus (MRSA) as the etiologic agent for severe community acquired pneumonia (CAP) in otherwise healthy young patients with influenza [15] [16] [17] . In another study that investigated the incidence of communityacquired MRSA pneumonia in H1N1 influenza patients, 50 patients of 4491 (1 %) laboratory-confirmed pandemic influenza A (H1N1) cases had a bacterial respiratory tract pathogen. The most commonly cultured organisms were S. pneumoniae (16 patients), S. aureus (13 patients) and Haemophilus influenzae (9 patients); MRSA was detected in only 2 patients [18] . In contrast, among 838 children and adolescents admitted to 35 intensive care units in the U.S. with confirmed or probable severe H1N1 influenza A infection, 48 % of the 71 patients with suspected diagnosis of early S. aureus coinfection had MRSA [19] . Non-seasonal influenza infections have specific clinical manifestations. Pneumonia related to the 2009 H1N1 influenza A pandemic was also found in many cases to be rapidly progressive, leading to respiratory failure and ARDS [20•, 21•] . Additionally, the risk for complications and death due to that (2) Including disorders of the brain, spinal cord, peripheral nerve, and muscle such as cerebral palsy, seizure disorders, stroke, intellectual disability, moderate to severe developmental delay, muscular dystrophy, or spinal cord injury Adapted from the recommendations of the Advisory Committee on Immunization Practices (ACIP) [40••] pandemic influenza was found to be underestimated by commonly used pneumonia severity scores [22•, 23] . Avian influenza (H5N1) frequently presents as severe primary pneumonia that often progresses rapidly to the acute respiratory distress syndrome (ARDS), having caused high rates of death, especially among infants and young children in Southeast Asian countries [24] . In certain situations, confirmation of etiology by laboratory testing is required in order to guide the initiation and duration of antiviral therapy, and for the implementation of infection control measures and surveillance. Other benefits of influenza virus detection are the reduction of inappropriate antibiotic use, decreased length of stay in emergency departments, and fewer additional laboratory studies, all leading to a reduction in health care costs [1••] . The Centers for Disease Control and Prevention (CDC) and the Infectious Diseases Society of America (IDSA) have published guidelines to better define patients who should undergo influenza testing [1••, 2••]. The available methods include immunological techniques (i.e. rapid antigen-based tests, immunofluorescence assays, serologic testing), molecular techniques (i.e. reverse-transcriptase polymerase chain reaction [RT-PCR]), and microbiological techniques (i.e. viral cultures). While RT-PCR has the highest sensitivity and specificity, rapid antigen and immunofluorescence testing, though very useful as initial screening tests, are considerably less sensitive. Besides, the sensitivity and specificity of any of these techniques will vary depending on the laboratory equipment, personnel expertise, the timing of recollection, and the appropriate handling of the samples [1••, 2••]. Respiratory specimens can be obtained by many different methods including throat swabs, nasal aspirates, and nasopharyngeal swabs, aspirates and washing [2••] . In mechanically ventilated subjects, more invasive maneuvers such as endotracheal aspirates and bronchoscopic or non-bronchoscopic bronchoalveolar lavage (BAL) may be required in order to obtain adequate lower respiratory tract samples. While rapid antigen tests, immunofluorescence and RT-PCR all can yield results quick enough to guide point-of-care clinical decisionmaking, serologic tests and viral cultures provide retrospective diagnosis, availability only in reference laboratories, and usefulness when confirming screening test. For this reason, they are normally reserved for epidemiological and research purposes (Table 2) . Rapid influenza diagnostic test (RIDT) are designed to detect influenza A and B nucleoproteic antigens in respiratory specimens, with results expressed qualitatively as positive or negative in no more than 15 min. Many different FDAapproved tests are available; while some assays are capable of distinguishing between influenza A or B viruses, none can distinguish between pandemic and seasonal strains of influenza A [1••, 2••]. Compared to reference methods (i.e. RT-PCR and viral culture), they have low sensibility (10 % to 80 %) and high specificity (95 % to 100 %), and there is also great variability between the different commercially available kits [25] [26] [27] . Sensitivity appears to be somewhat lower for influenza B than that seen for influenza A [26] . Also, the reported sensitivity for rapid tests is higher for nasopharyngeal samples than for throat swabs [27] . There are two main factors that influence RIDT's negative predictive value. First, negative results obtained during periods of high viral activity in the community are more likely to be false negatives; alternatively, false positives, though much less frequent, are more likely during periods of less viral circulation. Second, respiratory samples collected within the first 48 to 72 h of symptom onset positively influence RIDT's sensitivity. For these reasons, when rapid tests are negative, confirmation by means of RT-PCR or viral culture should be considered [1••, 2••]. Compared to RT-PCR, the sensitivity of RIDTs for detecting novel influenza A (H1N1) was equal to or lower than the sensitivity to detect seasonal influenza viruses [29] [30] [31] [32] , so RIDT results need to be interpreted with caution when evaluating patients suspected of having pandemic H1N1 influenza A. Immunofluorescence Direct (DFA) or indirect (IFA) immunofluorescence antibody staining techniques are capable of detecting influenza A and B viruses, and distinguish the viruses from each other as well as from other respiratory viruses [1••, 2••]. They have levels of sensitivity and specificity that come close to those of RT-PCR [28, 33] , and results are often available in a few hours [1••, 2••]. Although these tests have improved sensitivity over RIDT, they are more technically complex and require expertise in obtaining quality respiratory samples ( Table 2) . This is the reference influenza detection method and has the highest sensitivity and specificity [34, 35] . Several modalities of RT-PCR have been designed: conventional gel-based PCR (cRT-PCR), multiplex PCR (mRT-PCR), and real-time RT-PCR (rRT-PCR). They can differentiate between influenza types (A or B) and subtypes (including pandemic H1N1 influenza and avian H5N1 influenza), and results are available in 2-6 h (although due to transportation of batched specimens to reference centers for processing, it may take longer for results to be available). During the 2009 H1N1 influenza A pandemic, it became clear that rapid case identification was essential for timely management of patients and for adequate public health actions to be taken. To answer to this threat, the CDC optimized the previously developed rRT-PCR procedures for detection of the A/H1N1 2009 pandemic influenza virus [36• ]. Yang et al. compared the performance of 12 rRT-PCR primer-probe sets designed for detecting the hemagglutinin (HA) or the neuraminidase (NA) gene of the pandemic influenza A/H1N1 2009 virus, using the primer-probe set developed at the CDC as reference. They found that although all primer-probe sets had specificity levels as high as 98.4 % to 100 %, some of the primer-probe sets had better specificity, sensitivity, and amplification efficiency than others, and that a combination of primer-probe sets targeted to the HA and NA genes had higher detection sensitivity than those targeting HA or NA individually [37] . In another study, Lam et al. showed that although rRT-PCR assays can be 10-fold more sensitive than cRT-PCR, newly developed cRT-PCR assays targeting the HA gene are a reliable alternative for laboratories where a rRT-PCR machine is not available [38] . Sensitivity and specificity of these assays also depend on the type of respiratory sample employed. For example, in a Spanish study that assessed the utility of rRT-PCR for the diagnosis of the novel influenza A/H1N1 virus, the authors reported that the diagnostic yield of combined nose and throat swabs was higher than that of nasopharyngeal aspirates [39] . Even the fastest viral cultures techniques can take days to demonstrate influenza cytopathic effects. Since they are not suitable for initial clinical management, their utility during an influenza outbreak is to confirm some negative test results from RIDT and immunofluorescence. Viral cultures also provide information about circulating influenza strains and its subtypes; this information is required for next season vaccine production, for surveillance of the emergence of new influenza A strains, and for the detection of antiviral resistance. Similarly, serologic tests (i.e., hemagglutinin inhibition, ELISA, complement-fixation) that demonstrate a four-fold increase in serum antibody titers between acute and convalescence phases of the disease, are only useful for retrospective diagnosis or research purposes ( Table 2) . Annual immunization is the most important preventive measure [40••] . However, two classes of antiviral drugs are available and play an important role in the treatment and prevention of influenza [41••] : the neuraminidase inhibitors (NI), oseltamivir and zanamivir, which are active against both influenza A and B viruses; and the M2 inhibitors, amantadine and rimantadine, which are active against all influenza A strains, but have no activity against influenza B viruses. In general, the duration for therapy with an NI is 5 days, and with the M2 inhibitors is three to 5 days. NI are sialic acid analogs that competitively inhibit neuraminidase on the surface of both influenza A and B, thus interfering with the release of virus from infected cells. Oseltamivir phosphate is an orally bioavailable prodrug that is rapidly absorbed from the gastrointestinal tract and is converted by hepatic esterases to the active metabolite, oseltamivir carboxylate. It has an elimination half-life of approximately 8 h, primarily . These side effects are usually mild and limited to the first days of treatment, although more serious side effects have been described. Oseltamivir has been linked to selfinjury and delirium in pediatric populations, although no causal association could be demonstrated [42, 43] . The use of inhaled zanamivir has been associated with bronchospasm, sometimes severe or fatal, particularly in patients with underlying airways disease such as chronic obstructive pulmonary disease (COPD) or asthma [44] . Amantadine and rimantadine target the M2 protein of influenza A, which forms a proton channel in the viral membrane that is essential for viral replication. Amantadine is primarily excreted unchanged in the urine. In patients older than 65 years and in those with an estimated creatinine clearance of less than 50 mL/min, the daily dose should be reduced. Rimantadine is extensively metabolized in the liver, and dose reduction is recommended for patients with severe hepatic dysfunction, renal failure (creatinine clearance <10 mL/min), and the elderly. Central nervous system (CNS) side effects are well described with amantadine, particularly in elderly patients. In comparison, rimantadine is associated with a considerably reduced rate of CNS side effects [45] . In mild to moderate uncomplicated disease, the reported benefits of early treatment (< 48 h of symptom onset) with NI have been a shorter duration and severity of flu-like symptoms, and a reduced duration of viral shedding [46] [47] [48] [49] . More importantly, several trials and a few systematic reviews have shown that treatment with NI may reduce illness severity and the rate of lower respiratory tract complications [47, [50] [51] [52] . In a meta-analysis of 10 randomizedcontrolled trials, Kaiser et al. found that therapy with oseltamivir was effective in reducing the incidence of influenzarelated lower respiratory tract complications that required antibiotic use (4.6 % for oseltamivir vs. 10.3 % for placebo, P<0.001), independent of the presence of risk factors for complications [51] . More recently, Hernán et al. conducted another meta-analysis of 11 controlled trials, most of which were included in the previous meta-analysis by Kaiser et al. They found that treatment with oseltamivir significantly reduced influenza-related lower respiratory tract complications by 37 % (CI 95 % ) [52] . NI have also been reported to reduce the duration of hospitalization in severely-ill cases [53] , and also to reduce influenza-related mortality [54] . During the 2009 H1N1 Influenza A pandemic, several observational studies of hospitalized and critically-ill patients reported that treatment with oseltamivir reduced disease severity, complications and mortality [55] [56] [57] . In a Chinese study of 1291 patients with confirmed H1N1 influenza A infection, oseltamivir reduced the risk of developing pneumonia, even when administered after the first 48 h of symptom onset (OR 0.12, 95 % CI [0.08-0.18]) [58] . In another retrospective study of 304 hematopoietic stem cell transplant recipients with influenza infection, 161 patients had H1N1 influenza A confirmed infection, and both early and delayed administration of antiviral therapy was shown to be beneficial in terms of decreased risk of lower respiratory tract compromise (OR00.04, 95 % CI [0-0.2] vs. OR00.14, 95 % CI [0-0.7]) [59] . An intravenous form of zanamivir is under development, and was made available during the 2009 H1N1 influenza A pandemic for severely ill patients with highly suspected or confirmed oseltamivir-resistant infection that could not tolerate inhaled zanamivir. Several studies reported favorable outcomes with the use of IV zanamivir [60] [61] [62] [63] . During the 2009 swine influenza pandemic, the FDA briefly authorized the emergency use of peramivir, an investigational NI that is administered intravenously [64] . In one study, a single dose of 300 mg or 600 mg of IV peramivir significantly reduced the time to alleviation of symptoms compared with placebo (HR 0.68 and 0.67 for the 300 mg and 600 mg doses respectively) [65] . Laninamivir octanoate, another long acting NI in development, was non-inferior to a five-day course of oral oseltamivir in adults with seasonal influenza in one study [66] . Experience with avian influenza H5N1, which has caused sporadic cases since 2004 with an elevated mortality rate, is much more limited and recommendations for its treatment are in most cases extrapolated from trials of seasonal influenza. There are, however, reports that suggest reduced severity and mortality with the administration of oseltamivir in patients with avian influenza [67] [68] [69] . In patients with pneumonia or with clinical failure to standard regime with oseltamivir, a higher dose of 150 mg bid for 10 days should be considered [24, 67] . In regions with adamantine-susceptible strains, combination therapy with a NI and an adamantine may also be considered with pneumonia or clinical worsening [24] . Therapy with oseltamivir should be administered even in the late course of influenza A (H5N1) infection since viral replication is more prolonged than with seasonal influenza [24] . The United States Advisory Committee on Immunization Practices (ACIP) and the CDC have recently published updated guidelines for the treatment of patients with confirmed or suspected influenza virus infections caused by either pandemic H1N1 or seasonal strains [41••, 70••] . According to these recommendations, those individuals with severe disease (requiring hospitalization or with evidence of lower respiratory tract infection) or those at high risk for complications (Table 1) should receive antiviral therapy even after 48 h of symptom onset. Adults with mild illness without high risk conditions who are younger than 65 years of age do not require treatment. If such individuals present within the first 48 h of illness, antiviral treatment can be considered in order to reduce the duration of illness. In all cases, decisions to administer antiviral medications for influenza treatment should be based on clinical and epidemiologic grounds, and never be delayed because of confirmatory laboratory tests [41••] . The usual dosing of oseltamivir for the treatment of influenza is 75 mg orally twice daily and of zanamivir is 10 mg (2 inhalations) twice daily. The recommended duration for antiviral treatment is 5 days, although longer treatment courses for patients who fail to improve after 5 days of treatment can be considered [70••] . Also, doubling the dose of oseltamivir to 150 mg orally twice daily has been suggested to be beneficial for some severely ill patients with H5N1 avian influenza [24, 67] . The FDA licensed oseltamivir for use in children 1 year of age and older in a liquid formulation at a dosage of 2 mg/kg/dose twice daily for 5 days. Zanamivir is not approved for use in children under 5. Oseltamivir and zanamivir are Pregnancy Category C drugs, but since influenza causes more severe disease and an increased rate of mortality among pregnant women, they should receive antiviral therapy with oseltamivir when indicated, since the potential benefit outweighs the theoretical risk to the fetus (Table 1) . Although oseltamivir-resistant seasonal H1N1 influenza A viruses have been identified since 2007 (the H274Y mutation) [71] , in 2009 the CDC reported that most circulating strains of the novel H1N1 influenza A virus were sensitive to the NI, oseltamivir and zanamivir, but that nearly all strains were resistant to the amantadines [41••, 72] . Consequently, the ACIP has advised against the use of M2 inhibitors for treatment of influenza, except in selected circumstances [41••] . It is recommended that patients with pandemic H1N1 influenza A who develop pneumonia be treated empirically for CAP according to published evidence-based guidelines, given the risk of secondary bacterial pneumonia [73] . In the presence of profound hypoxemia that has been refractory to routine mechanical ventilation, salvage therapies include neuromuscular blockade, inhaled nitric oxide, high-frequency oscillatory ventilation, extracorporeal membrane oxygenation (ECMO), and prone positioning ventilation [74•, 75] . Corticosteroids should not be used routinely, but may be considered for septic shock with suspected adrenal insufficiency requiring vasopressors [76] . Therapy for influenza-associated ARDS should be based upon published evidence-based guidelines for sepsis-associated ARDS, specifically including lung protective mechanical ventilation strategies [76] . The CDC recommends routine annual influenza vaccination for all persons 6 months of age and older. When vaccine supply is limited, vaccination efforts should focus on those groups with health conditions associated with increased risk of influenza complications [40••] . Antiviral drugs should not be used as a substitute for influenza vaccination. Their adjunctive use is appropriate in certain targeted populations at high risk for complications of influenza who are close contacts with suspected or confirmed cases [ Table 1 ]. Postexposure prophylaxis should only be used when antivirals can be started within 48 h of the most recent exposure. Recommended duration is 7 days after exposure, and the CDC recommends a minimum of 2 weeks for control of influenza outbreaks in long-term care facilities (i.e., nursing homes with elderly) and hospitals [70••] . The choice to offer post-exposure prophylaxis to otherwise healthy unvaccinated adults should be weighed against the risk of promoting antiviral drug resistance [41••, 1••]. A potentially fatal complication of influenza infection is the involvement of the lower respiratory tract caused directly by the influenza virus, and the development of secondary bacterial pneumonia. In these cases, and in patients who are at an increased risk for influenza infection complications, confirmation of etiology by laboratory testing is required in order to guide the initiation and duration of antiviral treatment, and for the implementation of infection control measures and surveillance. Recently, the emergence of novel influenza A strains that carry the risk for more severe disease regardless of age or previous health status, has prompted the development of quick and reliable laboratory tests in an effort to optimize their management and reduce morbidity and mortality. Pneumonia related to the 2009 influenza A pandemic was found in many cases to be rapidly progressive, leading to respiratory failure which in many cases was underestimated by commonly used pneumonia severity scores. Given the limited sensitivity of RIDT and immunofluorescence assays, confirmation of pandemic H1N1 influenza A infection can only be made by rRT-PCR or viral culture. Although annual immunization is the most important preventive measure, NI are the agents of choice for chemoprophylaxis in selected high risk patients, and for treatment. Treatment with NI beyond 48 h of symptoms should be considered only for patients with severe disease. Disclosure No potential conflicts of interest relevant to this article were reported. Open Access This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited. Papers of particular interest, published recently, have been highlighted as: • Of importance, •• Of major importance

1,3-Diphenyl-4,5-dihydro-1H-pyrazol-5-one

In the title pyrazolone derivative, C(15)H(12)N(2)O, the five-membered ring is approximately planar (r.m.s. deviation = 0.018 Å), and the N- and C-bound benzene rings are inclined to this plane [dihedral angles = 21.45 (10) and 6.96 (10)°, respectively] and form a dihedral angle of 20.42 (10)° with each other. Supra­molecular layers are formed in the crystal structure via C—H⋯O and C—H⋯N inter­actions, and these are assembled into double layers by C—H⋯π and π–π inter­actions between the pyrazole and C-bound benzene rings [ring centroid–centroid distance = 3.6476 (12) Å]. The double layers stack along the a axis being connected by π–π inter­actions between the N- and C-bound benzene rings [ring centroid–centroid distance = 3.7718 (12) Å].

In the title pyrazolone derivative, C 15 H 12 N 2 O, the fivemembered ring is approximately planar (r.m.s. deviation = 0.018 Å ), and the N-and C-bound benzene rings are inclined to this plane [dihedral angles = 21.45 (10) and 6.96 (10) , respectively] and form a dihedral angle of 20.42 (10) with each other. Supramolecular layers are formed in the crystal structure via C-HÁ Á ÁO and C-HÁ Á ÁN interactions, and these are assembled into double layers by C-HÁ Á Á andinteractions between the pyrazole and C-bound benzene rings [ring centroid-centroid distance = 3.6476 (12) Å ]. The double layers stack along the a axis being connected byinteractions between the N-and C-bound benzene rings [ring centroid-centroid distance = 3.7718 (12) Å ]. For the therapeutic importance of pyrazoles, see: Sil et al. (2005) ; Haddad et al. (2004) . For their diverse pharmacological activities, see: Bekhit et al. (2012) ; Castagnolo et al. (2008) ; Ramajayam et al. (2010) . For background to the synthesis, see: Nef (1891) ; Katritzky et al. (1997) ; Wardell et al. (2007) ; de Lima et al. (2010) . For evaluation of tautomeric forms using NMR MO calculations and crystallography, see: Feeney et al. (1970) ; Hawkes et al. (1977); Freyer et al. (1983) ; Dardonville et al. (1998) ; Kleinpeter & Koch (2001) ; Bechtel et al. (1973a,b) ; Chmutova et al. (2001); Wardell et al. (2007) ; Gallardo et al. (2009); Ding & Zhao (2010) . For a previous synthesis, see : Kimata et al. (2007) . For a recently reported structure, see: Wardell et al. (2012) . Table 1 Hydrogen-bond geometry (Å , ). Cg1 is the centroid of the C10-C15 ring. Symmetry codes: (i) x; Ày þ 1 2 ; z þ 1 2 ; (ii) x; Ày þ 3 2 ; z þ 1 2 ; (iii) Àx þ 1; Ày þ 1; Àz þ 2. Data collection: COLLECT (Hooft, 1998) ; cell refinement: DENZO (Otwinowski & Minor, 1997) and COLLECT; data reduction: DENZO and COLLECT; program(s) used to solve structure: SHELXS97 (Sheldrick, 2008) ; program(s) used to refine structure: SHELXL97 (Sheldrick, 2008) ; molecular graphics: ORTEP-3 (Farrugia, 1997) and DIAMOND (Brandenburg, 2006) ; software used to prepare material for publication: publCIF (Westrip, 2010) . Pyrazoles are key structures in numerous compounds of therapeutic importance (Sil et al., 2005 , Haddad et al., 2004 . Compounds containing this ring system are known to display diverse pharmacological activities, for example as antimalarial agents (Bekhit et al., 2012) , anti-tuberculosis agents (Castagnolo et al., 2008) , and as SARS-coronavirus protease inhibitors (Ramajayam et al., 2010) . A general route to pyrazole derivatives involves reaction of an arylhydrazine, ArNHNH 2 , with a β-dicarbonyl compound, R/COCH 2 COX.. This reaction provides initially a hydrazone derivative, RNHN=CR/CH 2 COX, I ( Fig. 1 ), which can be isolated but which readily undergoes cyclization to a pyrazone derivative, II ( Fig.1 ), (Nef, 1891; Katritzky et al., 1997; Wardell et al., 2007; de Lima et al., 2010) . Equilibrium involving tautomers of II in solution have been variously studied using NMR and IR spectroscopy and ab initio calculations. (Feeney et al., 1970; Hawkes et al., 1977; Freyer et al., 1983; Dardonville et al., 1998; Kleinpeter & Koch, 2001) . Crystal structures of various pyrazone compounds of forms IIa, IIb and IIc have been reported (see for example, Bechtel et al., 1973a; Bechtel et al., 1973b; Chmutova et al., 2001; Wardell et al., 2007; Gallardo et al., 2009; Ding & Zhao, 2010) . In continuation of recent studies (Wardell et al., 2012) , herein, the isolation the title compound [i.e. form IIc ( Fig. 1) ] from the reaction between PhNHNH 2 and PhCOCH 2 CO 2 Et in EtOH is described as is its crystal structure. The same tautomer was also isolated in the reaction between PhNHNH 2 and PhCOCH 2 CONHPh in EtOH. In the title compound, Fig. 2 , crystallography proves the IIc tautomer in the solid-state. The pyrazole ring is planar with a r.m.s. deviation for the fitted atoms of 0.018 Å; the maximum deviations from this plane are 0.015 (1) Å (for the N1 atom) and -0.015 (1) Å (C8). The N-and C-bound benzene rings are inclined to this plane forming dihedral angles of 21.45 (10) and 6.96 (10)°, respectively; the dihedral angle between the benzene rings is 20.42 (10)° consistent with a nonplanar molecule. In the crystal structure, supramolecular layers are formed in the bc plane through C-H···O and C-H···N interactions, Fig. 3 and Table 1 . Large 21-membered {···NC 4 H···N 2 CO···HC 2 O···HC 5 H} synthons are formed through these interactions. Layers are connected into double layers by C-H···π, Table 1 , and π-π interactions formed between the pyrazole and C-bound benzene rings [ring centroid···centroid distance = 3.6476 (12) Å, angle of inclination of 6.96 (10)° for symmetry operation 1 -x, 1 -y, 2 -z]. The double layers stack along the a axis being connected by π-π interactions between the N-and C-bound benzene rings [ring centroid···centroid distance = 3.7718 (12) Å, angle of inclination of 21.45 (10)° for symmetry operation -x, 1 -y, 1 -z]. A solution of PhNHNH 2 (1 mmol) and PhCOCH 2 CO 2 Et (1 mmol) in EtOH (15 ml) was refluxed for 1 h. The reaction mixture was maintained at room temperature and crystals of the titled compound were collected after 2 days, M.pt: 408-410 K: lit. value 409-411 K (Kimata et al., 2007) . IR and NMR spectra are in agreement with published data (Castagnolo The C-bound H atoms were geometrically placed (C-H = 0.95-0.99 Å) and refined as riding with U iso (H) = 1.2U eq (C). Owing to poor agreement one reflection, i.e. (3 1 1) , was removed from the final cycles of refinement. The maximum and minimum residual electron density peaks of 0.74 and 0.20 e Å -3 , respectively, were located 1.02 Å and 0.58 Å from the H8a and O1 atoms, respectively. Data collection: COLLECT (Hooft, 1998) ; cell refinement: DENZO (Otwinowski & Minor, 1997) and COLLECT (Hooft, 1998) ; data reduction: DENZO (Otwinowski & Minor, 1997) and COLLECT (Hooft, 1998) ; program(s) used to solve structure: SHELXS97 (Sheldrick, 2008) ; program(s) used to refine structure: SHELXL97 (Sheldrick, 2008) ; molecular graphics: ORTEP-3 (Farrugia, 1997) and DIAMOND (Brandenburg, 2006) ; software used to prepare material for publication: publCIF (Westrip, 2010) . The molecular structure of (I) showing the atom-labelling scheme and displacement ellipsoids at the 50% probability level. A view in projection down the c axis of the crystal packing in (I). The C-H···O, C-H···N, C-H···π and π-π interactions shown as orange, blue, brown and purple dashed lines, respectively. where P = (F o 2 + 2F c 2 )/3 (Δ/σ) max < 0.001 Δρ max = 0.74 e Å −3 Δρ min = −0.20 e Å −3 Special details Geometry. All s.u.'s (except the s.u. in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell s.u.'s are taken into account individually in the estimation of s.u.'s in distances, angles and torsion angles; correlations between s.u.'s in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell s.u.'s is used for estimating s.u.'s involving l.s. planes. Refinement. Refinement of F 2 against ALL reflections. The weighted R-factor wR and goodness of fit S are based on F 2 , conventional R-factors R are based on F, with F set to zero for negative F 2 . The threshold expression of F 2 > 2σ(F 2 ) is used only for calculating R-factors(gt) etc. and is not relevant to the choice of reflections for refinement. R-factors based on F 2 are statistically about twice as large as those based on F, and R-factors based on ALL data will be even larger. x y z U iso */U eq O1 0.22694 ( (9) 0.0201 (9) −0.0037 (6) 0.0061 (7) 0.0006 (7) C2 0.0271 (9) 0.0236 (9) 0.0244 (9) −0.0011 (7) 0.0113 (7) 0.0003 (7) (7) C5 0.0235 (9) 0.0233 (9) 0.0294 (10) −0.0014 (7) 0.0053 (7) 0.0042 (7) C6 0.0216 (9) 0.0230 (9) 0.0243 (9) −0.0013 (7) 0.0086 (7) −0.0014 (7) C7 0.0217 (9) 0.0170 (8) 0.0221 (9) −0.0005 (7) 0.0099 (7) −0.0007 (7) C8 0.0231 (9) 0.0174 (9) 0.0207 (8) 0.0006 (7) 0.0094 (7) 0.0010 (7) C9 0.0170 (8) 0.0201 (9) 0.0195 (8) 0.0000 (6) 0.0079 (7) 0.0015 (7) C10 0.0191 (8) 0.0230 (9) 0.0214 (9) −0.0013 (7) 0.0106 (7) 0.0000 (7) C11 0.0214 (9) 0.0241 (9) 0.0233 (9) 0.0026 (7) 0.0086 (7) 0.0018 (7) C12 0.0272 (9) 0.0235 (9) 0.0294 (10) 0.0012 (7) 0.0140 (8) −0.0040 (7) C13 0.0286 (10) 0.0316 (11) 0.0217 (9) −0.0024 (8) 0.0108 (8) −0.0062 (7) C14 0.0273 (10) 0.0305 (10) 0.0199 (9) 0.0018 (8) 0.0093 (7) 0.0035 (7) C15 0.0257 (9) 0.0205 (9) 0.0232 (9) 0.0012 (7) 0.0120 (7) 0.0016 (7) Geometric parameters (Å, º) Symmetry codes: (i) x, −y+1/2, z+1/2; (ii) x, −y+3/2, z+1/2; (iii) −x+1, −y+1, −z+2.

The CEA/CD3-Bispecific Antibody MEDI-565 (MT111) Binds a Nonlinear Epitope in the Full-Length but Not a Short Splice Variant of CEA

MEDI-565 (also known as MT111) is a bispecific T-cell engager (BiTE®) antibody in development for the treatment of patients with cancers expressing carcinoembryonic antigen (CEA). MEDI-565 binds CEA on cancer cells and CD3 on T cells to induce T-cell mediated killing of cancer cells. To understand the molecular basis of human CEA recognition by MEDI-565 and how polymorphisms and spliced forms of CEA may affect MEDI-565 activity, we mapped the epitope of MEDI-565 on CEA using mutagenesis and homology modeling approaches. We found that MEDI-565 recognized a conformational epitope in the A2 domain comprised of amino acids 326–349 and 388–410, with critical residues F(326), T(328), N(333), V(388), G(389), P(390), E(392), I(408), and N(410). Two non-synonymous single-nucleotide polymorphisms (SNPs) (rs10407503, rs7249230) were identified in the epitope region, but they are found at low homozygosity rates. Searching the National Center for Biotechnology Information GenBank® database, we further identified a single, previously uncharacterized mRNA splice variant of CEA that lacks a portion of the N-terminal domain, the A1 and B1 domains, and a large portion of the A2 domain. Real-time quantitative polymerase chain reaction analysis of multiple cancers showed widespread expression of full-length CEA in these tumors, with less frequent but concordant expression of the CEA splice variant. Because the epitope was largely absent from the CEA splice variant, MEDI-565 did not bind or mediate T-cell killing of cells solely expressing this form of CEA. In addition, the splice variant did not interfere with MEDI-565 binding or activity when co-expressed with full-length CEA. Thus MEDI-565 may broadly target CEA-positive tumors without regard for expression of the short splice variant of CEA. Together our data suggest that MEDI-565 activity will neither be impacted by SNPs nor by a splice variant of CEA.

Carcinoembryonic antigen (CEA; CEACAM5; CD66e) is a glycosylated human oncofetal antigen that belongs to the CEArelated cell adhesion molecule (CEACAM) family of the immunoglobulin (Ig) gene superfamily [1, 2] . CEA is closely related to CEACAM1, CEACAM3, CEACAM4, CEACAM6, CEACAM7, and CEACAM8. Carcinoembryonic antigen has been suggested to mediate cell-cell adhesion, facilitate bacterial colonization of the intestine, and protect the colon from microbial infection by binding and trapping infectious microorganisms [3] . CEA is expressed at low levels in normal tissues of epithelial origin in a polarized manner; found only at the luminal portion of the cell but not at the basolateral surface [3] . In contrast, expression of CEA is frequently high in carcinomas, including colon, pancreatic, gastric, esophageal, lung, breast, uterine, ovarian, and endometrial cancers [3] . Cancer cells not only lose polarized (luminal) expression of CEA, but actively cleave CEA from their surface by phospholipases, an action that results in serum concentrations of CEA that can approach 5 mg/mL [3, 4, 5] . Serum CEA levels may be monitored to detect a response to anticancer therapy, or disease recurrence in colorectal cancer [6] , and serves as a prognostic indicator in patients with gastrointestinal cancers, where elevated levels indicate a poor prognosis and correlate with reduced overall survival [7, 8, 9] . Cell-bound CEA has served as a target for tumor imaging and anti-cancer therapies. Clinical studies have demonstrated that radiolabeled anti-CEA antibodies and antibody fragments, such as the Food and Drug Administration-approved, Tc-99m-labeled, anti-CEA Fab arcitumomab (CEA-ScanH), can be successfully used as imaging reagents to specifically localize CEA-expressing solid cancers [10, 11, 12] . Anti-CEA radio-immunoconjugate anti-bodies have also been shown to be potentially efficacious for the treatment of patients with metastatic colorectal cancer [13] . In addition, CEA-specific antibody-directed enzyme prodrug therapy and CEA-based vaccine approaches have been developed to treat cancers [14, 15] . As a novel CEA-directed therapy, we have developed a CEAtargeting bispecific single-chain antibody of the bispecific T-cell engager (BiTE) class termed MEDI-565 (also known as MT111) [16, 17, 18] . BiTE antibodies are a unique subclass of bispecific antibodies that contain one single chain variable fragment (scFv) with specificity for a tumor associated antigen molecularly fused to another scFv with specificity for CD3 on T cells [19] . Highly potent and specific tumor cell lysis is triggered only when BiTE antibodies bind both epitopes simultaneously, resulting in directing T cells to the tumor cells and activating the T cell through the CD3/T cell receptor (TCR) complex [20] . Notably, activation of T cells is independent of TCR specificity, peptide antigen presentation, and co-stimulatory signals [21] . As a result of T cell activation, granzymes and perforin are mobilized to the tumor cell-T cell interface and mediate an apoptotic killing of target cells; FAS ligand expression may also contribute to the induction of apoptosis through engagement of FAS on tumor cells [22, 23, 24] . BiTE antibodies activate both CD4+ and CD8+ T cell subsets [23, 25, 26, 27, 28] ; both subsets of T cells contribute to tumor cell killing at relatively low effector T cell:target tumor cell (E:T) ratios [22, 29] . Cytokine secretion and the upregulation of cell surface IL-2 receptor by T cells (measured experimentally using antibodies recognizing the IL-2Ra chain/CD25) occur simultaneously and accompany tumor cell killing [18, 22, 25] . Both potentiate further T cell activation and proliferation, facilitating the subsequent engagement and killing of additional tumor cells bound by a BiTE antibody. BiTE antibodies not only induce the potent lysis of human cancer cell lines in vitro, but have also been shown to mediate the killing of primary human tumor cells by cancer patient T cells ex vivo [17, 25, 30] . Furthermore, BiTE antibodies promote in vivo tumor regression in numerous pre-clinical animal models [16, 31, 32, 33, 34, 35, 36] and have shown signs of clinical benefit in patients with cancer [37, 38, 39] . MEDI-565 itself is composed of a humanized single-chain antibody recognizing human CEA connected by a short flexible linker to a de-immunized single-chain antibody specific for CD3, and shares the characteristics of the BiTE antibody class as described above [16, 17, 18] . It specifically binds to CEA and not to other CEACAM family members [16] . Upon concurrent binding to CEA on cancer cells and CD3 on T cells, MEDI-565 mediates T cell activation and the subsequent killing of CEA-expressing target cells in a perforin-and granzyme-dependent manner that is insensitive to concentrations up to 5 mg/mL of soluble CEA [16, 17, 18] . Furthermore, intravenous and subcutaneous administration of MEDI-565 had anticancer activity in severe combined immunodeficient mouse xenograft models of various human cancers [18] ; growth inhibition of the cancers was contingent on both the presence of un-stimulated human T cells and the expression of CEA on the cancer cells [18] . MEDI-565 is currently in phase I clinical trials (clinicaltrials.gov identifier: NCT01284231) for the treatment of gastrointestinal adenocarcinomas. To understand the molecular basis of human CEA recognition by MEDI-565 and the impact of amino acid polymorphisms and splice variants of CEA on the activity of MEDI-565, we sought to identify the CEA domain and the critical residues involved in MEDI-565 binding. The CEA-specific arm of MEDI-565 is a humanized version of the murine antibody A5B7 [40, 41] the binding epitope of which is unknown [40, 42] . We mapped its epitope using mutagenesis and computational modeling approaches. We further investigated what impact a CEA splice variant which lacks a large portion of the identified epitope could have on the binding and activity of MEDI-565. The plasmid containing the full-length CEA sequence was purchased from Open Biosystems (Thermo Fisher Scientific, Huntsville, AL). The CEA complementary DNA (cDNA) was subcloned using sequence specific primers into a modified version of the lentiviral vector pCDH1-CMV-MCS-EF1-Puro (System Biosciences, Mountain View, CA) in which the puromycin-resistance casette was replaced with a blasticidin-resistance cassette (pCDH1-CMV-MCS-EF1-Blast) and the CEA sequence was verified by DNA sequencing. The plasmid containing the sequences of the CEA splice variant (cDNA clone DKFZp781M2392) was purchased from ImaGenes GmbH (Source BioScience UK Limited, Nottingham, UK) in association with B Bridge International (Cupertino, CA). The CEA splice variant was cloned into the puromycin resistance lentiviral vector pCDH1-HCS1-EF1-Puro (System Biosciences) using sequence-specific primers and the resulting clones were sequence-verified. FreeStyle human embryonic kidney (HEK293) F cells (American Type Culture Collection, Manassas, VA) were cultured in FreeStyle 293 expression medium (Life Technologies/Invitrogen, Carlsbad, CA). Dihydrofolate reductase deficient (DHFR-) Chinese hamster ovary (CHO) cells or CHO DHFR-cells (American Type Culture Collection, Manassas, VA) were cultured in RPMI 1640 media (Invitrogen) containing 10% fetal bovine serum (FBS; Invitrogen) in a humidified cell culture incubator at 37uC and 5% CO 2 . CHO cells expressing the CEA splice variant (CHO SV CEA) were created by infection of CHO DHFR-cells with the pCDH1-HCS1-EF1-Puro-CEA splice variant lentiviral vector, and selected by culturing in medium containing 5 mg/mL puromycin (EMD Chemicals Group, Merck KGaA, Darmstadt, Germany) for 72 hours. CHO cells expressing the full-length CEA sequence (CHO FL CEA) were created by infecting CHO DHFRcells with the lentiviral expression vector pCDH1-CMV-MCS-EF1-Blast-CEA, and selected by culturing in medium containing 10 mg/mL blasticidin (Invitrogen) for 10 days. Likewise, cells expressing both the CEA splice variant and full-length CEA (CHO FL+SV CEA) were created by infecting CEA splice variantexpressing cells with the pCDH1-CMV-MCS-EF1-Blast-CEA lentiviral vector followed by blasticidin selection. The anti-CEACAM5 specific monoclonal antibody (mAb) clone 26/3/13 was purchased from Genovac GmbH (Aldevron, Freiburg, Germany). Propidium iodide (PI) was purchased from Sigma-Aldrich (St. Louis, MO). Construction and production of MEDI-565 and the Control BiTE (MEC14 BiTE) has been described [17] . Twenty pancreatic primary tumors were purchased from Asterand, Inc (Detroit, MI). The panel included 13 adenocarcinomas (stage I-IV), 5 pancreatic endocrine tumors and 2 benign pancreatic adenomas. Four of the 20 pancreatic tumor tissue samples had matched normal adjacent pancreatic tissue samples. Patient age ranged from 23 to 77 years. All samples were freshly frozen and collected before the initiation of any cancer treatment. Tumor samples were macrodissected to remove normal tissue, and tumor purity in all samples was greater than 85%. Normal samples were macrodissected to remove non-glandular tissue. Cancer tissue cDNA arrays: HCRT101 (Colon Cancer), HGRT101 (Gastroesophageal Cancer), PNRT101 (Pancreatic Cancer), HLRT101 (lung cancer), and BCRT101 (breast cancer) were purchased from OriGene Technologies (Rockville, MD). Each array contains cDNAs from 5 to 8 normal tissues and 19 to 42 cancer tissues. The tumor stage ranged from stage I to IV and the tumor tissues were comprised of 50-90% tumor. Cells were lysed in RIPA lysis buffer (Boston Bioproducts, Ashland, MA) and the resulting protein was quantified using the BCA protein assay kit (Thermo Fisher Scientific, Huntsville, AL). Cell lysates were mixed with NuPAGEH LDS 4X LDS sample buffer (Invitrogen) to 1X concentration; ten micrograms of total protein from each lysate was analyzed on a pre-cast NuPAGEH NovexH 10% polyacrylamide Bis-Tris gel (Invitrogen) together with the NovexH Sharp Pre-stained Protein Standard (Invitrogen) using NuPAGEH Tris-acetate SDS running buffer (Invitrogen). Protein from each gel was transferred to a polyvinylidene fluoride membrane using an iBlotH Dry Blotting System (Invitrogen) and 2X NuPAGEH transfer buffer (Invitrogen) containing 20% methanol and blocked with 3% bovine serum albumin (Sigma) in PBS pH 7.4 (Invitrogen). To detect CEA, each blot was probed overnight with 1 mg/mL of the CEACAM5-specific mAb clone 26/3/13 (Genovac), washed with PBS pH 7.4 containing 0.1% Tween-20, and subsequently probed for 1 hour with a horseradish peroxidase-conjugated donkey anti-mouse IgG secondary antibody (Jackson ImmunoResearch Laboratories, West Grove, PA). After 5 washes with PBS pH 7.4 containing 0.1% Tween-20, each blot was exposed to SuperSignal West Pico Chemiluminescent Substrate (Thermo Fisher Scientific) and exposed to BioMax MS Kodak film (Sigma) for detection of chemiluminescence resulting from bound anti-CEACAM5 antibody. Equal amounts of protein loaded into each lane of the gel were controlled by detecting the glyceraldehyde 3-phosphate dehydrogenase (GAPDH) protein using a rabbit anti-human GAPDH polyclonal antibody (Trevigen, Gaithersburg, MD) followed by a horseradish peroxidaseconjugated donkey anti-rabbit IgG (Jackson ImmunoResearch). The cDNA encoding the CEA deletion variants DelA1-A2, Del A1, DelB1, DelA2, and DelB2-A3 and the swap variants KO_A, KO_B, KO_C, KI_A, KI_B, and KI_C (see Figures 1, 2, 3, and 4 for domain definition and variant nomenclature) were assembled and amplified by overlapping extension polymerase chain reaction (PCR) using an in-house full-length CEA cDNA clone, pNEO human CEA-GPI, as a template. The CEA variants and the fulllength CEA cDNA were cloned into a mammalian expression vector pcDNA3.1 (Invitrogen). Expression vectors encoding the CEA deletion and swap mutants were transiently transfected for expression of glycophosphatidyl inositol (GPI)-anchored proteins using FreeStyle HEK293 F cells (Invitrogen). One day prior to transfection, HEK293 F cells were seeded at a density of 0.7610 6 cells/mL. Three and a half micrograms of each expression vector were transfected into 5 mL of a suspension of HEK293 F cells using 5 mL of 293fectin transfection reagent (Invitrogen). Transfected cells were incubated with 5 mg/mL of MEDI-565 for 1 hour. For the detection of bound MEDI-565, which possesses a six-histidine tag at its carboxy-terminus, the cells were washed three times with PBS and incubated with 1 mg/mL of an Alexa FluorH 488conjugated anti-penta-His antibody (Qiagen, Valencia, CA) for 30 minutes, and then analyzed for binding using the LSRII flow cytometer (BD Biosciences, San Jose, CA). Expression of CEA deletion mutants was monitored with a goat anti-CEA polyclonal antibody (R&D Systems, Minneapolis, MN) that was detected by an R-phycoerythrin-conjugated anti-goat IgG (Invitrogen). Site-directed mutagenesis and computational homology modeling were implemented to identify amino acids within CEA that are critical for MEDI-565 binding. The following amino acids in the epitope-containing segments A and C of the A2 domain (see Results and Discussion section) that differ from those in the A3 domain were mutated in clusters or individually to encode the corresponding amino acids of the [44] . This model was used as a template to engineer two additional mutants. These two mutants 1) substituted clusters V 388 G 389 P 390 E 392 and I 408 N 410 of the A2 domain with the corresponding clusters of the A3 domain (KO_VGPE+IN), and 2) grafted clusters F 326 T 328 N 333 , V 388 G 389 P 390 E 392 , and I 408 N 410 of the A2 domain into the A3 domain (KI_FTN+VGPE+IN), respectively. All mutants were assembled by overlapping extension PCR using the truncated mutants A2 or A3 as a template, which encodes for the N-terminal domain, the A2 or A3 domain, and the GPI region. They were cloned into the mammalian expression vector pcDNA3.1 (Invitrogen). Expression of these mutants and binding analyses were conducted as listed above for the deletion and swap mutants. Full-length mRNA transcript sequences for CEA (NM_004363.2) and the CEA splice variant (DKFZp781M2392) were retrieved from the NCBI Reference Sequences database. For the full-length CEA assay we targeted the splice junctions of exons 3 and 4 to design the gene specific probe. For the CEA splice variant assay we targeted the splice junctions of exons 2 and 5 exon to design the gene specific probe. All primer/probes were imported into the Primer Express (Applied Biosystems, Foster City, CA) software tool to ensure the optimal design for utilization in the TaqMan Gene Expression assay procedure. All probes were designed to incorporate a minor groove binding moiety (MGB), and were labeled with a fluorescent dye (FAM) for detection and as a non-fluorescent quencher. Primers and probes were custom ordered from Applied Biosystems. Sequences for all primer/probe combinations are as follows: CEA full-length: The sequence of the probe is 59-CAGGCG-CAGTGATTCA-39. The sequence of the forward primer 59-GAAACCCAGAACC-CAGTGAGT-39. The sequence of the reverse primer is 59 GCCATAGAGGA-CATTCAGGATGAC-39. CEA splicing variant: The sequence of the probe is; CEA 59-ATGCATCCCTGCTGATCC-39. The sequence of the forward primer is 59-CGCATA-CAGTGGTCGAGAGATAATA-39. The sequence of the reverse primer is 59-CGCTGTGGTCAA-CACTTAATTTGT-39. Real-time qPCR was conducted with the BioMark TM Dynamic Array Microfluidics System (Fluidigm Corporation, South San Francisco, CA) using RNA isolated from frozen pancreatic tissues. First, total RNA was extracted from frozen slide sections of pancreatic tissue classified as normal (NAT, normal adjacent to tumor), adenocarcinoma, benign adenoma, or endocrine tumor using the ZR RNA MicroPrep kit (Zymo Research, Orange, CA). RNA purity and concentration were determined spectrophotometrically. RNA quality was assessed on an Agilent 2100 Bioanalyzer using the RNA 6000 Nano LabChipH (Agilent Technologies, Santa Clara, CA). Single stranded cDNA was generated from total pancreas RNA using the SuperScriptH III First-Strand Synthesis SuperMix (Invitrogen). cDNA samples were pre-amplified using TaqMan Pre-Amp Master Mix (Invitrogen), according to the manufacturer's instructions. Reactions were cycled with the recommended program for 14 cycles and then diluted 1:5 with TE buffer. Preamplified cDNA was either utilized immediately or stored at 220uC until processed. Samples and TaqMan Gene Expression assays (Applied Biosystems, Carlsbad, CA) were load onto a 96.96 dynamic array (Fluidigm) according to the manufacturer's instructions. The prepared array was loaded on the BioMark TM Real-Time PCR System (Fluidigm) for thermal cycling (10 min at 95uC followed by 40 cycles of 95uC for 15 sec and 1 min at 60uC). Upon completion of the qPCR, Fluidigm Real-Time PCR Analysis software was used to generate threshold cycle (Ct) values. A Ct cutoff value of 24, below which the samples were considered to be positive and above which the samples were considered to be negative for transcript expression, was empirically selected. TissueScan TM Disease Tissue quantitative polymerase chain reaction (qPCR) arrays (Origene Technologies, Rockville, MD) were employed to determine the expression of full-length and CEA splice variant (SV) transcripts in normal and cancerous tissues of various stages and grades. For each tissue cDNA array, lyophilized cDNA was resuspended in 2.5 mL of ribonuclease-free water. Plates were sealed, vortexed, and centrifuged (1200 g for 1 min) to ensure resuspension of the full cDNA sample. A 0.2X mixture of the TaqMan assays (Applied Biosystems) was then created according to protocols supplied by the manufacturer for pre-amplification (one cycle of 95uC, 10 min followed by 14 cycles of 95uC 15 sec/60uC 4 min). Subsequent amplification and transcript detection (one cycle of 95uC, 20 sec followed by 40 cycles of 95uC, 1 sec/60uC, 20 sec) of the specific target transcripts was carried out using an ABI 7900HT Fast Real Time PCR Instrument (Applied Biosystems). Upon completion of the qPCR, SDS 2.2 Software (Ericsson, Stockholm, Sweden) was used to generate threshold cycle (Ct) values. A Ct cutoff value of 30, below which the samples were considered to be positive and above which the samples were considered to be negative for transcript expression, was empirically selected. As positive controls, cDNAs from CHO cells expressing either full-length CEA (CHO FL CEA) or the CEA splice variant (CHO SV CEA) were used. Complementary DNA isolated from DHFR-deficient CHO cells lacking CEA expression was used as a negative control for qPCR. To measure the killing of human tumor cell lines by MEDI-565 or the control BiTE, cell lines were co-cultured with enriched CD3+ T cells and BiTE antibody as described above, and cell lysis was measured by the release of caspase-3 into the tissue culture media via a caspase-3 specific electrochemiluminescence assay (Meso Scale Discovery, Gaithersburg, MD). The statistical significance of differences in affinity (apparent K D ) and cytotoxicity (EC 50 ) values was calculated using two-tailed, parametric t tests calculated in GraphPad Prism software version 5.01 for Windows (GraphPad Software). MEDI-565 recognizes full length CEA; however, the corresponding epitope is unknown. CEA deletion mutants were generated to identify the domain of CEA to which MEDI-565 binds. The full-length CEA transcript (NCBI accession number NM_002483) contains 10 exons and 9 introns encoding a 702 amino acid (aa) protein composed of a 34 aa processed leader sequence, one IgV-like N-terminal domain, six immunoglobulin constant (IgC)-like domains denoted A1, B1, A2, B2, A3 and B3, and a C-terminal 17 aa peptide which is removed during GPI linkage [1, 2, 45, 46] (Figure 1, 2) . Five deletion mutants were constructed by removing the following IgC-like domains: A1, B1 and A2 domains (DelA1-A2), A1 domain (DelA1), B1 domain (DelB1), A2 domain (DelA2), and B2 and A3 domains (DelB2-A3) as shown in Figure 3A . The mutants were transiently expressed as GPI-anchored proteins on HEK293 F cells. The binding of MEDI-565 or the anti-CEA polyclonal control antibody to each of the deletion mutants was analyzed using flow cytometry. MEDI-565 did not recognize any of the deletion mutants which lack the A2 domain (DelA1-A2 and DelA2, Figure 3B ), but bound well to all mutants that contained it (DelA1, DelB1 and DelB2-A3, Figure 3B ). Additionally, when all but the A2 IgC-like domains were removed (A2; Figure 4B ), MEDI-565 still recognized the corresponding truncated CEA protein at a level similar to that measured for full-length CEA protein ( Figure 4C ). These results suggested that the epitope of MEDI-565 is localized in the A2 domain of CEA. The A3 domain is not involved in MEDI-565 binding despite being highly homologous with the epitope-containing A2 domain; therefore, swap-mutants were constructed by exchanging short segments of the A2 domain with the corresponding portions of the A3 domain to further refine the MEDI-565 binding epitope. Two truncated CEA mutants (A2 and A3) were engineered and used as templates for the construction of swap mutants. The A2 or A3 mutant is comprised of the N-domain, the A2 or A3 domain, and the GPI region. An alignment of the amino acid sequences of the A2 and A3 domains identified differences in their aa composition. These regions of sequence diversity were divided into three shorter segments labeled as A (aa 326 to 349), B (aa 360 to 382) and C (aa 388 to 410) ( Figure 4A ). Six swap mutants were generated by swapping segments A, B or C of the A3 domain into the A2 domain (KO mutants), or by swapping segments of the A2 domain into the A3 domain (KI mutants) as shown in Figure 4B . MEDI-565 did not bind to any of the KO or KI mutants lacking either the A or C segment of the A2 domain (KO_A, KO_C, KI_A, KI_B and KI_C, Figure 4B , C), but bound well to the swap variant which encoded both the A and C segments of the A2 domain (KO_B, Figure 4B , C). Some residual binding of MEDI-565 to the KO_C mutant could be detected. Together with . Swap-mutants of human CEA. A, amino acid sequence alignment of the A2 and A3 domains of CEA. Sequence homology analysis revealed 21 amino acids that differed between these two domains (amino acids boxed). Three segments, A, B, and C, were defined in the A2 and A3 domains to generate swap-mutants. B, a schematic display of swap mutants that were constructed by exchanging segments A, B, or C between the A2 (open boxes) and A3 (grey boxes) domains using the truncated mutant A2 or A3 as a template which encodes the N-domain, the A2 or A3 domain, and the GPI region. C, flow cytometry analysis of binding of MEDI-565 to deletion mutants expressed on the surface of HEK293 F cells. All mutants were expressed well as monitored by anti-CEA polyclonal antibody. MEDI-565 did not bind well to any of the knock-out (KO) or knock-in (KI) mutants which lack either the A or C segment of the A2 domain (KO_A, KO_C, KI_A, KI_B and KI_C), but bound well to the one variant which encoded both the A and C segments of the A2 domain (KO_B). doi:10.1371/journal.pone.0036412.g004 MEDI-565 lack of binding to KO_A mutant, this data indicated that segment C significantly contributed to MEDI-565 binding, but to a lesser degree than segment A. Due to the high degree of identity between segment B in the A2 and the A3 domains, direct involvement of this segment in MEDI-565 binding could not be entirely ruled out using a swap mutant-based approach. It is however unlikely for the following reasons 1) MEDI-565 did not bind to the mutants encoding the B segment of A2 domain in conjunction with either A or C segment of the A2 domain (KO_A or KO_C, Figure 4 ); 2) The binding level of MEDI-565 to the mutant encoding both the A and the C segments of A2 domain (KO_B) was comparable to the signal of both the full-length CEA and the A2 deletion mutant (Figure 4) ; and 3) The modeled structure of the A2 domain using murine CEACAM1 as a template revealed that segment B was spatially distal from critical residue N 333 (see following section). Therefore, the swap-mutants revealed that MEDI-565 bound to a nonlinear epitope in the A2 domain of CEA, which is comprised of two segments, namely 326-349 (segment A) and 388-410 (segment C). Site-directed mutagenesis and computational homology modeling were employed to identify critical residues within segments A and C that are essential for the binding of MEDI-565. The amino acids of segments A and C of the A2 domain that differed from the A3 domains were replaced with the corresponding A3 residues encoding several substitutions at a time: Figure 4A ). The binding of MEDI-565 was substantially decreased to the variants in which residue N 333 was mutated (KO_FTN and KO_NE), but bound well to the other mutants ( Figure 5A ). Furthermore, replacing only residue N 333 with either its counterpart residue Lys in the A3 domain (KO_N) or with Ala (N 333 to A) abolished MEDI-565 binding ( Figure 5C ). Mutating the residue F 326 or T 328 to Ala substantially decreased the binding of MEDI-565 (F 326 to A, T 328 to A), suggesting that they were also involved in the interaction with MEDI-565 but to a lesser extent than residue N 333 ( Figure 5C ). Taken together, our data demonstrated the importance of F 326 , T 328 , and N 333 . Finally, since E 338 is encoded in the KO_ELI variant which binds well to MEDI-565, we concluded that this residue is not energetically involved in MEDI-565 epitope. We had previously noted that grafting only segment A that includes the F 326 , T 328 , and N 333 residues of the A2 domain into the A3 domain (KI_A) did not result in MEDI-565 binding. This suggested that MEDI-565 binding epitope comprises additional critical residues and pointed towards it being nonlinear and conformational. The modeled structure of the A2 domain also revealed two clusters of amino acids V 388 G 389 P 390 E 392 and I 408 N 410 in segment C that were spatially close to the already identified critical residue N 333 in segment A ( Figure 5B ). This observation suggested that these amino acids could contribute to the binding of MEDI-565 in concert with N 333 . Knocking-out amino acids V 388 G 389 P 390 E 392 (KO_VGPE) or I 408 N 410 (KO_IN) separately ( Figure 5A ), or knocking-in the entire segment C which includes both clusters of V 388 G 389 P 390 E 392 and I 408 N 410 (KI_C) ( Figure 4C ) had no effect in abolishing or restoring MEDI-565 binding, respectively. However, knocking-out these amino acids together reduced (but did not completely abolish) MEDI-565 binding (KO_VGPE+IN, Figure 5C ). Thus, amino acids V 388 , G 389 , P 390 , E 392 , I 408 , and N 410 in segment C of CEA could also be involved in MEDI-565 binding. It is possible that some knock-out CEA variants exhibited an incorrect fold in or near the MEDI-565 epitope region, thereby losing their binding capacity. Therefore, we further confirmed the potential epitope regions identified with knock-out variants by using gain of function (knock-in) mutants. Indeed, ''knocking in'' segments A (326-349) and C (388-410) of the A2 domain into the A3 domain (KO_B/KI_A+C, KO_B encodes the same amino acids as KI_A+C) resulted in MEDI-565 binding comparable to full length CEA ( Figure 4C, 5C) . Furthermore, only grafting the three amino acids F 326 T 328 N 333 of segment A with the six amino acids V 388 G 389 P 390 E 392 and I 408 N 410 of segment C into the A3 domain (KI_FTN+VGPE+IN) also resulted in MEDI-565 binding comparable to full length CEA ( Figure 5C ). In summary, these results demonstrate that the epitope of CEA bound by MEDI-565 is a nonlinear, conformational epitope located in the A2 domain of CEA; it is comprised of two segments of amino acids 326 to 349 and 388 to 410 with critical amino acids F 326 , T 328 , N 333 , V 388 , G 389 , P 390 , E 392 , I 408 , and N 410 . The residue N 333 may contribute more to the binding of MEDI-565, since mutating it alone completely disrupted the interaction ( Figure 5C ). MEDI-565 binds to mature full-length CEA. However, polymorphisms and/or isoforms of CEA may alter the binding epitope and negatively affect the ability of MEDI-565 to bind. Once the epitope of CEA bound by MEDI-565 had been identified, the ability of MEDI-565 to recognize cancerous cells expressing polymorphisms and isoforms of CEA could be evaluated. Polymorphisms of CEA were surveyed using the NCBI singlenucleotide polymorphism (SNP) database (http://www.ncbi.nlm. nih.gov/projects/SNP). Two non-synonymous coding SNPs of CEA (rs10407503, rs7249230) were identified in the binding epitope of MEDI-565 (shown in Figure 2 ). The single-nucleotide C to A change in the SNP rs10407503 resulted in the amino acid change of Ala to Asp at the aa position 340. The single-nucleotide A to G change in the SNP rs7249230 encoded an aa change of Glu to Lys at the aa position 398. According to the SNP database, the minor allele frequencies in the population for rs10407503 are 0.014,0.267 and minor allele frequencies for rs7249230 are 0.03,0.3 in the population, respectively. However, the minor allele homozygosity rate for rs10407503 is close to 0 in both European and Asian populations and is 0.083 in Sub-Saharan African populations. The minor allele homozygosity rate for rs7249230 was from 0 to 0.068 in different populations. Since the homozygosity rates of both SNPs are very low, we anticipated the identified CEA SNPs to have little or no impact on MEDI-565 binding to CEA for these populations. In addition, the National Center for Biotechnology Information (NCBI) GenBankH database (http://www.ncbi.nlm.nih.gov/ genbank) was searched for splice variants of CEA. A single splice variant (NCBI accession number CR749337) from colon cancer tissue was identified. This transcript uses an alternative splice donor site in exon 2 and skips exons 3 and 4; thus, the translation of the transcript results in a 420 aa protein with an in-frame truncation from amino acids 116 to 396 of the full-length CEA. This truncation deletes a small portion of the N-terminal domain, the entire A1 and B1 domains, and a large portion of the A2 domain (Figure 1, 2) . The splice variant sequence codes for the same 34-aa processed leader sequence and C-terminal 17-aa peptide which is expected to be removed during GPI linkage in a similar manner as full-length CEA, and, after undergoing similar post-translational modifications as full-length CEA, is predicted to be a mature, GPI-anchored membrane protein of 369 amino acids. Another putative CEA splice variant involving novel splicing of exons 9, 10 and the intervening intron sequence has been detected in the peripheral blood of colon cancer patients by reverse transcription PCR [47] . However, sequences within the middle part of a 152 bp PCR product (base pairs 26-99) of the putative CEA isoform isolated from white blood cells did not match any part of the genomic sequence of CEA. Therefore, this PCR product more likely represents a PCR artifact and not a real CEA splice form. Consequently, we did not investigate this putative CEA splice variant further. The biological function and distribution of the CEA splice variant in normal and cancerous tissues is unknown. To determine the expression frequency of full-length and CEA splice variant transcripts in normal and cancerous human tissues, real-time qPCR was performed using both RNA isolated from frozen primary human pancreatic tissues and cDNA generated from frozen colorectal, lung, breast and pancreatic tissues purchased in the form of multi-tissue cDNA arrays. Each analysis used sequence-specific primers that specifically amplified either fulllength CEA or CEA splice variant sequences. Results (Table 1 ) of qPCR analysis using frozen primary pancreatic tissue specimens demonstrated that full-length CEA transcript was detected in normal human pancreatic tissues (3 of 4). Among diseased tissue, the full length transcript was found frequently in pancreatic adenocarcinoma (12 of 13) and less frequently in benign adenomas (1 of 2) and in pancreatic endocrine tumors (3 of 5), although the total number of specimens representing the latter two categories were small. Transcripts of the CEA splice variant were not detected in normal pancreatic tissues (0 of 4), benign adenomas (0 of 2) nor in endocrine tumors (0 of 5), and were rarely detected in pancreatic adenocarcinomas (1 of 13). Expression of the CEA splice variant transcript in the single positive adenocarcinoma specimen was concordant with full-length CEA transcript expression (Table S1) . Results from qPCR analysis using human tissue cDNA arrays ( Table 2) (Table S1 ). Thus, expression of the CEA splice variant transcripts varied in different cancers; however, it was always coexpressed with fulllength CEA transcripts. Although the CEA splice variant was infrequently found in pancreatic tumors, it was found at a high frequency in colorectal (98%) and gastroesophageal (50%) cancers, and to a lesser degree in lung (30%) and breast (12%) tumors. Due to its concordant expression with full length CEA in human tumors, we sought to understand the binding of MEDI-565 to the CEA splice variant and the role that this CEA isoform might play in targeting CEApositive tumors. The amino acids in full-length CEA important for MEDI-565 binding were found to be largely absent in the CEA splice variant, except amino acids I 408 and N 410 in segment C ( Figure 2 ). This observation suggested that binding of MEDI-565 to the CEA splice variant was unlikely to occur. To test this hypothesis, CHO cells were infected with lentivirus constructs directing the expression of full-length CEA (CHO FL CEA) or the CEA splice variant (CHO SV CEA), or both concurrently after sequential infection of cells with the CEA splice variant then the full-length CEA (CHO FL+SV CEA); full-length CEA and CEA splice variant protein expression were verified by western blotting ( Figure S1 ). As anticipated, MEDI-565 bound to cells expressing full-length CEA but not to cells expressing the CEA splice variant ( Figure 6A ). We note that a higher level of MEDI-565 binding was observed for cells expressing both the CEA splice variant and full length CEA together relative to cells expressing only full length CEA. This is likely due to different levels of expression efficiency in each independently infected cell line, although effects of the CEA splice variant protein on full length CEA protein levels or epitope accessibility in cells cannot be ruled out. Zhou et al [48] showed that the both the N-terminal domain and A3 mediated the interaction between CEA molecues on apposing cell surfaces; both domains are present within the CEA splice variant. Thus, we tested the hypothesis that co-expression of the CEA splice variant and full-length CEA proteins may, through heterophilic interactions, result in the MEDI-565 binding epitope being masked in the full-length form and subsequently prevent or reduce MEDI-565 binding. Co-expression of the CEA splice variant with full-length CEA on the same cells did not significantly affect the apparent binding affinity of MEDI-565 to full-length CEA (CHO FL CEA, apparent K D = 5.060.15 nM; CHO FL+SV CEA, apparent K D = 5.663.0 nM; p = 0.86). These results were expected since homophilic interactions between fulllength CEA proteins also do not appear to prevent the binding of MEDI-565 [17, 18] or other CEA-specific BiTE antibodies [16] . Consistent with the MEDI-565 binding data, CHO SV CEA did not trigger the activation of T cells from healthy donors in the presence of MEDI-565, as measured by the up-regulation of the IL-2Ra chain/CD25 protein on either CD8 or CD4 T cells ( Figure 6C and D) . This T cell activation marker has been shown previously to correlate temporally with the release of cytokines from T cells activated by MEDI-565 [18] and other BiTE antibodies [23, 25] . Likewise MEDI-565 also did not mediate the lysis of target cells expressing the splice variant ( Figure 6B ). In contrast, MEDI-565 activated T cells and induced killing of CHO FL CEA or CHO FL+SV CEA cells with similar levels of potency (EC 50 values: CHO FL CEA, EC 50 = 75635 ng/mL; CHO FL+SV CEA, EC 50 = 59643 ng/mL; p = 0.79). Somewhat higher maximal T cell activation and maximum achievable cytotoxicity levels were observed with CHO FL+SV CEA cells relative to CHO FL CEA cells, and may be explained by the higher full length CEA expression in the former, as described above. Studies with full length CEA and the splice variant were carried out in CHO cells due to their ease of infection with lentiviral constructs and their subsequent selection of CEA-expressing cells using flow cytometry. To demonstrate the killing specificity of MEDI-565 for CEA positive human tumor cells, we examined numerous human cancer cell lines for CEA expression and for their ability to be killed by MEDI-565 activated T cells. Consistent with the specificity observed for CHO cells, MEDI-565 mediated the killing of human tumor cells that expressed CEA, but not those that did not express cell surface CEA ( Figure S2 ). Additionally, a control BiTE did not induce T cell lysis of CEA positive or negative cell lines. To provide insights for CEA-targeted diagnostics and therapy and understand the specificity of anti-CEA mAbs, many efforts have been made to characterize the binding epitopes of these antibodies using various approaches [49, 50, 51, 52] . The most extensive investigations involve the classification of 52 anti-CEA mAbs into five distinct epitopes, Gold epitopes 1-5, by competitive binding analysis [50] . Further studies have correlated the Gold epitopes to the domains of CEA using different CEA fragment constructs [51, 53, 54] , however there is no precise localization of the epitopes to the amino acid sequence level. An attempt to identify the amino acid sequences corresponding to Gold epitopes using synthetic overlapping fifteen-mer peptides has failed [55] , suggesting that the Gold epitopes are conformational. In addition, immunohistochemistry and immuno flow cytometry were used to demonstrate a good correlation between the Gold epitope groups and their binding specificity: 1) mAbs that bind to Gold epitopes 2 and 3 were generally specific to CEA, reacting only with colon carcinoma, normal colon mucosa and normal gastric foveola; 2) mAbs in epitopes 4 and 5 were highly cross-reactive with different normal tissues possibly due to binding to CEA-related antigens; and 3) both specific and cross-reactive mAbs were found in epitope 1 [56] . We have mapped the epitope of the CEA-specific arm of MEDI-565 to the A2 domain comprised of two stretches of amino acids, 326-349 and 388-410. Interestingly, the MEDI-565 epitope on CEA belongs to the Gold epitope 2, which has been localized in the A2-B2 domains [53] and identified as a CEA-specific epitope group [56] . CEA comprises highly repetitive immunoglobulin domains of A1-B1, A2-B2, and A3-B3. Some anti-CEA mAbs bind repetitive epitopes of CEA [51] and some are highly cross-reactive, lacking specificity to CEA [56] . The anti-CEA arm of MEDI-565 is a humanized version of the murine antibody A5B7, which has been shown to specifically bind to CEA [40, 56] . Characterizing the epitope of MEDI-565 has led us to propose that the critical residues F 326 , T 328 , and N 333 mediate the fine specificity to CEA of MEDI-565 due to their uniqueness in the A2 domain of CEA when compared with all other immunoglobulin domains of CEA and other related molecules of the CEACAM family, including CEACAM1, CEACAM3, CEACAM4, CEACAM6, CEACAM7, and CEACAM8. After identifying the binding epitope of MEDI-565, we evaluated the impact of polymorphisms and isoforms in the A2 domain of CEA on the activity of MEDI-565. Two nonsynonymous SNPs were identified in the binding epitope of MEDI-565 but occurred at a very low frequency in the general population and were considered to have little or no impact on MEDI-565 binding to CEA. In addition, a single splice variant was identified lacking a portion of the A2 domain critical for MEDI-565 binding. Efforts to understand the association of splice variant expression to disease found that its expression occurred in a substantial percentage of primary human colorectal and gastro-esophageal tumors, and to a lesser extent in lung and breast tumors in a pattern that was concordant with full length CEA expression. However, since MEDI-565 did not bind the CEA splice variant due to loss of the antibody-binding epitope, it was clear that the splice variant itself is not a target for MEDI-565 in primary human tumors that also express full length CEA. Because CEA could participate in homotypic interactions on adjacent cells, it remained possible that expression of the CEA splice variant may interfere with the binding of MEDI-565 and the subsequent tumor cell lysis through its interaction with full length CEA. By coexpressing the CEA splice variant with full length CEA, we formally demonstrated that this was not the case, as MEDI-565 binding to full length CEA and potency of CEA-directed lysis were not significantly affected by simultaneous expression of the splice variant and full length CEA. Therefore, the expression of the CEA splice variant by primary human tumor cells is not anticipated to interfere with MEDI-565 binding to full-length CEA, nor should it inhibit MEDI-565-mediated T-cell killing of tumor cells expressing full-length CEA. However, discrimination of full-length CEA from the CEA splice variant may be important while monitoring the status of CEA positive tumors via changes in serum CEA levels in clinical study patients that receive MEDI-565. Figure S1 Western blot of CEA protein expressed by CHO cell lines. The full-length CEA protein is indicated by a filled arrowhead, and the CEA splice variant by an open arrowhead; both proteins were detected using a CEACAM5-specific mAb. Molecular weights (kilodaltons; KDa) of the protein standard are indicated to the left of the image. Lanes of CHO FL and CHO FL+SV cell lysates contain a band at ,100 kDa that is presumed to be a non-glycosylated form of CEA. Equal amounts of protein loaded into each lane of the gel were controlled by detecting

High Throughput Screening for Small Molecule Enhancers of the Interferon Signaling Pathway to Drive Next-Generation Antiviral Drug Discovery

Most of current strategies for antiviral therapeutics target the virus specifically and directly, but an alternative approach to drug discovery might be to enhance the immune response to a broad range of viruses. Based on clinical observation in humans and successful genetic strategies in experimental models, we reasoned that an improved interferon (IFN) signaling system might better protect against viral infection. Here we aimed to identify small molecular weight compounds that might mimic this beneficial effect and improve antiviral defense. Accordingly, we developed a cell-based high-throughput screening (HTS) assay to identify small molecules that enhance the IFN signaling pathway components. The assay is based on a phenotypic screen for increased IFN-stimulated response element (ISRE) activity in a fully automated and robust format (Z′>0.7). Application of this assay system to a library of 2240 compounds (including 2160 already approved or approvable drugs) led to the identification of 64 compounds with significant ISRE activity. From these, we chose the anthracycline antibiotic, idarubicin, for further validation and mechanism based on activity in the sub-µM range. We found that idarubicin action to increase ISRE activity was manifest by other members of this drug class and was independent of cytotoxic or topoisomerase inhibitory effects as well as endogenous IFN signaling or production. We also observed that this compound conferred a consequent increase in IFN-stimulated gene (ISG) expression and a significant antiviral effect using a similar dose-range in a cell-culture system inoculated with encephalomyocarditis virus (EMCV). The antiviral effect was also found at compound concentrations below the ones observed for cytotoxicity. Taken together, our results provide proof of concept for using activators of components of the IFN signaling pathway to improve IFN efficacy and antiviral immune defense as well as a validated HTS approach to identify small molecules that might achieve this therapeutic benefit.

There has been significant progress in the development of vaccines and therapeutics against viruses, but there are still major gaps in medical therapy for some of the most common types of viral infections. For these types of infections, vaccines can still be ineffective due to new and emergent strains and can exhibit significant off-target effects [1, 2] . Similarly, the efficacy of antiviral therapeutics can often be limited by pathogen resistance as another sign of the difficulty in keeping up with rapidly evolving viral genomes [3] [4] [5] [6] [7] [8] [9] . An alternative to agents that specifically and directly target the virus itself is the possibility of improving natural host defense against a broad range of viruses. Although antiviral defense exhibits significant complexity and redundancy, one system that stands out as a useful target for improvement is the one based on the action of interferons (IFNs). And within this IFN system, which is similarly complex, the STAT1 transcription factor is remarkable as a central component that is critical for the functional activity of each type of IFN ( Figure 1 ). Consequently, genetic loss of STAT1 function causes a marked susceptibility to viral infection in mice and humans [10] [11] [12] . Moreover, modification of STAT1 to a form) that improves the efficiency of IFN signal transduction can result in improved control of viral infection [13] . These observations indicate that the IFN-signaling pathway is subject to a so-called ''rheo-STAT'' adjustment wherein downregulation causes increased susceptibility to viral infection whereas up-regulation might lead to increased efficiencies for IFNstimulated gene (ISG) expression and control of infection [14] . In the present study, we aimed to mimic the beneficial actions of STAT1 modification with a small molecule that also enhances the activity of the IFN signaling pathway. We describe here the development of a high-throughput screening (HTS) system for novel small molecular weight compounds (so-called ''small molecules'') that might increase ISG expression and antiviral activity. To develop this screening system, we generated cell lines that stably express the human interferon-stimulated response element (ISRE) driving a luciferase reporter gene. The ISRE gene promoter element is responsible for type I IFN signaling that mediates host defense against a wide range of viruses [15, 16] . After establishing that the ISRE-reporter cell line responded linearly to IFN-b concentration and treatment time, we converted the assay to an automated format for a screen against already approved or approvable drugs. We also screened a library of phosphatase inhibitors that might mediate increased STAT1 phosphorylationactivation. Our analysis identified a series of diverse compounds capable of significantly increasing ISRE activity. One compound in particular, the anthracycline antibiotic idarubicin hydrochloride, was used to explore mechanism of action and to validate the proposal that small molecules can enhance ISRE activity to drive higher levels of ISG expression and improved control of viral level. The findings provide for the concept that current antiviral therapeutics act directly and specifically on viral proteins whereas next-generation antivirals might act to enhance host immunity against a broad range of viruses. Either alone or together, these approaches might better address the current need for more effective treatment against common as well as new and emergent viral infections. Based on the observation that STAT1-CC-expressing cells show increased activity of the endogenous ISRE promoter element [13] , we established cell lines that stably expressed an ISRE-containing gene promoter driving a click beetle luciferase reporter gene CBG99luc (Figure 2A ). We used 2fTGH cells (the parental line for STAT1-deficient U3A cells) as well as HEK293T cells that both express endogenous STAT1. Clonal lines showing IFN-b-inducible ISRE-promoter driven luciferase activity were designated 2fTGH-or HEK293T-ISRE-CBG99 cells. For initial assay development, we used the 2fTGH-and HEK293T-ISRE-CBG99 cells to establish an optimal luciferase light reaction time for both cell-lines. Based on luminescence signal stability over the time course of the reaction, an optimal readout window of 40-70 min after the start of the reaction was chosen for subsequent experiments ( Figure 2B ). After optimization of cell growth time, response to various IFN-b treatment times and concentrations were tested in 2fTGH and HEK293T cell lines. Each cell line exhibited a distinct IFN-b treatment time for maximal signal: 7-12 h for 2fTGH-ISRE-CBG99 cells and 14-24 h for HEK293T-ISRE-CBG99 cells. Although a lower signal magnitude was obtained with 2fTGH-ISRE-CBG99 cells compared to HEK293T-ISRE-CBG99 cells, the 2fTGH-ISRE-CBG99 cells show more specificity for IFN-b (compared to IFN-c) treatment at all IFN-b treatment time periods tested ( Figure 2E ) and over a range of IFN-b (and IFN-c) concentrations ( Figure 2F , G). Thus, the 2fTGH-ISRE-CBG99 cells were chosen for further assay development. To achieve assay automation and miniaturization, the ISRE activity assay was first automated in 96-well plates and then reformatted for 384-well plates. In the 384-well format, the assay exhibited a near-maximal signal at 8000 cells per well and consistent well-to-well and plate-to-plate reproducibility (Figure S1). In 96-and 384-well formats, signal to background (S/B) ratios, coefficients of variance, and Z9-factors achieved excellent performance in comparison to published standards [17, 18] . Representative results for 2fTGH-ISRE-CBG99 cells treated with IFN-b (1000 U/ml) for 7 h compared to 1% DMSO vehicle alone are provided in Table 1 . The results indicate the development of a quantitative and specific cell-based HTS assay of IFN-responsive gene promoter activity. We used the automated ISRE-activity assay to perform a screen of a 2240 chemical compound library. This library consisted of 2160 compounds from the Johns Hopkins Clinical Compound Library (JHCCL) of FDA approved or approvable drugs [19, 20] . In addition, we included 33 compounds from the Screen-Well Phosphatase Inhibitor Library based on the observation that the improvement in IFN signaling in STAT1-CC-expressing cells correlated with prolonged phosphorylation of STAT1 and STAT2 [13] . Each compound was tested at 4 different concentrations (0.24, 1.2, 6 and 30 mM) and simultaneous treatment with IFN-b at 5 U/ml, the concentration at the initial inflexion of the IFN concentration-response curve (as shown in Figure 2F ). These treatment conditions were duplicated on a second plate. Each assay plate also contained control wells containing a range of IFN- Figure 1 . Scheme for IFN signal transduction. Type I IFN signaling starts by activation of the IFN-a/b receptor (IFNAR) and subsequent activation of the IFNAR1-associated TYK2 and IFNAR2-associated JAK1, with consequent recruitment of STAT2. Phosphorylation of STAT2 enables reruitment of STAT1 and release of the phosphorylated STAT1-STAT2 heterodimer bound to IRF-9. This complex binds to the IFN stimulated response element (ISRE) and in concert with recruited transcriptional co-activators such as p300/CBP then drives IFN-stimulated gene (ISG) transcription. doi:10.1371/journal.pone.0036594.g001 b concentrations (0-200 U/ml) in quadruplicate ( Figure 3A ). This arrangement achieved excellent signal reproducibility between duplicate compound plates as well as signal consistency through the full screening run of 56 assay plates (28 duplicate pairs) ( Figure 3B , C). After raw data were normalized, scaled to z-scores, and summarized, we found that 321 data points (out of a total of 8960 data points representing the 2240 compounds tested at 4 concentrations) had an ISRE activity z-score $2 ( Figure 4A ). This data set represented 285 individual compounds, as some compounds had an ISRE activity z-score $2 at more than one dose. Of these 285 compounds, 64 hit compounds (2.9% of the total compound library) were selected for validation based on a combination of dose-response characteristics and inter-replicate reproducibility. This approach captured all 20 of the compounds with the highest z-scores. In support of re-purposing as a drug discovery strategy, the 64 screening hits were found in a broad range of drug classes ( Figure 4B ). Each of the 64 primary hits was subjected to primary validation for ISRE activity over a broader range of concentrations of drug and IFN-b (0-15 U/ml). Among the primary and confirmed screening hits, idarubicin hydrochloride ranked highest in potency for enhancing ISRE activity (i.e., idarubicin exhibited a significant effect at a lower concentration than other compounds). During the ISRE validation, we found that idarubicin caused a concentrationdependent increase in ISRE activity over a range of IFN-b treatment concentrations, with highly significant effects as low as 25 nM idarubicin in combination with 15 U/ml IFN-b ( Figure 5A) . The structure for idarubicin shows characteristic features of an anthracycline antibiotic unrelated to any other antiviral compound in clinical use. To further validate the effect of idarubicin on ISRE activity, we tested three other anthracyclines (daunorubicin, doxorubicin, epirubicin) with very similar chemical structures to idarubicin. Each of these compounds also showed a capacity to significantly increase ISRE activity ( Figure 5B ). In addition, we found that the immune activators DMXAA (Vadimezan) and Imiquimod did not cause any increase in ISRE activity in the same concentration range ( Figure 5C ). These compounds appear to directly activate immune cells (including increased IFN production) [21, 22] . However, for the present work, we specifically studied non-hematopoietic cells since that population appears critical for STAT1-mediated defense against at least some types of viruses [11] . We also found a cytotoxic effect of idarubicin that is consistent with previous observations [23, 24] . Under the present conditions, the major effect of idarubicin on cell viability was detected at drug concentrations .3 mM, so that the ISRE-activating effect of idarubicin occurred at concentrations below those that cause a major effect on cell viability ( Figure 5D ). Nonetheless, idarubicin is best known as an anti-neoplastic agent that acts via DNA intercalation and topoisomerase II inhibition [25] . To determine whether the effect of idarubicin on ISRE activity is related to topoisomerase II inhibition, we tested three other potent topoisomerase inhibitors Etoposide, Hu-0331, and ICRF-193 up to concentrations known to cause topoisomerase II inhibition [26] [27] [28] . In contrast to idarubicin, these other compounds caused no significant increase in ISRE activity ( Figure 5E ). In fact, two of the topoisomerase inhibitors (Etoposide and Hu-0331) caused a decrease in ISRE activity in concert with cytotoxic effects at higher concentrations. These findings indicate that the capacity of idarubicin to activate the ISRE component of the IFN signaling pathway occurs independently of topoisomerase inhibition. Together, the findings provide evidence of idarubicin capacity to increase ISRE activity independent of the anti-neoplastic properties of the drug. We also observed that the effect of idarubin and the other anthracyclines on ISRE activity occurred at a lower concentration of drug when IFN-b was co-administered, particularly at the highest concentration of IFN-b (15 U/ml) ( Figure 5A ). For example, the EC 50 for idarubicin decreased as the concentration of IFN-b increased ( Table 2 ). These findings suggested that idarubicin might somehow interact with the IFN signaling pathway. In that regard, we also found that the effect of idarubicin on ISRE activity was observed under baseline conditions when there was no detectable production of endogenous IFN-b and no administration of exogenous IFN-b (data not shown and Figure 5A ). These results suggested that the effect of idarubicin on ISRE activity is independent of IFN production or action. Indeed, we also found that the effect of idarubicin on ISRE activity persisted without change during effective IFN-a/b receptor 2 (IFNAR2) blockade ( Figure 6 ). Together, the findings indicate that idarubicin causes an increase in ISRE activity independent of IFN production or IFN-IFN-receptor interaction and instead acts downstream of ligandreceptor binding in the IFN signaling pathway. We subjected idarubicin to further validation as an ISRE activator in assays of ISG expression and antiviral activity. For ISG expression, U3A (STAT1-null) and U3A-STAT1 cells were treated with a range of concentrations of idarubicin and IFN-b and then harvested for gene expression using quantitative realtime PCR assay. We found that idarubicin increased the expression of the antiviral gene 29,59-oligoadenylate synthetase 1 (OAS1), particularly with IFN-b treatment ( Figure 7 ). There was no effect of drug on ISG expression in STAT1-null U3A cells, indicating that the drug is specific for STAT1-dependent gene expression. We found similar results for the antiviral ISG guanylate-binding protein 1 (GBP1) and three other ISGs (MX1, PARP9, and IRF1) ( Figure 6 and data not shown). For antiviral activity, 2fTGH cells were treated with idarubicin along with or without IFN-b and then assessed for control of encephalomyocarditis virus (EMCV) levels and virus-induced cytopathic effect. We selected EMCV since it was previously found to be sensitive to STAT1-CC-dependent improvement in IFN signaling [13] . In the present experiments, we found that idarubicin treatment (at a relatively low concentration of 25 nM) caused a significant decrease in EMCV titer at baseline and with IFN-b treatment (at a relatively low concentration of 5 U/ml) ( Figure 8A ). In addition, we observed that idarubicin-dependent improvement in viral control translates into a significant decrease in viral cytopathic effect under those same treatment conditions ( Figure 8B ). Higher concentrations of idarubicin in combination with IFN-b treatment caused a significant cytotoxic effect (data not shown), consistent with the antineoplastic properties of the drug. Nonetheless, the results with a relatively low concentration of idarubicin provide proof-of-concept that a small molecule activator of the ISRE component of the IFN signaling pathway will allow for increased ISG expression and improved control of viral level. The present study was undertaken to discover antiviral therapeutics that broadly increase host defense. We focused on the IFN system that is central to the antiviral response, although we recognized that other labs have pursued this target with limited success in the past. Some of these previous investigators have used administration of IFN itself to increase the antiviral response, but for this therapeutic goal and others, the side effects of IFN administration have proven to be rate-limiting [29] . Similarly, other investigators have attempted to boost IFN production, e.g., through administration Toll-like receptor (TLR) agonists CpG or Imiquimod, but these agents have also caused similar side effects [30] [31] [32] [33] . The small molecule DMXAA activates multiple immune pathways (NF-kB, TBK1/IRF3, NOD, and MAP kinase) but was ineffective as an antiviral unless it was administered before infection [34] [35] [36] [37] . To circumvent at least some of these issues with IFN production, toxicity, and specificity, we therefore pursued the goal of antiviral drug discovery with a novel screening approach for identifying small molecule enhancers that might selectively boost the activity or efficiency of the IFN signaling pathway. Our specific approach was based on previous success with the use of a modified STAT1 signaling pathway. In this work, we demonstrated that a designer form of STAT1 (designated STAT1-CC based on double-cysteine substitutions) was able to enhance IFN signaling and better control viral replication [13] . Although STAT1-CC gene expression would be challenging to translate to practical application, the mechanism of action served as a guide to design a screening strategy to identify small molecules that could mimic the antiviral benefit. In that regard, we also recognized that phenotypic screening approaches have proven to be more effective than target-based approaches for the discovery of first-in-class small molecule drugs [38] . Thus, target-based approaches (defined as direct drug action on a particular target) allow for analysis and refinement of structure and function but can also waste resources when the molecular hypotheses used to design screening assays may not be relevant to the disease. Meanwhile, phenotypic screening can take longer in terms of hit-to-lead development but provide more proteins in the pathway to be targeted and do not require prior knowledge of molecular mechanisms of action. Most importantly, the activity found in phenotypic-based approaches is often more effectively translated into therapeutic impact in disease models. For the present work, we take advantage of both of these strategies to some extent and devise a screen that incorporates molecular mechanism (i.e., enhancing a specific type of IFN signaling pathway) and the need to achieve phenotype (i.e., identifying any compound that could increase this type of signaling pathway regardless of specific mechanism). Considering these factors and observations in the STAT1-CC model system, we designed a cell-based luciferase reporter assay for measuring type I-dependent ISRE activity. This assay proved to yield excellent signal-to-background and Z factors, specificity for IFN-b treatment over IFN-c treatment, and suitability for automation and screening. Furthermore, because the construct design uses the Click Beetle Green luciferase, the assay can be paired with other luciferase reporter genes to develop dual color assays to report activity of other signaling pathways, including the type II IFN-c activated sequence (GAS) promoter activity that mediates defense against intracellular bacteria. A related approach was used to screen for small molecules that increase GAS activity for anti-proliferative and pro-apoptotic effects in cancer cells [39] . Others screened for compounds that inhibit Type I IFN production and signaling [40] . An ISRE-RFP reporter system has also been used to screen for assessing the effects of immunostimulatory RNA [41] . Others have used a less directed approach to screen for compounds that might use any mechanism to decrease viral levels [40, [42] [43] [44] [45] . However, to our knowledge, the present study conducts the first semi-quantitative screen measuring ISRE activity to discover small molecule enhancers of the type I IFN signaling pathway as broad-spectrum antiviral therapeutics. Our primary screen identified idarubicin on the basis of its capacity to significantly increase ISRE activity. Subsequent validation assays demonstrated that idarubicin facilitates STAT1-dependent ISG expression and STAT1-directed control of viral replication and cytopathic effect. While others previously reported the antiviral properties of anthracyclines some time ago, no mechanism of their antiviral action was elucidated [46] [47] [48] . In the present study, we observed drug-induced cytotoxicity in a dose-range similar to those reported previously [23, 24] , however, we establish that the effect of idarubicin on the antiviral IFN pathway is independent of cytotoxicity and topoisomerase inhibition. Because idarubicin enhances the IFN signaling pathway output, we questioned whether the drug might also cause IFN-driven cell death. However, we found no increase in cytotoxicity in cells treated with idarubicin and IFN together compared to cells treated with idarubicin alone. We also found that the idarubicin concentrations for activating the ISRE component of the IFN signaling pathway were significantly less than those required for major cytotoxicity. Thus, we conclude that idarubicin effect on IFN signaling is distinct from the effect on DNA-based cytotoxicity. The dose dependency of these effects underlines the need to conduct screening at multiple concentrations of test compounds, particularly lower concentrations that prevent false negative hits due to cytotoxic effects. In a further analysis of drug mechanism, our study demonstrates that the antiviral activity of idarubicin and other closely related anthracyclines is derived from enhancing the activity of the type I IFN signaling pathway. Our data further show that the enhancing effect of idarubicin is based on ISRE activation and ISG expression independent of IFN production or IFN-IFN-receptor interaction, since the effect of idarubicin is unchanged by IFNreceptor blockade. These findings suggest that idarubicin activation of the ISRE is due to an action in the IFN signaling pathway distal to ligand-receptor binding, e.g., at the level of receptorassociated JAK kinases or further downstream at the formation, transport, binding, and/or assembly of the ISRE transcriptional complex. In that regard, anthracyclines are known to inhibit DNA and RNA synthesis by intercalating between base pairs of the DNA/RNA strand to prevent replication, but whether this mechanism can affect ISRE or other gene promoter elements still needs to be defined. The present screening approach overcomes the uncertainty in molecular mechanism by using a phenotypic (rather than a target-based) screening approach and thereby captures compounds that increase the activity of the IFN signaling pathway by either established or undefined mechanisms. In sum, we describe and validate a phenotypic screening strategy to identify small molecules that enhance the activity of the type I IFN signaling pathway and consequently improve antiviral host defense. This approach is designed to lead to discovery of drugs with activity against a broad range of viruses for clinical application as well as experimental tool compounds to further understand IFN-dependent immune mechanisms. Current approaches to defining the basis for IFN signal transduction, particularly in vivo, often rely on complex transgenic and gene targeting approaches. Thus, the use of small molecule enhancers (SMEs) of the IFN signaling pathway may provide much greater flexibility and ease of application to achieve transient adjustment of IFN-related actions and consequent scientific and clinical benefit. Our approach should thereby prove useful to discover drugs with activity against a broad range of viruses as well as effectiveness in other conditions (e.g., multiple sclerosis and melanoma) where the efficacy of IFN treatment might benefit from enhancing the IFN signaling pathway. Stimulating agents and chemical compounds IFN-b and IFN-c were obtained from PBL Interferon source (Piscataway, NJ), diluted and aliquoted according of manufacturer's recommendation, and stored at 280uC. The Johns Hopkins Clinical Compound Library (JHCCL) was obtained from Dr. David Sullivan at the Johns Hopkins University [19, 20] . The Screen-Well Phosphatase Inhibitor Library was obtained from Enzo Life Sciences (Farmingdale, NY). All other chemical compounds were obtained from Sigma Aldrich (St. Louis, MO). To construct the pISRE-CBG99 vector, we first generated a 5xrepeat of the ISRE sequence and a TATAA box (5xISRE-TATAA) in the pUCMinusMCS vector from Blue Heron Biotechnology (Bothell, WA) and then cloned this sequence into the Chroma-Luc pCBG99-Basic reporter vector from Promega (Madison, WI) and ligated into Xma1 and Nco1 sites using T4 DNA ligase from Life Technologies (Carlsbad, CA). The DNA sequence of the resultant pISRE-CBG99 vector was confirmed by carrying out BigDye Terminator v3.1 sequencing reactions (Life Technologies) on an ABI capillary sequencer. This vector and the pPUR selection vector from Clontech (Mountain View, CA) were co-transfected at a 9:1 ratio into 2fTGH or HEK293T cells to increase the likelihood that cells tolerating puromycin selection (0.5 mg/ml) contained one or more copies of pISRE-CBG99 in addition to pPUR. The 2fTGH cells [49] were obtained from G. Stark (Cleveland Clinic), and HEK293T cells [50, 51] were obtained from T. Brett (Washington University). Transfection was performed using Fugene6 transfection reagent from Roche Applied Science (Indianapolis, IN) . Limiting dilution was used to obtain individual cell clones that were then screened for luciferasemediated luminescence after treatment with IFN-b (1000 U/ml) on a BioTek Synergy 4 multimode plate reader (BioTek, Winooski, VT). Clonal cells exhibiting stable expression were then used for further assay development. To optimize the luciferase light reaction, we tested a series of flash and glow luminescent substrate systems in both lysed and live 2fTGH-ISRE-CBG99 and HEK293T-ISRE-CBG99 cells, including D-luciferin (Fisher Scientific, Pittsburgh, PA), the Chroma-Glo Luciferase assay system from Promega and the steadylite plus reporter gene assay system from Perkin Elmer (Waltham, MA) under a range of incubation conditions. The steadylite plus system was selected based on kinetic profile. Effects of IFN-b concentra-tion and treatment time were assessed with all other variables constant. The assay was automated in a 96-well format with a customized and fully integrated robotic system. The system equipment included: a Caliper Sciclone ALH 3000 workstation (Perkin Elmer) and a EL406 washer (BioTek) for liquid handling, an automated Liconic incubator (Thermo Scientific) for cold storage of plates, an automated Cytomat incubator (Thermo Scientific) for cell culture environment, a separate hotel for storage of plates at room temperature, a Synergy 4 plate reader, a Flexiseal plate heat sealer (K Biosciences, Beverly, MA), a Caliper Twister II, and a Beckman Sagian Orca robotic arm on a linear rail (Beckman Coulter, Fullerton, CA). Construction allowed for transfer of plates, reagents, and plasticware between all instruments, so that there was no need for any manual interference during screening assays. This entire system was enclosed in a custom-made laminar flow hood to allow for HTS screening capability under BSL2 sterile conditions. After the system demonstrated satisfactory performance in a 96-well format, the assay was miniaturized to a 384-well format and re-tested for reproducibility and stability under IFN-b and vehicle (1% DMSO) treatment conditions. To achieve simultaneous treatment of cells with IFN-b and various compound concentrations and to avoid reagent degradation over time, the screen was run in a modular manner with a precise timeline ( Figure S2 ). The first step included production of plates with appropriate concentrations of compound and IFN-b and then storage at 4uC. A separate plate was made for each of the four compound concentrations (0.24, 1.2, 6 and 30 mM). The Twister II, Sciclone, Orca, and Liconic cold storage incubator handled this step. For the second step, cells were plated at 8000 cells per well in 384-well assay plates (n = 56). This step was accomplished in seven batches (8 assay plates per batch) using the Sciclone. A uniform suspension of cells was maintained by intermittent mixing on the Sciclone deck between cell plating. Simultaneously, the compound stock plates were sealed using the Flexiseal and stacked back into a Twister II rack for storage. For the third step, cells were allowed to grow for 11 h and then were treated with compound and IFN solutions. This step required that a plate containing cells be brought from the Cytomat incubator to the Sciclone deck in concert with a set of compound/IFN dilutions plates from the cold storage incubator. Cell treatments were timed so that each assay plate would be incubated for 10.3 h before the final step of performing the luciferase assay. For this last step, robotics were programmed so that each assay plate developed the luciferase light reaction for 40 min at 25uC in the plate hotel and then was delivered to the Synergy 4 plate reader for determination of luminescence. For this assay, the BioTek EL406 washer was used for aspiration of media and dispersion of substrate. In entirety, the screen took 41.6 h to complete. The raw data from the HTS assay were subjected to statistical analysis using cellHTS2 [52, 54] , a software package designed for the analysis of HTS data as part of the Bioconductor project for statistical computing [55] . Raw data were normalized using the plate median method [52, 56] . Next, a z-score transformation was applied to center and scale the data across the experiment. Replicates for a given compound at a given dose (N = 2 for each dose/compound combination) were then mean summarized. A zscore threshold of $2 was chosen to identify potential hits. Thereafter, to reach a smaller and tractable set of hits to validate experimentally, we took advantage of testing each compound at four concentrations. Specifically, we used self-organizing maps analysis to cluster hit compounds by shape of the dose-response curve. The significance of change from dose to dose (0.24 to 1.2, 1.2 to 6, and 6 to 30) was also analyzed using linear models and moderated F-statistics as implemented in the limma package [57] in Bioconductor [55] . The concentration-response curves for each compound were then visually inspected, using scatter plots generated in TIBCO Spotfire DecisionSite (TIBCO, Palo Alto, CA), with respect to the shape of the curve and reproducibility between replicates. Compounds showing an erratic concentrationresponse (e.g. increase, then decrease, and increase again in ISRE activity with increasing concentration) were rejected. Compounds with a consistent increase or decrease in response with increasing drug concentration or good efficacy at any concentration were included for further validation. This approach led to selection of 64 compounds for further validation, including compounds with the 20 highest z-scores. Hits from the primary screen were validated using the ISRE activity-luciferase reporter assay over a broad range of compound concentrations (0.01-25 mM) in the absence or presence of IFN-b (1, 5, and 15 U/ml). To determine drug potency, as defined by halfmaximal effective concentration (EC 50 ), this data was fit to a fourparameter concentration-response curve as described previously [58] using the log agonist concentration versus response, variable slope algorithm in GraphPad Prism 5 software (La Jolla, CA) where Y = Bottom + (Top-Bottom)/(1+10((LogEC50-X)*HillSlope)). To determine whether compound effect depended on IFN production, the ISRE activity-luciferase reporter assay was also performed in the presence of mouse anti-human IFN-a/b receptor chain 2 (IFNAR2) blocking mAb (clone MMHAE-2; Millipore, Billerica, MA) at a concentration of 4 mg/ml. A resazurin (Alamar Blue) metabolism assay was used to assess cell toxicity during compound treatment [59] . For these experiments, cells were treated with compound or an equivalent concentration of vehicle (DMSO) for 12 h, and the medium was replaced with fresh medium containing resazurin (80% dilution of Tox-8 kit, Sigma-Aldrich, St. Louis, MO). After 1.5 h at 37uC under standard culture conditions, the fluorescence of the resultant product resorufin was measured using the Synergy 4 plate reader. Wells with cells containing no compound (DMSO alone) and wells containing no cells were used as 100% and 0% viability controls. Data were normalized to calculate percentage viability. Expression of ISG's was assessed with real-time quantitative PCR assay for the corresponding mRNA level. For these experiments, U3A and U3A-STAT1 cells were first treated with the programmed combination of compound and IFN-b for 12 h and then were washed twice with cold Dulbecco's PBS followed by lysis with Cells-to-cDNA II lysis buffer (Life Technologies) and treatment with DNase according to the manufacturer's instructions. The U3A cells were obtained from G. Stark (Cleveland Clinic) and complemented with STAT1 to generate U3A-STAT1 cells as described previously [13] . A 25-ml aliquot of the cell lysate was used to generate cDNA using the High Capacity Reverse Transcription Kit (Life Technologies). The resulting cDNA was quantified using the Quant-iT OliGreen ssDNA kit (Life Technologies). Average cDNA concentration was 71625 ng/ml. Subsequent PCR assays were performed in adherence with MIQE guidelines [60, 61] , including the design of assays for ISGs (OAS1 and GBP1) and normalizer gene ornithine decarboxylase antizyme (OAZ1). The normalizer gene OAZ1 was selected and validated for cell samples treated with and without IFN-b. In brief, candidate normalizer genes were selected from a combination of invariant genes selected from previous microarray data [11, 62, 63] and prior large-scale analyses of publicly available microarray data [64, 65] . These candidates were then tested using real-time quantitative PCR assays. Comparison of candidate normalizer gene expression between various IFN treatment and infection conditions using multiple software packages [66 68] led to the selection of the OAZ1 as the normalizer gene. Primers and probes for real-time quantitative PCR assays were designed using the ProbeFinder design algorithm (Roche Applied Science). For OAS1, 59ggtggagttcgatgtgctg-39 and 59-aggtttatagccgccagtca-39 were used as forward and reverse primers, along with UPL probe #37 (Roche Applied Science). A plasmid containing OAS1 transcript variant 2 cDNA (Ref ID NM_002534, Origene, Rockville, MD) was used as a standard for absolute quantitation of OAS1 copy number. For GBP1, 59-ttccaaaactaaaactctttcagga-39 and 59tgctgatggcattgacgtag-39 were used as forward and reverse primers, along with UPL probe #85. A plasmid containing the GBP1 cDNA (Clone ID: 3606865, Thermo Open Biosystems, Huntsville, AL) was used as a standard. The IDT PrimeTime pre-designed assay Hs.PT.42.328511.g (Integrated DNA Technologies, Coralville, IA) was used for OAZ1. A cDNA vector was used for OAZ1 as well (Clone ID: LIFESEQ913650, Thermo Open Biosystems). Data were collected on a LightCycler 480 instrument (Roche Applied Science). Quantification cycle values were calculated using a second derivative maxima algorithm as implemented in the Lightcycler 480 software. Cells were cultured overnight and then treated with compound and IFN-b for 6 h. Thereafter, cells were washed and then were inoculated with EMCV (strain VR-129B, ATCC, Manassas, VA) for 1 h at MOI 1 as described previously [13] . Cells were then washed twice and cultured in medium containing 2% fetal bovine serum for 28 h. At that time, cell supernatants were used to determine viral titer based on real-time quantitative PCR assay for EMCV RNA with 59-ctgccttcggtgtcgc-39 (forward primer), 59tgggtcgaatcaaagttggag-39 (reverse primer), and 59caaggttttgagcgtgtctacgatgtgg-39 (probe). A TA plasmid containing the EMCV 3D protein was used as a standard for absolute quantitation of viral copy number. In addition, cell viability was determined using the Cellomics Arrayscan VTi high content imager (Thermo Scientific). For this assay, 15 images per well were obtained with a 10x objective. After background subtraction, cells were identified by nuclei stained by cell permeable dye Hoechst 33342. Propidium iodide fluorescence was quantified by defining a boundary of 2 pixels around the nuclei and then gating on a cell population that showed higher staining. For each sample replicate, cytotoxicity was calculated as the percentage of cells that showed increased propidium iodide staining based on samples of at least 5000 cells per well.

Feline Immunodeficiency Virus in South America

The rapid emergence of AIDS in humans during the period between 1980 and 2000 has led to extensive efforts to understand more fully similar etiologic agents of chronic and progressive acquired immunodeficiency disease in several mammalian species. Lentiviruses that have gene sequence homology with human immunodeficiency virus (HIV) have been found in different species (including sheep, goats, horses, cattle, cats, and several Old World monkey species). Lentiviruses, comprising a genus of the Retroviridae family, cause persistent infection that can lead to varying degrees of morbidity and mortality depending on the virus and the host species involved. Feline immunodeficiency virus (FIV) causes an immune system disease in domestic cats (Felis catus) involving depletion of the CD4+ population of T lymphocytes, increased susceptibility to opportunistic infections, and sometimes death. Viruses related to domestic cat FIV occur also in a variety of nondomestic felids. This is a brief overview of the current state of knowledge of this large and ancient group of viruses (FIVs) in South America.

Feline immunodeficiency virus (FIV) is a Lentivirus, closely related to HIV and SIV, which infects members of Felidae family. FIV is an important viral pathogen worldwide in the domestic cat (Felis catus), causes a slow progressive degeneration of immune functions that eventually leads to a disease. FIV is unique among the nonprimate lentiviruses because in its natural host species it induces a disease similar to AIDS in humans infected with human immunodeficiency virus type 1 (HIV-1), characterized by a progressive depletion of CD4 + T lymphocytes [5, 28, 35, 48, 77] . Species-specific strains, related to domestic cat FIV, have been isolated from a variety of nondomestic Felidae [11, 43] . Like HIV, FIV can be transmitted via mucosal exposure, blood transfer, and vertically either prenatally or postnatally [26] . For these reasons, FIV has been studied widely as both an important veterinary pathogen and an animal model for HIV/AIDS. Although FIV was first recognized in 1993 in Brazil [23] and in 1994 in Argentina [65] , there are few data describing the prevalence, ecology, clinical aspects, or genetic analyses of FIV in South America ( Figure 1 ). The prevalence of FIV within the continent is summarized in Table 1 . A better characterization of FIV strains circulating within South America will be required to augment our understanding of the importance of this lentivirus in felids. This paper provides an overview of the current state of knowledge of this large and ancient group of viruses (FIVs) in South America, grouped according to domestic and nondomestic felids. The data obtained allow a better understanding on FIV epidemiology and distribution. Efforts were made to gather and review all of the available information for each country. FIV infection, in domestic cats, causes a variable immunodeficiency syndrome characterized by recurrent gingivitis-stomatitis, cachexia, wasting, neurology, and an increased incidence of tumor development [1, 4, 48, 76] . In contrast, the ungulate lentiviruses cause diseases reminiscent of chronic inflammatory conditions while infection with the bovine lentivirus seems to be inapparent [71] . The rate of progression of the disease can depend on the genotype of the infecting FIV and is also likely influenced by undefined genetic determinants of the particular host [16] . FIV infection in domestic cats is associated with early robust humoral and cellular anti-viral immune responses, followed by a progressive immune suppression that results eventually in AIDS. The outcome of infection depends on the balance between the viral destruction of the immune system and the ability of the remaining immune system to eliminate the virus. Although the decrease in numbers of CD4+ cells is the hallmark of FIV infection, the virus has been shown to infect a variety of cell types in their respective hosts including CD4 + and CD8 + lymphocytes, B lymphocytes, cells of neuronal lineage and monocyte/macrophage lineage [15, 17, 29] . Joshi et al. (2005) have characterized feline CD4 + CD25 + T regulatory cells that support FIV replication. Recently, Reggeti, Ackerley and Bienzle (2008) have shown that feline dendritic cells express specific viral receptors and are infected productively by FIV [53] . FIV shares a similar pattern of receptor usage to HIV-1; however, CD 134 rather than CD4 is the primary binding partner, and subsequent interaction with the secondary receptor CXCR4 permits cells entry [58, 72, 73] . Differences in pathogenicity have been demonstrated among genetically distinct subtypes of FIV that circulate in domestic cats [14, 16, 49, 63, 68, 73] . On the basis of the analysis of envelope glycoprotein (Env), focusing on the third to fifth variable regions (V3-V5), FIV has been classified into five subtypes [30, 46, 60] a number that should be expected to increase as further studies reveal additional diversity. Recent studies identified distinct groups of FIV isolates from the United States and New Zealand [24, 69] (Figure 2 ). Data regarding FIV infection in domestic felids in South America are sparse and have not been well evaluated. Expanded surveys of South American isolates will be required to determine the FIV isolates in the continent since only few studies have been published. Although there are no doubts about the presence of FIV in South America, prevalence data obtained using different techniques cannot be compared amongst countries or studies (Table 1) . Knowing the prevalence and variability of FIV is important for designing and testing vaccines under field conditions [27, 77] . Also, identification of circulating subtypes is essential to develop strategies for molecular diagnosis, since the genetic diversity of this virus is high [44, 54] which may lead to false negative diagnoses if inappropriate primers are used. In South America, only subtype B and E viruses have been found. It is important to remember that subtype B viruses are distributed worldwide and that the subtype E viruses have been more consistently identified only in Argentina ( Figure 1 ). Preliminary studies suggested that FIV infection is widespread in the domestic cat population of Brazil [9, 12, 40, 52, 61, 64 ] . A published review indicated that subtype E was the only prevalent in Brazil [77] . Nevertheless, all studies indentified B as the only subtype circulating in FIV positive animals in Brazil, [12, 32, 40, 62] . Here an analysis was conducted with 473 bp of sequence encoding 157 amino acids comprising the V3-V4 region of the envelope glycoprotein from different subtypes, including those reported previously from South America (Figure 2 ). In this study we used this region of env in order to permit us to include more samples from South America. For this phylogenetic tree, the GenBank accession numbers, names, country and subtype for the FIV env sequences included were: M25381. It is important to state that all phylogenetic studies carried out in Brazil were performed in the same area, namely the south-east, and that Brazil is a huge country (Figure 1 ). More widespread surveys of Brazilian isolates are required to determine whether a single subtype of FIV predominates in Brazil. In Brazilian domestic cats, FIV infected cats have been observed over a prolonged period. During this time, few clinical signs were observed, although the virus was replicating and inducing changes in the immune system, leading to a progressive decline in immune function and the development later of overt clinical signs [51, 78, Hagiwara and Teixeira, unpublished data]. Previously, Brazilian studies established relationships between FIV infection and Toxoplasma gondii and Mycoplasma haemofelis [37, 39] . Otherwise, no association with disease has been recorded in cases of Brazilian FIV infection. It has been suggested that clade B viruses may be more ancient and relatively host adapted and thus may be less virulent [2, 50, 63] . Preliminary seroepidemiological studies carried out on clinical cases suggested that FIV infection is widespread in the domestic cat population of Argentina [65] . The genetic diversity of FIV isolates from Argentine domestic cats has been well characterized [47, 75] . FIV isolates were isolated from peripheral blood mononuclear cells of four domestic cats. Phylogenetic analysis revealed that one isolate clustered with subtype B and the others formed subtype E [47] , prototype sequence for this group (Figure 2) . In the north of the continent a single study was performed in 52 domestic cats on Isabela Island, Galapagos, Ecuador's coast. It was demonstrated using serological methods that none of the tested animals was infected with FIV [36] . Viruses related to domestic cat FIV occur also in nondomestic felids, indeed FIV strains have been present in the nondomestic cat population for longer than domestic cats [45] . Carpenter et al. (1996) comment that members of at least eighteen of the 37 species in the family Felidae carry an FIV-related virus, as has been shown by the presence in their sera of antibodies which react with FIV antigens. A further twelve species were reported in another study that employed a three-antigen Western blot screening (cat, puma and lion FIV antigens) and a multigene PCR amplification of FIV genes [66] . In South America, 12 [74] . Lentiviruses in eight of these species have been detected in South America [6, 10, 19, 20, 33, 55, 66] . Data regarding FIV infections in South American wild felids are sparse and studies have concentrated primarily on Brazil. The presence of antibodies against FIV in puma, detected by Western blotting, was found in Argentina (5 in 22, 23%), Bolivia (5 in 5, 100%), Brazil (2 in 13, 15%), Peru (1 in 5, 20%) and Venezuela (4 in 8, 50%) [10] . Further studies have reported antibodies recognizing FIV and the puma lentivirus (PLV in Brazilian free-ranging puma) [6, 19] . Troyer et al. (2005) concluded that most of the South American felids maintain a low level of FIV infection throughout their population. Within wild populations, the seroprevalence in South American felids varies from 5 to 28%. Unfortunately, the authors did not describe the regions of the continent where the samples originated. FIV pol genes from a Peruvian and a Brazilian zoo puma have been sequenced, the former being classified as subtype B and the latter as a distinct group, neither A nor B [10] . Additionally, FIV provirus has been reported in Brazilian jaguars (Panthera onca), pumas, jaguarondis (Puma yagouaroundi), oncelots (Leopardus pardalis), margays (Leopardus wiedii), pampas cat (Leopardus colocolo), geoffroy's cat (Leopardus geoffroyi) and little spotted cats (Leopardus tigrinus) [20, 33, 55] . The finding of these FIV infected species highlights the need for additional monitoring. Although the implications of these infections for wild felid conservation are difficult to assess, it is generally accepted that monitoring these infections is an important component for the management of endangered populations [13] . It is important to emphasize that FIV strains infecting 9 species of the Felidae have been at least partially sequenced and molecularly characterized [3, 10, 11, 25, 34, 38, 42, 43, 66] . Genetic analysis indicates that different felid species are infected by different strains of FIV [8, 11] . Analysis of pol gene sequence of FIV from lions (Panthera leo), pumas (Puma concolor) and domestic cats indicated that each species has a specific strain of FIV and that the strains are related but distinct [7, 43] . Also, strains from African lions (subtype B and E) differ in their abilities to replicate in feline cell lines [59] , their sensitivity to receptor antagonists [71] , and their requirement for ectopic expression of CD134, the primary cellular receptor, for productive infection [41] . It remains to be demonstrated that FIV-related viruses cause severe disease in species other than the domestic cat [6, 38] . The apparent absence of clinical signs in pumas and lions may reflect a longer period of coevolution between virus and host in these species, whereas in the domestic cat, the virus and host have not yet had time to reach a similar state of nonpathogenic coexistence [6, 7, 57] . However, it is by no means certain that FIV does not cause disease in non-domestic cats. Not long ago, reports have shown immune depletion associated with FIV infection in lions and pumas [56, 57] and another recent study reported evidence of immune suppression in the Pallas' cat (Otocolobus manul), including histopatological changes [8] . In addition, interspecies transmission (although is rare) may occur [22, 67] . For example, a leopard cat (Felis bengalensis) was found to be infected with a domestic cat virus [42] and FIV infecting one puma was more characteristic of domestic cat FIV rather than puma FIV [10] . The prevalence of FIV infection is South America has not been well evaluated and regional variations remain largely unexplored in domestic and wild cats. Considering that FIV has been detected in domestic cats in South America and that wild and domestic cats have overlapping territories in the communities and buffer zone, there is the potential for domestic felids to transmit this virus to naive wild felids in zoologic as well as free-range settings. The isolation and molecular characterization of these pathogens, both in domestic and a variety of wild felines, would be helpful and may provide important baseline data to develop effective programs aimed at infectious disease prevention. We believe that the feline population should be continually monitored for FIV infection and that clinical correlates to FIV infection should be further investigated. As recently proposed [70] , researchers could consider early surveillance programs across defined populations and detailed, cohort studies of naturally infected animals to provide further insights. Such studies would provide an opportunity to track retrospectively the pattern and consequences of an ongoing epizootic. There are technical reasons that hinder such studies, there is an urgent need for increased capacity in South American laboratories in order to conduct FIV screening and the apparent absence of FIV infection in some countries of the continent may merely reflect an absence of investigations. In addition, it is not easy to study FIV in wild cats as it is difficult to obtain samples from wild populations and only when these difficulties are overcome will it be possible to analyze and characterize FIV strains from the continent.

Chinese Medicine Shenfu Injection for Heart Failure: A Systematic Review and Meta-Analysis

Objective. Heart failure (HF) is a global public health problem. Early literature studies manifested that Shenfu injection (SFI) is one of the most commonly used traditional Chinese patent medicine for HF in China. This article intended to systematically evaluate the efficacy and safety of SFI for HF. Methods. An extensive search was performed within 6 English and Chinese electronic database up to November 2011. Ninety-nine randomized controlled trails (RCTs) were collected, irrespective of languages. Two authors extracted data and assessed the trial quality independently. RevMan 5.0.2 was used for data analysis. Results. Compared with routine treatment and/or device support, SFI combined with routine treatment and/or device support showed better effect on clinical effect rate, mortality, heart rate, NT-proBNP and 6-minute walk distance. Results in ultrasonic cardiography also showed that SFI combined with routine treatment improved heart function of HF patients. There were no significant difference in blood pressure between SFI and routine treatment groups. Adverse events were reported in thirteen trails with thirteen specific symptoms, while no serious adverse effect was reported. Conclusion. SFI appear to be effective for treating HF. However, further rigorously designed RCTs are warranted because of insufficient methodological rigor in the majority of included trials.

Heart failure (HF) is a leading cause of death, hospitalization, and rehospitalization worldwide. Despite advances in the treatment of HF, including use of drugs, devices, and heart transplantation, the condition remains associated with substantial morbidity and mortality [1] . International cooperation research program on cardiovascular disease in Asia showed that, on a total of 15,518 Chinese adults (35-74 years old) survey, the prevalence of HF was 0.9%, 0.7% for the males, and 1.0% for the females [2] . In the United States, HF incidence approaches 10 per 1,000 of the population over 65 years of age [3] . A report from the European Society of Cardiology (ESC) indicated at least 10 million patients with HF in these representing countries with a population of over 900 million. Half of the HF patients will die within 4 years, and more than half of those with severe HF will die within 1 year [4] . At present, the conventional therapeutic approaches in HF management include angiotensin-converting enzyme (ACE) inhibitors, β-blockers, and diuretics. Although several of them have led to an important effectiveness, HF remains the leading cardiovascular disease with an increasing hospitalization burden and an ongoing drain on health care expenditure [5] . Therefore, it remains necessary to search alternative and complementary treatment, in which Traditional Chinese Medicine takes a good proportion [6] . In TCM theory, pathogenesis of HF is related to deficiency of heart yang and heart qi and stasis of blood and excessive water (fluid), as well as interaction within these pathological factors. Under physiological conditions, yang can promote water metabolism, while qi can accelerate blood circulation, so yang and qi are the vital elements for human body to maintain life activity. TCM theory holds that patients suffered from HF are in deficiency of heart yang and qi for a long course, which directly leads to excessive fluid retention and blood stasis (Figure 1 ). Two Chinese herbal medicines, namely, Radix Ginseng (ginseng) and Radix Aconiti Lateralis Preparata (prepared aconite root), are used in treating HF over 2000 years. Ginseng invigorates qi, while prepared aconite root can warm and strengthen yang and lead to diuresis. Long-term clinical practice has proved that compatibility of ginseng and prepared aconite root can effectively ameliorate patients' symptom of HF and improve quality of life ( Figure 1 ). Shenfu injection (SFI) has been used in treating cardiac diseases for a long time in China [7] . The main active components of SFI are extraction of traditional Chinese herbs, namely, ginsenosides and higenamine. Modern pharmacological research shows that ginsenosides can improve ischemic myocardium metabolism, scavenge free radicals, protect myocardial ultrastructure, and reduce Ca 2+ overload, and higenamine can enhance heart contractility, improve coronary circulation, and decrease the effect of acute myocardial ischemia [8] . Currently, SFI used alone or integrated with routine treatments has been widely accepted as an effective method for the treatment of HF in China. Many clinical studies reported the effectiveness ranging from case reports and case series to controlled observational studies and randomized clinical trials, but the evidence for its effect is not clear. This paper aims to evaluate the beneficial and harmful effects of SFI for treatment of HF in randomized controlled trials. (Shenfu injection or Shenfu or Shen-fu) AND (heart failure or cardiac dysfunction or cardiac inadequacy or cardiac failure or congestive heart failure). All of those searches ended before November 2011. And the bibliographies of included trials were searched for thorough references, irrespective of languages. All the randomized controlled trails (RCTs) of SFI compared with routine or conventional treatment (control group) in adult patients with HF were included. RCTs combined SFI with conventional treatment and/or invasive respiratory support (SFI group) compared with conventional treatments and/or invasive respiratory support (control group) were included. Both acute heart failure and chronic heart failure were included. Outcome measures include clinical effect rate, death and adverse events, ultrasonic cardiography, heart rate and blood pressure, and quality of life. Wen-Ting and C. Fa-Feng) extracted the data from the included trials independently, based on the inclusion criteria outlined above. Nonrandomized evaluations, pharmacokinetic studies, animal/laboratory studies, and general reviews were excluded, and duplicated publications reporting the same groups of patients were also excluded ( Figure 2 ). Extracted data was entered into an electronic database by two authors, S. Wen-Ting and C. Fa-Feng independently. The methodological quality of RCTs was assessed by using criteria from the Cochrane Handbook for Systematic Reviews of Interventions, Version 5.0.1. The quality of trials was categorized into low risk of bias, unclear risk of bias, or high risk of bias according to the risk for each important outcome within included trials, including adequacy of generation of the allocation sequence, allocation concealment, blinding, whether there were incomplete outcome data or selective outcome, or other sources of bias. The statistical package (RevMan 5.0.2), which is provided by The Cochrane Collaboration, was used to analyze collected data. Dichotomous data was presented as risk ratio (RR), with 95% confidence intervals (CIs). Continuous outcomes were presented as mean difference (MD), with 95% CI. Analyses were performed by intention-to-treat where possible. Heterogeneity between trials results was tested, and heterogeneity was presented as significant when I 2 is over 50% or P < 0.1. Random effect model was used for the meta-analysis if there was significant heterogeneity, and fixed effect model was used when the heterogeneity was not significant [21] . Publication bias was explored via a funnelplot analysis. According to the search strategy, we screened out 903 potentially relevant studies for further identification ( Figure 2 ). By reading titles and abstracts, we excluded 701 studies that were obviously ineligible, including review articles, case reports, animal/experimental studies, and nonrandomized trials. 202 studies with full text papers were retrieved. After the full text reading, 6 studies were excluded because of duplicated publication. 84 studies were excluded due to lack of clinical effect rate which is the primary outcome evaluated in present study. 4 studies were excluded because the reported groups of participants were same as previous trials. In 108 RCTs, 11 studies were excluded due to other herbal intervention which was combined with SFI as treatment arm. Thus, 97 RCTs [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] were included for systematic review. Ninety seven RCTs involved a total of 8,202 patients with HF, including 92 trails (7854 patients) of chronic HF and 5 trials (348 patients) of acute HF. The sample size varied from 24 to 248 participants, with an average of 42 patients per group. Since RCTs of HF on children were excluded, patients are adults (ranged from 28 to 89 years old). More males were included than females (52% males and 48% females). Disease duration was reported in 31 trials, ranging from 3 months to 26 years. 49 trials were observed in inpatients, 5 outpatients [22] [23] [24] [25] [26] , 5 both inpatients and outpatients [27] [28] [29] [30] [31] , and 39 unclear. All studies were published in Chinese. Mortality was reported in eleven studies, while the rest of the eighty eight trials did not mention death. Effect rate was assessed in all the trials, based on the improvement of heart function. Ninety one trials used New York Heart Association (NYHA) Classification of Clinical Status, and six trials used Killip's Rating Standards [22, 25, 26, [33] [34] [35] for diagnosing HF and rating the patients. Patients in fifty one trails ranged from II to IV, seven trials II to III, twenty one trials III to IV, and five trials IV according to NYHA Classification; patients in five trials ranged from II to IV and one trial IV according to Killip's Standard. Results of ultrasonic cardiography were reported in 61 trails (5135 patients) with left ventricular ejection fraction (LVEF) as main parameter. Other parameters such as left ventricular diastolic diameter (LVDd), cardiac output (CO), cardiac index (CI), stroke volume (SV), and A peak E-wave velocity ratio (E/A) were reported in 16, 17, 20, 18 , and 11 trials, respectively. N-terminal pro-B-type nature tripeptide (NT-proBNP) level in blood was reported in 12 studies of 887 patients, and 6-minute walk distance (6-MWD) was reported 4 Evidence-Based Complementary and Alternative Medicine in 8 trials of 630 patients. Heart rate, systolic blood pressure (SBP), and diastolic blood pressure (DBP) were reported in 27, 15, and 13 trials, respectively (Table 1) . According to our predefined quality assessment criteria, all of 97 included trials were evaluated as having unclear risk of bias (Table 2, Figure 3 ). None of the 97 trials reported sample size calculation. Eleven trials described randomization procedures, nine trials [9-11, 20, 30, 38-41] used a random number table, one drew lots [19] , and one trial separated patients by odd and even number of patient ID as a quasirandomization [42] . Only one trial [43] blinded both patients and outcome assessors, and three trials [44] [45] [46] blinded patients. None of the trials reported adequate allocation concealment. Five out of ninety seven trials mentioned that followup ranged from 3 months to 12 months after treatment. One trial [47] followed all the patients for 12 months, one trail [38] for 6 month, and the rest [9, 11, 12] for 3 months. However, neither of them used intention to treat method. The primary outcomes were effect rate and mortality. Secondary outcome measures included LVEF, LVDd, SV, CO, CI, HR, systolic blood pressure (SBP), diastolic blood pressure (DBP), NT-proBNP, and 6-MWD. Effect Rate. All the trials reported clinical effect rate to evaluate the outcome, which was based on NYHA Classification of Clinical Status and Killip's Rating Standards. Killip's Rating Standards were used by six trials with patients of myocardial infarction-induced HF, while other trials used NYHA Classification. Most of trails used three categories to evaluate treatment effect including markedly effective (an improvement of two classes on the classification), effective (an improvement of one class), and ineffective (no improvement, deterioration or death), and others only reported total effect. Total effect rate is the combination of markedly effect rate and effect rate. Trials of myocardial infarction-induced HF and nonmyocardial infarction-induced HF were separated into two subgroups. The meta-analysis showed a total significant difference between SFI and control groups on total effect rate (RR: (Figure 4) . Death. Eleven studies reported mortality data, and total death number was 142 out of 978. Two trials [12, 38] assessed the mortality with 3-and 6-month followup, respectively, and other trials reported death at the end of treatment course. Trials were also separated into two subgroups depending on whether HF was induced by myocardial infarction. The result of meta-analysis indicated that SFI can significantly reduce mortality of patients of myocardial Figure 5 ). NT-proBNP. NT-proBNP level is used for screening and diagnosis of acute HF and may be useful to establish prognosis in HF, as it is typically higher in patients with worse outcome [109] . It was reported in 12 studies [20, 22, 38, 45, 49, 52, [54] [55] [56] [57] [58] [59] Heart Rate and Blood Pressure. Heart rate and blood pressure were reported in 27 and 15 trials, respectively. Metaanalysis showed that there was statistical significance between SFI group and control group (WMD: 6.31; 95% CI [5.18, 7.44] , P < 0.01) (see Supplementary Figure 1 Results of Ultrasonic Cardiography. LVEF is the ratio of the stroke volume and the left ventricular end-diastolic volume [107] . It is usually used for the assessment of HF and drug efficacy. Sixty-one studies reported the outcomes for LVEF. Meta-analysis showed that SFI group was better than control group in increasing LVEF (WMD: 6.31; 95% CI [5.18, 7.44] , Figure 4) . SV is the volume per stroke by left ventricle, and CO is the volume of blood being pumped by the heart in the time interval of one minute [107] . CI is a vasodynamic parameter that is relating CO to body surface area [107] . All the three parameters indicate left ventricular systolic function, as LVEF does. This paper made meta-analysis of these outcomes, respectively; results showed that SFI group was better than control group in these three parameters: SV (WMD: 7.25; 95% CI [4.60, 9. E/A ratio is widely accepted as a clinical marker of diastolic HF, and E/A ratio is reduced in diastolic dysfunction [108] . The result of meta-analysis of E/A ratio was WMD: 0.15; 95% CI [0.08, 0.22], P < 0.01, which indicated that SFI better improved diastolic function of heart on HF patients Evidence-Based Complementary and Alternative Medicine 5 than conventional medicine treatment did (Supplementary Figure 8) . LVDd is the end-diastolic dimension of the left ventricle. There was no statistical significance between SFI combined with conventional medicine treatment and conventional medicine treatment groups (WMD: −1.59; 95% CI [−5.29, 2.12], P = 0.40) (Supplementary Figure 9) . None of the trials reported quality of life. Funnel plots based on the data of effect rate were elaborated in Figure 8 . The figure was asymmetrical, which indicated that potential publication bias might influence the results of this paper. Although we conducted comprehensive searches and tried to avoid bias, since all trials were published in Chinese, we could not exclude potential publication bias. 3.6. Adverse Effect. Thirty seven out of ninety seven trials mentioned the adverse effect except in sixty-two trials which was unclear. Thirteen trials [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] 60] reported the following thirteen specific symptoms of side effects including dry mouth, dryness heat, fullness of the head, insomnia, dysphoria, skin itching, tachycardia, feverish dysphoria, flushing of face, tidal fever, dizziness due to low blood pressure, gastrointestinal discomfort, and palpitation. Among these side effects, dry mouth and fullness of the head were reported in 4 trails with 14 and 10 cases, respectively. These symptoms were regarded to be mild and recovered spontaneously after SFI withdrawal. Twenty four trials reported that no side effects were observed in the SFI group (Table 3) . The above side effects might be related to higenamine, which is the active ingredient of prepared aconite root. In TCM books and papers, prepared aconite root is frequently mentioned with adverse effects as dry mouth, dryness heat, fullness of the head, and dysphoria due to its strong effect of strengthening yang. In many years, western medicine has made tremendous progress and has become the dominating medical treatment worldwide. However, it has been increasingly recognized that Events 27 26 31 37 32 28 79 49 122 37 58 36 25 30 51 26 33 71 46 33 55 17 42 36 36 40 57 58 68 28 47 56 33 54 32 46 63 23 24 68 45 36 36 40 31 44 29 40 54 47 28 77 23 56 33 39 18 36 34 13 52 49 45 18 26 22 54 37 34 41 28 28 37 27 50 37 37 41 36 54 29 11 45 28 70 53 17 52 28 55 18 3678 22 24 21 22 40 24 153 3831 Total 30 30 35 40 35 30 90 56 127 41 62 40 30 40 55 31 36 76 48 35 60 18 48 38 38 46 62 64 80 30 50 60 42 58 35 48 66 25 28 74 50 42 40 43 32 50 31 42 60 50 32 85 24 60 34 45 22 38 37 16 62 58 50 20 31 24 58 40 40 44 33 30 40 30 56 39 40 46 40 60 30 12 50 30 78 56 19 56 30 55 20 4047 35 36 23 28 54 36 212 4259 Events 20 22 22 32 26 22 61 45 97 14 48 30 16 12 15 23 19 42 25 22 44 15 26 28 27 28 56 47 40 23 33 54 21 41 24 41 52 17 14 63 39 33 33 32 29 38 21 24 40 46 19 67 21 50 28 33 4 22 29 10 31 25 22 15 24 20 43 30 24 27 18 22 32 21 21 27 28 28 30 38 22 8 34 22 59 40 13 46 22 40 13 2770 17 16 14 12 29 16 104 2874 Total 39 28 33 40 35 27 90 56 121 22 62 40 30 20 30 28 26 76 30 30 60 18 39 38 38 40 62 64 60 30 40 60 41 54 32 48 64 25 20 78 50 40 40 42 32 50 30 42 52 50 30 85 24 60 34 42 10 30 36 16 61 38 26 20 31 24 58 40 34 44 29 30 40 30 30 37 40 32 40 50 26 12 50 30 80 56 18 56 30 50 20 3731 35 38 22 23 western medicine may sometimes fail to treat an illness, whereas such illness is reportedly improved by the so-called complementary medicine based on a different theory [110, 111] . Although conventional therapeutic approaches were used in HF, it remained a cardiovascular disease with an increasing hospitalization burden and an ongoing drain on health care expenditures [2] . TCM plays an important role in treating HF in China. SFI was a traditional Chinese Patent Medicine based on TCM theory, which was approved by the Chinese State Food and Drug Administration. In recent 10 years, it has been widely used for HF in many hospitals and clinics. However, few RCTs of SFI were reported in English journals, and it was difficult for western doctors to accept SFI as an alternative medicine. Although there were two systematic reviews about SFI for HR published in Chinese journal [112, 113] , only 16 and 8 trials were included in their study. Therefore, the present study aimed to systematically assess the efficacy and safety of SFI for HR. Data from the 97 RCTs demonstrated that SFI combined with conventional medication may be more effective on HF than conventional medication only. With improvement of cardiofunction of patients, based on NYHA Classification of Total (95% CI) Test for overall effect: Z = 7.14 (P < 0.00001) Mean 443 250 497 329 218 216 445 330 SD 66 112 74 64 17 18 65 18 Total 40 35 36 38 50 42 40 46 327 Mean 395 200 413 280 211 203 395 316 SD 64 90 67 52 15 16 63 17 Total 40 35 26 38 50 42 40 Clinical Status and Killip's Rating Standards, the effect rate of SFI group was, on average, 17 percent more than control group (RR, 1.19; 95% CI, 1.17 to 1.21). Mortality data was another primary outcome. In eleven trials in which death was recorded, meta-analysis showed that mortality was significantly lower in SFI group than control group. This result was mainly contributed by subgroup of HF induced by myocardial infarction, for patients in this subgroup were more vulnerable. Ultrasonic cardiography is widely used in inspection for HF patients. From results of ultrasonic cardiography, the systolic and diastolic functions of heart can be interpreted. LVEF, CO, CI, SV, LVDd, and E/A were reviewed by us, respectively. There was significant difference between SFI group and control group in all of the outcomes except LVDd. Since SV, CO, CI, and LVEF indicate heart systolic function, and E/A indicate heart diastolic function, conclusion can be drawn that SFI benefits both systolic and diastolic functions of heart. But it did not have significant effect on expansion of heart. NT-proBNP level in serum of SFI group was significantly lower than the control group, which is inconsistent with effect rate. 6-MWD results of patients of SFI group also are better than thos of control group. It indicates that SFI had a tendency to improve life status. Furthermore, heart rate was obviously reduced in SFI group, which could be related to alleviation of HF. Meta-analysis on LVEF, CO, CI, SV, LVDd, E/A, heart rate, and NT-proBNP all showed significant heterogeneity. Several possible explanations can be given, for example, different complications, different instruments employed for test, and difference in methodological rigor. However, we should consider the following limitations before accepting the findings of this paper. Firstly, the methodological quality of the included studies is generally poor. Although all trials claimed to perform randomization, only eleven trials reported the procedure to generate the sequence, while the rest of trials did not give any details of the randomization method. Thus, whether randomization was effectively conducted in these trials was doubtful. Blinding was mentioned in four trials, with one trial blinded patients and outcome assessors [43] and three blinded patients only [44] [45] [46] . Neither of them described the methods of allocation concealment. Dropouts account and intention to treat analysis were not mentioned in all the trails. Due to inadequate reporting of methodological design, it was possible that there was performance bias and detection bias due to patients and researchers being aware of the therapeutic interventions for the subjective outcome measures. Therefore, we cannot draw a confident conclusion that there were significant beneficial effects of SFI combined with conventional medicine treatment compared with conventional medicine treatment. Secondly, limited outcomes were reported, especially death and adverse events. Since HF is a disease with high mortality, death is the most important primary outcome. However, only eleven studies out of ninety seven trials reported death, and most of the eleven trials assessed mortality at the end of treatment, without followup. Another outcome was adverse events, to which more attention should be attached. Only 37.4% of the trials described the occurrence of adverse events, indicating an incomplete evaluation of the safety profile of SFI, as well as poor quality of reporting. In most trials, the duration of therapy and followup was 22 Evidence-Based Complementary and Alternative Medicine too short to achieve conclusive results, except that only one trial had a treatment of 10 months [47] . Only 6 included trials had a followup period (ranged from 3 to 12 months), while in rest of studies, the outcomes were evaluated at the end of the treatment (mostly range from 14 to 21 days). In order to evaluate drug efficacy for chronic HF, long-term improvement (at least 6 months) of chronic HF-specific clinical symptoms is needed [114] , because some drugs have shown to increase mortality in the long-term application despite a short-term improvement in clinical symptoms [115] . In addition, long-term toxicity assessment was also important for drug safety evaluation. Next, although irrespective of languages, all the trials included in this paper were published in Chinese journals, Zhang et al. and Liu et al. [115, 116] found that some Asian countries including China unusually publish high proportions of positive results. Wu et al. [117] and Jin et al. [118] accounted that RCTs in Chinese journals often had problems of low methodological quality and selective publication of positive results. Considering that all of the ninety seven trials were published in Chinese, the publication bias possibly existed. Additionally, none of the ninety seven trials reported sample size calculation, and in most trials, the sample size was limited. Further high-quality studies with larger sample size are needed to confirm the effectiveness of SFI in treating HF. Quality of life was not reported in all the including trials. Although 6-MWD showed a tendency of SFI to improve life status for HF patients, we advise future RCTs to select outcomes of life quality according to international practice. Considering that there was no sufficient amount of highquality trials on SFI treating patients with HF, the effectiveness and safety of SFI need further rigorous trials to prove, which should be consistent with the CONSORT statement on the reporting of the results of randomized trials (http:// www.consort-statement.org/). The preliminary conclusion of the current study suggests that SFI might be beneficial to patients with HF. More rigorously designed trails with high methodological quality are necessary for further proof.

Genome Stability of Pandemic Influenza A (H1N1) 2009 Based on Analysis of Hemagglutinin and Neuraminidase Genes

Influenza A virus (H1N1), which arose in 2009, constituted the fourth pandemic after the cases of 1918, 1957, and 1968. This new variant was formed by a triple reassortment, with genomic segments from swine, avian, and human influenza origins. The objective of this study was to analyze sequences of hemagglutinin (n=2038) and neuraminidase (n=1273) genes, in order to assess the extent of diversity among circulating 2009-2010 strains, estimate if these genes evolved through positive, negative, or neutral selection models of evolution during the pandemic phase, and analyze the worldwide percentage of detection of important amino acid mutations that could enhance the viral performance, such as transmissibility or resistance to drugs. A continuous surveillance by public health authorities will be critical to monitor the appearance of new influenza variants, especially in animal reservoirs such as swine and birds, in order to prevent the potential animal-human transmission of viruses with pandemic potential.

Influenza A viruses belong to the Orthomyxoviridae family, and have a genome composed of eight segments of single-stranded, negative-sense RNA. Their surfaces are composed by a lipid envelope, originated from the plasmatic membrane of infected epithelial cells, and two antigenic proteins: Hemagglutinin (HA) and Neuraminidase (NA); these two antigens exhibit higher variability compared with their remaining proteins [1] . Depending on the extent of variability of two surface proteins, until now are known 16 HA (H1-H16), and 9 NA genotypes (N1-N9), respectively, which can be combined in different combinations [1, 2] . In early April 2009, authorities from the Mexican public health observed a high number of influenza-like illnesses in their territory, and informed about this outbreak to the regional office of the World Health Organization (WHO). In mid April, the Centers for Disease Control from USA identified the new virus in two cases from California. The new virus spread rapidly throughout the world, and as a consequence the WHO authorities declared the "Pandemic (H1N1) 2009" on June 11, 2009 [3] . It is thought that the new 2009 H1N1 pandemic virus (from here, 2009 H1N1pdm) has emerged through at least four reassortment and transmission events among swine, avian and human H1N1 lineages, probably in Asia and North America [4] . Particularly, the HA segment of 2009 H1N1pdm was originated from American swine lineage, whereas the NA segment derived from the European swine lineage [5, 6] . It is believed that the ancestors of this pandemic strain remained undetected for approximately one decade due to lack of a *Address correspondence to this author at the Rio de la Plata y Lagerenza, CP 1120 Asunción, Paraguay; Tel: +595 21 424 520; Fax: +595 21 480 185; E-mail: emilioespinola@hotmail.com surveillance system in pigs, the historical "mixing vessel" for new influenza viruses. Furthermore, the closest ancestors of the new pandemic strains emerged probably in January 2009 [4] . The objective of this study was to analyze a dataset of complete nucleotide (nt) sequences of HA and NA genes, in order to assess the extent of diversity among circulating 2009-2010 strains, estimate if these genes evolved through positive, negative, or neutral selection models of evolution during the pandemic phase, and analyze the worldwide percentage of detection of important amino acid mutations that could enhance the viral performance, such as transmissibility or resistance to drugs. Complete CoDing Sequences (CDS) of HA (1701 nt) and NA (1410 nt) genes corresponding to 2009 H1N1pdm, isolated from humans, were downloaded from the Influenza Virus Resource (http://www.ncbi.nlm.nih.gov/genomes/FLU /SwineFlu.html) from the National Center for Biotechnology Information, by the year of sequence repository. The first dataset consisted of 3765 HA and 2996 NA sequences, respectively, which were reported in the period 2009-2010. After discarding exact duplicates in sequence using a Perl script, we obtained 2038 HA and 1273 NA sequences, respectively; these sequences were different in at least one nucleotide among all representatives. Reassortant strains were discarded, as well as incomplete CDS sequences. Nucleotide sequences were manually edited in FASTA format, using BioEdit v7.0.5 [7] , and aligned with CLUSTAL W [8] . Sequence information (GenBank accession number, strain, and year of isolation) for each sample used in this study are available for HA (Table S1) and NA genes (Table S2) , respectively. Pairwise distances were calculated with MEGA v5 [9] . The percentages of identities were calculated by applying the formula 100 -(pairwise distance value x 100). A graph was constructed by plotting the percentage identities in the abscissa (x axis) vs the frequency of each of the calculated pairwise identities in the ordinate (y axis). The graphs were prepared in the R environment, using ggplot2 package (www.r-project.org). The models of nucleotide substitution that best fitted each dataset were determined with MEGA v5 [9] , and were: GTR+I model for HA genes, and T92+G model for NA genes, respectively. Phylogenetic relationships were reconstructed by the Neighbor-Joining method [10] , with the appropriate models of nucleotide substitution for each dataset (as described above) and bootstrap analysis of 1000 replicates, as incorporated in MEGA v5 [9] . Outgroup sequences for HA and NA genes corresponded to strain A/Puerto Rico/8/1934. Mutations in each CDS were analyzed by the method of Nei and Gojobori [11] . Codon aligned sequences for each dataset were analyzed using the Perl-based SNAP program (http://www.hiv.lanl.gov/content/sequence/SNAP/SNAP.htm l) [12] in order to calculate the variability of each CDS. The selective pressure was measured by comparing the rate of non-synonymous nucleotide substitutions per nonsynonymous site (d N ) against that of synonymous substitutions per synonymous site (d S ). The ratio d N /d S was used as an index to assess positive selection. A ratio d N /d S >1 means positive (diversifying) selection, =1 means neutral selection, and <1 means negative (purifying) selection. The analysis of pairwise identity frequencies showed high percentage of similarities among circulating 2009-2010 pandemic influenza strains (Fig. 1) . The average percentage of identity was 99.7% for both HA and NA genes. Thus, in this period of pandemic circulation, both genes did not segregate into different clusters, but on the contrary showed a constant and stable evolution. The high percentages of nucleotide identity were in accordance with the single clustering of all 2009-2010 strains in the phylogenetic tree of HA and NA genes (Fig. 2) , without temporal or geographical distribution. It is interesting to note that the overall genetic diversity among 2009 H1N1pdm was less than typically observed among seasonal influenza. This is in accordance with its short period of time of circulation in humans [13] . The single clustering of 2009 H1N1pdm observed in this report, however, is in contrast with other studies [14] , in which the authors observe differences by using small datasets of sequences. The single clustering of 2009 H1N1pdm, furthermore, agrees with serological data in which it was observed that antigenically, the new pandemic viruses were all similar [6] , and thus not requiring a new update of the vaccine (strain A/California/07/2009) until now. Given that 2009 H1N1pdm constituted a homogeneous phylogenetic group, it was hypothesized that the diversity in nucleotide sequences localized (in average) in the 0.3% of differences within each analyzed gene. Taking into account the complete CDS for HA and NA genes, this percentage of differences constitutes approximately four to five nucleotide random variations among circulating strains. Calculation of average d N /d S rates of evolution showed that both HA and NA genes evolved through negative (purifying) selection (Table 1) , with d N /d S values of 0.2762 and 0.1939, respectively. Even though in general, both genes underwent negative selection, some positions can evolve through positive selection. For example, an early study showed that two sites involved in receptor binding specificity of HA (220 and 278) were under positive selection, and these sites were not found in swine or seasonal H1N1 viruses [15] . Thus, changes in receptor binding sites could lead to alterations in receptor binding specificities. In other viruses such as SARS-CoV, it was observed that they can develop through positive selection through the Fig. (1) . Pairwise identity frequencies for (a) Hemagglutinin and (b) Neuraminidase genes, respectively. cross-species transmission in early epidemics, and negative selection during late epidemics [16] . It is possible that the same mechanism was the driven force of evolution of 2009 H1N1pdm, with positive selection at least during crossspecies transmission. A number of different amino acid mutations that could confer new functionalities to the new virus were reported worldwide, including those related to increased pathogenicity or antiviral resistance ( Table 2) . Polymorphism at position 239 in HA has been associated with severe clinical outcomes, especially in immunocompromised patients; in particular, substitution 239G was found to correlate with fatal outcomes in different countries [17, 18] . Furthermore, this mutation can arise de novo from wildtype (D239) virus in the same patient throughout the disease course [19] . Mutation at position 239 can induce alterations in the receptor binding site, and 239G mutants bind a broader range of 2-3-linked sialyl receptors sequences expressed on cells from the lower respiratory tract, which suggested that its presence could be responsible for the exacerbation of disease [20] . Mutants 239E target mainly non-ciliated cells. We found no significant difference between sequences bearing mutations 239G (2.6%) and 239E (5.5%). The low percentage of global circulation of mutants 239G found in this study is in accordance with its lower potential to transmit to other individuals [21] . Positions 239 and 220 are localized within the HA antigenic site called Ca. The amino acid S220, though not exposed to the surface, is localized in the receptor binding domain (RBD), and its change could affect the transmissibility and infectivity of H1N1 in humans. The fixed mutation, S220T, has been found at high percentage (76.7%) in this study. To test whether change 220T could contribute to antigenic drift, it would be interesting to compare its antigenic profile against a wildtype isolate (S220). This mutation, probably, has become fixed in all pandemic strains through optimization of viral fitness, rather than immune selection or adaptation to the host. Substitution S101N has been proposed previously as a reversion to the seasonal H1N1 residue 101N and thus possibly an adaptation to the human host, being found in some studies at high frequencies. Its global impact, however, is controvertible because it was found in only 0.2% of our sequences. Substitution E391K, found at 15.6% in our study, has been identified as part of a highly conserved epitope in the 1918 H1N1 virus with a possible role in membrane fusion [22] . Another proximal substitution found in other studies, N387H, was found in only 1.7% of our sequences. In the NA gene, it was showed that mutations V106I and N248D were present in samples at increasing numbers through early pandemic month (April to December 2009) [23] . We found both mutations at high percentages, 85.1% and 85.9% respectively, in our dataset. Change 106I was present in the 20th century cases of H1N1 (in 1918 [pandemic] , and 1977), as well as 248D (in 1977). Since residue at position 248 is located at the drug target domain (DTD) region, as residue 275, it could potentially affect the sensitivity to NA inhibitors. Another substitution of possible interest in NA sequences is D199N, which was previously associated with an increase in oseltamivir resistance in both seasonal and H5N1 virus strains [24] . We found, however, only 4 out of 1273 NA sequences (0.3%) containing this change. The rare substitution I223R, which was reported in association with resistance to oseltamivir, zanamivir, and peramivir [25] , was also found in only 2 out of 1273 NA sequences (0.2%). Substitution H275Y has been related to oseltamivir resistance, especially in immunocompromised or severely ill persons [26] . It was found, however, in sporadic cases in most of the countries at low frequencies (~1%) [27] . In our study, we found 2% of sequences containing this change. In conclusion, the stable evolution of 2009 H1N1pdm offers an opportunity to control its spread and prevent infections. Reports about new mutations, however, will still be important if those changes can confer an enhanced transmissibility or resistance to drugs. Furthermore, a continuous surveillance by public health authorities will be critical to monitor the appearance of new influenza variants, especially in animal reservoirs such as swine and birds, in order to prevent the potential animal-human transmission of viruses with pandemic potential. This work (project code: INV11) was supported by the Consejo Nacional de Ciencia y Tecnología (CONACYT) Programa de Apoyo al Desarrollo de la Ciencia, Tecnología e Innovación en Paraguay (BID 1698/OC-PR). Declared none. Supplementary material is available on the publisher's web site along with the published article.

Severe imported malaria in an intensive care unit: a review of 59 cases

BACKGROUND: In view of the close relationship of Portugal with African countries, particularly former Portuguese colonies, the diagnosis of malaria is not a rare thing. When a traveller returns ill from endemic areas, malaria should be the number one suspect. World Health Organization treatment guidelines recommend that adults with severe malaria should be admitted to an intensive care unit (ICU). METHODS: Severe cases of malaria in patients admitted to an ICU were reviewed retrospectively (1990-2011) and identification of variables associated with in-ICU mortality performed. Malaria prediction score (MPS), malaria score for adults (MSA), simplified acute physiology score (SAPSII) and a score based on WHO's malaria severe criteria were applied. Statistical analysis was performed using StataV12. RESULTS: Fifty nine patients were included in the study, all but three were adults; 47 (79,6%) were male; parasitaemia on admission, quantified in 48/59 (81.3%) patients, was equal or greater than 2% in 47 of them (97.9%); the most common complications were thrombocytopaenia in 54 (91.5%) patients, associated with disseminated intravascular coagulation (DIC) in seven (11.8%), renal failure in 31 (52.5%) patients, 18 of which (30.5%) oliguric, shock in 29 (49.1%) patients, liver dysfunction in 27 (45.7%) patients, acidaemia in 23 (38.9%) patients, cerebral dysfunction in 22 (37.2%) patients, 11 of whom with unrousable coma, pulmonary oedema/ARDS in 22 (37.2%) patients, hypoglycaemia in 18 (30.5%) patients; 29 (49.1%) patients presented five or more dysfunctions. The case fatality rate was 15.2%. Comparing the four scores, the SAPS II and the WHO score were the most sensitive to death prediction. In the univariate analysis, death was associated with the SAPS II score, cerebral malaria, acute renal and respiratory failure, DIC, spontaneous bleeding, acidosis and hypoglycaemia. Age, partial immunity to malaria, delay in malaria diagnosis and the level of parasitaemia were not associated with death in this cohort. CONCLUSION: Severe malaria cases should be continued monitored in the ICUs. SAPS II and the WHO score are good predictors of mortality in malaria patients, but other specific scores deserve to be studied prospectively.

The number of malarial infections acquired by international travellers, thought to be nearly 25,000 cases annually, remains very small compared with the annual global incidence of nearly 250 million malaria cases and about 700,000 related deaths in endemic areas, mainly among young children in Africa. Malaria in non-endemic countries is a challenge since, due to the fact that it is infrequent, experience in diagnosis and treatment is scarce and delays in diagnosis and adequate treatment may be fatal to the patient [1 -3] . The diagnosis of malaria requires a high index of suspicion, as symptoms can mimic many other diseases. Portugal maintains a close relationship with African countries for several reasons and so malaria must be suspected in travellers that return from those countries. The present study includes expatriates (emigrants), people that were born and live in Africa travelling to Portugal to visit friends and relatives, immigrants (people that were born in an endemic area and that live in Portugal) and those travelling for business or tourism. Malaria is a notifiable disease in Portugal; about 50 cases are reported annually to the Public Health System, but cases are under-reported [4] . The notification implies filling-in a form that needs to be posted, which leads to forgetfulness. A person from a non-endemic country (emigrant or expatriate) who stays in malariaaffected areas for less than two years is considered nonimmune, and also at greater risk of more severe forms of the disease [5] . Non-immune people can develop symptoms of falciparum malaria within a month after leaving the endemic area (average 10 days) but, sometimes, after longer periods as can be observed in pregnant women, immigrants, people that were prescribed mefloquine and HIV-infected people [6] [7] [8] . Severe forms of malaria are almost always caused by Plasmodium falciparum; though rare, vivax malaria can also cause severe disease. The object of this study is falciparum malaria solely. Severe forms of malaria should be regarded as a medical emergency and managed in intensive care units (ICU) [9] [10] [11] . Clinical deterioration usually develops three to seven days after fever onset, but it can sometimes develop in the first 24 hours. A high standard of care and continued monitoring in the acute stage is very important to reduce mortality [12, 13] . In 1990, the World Health Organization (WHO) established the criteria for severe malaria which were revised in 2010 (Table 1) , with the purpose of identifying individuals at risk of dying and specifying the risk factors of severe forms of the disease. The guidelines for the treatment of malaria published in 2006 were also revised in 2010 [14] [15] [16] . In general, there is a correlation between the parasite density in the peripheral blood and the severity of the disease and its complications, especially among nonimmune people. In such cases, the patients' clinical condition can deteriorate even after initial appropriate treatment due to exacerbation of systemic inflammatory response, leading to organ dysfunction [17] [18] [19] . The aim of this retrospective study is to describe the clinical spectrum of severe malaria cases admitted to an ICU and identify factors associated with in-ICU mortality. Parasitological diagnosis was done by thin smear and light microscopic observation; parasitaemia quantification was done whenever possible. Daily smears were done until the Plasmodium smear became negative. Since 2000, a immunochromatographic assay for the qualitative detection of Plasmodium antigens were also used at the emergency department (Malaria Now-Binax ® ). Clinical and laboratory parameters were analysed in order to determine which of them were associated with ICU mortality. To evaluate severity at the ICU admission, the Simplified Acute Physiology Score II (SAPS II) was applied. The degree of immunity to malaria was estimated as follows: a) people who had been living in an endemic area for at least two years (emigrant) at the time of the diagnosis were presumed semi-immune, as well as adult migrant Africans that come to Portugal to visit friends or relatives; b) Europeans who travelled occasionally to endemic areas were considered non-immune. WHO's definition for severe malaria was applied and major and minor indicators considered according to the definitions presented in Table 1 . The number of severity criteria was quantified and analysed. Two prognostic scores of malaria were applied: (1) Malaria Prediction Score (MPS) determined by: 2.13 + 0.02 × (age) + 0.25 × (creatinine) -0.24 × (haemoglobin) + 3.05 (malaria cerebral criteria) + 0.8 (presence of pregnancy) + 0.8 (ventilated) (where age = age in years; creatinine is in mg/dl, haemoglobin in g/dl; presence of pregnancy, cerebral malaria or ventilatory support, when present = 1, when absent = 0) [20] .; (2) Malaria Score for Adults (MSA) was applied to all but three children and the score was determined by: 1 (severe anaemia) + 2 (acute renal failure) + 3 (respiratory distress) + 4 (cerebral malaria). The MSA ranges from 0 to 10 [20] . Variables' definitions were: severe anaemia (Hgb < 5 g/dL), thrombocytopaenia (platelets < than 100 x10 5 / uL); acute respiratory distress syndrome (ARDS) (acute respiratory failure defined by an acute hypoxaemia with the ratio of the partial pressure of oxygen in the patients' arterial blood (PaO 2 ) to the fraction of oxygen in the inspired air (FIO 2 ) (PaO 2 /FIO 2 ratio), less than 200 after exclusion of cardiogenic pulmonary oedema by clinical criteria or by a pulmonary capillary wedge pressure (PCWP) < 18 mm Hg in patients with a pulmonary artery (Swan-Ganz catheter); cerebral malaria (unrousable coma), acute renal failure (ARF) (creatinine > 3 mg/ dl or urine output < 400 mL/day in adults), liver dysfunction (ALT at least two times the normal value IU/ mL), hyperbilirubinaemia (bilirubin value > 2.5 mg/dL), hyponatraemia (Na < 125 mEq/mL), acidaemia (pH < 7.25), and hypoglycaemia (blood glucose < 40 mg/dL). Community-acquired co-infection was defined as any infection diagnosed within the first two days of hospitalization. Infections occurring later than this were considered nosocomial. ICU admission criteria other than high parasitaemia were any cerebral dysfunction (GCS < 12), haemodynamic instability (SBP < 80 mmHg), respiratory distress (respiratory rate > 20/min or pCO2 < 32 mmHg), jaundice (bilirubine > 2.5 mg/dl) or renal dysfunction (oliguria or creatinine > 3 mg/dl). Before 1992, the only available treatment for malaria was oral triple therapy (quinine sulphate plus pyrimethamine plus sulphadiazine), whereas after 1992, intravenous quinine dihydrochloride plus clindamycin or doxycycline has been chosen for the treatment of severe cases of malaria. Artemisinin derivatives are not yet available in Portugal. Treatment support includes: blood products to treat severe anaemia and coagulation disorders, intravenous glucose for hypoglycaemia, acetaminophen for hyperpyrexia, if moderate hypoxaemia oxygen by facial mask, antibiotics for bacterial co-infection; haemodynamic and life support according to SSC guidelines since 2005 [21] . Endotracheal intubation and ventilatory support would be done if the patient had respiratory failure or neurologic dysfunction, needing airway protection; if acute oliguric renal failure or severe metabolic acidosis and/or severe electrolytic imbalance were present, renal support, usually a continuous technique, would be started. Statistical analysis was performed with Stata V12. Descriptive statistics included frequency analysis (percentages) for categorical variables and median and interquartile ranges (IQR) for continuous variables. The differences in characteristics according to the outcome were tested using Fisher exact tests or chi-squared test for categorical variables and Wilcoxon tests for continuous variables. From January 1990 to May 2011, a total of 284 patients suffering from malaria were admitted. Fifty nine (20.7%) patients were within the criteria of severe malaria and were admitted in the ICU-ID, a level III medical-surgical unit. Three patients were children aged one, nine and 11. Adults' ages varied from 20 to 71, with a median value (IQR) of 42 (33) (34) (35) (36) (37) (38) (39) (40) (41) (42) (43) (44) (45) (46) (47) (48) (49) (50) , and six were older than 60; 47 (79.6%) were male. Forty one (69%) were Portuguese emigrants working in Africa (for a period of two or more years in 20 patients and varying from two to 34 years; for a period of up to two years in14 patients and for a period that has not been defined in 7); seven patients were tourists and other seven lived in Africa and came to Portugal (five were Africans that came to visit friends and relatives and two were immigrants), and four worked in boats that had made several stops in seaports along the way. All but three had returned from sub-Saharan Africa: 48 (81.3%) from former Portuguese colonies in Africa -Angola (31), Mozambique (9), Guinea-Bissau (7) and Sao Tome and Principe (1); the others had returned from Senegal (2) and Gabon, Malawi, Burkina Faso, Central African Republic, Nigeria, and Côte d'Ivoire, one patient from each country; two other patients had returned from Thailand and one from the Philippines. Fifty one patients (86.4%) were white, six black and two Asian. Malaria immune status was evaluated in 54 (91.5%) patients: 25 (46.3%) were nonimmune and 29 (53.7%) semi-immune, according to the defined criteria. Chemoprophylaxis was prescribed to 14 (23.7%) patients, but discontinued or not taken properly, except from one patient with cerebral malaria that was on mefloquine chemoprophylaxis. Co-morbidities were present in 15 (25.4%) patients: alcoholic hepatic disease, diabetes mellitus and ischemic cardiac disease in five patients; sickle cell anaemia, nephrolithiasis, Still disease and hepatitis C were observed in one patient each. One had been bone marrow transplanted seven years before and was asymptomatic, with no immunomodulatory therapy. An HIV primary infection was diagnosed simultaneously with malaria in one patient, being all other patients HIV negative. Some patients presented more than one comorbidity. The diagnosis was made between the day of return from the endemic area up to the 47th day in 54 patients; in one patient, diagnosis was done on the 120th day after leaving Angola, where he had been for 16 consecutive months; in other four patients, the gap was impossible to determine. Symptoms, namely fever before admission, varied from one day to three weeks, the median (IQR) duration being more than seven days in 44% of the patients. SAPS II was registered in 56 patients and the median value (IQR) was 33 . Malaria diagnosis was established by positive smears for asexual P. falciparum forms. In 48 (81.3%) patients, parasitaemia quantification was ≥ 2% and varied from 2-90%, with a median (IQR) of 15 . Parasitaemia was considered high but was not quantified in 13 (22%) patients. The immunochromatographic assay in whole blood for the qualitative detection of Plasmodium antigens, used at emergency department, was positive in all the samples tested. The most common analytic changes observed in these patients were: thrombocytopaenia in 54 (91.5%) patients associated with DIC in seven (11.8%) and bleeding dis- Considering the major WHO criteria of severe malaria, 29 (49.1%) patients presented five or more dysfunctions. Forty one (69.5%) patients were treated with quinine dihydrochloride, with a 20 mg/Kg loading dose followed by 10 mg/Kg tid, in a four-hour infusion associated with endovenous clindamycine. The other 18 patients (30.5%) were treated with oral quinine sulphate associated with doxycycline or pyrimethamine plus sulphadiazine in the first period of the study. Blood smears became negative for Plasmodium from the second to eighth day of treatment, except in a patient whose parasitaemia was 90% and who died with positive smears in the second day after admission. All patients admitted to the ICU were close monitored and had treatment support according to the dysfunctions presented, namely transfusions of packed red cells whenever haemoglobin was lower than 7.5 g/ dL, associated with platelets and fresh frozen plasma in patients with bleeding disorders; blood products were also given before an invasive technique whenever platelets were lower than 50 × 10 5 /uL or if coagulopathy was present. Fluids were administered to correct hypovolaemia monitored by central venous pressure, mean arterial pressure and urinary output. Even so, 29 (49.1%) patients needed vasopressor support, with noradrenaline or dopamine or both. Ventilatory support was needed in 22 (37.3%) patients, for a period varying from two to 74 days (median 10 days). Two patients with severe ARDS with refractory hypoxaemia had been put on venovenous Extracorporeal Membrane Oxygenation (ECMO). In one of the patients, ECMO was completely successful and the patient is now doing well, without significant respiratory sequels; the other patient died on ICU in refractory shock. Eight (13.6%) patients required renal replacement therapy. Six patients had upper gastrointestinal bleeding and did an endoscopy for further study. All patients began enteric nutrition in the first 48 hours, if not contraindicated. Glycaemia control was monitored, being hypoglycaemia registered and treated in 18 patients (30.5%). Other simultaneous infections occurred in some patients. Thirteen (22.0%) patients had pneumonia: in two of them, pneumonia was considered communityacquired; Pneumococcus and Klebsiella spp. were isolated from the tracheal aspirate, in each patient. In the remaining 11 cases, pneumonia was nosocomial and, apart from one isolate (an MDR Pseudomonas aeruginosa), no agent was identified; six of these patients were ventilator associated pneumonia. All patients but one recovered. Another seven (11.9%) patients developed acute acalcoulous cholecystitis, which was managed with medical treatment in five and, in the other two, a percutaneous drainage was done, with a favourable evolution in all of them. Two (3.3%) patients had nosocomial central venous catheter-associated sepsis, having Acinetobacter baumanni been identified (from central catheter and peripheral venipuncture) in one patient and Pseudomonas aeruginosa MDR in the other; both were treated with antibiotics and had a favourable outcome. The UCI stay ranged from one to 81 days, being the average (IQR) four days (two to twelve) in survivors and eight days (three to twenty seven) in the patients that died (p = 0.185). All patients who came out alive from the ICU were discharged from the hospital. In two of them, an acute disseminated encephalomyelitis (ADEM) was diagnosed based on neurologic dysfunction and cerebral resonance magnetic image; both had a favourable outcome. All patients were also evaluated after hospital discharge for follow-up of residual dysfunctions and no other sequelae were detected; in those whose rapid malaria test had been consistent with coinfection with P vivax, primaquine was prescribed after excluding Glucose-6-Phosphate Dehydrogenase deficiency. The global case fatality rate of the infectious diseases department was 3.1%, but the ICU fatality rate was 15.2% (nine patients). Eight were workers: seven returned from Angola, one from Thailand and the 9th was a tourist who returned from Mozambique. Death occurred between the second and 28th day. In all but three patients death occurred in the acute phase of the illness. One of them was a male nurse who, despite compliance to mefloquine chemoprophylaxis, had malaria with 30% of parasitaemia; he was conscious when he was admitted but had a rapid neurologic deterioration in some hours with severe intracranial hypertension and an irreversible brain swelling on cerebral computerized tomography image, with brain death in the first 24 hours. In the other three patients, who died later on, death was related to ARDS complications. The variables associated with death ( ; p = 0.08) and hyperbilirubinaemia (77.8% vs 40.4%; p = 0.066), although more frequent in patients who died, did not reach statistical significance. The duration of the symptoms before admission has no statistical significance when considering the risk of death (p = 0.82), but the same was not true when we analysed the time since the patients' arrival from the endemic area to diagnosis, which was statistically associated with a worse prognosis (p = 0.02). The number of organ dysfunctions as described in septic shock was correlated with prognosis and this is well documented by the presence of more than four WHO major criteria of severe falciparum malaria and the risk of death (p < 0.001). Renal replacement (88.9% vs 46.0%; p = 0.027) and mechanical ventilation (88.8% vs 28.0%; p < 0.001) were also significantly more frequent in those patients who died. The median (IQR) duration of the ICU stay increased, but without reaching statistical significance in the nine patients who died vs 4 (2-13); p = 0.575). The MPS ranged from 1.78-5.20 (median, IQR: 1.8; 0.31-3.0) and the MSA from 3-9 (median; IQR: 2; 0-5). Comparing the median value of the two prognostic scores, both were good predictors of death in this cohort (Table 2) . When the four scores were compared, MPS (AUC 0.77; IC95% 0.64-0.90), MSA (0.84; IC95% 0.70-0.98), SAPS II (0.90; IC95% 0.81-0.99) and WHO score (0.91; IC95%0.82-1.0), based on the area under the ROC curve, the best death predictors were SAPS II and the WHO scores ( Figure 3 ). In this retrospective study 59 consecutive patients with severe malaria admitted in a Portuguese infectious disease UCI over a 22 year period were considered. Obviously, the longer the period, the greater the impact of lost data, but the sample did not change over time and it includes mostly adult white males, middle-aged, travelling for work reasons. Malaria is a complex disease and it is well known that the higher the parasitaemia the greater the systemic inflammatory response and the production of cytokines and so, a major risk of organ dysfunctions, but some aspects concerning severe forms need to be clarified [21] [22] [23] . The WHO has redefined the criteria of severe malaria and pointed out a list of complications, but prognostic importance of each complication has not been well established. So, parameters usually associated with poor prognosis should not be forgotten and the patients' monitoring according to their presence or possibility, including the admission to an ICU [22] . In this paper, the ICU admission was based on the presence of at least a life-threatening organ dysfunction, haemodynamic instability, as well as high parasitaemia, which often precedes clinical deterioration. Delay in diagnosis of malaria has been associated with a worse prognosis [24, 25] , but this was not found in this review, where 49% of the patients had the symptoms for more than a week before diagnosis, without any correlation with a worse prognosis. It is difficult to explain the reasons why. The validity of the variable (duration of symptoms) may be questioned because the time since arrival to diagnosis was statistically significant for the risk of death. A delay in diagnosis can be caused by patient ignorance about the potential severity of malaria, auto medication for fever with antipyretics before seeking medical attention or the doctor's delay in considering the diagnosis, only overcome with proper training and experience. In this review, the case fatality rate was 15.2% (nine patients) and factors statistically associated with death were time since arrival to diagnosis (p = 0.027), SAPS II score (p < 0.001), cerebral malaria (p = 0.019), acute renal failure (p < 0.001), ARDS (p < 0.001), hypoglycaemia (p = 0.02), DIC or spontaneous bleeding (p = 0.007) and acidaemia (p < 0.001) ( Table 2) ; the need for renal replacement (p = 0.027) and invasive ventilation (p = 0.027) was significantly more frequent in those who died. These data are in agreement with literature review, where the fatality rate is higher in cerebral malaria and acute respiratory distress syndrome (ARDS). Two of the patients were put on ECMO, used as a rescue therapy [26] [27] [28] . They had been working in Angola and Mozambique for some years, had parasitaemia 2% and 3% respectively and severe ARDS and multiorganic dysfunction. One of them survived and recovered without respiratory sequelae, the other died in refractory shock. These cases show the importance of a clinical surveillance of all patients, because the evolution is sometimes unfavourable in cases that initially seemed benign. The patient on mefloquine chemoprophylaxis that died had a parasitaemia of 30% and a fulminant evolution. Unfortunately he was not autopsied and it was impossible to investigate if it was a problem of resistance to mefloquine. The number of organ dysfunctions may be directly correlated with mortality as was observed in this cohort, where the presence of more than four WHO major criteria for severe falciparum malaria significantly increased the risk of death (100% vs 40%; p < 0.001) [11, 29, 30] . Unexpectedly, age, duration of symptoms, immune status, parasitaemia level, bilirubin value and the presence of shock were not statistically associated with mortality, even though hyperbilirubinaemia and shock were more frequent in patients who died, but diverse results are found in different studies [31] . These results should be interpreted with some caution because the study covers a small number of patients over a long period of time, although this does not necessarily explain the discrepancies between this and other studies. Previous papers reporting malaria parameters associated with mortality, mainly retrospective, have different aims and different degrees of severity and only a few of them take ICU cohorts into consideration [12, 13, 32, 33] . This is the case of a study reported in 2003 from a French group that evaluated 188 ICU patients with imported falciparum malaria over a 12-year period [10] with an overall mortality rate of 5.3% but, in most severe patients, the mortality rate was 11%. Factors associated with death were SAPS II, shock, acidosis, coma, pulmonary oedema and coagulation disorders. No predictive malaria score system had been used as routine. Two malaria prognostic scores (MSA and MPS) used in this cohort were good death predictors [20] . When the four scores applied, SAPS II, MPS, MSA and WHO score were compared, SAPS II and the WHO score were the most sensitive to predict death ( Figure 1) . Patients with severe forms of malaria are highly susceptible to bacterial infections and WHO recommends concomitant antibiotic therapy. In this review, some infections were documented: 13 patients had pneumonia, two of them community-acquired and 11 nosocomial, seven had acute acalculous cholecystitis and, in another two patients, central venous catheter-associated sepsis occurred. Although malaria acute acalculous cholecystitis is rarely reported, seven cases (11.9%) were found in this cohort [34, 35] . The HIV primary infection was considered as the patient has been tested six months before, when he was treated for a cellulitis, having on that date a negative test for the HIV virus. The patient had a favourable evolution without concomitant antiretroviral therapy. There is no consensus about severity or complicated malaria in HIV co-infected adults but the fact that fever is a common symptom to both diseases can delay both diagnoses [7] . Post-malaria neurological syndrome, another rare complication, was reported in two patients who recovered completely. This neurological syndrome has been defined as the acute neuropsychiatric manifestation in patients recently recovered from malaria [36] . The prevention of malaria is essential with easy interventions such as different barrier methods and chemoprophylaxis whenever justified. Pre-travel advice and general practitians should stress prevention measures. Although the study does not point out that a delay in diagnosis is directly connected with a worsening of malaria, people at risk should be informed about the malaria symptoms, allowing an early diagnosis and treatment. Although anti-malarial drug medication is the only intervention of proven efficacy to treat severe malaria, it is very important to monitor and treat severe forms with organic dysfunctions in ICU [37] .

Use of controlled low dose gamma irradiation to sterilize allograft tendons for ACL reconstruction: biomechanical and clinical perspective

As reviewed here, numerous biomechanical and clinical studies support the use of controlled, low temperature irradiation of allograft tendons, to provide both excellent clinical results and medical-device grade sterile allografts with minimal risk of disease transmission.

The preferred method of terminal sterilization for tendon allografts, gamma irradiation, remains a concern to some surgeons. While some older studies have shown that higher, uncontrolled doses of gamma irradiation ([3 Mrad) can have detrimental effects on the strength of allograft tissue, numerous studies suggest that the currently used practices of low dose, controlled, low temperature gamma irradiation are effective to achieve terminal sterilization without detrimental impact on allograft tissue strength. In this review, irradiation methods are presented as well as biomechanical, clinical, and safety assessments of irradiated tendons used for ACL reconstructive procedures. Prior to assessing studies regarding irradiated tendons, it is important to understand how irradiation methods are reported. Key variables of irradiation include: • Target dose • Dose range • Temperature of irradiation • Tissue treatment prior to irradiation The targeted dose is that (e.g., 22 kGy) which the tissue is intended to be exposed. However, the manner in which tissue is irradiated in its container or chamber does not allow all tissues to receive an exact similar dose. If only a single targeted dose is reported for tissue treatment, then that is likely the minimal exposure dose. For example, an exposure reported as 25 kGy (2.5 Mrad) likely indicates that all materials received at least 25 kGy exposure and that some may have received a much higher dosage (e.g., the outer grafts in a container). Thus, when only a single dose is reported it is fair to consider that to represent the minimal exposure and that some or most tissues will receive a higher dose. The most accurate way to report an irradiation dose is as a dose range, e.g. 15-18 kGy (1.5-1.8 Mrad), which should indicate both the minimum and maximum dose exposure throughout the irradiated container. In order to minimize any negative material impact, it is important that both low doses and tightly controlled dose ranges be employed. It is very difficult to interpret any study that does not include a dose range. Again, if only a single dose is given, it should be considered the minimum exposure. The third variable, temperature has been shown to be important as low (dry ice) temperatures serve to minimize any free radical generation and subsequent tissue damage (Anderson et al. 1992; Hamer et al. 1999) . Prior to knowledge of the importance of a controlled irradiation dose range at low temperatures, it is difficult to rely on the findings of studies where the details are not reported. Ideally a low average dose, at a narrow dose range, and at low temperatures are all factors in minimizing any irradiation-mediated alteration of material properties. In addition, the treatment of any tendon prior to irradiation could play a role in how the tissue is potentially impacted. There are certain cleaning and disinfection methods that involve harsh chemicals or physical forces on the tissue. These may damage grafts independent of irradiation or pre-dispose grafts to alteration by irradiation. To best interpret study results involving irradiated tissue, it is important to know how a tissue was treated prior to irradiation and exactly how it was irradiated. In testing biomechanical properties of irradiated tendons, Balsly et al. (2008) reported no change in graft strength or elastic modulus for bone-patellar tendon-bone (BPTB) grafts, anterior tibialis tendons, semitendinosus tendons, or fascia lata soft tissue grafts when treated with sterilizing, low dose (18.3-21.8 kGy) gamma irradiation at dry ice temperatures. Likewise, Greaves et al. (2008) investigated the biomechanical properties of low dose (14.6-18.0 kGy) gamma irradiation on tibialis tendon allografts. In this matched pair study, 63 tibialis tendons were irradiated on dry ice while the contralateral tendons from the same respective donors were not irradiated. The study found that low dose irradiation did not significantly affect the failure load of either single stranded or double stranded tibialis tendon grafts (Table 1 ). In a similar study, Roche et al. (2005) investigated the ultimate tensile strength of low dose gamma irradiation (15.4-15.5 kGy) on patellar ligaments and fascia lata allografts. Each irradiated allograft was matched with a control non-irradiated graft from the same donor to limit biomechanical variability resulting from different donors. The study did not find a statistical difference in the tensile strength between the matched low dose irradiated and non-irradiated allografts. In addition, Gibbons et al. (1991) showed in a biomechanical study that maximum stress, maximum strain, and strain energy density to maximum stress was not significantly reduced in goat BPTB grafts irradiated with 20 kGy of gamma irradiation. Further, Goertzen et al. (1995) found no significant difference in strength between canine BPTB grafts irradiated with 25 kGy and non-irradiated BTPB grafts after being implanted in an ACL reconstruction for (Haut and Powlison 1990; Mae et al. 2003; Maeda et al. 1993 Maeda et al. , 1998 Smith et al. 1996) of irradiated tendons indicate that treatment below 20-25 kGy have minimal impact on biomechanical properties. There is also clinical evidence supporting the utility of low-dose irradiated tendons, which also have the advantage of minimizing any risk of disease transmission. Fanelli et al. (1996) compared irradiated BPTB and Achilles tendon allografts versus BPTB autografts for ACL reconstruction in patients who had combined ACL/PCL instability. No irradiation levels were given, so the presumption is of at least a 15-25 kGy dose. Although the sample size was low (20 patients), the prospective study was the largest study to date (1996) that evaluated ACL/PCL instability. This clinical study found equivalent results between irradiated allografts and autograft tendons. In a technique article, Harner and Elkousy, along with lead author Sekiya et al. (2002) , noted they only use patellar tendon allografts that have been irradiated for ACL reconstruction. In further support, Rihn et al. (2006) investigated the irradiation variable for ACL reconstruction in a clinical study involving 102 patients with an average follow-up of 4.2 years. The study found that 2.5 Mrad of gamma irradiation on BPTB allografts, which is effective in eliminating bacteria, does not compromise the clinical effectiveness of the grafts ( Table 2 ). The authors concluded ''[t]hese data suggest that irradiation can be used to sterilize BPTB allograft without adversely affecting clinical outcome.'' Taken together, the above articles suggest that a sterilizing dose of irradiation may not be of clinically significant concern when using allografts for ACL replacement. While three reports in particular have presented higher failure rates for irradiated allografts in ACL reconstructions (Rappe et al. 2007; Sun et al. 2009; Prodromos et al. 2007) , there are significant questions regarding these studies or data analysis. Those study limitations are detailed here. The most significant issues with the first study (Rappe et al. 2007) include questionable follow-up methodology and a lack of information regarding the irradiation process used. The study followed up with *73% of irradiated graft patients compared with *93% of non-irradiated graft patients. It is unclear why a substantially fewer number of patients in the irradiated graft group returned for follow up care especially since they reportedly had significant graft failure (33%). Also, the method used to irradiate the grafts is unknown, e.g., the temperature of irradiation. The irradiation dose is given as 20-25 kGy and it is unknown if this is a true dose range and what the dose range of irradiation exposure actually was. In addition, other processing methods are unknown, including whether these grafts were also exposed to harsh solvents or cyclic pressures as performed by some banks, such that they may have been prone to damage from the higher irradiation dose used. Also, one surgeon reported twice the failure rate as the other in the study and the recipient age in relation to failures in both groups was not reported, making interpretation difficult. Finally, many tissue providers will routinely treat 'non-irradiated' or 'non-sterilized' grafts with 10-15 kGy doses of pre-processing irradiation for safety reasons so it remains unknown whether the control grafts were truly non-irradiated and also whether the irradiated grafts were double-dosed. There are also numerous concerns with the Sun et al. study (2009) . Only the target dosage is given (25 kGy) with unknown dose range. Also, these grafts were irradiated at room temperature and the grafts were not disinfected prior to irradiation (author, private communication). In addition, grafts were soaked in iodine prior to use. These graft treatments compromise any meaningful interpretation of the results. As significantly, the study has inconsistent results that possibly indicate an issue in either the surgery or measuring techniques. Unfortunately, the allograft patients exhibited a significant increase in duration of post-operative fever over autograft patients. The average duration of fever for an irradiated allograft patient was over 1 week (8.8 days) versus 4.7 days for the autograft patients. In the discussion, the authors state that this high fever rate ''was associated with…[different possibilities, or]…the real ability of tissue banks in our country [China] to process allografts''. The study also concludes that irradiated bone patellar-tendon bone (BPTB) allografts are clinically inferior to both nonirradiated BPTB allografts and BPTB autografts because of the laxity measurements with a KT-2000 arthrometer. The study reports only 31.3% of irradiated BPTB allografts had less than 3 mm of laxity while 85.3% of the non-irradiated and 87.8% of the autograft group had less than 3 mm of laxity. This is a surprising difference made even larger by the report of 34.4% of the irradiated group exhibiting more than 5 mm of laxity (defined as graft failure by the authors). These extreme percentages should indicate noteworthy irradiated graft patient dissatisfaction as well as significantly different results in other subjective and objective tests. In contrast, however, there were no significant differences in the overall IKDC scores between any of the three groups. Irrgang et al. (1998) noted that the IKDC was an especially rigorous evaluation tool because the lowest score received in any given area becomes the overall score instead of combining the averages like other evaluation systems. This makes the validity of the Sun et al. study even more uncertain since laxity measurement is a component of the overall IKDC score. It is unclear how there was no significant difference in the overall IKDC score among the 3 groups when the irradiated graft group had such extreme laxity measurements. Furthermore, the objective range of motion (ROM), vertical jump, and one-leg hop tests found no significant difference in any of the groups. There were also no significant differences among the 3 groups for mean Lysholm, Tegner, or Cincinatti knee scores. Moreover, there was not a significant difference in patients' satisfaction with their ability to participate in sports in any of the groups. One would expect significant differences in all or most of these tests and responses if 68.7% of irradiated allograft patients had greater than 3 mm laxity. The IKDC system is one of the best evaluation tools to measure ACL reconstruction results (Foster et al. 2010 ) and the results of this test and all the others should have been balanced against the arthrometer measurements. This balance is particularly important because there may be no correlation between laxity measurements and functional outcome (Mirzayan 2005) . Pollet et al. (2005) investigated this correlation in a prospective, clinical study of 29 ACL deficient patients with an average 33 month followup. After comparing anterior knee laxity, questionnaire based on IKDC score, sports activity rating scale (SARS), activities of daily living (ADL), and other tests, the study found ''no correlation between the joint laxity and the functional outcome score.'' This lack of correlation is actually supported by the Sun et al. (2009) study in which, according to the authors, almost 4 times as many irradiated allograft patients had graft failure based on laxity measurements but there was no statistical difference in patients' satisfaction in their postoperative sports activity or overall IKDC score. At the very least, the inconsistent results and non-standard tissue treatment methods should have given the authors pause before making the recommendation to completely discontinue use of irradiation to sterilize allografts. Prodromos et al. (2007) performed a meta-analysis on stability of autografts and allografts for ACL reconstruction. This included a sub-analysis of nonirradiated vs. irradiated grafts. They came to the following conclusion: ''The direct deleterious effects of graft radiation are an additional area of concern. The stability rate in the radiation-sterilized grafts in this study was strikingly low.'' However, the data used to draw this conclusion is heavily skewed by one particular study. In detail, to examine irradiated tendons, the authors included three studies, here called Noyes, Gorschewsky, and Rihn. They based conclusions on normal stability rate (which was 43% for irradiated vs. 63% for non-irradiated allograft) and abnormal stability rates (which was 31% for irradiated and 12% for non-irradiated allografts). This certainly appears negative for irradiated allografts. However, the irradiated group included in the Gorschewsky study included more grafts, and thus was more heavily weighted, than the other two studies combined and, most importantly, included a process method with steps containing acetone, sodium hydroxide, and hydrogen peroxide. These harsh chemicals can be quite damaging to soft tissues and no conclusion can be drawn from the fact that these grafts were also irradiated unless the proper controls were included (treatment with these chemicals without irradiation). If this single study is removed from the meta-analysis, then the comparisons become: normal stability rates of 62% for irradiated vs. 63% for non-irradiated allografts and abnormal stability rates of 15% for irradiated and 12% for non-irradiated allografts, respectively. Thus, the exclusion of the harsh chemically treated graft data set, yields results suggesting equivalent performance for irradiated vs. non-irradiated grafts. Further, note that this study that was, in fact positive regarding irradiation reported on tissue irradiated with 2.5 Mrad, which is even above commonly used levels of 13-18 kGy, further suggesting the utility of terminal sterilization via gamma irradiation. It appears that there still exists confusion as to the definitions of sterility and processing methods. Sterile or aseptic tissue recovery by itself is mistakenly considered as a method that will result in the supply of sterile grafts hence making terminal sterilization unnecessary (Marrale et al. 2007) . Recovery under aseptic conditions seeks to ensure that no further bioburden is introduced from the environment but does not remove existing bioburden in the tissue (Vangsness 2004; Vangsness et al. 2003) . Tissue banks must use disinfection steps and/or terminal sterilization to accomplish sterilization of existing bioburden. Furthermore, aseptic recovery occurring in a surgical operating room can only result in, at best, a sterility assurance level (SAL) of 10 -3 , and then only if properly validated, versus a terminal sterilization SAL of 10 -6 . Sterility assurance level gives the probability of there being viable microorganisms on a particular graft unit, instrument, etc. A SAL of 10 -6 indicates there is only at most a 1 out of 1,000,000 chance that a viable organism exists with any single graft compared with an SAL of 10 -3 which indicates a 1 out of 1,000 chance (Vangsness et al. 2003) . Some tissue banks choose to use terminal sterilization to increase the likelihood of the safety of their tissue. If the allograft can be guaranteed to a SAL of 10 -6 then it may possibly possess an even lower degree of infection risk than an autograft procedure (Bryans et al. 2010; Katz et al. 2008) . While the potential risk of viral transmission is extremely serious, it should be kept in perspective that this risk is virtually non-existent if the allograft is procured from a bank using intensive donor screening, tissue disinfection procedures, and terminal sterilization methods that inactivate viruses. While some studies reported that at least 30 kGy of gamma irradiation is needed to inactivate HIV, these studies have assumed HIV is present in high density levels (Fideler et al. 1994; Hernigou et al. 2000) . At least 30 kGy may be necessary to inactivate high density amounts of HIV but is excessive for lower density levels of the virus. If in the extremely unlikely event that HIV is present at all, the donor screening and tissue disinfection procedures help ensure that the virus will only be present in extremely low density amounts. The low 10-20 kGy dosage used for terminal sterilization is able to deactivate 99.9% of any remaining low-density HIV in allograft tissue (Moore 2010 ). The preferred method of terminal sterilization for allografts, gamma irradiation, remains a concern to some surgeons. However, while some studies have shown high dose gamma irradiation ([3 Mrad) can have detrimental effects on the strength of allograft tissue, numerous studies have shown that the currently used controlled and low doses of gamma irradiation are effective in terminal sterilization and have no detrimental effect on allograft tissue strength. Rihn et al. (2006) determined that not only is using 25 kGy of gamma irradiation on BPTB allografts effective in preventing bacterial infection but it does not compromise the clinical effectiveness of the graft. The results are also comparable for soft tissue allografts. Balsly Cell Tissue Bank (2012 ) 13:217-223 221 et al. (2008 reported no change in graft strength for patellar tendons, anterior tibialis tendons, semitendinosus tendons, and fascia lata soft tissue grafts when low dose gamma irradiation was used to terminally sterilize at low temperatures. Greaves et al. (2008) found low dose gamma irradiation did not affect the strength or stiffness of soft tissue tibialis tendon allografts. The terminal gamma irradiation is necessary in order to provide allograft tissue with a SAL of 10 -6 which is equivalent with implantable medical devices. Low dose gamma irradiation (10-15 kGy) in combination with donor screening and tissue processing procedures allows for thorough bactericidal treatment while maintaining intrinsic biomechanical properties and ensuring successful clinical performance (Block 2006) . As reviewed here, numerous biomechanical and clinical studies support the use of controlled, low temperature irradiation of allograft tendons, to provide both excellent clinical results and medical-device grade sterile allografts with minimal risk of disease transmission. Open Access This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.

DC-SIGN (CD209) Promoter −336 A/G (rs4804803) Polymorphism Associated with Susceptibility of Kawasaki Disease

Kawasaki disease (KD) is characterized by systemic vasculitis of unknown etiology. High-dose intravenous immunoglobulin (IVIG) is the most effective therapy for KD to reduce the prevalence of coronary artery lesion (CAL) formation. Recently, the α2, 6 sialylated IgG was reported to interact with a lectin receptor, specific intracellular adhesion molecule-3 grabbing nonintegrin homolog-related 1 (SIGN-R1) in mice and dendritic cell-specific intercellular adhesion molecule-3 grabbing nonintegrin (DC-SIGN) in human, and to trigger an anti-inflammatory cascade. This study was conducted to investigate whether the polymorphism of DC-SIGN (CD209) promoter −336 A/G (rs4804803) is responsible for susceptibility and CAL formation in KD patients using Custom TaqMan SNP Genotyping Assays. A total of 521 subjects (278 KD patients and 243 controls) were investigated to identify an SNP of rs4804803, and they were studied and showed a significant association between the genotypes and allele frequency of rs4804803 in control subjects and KD patients (P = 0.004 under the dominant model). However, the promoter variant of DC-SIGN gene was not associated with the occurrence of IVIG resistance, CAL formation in KD. The G allele of DC-SIGN promoter −336 (rs4804803) is a risk allele in the development of KD.

Kawasaki disease (KD), mucocutaneous lymphnode syndrome, is a systemic vasculitis that predominantly affects children under the age of five years. Although the cause is still unknown, KD is the most common cause of acquired heart disease during childhood in the developed countries at this time. Coronary artery lesions (CAL) are the major complications of KD. There is a 15-25% chance of CAL developing in KD patients without early treatment [1] . Although the exact therapeutic mechanisms have not been fully established, high-dose intravenous immunoglobulin (IVIG) is the most effective therapy for KD to reduce the prevalence of CAL [2] . Many potential mechanisms of action for IVIG have been suggested [3] . Of them, at least three main mechanisms are suggested to explain the anti-inflammatory function of high dose IVIG: first, high-dose IgG saturates the neonatal FcRs (FcRn) and leads to the increased catabolism of autoantibodies; second, high-dose IgG saturates the activating Fcγ receptors (FcγRs) and prevents autoantibody-mediated activation of leukocytes; third, high-dose IgG increases the cell surface expression of inhibitory Fcγ receptors [4] . A single, N-linked glycosylation site exists at the amino acid 297 in the heavy chain of all IgG subclasses with approximately 10% terminating in sialic acid [5] . Recently, the α2, 6 sialylated IgG was reported to interact with a lectin receptor, specific 2 The Scientific World Journal intracellular adhesion molecule-3 grabbing nonintegrin homolog-related 1 (SIGN-R1) in mice and dendritic cellspecific intercellular adhesion molecule-3 grabbing nonintegrin (DC-SIGN) in humans, and to trigger an anti-inflammatory cascade that promotes the upregulation of inhibitory FcγRs on inflammatory macrophages [6] . There has been some evidence demonstrating the role of DC-SIGN promoter variants in the susceptibility to or the protection against various infectious diseases, such as dengue fever, tuberculosis, and AIDS [7] [8] [9] . However, whether DC-SIGN promoter variants have effects on susceptibility to KD is still unknown. Since DC-SIGN, also known as CD209, is so important for the anti-inflammatory functions of IVIG, it is reasonable to hypothesize that a functional single nucleotide polymorphism (SNP) in the CD209 molecule will be involved in the pathogenesis of KD or response to IVIG treatment. We hypothesized that the SNP rs4804803 of DC-SIGN promoter may be involved in the susceptibility to KD, CAL formation, coronary artery fistula formation, and IVIG treatment response in KD patients. To test this hypothesis, we conducted a case-control study involving 278 patients with KD and 243 controls. All patients studied were children who fulfilled the diagnostic criteria for KD and were admitted for IVIG treatment at Chang Gung Memorial Hospital-Kaohsiung Medical Center, from 2002 and 2009. All patients were treated with a single high-dose IVIG (2 g/kg) over a 12hour period [10] [11] [12] . This study was approved by the Institutional Review Board of Chang Gung Memorial Hospital with written consent statement. We excluded patients who did not fit the diagnostic criteria of KD. CAL was defined by the internal diameter of the coronary artery being greater than 3 mm (4 mm, if the subject was over the age of 5 years) or the internal diameter of a segment being at least 1.5 times that of an adjacent segment, as observed in the echocardiogram [13, 14] . KD patients with coronary artery ectasia or dilatation which was disappearing within the initial 6-8 weeks after the onset of illness was defined as transient CAL [3] . The diagnosis of coronary artery fistula (CAF) was made mainly by pulsed Doppler and color flow imaging [15] . IVIG treatment responsiveness was defined as defevrscence 48 hrs after the completion of IVIG treatment and no fever (temperature, >38 • C) recurrence for at least 7 days after the initial IVIG treatment with marked improvement of inflammatory signs [16] . Patients with IVIG resistance received another dose of IVIG (1-2 g/kg) or other anti-inflammatory regiments. Children who were admitted for upper and/or lower respiratory tract infections (including acute bronchiolitis, acute pharyngitis, acute bronchitis, croup, and acute tonsillitis) were also collected as control subjects for comparison during the same study period, as we have previously described [12] . Genotyping of CD209 rs4804803 SNP. Genomic DNA was isolated from heparin-anticoagulated blood samples using a standard phenol-chloroform extraction followed by 70% alcohol precipitation. Genotyping for the CD209 variant (−336 A/G; rs4804803) was carried out using Custom TaqMan SNP Genotyping Assays (Applied Biosystems, Foster City, CA, USA). The primer sequences were 5 -GGACAGTGCTTCCAGGAACT-3 (forward) and 5 -TGTGTTACACCCCCTCCACTAG-3 (reverse). The Taq-Man minor groove binder probe sequences were 5 -TACCTGCCTACCCTTG-3 and 5 -CTGCCCACCCTTG-3 . The probes were labeled with the TaqMan fluorescent dyes VIC and FAM, respectively. The PCR was conducted in total volume of 15 μL using the following amplification protocol: denaturation at 95 • C for 10 min, followed by 40 cycles of denaturation at 94 • C for 20 s, followed by annealing and extension at 60 • C for one minute. After the PCR, the genotype of each sample was determined by measuring the allele-specific fluorescence in the ABI Prism 7500 Sequence Detection System, using SDS 1.1 software for allele discrimination (both Applied Biosystems). To validate the genotyping by real-time PCR analysis, 100 PCR products were subject to restriction fragment length polymorphism (RFLP) analysis with MscI restriction enzyme (New England Biolabs, Beverly, MA, USA) and showed a 100% identical result between these two genotyping systems as noted in our previous report [17] . The Hardy-Weinberg equilibrium was first checked. The statistical differences between case and control in genotype and allele frequency were assessed by chi-square test. The statistical differences in the genotype and allele frequency of KD patients with and those without CAL formation, aneurysm formation, patients responding to IVIG, and those showing resistance were assessed using chisquare test. SAS 9.1 for Windows was used for data analysis. Associated with the Susceptibility of Kawasaki Disease. In this study, a total of 278 KD patients were included, of which 35 patients (12.6%) were resistant to initial IVIG treatment, 42 patients (15.1%) had CAL formation and 13 patients (4.7%) developed coronary artery fistula. In this study, as shown in Table 1 , the difference of rs4804803 genotype between KD patients and controls was statistically significant (P = 0.004, dominant model, Table 1 ). Minor G allele of rs4804803 was over represented in the KD patients as compared with the controls (8.1 versus 3.5%). We further evaluate the relationship between rs4804803 and the risk of IVIG resistance or CAL formation. As shown in the Tables 2 and 3, the frequency of AA genotype was higher in the patients with IVIG responsiveness (86.0 versus 77.1%) and without CAL formation (85.2 versus 83.3%). The genotype or allele frequency of rs4804803, however, was not statistically associated with IVIG resistance ( Table 2 ) or CAL formation (Table 3) . To further identify the role of rs4804803 of CD209 in the pathogenesis of coronary artery fistula in KD patients, we performed a subset analysis in cases that were reported as having fistula formation (13/278, 4.7%). Subset analysis between cases with coronary artery fistula and rs4804803 did not yield any significant results (Table 4 ). DC-SIGN is a transmembrane lectin receptor on dendritic cells with multiple immune modulation function [18] DC-SIGN can recognize many pathogens, such as viruses (HIV-1, dengue, and measles virus) [19] [20] [21] , bacteria (Helicobacter pylori, Mycobacterium tuberculosis) [22] , and fungi (Candida albicans and Aspergillus fumigatus) [23] contributing to generation of pathogen-tailored immune responses and immunosuppressive response by the MAPK pathway in DCs [24] . The real cause of KD remains unknown. It is generally accepted that KD results from an undefined infectious process trigger in a genetically predisposed individual [25] . A genetic predisposition is suggested based on clinical and epidemiologic features [1, 26] . In this study, we investigated whether the polymorphism of DC-SIGN (CD209) promoter −336 A/G (rs4804803) was associated with susceptibility and CAL formation in KD. Our study showed that the allele −336G was associated with susceptibility to KD. To the best of our knowledge, this is the first study to explore the association between DC-SIGN polymorphisms and susceptibility to KD. Immunoglobulin is well known for its defensive role in pyogenic infection. In addition to its protective role, immunoglobulin G (IgG) was also noted to have anti-inflammatory effects at high doses. Recently, the α2, 6 sialylated IgG was reported to interact with a lectin receptor, SIGN-R1 in mice and DC-SIGN in humans, and to trigger an antiinflammatory cascade. Thus it is reasonable to hypothesize that a functional SNP in the DC-SIGN molecule will be involved in the response to IVIG treatment. However, in our study, we found the variant and haplotype of −336A/G in the DC-SIGN gene did not associate with the occurrence of IVIG resistance or CAL formation in KD. Because of its highly polymorphic nature and numerous SNPs of DC-SIGN gene 4 The Scientific World Journal [27] [28] [29] , further investigation into other candidate SNPs contributing to KD morbidity is needed. Besides dendritic cells, IVIG was observed to affect many other cells, including endothelial cells, monocytes, neutrophils, and T and B cells [30] [31] [32] . The numerous effects of IVIG therapy also partly explain there not being an association of −336A/G SNP of the DC-SIGN gene with IVIG resistance in KD. There were some limitations with regards to this study. First, the relatively small sample size of this study might prevent some of the detected associations from being statistically significant. Second, our study results need to be validated across different populations. Since the incidence of KD in Asian populations is much greater than among Caucasians [1] , the host's genetic background must be considered in the study of KD. In our control group, the frequency of the −336G DC-SIGN gene allele was 3.5%. This result agreed with previous reports showing very low −336G DC-SIGN gene allelic frequency in Asians [33, 34] . The highest −336G allelic frequency was found in African populations (35-48%), next in Caucasian populations (20%), and the lowest was observed in Asians [8, 33] . In conclusion, from our study, we found the G allele of DC-SIGN promoter −336 (rs4804803) to be a risk allele in the development of KD in a Chinese population. Further studies to explore the effects of other SNPs of DC-SIGN or a combination of genes are needed.

Interferon-Induced Ifit2/ISG54 Protects Mice from Lethal VSV Neuropathogenesis

Interferon protects mice from vesicular stomatitis virus (VSV) infection and pathogenesis; however, it is not known which of the numerous interferon-stimulated genes (ISG) mediate the antiviral effect. A prominent family of ISGs is the interferon-induced with tetratricopeptide repeats (Ifit) genes comprising three members in mice, Ifit1/ISG56, Ifit2/ISG54 and Ifit3/ISG49. Intranasal infection with a low dose of VSV is not lethal to wild-type mice and all three Ifit genes are induced in the central nervous system of the infected mice. We tested their potential contributions to the observed protection of wild-type mice from VSV pathogenesis, by taking advantage of the newly generated knockout mice lacking either Ifit2 or Ifit1. We observed that in Ifit2 knockout (Ifit2 (−/−)) mice, intranasal VSV infection was uniformly lethal and death was preceded by neurological signs, such as ataxia and hind limb paralysis. In contrast, wild-type and Ifit1 (−/−) mice were highly protected and survived without developing such disease. However, when VSV was injected intracranially, virus replication and survival were not significantly different between wild-type and Ifit2(−/−) mice. When administered intranasally, VSV entered the central nervous system through the olfactory bulbs, where it replicated equivalently in wild-type and Ifit2 (−/−) mice and induced interferon-β. However, as the infection spread to other regions of the brain, VSV titers rose several hundred folds higher in Ifit2 (−/−) mice as compared to wild-type mice. This was not caused by a broadened cell tropism in the brains of Ifit2 (−/−) mice, where VSV still replicated selectively in neurons. Surprisingly, this advantage for VSV replication in the brains of Ifit2(−/−) mice was not observed in other organs, such as lung and liver. Pathogenesis by another neurotropic RNA virus, encephalomyocarditis virus, was not enhanced in the brains of Ifit2 (−/−) mice. Our study provides a clear demonstration of tissue-, virus- and ISG-specific antiviral action of interferon.

Virus infection of mammals induces the synthesis of type I interferons (IFN), which, in turn, inhibit virus replication. The high susceptibility of type I IFN receptor knockout (IFNAR 2/2 ) mice to infection by a variety of viruses [1] [2] [3] provides strong evidence for the major role of the IFN system in protecting from viral pathogenesis. In these mice, although IFN is induced by virus infection, it cannot act on target cells. Similarly, in genetically altered mice that are defective in IFN production due to the absence of specific pathogen-associated pattern recognition receptors, signaling proteins or specific transcription factors, viral pathogenesis is enhanced [4] [5] [6] . Although the critical importance of the IFN system in regulating viral pathogenesis is now well established, in many cases it is still unclear how IFN inhibits the replication and spread of a specific virus in vivo. In this context, activation of different components of the immune system plays a major role in controlling viral diseases that are relatively slow to develop [7] [8] [9] . In contrast, in acute infection by viruses that cause severe pathogenesis and death within a few days after infection, protection is primarily provided by the intrinsic antiviral actions of IFN-induced proteins encoded by the hundreds of IFN-stimulated genes (ISGs) [10] [11] [12] , several of which often contribute to the overall effect of IFN against a given virus. Our knowledge of the antiviral and the biochemical properties of individual ISG products is mostly limited to a few intensively studied examples such as PKR, OAS/RNase L or Mx [13] . However, recent systematic investigation of the antiviral functions of the entire family of ISGs has started producing exciting new information [14] . In the above context, we have been investigating the biochemical and biological functions of the members of the Ifit family of ISGs, which are very strongly induced by IFN. There are three members of this family of genes in mice: Ifit1/ISG56, Ifit2/ ISG54 and Ifit3/ISG49; all of the encoded proteins contain multiple tetratricopeptide repeats (TPR), which mediate proteinprotein and protein-RNA interactions [15] . In vitro, P56 and P54, the products of Ifit1 and Ifit2, respectively, bind to the translation initiation factor eIF3 and inhibit protein synthesis [16] . The third member, P49, the product of Ifit3, does not share this property [17] . Recently, it has been reported that Ifit proteins form a multiprotein complex that can bind to the triphosphorylated 59 end of RNAs, an RNA-species produced during the replication of some, but not all, viruses [18] . In vivo, these genes are strongly induced in brains of mice infected with West Nile virus (WNV) or Lymphocytic choriomeningitis virus (LCMV); surprisingly, different Ifit genes are differentially induced in different regions of the brain, suggesting non-redundant functions [19] . To further explore the antiviral properties of the Ifit proteins, we generated Ifit1 knockout (Ifit1 2/2 ) mice and challenged them with different viruses. We observed that Ifit1 2/2 mice were particularly susceptible to a WNV mutant that is defective in its mRNA cap 29-O methylation; the mutant virus killed Ifit1 2/2 mice but not the wild-type (wt) mice [20] . Here, we report on the antiviral properties of the newly generated Ifit2 2/2 mice; these mice, but not Ifit1 2/2 mice, were highly susceptible to neuropathogenesis after intranasal infection with vesicular stomatitis virus (VSV), a negative sense, singlestranded RNA rhabdovirus. VSV replication is highly sensitive to the inhibitory action of IFN and is routinely used to assay the antiviral activity of IFN in vitro [21] . As expected, IFNAR 2/2 mice are highly susceptible to VSV pathogenesis and the same is true for mice that specifically lack expression of IFNAR on the cells of their central nervous system (CNS) [1] . In spite of these observations, little is known about how IFN inhibits VSV replication in vivo. Our new results indicate that in the brain, but not in other organs, Ifit2 is a major mediator of IFN's protective effect against VSV. In contrast, Ifit2 could not protect mice from neuropathogenesis caused by encephalomyocarditis virus (EMCV), a picornavirus. Thus, we have uncovered a virusspecific, tissue-specific and ISG-specific antiviral effect of the IFN system. Generation of Ifit2/ISG54 and Ifit1/ISG56 knockout mice Ifit2 gene knockout (Ifit2 2/2 ) mice were generated by deleting the entire protein-encoding region of the gene, which was achieved by flanking exons 2 and 3 with frt recombinase sites in C57BL/6 embryonic stem cells and excising the flanked region with Flp recombinase ( Figure 1A ). Ifit2 2/2 mice were bred to homozygosity ( Figure 1B) , and deficiency for induced expression of Ifit2 protein was confirmed in lysates of IFN-b-treated primary murine embryonic fibroblasts (MEF) ( Figure 1C ). Mice deficient for Ifit1 (Ifit1 2/2 ) were derived from C57BL/6 embryonic stem cells lacking the entire Ifit1 coding region ( Figure 1A ). Genotypic homozygosity of the Ifit1 2/2 mice and deficiency for Ifit1 protein induction were confirmed ( Figure 1B and 1C ). Both knockout mouse lines were healthy and fertile. Moreover, deletion of one gene within the Ifit locus did not alter the pattern of induction of other adjacent gene family members, as compared to wild-type (wt) mice ( Figure 1C ). To determine the impact of Ifit2 on the outcome of viral infections in vivo, we compared susceptibilities of Ifit2 2/2 and wt mice to VSV infection, using IFNAR 2/2 mice as positive controls of enhanced susceptibility. Virus was administered at a low dose [4610 2 plaque forming units (pfu)], intranasally, reflecting a natural route of infection for VSV [22] . As seen previously, 100% of IFNAR 2/2 mice rapidly succumbed to VSV infection within 2 days ( Figure 2A , and [1] ), after suffering symptoms of lethargy. On the other hand, 79% of wt mice survived, the remaining 21% succumbed to VSV, and this occurred later, at 7-10 days post infection (d.p.i.). In contrast, 100% of Ifit2 2/2 mice died by 7 d.p.i. (Figure 2A ), with most succumbing by 6 d.p.i.; thus, we observed uniform and more rapidly occurring death of Ifit2 2/2 compared to wt mice after VSV infection. Within 24 h before death, both wt and Ifit2 2/2 mice developed neurological signs including ataxia, hind limb paralysis, and hyper-excitability. Ifit2 +/2 mice displayed an intermediate survival curve, demonstrating a gene dosage effect ( Figure 2B ). Next, the role of a related gene, Ifit1, in VSV pathogenesis was evaluated by infecting Ifit1 2/2 mice. Unlike the results observed with Ifit2 2/2 mice, no statistically significant increase in mortality was observed in Ifit1 2/2 mice ( Figure 2B , 21% death for wt versus 36% for Ifit1 2/2 , respectively; p.0.25). Consistent with this, survival kinetics of Ifit1 2/2 and wt mice were similar. Increasing the virus dose by 10,000-fold (to 4610 6 pfu) did not appreciably change the survival curves of wt, Ifit1 2/2 , or Ifit2 2/2 mice ( Figure 2C ). These results demonstrate functional differences between the two closely related proteins encoded by Ifit1 and Ifit2. The virus-specificity of the antiviral action of Ifit2 was evaluated by infecting Ifit2 2/2 mice with EMCV, an unrelated neurovirulent positive-strand RNA virus of the picornavirus family ( Figure 2D ). IFNAR 2/2 mice were highly susceptible to EMCV infection with all mice succumbing within 2 d.p.i.; in contrast, wt mice died with a slower kinetics and at a rate of only 80%. Notably, Ifit2 2/2 mice behaved similarly to the wt mice, without enhanced or accelerated mortality ( Figure 2D ). The same conclusion was true for a lower dose of EMCV ( Figure S1 ). The survival pattern of EMCV-infected Ifit1 2/2 mice also was similar to that of the wt mice ( Figure 2D ). Mice of all genotypes either succumbed after developing neurological symptoms, mainly hind limb paralysis, or survived without symptoms. These results demonstrate that the antiviral action of Ifit2 is both virus-and Ifitspecific. The uniform penetrance of neuropathogenesis and lethality of VSV-infected Ifit2 2/2 mice, even at a low virus dose, prompted us to examine viral spread along its route from the nasal cavity into the CNS ( Figure 3A ). After intranasal administration, VSV infects In mammals, the first line of defense against virus infection is the interferon system. Viruses induce synthesis of interferon in the infected cells and its secretion to circulation. Interferon acts upon the as yet uninfected cells and protects them from oncoming infection by inducing the synthesis of hundreds of new proteins, many of which interfere with virus replication. Vesicular stomatitis virus (VSV), a virus similar to rabies virus, is very sensitive to interferon but it is not known which interferon-induced protein inhibits its replication. Here, we have identified a single interferon-induced protein as the protector of mice from death by VSV infection. Knocking out the gene encoding this protein, Ifit2, made mice very vulnerable to neuropathogenesis caused by VSV infection; a related protein, Ifit1, did not share this property. Moreover, Ifit2 failed to protect mice from another neurotropic virus, encephalomyocarditis virus, nor was it necessary for protecting organs other than brain from infection by VSV. Our observation that a single IFN-induced protein protects a specific organ from infection by a specific virus revealed an unexpected degree of specificity of the antiviral action of IFN. shared wt mice (n = number of animals used). C, survival of Ifit2 2/2 , Ifit1 2/2 and wt mice after intranasal infection with a higher dose of VSV (4610 6 pfu). D, survival of Ifit2 2/2 , Ifit1 2/2 , IFNAR 2/2 and wt mice after infection with 5610 2 pfu of EMCV. In A-D, data are cumulative from at least two independent experiments (exceptions: Figure 2B , Ifit2 +/2 mice and Figure 2D , Ifit1 2/2 mice infected in a single experiment). Statistical significance of survival differences relative to wt mice is indicated by p-values; n.s., not significant; i.n., intranasal. doi:10.1371/journal.ppat.1002712.g002 the nasal epithelia including olfactory sensor neurons, which project to the outer layer of the olfactory bulbs (OB) [23] . This represents the entry step into the CNS, which we examined by immunostaining of OB sections. In wt mice, VSV P protein was detected exclusively within the glomeruli of the OB at 2 d.p.i. ( Figure 3B , upper right panel and [1] ), whereas in IFNAR 2/2 mice, VSV antigen had spread into deeper layers of the OB ( Figure 3B , lower left panel). In Ifit2 2/2 mice OB, viral antigen was restricted to the glomeruli, as seen in wt mice ( Figure 3B , lower right panel). This similar pattern of viral antigen expression between wt and Ifit2 2/2 mice was reflected in the equivalent levels of viral RNA in OB at 2 d.p.i. ( Figure 3C ). In contrast, ,10 times more VSV RNA was present in OB of IFNAR 2/2 mice ( Figure 3C , right panel, p,0.05). A comparison of the infectious viral burden between wt and Ifit2 2/2 mice in the OB confirmed these findings: at 2 d.p.i., ,10 6 pfu/g of VSV was present in both wt and Ifit2 2/2 mice ( Figure 3D , p = 1.0). However, later in the course of infection, by day 6, viral OB titers in Ifit2 2/2 mice were not significantly changed, whereas in wt mice average titers of infectious VSV as well as viral RNA levels had decreased by ,10-fold ( Figure 3C and D, both p,0.05). These results suggest that VSV initially enters and replicates with similar efficiency in both wt and Ifit2 2/2 OB before spreading into the rest of the brain. The efficiency of VSV replication in the brain, excluding the OB, was examined by quantifying infectious VSV as well as viral RNA. Early after infection, at 2 d.p.i., virus titers in brains were low (,10 4 to 10 5 pfu/g) and roughly equivalent in wt and Ifit2 2/2 mice ( Figure 4A , p.0.25). Similarly, viral RNA levels at the same Figure 3 . Ifit2 does not inhibit VSV entry and replication in olfactory bulbs. A, schematic entry route of VSV into the central nervous system of wt mice after intranasal infection, and VSV spread within brain, as reported in the literature. OB, olfactory bulbs; CX, cortex; MB, midbrain; CB, cerebellum; BS, brain stem; SC, spinal cord. B, VSV P protein in OB of VSV-infected wt, Ifit2 2/2 and IFNAR 2/2 mice at 2 d.p.i., detected by immunohistofluorescence. C, VSV RNA levels in OB of uninfected or VSV-infected wt, Ifit2 2/2 and IFNAR 2/2 mice at 1, 2 or 6 d.p.i., plotted as mean+SD on log scale; ND, none detected. D, infectious VSV titers in wt and Ifit2 2/2 OB at 2 and 6 d.p.i.; plotted as pfu/g with mean on log scale; dashed line depicts threshold of detection. In C and D, n = 4-8 mice per infected group accumulated from three independent experiments; in B, n = 2 mice from two independent experiments. All infections were 4610 2 pfu of VSV administered intranasally. Asterisks indicate statistical significance: ** p = 0.006, * p,0.05; n.s.: not significant. doi:10.1371/journal.ppat.1002712.g003 time were low and comparable between wt and Ifit2 2/2 ( Figure 4B , p.0.25). However, at the same time, levels of VSV RNA (380-fold, p,0.05) were much higher in the brains of IFNAR 2/2 mice ( Figure 4B , right panel). Later in the course of infection (6 d.p.i.), brains of wt mice accumulated only ,5-fold more infectious VSV, with occasional clearance of the virus. In contrast, we detected markedly higher VSV titers in the brains of Ifit2 2/2 mice (,350-fold higher compared to wt mice, p = 0.0009), reaching ,10 8 pfu/g ( Figure 4A ); the high virus load likely caused the pronounced lethality. Differences in viral RNA levels in brains of wt and Ifit2 2/2 mice at 6 d.p.i. correlated well with levels of infectious VSV ( Figure 4B ). To determine whether Ifit2 selectively restricts replication of VSV in particular regions of the brain, we measured viral RNA levels in cortex, midbrain, cerebellum and brain stem at 6 d.p.i. In wt mice, VSV RNA was present prominently in the cortex, midbrain and brainstem, but not in the cerebellum ( Figure 4C ), which is consistent with published results [24] . However, in Ifit2 2/2 mice, viral RNA was 200-fold or more (p,0.05) abundant in all regions of the brain examined, including the cerebellum. The increase of VSV replication in Ifit2 2/2 brains was not due to a broadened cell tropism of the virus; immunostaining for viral P protein showed exclusive localization to neurons and not other cell types, such as astrocytes ( Figure 4D ). From the above observations, we conclude that after intranasal infection by VSV, Ifit2 protects mice from neuropathogenesis by suppressing replication or spread of the virus in brain neurons. Ifit2 and Ifit1 are induced in VSV-infected regions of OB and brain The protective effect of type I IFN signaling and in particular, Ifit2, against VSV neuropathogenesis prompted us to confirm its expression in OB and brain of wt mice, and whether it was induced in a type I IFN-dependent manner. In wt OB, Ifit2, Ifit1, and IFN-b mRNA was induced strongly by 2 d.p.i., and Ifit2 and Ifit1 RNA remained abundant until day 6 d.p.i. (Figure 5A ). The induction of these genes was dependent on type I IFN receptor in OB as well as in brain ( Figure 5B and 5E, and data not shown). Ifit2 suppresses VSV replication in the brain after intranasal infection. A, infectious VSV titers in wt and Ifit2 2/2 brains at 2 and 6 days after intranasal infection, plotted as pfu/g with mean on log scale; dashed line depicts threshold of detection. B, VSV RNA levels in brains of uninfected or VSV-infected wt, Ifit2 2/2 and IFNAR 2/2 mice at 2 or 6 d.p.i., plotted as mean+SD on log scale. C, VSV RNA levels in different regions of the brains of uninfected or VSV-infected wt and Ifit2 2/2 mice at 6 d.p.i., plotted as mean+SD on log scale. D, VSV P protein in midbrain neurons of Ifit2 2/2 mice at 6 d.p.i.; detection by immunohistofluorescence-labeling of VSV-P (red) and neuron (NeuN) or astrocyte (GFAP) markers (green); in A and B: n = 4-8 mice per infected group accumulated from three independent experiments; in C: n = 4 mice per infected group; in D: n = 2 mice per infected group; all infections in A-D were intranasal with 4610 2 pfu of VSV. ND, none detected. Brains in A and B were separated from OBs assayed in Figure 3D and 3C, respectively. Asterisks indicate statistical significance: *** p#0.0009; n.s.: not significant. doi:10.1371/journal.ppat.1002712.g004 Furthermore, expression of Ifit2 mRNA in wt OB coincided with the presence of detectable levels of the encoded Ifit2 protein ( = P54) at 2 d.p.i. and 6 d.p.i., as seen by immunohistochemistry ( Figure 5C , and data not shown). Ifit2 protein staining was observed in VSV-infected cells within OB glomeruli as well as in surrounding and distant viral antigen-free cells, consistent with a remote IFN-dependent induction of Ifit2 expression ( Figure 5C , arrowheads in magnified images of right panel). Ifit1 and IFN-b mRNAs were induced as strongly in OB of Ifit22/2 as in wt mice, which correlated well with similar abundance of VSV RNA in wt and Ifit2 2/2 OB (Figure 5A compared to Figure 3C ). In brains, at 6 d.p.i., in contrast to OB, induction of Ifit1 and IFN-b mRNAs was considerably stronger in Ifit2 2/2 mice compared to wt mice ( Figure 5D , 5-fold and 27-fold, respectively, both p,0.005). The enhanced gene induction in VSV-infected Ifit2 2/2 mice was not restricted to specific regions of the brain ( Figure S2 ). Enhanced cellular gene expression also was observed for several virusinduced cytokine and chemokine genes, as measured by quantitative RT-PCR ( Figure S3A ). Gene expression profiling of brain tissue at day 6 d.p.i., using microarray analysis, revealed that many other genes, including ISGs, were also more strongly induced (Table S1 ). These results demonstrated that enhanced virus replication in the brains of Ifit2 2/2 mice led to enhanced type I IFN, other cytokines and ISG induction, which nevertheless failed to restrict VSV replication in the absence of Ifit2. Wt mice are as susceptible as Ifit2 2/2 mice to intracranial VSV infection Our results from intranasal VSV infection indicated that Ifit2 induction in the brain was mediated by type I IFN that was, in all likelihood, produced by infected cells in the OB ( Figure 5A ). Virus replication and resultant IFN induction at 2 d.p.i. were similar in the OBs of wt and Ifit2 2/2 mice (Figs. 3C, 3D and 5A); presumably, the newly produced IFN diffused into the rest of the brain and induced local Ifit2 expression in the wt mouse brains, prior to the arrival of the infectious virus. If this were the case, one would anticipate that direct infection of the brain, without prior action of IFN produced in infected OB, would minimize the difference between the phenotypes of wt and Ifit2 2/2 mice. To test this idea, we injected a very low dose (10 pfu) of VSV intracranially. As hypothesized, wt and Ifit2 2/2 mice were now equally susceptible; almost all mice died by 3 d.p.i. even at this low dose ( Figure 6A ) and there were equally high virus titers and viral RNA levels in the brains of mice of both genotypes ( Figure 6B and 6C). Concomitant with virus replication, there was similar induction of Ifit1 and IFN-b ( Figure 6C ) and other cytokines and chemokines ( Figure S3B ). These results indicate that in the absence of prior induction of Ifit2 by IFN, brain neurons are highly susceptible to VSV infection. Unlike the brain, other organs of Ifit2 2/2 mice are not more susceptible to intranasal VSV infection IFNAR 2/2 mice succumbed within two days after VSV infection without accumulating very high VSV RNA levels in the brain ( Figure 4B ). These mice did not develop CNS-related signs of disease, but showed severe lethargy before death, suggesting that death was due to efficient replication of the virus in peripheral organs, due to the absence of an otherwise effective type I IFN-mediated antiviral protection of the same organs in wt mice. To test this, we assessed the kinetics of VSV accumulation in brains, livers and lungs of wt, IFNAR 2/2 and Ifit2 2/2 mice (Figure 7) . At 2 d.p.i., VSV titers were very high in the liver of IFNAR 2/2 mice, reaching 10 9 pfu/g ( Figure 7A ). In contrast, no or little infectious virus was detected in the liver of wt mice at 2 or 6 d.p.i., indicating efficient IFN-dependent suppression of VSV replication; intriguingly, this was also observed in Ifit2 2/2 mice, demonstrating that Ifit2 did not mediate the anti-VSV effects of type I IFN in the liver. In lungs, which directly received a part of the virus inoculum from intranasal inhalation of VSV, the virus also replicated efficiently in IFNAR 2/2 mice, reaching 10 8 pfu/g before death ( Figure 7B ). In comparison, lungs of wt and Ifit2 2/2 mice exhibited much lower levels of VSV at 2 and 4 d.p.i. (3,000 to 10,000-fold lower for wt and Ifit2 2/2 compared to IFNAR 2/2 mice, all p,0.05). By days 5 and 6 d.p.i., the virus was cleared from the lungs of a subset of wt and Ifit2 2/2 mice. In contrast, in brains from the same animals, 10 to 100-fold higher average titers (p,0.05) of VSV accumulated in Ifit2 2/2 compared to wt mice at all time points between 2 and 6 d.p.i. ( Figure 7C ). As expected, in wt mice, both Ifit1 and Ifit2 were induced not only in brains ( Figure 5D ), but also in livers ( Figure 7D ) and lungs ( Figure 7E ); IFN-b was also induced in lungs, but not livers. Ifit1, Ifit2 and IFN-b mRNAs were also induced in the brains of EMCV-infected wt mice ( Figure S3C ). These findings demonstrate an unexpected brain-restricted and virus-restricted function of Ifit2 in the context of the type I IFN-mediated antiviral response to VSV infection. They also indicate that in Ifit2 2/2 mice, other ISGs, which presumably protect the peripheral organs of VSV-infected wt mice, are either not induced in neurons or insufficient to protect them. IFNs are defined by their antiviral activities. They inhibit the replication of many, if not all, viruses mostly by direct inhibition of replication in the infected cells but also by promoting the ability of immune cells to recognize and eliminate the virus-infected cells [25] . The direct effects are mediated by ISGs, which number in the hundreds, and different ISGs are thought to have more potent antiviral activities toward different families of viruses [13] . However, in most cases, it is not known which ISG inhibits the replication of a given virus; the rare exception is the Mx-mediated inhibition of influenza viruses, the underlying effect which allowed for the discovery of IFNs [26] . The task of connecting a specific IFN-induced protein to a specific antiviral action is compounded by the fact that often several IFN-induced proteins act in concert to inhibit the same virus at different stages of its life cycle. Moreover, a specific IFN-induced protein may be more relevant for inhibiting a virus in one specific cell-type than another. Recent systematic investigation of the specific antiviral effects of different ISGs has started providing significant insight into this problem [14] . Such findings are complemented by the analyses of the spectra of the antiviral effects of a specific ISG or a family of ISGs [27] . We have undertaken an investigation of the Ifit family of mouse ISGs. The corresponding human proteins are known to have antiviral activities against human papillomavirus (HPV) and hepatitis C virus (HCV), neither of which replicate in mouse cells. The anti-HPV activity of human IFIT1 ( = P56) has been attributed to its ability to bind HPV E1 protein and to inhibit its helicase activity, which is essential for HPV DNA replication [28, 29] . The antiviral effect on HCV, on the other hand, is manifested at the level of inhibiting viral protein synthesis as a consequence of the ability of IFIT1 to bind the translation initiation factor eIF3 and inhibit its various actions in translation initiation [30] . It has been reported recently that the IFIT1 protein can form a complex and bind to RNAs with triphosphorylated 59 ends, presumably providing another means to inhibit specific viruses that produce such RNAs [18] . The Ifit genes are clustered at a single locus in both human and mouse. In the latter species, two alleles of Ifit3 genes are flanked on two sides by one allele of Ifit2 and one allele of Ifit1 [15] . To identify their physiological functions, we have separately deleted the entire coding regions of Ifit1 or Ifit2 genes. The Ifit1 2/2 mice exhibited an interesting phenotype in allowing the replication of and resultant pathogenesis by a WNV mutant, which failed to replicate in wt mice [20] . Because this mutant is defective in 29-O methylation of the cap structure of viral mRNAs, its rescue in the Ifit1 2/2 mouse indicates that this antiviral protein recognizes the 59 ends of mRNAs, a conclusion that is consistent with the observation that, in vitro, it can bind to RNAs having specific structures at the 59 ends [18] . It remains to be seen whether the proposed property of Ifit proteins to recognize 59 ends of RNA is connected in any way to their ability to inhibit the functions of eIF3 [16] , which participates in several steps of translation initiation taking place at or near the 59 ends of mRNAs. Replication of VSV is highly sensitive to the antiviral activity of IFNs, and VSV is widely used to determine the specific activities of IFN preparations quantitatively [21] . In spite of this strong connection, it is unclear how IFN inhibits VSV replication. An early report indicated that viral primary transcription is inhibited by IFN, but it is not known which IFN-induced protein mediates this inhibition [31] . The observed sensitivity of VSV replication in vitro is reflected in vivo. IFNAR 2/2 mice are extremely susceptible to VSV infection; they rapidly die within 2 days after infection and the virus replicates to very high titers in many organs of the infected mice. The extreme sensitivity of IFNAR 2/2 mice to VSV infection suggests that type I IFN provides the majority, if not all, of the protective innate immune defense. Eventually, protection may be facilitated by immune cell-mediated antiviral actions, but this is a slow process that does not appear to function before 6-10 days post-infection [32, 33] . Thus, it is likely that one or more ISGs directly inhibit VSV replication in vivo. In this context, it has been reported that mice lacking PKR, a well-studied ISG, display higher susceptibility to VSV pathogenesis [34] . However, detailed investigation of the underlying mechanism revealed that PKR did not execute IFN's antiviral action; rather, it was required for efficient induction of IFN-a/b in the infected mice [35] . In vivo VSV-infection induces IFN synthesis in many cell types, using either the cytoplasmic RIG-I pathway or the endosomal TLR7 pathway [4, 36] ; however, it is unclear how PKR aids this process. Our results show that Ifit2 2/2 mice are highly susceptible to intranasal VSV infection and the effect is gene dosage-dependent: Ifit2 +/2 mice had an intermediate susceptibility phenotype. Infected Ifit2 2/2 mice displayed symptoms of severe neuropathogenesis late after VSV infection accompanied by efficient replication of the virus in many regions of the brain. However, virus replication was restricted to neurons and did not spread to other types of cells in the brain, such as astrocytes. Our results are consistent with the hypothesis that prior, IFN-induced, Ifit2 expression in the brain restricts VSV replication. Supporting genetic evidence for the requirement of IFN action is provided by the high susceptibility of the IFNAR 2/2 mice, which possess the functional Ifit2 gene but Ifit2 is not induced by VSV infection because these mice cannot respond to type I IFN. Additional evidence comes from a previous study using brain-specific IFNAR 2/2 mice, which displayed a pattern of susceptibility to intranasal VSV infection similar to that of our Ifit2 2/2 mice [1] . In our experimental system, the source of the IFN production was Figure 5C ). Ifit2 was also induced at this time in the rest of the brain, without any induction of IFN mRNA ( Figure 5D ) suggesting that the source of IFN was the OB. In accord with the well-established concept of IFN action, pre-induction of Ifit2 in neurons, before the onset of infection, was essential for the antiviral effect. In comparison, induction of IFN and Ifit2 that was concomitant with VSV infection failed to have an appreciable antiviral effect, as manifested by robust virus replication at directly infected sites, such as the OBs of wt mice infected intranasally ( Figure 3D ) or the brain of wt mice infected intracranially ( Figure 6B ). High mortality of the infected mice correlated with high virus titers in the brains of intranasally infected Ifit2 2/2 mice or intracranially infected wt and Ifit2 2/2 mice. In the intranasally infected Ifit2 2/2 mice, death was not preceded by widespread apoptosis in the brain ( Figure S4 ). However, as expected with high viral loads, IFN and other cytokines and chemokines were strongly induced ( Figures 5D, S2 and S3A) ; consequently, many ISGs, except Ifit2, were also induced (Table S1) . Pre-induced Ifit2 prevents efficient VSV replication in the brain, most probably by blocking one or more essential step of the viral life cycle including viral entry, uncoating, primary transcription, viral protein synthesis, RNA replication, virion assembly or egress. It also might block trans-synaptic spread of the virus, although unlike another rhabdovirus, rabies virus, VSV is not known to depend on transit from neuron to neuron. In this context, it is important to note the observations made by Iannacone et al. [37] using a footpad VSV infection model. They concluded that type I IFN, produced by infected macrophages and plasmacytoid dendritic cells in infected mice, blocked infection of peripheral neurons resulting in lowered infection of the CNS and prevention of neuropathogenesis. It is worth noting that in our studies, the absence of Ifit2 did not affect IFN induction by VSV (Figures 5A and 6C ). Further investigation of the biochemical mechanism behind the observed in vivo effect of Ifit2 2/2 is hampered by the absence of a suitable cell culture model of the phenomenon. For example, Ifit2 was not required for mediating the anti-VSV effect of IFN in mouse embryonic fibroblasts ( Figure S5 ), in primary fetal neurons or in Ifit2-ablated neuroblastoma cells (data not shown), results that are not surprising given the strong tissuespecificity of Ifit2 action observed in vivo (Figure 7) . Specific RNAbinding properties of Ifit proteins have been recently reported [18] . Following this lead, we examined the RNA-binding properties of recombinant murine Ifit1 and Ifit2 using VSV leader RNA as the probe in an electrophoretic mobility shift assay: Ifit1 bound RNA with a 59-ppp end but not with a 59-OH end; however, Ifit2 bound neither ( Figure S6 ). To obtain meaningful leads, future investigation of this kind may require using brain extracts from infected mice to detect protein-viral RNA complexes that may contain Ifit2 along with adult neuronspecific proteins. Our results revealed several layers of specificity of IFN action, some of which were not anticipated. First, compared to Ifit2 2/2 mice, Ifit1 2/2 mice were much less susceptible to intranasal VSV infection; this was true for both low and high doses of virus. This finding was surprising in view of a recent report on VSV susceptibility of Ifit1 2/2 mice [18] and the observation that Ifit1, but not Ifit2, could bind VSV leader RNA in vitro ( Figure S6) . The above results demonstrate that different Ifit proteins have nonredundant functions in vivo. The second layer of specificity was directed toward the nature of the infecting virus. Although both VSV and EMCV caused neuroinvasive disease, induced IFN-b, Ifit1 and Ifit2 in the brain and type I IFN action was required for protection against both viruses, Ifit2 was critical only for protection against VSV; the absence of either Ifit1 or Ifit2 did not exacerbate susceptibility to EMCV. The third layer of specificity was revealed by the organ-specific action of Ifit2. In the complete absence of type I IFN action in the IFNAR 2/2 mice, intranasally infected VSV replicated vigorously not only in brains, but also in livers and lungs ( Figure 7A-C) . In contrast, in Ifit2 2/2 mice, efficient VSV replication was restricted to the brain suggesting that Ifit2 does not act as an anti-VSV ISG in the liver or the lung because its absence did not impact virus titers, even though Ifit2 was induced in these organs of infected wt mice ( Figure 7D and 7E) . The efficient VSV replication in livers and lungs of IFNAR 2/2 mice, but not wt and Ifit2 2/2 mice, indicates that other ISGs must have anti-VSV effects in those organs. Further investigation is needed to determine the basis of neuronal specificity of Ifit2 action and the identities of other ISGs that inhibit VSV replication in other organs. All animal experiments were performed in strict accordance with all provisions of the Animal Welfare Act, the Guide for the Care and Use of Laboratory Animals, and the PHS Policy on Humane Care and Use of Laboratory Animals. The protocol was approved by the Cleveland Clinic Institutional Animal Care and Use Committee (IACUC), PHS Assurance number A3047-01. All experimental manipulations or intranasal instillations of mice were performed under anesthesia induced by pentobarbital sodium or isofluorane, respectively, and all efforts were made to minimize suffering. All mice used were of C57BL/6 background and of both sexes; Ifit2 2/2 mice were custom-generated by Taconic Farms, Inc. by flanking exons 2 and 3 of Ifit2, encompassing the complete protein-encoding region, with frt sites in C57BL/6 embryonic stem (ES) cells, and deleting the flanked region by transfection of Flp recombinase. ES cell clones were injected into BL/6 blastocysts, and heterozygous offspring mice were crossed to homozygosity. Ifit1 2/2 mice were generated from C57BL/6 ES cells lacking the whole coding region of Ifit1 (20); ES cells were obtained from the NIH Knockout mouse project (KOMP, allele Ifit1 tm1(KOMP)Vlcg ). The same ES cell line was independently used to generate mice in another study [18] . IFNAR 2/2 mice (lacking Ifnar1) were a gift of Murali-Krishna Kaja (Emory University, Atlanta, GA). Congenic wild-type mice were obtained from Taconic Farms. Vesicular stomatitis virus (VSV) Indiana was a gift from Amiya K. Banerjee, Lerner Research Institute, Cleveland, Ohio. For intranasal infections, between 4610 2 and 4610 6 pfu of VSV in 35 ml of endotoxin-free PBS were inhaled by isofluorane-anesthetized 8-12 week-old mice, with PBS-only as control. For intracranial infections, 10 pfu of VSV in 30 ml of endotoxin-free PBS were injected into the brains of 6-7 week-old mice, with PBSonly as control. Thereafter, mice were monitored daily (twice daily after i.c. injection) for weight loss and symptoms of disease. Encephalomyocarditis virus (EMCV) K strain was a gift from Robert H. Silverman, Lerner Research Institute, Cleveland, Ohio. For intraperitoneal infections, between 25 and 5610 2 pfu of EMCV in 500 ml of PBS were injected into the peritoneal cavity of mice. Mice were monitored daily for weight loss and symptoms of disease. Mice were anesthetized with pentobarbital (150 mg/kg) and blood was removed from organs by cardiac perfusion with 10 ml of PBS, followed by perfusion with 10 ml of 4% paraformaldehyde/PBS for fixation. Brains were placed in 4% paraformaldehyde overnight for complete fixation, submerged in 30% sucrose/ PBS overnight for cryoprotection, and frozen in O.C.T. compound (Sakura Finetek USA, Torrance, CA, USA). 10 mm sagittal sections were cut at 220uC in a Leica CM1900 cryostat, mounted on coated slides (Superfrost Plus, Fisherbrand, Fisher Scientific); membranes were permeabilized by 0.2% Triton X-100/PBS treatment for 15 min. For immunohistochemistry, the Envision+ DAB kit (Dako, Carpinteria, CA) was used with antimouse Ifit2/P54 [38] or anti-VSV-P protein (a gift from Amiya K. Banerjee, Lerner Research Institute, Cleveland, Ohio) as primary antibodies. For immunohistofluorescence, anti-VSV-P or anti-NeuN (Chemicon Intl./Millipore, Billerica, MA) or anti-GFAP (Sigma-Aldrich, St. Louis, MO) were used; labeled brain sections were stained with AlexaFluor-594 secondary antibody (Invitrogen/Molecular Probes, Carlsbad, CA). For detection of apoptotic cells in brain sections, the DeadEnd fluorometric TUNEL system (Promega) was used according to manufacturer's instructions. All objects were then mounted with VectaShield (with DAPI, Vector Labs, Burlingame, CA), and examined with a Leica DRM fluorescence microscope. Mice were anesthetized with pentobarbital (150 mg/kg) and blood was removed from organs after cardiac perfusion with 10 ml of PBS. Brains were separated into olfactory bulbs and the remainder of the brain, snap-frozen in liquid nitrogen (as well as livers and lungs) and RNA was extracted using TRIzol reagent (Invitrogen). DNase I treatment (DNAfree, Applied Biosystems/ Ambion) and reverse transcription with random hexamers (ImProm-II, Promega) were performed according to manufacturer's instructions. 0.5 ng of RNA was used in 384 well-format realtime PCRs in a Roche LightCycler 480 II using Applied Biosystem's SYBR Green PCR core reagents. PCR primers for murine ISG49/Ifit3, ISG54/Ifit2, ISG56/Ifit1 and 18S rRNA have been published previously [17] ; primers targeting murine Ifnb1 [59-CTTCTCCGTCATCTCCATAGGG-39 [39] , with the alternative reverse primer: 59-CACAGCCCTCTCCATCAACT-39], VSV N RNA [40] or EMCV 3D polymerase genomic region [41] were described previously. Primers for Ccl2, Il1b, Il6, Tnf, Il12b and Nos2 have been described previously [42, 43] . Average expression levels, relative to 18S rRNA and normalized by use of calibrator samples, were graphed with Prism 5.02 software. For analysis of different regions of the brain, brains without OB of perfused mice were separated into cortex, cerebellum, brain stem and remaining ''midbrain'', and tissue was submerged into RNAlater stabilizing reagent (Qiagen) overnight and frozen. RNA was then extracted via TRIzol and further processed and assayed by realtime RT-PCR as described above. For microarray analysis, TRIzol-extracted and DNase I-treated RNA was additionally purified using spin columns (RNeasy Mini kit, Qiagen) before subjection to mRNA expression microarray analysis via Illumina Mouse Ref-8 V2 beadchip and GenomeStudio software V2010.2 (Illumina, Inc.); RNA hybridization to chips was performed by the Lerner Research Institute Genomics Core at the Cleveland Clinic. Microarray raw data were deposited in the NCBI Gene Expression Omnibus (GEO), accession number GSE33678. For quantification of infectious VSV in organs, mice were anesthetized with pentobarbital (150 mg/kg) and blood was removed from organs by cardiac perfusion with 10 ml of PBS. Organs were snap-frozen in liquid nitrogen, weighed, pestle/tubehomogenized (Kimble/Kontes) in 1 ml of PBS per brain or peripheral organ or 0.1 ml per pair of olfactory bulbs, and virus was titered in 10-fold serial dilutions on Vero cells by plaque assay. Results are expressed as plaque-forming units (pfu) per gram of tissue. For quantification of infectious VSV yields in MEF, cells (2/+IFN-b pretreatment as indicated) were infected with VSV inoculum for 1 h, and after another 12 h, cells were freeze/ thawed, and cleared supernatants of lysates were assayed for VSV by plaque assay on Vero cells. Primary murine embryonic fibroblasts (MEFs) were stimulated with 1000 U/ml murine IFN-b (PBL, Inc., Piscataway, NJ) for 16 h and lysed in lysis buffer [50 mM Tris pH 7.6, 150 mM NaCl, 0.5% Triton X-100, 1 mM sodium orthovanadate, 10 mM sodium fluoride, 5 mM sodium pyrophosphate, 10 mM bglycerophosphate and 16 complete EDTA-free protease inhibitor (Roche, Indianapolis, IN)]. 10 mg of whole cell extract were separated via 10% SDS-PAGE, transferred to PVDF membranes, blocked with 5% dry milk in Tris-buffered saline/0.05% Tween-20 overnight and labeled with anti-Ifit3/P49, anti-Ifit2/P54 or anti-Ifit1/P56 polyclonal rabbit sera [17, 38] . Single-stranded VSV leader RNA (nucleotides 1-18) was T7 polymerase-transcribed in presence of [a-32 P]-CTP, yielding radiolabeled 59-triphosphorylated (ppp-) RNA, followed by alkaline phosphatase treatment for generation of 59-hydroxyl (HO-) RNA. ppp-RNA or HO-RNA were added to bacterially expressed and purified 6xHis-tagged Ifit1 or Ifit2 protein in reaction buffer (50 mM Tris pH 8.0, 100 mM NaCl, 1 mM EDTA, 2 mM DTT, 0.05% Triton X-100, 10% glycerol) and incubated for 30 min on ice. Reaction products were separated by 6% native polyacrylamide gel electrophoresis followed by exposure to film. Statistical significance of mouse survival differences was calculated by Mantel-Cox log rank test. To assess significance of differences in gene expressions or virus titers, the two-tailed Mann-Whitney test was used. All calculations were performed using GraphPad Prism 5.02 software. Previously published transcript sequences in the NCBI Entrez Nucleotide database: Ifit2, NM_008332; Ifit1, NM_008331; Ifit3, NM_010501; Ifnb1, NM_010510; Ifnar1, NM_010508. Figure S1 Survival of wt and Ifit2 2/2 mice after infection with low EMCV dose (25 pfu). Statistical significance of survival differences is indicated by p-value; n.s., not significant. (PDF) Figure S2 Enhanced ISG and IFN-b induction in intranasally VSV-infected Ifit2 2/2 brain regions. IFN-b-, and Ifit3/2/1 mRNA levels in different regions of brains of uninfected or VSV-infected wt and Ifit2 2/2 mice at 6 d.p.i., plotted as mean+SD. n = 4 mice per infected group; ND, not done. Infections were intranasal with 4610 2 pfu of VSV. (PDF) Figure S3 Gene induction in brains after VSV or EMCV infections. A, mRNA levels of select genes in brains (without OBs) of uninfected or intranasally VSV-infected wt and Ifit2 2/2 mice at 6 d.p.i., plotted as mean+SD; n = 3 mice per infected group; infection was intranasal with 4610 2 pfu of VSV. B, mRNA levels of select genes in brains (without OBs) of uninfected or intracranially VSV-infected wt and Ifit2 2/2 mice at 24 h post injection, plotted as mean+SD; n = 4 mice per infected group; infection was intracranial injection with 10 pfu of VSV. C, Ifit2, Ifit1, IFN-b and EMCV RNA levels in brains 4 days after EMCV infection (5610 2 pfu, n = 3 mice per infected group). (PDF) Figure S4 Region-selective induction of apoptosis in brains of intranasally VSV-infected Ifit2 2/2 mice. Ifit2 2/2 mice were i.n. infected with 4610 2 pfu of VSV; at 6 d.p.i., adjacent sections of fixed brains were labeled to detect apoptotic cells (TUNEL) or VSV P protein (immunohistofluorescence), n = 2 mice; only few regions such as striatum show positive TUNEL; infected wt brains and uninfected control brains of either genotype did not show appreciable signals, hence data not shown). (PDF) Figure S5 VSV yields from infected wt and Ifit2 2/2 MEF. Immortalized MEF were treated for 16 h with 10 U/ml IFN-b and infected with VSV at moi 10. 12 hours after infection, virus yields were determined by plaque assay. Results are plotted as mean+SD on log scale, representing one of two independent experiments. (PDF) Figure S6 Murine Ifit2 protein does not bind ppp-RNA. Single-stranded radiolabeled VSV leader RNAs (nt 1-18) with either 59-triphosphorylated or free 59-hydroxyl-ends (ppp-RNA or HO-RNA) were in vitro incubated with purified murine Ifit1 ( = P56) or Ifit2 ( = P54) proteins; formation of protein/RNA complex was detected by electrophoretic mobility shift assay. (PDF) Table S1 Enhanced gene expression in brains incl. OBs of intranasally VSV-infected Ifit2 2/2 versus wt mice at 6 d.p.i. Wt or Ifit2 2/2 mice were intranasally VSV-infected with 4610 2 pfu, and at 2 or 6 d.p.i., brain (incl. OB) RNA expression profiles were obtained by microarray. Genes are ranked by their ''fold expression level in Ifit2 2/2 over wt at 6 d.p.i.''. Only genes with at least 3-fold higher expression level in Ifit2 2/2 are included. Note: The Ifit1/ISG56 probe of the Illumina mouse Ref-8 chip is defective and therefore the gene is not included in this list. (PDF)

Immunogenetic Mechanisms Driving Norovirus GII.4 Antigenic Variation

Noroviruses are the principal cause of epidemic gastroenteritis worldwide with GII.4 strains accounting for 80% of infections. The major capsid protein of GII.4 strains is evolving rapidly, resulting in new epidemic strains with altered antigenic potentials. To test if antigenic drift may contribute to GII.4 persistence, human memory B cells were immortalized and the resulting human monoclonal antibodies (mAbs) characterized for reactivity to a panel of time-ordered GII.4 virus-like particles (VLPs). Reflecting the complex exposure history of the volunteer, human anti-GII.4 mAbs grouped into three VLP reactivity patterns; ancestral (1987–1997), contemporary (2004–2009), and broad (1987–2009). NVB 114 reacted exclusively to the earliest GII.4 VLPs by EIA and blockade. NVB 97 specifically bound and blocked only contemporary GII.4 VLPs, while NBV 111 and 43.9 exclusively reacted with and blocked variants of the GII.4.2006 Minerva strain. Three mAbs had broad GII.4 reactivity. Two, NVB 37.10 and 61.3, also detected other genogroup II VLPs by EIA but did not block any VLP interactions with carbohydrate ligands. NVB 71.4 cross-neutralized the panel of time-ordered GII.4 VLPs, as measured by VLP-carbohydrate blockade assays. Using mutant VLPs designed to alter predicted antigenic epitopes, two evolving, GII.4-specific, blockade epitopes were mapped. Amino acids 294–298 and 368–372 were required for binding NVB 114, 111 and 43.9 mAbs. Amino acids 393–395 were essential for binding NVB 97, supporting earlier correlations between antibody blockade escape and carbohydrate binding variation. These data inform VLP vaccine design, provide a strategy for expanding the cross-blockade potential of chimeric VLP vaccines, and identify an antibody with broadly neutralizing therapeutic potential for the treatment of human disease. Moreover, these data support the hypothesis that GII.4 norovirus evolution is heavily influenced by antigenic variation of neutralizing epitopes and consequently, antibody-driven receptor switching; thus, protective herd immunity is a driving force in norovirus molecular evolution.

Noroviruses (NoVs) are the leading cause of severe viral gastroenteritis and are responsible for 50% of all acute gastroenteritis outbreaks in the United States and Europe [1] . Although the severity of disease is usually moderate, lasting 1-3 days, infection can be especially virulent in young children, the elderly, and the immunocompromised, with the latter group experiencing chronic diarrhea and virus shedding for over a year [2] [3] [4] [5] [6] [7] [8] . Importantly, it is estimated that 200,000 people die each year from norovirus infections, primarily children in the developing world [9] . An effective vaccine would be particularly advantageous for the very young and aged populations, military personnel, children and healthcare providers, food handlers, cruise ship passengers, and populations of the developing world [10] . Immunotherapeutics are especially needed for treating immunosuppressed populations experiencing long-term infections with chronic diarrhea. The lack of understanding of the extensive antigenic relationships among the large number of norovirus strains and the complex relationship between host protective immunity and virus antigenic heterogeneity are the primary obstacles to norovirus vaccine development. Noroviruses are ,38 nm icosahedral viruses with a ,7.5 kb single-stranded, positive-sense RNA genome that contains three large open reading frames (ORFs). ORF1 encodes the non-structural proteins, while ORFs 2 and 3 encode the major and minor capsid proteins respectively. Expression of the major capsid protein (ORF2) in Venezuelan equine encephalitis (VEE) virus or baculovirus results in the formation of virus-like particles (VLPs) composed of 90 copies of the major capsid protein dimer [11] . Noroviruses are grouped by the amino acid sequence of the major capsid protein: viruses with less than 14.3% difference are classified as the same strain, 14.3-43.8% difference as the same genotype, and 45-61.4% difference as the same genogroup [12] . Currently, noroviruses are grouped into five genogroups (GI-GV). Genogroups GI and GII are responsible for most human infections and are further subdivided into 8 and 21 different genotypes, respectively [1, 12] . Structurally, the capsid monomer is divided into three domains. The shell domain (S) forms the core of the particle and the protruding domain (P) extends away from the core. The P domain is further subdivided into the P1 subdomain (residues 226-278 and 406-520) and the P2 subdomain (residues 279-405) [11] . The P2 subdomain is the most exposed region of the viral particle and is well positioned to interact with potential neutralizing antibodies and histoblood group antigen (HBGA) ligands [13] [14] [15] [16] [17] . Previous studies have shown that the P2 subdomain of the major capsid protein of GII.4 strains is evolving rapidly, resulting in new epidemic strains with altered carbohydrate ligand binding properties and antigenicity [13, [18] [19] [20] [21] [22] [23] . For the past two decades, the majority of norovirus outbreaks have been caused by strains within the genogroup II, genotype 4 (GII.4 strains) subcluster. Between 1995 and 2006, four major norovirus pandemics associated with GII.4 strains were characterized using molecular epidemiologic methods. During the mid-1990's [24] strain US95/96 was responsible for ,55% of the norovirus outbreaks in the USA and 85% of the outbreaks in the Netherlands [25] . In 2002, the US95/96 strain was replaced by the Farmington Hills strain [26] , which was associated with ,80% of norovirus outbreaks [27] in the USA. In 2004, the Hunter GII.4 variant was detected in Australia, Europe, and Asia [28] [29] [30] . Hunter strains were largely replaced in 2006 by two new cocirculating GII.4 variants in the USA and Europe, Laurens (2006a) and Minerva (2006b) [5, 29, 31] . Although similar to Minerva, Apeldoorn317 (GII.4.2008, GenBank accession no. AB445395) represents a new evolutionary cluster in the phylogeny of the GII.4 viruses. Most recently, a new GII. 4.2006 4. variant, GII.4.2009 New Orleans, has been the predominate outbreak strain, although GII.4.2006 Minerva continues to circulate at low levels [1, 32] . A variety of studies using time ordered human outbreak sera and mouse monoclonal antibodies support the hypothesis that the GII.4 noroviruses are undergoing antigenic variation and that this variation contributes to the emergence of new outbreak strains over time [13, 22, 33, 34] . However, the lack of a cell culture or small animal model for human norovirus cultivation restricts study of neutralization antibodies and epitopes. To circumvent this problem, highly informative in vitro assays have been developed that measure the ability of an antibody to ''block'' binding of a VLP to a carbohydrate ligand [13, 35, 36] . This assay is highly sensitive, as it differentiates between norovirus strains too similar to be distinguished by enzyme immunoassay (EIA). The clinical relevance of the blockade assay, as a surrogate neutralization assay, has been confirmed in both infected chimpanzees [37] and Norwalk virus-infected humans [10, 38] . For NoV strains that hemagglutinate RBCs, high blockade antibody titers that prevent hemagglutination also correlate with protection from disease in humans [39] . Using human norovirus outbreak sera, VLPimmunized mouse sera, and mouse mAbs [13, 33, 34] , the early GII. 4 strains (1987 and 1997) were antigenically indistinguishable from each other by EIA and surrogate neutralization assays. VLPs of strains circulating post 2002 had significantly less reactivity with sera directed against earlier strains and no reactivity to mouse mAbs directed to GII.4.1987 . Conversely, select mouse mAbs generated against GII.4.2006 reacted with VLPs that circulated only from 2002 or later. No blockade epitopes were found to be in common between GII. 4.1987 and GII.4.2006. Prior to this work, we and others had predicted GII.4 antibody epitopes using bioinformatic techniques. As expected, discreet amino acids are repeatedly identified as potential evolving epitopes. In particular, residues 296-298 and 393-395 are consistently identified as putative epitopes that change between epidemic GII.4 strains. Additional surface residues at 333, 340, 356, 368, 372, 407 and 412-413 were also predicted as potential antibody epitopes [13, 18, 19, 22, 28, [40] [41] [42] . These amino acids tend to cluster on loops and ridges of the P2 subdomain where antibody interaction would be most accessible. Beyond bioinformatics predictions, only a few studies have shown empirical evidence mapping GII.4 antigenic change. Allen et al. [42] compared the reactivity of five monoclonal antibodies against a pre and post-2002 epidemic GII.4 strain, and identified conformational epitopes composed of residues 294-296 and 393-395. The carbohydrate blockade potential of these antibodies was not reported, so the role of these sites in escape from herd immunity was unclear. However, the finding that residues 393-395 were antigenically important supported our previously published work identifying amino acid 395 as an antigenic determinant in the GII.4.2002 Farmington Hills strain [13] . Using mouse mAbs and molecular biology approaches to exchange predicted epitopes between GII.4 strain backbones, we have clearly identified amino acids 294, 296-298, 368 and 372 to comprise an evolving blockade epitope, as exchange of these amino acids from GII.4.1987 into GII.4.2006 conferred binding of mAbs that recognize GII.4.1987 but not GII.2006 [43] . Extending this approach, we have also confirmed amino acids 407, 412 and 413 to constitute a GII.4.2002 Farmington Hills-specific blockade epitope [17] . These empirical studies support the validity of using computational analysis to guide norovirus epitope studies. Comparing reactivity of polyclonal sera collected from immunized mice and infected humans suggested antigenic variation within the GII.4 noroviruses [13, 33] . The development of mouse mAbs to different time-ordered GII.4 VLPs has greatly facilitated progress towards understanding the complex antigenic relations among these strains by clearly demonstrating antigenic variation over time and epidemic strain [34, 42, 43] . However, to maximally define the mechanistic relationships that exist between antigenic variation, immunity, and HBGA binding patterns noted in the GII.4 noroviruses in the context of natural infection history, the cross reactivity patterns, blockade responses, and epitope targets of human anti-GII.4 monoclonal antibodies are needed. Robust approaches exist for the isolation of human monoclonal antibodies that are elicited following virus infection. Using human PBMCs as Noroviruses are the principal cause of epidemic gastroenteritis worldwide with GII.4 strains accounting for 80% of infections. The major capsid protein of GII.4 strains is evolving rapidly, resulting in new epidemic strains with altered antigenic sites. To define these sites we prepared the first human monoclonal antibodies (Hu mAbs) against GII.4 noroviruses by immortalizing memory B cells and characterizing antibody reactivity and carbohydrate blockade responses across a ,20 year panel of time-ordered GII.4 virus-like particles (VLPs). Reflecting the complex exposure history of the patient, human anti-GII.4 mAbs grouped into three VLP reactivity patterns: broad (1987-2009), contemporary (2004) (2005) (2006) (2007) (2008) (2009) , and ancestral (1987) (1988) (1989) (1990) (1991) (1992) (1993) (1994) (1995) (1996) (1997) (1998) (1999) (2000) (2001) (2002) . We also identified the location of several defined epitopes which evolve over time and drive antigenic change. Our data indicate that antibodies targeting these sites block carbohydrate binding and likely select for the emergence of new strains that escape herd immunity and recognize unique carbohydrates for entry, resulting in new outbreaks of disease in vulnerable human populations. Importantly, these studies critically inform the rational design of broadly active vaccines and immunotherapeutics for the treatment of norovirus disease. a source of memory B cells, we created a panel of human mAbs directed against GII.4 strains and compared the reactivity of these mAbs to a panel of time-ordered GII.4 VLPs using EIAs and surrogate neutralization assays. We identified one novel, broadly cross reactive antibody that differentially blocks GII.4.1987 through 2009 VLP interactions with carbohydrate ligands, a potential immunotherapeutic for the treatment of acute or chronic GII.4 disease. We also defined unique antibody interactions with two different surface exposed epitopes that evolve over time. Importantly, antigenic variation in one of these epitopes correlated with changing carbohydrate ligand binding patterns over time, supporting the proposed relationship between epitope escape from human herd immunity and changing HBGA usage for virus docking [13] . In addition to defining the first human monoclonal antibodies with therapeutic potential for treating acute and chronic NoV GII.4 infections, these data support the hypothesis that GII.4 norovirus evolution results in antigenic drift of neutralizing epitopes and consequently, antibody-driven HBGA receptor switching; thus, protective herd immunity is a driving force in norovirus evolution. The development of mouse mAbs against different time-ordered GII.4 VLPs has greatly facilitated understanding of the complex antigenic relations between these strains by clearly demonstrating antigenic variation over time and by epidemic strain [34] . However, understanding of the human anti-GII.4 norovirus antibody response is essential not only for understanding the complex relationships between host immunity and virus antigenic change, but also for rational vaccine design based on defined neutralizing epitopes. Therefore, we developed a panel of human anti-GII.4 norovirus monoclonal antibodies to begin to characterize GII.4 antibody reactivity in the native virus host under natural infection conditions, noting that the norovirus preexposure histories in human volunteers are unknown and can only be inferred by human sera cross-reactive antibody binding and blockade patterns using time-ordered VLPs representing different outbreak and pandemic strains (Table 1) [13, 33] . Plasma and PBMC samples from 63 healthy individuals were collected in early 2009 and plasma binding titers (ED50) were measured by EIA against a panel of 6 different norovirus VLPs representing GI and GII strains ( Figure 1 ). The majority of tested samples reacted by EIA with variable ED50 titers to the panel of VLPs tested ( Figure S1 ). One sample (Donor NVB) was shown to react strongly with GII VLPs and was therefore chosen for further character- Figure 3 , and Figure S2 ). Significantly more antibody was needed to block GII.4.1997-PGM binding (EC50 0.4152 mg/ml) than GII.4.1987-PGM binding (EC50 0.3414 mg/ml) ( Figure 3B ) (p,0.05), supporting the hypothesis that subtle antigenic differences exist between these strains. The difference in blockade sensitivity of GII. 4.2006 and GII.4.2009 to NVB 97 provides the first evidence of subtle antigenic divergence between two Minerva variants, each of which caused widespread outbreaks globally [1] . This observation is further supported by Human mAbs NVB 111 and NVB 43.9 reactivity profiles. By single-dilution EIA, NVB 111 specifically reacted to 2006 but minimally with the 2009 variant of Minerva and other tested VLPs (Table 2 and Figure S2 ). Accordingly, NVB 111 required 13-fold more antibody to block GII.4.2009-PGM interaction (EC50 9.953 mg/ml) than it required to block GII.4.2006-PGM interaction (EC50 0.7376 mg/ml) ( Figure 5A and B) (p,0.05). In comparison, NVB 43.9 specifically recognized both the 2006 and 2009 Minerva variants by EIA (Table 2 and Figure S2 ). The interaction of both variants with PGM ligand was efficiently blocked by NVB 43.9 ( Figure 5C (Table 2 and Figure S2 ). Despite broad EIA reactivity, NVB 37.10 Figure 6D ). These three human mAbs indicate the existence of conserved GII.4 epitopes over the past twenty-five years and across three pandemic strains that could serve as targets for broad-based vaccine design. Importantly, NVB 71.4 represents the first potential, broad spectrum immune-therapeutic for any NoV. In addition to VLP-PGM interaction blockade assay, human mAbs were also tested for blockade of VLP-synthetic biotinylated HBGA (Bi-HBGA) interaction and ability to block VLP hemagglutination of O+ RBCs. Regardless of substrate (PGM or Bi-HBGA), the dose-response profiles for all blockade antibodies and VLPs were similar (compare Figure 7E and F), although EC50 titers were relatively high (EC50 0.9753 and 1.581 mg/ml for NVB 37.10 and 61.3, respectively), compared to the amount of antibody needed to block GII.4.2009 by the strain-specific mAbs (EC50 0.1140 and 0.1732 mg/ml for NVB 43.9 and 97). Repeated testing with PGM as substrate did not replicate the findings with synthetic carbohydrate substrates ( Figure 6A and B ). An additional measurement of antibody blockade ability uses RBCs as the VLP binding substrate. Previous work has demonstrated that Norwalk virus VLPs hemagglutinate (HA) O+ RBCs, that this interaction can be disrupted by antibodies found in polyclonal serum (hemagglutination inhibition; HAI), and that the HAI titer of serum correlates with antibody blockade of VLP-Bi-HBGA interaction [38, 39, 45] . To determine if these findings could be extended to study GII. 4 Our previous work with mouse-derived anti-norovirus mAbs suggested that blockade epitopes are conformation dependent [17, 34] . To test the effect of protein conformation of human mAb binding, we used both Western blot and EIA analysis to compare antibody binding to GII.4.2006 VLPs and P proteins. P proteins of GII.4.2006 are composed of the C-terminal portion of the major capsid protein (amino acids 221-531) [21] . Expression of the P protein in E. coli results in small particle formation estimated to consist of 12 P dimers that reportedly maintains VLP characteristics in carbohydrate and antibody binding studies [46, 47] . None of the human anti-GII.4 mAbs recognized either the denatured VLP or P protein by Western blot analysis, suggesting that the epitopes for these antibodies are conformation dependent (data not shown). Surprisingly, only half of the mAbs that recognized GII.4.2006 VLP ( Figure 8A ) by EIA also recognized GII.4.2006 P protein by EIA ( Figure 8B ). NVB 71.4 and 61.3 extended their broad reactivity to P proteins, whereas NVB 37.10 did not, indicating that a minimum of three GII.4 cross-reactive epitopes must exist. NVB 97 also detected P protein by EIA. Neither of the Minerva variant mAbs recognized P protein even at protein concentrations 8-fold above standard EIA conditions (1 mg/ml coating protein). Further, all seven mAbs detected increasing concentrations of VLP in a linear dose response with signals saturating at 4 mg/ml of VLP when the mAb concentration was held at 1 mg/ml, which is our standard EIA antibody titer ( Figure 8A ). Antibody reactivity to the P protein saturated at a lower protein concentration than VLP and at optical densities below the linear range of the assay (compare Figure 8A and 8B), suggesting that even among the mAbs that bind to P proteins conformation-based epitopes may be limited in a way not observed with VLPs. These data suggest two important points. First, some of the mAb epitopes are highly sensitive to conformation, and secondly, that the principle P protein conformation is not identical to VLPs at least for some critical blockade epitopes. The evolution of the GII.4 noroviruses was assessed over a 36year period of time by comparing strains from 1974 to 2010. In comparing these sequences, sites of variation in the P2 subdomain were noted, and these sites were mapped onto the crystal structure of the P-domain dimer for the 1997 strain VA387. Surfaceexposed sites of variation were then examined to determine which residues may be close enough to constitute a single epitope, and five epitopes were predicted based upon this variation ( Figure 9A , and [40] ). Epitope A encodes significant amino acid changes over time and has also been demonstrated to be an evolving GII.4 blockade epitope using mouse mAbs ( Figure 9A, 9B and [43] ). Epitope A is conformational and is located on the top of the capsid proximal to the HBGA binding pocket. Six variable sites were close to each other in the region of this putative epitope, suggesting that these residues may work in concert to change the local structure of Epitope A. The variable, surface-exposed residues include positions 294, 296-298, 368 and 372. Of note, Epitope A is continuing to evolve in extant strains, whereby the amino acid at position 294 seems to vary extensively in strains from 2008-2010 (amino acid replacements P294A, P294S and P294T have been observed at this position). Epitope B was predicted based upon two variable residues at positions 333 and 382. While these residues are buried in the dimer interface between two chains, the patterns of variation at these sites suggest that they play an important role in the evolution of novel strains, perhaps by evolving replacements that allow the more surface exposed residues in other surface exposed epitopes to dramatically change the physiochemical properties of the amino acid replacements. Residues 340 and 376 make up the variable residues of putative Epitope C. This putative conformation dependent epitope is on the surface and lateral edge of the capsid and is directly proximal to the HBGA binding pocket, suggesting that this epitope may play a role in receptor switching along with Epitope D. Epitope D is comprised Table S2 . doi:10.1371/journal.ppat.1002705.g007 of three variable residues from positions 393-395. In the first reported crystal structure for the GII.4 noroviruses, this region was reported to be a secondary HBGA binding site [16] . However, the location of this epitope on the surface of the capsid, directly proximal to the HBGA binding site, suggests that it likely plays a role in both receptor switching and in escape from herd immunity and perhaps both, simultaneously [13, 21, 40, 43] . Epitope D is close enough to the HBGA binding pocket to contribute to or inhibit carbohydrate binding, and yet variable enough to suggest that it is targeted by the immune response. Putative Epitope E is comprised of variable residues 407, 412 and 413, which are surface exposed regions lateral to the HBGA binding pockets and the other epitopes. These residues vary with every major epidemic strain after 2002, suggesting that it is a hot spot for the emergence of immunologically novel GII.4 strains. Epitope E is a GII.4.2002 blockade antibody epitope [17] . This putative epitope is lateral to the HBGA binding pockets indicating that antibodies are targeting interior regions below the capsid surface, which suggests that other epitopes may be present in the P1 subdomain. A few variable residues do not necessarily identify the boundaries of a putative epitope. Moreover, it is nearly impossible to predict the surface area of a putative epitope by sequence analysis alone. Therefore, we expanded the putative epitopes to include residues within 8 Å of the variable sites from which the epitopes were predicted ( Figure 9B ). The described mAbs indicate at least five unique or overlapping GII.4 blockade epitopes with different specificities: 1) early GII.4 strain specific, 2) contemporary GII.4 strain specific, 3) Minervavariant strain specific, 4) genogroup II strain specific, and 5) GII.4 strain specific. Using capsid sequences as a guide, mutant VLPs were designed to contain chimeric combinations of the predicted evolving GII.4 epitopes ( Figure 10) All epitope exchange VLPs were morphologically intact by electron microscopy visualization and retained the ability to bind PGM ( Figure 10A and B, [43] ), confirming chimeric VLP structural integrity. Epitope mutant VLPs were compared to wild type strain VLPs for reactivity to the donor plasma sample. Consistent with high EIA reactivity to GII.4.1987 and 2006 VLPs (Table 2 and Figure S2 ), donor plasma reacted across the panel of epitope-exchange mutant VLPs ( Figure 10B) . Donor plasma was able to block each epitope-exchanged VLP binding to PGM ( Figure 11A and C) . Exchange of all of the epitopes, except Epitope D into either backbone and GII.4.1987 C into GII.4.2006 resulted in significantly different EC50 values compared to the parental strains ( Figure 11B , 11D, and Table S4 ) (p,0.05). Only the exchange of Epitope A between the backbones resulted in an exchanged blockade phenotype, as observed with epitope-specific mAbs [43] . Exchange of Epitope A between the two parental backbones resulted in a chimeric VLP (GII.4.1987/2006A) that was blocked with significantly less plasma than the parental GII.4.1987 (EC50 0.0167 mg/ml compared to 0.0673 mg/ml) (p,0.05) and a chimeric VLP (GII.4.2006/1987A) that was blocked with signif- Figure S2 and Table 2 ). Bars are SEM. doi:10.1371/journal.ppat.1002705.g008 Table S4 ) (p,0.05). In this individual, these data suggest that Epitope A may be an important evolving GII.4 neutralization epitope as the blockade response is significant enough to be detected in the polyclonal antibody response. Agreeing with the assumption that epitope-exchange mutants are unlikely to identify epitopes of cross-reactive mAbs, NVB 37.10, 61.3 and 71.4 reacted with the entire panel of chimeric VLPs by EIA ( Figure 10B ). In contrast, each of the strain-specific mAbs displayed differential EIA reactivity to exchanged epitopes Table S4 ). Exchange of the other GII.4.1987 epitopes did not eliminate NVB 114 blockade potential ( Figure 12A and B) Figure 13A and B) . Binding was not restored to wild type levels as the EC50 of GII.4.1987/2006D was 0.6349 mg/ml, significantly higher than the blockade EC50 for GII.4.2006 -PGM (0.1195 mg/ml) ( Figure 13B and Table S4 ) (p,0.05). Both EIA and blockade data clearly indicate that Epitope D is critical for the binding of NVB 97 and suggest that amino acids 393-395 are important components of a GII.4 evolving blockade epitope in addition to modulating VLP-carbohydrate ligand binding [13, 21] . Interestingly, Epitope D has a single amino acid change in GII.4.2004, explaining the highly conserved binding and blockade responses noted across GII.4.2005 to 2009 VLPs, while ancestral strains display significant antigenic variation across these residues. Together, these data map the GII.4 evolving blockade epitopes recognized by each of the four strain-exclusive human mAbs described in this study ( Figure 14 ). Noroviruses are recognized as a leading cause of viral foodborne gastroenteritis. With the successful vaccination program being developed against rotavirus, focus is shifting to norovirus as the primary causative agent of severe childhood diarrhea resulting in a yearly estimate of 1.1 million episodes of pediatric gastroenteritis in developed nations and 218,000 deaths in developing nations [9] . The elderly and immunocompromised also suffer sometimes life-threatening or chronic long-term norovirus diarrheal disease characterized by malnutrition and dehydration [48] . In some HIV infected patients, chronic norovirus diarrheal disease is associated with persistent norovirus infection [49] . The economic disease burden of a norovirus outbreak within a care facility has been estimated at over $657,000 for a single event [50] . These statistics emphasize the critical need for a norovirus vaccine. Although recent reports strongly support the development and use of an efficacious norovirus vaccine in humans [38, 51] , a primary obstacle to a successful vaccine is the lack of a definitive correlate to protection coupled with the extreme antigenic variation across the many norovirus strains. In fact, the existence of long-term protective immunity to norovirus infection remains controversial within the field [52] . Human challenge studies conducted before molecular diagnostics of infection and refined immune response assays had indicated that some volunteers could be reinfected with the same norovirus strain, suggesting that norovirus infection did not induce long-term protection [53] . However, recent reports identifying immune responses in norovirus-challenged but uninfected volunteers [44, 54] have necessitated qualification of these early observations to acknowledge that the findings may be compromised by assay limitations and/or the overwhelming challenge dose in comparison to the very low norovirus infectious dose [44, 54] . Clearly defining the relationships between pre-exposure history, blockade antibody responses, T cell immunity, virus evolution, and the components of protective immunity represent key challenges for future vaccine and therapeutic design. In addition to the clinical applications of therapy and diagnosis, monoclonal antibodies have also proven to be superior tools for studying viral antigenicity, evolution, and for the treatment of acute viral disease in humans [13, 34, 55, 56] . Characterization of neutralizing mAb escape-mutants has been fundamental to identifying epitopes associated with virus receptor usage, pathogenesis, and fitness [57] [58] [59] . In this manuscript, we isolated and characterized the first human mAbs against noroviruses, derived from a healthy donor whose pre-exposure history was unknown. The number of unique GII.4 human monoclonal antibodies in this patient accurately reflects the high prevalence of GII.4 norovirus infection seen in human populations over the past 25 years. Moreover, distinct antibody cross reactivity patterns support the hypothesis that the GII.4 genotype is undergoing antigenic variation which not only correlates with loss of antibody blockade activity and emergence of new epidemic norovirus strains, but also changing carbohydrate ligand binding patterns over time. Importantly, antibody-mediated antigenic drift of GII.4 strains coupled with both mucosal IgA and T cell responses in challenged but uninfected volunteers strongly support the existence of longterm protective immunity against norovirus strains. In fact, the unique antibody reactivity patterns characterized herein are most likely explained as an archeological immune record of successive waves of contemporary GII.4 infections in this individual over time, implying that successful vaccine design is possible, as previously proposed by our group and others [37, 44, 54] . Previously, we have identified two immunological responses associated with protection from infection in norovirus-challenged volunteers. An early (day 1-3) post-exposure salivary IgA response in genetically susceptible volunteers correlated with protection from Norwalk virus infection [54] and an early T h 1 response correlated with protection from Snow Mountain virus infection [44] . Historically, the role of IgG in norovirus protection has been unclear. By adulthood, .90% of the population [48] is positive for anti-norovirus IgG, and norovirus strains within a genogroup share a high degree of antigenic cross-reactivity as measured by EIA [44, 60] . These facts likely skew functional interpretations of the role of serum IgG titers on susceptibility to infection and/or infection outcomes. Although carbohydrate ligand blockade antibody responses have been suggested as correlates of protective immunity [13, 35, 36] , it wasn't until recently that these blockade responses were correlated definitively with protection from clinical disease and infection [10, 38] . Importantly, our surrogate neutralization assay is specific enough to differentiate GII.4 norovirus strains too similar to be distinguished by EIA but different at key antigenic sites. While human re-challenge studies using the same viral inoculum are necessary to confirm an association between a blockade IgG response and protection from repeat norovirus infection, our findings support the clinical relevance of the antibody blockade assay as a correlate of protective immunity [38] . Monoclonal antibodies coupled with the blockade assay are powerful tools for elucidating the antigenic relationship between GII.4 strains. While mouse mAbs provide insight into GII.4 antigenic structure, data in this manuscript argues that human mAbs offer considerable advantages, including: a) immunologic record of B-cell immunity following repeat GII.4 infection, b) relationships between antibody blockade responses and antigenic variation, c) relationships between immune selection and carbohydrate ligand reactivity, d) relationships between early infection and downstream immunity, and e) epitope mapping. We focused on anti-GII.4 mAbs because of the clinical relevance of the GII.4 strains. One key finding is a direct relationship between anti-GII. (Figure 3) . Given the lack of cross reactivity with later strains, the most likely explanation is that NVB 114 was derived from a long-term memory plasma cell that had been elicited some 12-22 years earlier. In support of this idea, examination of human monoclonal antibodies against 1918 H1N1 influenza also identified antibody variants that recognized ancestral or contemporary isolates [61] . The exclusive blockade reactivity of NVB 114 with GII.4.1987 and 1997 supports potential long-term protective immunity. Further, NVB 114 is the first antibody identified to clearly demonstrate antigenic difference between GII. 4.1987 and GII.4.1997 , suggesting that antigenic variation may have subtly contributed to the emergence of the GII.4 US 95/96 pandemic strain from ancestral strains. Human monoclonal antibodies clearly identified two evolving epitopes on the surface of the GII.4 VLP. Epitopes A and D were both confirmed as evolving GII.4 antibody blockade epitopes using chimeric VLPs containing a mixture of epitopes derived from early or contemporary strains. We recognize that Epitopes A and D have not been structurally defined as epitopes, supporting the need for structural studies to define the exact mAb binding site on the VLPs. Moreover, we focused on discrete regions of varying residues and all of the residues within 8 Å (representing approximately 201 Å 2 ) to the primary sites that may be influenced directly by an amino acid replacement ( Figure 9B ). However, Ab binding sites have been reported to be much larger (700-800 Å 2 ), suggesting that the epitopes that we have predicted may actually work in concert to form larger Ab recognition sites. By expanding the putative epitopes, we identified other residues that were less variable; however, the exact role that these varying residues play in evolution is less clear. In some cases variation may be required to encode changes that are necessary for the replacement at a primary site. In addition, all of our structural analyses have been conducted using models of the P dimer, representing about 1/90 th of the VLP structural surface. Interactions between the epitopes in the context of the superstructure have not been determined. Therefore, these observations indicate that the complex nature of the NoV Ab epitope requires further research to define the specific boundaries and residues that regulate Ab binding. The prediction of five putative epitopes allowed us to gain several important insights into GII.4 norovirus evolution: 1) Discrete sites of variation occur on the GII.4 norovirus capsid, either directly on the surface or lateral to the HBGA binding sites; 2) Secondary variable sites are within 8 Å of the primary variable sites, and these secondary sites could also contribute to epitope remodeling; 3) Many of the putative expanded epitopes overlap, suggesting that two or more highly variable epitopes may work in concert to escape from an antibody response; 4) Putative epitopes that are buried may exert an effect on the structure by altering the interior fold space, allowing unconventional replacements to be tolerated; 5) An underlying amino acid network likely preserves the functional core of the capsid proteins by regulating the variable residues above them; and 6) Escape and HBGA binding may be intimately linked via the underlying regulatory network of amino acids that preserve the functional integrity of the capsid core. Epitope A, which likely includes varying amino acid residues 294, 296-298 and 368 and 372 and potentially other undefined nearby residues, has been mapped as a blockade epitope in both GII Figure 9A ). At this time it is unclear how many and which amino acids in Epitope A are needed to mediate an escape mutant phenotype that is completely resistant to GII.4.2006 antibody blockade. However, it is clear that Epitope A varies, and that the site is conserved as a major target for blockade antibody response between 1987 and 2009. Although correlative, comparisons of Epitope A variation along with residues that are proximal to those that appear to be evolving over time suggest that changes at positions 292, 293, 294, 295, 296, 297, 298, 300, 365, 367, 368 and 372 might contribute to an escape phenotype, with residues 294, 296, 297, 298, 300, 368 and 372 playing direct roles in this variation ( Figure 9 ). However, the minor replacements at other positions are likely essential for remodeling the local structural neighborhood such that more profound changes can be tolerated. Supporting the sensitivity of epitopes to the local environment, mAbs that recognized Epitope A differentiated between GII.4.2006 VLPs and GII.4.2006 P dimers. P protein (P dimer) is a dimeric, truncated form of the major capsid protein composed of residues 214-539 [62] . P-dimers have been widely used to determine the crystallographic structure of NoV-HBGA interactions and are considered accurate reflections of the VLP surface topology [16, 21, 63] . P -particles can assemble as higher ordered structures composed of varying copies of the P dimer [46, 64] . These subviral particles are reported to have similar characteristics to VLPs [62, [65] [66] [67] and have been proposed as a candidate vaccine platforms [68] . To our knowledge, this is the first immunologic characterization of P-dimers vs. VLPs using monoclonal antibodies derived from human infections. Here, P-dimers derived from GII.4.2006 lost binding of mAbs NVB 111 and 43.9, the mAbs that recognize Epitope A in GII. 4.2006 . P dimer binding to the Epitope D binding mAb NVB 97 and the broadly cross-blockade mAb NVB 71.4 were retained. Recently, P-particle vaccines were shown to be less robust at inducing strong blockade responses, as compared with intact VLP [69] , perhaps because of the loss of blockade Epitope A in this higher ordered structure. While speculative, it is recognized that virions ''breathe'' suggesting that the possibility that P-dimers and P-particles may become ''locked'' in a slightly less immunologically reactive state that affects some but not all blockade epitopes on the virion surface [70, 71] . These data absolutely underscore the critical importance of determining the structures of several of these human mAbs with their appropriate GII.4 VLP epitopes by either cryoEM or crystallography, for informing targeted mutagenesis to identify the role of key residues in regulating antigenicity and antibody escape. Epitope D is a conformational epitope comprised of varying amino acids 393-395 and likely other nearby residues that are less clearly defined; however, additional mapping and crystallographic studies will be needed to clarify this epitope structure. With the emergence of the pandemic GII. 4 (Figure 4) . Previously, these residues have been implicated in regulating norovirus VLPcarbohydrate ligand binding interactions [13, 21, 43] . The identi-fication of Epitope D as a human antibody blockade epitope that changes over time provides direct support for our previous hypothesis that escape from protective herd immunity may drive changes in carbohydrate ligand binding affinities over time and potential retargeting of virus infection in different human populations [13, 40] . These data indicate that like influenza, a successful norovirus vaccine regimen will require periodic population sampling to identify future strains for inclusion into the next year's vaccine formulation not unlike the strategy employed by the Influenza Virus Global Surveillance Program. Norovirus population sampling has already begun as monitoring systems for detection of NoV infections have been established in the United States, Europe, and Japan. Identification of GII.4 evolving antigenic epitopes furthers our understanding of norovirus pathogenesis and provides target epitopes that may be useful for surveillance and prediction of new strain emergence. Identification of GII.4 conserved epitopes also informs diagnostic and potential therapeutic reagent development and design. Monoclonal antibodies NVB 37.10, 61.3 and 71.4 all recognize epitopes conserved among the GII.4 strains from 1987 through 2009. NVB 37.10 and 61.3 have enhanced GII VLP recognition, binding not only GII.4 VLPs but also other GII VLPs, but are unable to block VLP-PGM interaction. The high conservation of the NVB 37.10 and 61.3 epitopes suggest that these epitopes are highly resistant to antigenic variation within the GII strains, making these mAbs potentially valuable diagnostic reagents as GII strains cause up to 95% of norovirus outbreaks [27, 72] . The unidentified epitope for NVB 71.4 is clearly different from the epitopes recognized by NVB 37.10 and 61.3 and is conserved throughout the GII.4 strains. NVB 71.4 did not recognize any non-GII.4 VLPs but, importantly, it exhibited blockade activity for the entire panel of timeordered GII.4 VLPs with PGM and Bi-HBGAs. Emphasizing the difference between the two quantitative blockade assays and the qualitative HAI assay, GII.4.2002 HA was not inhibited by NVB 71.4. Noting that blockade assays are not true measurements of neutralization, NVB 71.4 has potential as a therapeutic reagent based on its broad GII.4 blockade potential and the fact that it is by nature a human antibody. Clearly, the effectiveness of NVB 71.4 at preventing or treating illness can only be determined empirically. Although one of the blockade-epitope specific mAbs with lower EC50 values/steeper Hill constants may be more effective at select strain neutralization, the breadth of strains neutralized is likely to be limited for these mAbs. There are a number of viral diseases currently being treated with mAbs including RSV, CMV and enterovirus; however, only the anti-RSV humanized mAb palivizumab has FDA approval for prophylactic use in humans [73, 74] . In outbreak settings or in chronically infected patients, an anti-NoV mAb that could be delivered before symptoms begin and protect from illness could be very useful in care facilities, the military and the cruise industry. Given the acute clinical disease window, it is less likely that therapeutic antibodies will provide relief in those individuals experiencing acute infections, however, therapeutic antibodies may offer opportunities for ameliorating symptomatic disease in chronic infections. The discovery of broadly cross reactive and cross blockade human GII.4 mAbs dictates the need for a new approach to map epitopes. Our current experimental approach was designed to identify GII.4 epitopes that change over time and provide insight into broadly conserved epitope locations. Because the identification of the epitopes recognized by NVB 37.10, 61.3, and 71.4 has important implications for successful vaccine design, new panels of mutated VLPs and other approaches will be needed to characterize these epitopes in the future. GII.4 NoVs are significant human pathogens that cause considerable morbidity and mortality, worldwide. The development of mouse mAbs to different time-ordered GII.4 VLPs has greatly facilitated progress towards understanding the complex antigenic relationships between these strains by clearly demonstrating antigenic variation over time and epidemic strain [34] . Here we have expanded these observations using human anti-GII.4 mAbs isolated from a healthy adult donor, who has likely experienced multiple norovirus infections throughout his/her lifetime. The identification of highly significant, varying antigenic epitopes that influence VLP-carbohydrate ligand interaction provides important new insights into vaccine design and the development of therapeutics that target norovirus virions. For example, these antibodies represent the first anti-norovirus human mAbs to be characterized, and they confirm findings from studies using mouse mAbs supporting antigenic drift and its linkage with varying carbohydrate ligand binding profiles within the GII.4 noroviruses. Further, we have demonstrated that the GII.4 NoV varying epitopes can be exchanged between time-ordered VLPs, providing a robust platform for expanding the antigenic and blockade cross reactivity of future vaccine candidates. Using this approach, we have identified two surface-exposed antibody blockade epitopes that vary over time and were differentially recognized by four of the seven human mAbs. We also identified three antibodies which recognize either overlapping or three unique highly conserved epitopes within the GII.4 VLP. These data continue to support the hypothesis that norovirus long-term protective immune responses are elicited following acute infection, a concept essential for effective vaccine design. We anticipate that a full understanding of the varying antigenic and blockade epitopes of GII.4 NoVs may not only help to predict the emergence of new epidemic strains but simultaneously identify key reformulations in vaccine design that will protect public health against contemporary and emerging epidemic strains in the future. A diverse panel of VLPs representing G1 and GII norovirus strains and epitope mutants was assembled as previously reported [13, 17, 75] . To design epitope exchange chimeric VLPs, we first identified surface exposed residue clusters that varied over time. Then, we synthesized a series of chimeric GII.4 ORF2 genes that exchanged ''putative'' epitopes between GII. 4.1987 and GII.4.2006 VLPs [43] . For all constructs except GII.4.2009 ORF2 [17] , the synthetically derived constructs were inserted directly into the VEE replicon vector for the production of virus replicon particles (VRPs) as previously described by our group. VLPs were expressed in VRPinfected BHK cells and purified by velocity sedimentation in sucrose and stored at 280uC. The GII.4.2009 (New Orleans [17] ) VLPs were expressed in the baculovirus system and purified by cesium chloride gradient centrifugation and were the kind gift of Dr. Jan Vinje, Centers for Disease Control and Prevention, Atlanta, GA. VLP protein concentrations were determined by the BCA Protein Assay (Pierce, Rockford, IL). VLP preparation purity averaged .80% by SDS-Page analysis. In early 2009, following written consent, blood samples from 63 donors were collected from adult healthy donors at the Lugano and Basel Blood banks (Switzerland). Peripheral blood mononuclear cells (PBMCs) and plasma were isolated and cryopreserved. On the day of use, PBMCs from Donor 302898 (Figure 1 ), an individual born in 1948, were thawed and IgG + memory B cells were isolated using CD22 microbeads (Miltenyi) followed by cell sorting, as described [76] . Cells were immortalized at 5 cells/well in multiple cultures using EBV in the presence of CpG oligodeoxynucleotide 2006 (Microsynth) and irradiated allogeneic PBMC. After 2 weeks, culture supernatants were screened for the presence of norovirus-specific mAbs by EIA against VLPs and positive cultures were cloned by limiting dilution. Antibodies were recovered from supernatants and purified using protein A affinity chromatography and finally desalted against PBS using a HiTrap FastDesalting column. Human mAb reactivity was determined by EIA, as reported [34] . Briefly, plates were coated at 1 mg/ml VLP in PBS before the addition 1 mg/ml purified IgG or donor plasma (0.2%). Primary antibody incubation was followed by anti-human IgG-alkaline phosphatase and color development with pNPP substrate solution (Sigma Chemicals, St. Louis, MO). Each step was followed by washing with PBS-0.05% Tween-20 and all antibodies were diluted in 5% dry milk in PBS-0.05% Tween-20. Data shown represent the average of at least three replicates and are representative of similar data from at least two independent trials. Establishment of EIAs using new mAbs included PBS-coated wells as negative controls and polyclonal anti-norovirus human sera as positive controls. Antibodies were considered positive for reactivity if the mean optical density after background subtraction for VLPcoated wells was greater than three times the mean optical density for PBS-coated wells [34] . For screening donor plasma samples, the binding titers of plasma to respective coated VLPs were determined by EIA as described above by measuring the plasma dilution required to achieve 50% maximal binding (ED50). EIA reactivity to GII.4.2006 P protein (amino acids 221-531 [21] ) was measured similarly to reactivity to VLP. GII.4.2006 P protein was the kind gift of B.V. Prasad, Baylor College of Medicine, Houston, TX). Pig Gastric Mucin Type III (PGM) (Sigma Chemicals) has been validated as a substrate for NoV VLP antibody-blockade assays [17] . PGM contains relatively high levels of H and A antigen and more moderate levels of Lewis Y antigen [17] . All of the GII.4 VLPs used in the blockade assays in this study bind to both PGM and synthetic Bi-HBGA, and binding to PGM is consistent with synthetic Bi-HBGA binding profiles for a-1,2-fucose (H antigen) and a-1,4-fucose (Lewis antigen) containing molecules [13, 17, 77] . For blockade assays, PGM was solvated in PBS at 5 mg/ml and coated onto EIA plates at 10 mg/ml in PBS for 4 hours and blocked over night at 4uC in 5% dry milk in PBS-0.05% Tween-20. VLPs (0.5 mg/ml) were pretreated with decreasing concentrations of test mAb or donor plasma for 1 hour at room temperature before being added to the carbohydrate ligand-coated plates for 1 hour. Bound VLP was detected by a rabbit anti-GII norovirus polyclonal sera made from hyperimmunization with either GII.4.2009 or a cocktail of GII. 4.1997 , GII.3.1999 , GII.1.1976 , and GII.2.1976 VLPs, followed by anti-rabbit IgG-HRP (GE Healthcare) and color developed with 1-Step Ultra TMB ELISA HRP substrate solution (Thermo-Fisher). The percent control binding was defined as the binding level in the presence of antibody pretreatment compared to the binding level in the absence of antibody pretreatment multiplied by 100. All incubations were done at room temperature. Each step was followed by washing with PBS-0.05% Tween 20 and all reagents were diluted in 5% dry milk in PBS-0.05% Tween-20. All antibodies were tested for blockade potential against the panel of GII.4 VLPs at two-fold serial dilutions ranging from 0.08 to 2 mg/ ml. Additional concentrations of blockade antibodies were tested if needed to complete the sigmoid dose-response curve. Blockade of synthetic Bi-HBGAs (Glycotech, Gaithersburg, MD) assays were done as described for PGM with the following exception. Bi-HBGAs were bound to Neutri-avidin coated plates (Pierce) at 10 mg/ml for one hour prior to the addition of 1 mg/ml VLP for 1.5 h. Reported mean % control binding reflects the results of at least two independent experiments with each dilution tested at least in duplicate. An antibody was designated as a ''blockade'' antibody for a VLP if at least 50% of control binding was inhibited by 2 mg/ ml antibody. Blockade data were fitted and EC50 values calculated using Sigmoidal dose response analysis of non-linear data in GraphPad Prism 5 (www.graphpad.com). EC50 values between VLPs were compared using the One-way ANOVA with Dunnett post test, when at least three values were compared or the unpaired t-test when two values were compared. A difference was considered significant if the P value was ,0.05. To test for antibody binding that prevents detector antibody from binding to the VLP instead of the VLP binding to the PGM, select blockade assays are performed without using a detector antibody and instead developed directly with an anti-human IgG-HRP. Antibodies tested this way give two responses; 1) a bell-shaped response curve for antibodies that are blockade and 2) a sigmoidal shaped curve for antibodies that are not blockade. These data indicate that it is the amount of human mAb that is directly blocking the VLP from binding to PGM. Of note, VLP concentrations in blockade assays are in the low nanomolar range and therefore cannot discriminate between antibodies with sub-nanomolar affinities. HAI assays were performed as reported [38, 39, 45] . VLPs at 50 ng/reaction were pretreated with antibody as described above for the blockade assays before addition to O+ RBCs at 4uC, pH 5.5. An HAI titer was determined as the lowest antibody concentration that completely prevented NoV VLP-induced HA by visualization. The amino acid sequences of GII. 4.1987 , GII.4.2002 , and GII.4.2006 capsids were individually aligned to the VA387 P domain sequence using Clustalx1.86 [78] , and the GII.4.2002 domain dimer X-ray crystal structure (PDB accession: 2OBT) [16] was used as a template for generating homology models. Homology models were generated using the program Modeller available via the Max Planck Institute Bioinformatics Toolkit (http://toolkit.tuebingen.mpg.de/). The structural models were analyzed and compared, and figures were generated using Mac Pymol (Delano Scientific).

Induction of GADD34 Is Necessary for dsRNA-Dependent Interferon-β Production and Participates in the Control of Chikungunya Virus Infection

Nucleic acid sensing by cells is a key feature of antiviral responses, which generally result in type-I Interferon production and tissue protection. However, detection of double-stranded RNAs in virus-infected cells promotes two concomitant and apparently conflicting events. The dsRNA-dependent protein kinase (PKR) phosphorylates translation initiation factor 2-alpha (eIF2α) and inhibits protein synthesis, whereas cytosolic DExD/H box RNA helicases induce expression of type I-IFN and other cytokines. We demonstrate that the phosphatase-1 cofactor, growth arrest and DNA damage-inducible protein 34 (GADD34/Ppp1r15a), an important component of the unfolded protein response (UPR), is absolutely required for type I-IFN and IL-6 production by mouse embryonic fibroblasts (MEFs) in response to dsRNA. GADD34 expression in MEFs is dependent on PKR activation, linking cytosolic microbial sensing with the ATF4 branch of the UPR. The importance of this link for anti-viral immunity is underlined by the extreme susceptibility of GADD34-deficient fibroblasts and neonate mice to Chikungunya virus infection.

During their replication in host cells, RNA and DNA viruses generate RNA intermediates, which elicit antiviral responses mostly through type-I interferon (IFN) production [1, 2] . Several families of proteins are known to sense double-stranded RNA (dsRNA), including endocytic Toll-like receptor 3 (TLR3) [3] , the dsRNA-dependent protein kinase (PKR) [4] and the interferoninducible 29-59-oligoadenylates and endoribonuclease L system (OAS/2-5A/RNase L) [5] . Viral dsRNA and the synthetic dsRNA analog polyriboinosinic:polyribocytidylic acid (poly I:C) are also detected by different cytosolic DExD/H box RNA helicases such as the melanoma differentiation-associated gene 5 (MDA5), DDX1, DDX21, and DHX36, which, once activated, trigger indirectly the phosphorylation and the nuclear translocation of transcription factors such as IRF-3 and IRF-7, resulting predominantly in abundant type-I IFN and pro-inflammatory cytokines production by the infected cells [1, 6, 7] . Alphaviruses such as Chikungunya virus (CHIKV) are small enveloped viruses with a message-sense RNA genome, which are known to be strong inducers of type-I IFN in vivo [8, 9] , a key response for the host to control the infection [10, 11, 12] . In vitro, however, response to RNA viruses is heterogeneous, since Sindbis virus (SINV), do not elicit detectable IFN-a/b production in infected of murine embryonic fibroblasts (MEFs) [13] . The specific points of blockage of type-I IFN production during infection are still not well delineated, but SINV and other alphaviruses could antagonize IFN production by shut-off of host macromolecular synthesis in infected cells [14, 15, 16] . Recently, human fibroblasts infection by CHIKV was shown to trigger abundant IFN-a/b mRNA transcription, while preventing mRNA translation and secretion of these antiviral cytokines [13, 15] . Contrasting with these reports, other groups using different CHIKV strains have observed abundant type-I IFNs release in the culture supernatants of CHIKV-infected human monocytes [17] , human lung cells , human foreskin fibroblasts and MEFs [10] . Type-I IFN stimulation of non-hematopoietic cells has also been shown to be essential to clear infection upon CHIKV inoculation in mouse, but CHIKV was found to be a poor inducer of IFN secretion by human plasmacytoïd dendritic cells [10] . Thus, great disparities regarding alphavirus-triggered IFN responses exist between viral strains and the nature of host cells or animal models. Once bound to their receptor on the cell surface (IFNAR), type-I IFNs activate the Janus tyrosine kinase pathway, which induces the expression of a wide spectrum of cellular genes including Pkr [18] . These different genes participate in the cellular defense against the viral infection. PKR is a serine-threonine kinase that binds dsRNA in its N-terminal regulatory region and induces phosphorylation of translation initiation factor 2-alpha (eIF2a) on serine 51 [19, 20] , leading to protein synthesis shut-off and apoptosis. PKR has been also been shown to participate in several important signaling cascades, including the p38 and JNK pathways [21] , as well as type-I IFN production [22, 23] . Inhibition of translation, IFN responses and triggering of apoptosis combine to make PKR a powerful antiviral molecule, and many viruses have evolved strategies to antagonize it [24, 25] . Interestingly, several positive RNA-strand viruses (eg. Togaviridae or Picornaviridae) have been shown to activate PKR, resulting in phosphorylation of eIF2a and host translation arrest [26] , while viral mRNA could initiate translation in an eIF2-independent manner by means of a dedicated RNA structure, that stalls the scanning 40S ribosome on the initiation codon [25] . Despite the existence of these viral PKR-evading strategies, the importance of PKR for type-I IFN production has been strongly debated over the years and even considered dispensable since the discovery of the innate immunity function of the DExD/H box RNA helicases [27, 28] . However, several PKR-deficient cell types have reduced type-I IFN production in response to poly I:C [23, 29, 30] , while PKR was demonstrated to be required for IFNa/b production in response to a subset of RNA viruses including Theiler's murine encephalomyelitis, West Nile (WNV) and Semliki Forest virus (SFV), but not influenza, Newcastle disease, nor Sendai virus [31, 32, 33, 34] . These studies raise therefore the possibility that some but not all viruses induce IFN-a/b in a PKRdependent and cell specific manner. Infection of PKR or RNAse L deficient mice demonstrated that these enzymes were not absolutely necessary for type I IFN-mediated protection from alphaviruses such as SFV or WNV, but still contributed to levels of serum IFN and clearance of infectious virus from the central nervous system [25, 35] . Similarly, deficient mice for both PKR and RNAse L showed no increase in morbidity following SINV infection, although, like during WNV infection, increased viral loads in draining lymph nodes were observed [35, 36] . These observations support a non-redundant and cell specific role for these enzymes in the amplification of type-I IFN responses to viral infection more than in their initiation [31, 32, 35] . Nevertheless, the exacerbated phenotypes observed upon alphavirus infection of mice deficient for type-I IFN receptor (IFNAR), underlines the limits of the individual contributions of PKR and RNAse L to the anti-viral resistance of adult animals [10, 35, 36] . In addition to dsRNA detection, different stress signals trigger eIF2a phosphorylation, thus attenuating mRNA translation and activating gene expression programs known globally as the integrated stress response (ISR) [37] . To date, four kinases have been identified to mediate eIF2a phosphorylation: PKR, PERK (protein kinase RNA (PKR)-like ER kinase) [38] , GCN2 (general control non-derepressible-2) [39, 40] and HRI (heme-regulated inhibitor) [41, 42] . ER stress-mediated eIF2a phosphorylation is carried out by PERK, which is activated by an excess of unfolded proteins accumulating in the ER lumen [38] . Activated PERK phosphorylates eIF2a, attenuating protein synthesis and triggering the translation of specific molecules such as the transcription factor ATF4, which is necessary to mount part of a particular ISR, known as the unfolded protein response (UPR) [43, 44] . Interestingly DNA viruses, such as HSV, that use the ER as a part of its replication cycle, have been reported to interfere with the ER stress response through different mechanisms, such as the dephosphorylation of eIF2a by the viral phosphatase 1 activator, ICP34.5 [45, 46] . We show here, using SUnSET, a non-radioactive method to monitor protein synthesis [47] , that independently of any active viral replication, cytosolic poly I:C detection in mouse embryonic fibroblasts (MEFs) promotes a PKR-dependent mRNA translation arrest and an ISR-like response. During the course of this response, ATF4 and its downstream target, the phosphatase-1 (PP1) cofactor, growth arrest and DNA damage-inducible protein 34 (GADD34, also known as MyD116 and Ppp1r15a) [48] , are strongly up-regulated. Importantly, although the translation of most mRNAs is strongly inhibited by poly I:C, that of IFN-ß and Interleukin-6 (IL-6) is considerably increased under these conditions. We further demonstrate that PKR-dependent expression of GADD34 is critically required for the normal translation of IFN-ß and IL-6 mRNAs. We prove the relevance of these observations for antiviral responses using CHIKV as a model: we show that GADD34-deficient MEFs are unable to produce IFN-ß during infection and become permissive to CHIKV. We further show that CHIKV induces 100% lethality in 12-day-old GADD34-deficient mice, whereas WT controls do not succumb to infection. Our observations demonstrate that induction of GADD34 is part of the anti-viral response and imply the existence of distinct and segregated groups of mRNA, which require GADD34 for their efficient translation upon dsRNA-induced eIF2a phosphorylation. We monitored protein synthesis in MEFs and NIH-3T3 cells after poly I:C stimulation, using puromycin labeling followed by immunodetection with the anti-puromycin mAb 12D10 [47] . Poly I:C delivery to MEFs and NIH-3T3, rapidly and durably inhibited protein synthesis, concomitant with increased eIF2a phosphorylation (P-eIF2a) ( Fig. 1A and Fig. S1A ). In MEFs, a strong eIF2a phosphorylation was observed after 4 h of poly I:C treatment, followed by a steady dephosphorylation at later times (Fig. 1A) . Protein synthesis arrest was confirmed in individual cells by concomitant imaging of poly I:C delivery, mRNA translation and P-eIF2a ( Fig. 1B and Fig. S1B ), and with a wide range of dsRNA concentrations (Fig. S1C ). Poly I:C-induced eIF2a phosphorylation and subsequent translation arrest were not observed in PKRdeficient MEFs ( Fig. 1C and 1D ), while eIF2a phosphorylation induced by the UPR-inducing drug thapsigargin (th) (an inhibitor of SERCA ATPases) or arsenite (as) was unchanged in PKR2/2 cells (Fig. 1C) . PKR is therefore necessary to induce protein synthesis inhibition in response to cytosolic poly I:C. Nucleic acids detection by multiple molecular sensors results in type-I interferon production, which protects cells and tissues from viral infections. At the intracellular level, the detection of double-stranded RNA by one of these sensors, the dsRNA-dependent protein kinase also leads to the profound inhibition of protein synthesis. We describe here that the inducible phosphatase 1 co-factor Ppp1r15a/ GADD34, a well known player in the endoplasmic reticulum unfolded protein response (UPR), is activated during double-stranded RNA detection and is absolutely necessary to allow cytokine production in cells exposed to poly I:C or Chikungunya virus. Our data shows that the cellular response to nucleic acids can reveal unanticipated connections between innate immunity and fundamental stress pathways, such as the ATF4 branch of the UPR. When levels of IFN-ß were quantified in culture supernatants and compared to total protein synthesis intensity, we found that most of the cytokine production occurred after 4 to 8 h of pIC delivery (Fig. 1E , WT, and S1D), a time at which mRNA translation was already considerably decreased ( Fig. 1A and S1E). We measured the amount of cytokine produced in NIH-3T3 cells at a time (7 h) at which translation was already strongly inhibited ( Fig. 1G and 1F ). To prove that IFN-b production truly occurred during this poly I:C-induced translation arrest, cells exposed for 7 h to poly I:C were washed and old culture supernatants replaced with fresh media for 1 h (with or without CHX), prior translation monitoring (Fig. 1F , right) and IFN-ß dosage (Fig. 1G , right). We observed that close to 30% of the total IFN-ß produced over 8 h of poly I:C stimulation is achieved during this 1 h period, despite a close to undetectable protein synthesis in the dsRNA-treated cells (Fig. 1F ). The neo synthetic nature of this IFN was further demonstrated by the absence of the cytokine in CHX-treated cell supernatants. IFN-b production in response to poly I:C is therefore likely to be specifically regulated and occurs to a large extent independently of the globally repressed translational context. As previously observed in MEFs, IFN-ß production in response to poly I:C was independent of PKR ( Fig. 1E ) [31] . This suggests that although its production occurs during cap-mediated translation inhibition, it does not directly depend on a specialized open reading frame organization, as described for the translation of the mRNAs coding for the UPR transcription factor ATF4 or the SV 26S mRNA upon eIF2a phosphorylation [26, 49] . This hypothesis is also supported by the ability of MEFs expressing the non-phosphorylatable eIF2a Ser 51 to Ala mutant (eIF2a A/A), to produce normal levels of IFN-ß in response to poly I:C (Fig. 1E) , while global translation was not inhibited by poly I:C in these cells (Fig. S2 ). We went on to investigate the molecular mechanisms promoting this paradoxical IFN-ß synthesis in an otherwise translationally repressed environment. Induction of eIF2a phosphorylation by PERK during ER stress promotes rapid ATF4 synthesis and nuclear translocation, followed by the transcription of many downstream target genes important for the UPR [50] . Similarly, in presence of PKR, nuclear ATF4 levels were found to be upregulated in MEFs responding to cytosolic poly I:C, albeit less importantly than upon a bona fide UPR induced by thapsigargin ( Fig. 2A) . One of the key molecules involved in the control of eIF2a phosphorylation is the protein phosphatase 1 co-factor GADD34, which relieves translation repression during ER stress by promoting eIF2a dephosphorylation [50, 51] , [52] . GADD34 is a direct downstream transcription target of ATF4 [53] . Expression of GADD34 was quantified by qPCR and immunoblot in WT and PKR 2/2 MEFs (Fig. 2B) . In WT cells GADD34 mRNA expression was clearly up-regulated (20 fold) in response to poly I:C, while GADD34 protein induction was equivalent in poly I:Cand thapsigargin-treated cells. GADD34 mRNA transcription and translation were not observed in PKR 2/2 cells responding to poly I:C, but occurred normally upon thapsigargin treatment, paragoning eIF2a phosphorylation (Fig. 2B, right) . We next investigated the importance of ATF4 for GADD34 transcription by monitoring the levels of GADD34 mRNA in ATF4-deficient cells. ATF4 2/2 MEFs displayed higher basal levels of GADD34 mRNA than WT cells. However, in absence of ATF4, MEFs were unable to efficiently induce GADD34 mRNA transcription in response to any of the stimuli tested (Fig. S3 ). GADD34 mRNA expression was induced only 2 fold in ATF4 2/2 MEFs exposed to poly I:C, suggesting that its transcription is mostly dependent on ATF4 in this context. We further investigated P-eIF2a requirement for GADD34 expression and found that eIF2a A/A expressing MEFs were incapable of up-regulating GADD34 in response to poly I:C (Fig. 2C ). Phosphorylation of eIF2a by PKR in response to cytosolic poly I:C induces therefore a specific integrated stress response (ISR), that allows ATF4 translation, its nuclear translocation and subsequent GADD34 mRNA transcription. GADD34 expression is required for global translation recovery in response to thapsigargin but not to poly I:C We next evaluated the relevance of GADD34 induction, by treating WT and GADD34 DC/DC fibroblasts with poly I:C or with drugs known to induce ER stress, such as thapsigargin and the Nglycosylation inhibitor tunicamycin [52] . As expected, in WT cells eIF2a phosphorylation was rapidly increased in response to all ISR-inducing stimuli and decreased concomitantly with the expression of GADD34 over time ( Fig. 3A and S4 ) [52] . Consequently eIF2a phosphorylation was greatly increased in GADD34 DC/DC MEFs in all the conditions tested ( Fig. 3A and S4A ). In thapsigargin-treated cells, protein synthesis was reduced in the first hour of treatment and rapidly recovered (Fig. 3B ) [54] . Poly I:C, however, nearly completely inhibited translation despite active eIF2a dephosphorylation. This was particularly obvious when poly I:C was co-administrated together with thapsigargin. Indeed, poly I:C dominated the response by preventing the translation recovery normally observed after few hours of drug treatment (Fig. 3B) . Surprisingly, in absence of functional GADD34, although eIF2a phosphorylation induction by poly I:C was augmented dramatically, no further decrease in protein synthesis was observed upon treatment of GADD34 DC/DC cells with the dsRNA mimic ( Fig. 3A and 3C ). The functionality of GADD34 in translation restoration was, however, fully demonstrated, when the same cells were treated with thapsigargin, and protein synthesis was completely inhibited by this treatment [52] control. Quantification of puromycin signal was quantified with ImageJ software and is represented above the immunoblot. Phosphorylation of eIF2a (P-eIF2a) was assessed in the same MEFs extracts. B) Immunofluorescence staining for puromycin, P-eIF2a and dsRNA of MEFs treated with poly I:C for 4 h and labeled with puromycin for 1 h. Scale bar, 10 mm. C) WT and PKR 2/2 MEFs were stimulated for 8 h with poly I:C (pI:C), thapsigargin (th) or arsenite (as). PKR and P-eIF2a were detected by immunoblot. D) WT and PKR 2/2 MEFs were stimulated for 8 h with poly I:C and protein synthesis was monitored like in (A). b-actin immunoblot is shown for equal loading control. E) IFN-b levels were measured, by ELISA, in cell culture supernatants of WT, PKR 2/2 , eIF2aA/A and control eIF2aS/S MEFs after 4 and 8 h of poly I:C stimulation. Data are mean 6 standard deviation of 3 independent experiments. F) Protein synthesis was measured in NIH3T3 cells by puromycin incorporation after 7 h of poly I:C treatment. Where indicated, a chase of 1 h with fresh media was performed prior to puromycin labeling and immunoblotting. Samples with cycloheximide (chx) and arsenite (as) added respectively 5 min and 30 min before the puromycin pulse are shown as controls. G) IFN-b was quantified by ELISA in culture supernatants in the conditions described above after 7 h of poly I:C stimulation or 7 h of poly I:C stimulation followed by 1 h with fresh media (chase). Data are mean 6 standard deviation of 4 independent experiments. doi:10.1371/journal.ppat.1002708.g001 (Fig. 3C) . Thus, cytosolic dsRNA delivery induces a type of protein synthesis inhibition, which requires eIF2a phosphorylation for its initiation, but conversely cannot be reverted by GADD34 induction and subsequent GADD34-dependent eIF2a dephosphorylation. The potential contribution of the OAS/2-5A/RNAse L system to this P-eIF2a-independent inhibitory process was evaluated by investigating RNA integrity in MEFs exposed to poly I:C. We used capillary electrophoresis to establish precise RNA integrity numbers (RIN) computed from different electrophoretic traces (pre-, 5S-, fast-, inter-, precursor-, post-region, 18S, 28S, marker) and quantify the degradation level of mRNA and rRNA potentially resulting from the activation of this well characterized anti-viral pathway. No major RNA degradation could be observed upon poly I:C delivery (Fig. S5 ), suggesting that global RNA degradation does not contribute extensively to the long term translation inhibition observed upon poly I:C delivery in our experimental system. We have observed that GADD34 expression counterbalances PKR activation by promoting eIF2a dephosphorylation, however it has little impact on reversing the global translation inhibition initiated by poly I:C. We next monitored the production of specific proteins and cytokines in WT and GADD34 DC/DC MEFs (Fig. 4) . . PKR is required for ATF4 and GADD34 expression in response to poly I:C. A) WT and PKR 2/2 MEFs were stimulated for 8 h with poly I:C (pI:C), or the UPR-inducing drug, thapsigargin (th) for 6 h. ATF4 protein expression was detected by immunoblot on nuclear extracts. Nuclear HDAC1 immunoblot is shown for equal loading control. * indicates unspecific band. B) GADD34 mRNA levels were quantified by qPCR after 6 h of poly I:C (pI:C) treatment in WT and PKR 2/2 MEFs. For the same cell extracts, immunoblots of GADD34 (middle panel), PKR and P-eIF2a (right panel) were performed. C) The same analysis was performed in eIF2a A/A and control eIF2a S/S MEFs. Treatment with thapsigargin (th), for 6 h was used as control to induce GADD34 and P-eIF2a. eIF2a and b-actin immunoblots are shown for equal loading control. Quantitative PCR data are the mean 6 standard deviation of 3 independent experiments. doi:10.1371/journal.ppat.1002708.g002 Cystatin C, a cysteine protease inhibitor was chosen as a model protein, since its secretion ensures a relative short intracellular residency time so that its intracellular levels directly reflect its synthesis rate [55] . This is confirmed by the N-glycosylated-and total Cystatin C accumulation in cells treated with brefeldin A (Fig. 4A , left panel). Cystatin C levels were found to follow a similar trend to that observed with total translation, being strongly reduced upon poly I:C exposure and not profoundly influenced by GADD34 inactivation (Fig. 4A, right panel) . Thapsigargin treatment induced a brief drop in cystatin C levels, prior to some levels of GADD34-dependent recovery. 6 hours of tunicamycin treatment affected more cystatin C accumulation than anticipated Figure 3 . GADD34 mediates eIF2a dephosphorylation but not global translation recovery in response to poly I:C. A) After treatment with poly I:C, protein extracts of WT and GADD34 DC/DC MEFs were immunobloted for GADD34 and P-eIF2a. B) Protein synthesis was analyzed in WT cells treated for 1 to 6 hours with poly I:C (pI:C) alone or together with thapsigargin (th). Controls are cells not treated with puromycin (co) and cells treated with cycloheximide (chx) 5 min before puromycin incorporation. C) Protein synthesis was analyzed in GADD34 DC/DC cells treated for 1 to 6 hours with poly I:C (pI:C) alone or together with thapsigargin (th). Tubulin or b-actin immunoblot are shown for equal loading control. In GADD34 DC/DC cells translation is strongly impacted by thapsigargin, but not poly I:C. doi:10.1371/journal.ppat.1002708.g003 Fig. 4A, right panel) , probably due to interference with the Nglycosylation and associated folding of this di-sulfide bridge containing protein [55] , thereby promoting its degradation by endoplasmic reticulum-associated protein degradation (ERAD) [56] . We next turned towards PKR, which displayed a pattern of expression completely different from cystatin C (Fig. 4B ). As expected from its IFN-inducible transcription, levels of PKR were increased in poly I:C-treated MEFs (Fig. 4B) , despite the strong global translation inhibition observed in these cells (Fig. 3) . GADD34 inactivation appeared to influence the accumulation of PKR, since the cytoplasmic dsRNA sensor levels were not upregulated and even decreased in poly I:C-treated GADD34 DC/DC MEFs (Fig. 4B ). Control treatment with tunicamycin and thapsigargin did not alter significantly PKR levels (Fig. 4B) , suggesting that ER stress did not influence the kinase expression. The absence of PKR up-regulation in the poly I:C-treated GADD34 DC/DC MEFs led us to investigate the capacity of these cells to produce anti-viral and inflammatory cytokines, which normally drive PKR expression through an autocrine loop. We ruled out any interference from the UPR in triggering IFN-ß production in our experimental system, since, as anticipated from PKR expression, tunicamycin and thapsigargin treatments were not sufficient to promote cytokine production in MEFs (Fig. S6 ) [43, 44] . We therefore investigated IFN-ß and IL-6 production in response to dsRNA in WT, GADD34 DC/DC and CReP 2/2 MEFs. CReP 2/2 MEFs were used as a control, since CReP (Ppp1r15b) is a non-inducible co-factor of PP1 and displays some functional redundancy with GADD34 [57] . Although basal levels of eIF2a phosphorylation were higher in CReP 2/2 , PKR expression and translation inhibition upon poly I:C delivery were equivalent in WT and CReP 2/2 MEFs ( Fig. S7A and S7B ). Quantification of IFN-ß and IL-6 levels in culture supernatants indicated that, although abundant and comparable amounts of these cytokines were secreted by WT and CReP 2/2 cells, they were both absent in poly I:C-treated GADD34 DC/DC MEFS ( Fig. 4C and S7C) . Quantitative PCR analysis revealed that, IFN-ß, IL-6 and PKR transcripts were potently induced in poly I:C treated GADD34 DC/ DC MEFs (Fig. 4D ), thus excluding any major transcriptional alterations in these cells, as confirmed by the normal levels of cystatin C mRNA, which remained constant in all conditions studied. Moreover, using confocal immunofluorescence microscopy, we could not detect intracellular IFN-b in poly I:C-stimulated GADD34 DC/DC MEFs, in contrast to WT cells, which abundantly expressed the cytokine, despite the global translation arrest (Fig. S8) . Thus, we could attribute the deficit in cytokine secretion of the GADD34 DC/DC MEFs to a profound inability of these cells to synthesize cytokines, rather than to a defect in transcription or general protein secretion. GADD34 induction by poly I:C is therefore absolutely necessary to maintain the synthesis of specific cytokines and probably several other proteins in an otherwise translationally repressed context. Importantly, GADD34 exerts its rescuing activity only on a selected group of mRNAs including those coding for IFN-ß and IL-6, but not on all ER-translocated proteins, since cystatin C synthesis was strongly inhibited by poly I:C in all conditions tested. Interestingly, in GADD34 DC/DC MEFs, PKR mRNA strongly accumulated in response to poly I:C (Fig. 4D) , despite the absence of detectable IFN-ß production and PKR protein increase (Fig. 4B) . This continuous accumulation of PKR mRNA in response to poly I:C suggests the existence of alternative molecular mechanisms, capable of promoting PKR mRNA transcription and stabilization independently of autocrine IFN-b detection. Nevertheless in these conditions PKR expression, like IFN-b, was found to be dependent on the presence of GADD34 for its synthesis (Fig. 4B) . Recent results indicate that PKR participates to the production of IFN-a/ß proteins in response to a subset of RNA viruses including encephalomyocarditis, Theiler's murine encephalomyelitis, and Semliki Forest virus [31] . Even though IFN-a/ß mRNA induction is normal in PKR-deficient cells, a high proportion of mRNA transcripts lack their poly(A) tail [31] . As GADD34 induction by poly I:C was completely PKR-dependent, we wondered whether the phenotypes observed in PKR 2/2 cells and GADD34 DC/DC MEFs could be related. Oligo-dT purified mRNA extracted from cells exposed to poly I:C were therefore analyzed by qPCR. PolyA+ mRNAs coding for IFN-ß and IL-6 were equivalently purified and amplified from WT and GADD34 DC/DC MEFs (Fig. S9 ). This confirms that albeit the phenotypes of PKR 2/2 and GADD34 DC/DC cells might be linked, mRNA instability is not the primary cause of the cytokine production defect observed in GADD34 DC/DC . Taken together these observations suggest the existence of a specific mRNAs pool, encompassing cardinal immune effectors such as IFN-ß, IL-6, and PKR, which are specifically translated in response to dsRNA sensing and increased levels of P-eIF2a. This mRNAs pool requires GADD34 for their translation during the global protein synthesis shut-down triggered by dsRNA detection. We verified that GADD34 inactivation, and no other deficiency, was truly responsible for the loss of cytokine production by complementing GADD34 DC/DC MEFs with GADD34 cDNA prior poly I:C delivery. IFN-ß secretion was partially restored in transfected GADD34 DC/DC cells while eIF2a was efficiently dephosphorylated in both WT and GADD34 DC/DC transfected MEFs (Fig. 4E) . To further demonstrate that the phosphatase activity of GADD34 controls cytokine production upon dsRNA detection, we treated WT MEFs with guanabenz, a small molecule, which selectively impairs GADD34-dependent eIF2a dephosphorylation [58] . Upon treatment with this compound, a dose dependent inhibition of IFN-ß secretion was observed in poly I:C-treated MEFs, confirming the importance of GADD34 in this process (Fig. S10) . as controls to induce GADD34. Immunoblot of tubulin is shown as equal loading control. C) Amount of IFN-b (left panel) and IL-6 (right panel) in cell culture supernatants of WT and GADD34 DC/DC MEFs after 6 h of poly I:C stimulation. Mock are samples treated with lipofectamine alone. Data are mean 6 standard deviation of five (IFN-b) and three (IL-6) independent experiments. D) Transcription of IFN-b, IL-6, PKR and Cystatin C was analyzed by qPCR in samples of WT and GADD34 DC/DC MEFs treated with poly I:C (pI:C). Mock represent samples treated with lipofectamine alone. E) WT and GADD34 DC/DC MEFs were transfected overnight with an expression plasmid carrying the murine GADD34 (G34) cDNA and then treated with poly I:C for 6 h. IFN-b production was quantified by ELISA, left panel, in cell culture supernatants and plotted as a ratio of IFN-b to total cell proteins to compensate for different cell mortality levels induced by the transfection. In the right panel immunoblots for GADD34 and P-eIF2a in the same experimental conditions are shown. One representative analysis of 3 independent experiments is shown. doi:10.1371/journal.ppat.1002708.g004 GADD34 is necessary for IFN production and to control Chikungunya virus infection Fibroblasts of both human and mouse origin constitute a major target cell of Chikungunya virus (CHIKV) during the acute phase of infection [59] . In adult mice with a totally abrogated type-I IFN signaling, CHIKV-associated disease is particularly severe and correlates with higher viral loads. Importantly, mice with one copy of the IFN-a/ß receptor (IFNAR) gene develop a mild disease, strengthening the implication of type-I IFN signaling in the control of CHIKV replication [59] . Recently, human fibroblasts infection by CHIKV was shown to induce IFN-a/ß mRNA transcription, while preventing mRNA translation and secretion of these antiviral cytokines. CHIKV was found to trigger eIF2a phosphorylation through PKR activation, however this response is not required for the block of host protein synthesis [15] . We tested the importance of PKR during CHIKV infection by infecting WT and PKR 2/2 MEFs with CHIKV-GFP, at a multiplicity of infection (MOI) of 10 and 50. Productive infection was estimated by GFP expression (Fig. 5A, left panel) , while culture supernatants were monitored for the presence of IFN-b (5A, right panel). PKR was found to be necessary to control CHIKV infection in vitro, since at least 60% of PKR-inactivated cells were infected after 24 of viral exposure, compared to only 15% in the control fibroblasts population. WT MEFs produced efficiently IFN-b, while the hypersensitivity to infection of the PKR 2/2 MEFs was correlated to a reduced type-I IFN production capacity after infection. Thus, during CHIKV infection, PKR is required for normal IFN production by MEFs. We also monitored protein synthesis in infected WT and PKR 2/2 fibroblasts using puromycin labeling followed by immunofluorescence confocal microscopy (Fig. 5B) . CHIKV-GFP positive PKR 2/2 MEFs were found to incorporate efficiently puromycin, while in their infected WT counterpart protein synthesis was efficiently inhibited. Thus CHIKV, in this experimental model, induces a PKR-dependent protein synthesis inhibition and is therefore particularly relevant to further confirm our observations on the role of GADD34 in controlling type-I IFN production during response to viral RNAs. GADD34 DC/DC MEFs were exposed to CHIKV-GFP (MOI of 10 or 50) for 24 and 48 h. Productive infection was estimated by GFP expression and virus titration (Fig. 6A) , and culture supernatants monitored for the presence of type-I IFN (Fig. 6B , left). Only minimal CHIKV infection (15%) could be observed at maximum MOI in WT MEFs (Fig. 6A, left) , while robust IFN-b amounts were already produced at the lowest MOI (Fig. 6B) . Contrasting with WT cells and regardless of the MOI used, a higher level of viral replication was observed in GADD34 DC/DC MEFs (Fig. 6A) . The GADD34-inactivated cells were clearly more sensitive to CHIKV, displaying a 50% infection rate after 24 h of infection (MOI 50) and a log more of virus titer in culture supernatants (Fig. 6A, right) . Correlated with their susceptibility to CHIKV infection, IFN-b production was nearly undetectable in GADD34 DC/DC MEFs (Fig. 6B) . Such observation confirms the incapacity of GADD34-deficient cells to produce cytokines in response to cytosolic dsRNA, a deficiency likely to facilitate viral replication. This interpretation is further supported by the abrogation of viral replication in both WT and GADD34 DC/DC MEFs briefly treated with IFN-b (Fig. 6C) . Thus, GADD34 inactivation does not favor viral replication per se, but is critical for type-I IFN production. Interestingly infection levels were found to be higher in PKR2/2 than in GADD34 DC/DC MEFs, although this difference could be attributed to clonal MEFs variation, it more likely suggests that PKR-dependent translation arrest could be key in preventing early viral replication in this system. In addition, the relatively lower permissivity of GADD34 DC/DC MEFs to infection at high MOI could indicate the existence of GADD34-dependent defense mechanisms, which could be independent from IFN production and eIF2-a dephosphorylation. To strengthen and generalize these observations, we treated a different strain of WT MEFs with guanabenz and examined the consequences for CHIKV infection. Biochemically, GADD34 expression was induced upon CHIKV infection, and guanabenz treatment resulted in a clear increase in eIF2a phosphorylation, demonstrating the importance of GADD34 in limiting this process during infection (Fig. 6D, right) . As observed with GADD34 DC/DC cells, pharmacological and RNAi inhibition of GADD34 was found to increase significantly the sensitivity of MEFs to infection, while reducing their IFN-b production ( Fig. 6D and S10) . Thus, induction of GADD34 and its phosphatase activity during CHIKV infection, in vitro, participates to normal type-I IFN production and control of viral dissemination. Several components of the innate immune response have been shown to impact on the resistance of adult mice and to restrict efficiently CHIKV infection and its consequences in vivo [10] . We decided to investigate the importance of GADD34 upon intradermal injections of CHIKV to WT (FVB) and GADD34 DC/DC mice. Neither strain of adult mice was affected by intradermal injections of CHIKV, with little statistically significant differences in the virus titers found in the different organs. Thus, GADD34 deficiency does not annihilate all the sources of type-I IFN in the infected adult animals, a situation exemplified by the capacity of GADD34 DC/DC bone-marrow derived dendritic cells to produce reduced, but measurable IFN-b in response to poly I:C [60] . This also infers that the light impact of GADD34 inactivation on mouse development [61] does not render these animals more sensitive to CHIKV infection. As in Humans, CHIKV pathogenicity is strongly agedependent in mice, and in less than 12 day-old mouse neonates, CHIKV induces a severe disease accompanied with a high mortality rate [59] . GADD34 function was therefore evaluated in this more sensitive context by injecting intradermally CHIKV to FVB (WT) and GADD34 DC/DC neonatal mice. As previously observed for C57/BL6 mice [59] , when CHIKV was inoculated to FVB neonates, a rate of 50% of mortality was observed 3 days after the infection of 9-day-old mice, while 12-day-old pups were found essentially resistant to the virus lethal effect (Fig. 7A ). Strongly contrasting with these results, all CHIKV infected GADD34 DC/DC neonates died within 3-5 days post inoculation whatever their age (Fig. 7A ). When infection was monitored 5 days post-inoculation of 12-day-old mice at, GADD34 DC/DC pups displayed considerably more elevated CHIKV titers (10-100 folds) in most organs tested, including liver, muscle, spleen and joints, the later being primarily targeted by the virus (Fig. 7B, left) . As expected, and in full agreement with the in vitro data, infected GADD34 DC/DC tissues showed a considerably reduced IFN-ß production (40-50%) compared to control tissues ( Figure 7B, right) , while serum levels were reduced by 20% (not shown). Although Infectious virus was poorly detected in the heart of WT animals, elevated titers of virus were observed in the heart of GADD34-deficient pups, matching the limited production of IFN in this organ. We further investigated the possible pathological consequences of cardiac tissue infection by carrying-out comparative histopathology. Hearts of infected GADD34-deficient animals displayed severe cardiomyocytes necrosis with inflammatory infiltrates by monocytes/ macrophages and very important calcium deposition (Fig. 8) , all being indicative signs of grave necrotic myocarditis. As a consequence, the left ventricles were strongly dilated, being probably the cause of acute cardiac failures and of the important death rate observed in GADD34 DC/DC infected pups. Histology of infected FVB mice hearts was, however, normal with only few inflammatory cells (mainly lymphocytes) observed in the close vicinity of capillaries. GADD34 expression is therefore necessary to allow normal type-I interferon production during viral infection and to promote the survival of young infected animals. We could circumvent the age-related acquisition of viral resistance in GADD34 DC/DC mice to 17 days, since mice inoculated at that age survived CHIKV inoculation. In these animals, 3 days post-infection, enhanced viral replication was observed in the spleen and muscles, matching the relatively low level of type-IFN production in these tissues ( Figure 7C ). Functional GADD34 is therefore required to mount a normal innate response against the virus, but in older mice type-I IFN production by non-infected innate cells is probably capable to gradually overcome GADD34-deficiency and limit viral proliferation in vital organs, such as the heart. Translation inhibition occurs in response to stress, when other cellular activities have to be reassigned or suspended momentarily. We demonstrate here that the activation of PKR by cytosolic dsRNA results in a stress response, leading to ATF4 and GADD34 induction. GADD34 expression has been observed during the infection of cells by different types of viruses [62] or intracellular bacteria such as Listeria monocytogenes [63] . Our observations demonstrate that GADD34 expression is a direct consequence of PKR activation and dsRNA sensing. Interestingly, although GADD34 induction by poly I:C promotes eIF2a dephosphorylation, this is not sufficient to prevent global protein synthesis arrest. The uncoupling of efficient eIF2a dephosphorylation from global translation recovery in response to cytosolic poly I:C implies therefore the existence of additional mechanisms inhibiting global translation. The 2-5A/RNAse L pathway does not seem to be sufficiently active in our experimental setting to explain this prolonged protein synthesis inhibition. The cleavage or the inactivation of other translation factors could work in concert with eIF2a to block or affect the efficiency of other individual steps of mRNA translation [64] . For instance, the phosphorylation of translation elongation factor 2 (eEF-2) is also controlled by eIF2a phosphorylation. Thus, Thr56 phosphorylation of eEF-2, which is known to inhibit its translational function by reducing its affinity for ribosomes, could contribute directly to the protein synthesis inhibition induced by PKR activation [65] . Independently of general protein synthesis inhibition, eIF2a dephosphorylation is necessary for the production of specific proteins upon dsRNA-induced translation inhibition. As demonstrated for ATF4, translation of a given mRNA during stress could rely on the structure and organization of its coding sequence, as well as the presence of multiple alternative initiation codons [49] . Surprisingly, functional GADD34 expression was found necessary for the translation of IL-6, IFN-b, and PKR. This observation points to the existence of a distinct group of mRNAs efficiently translated upon dsRNA detection and dependent on GADD34 activity. GADD34 is extremely short lived and has been shown to accumulate on the ER, when over-expressed [51] . GADD34 could mediate its activity at the ER level and influence differently eIF2a sub-cellular distribution according to the type, localization, and level of activity displayed by the different eIF2a kinases. The strong eIF2a phosphorylation mediated by PKR in response to poly I:C or viral infection and leading to the initiation of translation inhibition, could be circumvented through GADD34 activity solely at the ER level, thereby allowing local cytokine production in absence of other functional protein synthesis. This selectivity for translation of several specific mRNAs among other ER-secreted molecules suggests further that GADD34 dependent mRNAs might display specific features allowing their efficient identification by GADD34 and associated molecules, as well as allowing their translation in presence of minimal levels of active guanine nucleotide exchange factor eIF2B. GADD34 and PKR are necessary to produce anti-viral cytokines during CHIKV infection, and probably other types of infection. PKR, ATF4 and GADD34 should therefore be considered as an essential module of the innate anti-viral response machinery. The importance of PKR in anti-viral type-I IFN responses has been the object of contradictory reports [30, 31, 66, 67] . Our observations, however, suggest that PKR function should be re-evaluated by integrating the impact of viral detection on cellular translation. In eIF2A/A and PKR 2/2 cells, cytokine transcription is induced normally following poly I:C detection by DExD/H box RNA helicases, while as expected in these cells, no eIF2a phosphorylation and subsequent host translation inhibition are observed. This lack of translation arrest in the absence of potent eIF2a phosphorylation allows for normal cytokine production during dsRNA detection, with no requirement for an operational GADD34 feedback loop. The importance of PKR and GADD34 for IFN-b and other cytokines production could therefore be directly linked to the efficiency of the cellular translation inhibition induced by RNA viruses, as exemplified here with CHIKV, which in MEFs strongly activates PKR and subsequent protein synthesis inhibition. GADD34 DC/DC neonates are extremely sensitive to CHIKV infection and display signs of acute myocarditis and ventricles dilatation probably causing recurrent cardiac failures. CHIKV cardiac tropism is not normally observed in WT mouse and inability of heart tissues to produce sufficient type-I IFN in GADD34 DC/DC could allow abnormally high viral replication, myocarditis and dilated cardiomyopathy. Interestingly many cases of myopericarditis induced by CHIKV and leading to dilated cardiomyopathies in infected patients have been reported since the Figure 7 . CHIKV infection in mouse neonates. A) Kaplan-Meier plots representing the survival of FVB (WT) and GADD34 DC/DC mouse neonates 9-day-old (n = 11 per group) (upper panel) or 12-day-old (n = 14 per group) (lower panel) after intradermal inoculation with 10 6 PFU of CHIKV and observed for 21 days. B) Left panel, viral titers in different tissues and serum of 12-day-old mice inoculated with 10 6 PFU of CHIKV via the intradermal route. Mice were sacrificed 5 days after infection and the amount of infectious virus in serum and tissues quantified by TCID50 (see methods) (n = 5). In addition of considerably increased levels of viral replication in CHIKV target tissues, GADD34 DC/DC neonates also display signs of heart infection. Right panel, Quantification of IFN-b for the same different tissues, CHIKV-infected target tissues of GADD34 DC/DC mice produced less IFN-b than WT. C) 17-day-old mice were infected with 10 6 PFU of CHIKV via the intradermal route, and sacrificed 72 h later. Quantification of viral titers and IFN-b/ viral titers ratio is presented for different tissues. A broken line indicates the detection threshold. In B and C represented data are arithmetic mean 6 standard deviation, n = 5. In B and C p values were calculated using a Student's t test, *p#0.1, **p#0.05. doi:10.1371/journal.ppat.1002708.g007 Figure 8 . CHIKV infection causes severe myocarditis in mouse neonates. Histological appearance of horizontal sections of the heart through left and right ventricles of 12-day FVB (A, C and E) and GADD34 DC/DC mice at D5 pi (B, D and F). Normal appearance of heart of FVB infected mice, at low magnification (A, 610) with normal cardiomyocytes (C, 6100) and exceptional small foci of lymphocytes (E, 6400). Numerous foci of necrosis in the heart of GADD34 DC/DC infected mice, at low magnification (B, 610) and extensive through the ventricular wall (D, 6100). Higher magnification shows few residual cardiomyocytes (arrow head) and inflammation mainly composed of monocytes as well as extensive deposition of calcium (F, 6400). The mice were inoculated with 10 6 PFU of CHIKV via the intradermal route. doi:10.1371/journal.ppat.1002708.g008 1970s after the different western Indian Ocean islands and Indian subcontinent disease outbreaks [68, 69] . These particular symptoms and complications might therefore be the consequences of great variation in the tissue-specific type-I IFN levels induced in CHIKV-infected patients, who might display particular polymorphisms in their innate viral sensing pathways increasing their peculiar susceptibility to viral dissemination in the heart. Importantly, our data reveal a link between pathogen-associated molecular patterns (PAMPs) and the UPR through the activation of the eIF2-a/ATF4 branch [70] . Similarly, several laboratories have reported that TLR stimulation activates the XBP-1 branch of the UPR and that XBP-1 production was needed to promote a sustained production of inflammatory mediators, including IL-6 [71, 72] . Here, we identify GADD34 as a novel functional link between ISR and PAMPs detection in MEFs, required for the production of cytokines including type-I IFN. It will now be important to explore the therapeutic potential of targeting GADD34 to reduce cytokines overproduction during inflammatory conditions. Puromycin labelling for measuring the intensity of translation was performed as previously described [47] . For immunoblots, 10 mg/ml puromycin (Sigma, min 98% TLC, cell culture tested, P8833, diluted in PBS) was added in the culture medium and the cells were incubated for 10 min at 37uC and 5% CO 2 . Where indicated, 25 mM cycloheximide (Sigma) was added 5 min before puromycin. Cells were then harvested, centrifuged at 4uC and washed with cold PBS prior to cell lysis and immunoblotting with the 12D10 antibody. Cells were lysed in 1% Triton X-100, 50 mM Hepes, 10 mM NaCl, 2.5 mM MgCl 2 , 2 mM EDTA, 10% glycerol, supplemented with Complete Mini Protease Inhibitor Cocktail Tablets (Roche). Protein quantification was performed using the BCA Protein Assay (Pierce). 25-50 mg of Triton X-100-soluble material was loaded on 2%-12% gradient or 8% SDS-PAGE before immunoblotting and chemiluminescence detection (SuperSignal West Pico Chemiluminescent Substrate, Pierce). Nuclear extraction was performed using the Nuclear Complex Co-IP kit (Active Motif). Rabbit polyclonal antibodies recognizing ATF4 (CREB-2, C-20), GADD34 (C-19), Lamin A (H-102) and eIF2-a (FL-315) were from Santa Cruz Biotechnology, as well as mouse monoclonal anti-PKR (B-10). GADD34/PPP1R15A (Catalog No. 10449-1-AP) rabbit polyclonal antibody was purchased from PROTEINTECH. Rabbit polyclonal anti-eIF2a[pS 52 ] and Cystatin C were from Invitrogen and Upstate Biotechnology, respectively. Mouse monoclonal antibodies for b-actin and HDAC1 (10E2) were purchased from Sigma and Cell Signaling Technologies. Secondary antibodies were from Jackson ImmunoResearch Laboratories. MEFs and NIH3T3 were grown on coverslips overnight and stimulated for the indicated time with poly I:C complexed with Lipofectamine 2000. Cells were fixed with 3% paraformaldehyde in PBS for 10 min at room temperature, permeabilized with 0,5% saponin in 5% FCS PBS with 100 mM glycine, for 15 min at room temperature and stained for 1 h with indicated primary antibodies. Anti-P-eIF2a was from BioSource; anti-dsRNA (clone K1) from English & Scientific Consulting Bt.; anti-IFN-b-FITCconjugated from PBL Interferon Source; anti-puromycin (clone 2G11, mouse IgG1) has been previously described [47] . Alexaconjugated secondary antibodies (30 min staining) were from Molecular Probes (Invitrogen). Coverslips were mounted on a slide and images taken with a laser-scanning confocal microscope (LSM 510; Carl Zeiss MicroImaging) using a 636 objective and accompanying imaging software. When PKR WT and PKR 2/2 were infected with CHIKV, protocol was performed as follows: cells were fixed with 4% paraformaldehyde in PBS for 20 min, then permeabilized for 30 min in 0.1% Triton 100X (Sigma) and blocked in 10% of normal goat serum (Vector Laboratories). Cells were stained with a mouse monoclonal antibody directed against CHIKV capsid coupled to Alexa-488 and a mouse antibody against puromycin coupled to Alexa-555 and a rabbit antibody anti-eIF2a[pS 52 ] (Invitrogen) and a Cyanin-3 secondary antibody, and finally counterstained with Hoechst (Vector Lab). Cells were observed with an AxioObserver microscope (Zeiss). Pictures and Z-stacks were obtained using the AxioVision 4.5 software. ELISA IFN-b and IL-6 quantification in culture supernatant was performed using the Mouse Interferon Beta ELISA kit (PBL InterferonSource) and Mouse Interleukin-6 ELISA kit (eBioscience) respectively, according to manufacturer instructions. Total RNA was isolated from cells using the RNeasy miniprep kit (QIAGEN) combined with a DNA digestion step (RNase-free DNase set, QIAGEN). cDNA was synthesized using the Superscript II reverse transcriptase (Invitrogen) and random hexamer primers. Quantitative PCR amplification was carried out using complete SYBR Green PCR master mix (Applied Biosystems) and 200 nM of each specific primer. 5 ml of cDNA template was added to 20 ml of PCR mix, and the amplification was tracked via SYBR Green incorporation by an Applied Biosystems thermal cycler. cDNA concentration in each sample were normalized by using HPRT. A nontemplate control was also routinely performed. The primers used for gene amplification (designed with Primer3 software) were the following: mRNA isolation from total RNA was performed with oligodT columns (Genelute mRNA miniprep kit (Sigma). Data were analyzed using the 7500 Fast System Appled Biosystems software. RNA integrity upon poly I:C stimulation was measured by capillary electrophoresis using the the Agilent RNA 6000 Pico Chip kit (Agilent Technologies) in an Agilent 2100 Bioanalyser, according to manufacturer instructions. GADD34 DC/DC and the corresponding WT control MEFs were infected at a multiplicity of infection (MOI) of 10 or 50 with CHIKV-GFP generated using a full-length infectious cDNA clone provided by S. Higgs [71] . By 24 h and 48 h post infection, 30 000 cells were analyzed in triplicate by FACS for expression of GFP. At the same time-points, culture supernatants were collected and IFN-b protein assessed by ELISA. In experiments with exogenous IFN-b, cells were treated with mouse IFN-b (PBL InterferonSource) for 3 h before infection with CHIKV-GFP. When guanabenz was used to specifically inhibit GADD34, MEFs cells were treated for 2 h with 10 mM of Guanabenz or DMSO and then infected in the same medium. Three hours post infection the inoculum was removed and fresh medium with Guanabenz or DMSO was added and maintained all along the experiment. RNAi for GADD34 was performed as described in [60] . FVB WT mice were obtained from Charles River Laboratories (France). GADD34 DC/DC FVB mice were obtained from L. Wrabetz (Milan). Mice were anesthetized and inoculated via the intradermal route with 10 6 PFU of CHIKV-21 isolate [72] . Viral titers in tissues and serum were determined as described before [59] , and expressed as tissue cytopathic infectious dose 50 (TCID50)/g or TCID50/ml, respectively. Organs including heart, liver, skeletal muscles and spleen were collected for histopathological procedures. organs were then fixed in 4% paraformalde-hyde solution, paraffin-embedded, sectioned coronally in 5-10 mm thickness and stained with hematoxylin-eosin. Figure S1 Poly I:C stimulation induces protein translation inhibition and IFN-b production in NIH3T3 cells. A) Protein synthesis was quantified in poly I:C-stimulated NIH3T3 using puromycin labeling followed by immunoblot with antipuromycin mAb 12D10. Protein synthesis was strongly reduced upon poly I:C stimulation. Immunoblot for phosphorylated (P-eIF2a) and total eIF2a were performed on the same NIH3T3 extract. Cycloheximide (chx) was added 5 min before puromycin incorporation. b-actin immunoblot is shown for equal loading control. B) Puromycin integration was analysed by immunofluorescence in NIH3T3 cells treated for 4 h with poly I:C and labeled with puromycin in the last 10 min. Figure S2 Protein translation in cells with non-phosphorylatable eIF2a. Protein synthesis was quantified in MEFs with non-phosphorylatable eIF2a, eIF2aA/A and the corresponding control cells, eIF2aS/S. After poly I:C or thapsigargin treatment, puromycin labeling followed by immunoblot, was performed. Puromycin labeling was quantified with ImageJ software and protein translation was depicted as percentage of steady state. Cycloheximide (chx) was added 5 min before puromycin incorporation. Tubulin immunoblot is shown for equal loading control. One of two independent experiments with similar results is shown. (TIF) Figure S3 GADD34 mRNA induction in ATF4-deficient MEFs stimulated with cytosolic poly I:C. The levels of GADD34 transcript were determined by qPCR in WT and ATF4 2/2 cells after 8 h of poly I:C stimulation. Treatment with tunicamycin and thapsigargin were used as positive controls for GADD34 induction. Results are displayed according to both WT internal reference (left) and ATF4 2/2 internal reference (right). (TIF) Figure S4 GADD34 mediates eIF2a dephosphorylation in MEFs stimulated with poly I:C. A) Wild-type and GADD34 DC/DC MEFs were treated for the indicated times with poly I:C (pI:C), tunicamycin (tun) or thapsigargin (th) and eIF2a phosphorylation was monitored by immunoblot. B) GADD34 expression was analyzed by immunoblot in samples treated for 1 or 6 hours with poly I:C alone or together with tunicamycin (tun) or thapsigargin (th). Data shown in (A) and (B) are representative of three independent experiments with similar results. (TIF) Figure S5 RNA integrity upon poly I:C exposure. WT MEFs were treated with poly I:C for the indicated times and RNA integrity evaluated by capillary electrophoresis (Agilent RNA 6000). RNA Integrity Numbers (RIN) between 8.2 and 9.2 were obtained, indicating a high level of RNA integrity. Data shown are representative of three independent experiments with similar results. (TIF) Figure S6 UPR-inducing drugs do not elicit IFN-b production. Cell culture supernatants of murine embryonic fibroblasts were tested for the presence of IFN-b, after treatment with poly I:C (8 h), tunicamycin and thapsigargin (6 h). The results shown are representative of 4 experiments. (TIF) Figure S7 Deletion of the constitutively-expressed PP1 co-factor, CReP, does not impact protein translation and IFN-b production in MEFs. A) WT and CReP 2/2 MEFs were treated with poly I:C (pI:C) for the indicated times and the levels of P-eIF2a and PKR were analyzed by immunoblot. Although basal levels of P-eIF2a were higher in CReP 2/2 MEFs, increase of phosphorylation upon poly I:C exposure was similar to the WT. PKR expression upon poly I:C treatment was equivalent in CReP 2/2 and WT MEFs. B) Protein synthesis was quantified using puromycin labeling followed by immunoblot with the antipuromycin mAb 12D10. Where indicated, cells were treated with cycloheximide (chx) 5 min before puromycin incorporation. No major differences were found between WT and CReP 2/2 cells at the level of translation inhibition following poly I:C exposure.

Targets and Intracellular Signaling Mechanisms for Deoxynivalenol-Induced Ribosomal RNA Cleavage

The trichothecene mycotoxin deoxynivalenol (DON), a known translational inhibitor, induces ribosomal RNA (rRNA) cleavage. Here, we characterized this process relative to (1) specific 18S and 28S ribosomal RNA cleavage sites and (2) identity of specific upstream signaling elements in this pathway. Capillary electrophoresis indicated that DON at concentrations as low as 200 ng/ml evoked selective rRNA cleavage after 6 h and that 1000 ng/ml caused cleavage within 2 h. Northern blot analysis revealed that DON exposure induced six rRNA cleavage fragments from 28S rRNA and five fragments from 18S rRNA. When selective kinase inhibitors were used to identify potential upstream signals, RNA-activated protein kinase (PKR), hematopoietic cell kinase (Hck), and p38 were found to be required for rRNA cleavage, whereas c-Jun N-terminal kinase and extracellular signal-regulated kinase were not. Furthermore, rRNA fragmentation was suppressed by the p53 inhibitors pifithrin-α and pifithrin-μ as well as the pan caspase inhibitor Z-VAD-FMK. Concurrent apoptosis was confirmed by acridine orange/ethidium bromide staining and flow cytometry. DON activated caspases 3, 8, and 9, thus suggesting the possible coinvolvement of both extrinsic and intrinsic apoptotic pathways in rRNA cleavage. Satratoxin G (SG), anisomycin, and ricin also induced specific rRNA cleavage profiles identical to those of DON, suggesting that ribotoxins might share a conserved rRNA cleavage mechanism. Taken together, DON-induced rRNA cleavage is likely to be closely linked to apoptosis activation and appears to involve the sequential activation of PKR/Hck →p38→p53→caspase 8/9→caspase 3.

The trichothecenes, a group of sesquiterpenoid mycotoxins produced by Fusarium that contaminate wheat, barley, and corn globally, are problematic because of their resistance to degradation during processing and their potential to adversely affect human and animal health (Pestka, 2010) . Among the over 200 trichothecenes discovered to date, deoxynivalenol (DON) is most frequently encountered in food, and human exposure to this toxin throughout the world has been well documented in a recent series of elegant biomarker studies (Hepworth et al., 2012; Turner et al., 2010a Turner et al., ,b, 2011 . A potential target for adverse health effects for DON is the innate immune system, with low doses of the toxin having immunostimulatory effects and high doses causing immunosuppression (Pestka, 2010) . At the mechanistic level, DON has been shown in vitro and in vivo to activate mitogen-activated protein kinases (MAPKs), including p38, c-Jun N-terminal kinase (JNK), and extracellular signal-regulated kinase (ERK) in macrophages and monocytes (Islam et al. 2006; Shifrin and Anderson, 1999; Zhou et al., 2003a) . These MAPKs mediate upregulation of proinflammatory cytokine and chemokine expression as well as apoptosis (Chung et al., 2003; Islam et al., 2006; Moon and Pestka, 2002) . Notably, DON concentration dependently induces competing survival (ERK/AKT/p90Rsk/Bad) and apoptotic (p38/p53/Bax/ mitochondria/caspase-3) pathways in the macrophage (Zhou et al., 2005a) . Accordingly, MAPK activation is critical to both stimulation and suppression of the innate immune system. Two signal transducers that have been identified to be upstream of DON-induced MAPK activation are doublestranded RNA-(dsRNA) activated protein kinase (PKR) (Zhou et al., 2003b) and hematopoietic cell kinase (Hck) (Zhou et al., 2005b) . PKR is a widely distributed constitutively expressed serine/threonine protein kinase that can be activated by dsRNA, interferon, proinflammatory stimuli, cytokines, and oxidative stress (Garcia et al., 2006; Williams, 2001) . PKR also activates p53, p38, JNK, Nuclear Factor-kappaB, signal transducer and activator of transcription, and interferon regulatory factor-1 (Williams, 1999) . Hck, a member of Src kinase family, is expressed specifically in myelomonocytic cell lineages and transduces extracellular signals that regulate proliferation, differentiation, and migration (Ernst et al., 2002; Tsygankov, 2003) . Upon DON exposure, both PKR and Hck are activated prior to the MAPKs and their respective inhibitors suppress downstream MAPK activation (Zhou et al., 2005b) . Although crosstalk between PKR and Hck is not clearly understood, PKR appears to be required for Hck interaction with the ribosome in the human monocyte U937 cell line (Bae et al., 2010) . A prominent consequence of DON exposure in macrophages is ribosomal RNA (rRNA) cleavage, which could negatively impact innate immune function (Li and Pestka, 2008) . Two fundamental questions arising from this work relate to the specificity and mechanisms for this cleavage. Using oligonucleotide extension, we identified one site of DON-induced rRNA cleavage is within the peptidyl transferase center of 28S rRNA; however, further identification of additional sites was precluded by the relatively poor resolution of conventional gel electrophoresis and limitations on this techniques imposed by RNA hairpin structures (Li and Pestka, 2008) . Regarding mechanisms, it is known that ricin and other ribosomeinactivating proteins (RIPs) cleave rRNA via a highly specific mechanism that involves N-glycosidase-mediated adenine depurination at highly conserved sarin/ricin (S/R) loop (Endo and Tsurugi, 1986; Hartley and Lord, 2004) . However, DON and other trichothecenes are low molecular weight chemicals and are devoid of inherent enzyme activities (Li and Pestka, 2008) . Induction of rRNA cleavage by chemicals or viruses has previously been suggested to be linked to apoptosis (Banerjee et al., 2000; Naito et al., 2009) raising the possibility that DONinduced rRNA cleavage might be similarly linked to this mechanism of cell death. The purpose of this study was to characterize DON-induced rRNA fragmentation with respect to (1) specific sites of 18S and 28S rRNA cleavage and (2) identity of upstream signaling events that mediate this process. The results demonstrate that DON promotes cleavage of 18S rRNA and 28S rRNA into a minimum of five and six fragments, respectively. Furthermore, DON-induced rRNA cleavage occurred concurrently with apoptosis was mediated by sequential activation of PKR, Hck, p38, p53, and several caspases. Three other translational inhibitors, satratoxin G (SG), anisomycin, and ricin, evoked comparable selective rRNA cleavage to DON, suggesting that a common mechanism might exist for other ribotoxins. Chemicals. DON, anisomycin, PKR inhibitor C-16, and Hck inhibitor PP1 were purchased from Sigma-Aldrich (St Louis, MO). Ricin was obtained from Vector Labs Inc. (Burlingame, CA). SG was purified as described previously (Islam et al., 2009) . The p38 inhibitor SB 203580, JNK inhibitor SP600125, ERK inhibitor PD 98059, RNase L Activator, p53 inhibitor pifithrin-a, and pifithrin-l were purchased from EMD Chemicals Inc. (Gibbstown, NJ). Pan caspase inhibitor Z-VAD-FMK was supplied by BD Biosciences (San Diego, CA). [c-32 P]ATP was purchased from PerkinElmer. All other chemicals and media components were obtained from Sigma-Aldrich, except where noted. Cell culture. RAW 264.7 cells (ATCC, Rockville, MD) were cultured in Dulbecco's modified Eagle's medium supplemented with 10% (vol/vol) heatinactivated fetal bovine serum (Atlanta Biologicals, Lawrenceville, GA), streptomycin (100 lg/ml), and penicillin (100 U/ml) at 37°C in a humidified incubator with 5% CO 2 . Cell number and viability were assessed by trypan blue dye exclusion using a hematocytometer. Prior to exposure of toxins (DON, SG, anisomycin, and ricin) or inhibitors, cells (2.5 3 10 6 ) were seeded and cultured in 100-mm tissue culture plates for 24 h to achieve approximately 80% confluency. RNA purification and fragmentation analysis. RNAs were extracted by TRIZOL (Invitrogen, Carlsbad, CA) following the manufacturer's protocol and their concentrations were measured using a NanoDrop reader (ThermoFisher, Wilmington, DE). Fragmentation of RNA was assessed by denaturing agarose gel electrophoresis (Li and Pestka, 2008) or capillary electrophoresis using an Agilent 2100 Bioanalyzer with a NanoChip (Agilent, Santa Clara, CA) according to manufacturer's instructions. Northern blot analysis. Northern blot analysis was performed by a modification of a previously described procedure (Li and Pestka, 2008) . Briefly, probes (Table 1) were labeled with c-32 P by DNA 5# end-labeling system (Promega, Madison, WI) and purified by Micro-Bio-Spin6 chromatography column (Bio-Rad, Hercules, CA). Label incorporation was measured using a TopCount NXT (PerkinElmer, Shelton, CT). Total RNA (10 lg per lane) was separated on a 1.2% (wt/vol) formaldehyde denaturing agarose gel and transferred to a Biodyne membrane (Pall Gelman Laboratory, Ann Arbor, MI). After ultraviolet (UV) crosslinking (Stratagene, Cedar Creek, TX), membranes containing immobilized RNA were prehybridized for 1 h at 68°C and then incubated with [c-32 P]-labeled probes (1 3 10 6 cpm/ml) in Quickhyb solution (Stratagene) containing 200 lg/ml of herring sperm DNA at 68°C for 2 h. Blots were washed twice with 23 Side Scatter (SSC) containing 0.1% (wt/ vol) SDS at room temperature and once for 15 min with 0.13 SSC containing 0.1% (wt/vol) SDS at 50°C. The membranes were assembled with the Hyblot autoradiography film (Denville, Metuchen, NJ) into an X-ray exposure cassette and the film was developed after 24 h. Immunoblotting. Western blot analyses were conducted using primary antibodies specific for murine forms of total/cleaved caspase 9, cleaved caspase 3 (Asp 175), total caspase 8, and cleaved caspase 8 (Asp387) (Cell Signaling, Beverly, MA). Mouse b-actin antibody (Sigma) was also used to verify equal loading. Cells were washed twice with ice-cold phosphate-buffered saline (PBS), lysed in boiling lysis buffer (1% [wt/vol] SDS, 1 mM sodium orthovanadate, and 10 mM Tris, pH 7.4), boiled for 5 min, and sonicated briefly, the resultant lysate centrifuged at 12,000 3 g for 10 min at 4°C and protein concentration measured with a BCA Protein Assay Kit (ThermoFisher, Pittsburgh, PA). Total cellular proteins (40 lg) were separated on Bio-Rad precast 4-20% polyacrylamide gels (Bio-Rad) and transferred to a polyvinylidene difluoride membrane (Amersham, Arlington Heights, IL). After incubating with blocking buffer (Li-Cor, Lincoln, NE) for 1 h at 25°C, membranes were incubated with murine and/or rabbit primary antibodies (1:1000 dilution in Li-Cor blocking buffer) to immobilized proteins of interest overnight at 4°C. Blots were washed three times of 10 min with Tris-buffered saline and Tween 20 (50 mM Tris-HCl, 150 mM NaCl, and 0.1% Tween 20, pH 7.5) and then incubated with secondary IRDye 680 goat anti-rabbit and/or IRDye 800CW goat anti-mouse IgG antibodies (Li-Cor) (1:2000 dilution in Li-Cor blocking buffer) for 1 h at 25°C. After washing three times, infrared fluorescence from these two antibody conjugates were simultaneously measured using a Li-Cor Odyssey Infrared Imaging System. Apoptosis measurement by acridine orange/ethidium bromide staining. Acridine orange/ethidium bromide (AO/EB) staining was carried out using an adaptation of a previously described procedure (Muppidi et al., 2004) . Briefly, slides were cleaned and sterilized by UV light, placed into 100-mm tissue culture plates, and cultured with RAW 264.7 cells (2.5 3 10 6 ) for 24 h to achieve approximately 80% confluency. DON-exposed RAW 264.7 cells were stained for 2 min with 100 lg/ml AO and 100 lg/ml EB in PBS. The slides were washed twice with cold PBS and then covered with coverslip and examined at 3400 under Nikon fluorescence microscope equipped with a wide-band fluorescein isothiocyanate filter. Cells ( 200) were classified based on their nuclear morphology (bright chromatin, highly condensed, or fragmented nuclei) in to four categories: viable normal (VN), DON-INDUCED RRNA CLEAVAGE viable apoptotic (VA), nonviable apoptotic (NVA), and nonviable necrotic (NVN). The apoptotic index was calculated as follows: Apoptosis measurement by flow cytometry. RAW 264.7 cells were treated with vehicle PBS or DON (1000 ng/ml) for 6 h. Total DON-treated cells and adherent DON-treated cells were collected separately. After washing twice with cold PBS, annexin staining was performed according to the manufacturer's protocol (BD Biosciences). Briefly, cells were resuspended to 13 binding buffer at a concentration of 2.5 3 10 6 cells/ml, and 100 ll of the cell suspension was transferred to fluorescence activated cell sorter tubes. Cells were incubated with 5 ll of annexin V and propidium iodide (PI) for 15 min at room temperature in dark. Finally, samples were resuspended to 400 ll volume with binding buffer, acquired on the Accuri C6 Flow Cytometer (Ann Arbor, MI), and analyzed using FlowJo software (Tree Star, Ashland, OR). Live Raw 264.7 cells were gated using Forward Scatter versus SSC. Statistics. Data were analyzed by Student's t-test using Sigma Stat 3.11 (Jandel Scientific, San Rafael, CA). Data sets were considered significantly different when p < 0.05. The capacity of DON to induce rRNA cleavage in RAW 264.7 macrophages was assessed by denaturing gel electrophoresis (Fig. 1A) and capillary electrophoresis (Figs. 1B and 1C). Distinct 18S and 28S rRNA bands were evident in both control and DON-treated RAW 264.7 cells, whereas over five additional rRNA fragment bands were detected in DON-treated cells. Based on its high resolution and reproducibility as compared with conventional electrophoresis, capillary electrophoresis was employed to monitor cleavage in subsequent experiments. When the kinetics of the response were measured, DON at 1000 ng/ml was found to cause rRNA cleavage as early as 2 h that was very robust by 6 h (Figure 2A ). Cleavage at 6 h was concentration dependent with 200 ng/ml of DON evoking modest rRNA cleavage as compared with more marked rRNA cleavage induced by 1000 ng/ml ( Figure 2B ). Based on these findings, incubation with DON at 1000 ng/ml for 6 h was uniformly employed for subsequent mechanistic studies. Northern blot analyses using 32 P-labeled oligonucleotide probes (Table 1) for 18S and 28S rRNAs were performed to identify and map sites of rRNA cleavage. Incubation with five probes complementary to 28S rRNA revealed six fragments (a-f) with sizes approximating 4000, 3200, 2800, 1. Detection of DON-induced rRNA cleavage in RAW 264.7 by agarose gel and capillary electrophoresis. Cells were treated with or without 1000 ng/ml DON for 6 h. RNAs were purified and analyzed either (A) on 1.2% formaldehyde denaturing agarose gel (10 lg) or (B and C) by capillary electrophoresis (300 ng). The x-axis indicates the size of the fragments in nucleotide (nts) and the y-axis indicates the relative peak intensity in fluorescence units (FU). The two major peaks represent 18S rRNA (~2000 nts) and 28S rRNA (~4000 nts). Arrows designate three significant cleavage peaks between 28S and 18S rRNA and two peaks below 18S rRNA. Rectangle in C indicates chart region to be shown in subsequent figures. 384 HE, ZHOU, AND PESTKA 1500, 1000, and 500 nts (Fig. 3A) . Based on rRNA fragment sizes, probe position and Northern blotting, putative cleavage sites of 28S rRNA were deduced ( Figure 3B ). DON appeared to cleave 28S rRNA into one of two pairs of fragments: a þ e (approximately 5000 nts) and b þ d (approximately 4700 nts). Because no complementary fragment (approximately 2000 nts) to fragment c was detected, it was likely to be a product of the subsequent degradation of a or b. Autoradiography following hybridization with three probes complementary to 18S rRNA indicated the presence of five fragments (A-E) with approximate sizes of 1000, 800, 600, 500, and 300 nts (Fig. 4 A) . The putative 18S rRNA cleavage pattern suggested that fragments A (approximately 1000 nts) and B (approximately 800 nts) are derived from intact 18S rRNA, whereas the remaining fragments (C, D, and E, approximately 600, 500, and 300 nts, respectively) resulted from further secondary cleavage of the two primary fragments (C and E from A and D from B) (Fig. 4B) . AO/EB staining of adherent RAW 264.7 cells revealed that DON exposure (1000 ng/ml, 6 h) induced significant apoptosis concurrent with rRNA degradation (Fig. 5A) . Because RAW 264.7 cells attach to the bottom of the cell culture plate under normal physiological conditions but detach during apoptosis, flow cytometry was additionally employed to measure apoptosis in adherent and suspended cells following DON treatment (Fig. 5B) . Both adherent and total populations contained markedly higher annexin V-positive cells (early apoptotic) (60 and 78%, respectively) than the control total cells (31%). Similarly, the percentage of annexin V/PI-double positive cells (late apoptotic) were much higher in adherent (4.8%) and total (10.8%) populations than that of the control (1.2%). Because DON-induced apoptosis involves activation of PKR, Hck, and MAPKs (Zhou et al., 2003b; Zhou et al., 2005b) , the role of these kinases in rRNA cleavage was determined using selective inhibitors. Both the PKR inhibitor C-16 (0.1 and 0.3 lM) (Fig. 6A ) and the Hck inhibitor PP1 (5 and 25 lM) (Fig. 6B) were found to concentration dependently suppress DON-induced rRNA cleavage. Although the p38 inhibitor (SB 203580; 1 and 5 lM) also inhibited DON-induced rRNA cleavage concentration dependently (Fig. 6C) , inhibitors of JNK (SP600125; 0.2, 1, and 5 lM) and ERK (PO 98059; 20 and 100 lM) did not (data not shown) have suppressive effects. Thus, as has been observed for DONinduced apoptosis, PKR, Hck, and p38 appeared to be key upstream elements for DON-induced rRNA cleavage. It was previously shown that p38 mediates the sequential activation of p53 and caspase 3 to induce apoptosis in RAW 264.7 cells (Zhou et al., 2005a) . Suppression of DON-induced rRNA cleavage was observed here both for pifithrin-a (80 and 100 lM) (Fig. 6D) , which can reversibly inhibit p53-dependent transactivation of p53-responsive genes and apoptosis and for pifithrin-l (10 and 25 lM) (Fig. 6E) , which blocks p53 interaction with Bcl-2 family proteins and selectively inhibits p53 translocation to mitochondria. The caspase inhibitor Z-VAD-FMK also caused concentration-dependent inhibition of DON-induced rRNA cleavage (Fig. 6F) . Accordingly, both p53 and caspase activation are additional upstream elements in the signaling pathway leading to DON-induced rRNA cleavage. Extrinsic and intrinsic apoptotic pathways activate caspase 3 through caspases 8 and 9, respectively. To discern possible contributions of these two pathways, RAW 264.7 cells were treated with DON for 3 and 6 h and the presence of cleaved caspases 8, 9, and 3 was determined by Western blot analysis. Cleavage of caspases 3, 8, and 9 was observed (Figs. 7A and 7B), suggesting that extrinsic and intrinsic pathways could potentially be involved in DON-induced rRNA fragmentation (Fig. 8) . Evoke rRNA Cleavages Lipopolysaccharide, a major component of the outer membrane of Gram-negative bacteria that can activate macrophages via the Toll-like receptor 4 receptor, did not affect rRNA integrity indicating that the macrophage activation per se was insufficient to induce RNA cleavage. Because DON is a translational inhibitor and causes rRNA cleavage indirectly, we questioned whether this might be a common effect for other ribotoxins. Cells were incubated with SG and anisomycin, which can freely diffuse through cell membrane and ricin, a RIP, which can enter the cells by endocytosis and retrograde translocation to ER and cytosol. Unlike SG and anisomycin that directly bind to ribosome to inhibit translation, ricin possesses inherent RNA N-glycosidase activity to depurinate RNA. Although these toxins have apparently different mechanisms for inhibiting translation, SG, anisomycin, and ricin induced a similar apoptosis-mediated cleavage profile to that of DON (Fig. 8) , suggesting that ribotoxins might share conserved pathway for mediating rRNA cleavage. Understanding how DON induces rRNA degradation will provide insight into the mechanisms by which trichothecenes and other ribotoxins incapacitate or kill immunocompetent cells. The results presented here establish for the first time that (1) DON-induced cleavage at a limited number of sites in both 18S rRNA and 28S rRNA, (2) these cleavages result from PKR-driven p38 activation, (3) these events closely parallel induction of apoptosis by this mycotoxin, and (4) DON's effects are shared with other ribotoxic agents. In higher eukaryotes, 28S rRNA contains 12 conserved but variable divergent domains (D1-D12), originating from evolutionary large-scale length and diversity expansions (Michot et al., 1984) . Although the functions of D domains are not clearly understood, D2 and D8 have higher divergency rates than other domains (Houge et al., 1995) . Mouse 28S rRNA cleavage sites have been mapped to be within D2 (approximately 400-1140 nts) and/or D8 domains (approximately 2700-3280 nts) (Houge et al., 1993; Houge et al., 1995; Houge and Doskeland, 1996; Naito et al., 2009) . Apoptosisassociated cleavage pathways were previously reported to target D2 and D8 followed by secondary cleavage in other domains. As demonstrated here, DON-induced rRNA cleavage fragments (b-f) were from the D8 region, whereas two other cleavage sites (a and e) were within D10 (approximately 3785-3822 nts) (Michot et al., 1984) rather than canonical D2. It might be speculated that DON treatment rendered D8 and D10 accessible to constitutive and/or inducible RNases, whereas D2 did not. We have previously reported that upregulated expression of several RNases in DON-treated RAW 264.7 cells occurs prior to or simultaneously with rRNA degradation (Li and Pestka, 2008) . Particularly notable was RNase L, an enzyme that can mediate inhibition of protein synthesis, apoptosis induction, and antiviral activity (Stark et al., 1998) . Although RNase L is activated by a number of stressors, its only known endogenous direct activator is 2-5A, a product of oligoadenylate synthetase (Pandey et al., 2004; Liang et al., 2006) . Upon binding to 2-5A, RNase L is activated by dimerization and possibly inhibits protein synthesis via the degradation of both messenger RNA and 28S rRNA (Clemens and Vaquero, 1978; Wreschner et al., 1981) . We found that under cell-free conditions in the presence of 2-5A, RNase L readily cleaved both FRET probe and purified rRNA; however, it did not degrade rRNA in The observation that p38 but not JNK mediated DONinduced rRNA cleavage is consistent with previous observations that p38 is a central regulator of cell fate in DON-exposed macrophages (Zhou et al., 2005a) . Our findings further demonstrate that PKR and Hck also mediate DON-induced rRNA cleavage. Although crosstalk between PKR and Hck is still not completely understood, these kinases are wellestablished upstream mediators of DON-induced p38 activation (Zhou et al., 2003b (Zhou et al., , 2005b . These data are consistent with our previous observations that 40S ribosome subunit-associated PKR, Hck, and p38 are activated upon DON exposure (Bae et al., 2010) . We have previously hypothesized that rRNA cleavage might yield a double-stranded (ds) hairpin fragments capable of activating ribosome-associated PKR through its dsRNAbinding sites (Li and Pestka, 2008) . The data presented here do not support this hypothesis. First, PKR and Hck are activated by DON within minutes (Zhou et al., 2003b (Zhou et al., , 2005b , whereas DON-induced rRNA cleavage was detectable only after 2 h and required relatively higher DON concentrations. Second, suppression of DON-induced rRNA cleavage by PKR, Hck, and p38 inhibitors suggests that their activation is an upstream rather than downstream event. We therefore propose an alternative hypothesis in which entry of DON into the cell rapidly disrupts the conformation of rRNA yielding accessible intact hairpin loops at the surface of ribosome that would be capable of activating PKR or other double-stranded RNAbinding proteins. These events would initiate a stress response that would, especially at high toxin concentrations, ultimately drive rRNA cleavage in concert with apoptosis. This hypothesis is highly consistent with the growing recognition that damage-associated molecular patterns (DAMPs) initiate noninfectious innate immune and inflammatory responses (Newton and Dixit, 2012) . Bulavin et al. (1999) have previously shown that p38 mediates p53 phosphorylation. It was thus notable that rRNA cleavage was found here to be inhibited by pifithrin-a, a compound that reversibly inhibits both p53-dependent transactivation of p53-responsive genes and apoptosis, as well as by pifithrin-l, which blocks p53 interaction with Bcl-2 family proteins and selectively inhibits p53 translocation to mitochondria. Relatedly, we have previously demonstrated that DON induces translocation of BAX (a member of Bcl-2 family) to mitochondria and release of cytochrome c leading to apoptosis (Zhou et al., 2005a) . Finally, DON-induced rRNA cleavage was completely suppressed by the pan caspase inhibitor Z-VAD-FMK, indicating that activation of caspases was a prerequisite of rRNA cleavage. Caspases, proteolytic enzymes that play important roles in inflammation and cell death, are independently activated via either an intrinsic pathway involving caspase 9 or an extrinsic pathway through caspase 8 (Fuentes-Prior and Salvesen, 2004) . It is thus important to note that DON activated caspases 8 and 9, both of which can activate caspase 3, suggesting the involvement of both apoptotic pathways, respectively, in rRNA cleavage. Caspases mediate degradation of some ribosomalassociated proteins, including eukaryotic translation initiation factor (eIF)2a, eIF3/p35, eIF4B, and eIF4G family (Clemens et al., 2000) . This proteolytic action might expose ribosome- FIG. 7 . DON induces cleavage of caspases 3, 8, and 9 in RAW 264.7. Cells were treated with DON (1000 ng/ml) for 3 and 6 h. Western blotting was used to detect (A) caspase 9, cleaved caspase 9, and cleaved caspase 3 and (B) caspase 8 and cleaved caspase 8. b-Actin was used as loading control. Data are representative of three separate experiments. FIG. 8. SG, anisomycin, and ricin but not lipopolysaccharide induce rRNA cleavage patterns identical to DON in RAW 264.7. Cells were treated with DON (1000 ng/ml), SG (10 ng/ml), anisomycin (25 ng/ml), ricin (300 ng/ml), and lipopolysaccharide (10 lg/ml) for 6 h. RNAs were purified and analyzed by capillary electrophoresis. Results are representative of three separate experiments. 388 HE, ZHOU, AND PESTKA embedded rRNA to constitutive and inducible RNases, resulting in the highly selective cleavage observed here. The DON concentrations used in this study (200-1000 ng/ ml) are consistent with those predicted to be in plasma and tissues (e.g., spleen, liver, and small intestine) of mice treated with immunosuppressive doses ( 2 mg/kg bw) of this toxin (Azcona-Olivera et al., 1995; Li et al., 2005 Li et al., , 2007 . Turner et al. (2010) estimated daily intake of DON in male Normandy farmers to range between 27 and 1088 ng/kg/day. Thus, the concentrations employed in the present study might only likely be encountered in humans that have consumed a large bolus dose of DON in a heavily contaminated cereal. The results presented here and previously demonstrate that DON-induced rRNA cleavage requires prior sequential activation of PKR/Hck, p38, p53, and caspases (Fig. 9) . It was remarkable that ricin, which possesses N-glycosidase enzymatic activity for 28S rRNA depurination, induced the identical rRNA cleavage profile to that of DON and two other low molecular weight translational inhibitors, SG and anisomycin. Possibly, in all four cases, toxin-mediated ribosome damage activated a canonical apoptosis-associated rRNA cleavage pathway. Future work should focus on understanding how DON mediates PKR activation, identifying critical executing caspases and RNases, ascertaining whether induction of rRNA cleavage by other ribotoxins or apoptosis-inducing agents is mediated by the same intracellular signaling pathways as well as validating these findings in animal models. Supplementary data are available online at http://toxsci. oxfordjournals.org/. FUNDING National Institutes of Health (PHS grant ES003358); United States Department of Agriculture (grant 59-0206-9-058).

Pilot Evaluation of RT-PCR/Electrospray Ionization Mass Spectrometry (PLEX-ID/Flu assay) on Influenza-Positive Specimens

The PLEX-ID/Flu assay has been recently developed to enable the detection and typing of influenza viruses based on the RT-PCR/electrospray ionization mass spectrometry technology. This novel assay was evaluated for typing performance on 201 positive influenza A or B nasopharyngeal swab specimens (NPS) detected by real-time RT-PCR during the 2010-2011 season. The PLEX-ID/Flu assay detected and characterized 91.3% and 95.3% of all influenza A and B samples, respectively. All non-typeable influenza A and B specimens by the assay showed low viral loads with threshold cycle values ≥ 33. Taken together, and although our results need to be confirmed by further prospective studies, the PLEX-ID/Flu assay detected positively and gave a typing result for 93% of all NPS detected positively by real-time RT-PCR, thus suggesting a potential role for influenza virus surveillance among other techniques.

The World Health Organization (WHO) Global Influenza Surveillance and Response System provides essential information on the types and variants of influenza viruses circulating worldwide. Direct detection and identification of influenza strains from clinical samples are mostly performed by using specific real-time PCR assays and, less frequently, by culture-based methods [1, 2] . Genotyping or phenotyping characterization of positive cases is performed by sequencing classical PCR amplicons (requiring multiple primer pairs), followed by phylogenetic analysis, or hemagglutinin inhibition assay, respectively. The influenza A(H1N1)pdm pandemic in 2009 resulted in the development of a number of assays, such as single/multiplex real-time RT-PCR [3] [4] [5] [6] or microarray systems [7, 8] , allowing the rapid detection and typing of influenza virus in human specimens. Sensitive and rapid diagnostic or typing assays are essential for appropriate patient management, particularly high risk patients, and the use of appropriate antiviral therapy. RT-PCR/electrospray ionization mass spectrometry (ESI-MS) assays were developed recently to enable the potential detection and typing of microbial agents, including influenza [9] [10] [11] [12] [13] [14] , during the same flow procedure [15] [16] [17] . This technology relies on the analysis of nucleotide base *Address correspondence to this author at the Laboratory of Virology, Division of Infectious Diseases, University of Geneva Hospitals, 4 Rue Gabrielle-Perret-Gentil, 1211 Geneva 14, Switzerland; Tel: ++41 22 3724079; Fax: ++41 22 3724097; E-mail: samuel.cordey@hcuge.ch composition signatures of highly variable selected regions based on the measurement of the molecular weight of PCR amplicons [12, 18] . A first version of the influenza assay (Abbott Molecular, Des Plaines, IL, USA) performed on the T5000 instrument (Ibis/Abbott, Carlsbad, CA, USA) was validated for influenza A (including A(H3N2) and H5N1) and B isolates collected between 1999 and 2006 [13] . Sensitivity and specificity were estimated to reach 97% and 98%, respectively. ESI-MS analysis was also capable of identifying viral reassortments or co-infections (i.e., a mixed population). More recently, this assay reported a sensitivity and specificity of 94.1% and 97%, respectively, for A(H1N1)pdm detection [11] . Since then, the assay has been redesigned and updated. This "PLEX-ID/Flu assay" includes one pan-influenza primer set targeting the PB1 segment, five pan-influenza-A primer pairs targeting individually NP, M1, PA, PB2 and NS1 genes, one pan-influenza B primer pair targeting the PB2 segment, and two additional primer pairs targeting two surface antigens HA (H1) and NA (N1) genes [12] . The PLEX-ID platform was used to perform a pilot evaluation for the detection and subtyping or lineage characterization of human influenza virus types A and B, respectively. For this, nasopharyngeal swab specimens (NPS) collected from a network of more than 80 practitioners participating actively to the clinical surveillance of influenza cases in Switzerland and screened for influenza during the 2010-2011 season were used as follows. After viral genome extraction using the NucliSENS easyMAG (bioMérieux, Geneva, Switzerland), the following four onestep real-time RT-PCR assays were applied for the routine screening: the CDC pan-influenza A specific real-time RT-PCR assay [19] considered as a reference assay for influenza type A surveillance by the WHO, and three in-house developed assays specific for A(H1N1)pdm (Swine H1 GE), influenza A(H3N2) (A/H3), and influenza B (InfB MP) detection (supplementary Table 1 ) all validated on WHO quality controls. Each week, a batch of 22 or 46 influenzapositive specimens (according to the number of available specimens) were tested with the PLEX-ID/Flu assay in parallel to the usual real-time RT-PCR assays performed by the Swiss National Reference Centre for Influenza. NPS were selected blind of the real-time RT-PCR threshold cycle (C T ) values and the type of influenza (influenza A or B). For each assay, specific positive and negative internal controls were included systematically in each run to rule out any potential PCR inhibitors or contaminations, respectively. and Wisconsin/01/10 strains, respectively). Of note, the PLEX-ID/Flu assay's subtyping or lineage characterization for A(H3N2) and type B specimens, respectively, is less accurate than for A(H1N1)pdm specimens since HA and NA genes are not included in the analysis. Taken together, the PLEX-ID/Flu assay detected positively and gave a typing/lineage characterization result for 93% of all NPS detected positively by real-time RT-PCR. Positive results obtained by both methods were always in agreement. All non-typeable influenza A and B specimens by the PLEX-ID/Flu assay showed relatively low viral loads with C T values 33 obtained with the screening real-time RT-PCR assays. The typing performance of the PLEX-ID/Flu assay versus the Sanger-based sequencing method was then compared for all influenza specimens with low viral loads (C T values 30; 19 influenza A and 18 influenza B-positive NPS; Table 2 ). Briefly, for the Sanger-based sequencing method, reverse transcription was performed in the presence of Uni12w or BUni11w primers updated from [20] and [21] , respectively, provided by Prof. Rod Daniels (MRC, London). Part of the influenza HA-1 gene was then amplified as follows: for influenza A viruses, a first PCR reaction was performed with cswHAF1 and cswHAR1264 primers, followed by two nested PCR reactions with cswHAF31/cswHAR873 and cswHAF451/cswHAR1264 primer pairs (supplementary Table 1 ). For influenza B viruses, BHA1F1 and BHA1R1 primers were used for the first PCR reaction, followed by a nested PCR using the BHA25/BHAF primer pair. Sequencing was performed with ABI Prism 3130XL DNA Sequencer (Applied Biosystems, Rotkreuz, Switzerland) and analyzed with the Geneious program (Biomatters, Auckland, New Zealand). The same 29 negative controls are used in both comparisons (marked *). ND: not done. Among these 37 specimens, the PLEX-ID/Flu assay was able to subtype 9/19 influenza A (47.5%) and characterize the lineage for 14/18 influenza B (77.8%). The Sanger-based sequencing method was able to subtype 13/19 influenza A (68.4%) and identify the lineage for 9/18 influenza B (50%). Therefore, for these selected specimens, the PLEX-ID/Flu assay demonstrates a sensitivity for influenza B virus lineage characterization that is at least similar to the Sanger-based method, which is directly dependent on conditions used for viral genome amplification by classical PCR. However, the latter appears more sensitive for influenza A subtyping under these conditions. This observation can be explained by the PLEX-ID/Flu assay algorithm that requires the analysis of nucleotide base composition signatures of a minimum of three positive PCRs out of eight independent primer pairs present in the assay, with the mandatory coupling of detection and subtyping processes, for influenza A specimen analysis. Therefore, the typing constraint of the PLEX-ID/Flu assay explains partially the higher influenza A subtyping performance of the Sanger-based method for low viral load specimens. Indeed, 9/10 influenza A NPS detected positive by real-time RT-PCR could not be subtyped with the PLEX-ID/Flu assay, although one or two positive PCR signals were observed. Although the minimal requirement of three independent PCR signals for subtyping was not obtained, specific influenza A PCR amplifications were detected, suggesting a sensitivity threshold close to the realtime RT-PCR. Based on a selection of positive NPS collected in Switzerland during the 2010-2011 influenza season, this study suggests that the PLEX-ID/Flu assay is a convenient platform for the detection, typing, and subtyping (lineage characterization for influenza B) of circulating influenza viruses. The detection rate observed with the PLEX-ID platform, as well as the subtyping or lineage characterization of human influenza virus types A and B tends to suggest that it might play a role in the near future in most routine laboratories, but this needs to be confirmed in prospective and large comparative studies. In addition, the sensitivity of the PLEX-ID/Flu observed in this study could be increased if any positive PCR results were considered and if the typing result is not a requirement. However, this pilot study has intrinsic limitations mainly related to the fact that only realtime RT-PCR-positive specimens were selected initially. This precludes any appropriate evaluation of the respective sensitivity or specificity of the method. Compared to sequence analysis, the PLEX-ID/Flu assay has also limitations in terms of identification of specific mutations potentially involved in antiviral resistance, as well as drifted strains. Finally, the lack of accurate typing concerning the HA and NA genes is a major drawback and possibly limits the use of this assay as a first line diagnostic tests.

Evaluation of atherosclerotic lesions using dextran- and mannan–dextran-coated USPIO: MRI analysis and pathological findings

Magnetic resonance imaging (MRI) can detect atherosclerotic lesions containing accumulations of ultrasmall superparamagnetic iron oxides (USPIO). Positing that improved USPIO with a higher affinity for atherosclerotic plaques would yield better plaque images, we performed MRI and histologic studies to compare the uptake of dextran- and mannan–dextran-coated USPIO (D-USPIO and DM-USPIO, respectively) by the atherosclerotic walls of rabbits. We intravenously injected atherosclerotic rabbits with DM-USPIO (n = 5) or D-USPIO (n = 5). Two rabbits were the controls. The doses delivered were 0.08 (dose 1) (n = 1), 0.4 (dose 2) (n = 1), or 0.8 (dose 3) (n = 3) mmol iron/Kg. The dose 3 rabbits underwent in vivo contrast-enhanced magnetic resonance angiography (MRA) before and 5 days after USPIO administration. Afterwards, all animals were euthanized, the aortae were removed and subjected to in vitro MRI study. The signal-to-noise ratio (SNR) of the aortic wall in the same region of interest (ROI) was calculated in both in vivo and in vitro studies. Histological assessment through measurement of iron-positive regions in Prussian blue-stained specimens showed that iron-positive regions were significantly larger in rabbits injected with DM- rather than D-USPIO (P < 0.05) for all doses. In vivo MRA showed that the SNR-reducing effect of DM- was greater than that of D-USPIO (P < 0.05). With in vitro MRI scans, SNR was significantly lower in rabbits treated with dose 2 of DM-USPIO compared with D-USPIO treatment (P < 0.05), and it tended to be lower at dose 3 (P < 0.1). In conclusion, we suggest that DM-USPIO is superior to D-USPIO for the study of atherosclerotic lesions in rabbits.

Atherosclerosis is a chronic inflammatory response to vessel wall injury, leading potentially, to acute coronary syndrome and cerebral vascular disorders, induced by plaque rupture. The noninvasive imaging of atherosclerotic plaque progression and of therapeutic response is very important. In atherosclerosis, macrophage accumulation initiates lesion development and triggers clinical events by producing several molecules that promote inflammation, plaque disruption, and subsequent thrombus formation. [1] [2] [3] [4] However, conventional anatomic measurements of advanced lesions (eg, plaque size or luminal narrowing) do not necessarily correlate with macrophage content. It is desirable that imaging of macrophages permits the identification of highly activated plaques, to help in the prediction of acute thrombotic events, and to facilitate assessment of the therapeutic effects of antiatherosclerotic drugs, such as statins. In hyperlipidemic rabbits and humans, atherosclerotic plaques have been investigated using magnetic resonance imaging (MRI), using ultrasmall superparamagnetic iron oxides (USPIO): iron oxide nanoparticles stabilized with low-molecular-weight dextran, having a mean diameter of 30 nm. [5] [6] [7] [8] As these relatively small particles are not immediately recognized by the hepatic and splenic mononuclear phagocytic systems (MPS), 9, 10 the prolongation of their intravascular half-life permits their uptake, via macrophages, by the whole body, including the lymph nodes, lungs, and the walls of atherosclerotic vessels. [5] [6] [7] [8] 11, 12 With respect to the excretion pathway, USPIO particles are internalized by receptor-mediated endocytosis, and metabolized via the lysosomal pathway. 13 On internalization by macrophages, the dextran coating is progressively degraded; 89% is eliminated in urine, and the rest is excreted in feces. The iron contained in USPIO is incorporated into the body's iron store and used in hemoglobin manufacture. Like endogenous iron, it is eliminated very slowly, predominantly via the feces. 14 Based on USPIO-associated T2-and T2*shortening effects, atherosclerotic lesions with accumulated USPIO can be detected on MRI scans. Higher doses tended to be administered in earlier animal studies. We posited that improved USPIO with higher affinity for atherosclerotic plaques would make it possible to obtain better images of plaques at lower doses. The effectiveness of iron oxide nanoparticles, which conjugate with various biomolecules, has been studied previously in investigations of novel therapeutic actions and drug delivery, for example, in blood purification therapy and drug treatments for cancer and hyperthermia. [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] According to Bouhlel et al, 26 the mannose receptors of M2 macrophages are more abundantly expressed in pathological tissues of atherosclerotic lesions than in the adjacent zone. Therefore, we focused on mannan-dextran-coated USPIO (DM-USPIO), prepared in-house by adding alkalitreated mannan to dextran-coated USPIO (D-USPIO). Our working hypothesis posited that the addition of mannan to D-USPIO would facilitate its uptake by macrophage-rich atherosclerotic plaques. Here, we performed MRI and histopathological studies to compare the uptake of DM-USPIO with that of D-USPIO in atherosclerotic plaques of rabbits. All experimental protocols were approved by our Animal Experimentation Committee. All experiments were conducted in accordance with the Animal Care Guidelines of Shiga University of Medical Science. We studied two kinds of iron oxide: DM-USPIO and D-USPIO (supplied by Meito Sangyo, Kiyosu, Aichi, Japan). Superparamagnetic iron oxide (SPIO) is a widely-used, safe, liver-specific contrast medium (Resovist ® ; Bayer Health Care Japan, Osaka, Japan); 27 carboxydextran covers the iron core of the iron oxide particles. The particle size of the D-USPIO used in this study is different from that of SPIO but its composition is similar. We prepared DM-USPIO by adding alkali-treated mannan to D-USPIO ( Figure 1) were the same for D-and DM-USPIO (28 nm, 15 mg/mL, 0.028 erg ⋅ gauss -2 ⋅ g -1 , respectively). To obtain D-USPIO, 84 g of carboxydextran (molecular weight: 2500, approximately) was dissolved in 150 mL of water. To this was added an aqueous solution obtained by dissolving 17 g of ferrous chloride tetrahydrate in 130 mL of a 1 M aqueous ferric chloride solution under a nitrogen gas stream. Then, 220 mL of a 3 N aqueous solution of sodium hydroxide was added, with heating and stirring. The mixture was adjusted to pH 7.0 by adding 6 N hydrochloric acid, and then refluxed for 1.5 hours. The reaction mixture was subjected to centrifugation at 2500 × g for 1 hour, and the supernatant liquid then subjected to ultrafiltration (Centramate™; Pall Corporation, Long Island, NY) (molecular weight cut-off [MWCO]: 100,000) to obtain 1 L of D-USPIO. To create DM-USPIO we added 500 mL of D-USPIO solution (iron concentration: 30 mg/mL) to 15 g of mannan prepared by extraction from beer yeast and refluxed for 5 hours. The reaction mixture was then subjected to ultrafiltration (MWCO: 100,000) to obtain 500 mL of DM-USPIO. We used six Watanabe heritable hyperlipidemic (WHHL) rabbits obtained from the Institute for Experimental Animals at Kobe University, Hyogo, Japan. Each was 9-12 months old and weighed approximately 3 kg. At this age, WHHL rabbits harbor active plaque formations in the aortic wall. 28 They were divided into two equal groups and injected intravenously with 0.8 mmol iron (Fe)/kg (dose 3) of D-or DM-USPIO. Gadolinium-enhanced magnetic resonance angiography (MRA) scans were obtained before and 5 days after the injection. For each MRI session, the rabbits were fully anesthetized with ketamine (Ketalar; Daiichi Sankyo, Tokyo, Japan), at 25 mg/kg body weight, and medetomidine (Domitor; Nippon Zenyaku Kogyo, Koriyama, Fukushima, Japan), at 0.1 mg/kg body weight, then injected intravenously with 2 mL gadopentetate dimeglumine (Magnevist ® ; Bayer Health Care Japan), diluted in 10 mL saline. All scans were obtained using a 1.5T MRI system, using a transmit-receive coil (Magnetom Sonata; Siemens Medical Solutions, Erlangen, Germany) (maximum amplitude: 40 mT m -1 ; slew rate: 200 mT m -1 msec -1 ). For imaging, we used a fast low-angle shot (FLASH) protocol. The parameters used were: TR, 4.3 ms; TE, 1.7 ms; flip angle, 25°; field of view (FOV), 320 × 200 mm; matrix size, 512 × 320; and slice thickness, 1 mm. The source images were made available for analysis on a workstation that allowed interactive multiplanar reformatting of the data sets. For quantitative analysis, we calculated the signal-to-noise ratio (SNR) in the same three regions of interest (ROI) (3 × 20 pixels) in the aortic wall of the lower intrathoracic aorta in three different coronal reformatted images. We compared the SNR yielded by both types of USPIO. Of the twelve WHHL rabbits, ten were divided into two equal groups, and injected with D-or DM-USPIO at three different doses: 0.08 (dose 1) (n = 1), 0.4 (dose 2) (n = 1), or 0.8 (dose 3) (n = 3) mmol Fe/kg. Two were untreated and served as controls. The animals were euthanized 5 days post-injection, and MRI scans of resected aortic specimens were obtained. The specimens were placed in centrifuge tubes filled with gadopentetate dimeglumine (1:50 dilution) and 10% gelatin by weight. MRI was performed using a 1.5T MRI scanner, using a 1-channel loop coil (Signa HDxt; GE Healthcare, Little Chalfont, UK). Imaging was achieved using the three-dimensional (3D) first-spoiled GRASS protocol. The parameters used were: TR, 11.8 ms; TE, 4.1 ms; flip angle, 30°; FOV, 5 × 5 cm; matrix size, 512 × 512; slice thickness, 1 mm. Source images were made available for analysis on a workstation that enabled interactive multiplanar reformatting of the data sets. For quantitative analysis we calculated the SNR at three ROIs (3 × 20 pixels) in the wall of the lower intrathoracic aorta in three different coronal reformatted images. We compared the results obtained after three different doses of D-and DM-USPIO. Aortic specimens were fixed in 10% paraformaldehyde. From five different lesions on the lower thoracic aorta, we cut paraffin-embedded 4 µm thick sections on the axial plane. These were stained with Prussian blue to identify the accumulation of iron oxide. We also stained macrophages immunohistochemically with RAM11, and compared Prussian blue-and RAM11-stained areas. For quantitative assessment, we measured the Prussian blue-stained area in one FOV (magnification, 200×) in three different lesions for each histological section of the lower intrathoracic aorta (using Image-Pro Plus; Media Cybernetics, Silver Spring, MD), and calculated the average values. In addition, the amount of iron per unit weight (300 µg) of the lower submit your manuscript | www.dovepress.com Dovepress Dovepress intrathoracic aortic specimens of dose 3 rabbits was measured by nuclear magnetic resonance (Minispec MQ20; Bruker Optics, Billerica, MA) (20 MHz, 0.47 T). To acquire additional data on the biodistribution of Dand DM-USPIO, we examined tissues from the liver, spleen, lung, kidney, and heart of rabbits which had undergone in vivo MRI study and had received dose 3 of either D-or DM-USPIO (n = 3, respectively); two rabbits treated with neither D-nor DM-USPIO were the controls. These specimens were fixed in 10% paraformaldehyde. Paraffin-embedded 4-µm-thick sections were cut in the axial plane and stained with Prussian blue. We assessed one section for each organ. Prussian blue-stained areas were measured and evaluated as described above. To compare the phagocytosis of D-and DM-USPIO, we used the murine macrophage cell line, J774.1 (RIKEN Cell Bank; Wako, Saitama, Japan), which is widely used in research on macrophages. [29] [30] [31] Cells (8 × 10 5 ) were seeded in 12 multiwell cluster plates (Corning Inc., Corning, NY) and grown at 37°C for 24 hours in 1 mL growth medium (RPMI; Nacalai Tesque, Kyoto, Japan), supplemented with 10% bovine fetal calf serum by volume (Invitrogen, Carlsbad, CA) and 1% by weight mixed penicillin and streptomycin solution (Nacalai Tesque). The medium was then replaced with fresh medium. D-USPIO or DM-USPIO (10 µg iron oxide and 100 µL, respectively) were added, and the plates incubated for 1 hour for cell labeling. The medium was again replaced and the wells incubated for 24 hours. To identify intracellular iron oxide accumulations, the cells were stained with Prussian blue. The amount of iron contained in cell lysates was measured by atomic absorption photometry (AA-6800; Shimadzu, Kyoto, Japan). Cell labeling was performed as described in a previous study. 32 We used SPSS for Windows software (SPSS Japan, Tokyo, Japan) for statistical analysis. Differences in the uptake of D-and DM-USPIO in the arterial wall, in the uptake of iron by cultured macrophages, and in the SNR values from MRA scans obtained in vivo and in vitro were determined using the one-tailed Student's t-test. A P-value of P , 0.05 was considered statistically significant. In rabbits treated with dose 3 of D-or DM-USPIO (n = 3, respectively), MRA images obtained in vivo showed irregularities, seen as spotty signal voids in the aortic wall. These were interpreted as indicating iron deposits. Before treatment with either agent, the aortic wall appeared smooth, without any evidence of atherosclerotic plaque formation. The SNR of the aortic wall was significantly lower after treatment with D-and DM-USPIO than before treatment. The difference between the SNR obtained before and after the injection of nanoparticles was significantly greater in rabbits treated with DM-rather than D-USPIO (P , 0.05) (Figure 2 ). The aortic walls of the control animals were smooth, and the signal emitted was relatively high. In contrast, signal strength decreased in a dose-dependent fashion in rabbits injected with D-or DM-USPIO. At dose 3, the SNR tended to be lower in rabbits injected with DM-than with D-USPIO (P , 0.1). At dose 2 DM-USPIO, the SNRs obtained from three ROIs on three different aortic images were significantly lower than those of the rabbit treated with an equivalent dose of D-USPIO (P , 0.05). At dose 1, there was no statistically significant difference (P . 0.1) (Figure 3) . These findings indicate that the uptake of either type of USPIO produced a decrease in SNR. Prussian blue-stained histopathological sections showed marked iron uptake by macrophages embedded in atherosclerotic plaques in the aortic walls of rabbits injected with either type of USPIO. This was not the case in the control animals. Iron uptake was dose-dependent ( Figure 4 ). Immunohistochemical staining (RAM11) showed that the localization of iron deposits and of macrophages coincided ( Figure 5 ). Quantitative analysis of iron-positive areas in the aortic wall revealed that they were significantly larger in rabbits treated with dose 3 DM-than in those treated with dose 3 D-USPIO (P , 0.05). In addition, in different aortic sections, these areas were significantly larger in the rabbit treated with dose 1 or dose 2 DM-USPIO than in that treated with an equivalent dose of D-USPIO (P , 0.05) (Figure 6 ). At dose 3 of D-and DM-USPIO, the amount of iron accumulated per unit weight of aortic specimen was not different (Figure 7) . There was no statistically significant difference in the biodistribution of D-and DM-USPIO in tissues from the spleen and kidney. The values determined by Prussian blue staining in tissues of the spleen were 40,088 ± 10,480 vs 43,607 ± 9456 for D-and DM-USPIO, respectively, and submit your manuscript | www.dovepress.com Dovepress Dovepress were 1711 ± 1508 vs 758 ± 583, respectively, in kidney tissues. In the liver and lung, these values tended to be higher in DM-than D-USPIO-treated rabbits (18,118 ± 5027 vs 14,421 ± 5920 in the liver, and 1261 ± 1363 vs 169 ± 245 in the lung). There was no significant iron accumulation in cardiac tissues. In the controls, only the spleen (the ironstoring organ) showed significant iron deposits. In both D-and DM-USPIO-treated rabbits these levels were increased. Microscopically, almost all cultured (J774.1) cells phagocytosed both D-and DM-USPIO. However, the amount of intracellular iron measured by atomic absorption photometry was significantly higher in cells treated with DM-rather than D-USPIO ( Figure 8 ) (P < 0.05). Our experimental study showed that, in WHHL rabbits, both D-and DM-USPIO particles were phagocytosed by macrophages embedded in atherosclerotic plaques. This leads to a susceptibility-induced signal reduction in the atherosclerotic vessel wall on T1-weighted 3D gradient echo (GRE) images. In the T1-weighted fast 3D GRE sequence, T1 shortening is a characteristic of lower USPIO concentration. This renders the signal bright, and produces a predominant T2/T2* shortening at higher concentrations, resulting in a completely dark signal. 33 Consequently, at the initial stage of USPIO administration, the signal of the aortic lumen is completely dark due to a high blood concentration of USPIO. According to Ruehm et al, 5 the uptake by the MPS of USPIO particles over a 4-to 5-day period creates an ideal situation for imaging the vascular wall. The iron concentration in inflammatory cells in plaques is sufficiently high for the production of dominant T2 and T2* effects, while its concentration in the blood pool is at levels at which T1 shortening effects predominate. In that study, iron particles were observed electron microscopically in actively phagocytosing cells, but not in inactive foam cells filled with fat vacuoles. These, and other observations, suggest that both D-and DM-USPIO are potential contrast agents for the diagnosis of atherosclerotic plaques -which exhibit high activity before the onset of luminal narrowing -and are potentially useful for the assessment of therapeutic effects of anti-atherosclerotic drugs, such as statins. Macrophages and dendritic cells can be targeted by mannosylated nanoparticles because immune cells, including alveolar and peritoneal macrophages, monocyte-derived dendritic cells, and Kupffer cells, constitutively express high levels of the mannose receptor (MR). Hashida and colleagues took the lead in the development of mannosylated liposomes to target macrophages and dendritic cells for delivery (in animal models) of the anti-inflammatory agent dexamethasone palmitate and the CpG DNA complex. 34, 35 The study showed that intratracheally administered mannosylated liposomes, with various ratios of mannosylated cholesterol derivatives, were preferentially taken up by alveolar macrophages. Inhibition studies showed that mediation occurred via mannose receptor endocytosis. Mannosylation significantly improved liposome internalization by macrophages. 36 Our histological analyses showed that the uptake of DM-was higher than that of D-USPIO in the atherosclerotic wall of rabbits, resulting in a greater reduction of signal from the aortic wall in MRI scans. As the amount of intracellular iron was higher in macrophage cells treated with DM-rather than D-USPIO, we posit that macrophages contained in atherosclerotic plaques are more sensitive to DM-than to D-USPIO, making it possible to acquire plaque images at lower doses of contrast agent. We also found that the distribution of DM-USPIO in the liver and lung tended to be higher than that of D-USPIO, suggesting that receptor-mediated endocytosis had a strong role in the internalization of DM-USPIO by macrophages. As mannose receptors are expressed in the liver and lung, [34] [35] [36] it is not surprising that the distribution of DM-USPIO was higher than that of D-USPIO in these organs. We cannot rule out the possibility that increased distribution in these organs results in a reduction in the amount of circulating USPIO particles, and a consequent reduction in their presence in the aortic wall. However, based on our current findings, we suggest that the affinity of DM-USPIO particles for the aortic wall affects their uptake positively, and that the effect of reduced circulation of DM-USPIO due to its distribution in the liver and lungs is not strong. This is because the circulation lifetime of USPIOs is sufficiently prolonged, in accordance with particle size. On the other hand, the amount of iron per unit weight was not different irrespective of the type of USPIO used. We attributed this to technical difficulties with our preliminary arrangements for making NMR measurements. For our NMR measurements, the uniformity of the samples was the most important issue because we used minute quantities of sample for measurement. Nevertheless, it was very difficult to obtain vessel homogenates because the vessels contained abundant fibers. Consequently, insufficient preparation of uniform homogenate samples may have impacted our NMR measurements and caused inconsistency between NMR measurements and other examinations, such as imaging and histological analysis. Our study has some limitations. First, the number of rabbits was too small to confirm the reproducibility of the observed aortic wall changes, and assessment of inter-individual differences was not possible. Second, the USPIO doses we used were too high for clinical applications; additional studies are necessary to identify appropriate human dose levels. Also, the imaging sequence was not optimized. (The use of a more T2*-weighted GRE sequence with longer echo times may enhance the sensitivity for iron-induced susceptibility effects. With such a sequence, even smaller iron accumulations can be detected. This may facilitate a further reduction in the contrast dose). Third, after the injection of D-and DM-USPIO particles, the signal from the aortic wall was very low and not uniform among the sites examined. ROI placement affects the results of imaging analysis, and ROI analysis of the aortic wall is subject to greater dispersions. In efforts to avoid this potential bias, we used the average result from three different ROIs in our SNR measurements. Fourth, the imaging effects we observed may be specific for WHHL rabbits. The similarity between atherosclerotic plaque formation in rabbits and humans has been documented elsewhere. [38] [39] [40] [41] We suggest that the observed differences in the uptake of DM-and D-USPIO by rabbit atherosclerotic lesions are attributable to the binding of mannan to USPIO. Further study is required to investigate the stability of bound mannan and to determine optimum dosage amounts. The trapping of DM-and D-USPIO particles in the atherosclerotic wall of rabbits was indicated by a remarkable decrease in SNR. Our histological and imaging analyses showed that DM-USPIO was taken up to a greater degree than D-USPIO. Based on our observations, we suggest that DM-USPIO is superior to D-USPIO for the study of atherosclerotic lesions.

IL-15 Participates in the Respiratory Innate Immune Response to Influenza Virus Infection

Following influenza infection, natural killer (NK) cells function as interim effectors by suppressing viral replication until CD8 T cells are activated, proliferate, and are mobilized within the respiratory tract. Thus, NK cells are an important first line of defense against influenza virus. Here, in a murine model of influenza, we show that virally-induced IL-15 facilitates the trafficking of NK cells into the lung airways. Blocking IL-15 delays NK cell entry to the site of infection and results in a disregulated control of early viral replication. By the same principle, viral control by NK cells can be therapeutically enhanced via intranasal administration of exogenous IL-15 in the early days post influenza infection. In addition to controlling early viral replication, this IL-15-induced mobilization of NK cells to the lung airways has important downstream consequences on adaptive responses. Primarily, depletion of responding NK1.1+ NK cells is associated with reduced immigration of influenza-specific CD8 T cells to the site of infection. Together this work suggests that local deposits of IL-15 in the lung airways regulate the coordinated innate and adaptive immune responses to influenza infection and may represent an important point of immune intervention.

Influenza virus is a major human pathogen that causes substantial morbidity and mortality-approximately 36,000 deaths annually in the United States alone [1] . Combined with the severe economic burden imposed from seasonal influenza outbreaks and growing concerns over potential imminent influenza pandemics, there is considerable need for a firm understanding of the disease pathology, prevention strategies, and mechanisms of host defense against the virus [2] . Influenza virus is primarily transmitted via inhaled aerosols and results in an infection localized to the upper respiratory tract, with viral replication largely limited to epithelial cells [3] . Mechanisms by which the immune system eliminates influenza have been well studied and are known to involve the coordinated actions of the innate and adaptive immune systems. Namely, the cytolytic action of influenza-specific CD8 T cells has been shown to be the primary mediator of complete viral clearance, but important roles have also been described for CD4 T cells [4, 5, 6] . In addition to T cells, a crucial role has also been established for innate immune effectors including natural killer (NK) cells, which provide short-term control of viral replication prior to T cell activation [7] . NK cells become activated following the loss of inhibitory signals coupled with positive activating signals resulting in direct (via release of cytotoxic granules and interferon c) or indirect (via activation of macrophages and dendritic cells) target cell lysis [8] . NK cells are vital in limiting influenza viral replication as depletion of NK cells dramatically increases morbidity and mortality in hamsters and mice [9] , and in humans severe infections with the 2009 pandemic H1N1 virus positively correlated with reduced numbers of NK cells in the lungs [10] . Studies have indicated that the natural cytotoxicity receptors NKp44 and NKp46, which recognize hemagglutinin proteins of several different influenza strains [11, 12] is one mechanism used by NK cells to protect against lethal viral challenge [13] . Secondarily, NK cells also aid in viral clearance indirectly through the production and secretion of cytokines which both amplifies local inflammation and recruits antigen-specific CD8 T cells to sites of inflammation [14] . Implicit in both of these functions is the ability of NK cells to accumulate within the respiratory tract to contact infected cells and provide a source of chemotactic signals to recruit recently activated CD8 T cells. Type I IFNs expressed within hours after viral infection have been documented to induce expression of the chemokines CXCL9 and 10 which function to recruit CXCR3 expressing NK cells to sites of infection [15] . However, Type I IFNs also modulate the expression of the common gamma chain cytokine interleukin 15 [16, 17, 18] , which we recently reported to be temporally and locally increased following influenza infection [19] . This expression of IL-15 in the respiratory tract facilitates the recruitment of antigen-specific CD8 T cells to the respiratory tract. However, it is unclear whether the chemotactic properties of IL-15 uniquely affect migratory CD8 T cells or could be extended to other IL-15-sensitive immune cells. NK and NKT cells are nearly absent in IL-15 2/2 animals [20] , highlighting the important role of IL-15 on NK cell development and homeostasis in the steady-state. Following viral infections, de novo production of IL-15 by dendritic cells results in the activation and proliferation of NK cells [21, 22] , and transient systemic stimulation of NK cells with soluble IL-15/IL-15Ra complexes also results in an accumulation of phenotypically and functionally mature NK cells [23, 24] . In addition to these roles, IL-15 also can stimulate the migration of NK cells in vitro and enhances their adhesion to cultured endothelial cells [25] . We therefore hypothesized that virally induced IL-15 functionally assists in the migration of NK cells into the lung airways. We show here that an IL-15 deficiency results in a site-specific reduction in NK cells from the lung airway and an exacerbation of viral load at early time points post influenza infection. Additionally, exogenous IL-15 induces the specific migration of NK cells in vitro and in vivo. This IL-15dependent enhanced mobilization of NK cells to the lung airways correlates with decreased viral loads. Importantly, in the absence of NK cells, antigen-specific CD8 T cells fail to accumulate at the site of infection, providing a possible link between IL-15-mediated migratory effects of both the innate and adaptive immune responses to influenza infection and suggest therapeutic possibilities regarding the use of IL-15 to simultaneously regulate both arms of the immune system for improved responses to viral infection. All animals were handled in strict accordance with good animal practice as defined by the American Association for Accreditation of Laboratory Animal Care as well as federal and state agencies. All animal work presented here was approved by Institutional Animal Care and Use Committee of University of Georgia (AUP No. A2009-6-114). Mice, Viruses, IL-15 Blocking, and NK Cell Depletion C57BL/6 mice were purchased from Charles River (Wilmington, MA) through the NCI program. Influenza A/HK-x31 (x31, H3N2) was generously donated by Dr. S. Mark Tompkins (University of Georgia, Athens, GA). Animals were infected intranasally (i.n.) with 10 3 PFU HKx31 diluted in 50 mL sterile PBS. IL-15 was blocked using 25 mg anti-IL-15 mAb (clone AIO3) (eBioscience, San Diego, CA) administered daily via intraperitoneal injection (i.p.) in 200 mL sterile PBS. IL-15 depletion was confirmed by reductions in frequencies of both NK and CD44 hi CD8 + T cell in antibody-treated animals compared to untreated animals 7 days after the initiation of mAb treatment. NK1.1+ cells were depleted via intravenous (i.v.) injections of 200 uL PBS containing 200 mg anti-NK1.1/mouse (clone PK136) every other day (UCSF monoclonal antibody core facility, San Francisco, CA), and depletion of NK cells was verified using an anti-NKp46 mAb (Clone 21A9.4, eBioscience) throughout the experiment and 2 days after the last injection of the NK1.1 depleting mAb. Lung airway-resident cells were harvested by bronchioalveolar lavage (BAL) with 3 consecutive washes of 1 mL PBS. To isolate cells from the lung parenchyma, lungs were perfused with ,25 mL PBS/heparin sodium solution, harvested, minced and incubated at 37uC for 30 minutes in 1.25 mM EDTA. The tissue was subsequently incubated in collagenase diluted in RPMI (6 mg/mL) at 37uC for 1 hour and passed through a 5 mm cell strainer. Isolated cells were subjected to separation via density gradient centrifugation by resuspending cells in 47% Percoll underlain with 67% Percoll. The gradients were then centrifuged at 2800 rpm for 20 minutes, and lymphocytes at the interface were collected. Spleen and lymph nodes were harvested from animals, homogenized, and then passed through a cell strainer. Spleen homogenate was depleted of erythrocytes by incubation in tris-buffered ammonium chloride. Single cell suspensions were stained with combinations of cocktails containing anti-CD3, NK1.1, NKp46, CD11c, CD11b, CD122, and CD132 (eBioscience, San Diego, CA) as indicated for 20 minutes at 4uC. Where indicated, cells were concurrently stained with or without biotinylated anti-IL-15Ra (R&D Systems, Minneapolis, MN) followed by 20 minute incubation with APCconjugated Streptavidin at 4uC. Influenza nuclear protein (NP) MHC class I tetramer [H2-D b /ASNENMETM] were generated by the National Institute of Allergy and Infectious Diseases Tetramer Facility (Emory University, Atlanta, GA). Tetramer staining was conducted at room temperature for 1 hour concurrently with anti-CD3, NK1.1, NKp46, CD8, and CD44 (eBioscience, San Diego, CA). Stained cells were analyzed using a BD LSRII digital flow cytometer (BD Biosciences, San Jose, CA) and either BD FACSDiva or FlowJo software (Tree Star, Inc., Ashland, OR). IL-15 complexes (IL-15c) were generated on the day of use by incubating 1.5 mg recombinant mIL-15 with 7 mg IL-15Ra Fcchimera (R&D Systems, Minneapolis, MN) at 37uC for 20 minutes followed by 4uC for at least 10 minutes. For Ra only controls, 7 mg IL-15Ra Fc-chimera (R&D Systems, Minneapolis, MN) was incubated similarly to complexes without addition of the cytokine. Complexes or Ra alone were administered via passive inhalation into both nostrils using a micropipette delivering 36.25 mL (for daily treatments) or 72.5 mL (for one time treatments) of complexes in sterile PBS. For assessment of cell proliferation, animals received 2 mg of BrdU (Sigma-Aldrich, St. Louis, MO) administered i.p. in a 200 mL volume of PBS. Cells were isolated from these animals 12 hours after treatment and stained with 20 mL aminoactinomycin D (7-AAD; BD Pharmingen, San Jose, CA) for 20 minutes at 4uC. Cells were surface stained as previously described and stained intracellularly with FITC-labeled anti-Ki-67 and APC-labeled anti-BrdU monoclonal antibodies (BD Pharmingen, San Jose, CA) according to manufacturer's instructions. In vitro migration assays were performed by placing bulk populations of lymphocytes containing predetermined numbers of NK cells (verified by FACs analysis) from the indicated tissues on the top insert of a 5 mm chemotaxis transwell (Fisher Scientific, Waltham, MA) in which the bottom well contained warm media alone or supplemented with 100 ng/mL IL-15c. IL-15c was generated by incubating 100 ng of IL-15 with 500 ng IL-15Ra Fcchimeric protein at 37u for 20 minutes and 4u for 10 minutes. Plates containing transwells were then incubated at 37uC with CO 2 exchange, and 90 minutes after plating, cells were harvested from bottom chambers and the percent migration of NK cells was calculated as the ratio of the number of NK cells in the bottom chamber compared to the number of NK cells determined in the input sample. Plaque assays were preformed as previously described [26] . Briefly, lungs from HK-x31 infected animals were collected and homogenized using a tissue lyser (Qiagen, Hilden, Germany). Monolayers of Madin-Darby kidney cells were incubated with 10fold serial dilutions of 10% homogenate in dilution media (16MEM, 1 mg/mL TPCK-treated trypsin) for 1 hour at 37uC. Cells were washed with 16 sterile PBS and overlaid with MEM containing 1.2% Avicel microcrystalline cellulose (FMC BioPoly-mer, Philadelphia, PA), 0.04 M HEPES, 0.02 mM L-glutamine, 0.15% NaHCO 3 (w/v), and 1 mg/mL TPCK-treated trypsin. After 72 hours, the overlay was removed, and the cells were washed with 16 sterile PBS, fixed by incubation with cold methanol/acteone (60:40%), and stained with crystal violet. Statistical significance was determined by Student's T test using Prism 5 software (GraphPad Software). Significance was determined to be any p-value where p,0.05. We and others have shown that following influenza infection, IL-15 message and protein is increased in the lung airways [19, 27] . Because this IL-15 expression was rapidly induced by influenza infection and reached significant levels as early as day 3 post infection (p.i.) [19] , we hypothesized that influenza-induced IL-15 expression may be an important mediator of NK cell responses to influenza infection. We therefore first sought to determine whether NK cells responding to influenza infection were capable of receiving signals from this locally produced IL-15. To this end, lymphocytes were isolated from the lung airways of influenzainfected animals via BAL, and CD3 2 , NK1.1 + NK cells were analyzed for the expression of IL-15 receptor components by flow cytometry. The IL-15 receptor is a heterotrimer, consisting of the common gamma chain (CD132), the shared IL-2/IL-15Rb chain (CD122), and the specific IL-15Ra chain [28] . To date the majority of biological effects of IL-15 on NK cells, however, are mediated through the paired co-expression of CD122 and CD132 [29, 30, 31] , while IL-15Ra is only required by accessory cells which present IL-15 to respondent cells, a mechanism referred to as trans-presentation [32] . However, some groups have suggested that IL-15Ra expression alone contains some signaling moieties which may participate in distinct biological functions [27] . Therefore it is important to establish the kinetics of IL-15 receptor component expression and IL-15 signaling potential on NK cells in our model. NK cells are known to respond rapidly to influenza infection and continue accumulating in the lung airways through day 5 p.i. ( [13] and data not shown). Since we wished to specifically evaluate recent NK cell immigrants responding to the airway inflammation resulting from influenza infection, we restricted our analyses of NK cell kinetics to before and up to day 4 p.i.. NK cells were first detected in the BAL at day 2 post influenza infection (albeit at a low frequency, ,0.1-0.6% of lymphocytes, Figure 1A and Figure 2A ) but were completely absent in control mock-infected animals (data not shown). Despite the low frequency of NK cells in the lung airway at this early time point, one fifth consistently expressed IL-15Ra. Additionally, 30-40% of them expressed CD122 and CD132 ( Figure 1B ). By day 3 p.i., when NK cells represented a much more discernible population (,2-3% of lymphocytes, Figure 1A & 2A), greater than 90% of these gated cells expressed CD122 and CD132, and expression levels of these receptors on a per-cell basis increased over time as indicated by a higher median fluorescence intensity by day 4 p.i. ( Figure 1B ) and consistent with evidence that expression of IL-15R components is induced by activating stimuli [33] . Expression of IL-15Ra however, was variable but consistently much lower that the expression levels of CD122 and CD132. This biased expression of CD122/132 receptor chains over IL-15Ra and enhanced IL-15 levels [19] following influenza infection, indicate that NK cells accumulating at the site of influenza infection are capable of responding to locally produced IL-15 via the trans-presentation pathway. Since co-expression of CD122 and CD132 render NK cells responsive to IL-15 signals, we next wished to determine whether virally induced IL-15 and subsequent signaling through these receptors affected the accumulation of these cells in the respiratory tract. Because IL-15 2/2 mice exhibit a severe developmental defect in both NK and NKT cell lineages and are nearly devoid of these cell populations [20] , we chose to use an anti-IL-15 blocking antibody to selectively deplete IL-15 concurrent with infection. Therefore, we monitored the influx of NK cells in animals receiving either PBS or anti-IL-15 mAb administered i.p. daily from day 0 through day 4 post influenza infection. In order to more accurately define bona fide NK cells, particularly in the lung airways harboring few NK cells at very early time points post influenza infection, we included NKp46 reactivity in our staining protocol and henceforth define NK cells as CD3 2 lymphocytes positive for both NK1.1 and NKp46. While NK cells accumulated in the lung airways of untreated mice as expected, by day 2 p.i. the overall frequency of NK cells in IL-15 blocked animals was reduced by half and remained this low through day 4 p.i. (Figure 2A) . Concordantly, total numbers of NK cells in IL-15blocked mice were partially reduced at day 2 p.i., and substantially (Figure 2A ). In fact, whereas the numbers of NK cells continued to accumulate in the lung airways of control-treated animals through day 4 p.i., numbers of NK cells in the lung airways of treated animals plateaued by day 3 p.i. (Figure 2A ). In the lung parenchyma of control-treated animals, NK cells also accumulated over time post infection, similar to those in the lung airways, but the frequencies of NK cells in this site remained unchanged in IL-15-blocked mice. Numbers of NK cells in the lung parenchyma of anti-IL-15 treated animals also remained similar to control animals with only a slight reduction at day 2 p.i. ( Figure 2B ). Importantly, while the numbers of NK cells were significantly reduced at the site of infection as a result of an IL-15 deficiency, anti-IL-15 treatment had little effect on the frequency or number of NK cells found in anatomical sites distal to the site of infection such as the spleen ( Figure 2C ). In order to determine whether an absence of IL-15 in the lung airways resulted in the reduced frequencies and numbers of NK cells specifically, numbers of other populations of innate cells in the airways at these early time points post infection were analyzed. CD11c 2 CD11b + cells (granulocytes) or CD11c + CD11b 2 (dendritic cells) in the airways following infection were mostly unaffected by the IL-15 deficiency, with only a small reduction observed at day 4 p.i. (Figure 2D ). In contrast, NK1.1 + CD3 + (NKT cells) were markedly reduced in the lung airways of IL-15blocked mice ( Figure 2D ), but overall, these cells represented a low proportion of the innate cells responding to influenza infection at these early time points following infection ( Figure 1A ). Together, these data indicated that short term blockade of IL-15 did not result in global defects in NK cell homeostasis or survival in peripheral tissues and the effects of IL-15 were largely specific to NK cells as blocking IL-15 selectively resulted in a significant loss of NK cells recovered from the site of infection. To test whether this local reduction in NK cells impairs the control of viral replication, we performed plaque assays on control-and anti-IL-15 treated mice to quantify viral load in the lungs of animals with intact or diminished IL-15 and NK cell responses. Viral load was quantified on days 1-5 and 7-8 p.i. to specifically look at control during the time frame of NK cell entry and accumulation in the lung airways and the kinetics of subsequent viral clearance. In IL-15 blocked animals, differences in viral load were apparent as early as d2 p.i. where viral titers were about 36 higher through day 3 p.i. ( Figure 2E) ; however, these animals seemed to regain control of viral replication by day 4 p.i., which perhaps corresponds with the early entry of cells of the adaptive immune response as anti-influenza specific CD8 T cells are first detectable in the lung airways by d6 post infection by flow cytometry ( [34] and data not shown). Thus, while viral elimination is not ultimately dependent on IL-15, early control of the virus is impaired in the absence of IL-15, which correlates with the arrival of a significant number of NK cells in the lung airways. We thus hypothesized that IL-15 was important for the migration of NK cells in the lung airways following influenza infection. We observed significant reductions in the numbers of NK cells in the lung airways of influenza-infected animals in which IL-15 was blocked at time points associated with their arrival at the site of infection, and failure of these cell populations to accumulate had implications in early viral control (Figure 2) . In order to determine whether IL-15 might be an important signal for NK cells in the migration to and/or the proliferation within the site of infection, we chose to provide exogenous IL-15 in an attempt to enhance any IL-15-dependent NK cell migration and/or in situ proliferation within the lung airways of influenza-infected animals. To this end, either PBS or recombinant IL-15/IL-15Ra fusion protein complexes (IL-15c) were administered intranasally to mice three days following influenza infection. To ensure that any biological effects of the IL-15c could be attributed to activity of the cytokine (which is merely stabilized by complexing to IL-15Ra), a control group of mice received the IL-15Ra only. Concurrent with treatment, mice received an i.p. pulse of the thymidine analog BrdU to identify proliferating cells. Twelve hours post treatment, the frequency and total number of NK cells in the BAL was quantified as well as the percentage of these cells incorporating BrdU. Isolated cells were simultaneously stained with 7-AAD as an indicator of cell viability, as only cells with disrupted membranes stain positive for this fluorescent dye. Importantly, neither the IL-15Ra alone nor the IL-15c affected cell viability, as cells isolated from animals receiving these treatments had similar percentages of 7-AAD + NK cells as those from PBS-treated mice (data not shown). Upon introduction of IL-15c to the lung airways, the overall frequency of NK cells isolated from the BAL was significantly increased ( Figure 3A and B) , and the total number of NK cells isolated from this site was nearly three times that of PBS-treated control animals ( Figure 3B ). Interestingly, the percentage of NK cells expressing CD122, the IL-2/15 Rb chain, was reduced in IL-15c-treated animals ( Figure 3B ), perhaps indicative of increased signaling through and subsequent internalization of this receptor complex by IL-15 responsive cells. Unlike T cells, which require large clonal bursts of proliferation to achieve effector status, the effector function of NK cells is more related to activation and mobilization to the site of inflammation [35] . Nevertheless, we wished to test whether the large increases in NK cell number in the lung airways following IL-15c administration could be attributed to IL-15-induced proliferation of NK cells at the site. To assess the potential role of proliferation in this observed increase in NK cell frequency and number in the BAL of treated mice, these NK cells were analyzed for BrdU incorporation. Concomitantly, cells isolated from the BAL were assessed for expression of the cellcycle-specific protein Ki-67. Since BrdU is incorporated into the DNA of only cells in S phase whereas Ki-67 is expressed by cells in any stage of the cell cycle, it was unsurprising that BrdU + cells were only a fraction of Ki-67 + cells ( Figure 3A) . Therefore, we considered only cells positive for both markers as cells undergoing proliferation at the time of treatment. Although the percentage of BrdU-incorporating cells was modestly increased in IL-15c-treated animals, the overall frequency of proliferating cells was low in untreated (,10%) and treated (,12%) animals ( Figure 3B ). These data suggest that IL-15c may trigger the proliferation of NK cells, but proliferation of this cell population in the lung airways is, in general, low and not likely to be solely responsible for the total number of cells extracted from the lung airways. IL-15Ra alone had no significant impact on the accumulation or proliferation of this cell population ( Figure 3A and B) . Finally, differences in cell of CD11c 2 CD11b + cells, CD11c + CD11b 2 cells, and CD3 + NK1.1 + cells from BAL are shown 6 SEM in PBS control (solid lines) or aIL-15-treated (dashed lines) mice (n = 3 mice/group). (E) At the indicated times p.i. with 10 3 pfu HKx31, lung viral titers from PBS and aIL-15-treated mice were determined by plaque assay. Mean viral titer is plotted 6 SEM (n = 2 mice/group on days 1 and 5 p.i. or 3 mice/group for remaining time points; day 2 p = 0.0503). Data are representative of two independent experiments. doi:10.1371/journal.pone.0037539.g002 frequencies, numbers, CD122 expression, and BrdU incorporation were specific to the lung airways (the sight of treatment), as NK cells isolated from spleens were similar in control and IL-15ctreated animals ( Figure 3C and data not shown). Together, these data demonstrate that exogenous IL-15 results in increased numbers of NK cells in the lung airways. Since this increase appeared to be independent of IL-15-mediated effects on cell survival, and proliferation was low, we hypothesized that IL-15 may be responsible for a substantial amount of migration of NK cells into the lung airways following influenza infection similar to its effects on CD8 T cells [19] . Intranasal administration of IL-15c resulted in increased numbers of NK cells in the lung airways that appeared to be due to increased migration into that site. Previous studies have indicated that IL-15 is indeed chemotactic for NK cells. In vitro checkerboard assays revealed that freshly isolated NK cells migrated to IL-15 gradients, and IL-15 stimulation increased LFA-1-dependent binding of NK cells to cultured endothelial cells [25] . IL-15 has also been shown to play a central role in the recruitment of CD16 2 human NK cells into the endometrium following ovulation [36] . To test the direct chemotactic potential of IL-15c for NK cells in our own system, we employed an in vitro chemotaxis transwell assay. Bulk lymphocytes (or purified splenic NK cells, data not shown) from the BAL, lung, and spleen of mice collected three days after infection with influenza were placed in the top chamber of a transwell filter support with either media alone or media supplemented with IL-15c in the bottom chamber. After 90 minutes of incubation, IL-15c significantly enriched NK cells isolated from the lung and spleen ( Figure 4A ). Consistent with our findings with CD8 T cells [19] , NK cells from the lung airways did not migrate to IL-15c, perhaps because this site represents the terminal destination for these cell populations. Unlike CD8 T cells, which lose expression of CD122 upon residence in the lung airways following influenza infection [19] [37] , nearly 100% of the NK cells residing in the BAL of influenza-infected mice express CD122 ( Figure 1A and B) . Nonetheless, these data indicate that NK cells in the lung parenchyma or the general circulation (as represented by the spleen) migrate to IL-15 in vitro. Intranasal administration of IL-15c during the innate phase enhances early viral control A temporal cessation in IL-15 bioavailability reduced the numbers of NK cells (Figure 2A ) in the respiratory tract resulting in increased viral titers ( Figure 2E) , presumably due to impaired NK cell responses. Conversely, exogenous IL-15c promoted the migration of NK cells ( Figure 4 ) and resulted in increased numbers of NK cells in the lung airways ( Figure 3B ). We therefore hypothesized that intranasal administration of IL-15c early after infection could be used to enhance the early innate immune response to influenza and augment viral control. To this end, influenza-infected animals received either PBS or IL-15c intranasally on days 1-4 p.i., a time frame corresponding to the migration of NK cells into the lung airways and limiting any confounding effects IL-15c might have on adaptive immune cells entering the lung airways at later time points. Every other day, from day 2-8 p.i., whole lungs were collected and viral titers were quantified via plaque assay ( Figure 5A ). Early (d2) p.i., we observed no difference in viral load between PBS and IL-15c-treated mice, but as viral replication reached more significant levels at day 4 p.i., animals receiving the IL-15c had 2.46 less viral load than animals receiving only PBS ( Figure 5B and C) . Although viral titers dropped 10 fold in both groups of animals by day 6 p.i., as expected, surprisingly, there remained a greater than 2 fold significant difference in viral load ( Figure 5B and D) . Although both groups of animals cleared the influenza virus completely by day 8 p.i. (Figure 5B ), our data suggest administration of IL-15c on days 1-4 p.i. can enhance early control of influenza virus by cells of the innate immune system. Because we have previously reported that IL-15 is important for the migration of influenza-specific CD8 T cells into the lung airways [19] , it was possible that those observed effects were secondary to the recruitment of NK cells to the lung. To test whether the accumulation of NK cells in the lung was required for the subsequent immigration of influenza-specific CD8 T cells to the respiratory tract, influenza-infected animals were assayed for the accumulation of anti-influenza specific CD8 T cells in NK deficient animals, generated by administration of the aNK1.1 depleting mAb PK136 every other day ( Figure 6A ). Flow cytometric analyses of NKp46 expression of day 4 p.i. lymphocytes isolated from the lung, BAL, spleen, and mediastinal lymph nodes (MdLN) of aNK1.1-treated animals revealed robust depletion of NK cells as the number of CD3 2 NKp46 + cells were reduced to less than one third of those observed in PBS-treated control animals ( Figure 6B and data not shown). On days 6 and 8 p.i., the numbers of influenza-specific CD8 T cells in these same tissues were quantified through the identification of cells staining positive for a tetrameric reagent loaded with the immunodominant peptide derived from the influenza nucleoprotein (NP). Tetramer positive CD8 T cells could first be detected in the lung airways at day 6 p.i., and although numbers were low, they were somewhat reduced in the BAL of PK136-treated mice ( Figure 6C ). No reduction in influenza-specific CD8 T cell numbers could be observed in any other tissue. In fact, CD8 T cells seemed to accumulate in other tissues of animals depleted of NK1.1expressing cells. By day 8 p.i., the frequency of influenza-specific CD8 T cells in the BAL of PK136-injected animals was less than half than that of control animals, and the total numbers were reduced nearly threefold ( Figure 6D ). Again, this effect was specific to the lung airways, since frequencies and numbers of NPtetramer + CD8 T cells were unchanged in other tissues examined ( Figure 6D ). Therefore, the largely tissue-specific nature of the dependence of influenza-specific CD8 T cells on the presence of NK1.1-expressing cells in the lung airways partially indicates that lung-resident NK and possibly NKT cells are requisite for the subsequent migration of influenza-specific CD8 T cells into this site. These data suggest that IL-15-mediated migration of CD8 T cells to the lung airways may be, at least partially, an indirect effect of NK1.1 + cells moving into the lung airway space in response to IL-15 and that subsequent tissue remodeling or production of an intermediate factor is responsible for the subsequent recruitment of the CD8 T cells. To our knowledge, these data for the first time describe a role for IL-15 in linking the innate and adaptive responses to influenza infection necessitating the further inquiry of IL-15 as a potential vaccine adjuvant. Here, we have demonstrated that NK cells, which accumulate in the lung airways early after influenza infection, are dependent on IL-15 for this accumulation and subsequent ability to control viral load. NK cells in the lung airways express high levels of the common gamma chain (CD132) and CD122, receptors responsible for imparting IL-15 responsiveness. Since IL-15 is known to be produced in the lung airways following influenza infection, we investigated the role of IL-15 in NK cell responses to influenza. In the absence of IL-15, NK cell frequencies and numbers were significantly reduced in the BAL, resulting in impaired early control of influenza virus. Although numbers of CD3 + NK1.1 + cells were also substantially reduced following anti IL-15 treatment, it is unclear whether this cell population plays a relevant role in controlling primary infections with influenza. While NK cells are thought to be very important participants in the control of influenza replication, evidence for NKT cells playing a similar role is more controversial [38] . Models of NKT cell activation using aGal-Cer revealed enhanced innate responses to influenza and improved disease outcome [39] , but challenge of CD1d 2/2 mice with influenza led to increased survival, implying an immuno-regulatory role for NKT cells in this model [40] . Whether or not these cells are critically involved in viral clearance, in our model, they represent a very small percentage (,1%) of the lymphocytes in the lung airways in the first five days following infection. Therefore, we believe that an abrogation of virallyinduced IL-15 in the lung airways most dramatically affects CD3 2 , NK1.1, NKp46 double positive NK cells responding rapidly to infection. By the same principle, exogenous IL-15 could be used therapeutically to increase NK cell populations in the BAL-primarily as a result of IL-15-induced migration-to enhance viral control. Therefore, these combined data indicate that influenza-induced IL-15 is an important signal for the migration of NK cells to the lung airways where they help limit viral replication. IL-15 is produced by a variety of cell types within and outside of the immune system, including dendritic cells, monocytes and macrophages, stromal cells, endothelial cells, and epithelial cells [41] , and pathogenic stimuli are known to induce this expression above constitutive levels [17, 42, 43, 44] . Although dendritic cells are known to be an important source of IL-15 at day 6 post influenza infection [27] , it is unclear whether DCs or other cell types produce IL-15 to facilitate specific immune responses to influenza. For example, lung epithelial cells constitutively express IL-15 and IL-15Ra [45] , and neutrophils and macrophages can be a major source of IL-15 that is produced during a variety of lung inflammatory diseases including sarcoidosis, tuberculosis, bronchitis, and asthma [46] . Following influenza infection, the IL-15producing cell type(s) is/are still unknown, but influenza-induced expression of IL-15 is clearly an important regulator of the NK and CD8 T cell responses to this virus [19] . Not only might there be different cellular sources of IL-15 at different times after infection, but IL-15 signaling in NK cells could also be regulated by the context in which IL-15 is presented. At least one isoform of IL-15 is known to enter the cell secretory pathway and could therefore be released as a soluble molecule [47, 48] . These data lend the idea that soluble IL-15 could bind heterotrimeric receptors on NK cells. However, IL-15 is also known to be transpresented to NK cells, and indeed, this mode of signaling appears to be most important for IL-15-mediated NK cell survival [49] . While IL-15Ra is dispensable for NK cell survival, IL-15Ra is thought to be an important component in signaling the migration of CD16 2 human NK cells into the endometrium, since this particular NK cell population expresses higher levels of this receptor component than CD16 + NK cells that do not migrate to IL-15 [36] . In our studies, IL-15Ra was only expressed on a relatively low proportion of NK cells in the lung airways of influenza-infected animals ( Figure 1 ). In contrast, CD122 signaling appears to be important for NK cell migration to IL-15 as this receptor chain is down regulated on NK cells exposed to IL-15c in vivo ( Figure 3 ). Thus, we consider it likely that signaling through IL-15Ra is not essential to IL-15-mediated of migration of NK cells to the lung airways. Therefore, future work is needed to identify the IL-15-producing cell population(s) and the mechanism of migration to this cytokine in order to target this response for eventual applications in influenza vaccines and treatments. A significant finding of our work is that NK cells immigrating into the lung airways are partially required for substantial trafficking of influenza-specific effector CD8 T cells to this location. We found that NK cells are necessary for the optimal accumulation of antigen-specific CD8 T cells, important effectors of eventual viral clearance [4, 5] , at the site of infection, since depletion of NK1.1-expressing cells resulted in a significant decrease in the number of influenza-specific CD8 T cells in the BAL. A subpopulation of CD8 T cells has been shown to express NK1.1 upon activation [50] , and while these cells were detected in the lung parenchyma of influenza-infected C57Bl/6 mice, they comprised less than 1% of the NP-tetramer+ CD8 T cells at this site alone and were never detected in the lung airways. This and the fact that the reduction of activated, influenza-specific CD8 T cells was observed only in the lung airways, and only at later time points post infection lead us to believe that direct depletion of NK1.1-expressing CD8 T cells cannot account for this reduction in treated mice. These data strongly suggest that NK1.1 + cells in the lung airways are requisite for the subsequent accumulation of influenza-specific CD8 T cells in the BAL and implicate IL-15mediated NK cell migration into the lung airways as a link between the innate and adaptive responses to influenza infection. We currently favor a model in which NK cells migrate directly to influenza-induced IL-15 in the lung airways, and, CD8 T cell trafficking, in turn, occurs to both IL-15 directly and to chemotactic factors produced by NK cells already present at the site. A positive feedback loop in which NK cells responding to IL-15 induce further expression of IL-15 by dendritic cells for the stimulation of CD8 T cells is an additional possibility [51] . Moreover, IL-15 has also been shown to modulate chemokine and chemokine receptor expression by NK cells and T cells [52, 53, 54] . As IL-15 deficiencies also cause reductions in CD8 T cell accumulation in the BAL [19] , chemotactic potential of IL-15 for NK cells presented here provides a possible link between IL-15-mediated effects of both the innate and adaptive immune responses to influenza infection. Future work will attempt to tease out the roles of direct and indirect migration of both NK cells and CD8 T cells to IL-15 following influenza infection. Regardless of mechanism, IL-15-induced trafficking of lymphocytes is an important part of our overall understanding of influenza immunobiology. The studies presented here emphasize the importance of IL-15 in mediating the innate response to influenza via the trafficking of NK cells, as viral titers were not as efficiently controlled early after infection in IL-15-blocked animals. These and previous studies also suggest that exogenous IL-15c could be used to modulate the immune response at both the innate and adaptive phases. Overall, we believe that IL-15 may be in important point of eventual immune intervention for treatment of primary infection and for adjuvanting vaccines.

Screening for Influenza A(H1N1)pdm09, Auckland International Airport, New Zealand

Entry screening for influenza A(H1N1)pdm09 at Auckland International Airport, New Zealand, detected 4 cases, which were later confirmed, among 456,518 passengers arriving April 27–June 22, 2009. On the basis of national influenza surveillance data, which suggest that ≈69 infected travelers passed through the airport, sensitivity for screening was only 5.8%.

T he virus that caused the 2009 infl uenza pandemic, hereafter referred to as infl uenza A(H1N1)pdm09, is mainly spread internationally by air travel (1) . To prevent or delay such spread, during the pandemic many countries initiated screening of air travelers arriving at airports, even though these measures have not been recommended by the World Health Organization (2) . On April 25, 2009, New Zealand was one of the fi rst countries outside the Americas to confi rm infl uenza A(H1N1)pdm09 in arriving airline passengers (3) . During April 27-June 22, 2009, at the direction of the Ministry of Health, the Auckland Regional Public Health Service began a screening program at Auckland International Airport. Protocols for border screening were updated throughout the pandemic and evolved as new information became available. Screening was initially applied to all passengers arriving or transferring on fl ights from countries where community transmission of infl uenza A(H1N1)pdm09 virus was occurring. The screening program included the following ( Figure) : • All fl ights notifi ed New Zealand of the health of passengers and crew on board before landing; if indicated, the aircraft was met by public health offi cials who triaged these travelers. • Cabin crew announced a scripted health message requesting passengers to identify themselves if symptomatic; after disembarkation, all passengers passed through a public health checkpoint where signage encouraged ill travelers to seek screening. • Information about infl uenza A(H1N1)pdm09 was available, those with symptoms could self-declare, and public health offi cials visually inspected arriving passengers and approached those with apparent symptoms. • Neither thermal scanning nor active screening of every arriving passenger was used. Each unwell passenger and crew member was screened for infl uenza-like illness by a nurse and assessed by a medical offi cer if illness met the defi nition of a suspected case. Those whose illness met the case defi nition had nasopharyngeal swabs taken, were offered oseltamivir, and were sent home or to a facility for isolation. Reverse transcription PCR (RT-PCR) was used to confi rm infection. Screening was escalated on April 29 to include all passengers arriving from other countries and stopped on June 22. We quantifi ed the results of entry screening for infl uenza A(H1N1)pdm09 at Auckland International Airport. Using the information generated during screening, we retrospectively estimated the number of infected travelers who actually passed through the airport. To estimate the sensitivity of screening, we then compared screening fi ndings with the expected number of infected travelers who passed through the airport. Ethical approval was received from the Northern X Regional Ethics Committee of the New Zealand Ministry of Health. The number of screened passengers was obtained from airport records. The numbers of crew members on inbound international aircraft were estimated by using averages for fl ights into Auckland. The number of travelers detected at each step and referred to the next step of the screening process was obtained from Auckland Regional Public Health Service records. Virologic test results were extracted from laboratory information systems. A confi rmed case was one that met the current case defi nition (as published on the Ministry of Health website, www.health.govt.nz) and one for which RT-PCR result was positive. We estimated the number of infected travelers screened as the total number of confi rmed cases in New Zealand during this period, multiplied by the proportion of overseas-acquired cases, and the proportion of international travelers arriving at the airport. On April 30, 2009, nonseasonal infl uenza A (H1N1) was made notifi able, (4). During the screening period, 456,518 international travelers were screened; 406 (0.09%) of these were referred for medical assessment. Of those, 109 (27%) met the case defi nition and received virologic testing. RT-PCR results were located for 89 (82%), among which 4 were positive. A total of 312 cases were confi rmed. The proportion of travelers who acquired the infection overseas was 32%. The proportion who passed through the airport was 69%. The expected number of infected travelers estimated to have passed through the border during the screening program was therefore 69, giving an estimated sensitivity of 5.8% (95% CI 2.3%-14.0%). The infl uenza A(H1N1)pdm09 screening program at Auckland International Airport had low sensitivity. This form of border screening is therefore unlikely to have substantially delayed spread of the pandemic into New Zealand in 2009. Limitations of infl uenza screening include the high proportion of asymptomatic infected travelers (5), incubation of infections acquired before or during a fl ight (3), reliance on self-identifi cation, limitations of case defi nitions, and limitations of thermal scanning (6) . Modeling data have shown that the ability of border screening to delay global pandemic infl uenza is closely linked to the effectiveness of the screening process or travel restriction used. To delay infl uenza spread by 1.5 weeks, border restrictions need to reduce imported infections by 90% (7) . The entry screening program we describe does not meet these standards. The potential effectiveness of screening arriving travelers to prevent or delay infl uenza epidemics has been debated. Mathematical models and literature reviews have argued for (7, 8) and against (9-11) this approach. Some authors have found that entry screening for respiratory conditions or infl uenza A(H1N1)pdm09 is insensitive and not cost-effective (12) . Border screening did not substantially delay local transmission of infl uenza A(H1N1) pdm09 (13) . This study has several limitations, particularly with regard to estimating the number of infected travelers who would have passed through the airport during the screening period. Most cases of illness acquired overseas would probably not have been notifi ed, particularly those in patients with mild illness who did not see a doctor or who saw a doctor but did not receive a diagnosis. The estimated proportion of overseas-acquired cases was based on data from the fi rst 100 cases and would have decreased as the pandemic progressed. The net effect of these factors is unknown, but they would probably have increased the estimated number of undetected infected travelers passing through screening, thereby further reducing the estimated sensitivity of screening. Border screening might be conducted for reasons other than preventing or delaying an epidemic. It might provide public assurance and confi dence that something is being done (14) . The communication of health information and advice on how to seek treatment is consistently recommended as a pandemic prevention strategy (12, 15) and is usually delivered as part of border screening programs. These benefi ts need to be balanced against the considerable resources used, opportunity cost (resources used for this activity and thereby unavailable for other activities), uncertain effectiveness, and inconvenience of border screening. To delay or prevent infl uenza entry at borders, infl uenza screening needs to be considerably more effective than the mostly passive program described here. We hope that during this interepidemic period, a major international review of the role of international air travel in the dissemination of emerging infectious diseases will be conducted to identify effective interventions. Such a review should consider systemwide approaches, including exit screening, standardized health declarations, active screening of individual passengers (including use of rapid laboratory tests and thermal scanning), passenger tracking, policies and practices that support sick travelers wishing to defer travel, and circumstances where airline travel should be suspended entirely.

Immunologic Changes during Pandemic (H1N1) 2009, China

We analyzed changes in immunologic values over time for 28 hospitalized patients with pandemic (H1N1) 2009. Levels of interleukin-6, interferon-γ, and interleukin-10 increased 1 day after illness onset and then decreased to baseline levels. Levels of virus-specific antibody were undetectable 1 day after illness onset and peaked 36 days later.

mm 3 in 9 (32.1%) of 28 patients and represented <55% of total lymphocyte counts in 4 (14.3%) patients (Table). CD4 T-lymphocyte counts were <400 cells/mm 3 in 13 (46.4%) patients and represented <31% of total lymphocyte counts in 8 (28.6%) patients. CD8 T-lymphocyte counts were <190 cells/mm 3 in 3 (10.7%) patients. B-cell counts were <90 cells/mm 3 in 1 (3.6%) patient. Natural killer cells represented >27% of lymphocyte counts in 4 (14.3%) patients. Fourteen (50.0%) patients had a CD4:CD8 ratio less than the standard reference ratio of 1.4. Flow cytometric results showing development of peripheral blood lymphocyte subsets during the disease course were divided into 3 groups on the basis of time of illness onset until date of blood sample collection: 1-3, 4-6, or 7-10 days. These groups were used because the longest time from onset of illness to hospitalization was 3 days for all patients, and the peak temperature for patients was observed 3 days after illness onset. Mean counts and percentages of all T-and B-lymphocyte subsets increased after 3 days of illness compared with results obtained during the fi rst 3 days of illness ( Figure 1 ). However, the increase in CD8 and B cells was not signifi cant. Another study showed a decrease in CD4, CD8, and B cells ≤2 days of symptom onset in patients with pandemic (H1N1) 2009 than in healthy persons (4). Our results show impaired adaptive immune responses and a gradual increase during recovery in mildly affected patients. We measured serum cytokine concentrations and hemagglutination inhibition (HAI) antibody titers in patients during hospitalization and the follow-up period ( Figure 2 ). We observed an increase in interleulin-6 (IL-6) levels 1 day after illness onset, which were 6.0-fold higher than the baseline level, and a 2.3-fold increase in interferon-γ (IFN-γ) levels. These levels decreased to baseline levels 5 days after illness onset, although the IL-6 level 5 days after illness onset was higher than levels 15 and 37 days after illness onset. The maximum IL-10 level 1 day after symptom onset was 3.2-fold higher than the baseline level. This level decreased to a value lower than the baseline level within 4 days, and then gradually increased to the baseline level 37 days after illness onset. Serum IL-6, IFN-γ, and IL-10 levels were not related to patient temperature 1 day after symptom onset, peak temperature during the disease, or period of fever. These levels showed minor differences that were not related to cough or sore throat in patients. Only 1 patient had an HAI antibody titer ≥10 (titer 20) 1 day after illness onset. The HAI geometric mean titer increased 5 days after symptom onset compared with that 1 day after symptom onset and continued to increase until it reached a peak level of 137.9 at 37 days after symptom onset (25.5-fold increase). Peak HAI antibody titers ≥40 and ≥4-fold increases were observed in 27 (96.4%) patients. Bermejo-Martin et al. reported increased serum levels of IL-6, IFN-γ, and tumor necrosis factor-α in patients with pandemic (H1N1) 2009 during the fi rst 5 days after symptom onset; no difference in levels of these 3 cytokines was observed in patients with mild disease and controls (5) . However, similar to another report (4), we detected increases of IL-6 and IFN-γ levels in patients with mild disease during the fi rst 3 days after symptom onset. These different patterns may be caused by different intervals from time of symptom onset to date of sample collection (5 days vs. 3 days) because IL-6 and IFN-γ levels in our study quickly decreased to baseline levels ≤7 days after symptom onset. These results suggest that serum IL-6 and IFN-γ , and IL-6 are medians (pg/mL). Serum HAI antibody titers were transformed by using the natural logarithm and are shown as means. Baseline cytokine concentrations on the y-axis are values for healthy persons. *p<0.05 when IL-6 or HAI antibody levels were compared with those at day 5; †p<0.05 when IL-10 level was compared with those at baseline; ‡p<0.05 when IL-10 level was compared with those at days 5 or 15; §p<0.05 when value was compared with that at any other time point; ¶p<0.05 when value was compared with those at days 5, 15, or 49. levels may be increased in patients with pandemic (H1N1) 2009 within the fi rst 3 days after symptom onset, followed by a decrease to baseline levels ≤5 days after symptom onset in patients with mild disease or a continuous increase in severely affected patients. IL-6 and IFN-γ are associated with antiviral immune responses during infl uenza infection (6) (7) (8) . However, continuous, excessive release of IL-6 three days after illness onset likely contributed to serious pulmonary infl ammation and tissue injury, as has been documented for severe acute respiratory syndrome and 1918 pandemic infl uenza, but this release could be tempered by production of IL-10 (6, 7, (9) (10) (11) . The proportion of persons 18-60 years of age with a ≥4-fold increase in HAI titer who received 1 dose (15 μg) of monovalent pandemic (H1N1) 2009 nonadjuvant vaccine was 96.2%, and the proportion with an increased HAI titer ≥40 was 97.1%, results similar to those of a recent study (12) . However, the geometric mean titer in healthy vaccinated persons was 237.8, a 34.5-fold increase over the prevaccination titer, which was greater than that for patients naturally infected with pandemic (H1N1) 2009 virus (12) . This fi nding may have resulted from impaired adaptive immune responses against pandemic (H1N1) 2009 virus in the initial phase, which included decreased numbers of CD4 and B lymphocytes and an increase in T regulatory cells (4) . In conclusion, our data indicated changes in cellular profi les during pandemic (H1N1) 2009 virus infection; showed that transient production of IL-6, IFN-γ, and IL-10 are main effectors of the early innate immune response against pandemic (H1N1) 2009 virus; and indicated that adaptive immune responses are impaired in the initial phase after infection. These factors may help clarify the pathogenesis of pandemic (H1N1) 2009 virus and provide new approaches in overcoming severe infections.

Pandemic (H1N1) 2009 Risk for Frontline Health Care Workers

To determine whether frontline health care workers (HCWs) are at greater risk for contracting pandemic (H1N1) 2009 than nonclinical staff, we conducted a study of 231 HCWs and 215 controls. Overall, 79 (17.7%) of 446 had a positive antibody titer by hemagglutination inhibition, with 46 (19.9%) of 231 HCWs and 33 (15.3%) of 215 controls positive (OR 1.37, 95% confidence interval 0.84–2.22). Of 87 participants who provided a second serum sample, 1 showed a 4-fold rise in antibody titer; of 45 patients who had a nose swab sample taken during a respiratory illness, 7 had positive results. Higher numbers of children in a participant’s family and working in an intensive care unit were risk factors for infection; increasing age, working at hospital 2, and wearing gloves were protective factors. This highly exposed group of frontline HCWs was no more likely to contract pandemic (H1N1) 2009 influenza infection than nonclinical staff, which suggests that personal protective measures were adequate in preventing transmission.

A ustralia was affected early in the (H1N1) 2009 infl uenza pandemic with 37,636 cases and 191 deaths reported. The state of Victoria was the fi rst to observe a substantial peak in the number of persons infected (1) . The pandemic was managed within the framework of the Australian Health Management Plan for Pandemic Infl uenza (2) . Guidelines for use of personal protective equipment (PPE) were established in the Victorian Health Management Plan for Pandemic Infl uenza (3) . Recommendations included use of N95 masks, gloves, protective eyewear, and longsleeved gowns. Infl uenza in health care workers (HCWs) is common, and acquisition in the workplace is well documented. An uncontrolled study found that after an infl uenza epidemic in Glasgow, Scotland, 120 (23.2%) of 518 HCWs seroconverted (4) . Early in 2009, twelve HCWs with probable or possible work place acquisition of pandemic infl uenza were reported in the United States. None had worn full PPE (5) . That HCWs may be concerned about attending work during a potentially serious infl uenza pandemic is not surprising. During the severe acute respiratory syndrome outbreak of 2003, some HCWs reportedly stayed at home for fear of becoming infected and transmitting infection to family members. A number of surveys have found that 16%-33% of HCWs may not report to work in the event of an infl uenza pandemic (6) (7) (8) (9) . HCWs need to know the transmission risks to make rational decisions about working during an infl uenza pandemic. Because HCWs are exposed in the community as well as the workplace, they should know about the additional risks for contracting infl uenza at work. This information is also imperative for pandemic workforce planning. We sought to determine whether frontline HCWs were at greater risk for contracting pandemic (H1N1) 2009 infl uenza than the control group. Additionally, we sought information on factors that may have increased or decreased the risk for infection. We conducted a cohort study, comparing frontline HCWs with intensive patient contact (clinical) and staff with no patient contact (nonclinical). Frontline HCWs were defi ned as those who worked >1 shift per week and had likely exposure to patients with pandemic infl uenza infection. These workers included doctors, nurses, and physiotherapists, as well as others in the emergency department, intensive care unit, infectious diseases units, and respiratory and other wards where patients with suspected pandemic infl uenza were housed. Staff members who had no clinical contact were chosen as a convenient surrogate for a community control group. These workers included university and hospital staff in nonpatient contact areas such as the library, information technology, and administration. This study was approved by the Human Research Ethics Committees at each of the hospitals and all participants gave written informed consent. The study was conducted from August 24, 2009, through December 16, 2009. Four tertiary referral hospitals in metropolitan Melbourne were involved: Royal Melbourne, St Vincent's, Austin, and Alfred Hospitals. At all sites, patients with suspected or confi rmed pandemic infl uenza infection were cared for in negative pressure isolation rooms when they were available, and in private rooms when they were not. Institutional infection control policies directed that gloves, gowns, goggles, and masks be used when caring for these patients. Use of N95 masks was initially recommended in all hospitals, although hospital 1 changed to surgical masks after June 16, 2009 . Hand hygiene with an alcohol-based product and respiratory etiquette were promoted at all hospitals. The progression of the pandemic in Victoria is shown in Figure 1 . The original research plan was to obtain 2 serum samples, 3 months apart, from all participants to test for seroconversion and also to obtain weekly nose swabs for pandemic infl uenza detection by using real-time PCR. By the time the study commenced, the pandemic was waning, infl uenza cases were decreasing in Victoria, and following the original study plan was not considered feasible. The plan was thus modifi ed. An initial serum sample was obtained from all participants to measure for pandemic infl uenza antibodies. At study entry, participants completed a Web-or paper-based questionnaire that requested information on demographic characteristics, known infl uenza exposures outside the workplace, and any history of fever or respiratory symptoms occurring during the pandemic but before the study. In addition, the clinical group was asked about work exposure to patients with suspected pandemic infl uenza and their usual use of PPE when caring for these patients. Participants were also asked about use of neuraminidase inhibitors (NIs) and specifi cally whether they received prophylaxis after exposure to a patient with confi rmed infl uenza. Participants were instructed to provide nose swab specimens for viral testing if they experienced signs and symptoms, including cough, sore throat, rhinorrhea, laryngitis, fever, myalgias, or headache. All were asked to complete a weekly questionnaire regarding symptoms, infl uenza exposure, and use of NIs. If a participant reported respiratory illness, a second serum sample was requested for antibody testing to document possible seroconversion. Serum was tested for antibodies to pandemic (H1N1) 2009 infl uenza virus by using the hemagglutination inhibition assay with A/California/7/2009 virus and turkey red blood cells (10) . A titer of <40 was defi ned as negative and >40 as positive. Nucleic acid detection was performed on nasal swabs by using reverse transcription PCR (RT-PCR) for infl uenza-specifi c and pandemic (H1N1) 2009 virus-specifi c sequences on swabs; kits were provided by the Centers for Disease Control and Prevention (Atlanta, GA, USA) (11) and an ABI-7500FAST instrument at the World Health Organization (WHO) Collaborating Centre for Reference and Research on Infl uenza in Melbourne. On the basis of early estimates of antibody positivity to pandemic infl uenza virus in the community, we assumed 20% infection rates in clinical staff and 10% rates in nonclinical staff. We calculated that 438 participants were required to achieve 80% power to detect this difference using a 0.05 two-tailed signifi cance level. The primary outcome was the presence of a positive antibody titer in the fi rst serum sample, indicating likely pandemic infl uenza infection. We performed 2 separate univariate and multivariate analyses to delineate putative risk and protective factors (1 included all participants and the other included clinical participants only) to investigate any association between health care-specifi c risk factors and pandemic infl uenza. Multivariate analysis was performed by using forward and backward stepwise logistic regression, including all variables in the model initially and a p value for removal The study took place from August 24, 2009, through December 16, 2009, largely before release of the pandemic infl uenza vaccine, and no participant was vaccinated during the study. Table 1 shows the number of patients who had confi rmed pandemic infl uenza infection (by PCR) and were treated in each of the hospitals. Characteristics of study participants are shown in Table 2 . The median participant age was 38 years (range 18-74 years); 27% were <30 years of age, 20% were 30-39 years of age, 25% were 40-49 years of age, and 20% were >50 years of age. Figure 2 shows the reverse cumulative distribution of fi rst serum antibody titers, according to age. We found no statistically signifi cant difference between the curves (p = 0.11 by ordinal logistic regression). On multivariate logistic regression, the only factor associated with a higher risk for pandemic infl uenza among all participants was younger age (OR 0.96, 95% CI 0.94-0.99) after adjustment for participant status (clinical vs. nonclinical), age, gender, hospital, seasonal infl uenza vaccination, confi rmed pandemic infl uenza, reported respiratory illness, community contact with infl uenza, oseltamivir prophylaxis, number of children in the household <18 years of age, and hours worked per week. On univariate analysis, the only factors that were signifi cantly associated with protection against infection in the clinical group were use of any mask (OR 0.16, 95% CI 0.03-0.97) and use of gloves (OR 0.09, 95% CI 0.02-0.5) for patients in droplet precautions. Adjusted odds ratios are shown in Table 3 . Of the 395 participants, 140 (35%) reported a respiratory illness and 46 had nose swabs taken. Seven were positive for pandemic (H1N1) 2009 virus by PCR, 1 for subtype H3N2 infl uenza, and 38 were negative. One of the 46 had 2 swabs taken during different illnesses; the fi rst was positive and the second was negative for pandemic (H1N1) 2009 virus. PCR cycle threshold values for swab specimens were from 30 to 40, indicating low viral loads. This fi nding may indicate that poor swabbing techniques were used, that the sample had been taken as infection was waning, or that level of infection was low (data not shown). For 87 participants, a second serum sample was taken because of a reported respiratory illness. The average number of days between the fi rst and second sample was 60 days (range 28 to 122 days, median 54) days. Thirtysix participants who had nose swabs performed also had a second serum sample taken. Seroconversion occurred in only 1/87 workers, with an initial titer of <10 and a subsequent titer of 40 (76 days later). This participant had a nose swab taken during a respiratory infection, which was negative for infl uenza virus. Seroconversion did not occur in any of the participants with a positive nose swab specimen. The mean number of days from obtaining a One participant with a positive nose swab sample did not have a second serum sample taken. None of the participants with a positive nose swab or seroconversion reported taking NIs in their weekly survey. Four of the 7 participants with a positive PCR result and the 1 in whom seroconversion occurred were in the clinical group (3 doctors, 1 pharmacist, 1 nurse, 1 physiotherapist). The participant who showed seroconversion was 29 years of age; participants with a positive PCR result ranged from 24-63 years of age. Two of the participants with a positive PCR result worked on the infectious disease ward, 2 in the emergency department, and 1 in the intensive care unit; seroconversion occurred in the participant who worked in a medical ward. Five of the participants with positive PCR results and the participant in whom seroconversion occurred had received the 2009 and previous seasonal infl uenza vaccines. None of the participants with confi rmed infl uenza reported taking oseltamivir for either prophylaxis or treatment. In total, 395 participants completed 1-13 weekly questionnaires each. Eighty-nine clinical and 51 nonclinical participants reported 139 and 91 respiratory illnesses, respectively. No participant reported having laboratoryconfi rmed pandemic (H1N1) 2009 infl uenza. Six reported community contact with someone who had laboratoryconfi rmed infection. One reported taking oseltamivir after contact with an infected person in the workplace. This person had 2 serum samples taken 88 days apart; both had an antibody titer of <10. In this study, we evaluated the risk for pandemic (H1N1) 2009 in HCWs compared with the risk for such infection in a control group, as well as the factors associated with infection. HCWs had slightly higher rates of seropositivity than nonclinical staff; however, this difference was not statistically signifi cant. Our data are supported by results of another recent study, which found that being a HCW was not a risk factor for serologically confi rmed seasonal infl uenza virus infection and that the risk of HCWs acquiring infl uenza was more strongly associated with household than workplace exposure (12) . That study found a seroconversion rate of 11.2% in HCWs and 10.3% in non-HCWs. However, it examined only doctors and nurses, whereas our study included other types of frontline HCWs. Another study reported a seroprevalence for pandemic (H1N1) 2009 of 26.7% in HCWs, which was not signifi cantly different from the seroprevalence of the general population (13) . Neither of these studies examined use of PPE. Overall, we found that 17.7% of participants had serologic evidence of pandemic (H1N1) 2009 virus infection after the peak of the outbreak. This proportion refl ects the observed 16% seroprevalence in adults in Melbourne (14) . These rates are lower, however, than the 31.7% antibody positivity found in South Australia during a prelicensure study of pandemic infl uenza vaccine in July 2009, which excluded subjects with confi rmed or suspected pandemic (H1N1) 2009 infl uenza (15) . This difference in titers may have refl ected geographic differences in infection rates or differences between the populations sampled. In the analysis of all participants, we found that older age was associated with lower rates of pandemic (H1N1) 2009 infl uenza infection. We did not observe higher levels of preexisting antibodies against pandemic (H1N1) 2009 infl uenza with increasing age, which has previously been reported. However, results of other studies examining the relationship between seroprevalence and increasing age are confl icting (15) (16) (17) (18) . Immune mechanisms other than typespecifi c antibodies may be providing protection for older participants. Other possibilities are that older persons have older children who may be less likely to acquire or transmit infl uenza or that older participants were more conscientious with respiratory etiquette and hand hygiene; attempts to measure these factors were not included in this study. Among the HCWs we studied, working at hospital 2 conferred protection against pandemic (H1N1) 2009 virus infection. This hospital was in a geographic area with fewer cases than the others, but if this were the explanation, then a similar fi nding might have been expected in the nonclinical group, which was not demonstrated. Furthermore, at least as many cases of confi rmed pandemic (H1N1) 2009 infl uenza were seen at hospital 2 as were seen at the other hospitals (Table 1 ). Factors such as reported compliance with PPE, were adjusted for in the multivariable analysis to reduce the effect of hospital type on infl uenza risk. The reason for the lower risk associated with hospital 2 has not been identifi ed but may relate to other unmeasured factors, such as compliance with hand hygiene procedures. Wearing gloves while caring for patients as part of droplet precautions was strongly associated with a lower risk of having had pandemic (H1N1) 2009 virus infection. Use of gloves was highly correlated with use of gowns, masks, and eye protection on logistic regression (results not shown). This fi nding confi rms the great importance of PPE in preventing transmission of respiratory viruses in the health care setting and may explain why HCWs with defi nite exposure to infl uenza in the workplace, in addition to probable exposure in the community, do not have higher rates of infection than those with only community exposure. The risk for pandemic (H1N1) 2009 virus infection increased with the number of children <18 years of age living in the participant's household, which has previously been reported as a risk factor (12) . In Victoria, the median age of persons with reported pandemic (H1N1) 2009 virus infection was 15 years, with 67% of all notifi ed casepatients being 5-17 years of age (1). Miller et al. also found that children were predominantly infected (17) . This fi nding, coupled with the diffi culties of maintaining good respiratory etiquette in young children, is a plausible explanation for the effect of child number on infection risk. Working in the ICU was also identifi ed as a risk factor for pandemic infl uenza; patients in ICU may be severely ill, with high viral loads, and staff may be heavily exposed during multiple aerosol-generating procedures. In addition, use of PPE and hand hygiene compliance may have been lower than in other wards or patients with pandemic infl uenza may have been unrecognized and therefore appropriate PPE not used. Exposure of HCWs to suspected or proven pandemic infl uenza in the community was protective against having a positive antibody test result. This fi nding is counterintuitive and diffi cult to explain. One hypothesis is that HCWs who knew that they had had community exposure may have been more attentive to hand hygiene and other infection control precautions while at work or were more likely to enact social distancing. We found only 1 instance of seroconversion among the 87 participants (including the 6 with PCR-confi rmed infection), each of whom had 2 serum samples taken for antibody measurement. Miller et al. reported that 89.1% of participants with pandemic (H1N1) 2009 had an antibody titer of >32 three weeks after infection, although a baseline serum sample was not taken; therefore, seroconversion could not be demonstrated (17) . None of the participants with positive PCR results reported taking NIs, and all had serum samples taken >2 weeks after the positive nose swab specimen, allowing suffi cient time for seroconversion. Our results are likely to be true positives, as all swabs were only taken when patients were symptomatic. Previously, virus isolation has been the gold standard for infl uenza detection but RT-PCR is now considered to be more sensitive and specifi c. A previous study by some of the current authors has shown that seroconversion occurs in 80%-90% of serum samples if they are tested a suffi cient time after infection (confi rmed by RT-PCR) (19) . Nasal swabs are a relatively peripheral type of sample (20) . If viral load is low in the nose, the sample may be insuffi cient as an antigenic stimulus to induce a detectable level of seroconversion in the serum. This may be the explanation for the lack of seroconversion seen in some PCR-positive cases in this study. Because the number of pandemic (H1N1) 2009 cases in Victoria was low by the time this study commenced, we used a single antibody measurement for diagnosis in most patients. This is not ideal, because some participants may have had preexisting cross-reactive antibodies, as reported by others (15, 16) . However, this cross-reactivity has been most commonly found in older persons >65 years of age, a population which was underrepresented in our study. The explanation given for presence of cross-reactive antibodies in older persons has been past exposure to other antigenically similar viruses or a lifetime exposure to infl uenza A virus (17) . Because this exposure could not have occurred in our younger study participants (median age 38 years) and serum samples were collected toward the end of the pandemic wave when many would have already been exposed, reactivity likely was specifi c to pandemic (H1N1) 2009. These factors support the use of a single antibody measurement for diagnosis. This study relied on self-reported symptoms and risk factors, including use of PPE, making it subject to recall bias. This is a particular problem potentially for recall of exposures (e.g., to others with infl uenza or for use of PPE). However, many of the predictor variables were not subject to recall bias (e.g., clinical or nonclinical status, work place, age, gender, occupation, and number of children in the household). In addition, in order to infl uence the results, the 2 exposure groups would have had to exhibit differential recall. Although it could be postulated that HCWs may have perceived that they were at greater risk for exposure and may have therefore been more conscientious when fi lling out questionnaires, we believe that because of the large amount of public awareness of pandemic (H1N1) 2009 at that time, it is unlikely that this group would have been more conscientious than the nonclinical group. In conclusion, we found that HCWs did not have a substantially increased risk of contracting pandemic (H1N1) 2009 in a health care setting with high availability of PPE. We conclude that use of PPE was highly protective against acquiring pandemic (H1N1) 2009 virus infection, and we therefore encourage its use, along with scrupulous hand hygiene and respiratory etiquette.

Increased Urinary Angiotensin-Converting Enzyme 2 in Renal Transplant Patients with Diabetes

Angiotensin-converting enzyme 2 (ACE2) is expressed in the kidney and may be a renoprotective enzyme, since it converts angiotensin (Ang) II to Ang-(1-7). ACE2 has been detected in urine from patients with chronic kidney disease. We measured urinary ACE2 activity and protein levels in renal transplant patients (age 54 yrs, 65% male, 38% diabetes, n = 100) and healthy controls (age 45 yrs, 26% male, n = 50), and determined factors associated with elevated urinary ACE2 in the patients. Urine from transplant subjects was also assayed for ACE mRNA and protein. No subjects were taking inhibitors of the renin-angiotensin system. Urinary ACE2 levels were significantly higher in transplant patients compared to controls (p = 0.003 for ACE2 activity, and p≤0.001 for ACE2 protein by ELISA or western analysis). Transplant patients with diabetes mellitus had significantly increased urinary ACE2 activity and protein levels compared to non-diabetics (p<0.001), while ACE2 mRNA levels did not differ. Urinary ACE activity and protein were significantly increased in diabetic transplant subjects, while ACE mRNA levels did not differ from non-diabetic subjects. After adjusting for confounding variables, diabetes was significantly associated with urinary ACE2 activity (p = 0.003) and protein levels (p<0.001), while female gender was associated with urinary mRNA levels for both ACE2 and ACE. These data indicate that urinary ACE2 is increased in renal transplant recipients with diabetes, possibly due to increased shedding from tubular cells. Urinary ACE2 could be a marker of renal renin-angiotensin system activation in these patients.

Angiotensin-converting enzyme 2 (ACE2) is a recently identified member of the renin-angiotensin system (RAS) that degrades angiotensin (Ang) II to the seven amino acid peptide fragment Ang-(1-7) [1, 2] . ACE2 is a homologue of angiotensin-converting enzyme (ACE), but is not blocked by ACE inhibitors. Although ACE2 is found in many tissues, expression is especially high in the kidney, particularly within cells of the proximal tubule [3] [4] [5] . In mice deletion of the ACE2 gene is associated with development of late-stage glomerulosclerosis, and acceleration of diabetic nephropathy, in the absence of hypertension [6, 7] . In spontaneously hypertensive rats, administration of human recombinant ACE2 reduces blood pressure [8] , and in diabetic mice, exogenous human ACE2 diminishes blood pressure and glomerular injury [9] . Thus, ACE2 may be an endogenous protector against the progression of chronic kidney disease (CKD). In kidney tubular epithelial cells, ACE2 is localized to the apical membrane and also appears in the cytoplasm [3, 10] . ACE2 is shed at its carboxy-terminus from the plasma membrane in cultured human embryonic kidney cells and airway epithelial cells, a process catalyzed by the enzyme ''a disintegrin and metalloproteinase-17'' (ADAM-17) [11] [12] [13] . Whether this process occurs in the proximal tubule is unclear, although soluble ACE2 has been detected in human urine [10] . In a recent study, urinary levels of ACE2 protein were significantly increased in humans with CKD (the majority with chronic glomerulonephritis), compared to healthy controls, as determined by enzyme-linked immunosorbent assay (ELISA) [14] . Urinary ACE2 was also higher in diabetics with CKD [14] . These results suggest that ACE2 may be shed into the urine, and could be a biomarker in CKD patients. However, the presence of urinary ACE2 has not been studied in renal transplant recipients, and the factors associated with elevated urinary ACE2 remain unclear. Accordingly, the principle objective of the present study was to determine if urinary ACE2 activity, protein, and mRNA can be detected in renal transplant patients, and to identify factors associated with the presence of ACE2. In addition, we examined factors associated with urinary ACE activity, protein and mRNA in these patients. Our data indicate that urinary ACE2 is increased in renal transplant recipients with diabetes, possibly due to increased shedding from tubular cells. This study involved recruitment of human subjects as described below, with written informed consent, and the study was conducted according to the principles expressed in the Declaration of Helsinki. The study protocols were approved by the Research The subjects in this study were 50 healthy controls (age .18 yrs), recruited from the hospital or research centre staff, with no history of kidney disease, hypertension, or diabetes, and 100 renal transplant recipients from The Ottawa Hospital Renal Transplant Program, age .18 yrs, and .3 months posttransplant. At the time of enrollment, half of the transplant subjects (n = 50) were also enrolled in an ongoing randomized controlled trial to determine the effect of the ACE inhibitor ramipril on transplant outcomes (ACE inhibition for the preservation of renal function and survival in kidney transplantation; International Standard Randomized Controlled Trial Number Registry ISRCTN-78129473), but had not yet received either placebo or ramipril. These subjects had documented significant proteinuria (.200 mg urinary protein/day) at baseline. The 50 other transplant recipients were patients without significant proteinuria (spot urine ACR in the normal range, or ,200 mg proteinuria/day). Subjects were excluded if they were taking ACE inhibitors, angiotensin receptor antagonists, or renin inhibitors, or if they were pregnant or currently had a urinary tract infection. After obtaining informed consent, a spot urine sample was collected from each subject, and a blood sample was drawn for measurement of serum creatinine (Cr) and determination of estimated glomerular filtration rate (eGFR), using the Modification of Diet in Renal Disease (MDRD) calculation [15] . Demographic information was obtained via interview with the subject and/or from the hospital chart. For transplant recipients, a diagnosis of diabetes mellitus and the primary renal diagnoses were obtained directly from survey of the hospital chart. Urine samples were placed on ice, aliquoted and then centrifuged at 12000 g for 5 min at 4uC. Measurements of urinary albumin were performed on supernatant fractions, using an ELISA kit (Exocell Inc., Philadelphia, PA, USA). Results were corrected for urinary Cr concentration, using a kit specific for human Cr (Creatinine Companion, Exocell Inc.). The enzyme activities of urinary ACE2 and ACE were measured using synthetic substrates, essentially as we previously reported [16] . All results were corrected for the Cr concentration in the urine samples. Details on the assays can be found in Text S1. The amount of ACE2 present in urine specimens was quantified using a commercial ELISA kit (Cat. No. AG-45A-0022EK-KI01, AdipoGen, Seoul, Korea) according to the protocol provided by the supplier (http://www.adipogen.com/ ag-45a-0022/ace2-human-elisa-kit.html). A standard curve was generated by performing 1:2 serial dilutions of human recombinant ACE2 (50 ng/ml), provided with the kit, with the limit of detection ranging from 0.391 to 25 ng/mL. In preliminary experiments, the average intra-assay coefficient of variation (CV) for the assay was 2.9%, and the average interassay CV value was 8.7% (n = 10). The amount of ACE2 obtained by ELISA was normalized to the subject's urine Cr concentration, and is reported as ng/mg Cr. Urine aliquots (supernatant fraction, 15 mL) were subjected to immunoblot analysis for ACE2 and ACE, using commercially available antibodies. To control for variations in urine concentration, the values obtained by densitometry were divided by the corresponding Cr concentration for that urine sample. Details on the assays can be found in Text S1. Peptide N-Glycosidase F (PNGase F) Treatment Urine aliquots (supernatant fraction, 20 mL) were subject to deglycosylation reaction using PNGase F (Cat No. P0704S, New England Biolabs, Ipswich, MA, USA). Deglycosylated urinary ACE2 protein fragments were detected by western analysis. Details of the assays can be found in Text S1. Urine samples (40 mL) were centrifuged at 1000 g for 20 min at 4uC. Total RNA was isolated from pellet fractions and then subjected to real-time RT-PCR for quantitation of ACE2 and ACE (see Text S1 for assay details). Urinary levels of Ang II were measured using a commercial peptide radioimmunoassay (RIA) kit that contains an Ang IIselective polyclonal antibody and 125 I-Ang II (Peninsula Laboratories, San Carlos, CA, USA), essentially as described [16, 17] . Ang-(1-7) levels were measured using a commercial peptide enzyme immunoassay (EIA) kit that contains an Ang-(1-7)-selective polyclonal antibody (Peninsula Laboratories) [16] . For assay details, see Text S1. Data are presented visually as box plots. Continuous variables are reported as the median value, with the interquartile range in parentheses. Statistical analysis was performed by the nonparametric Mann-Whitney U test for unpaired samples to compare data in healthy controls and transplant recipients, and diabetic vs non-diabetic transplant recipients. A linear regression model was constructed to identify potential explanatory variables for urinary ACE2 and ACE levels in the 100 transplant subjects. The dependent variable was the measure under study (e.g. ACE2 activity) while the following explanatory variables were entered into the model: age, gender, diabetes, eGFR, albuminuria, hypertension, and use of calcineurin inhibitors. Variables were forced in simultaneously and removed from the model if not statistically significant. Standardized regression coefficients (b) are presented for each dependent variable in the model. Data were analyzed using SigmaStat (version 3.01 A; SYSTAT) and JMP (version 8.0.1, SAS Inc.). For all data, a p value ,0.05 was considered significant. The characteristics of the 50 healthy controls and 100 transplant subjects are depicted in Table 1 . The median age for control subjects was 45 [interquartile range (IQR), 36-50] yrs, and 26% were male. The median age for transplant subjects was 54 (IQR, 42-62) yrs, 65% were male, 38% had diabetes, and 88% had hypertension. Urinary ACE2 Activity, Protein, and mRNA As shown in Figure 1A , urinary ACE2 activity was significantly increased in transplant patients, compared to control subjects [control; median: 1.90 (IQR, 1.04-2.70) ng-eq/mg Cr610 2 vs transplants; median: 3.65 (IQR, 1.13-9.35) ng-eq/mg Cr610 2 ; p = 0.003]. Similarly urinary ACE2 protein by ELISA was significantly increased in transplant patients, compared to controls ( Figure 1B ) [control; median: 1.41 (IQR, 0.00-4.49) ng/mg Cr vs transplants; median: 8.04 (IQR, 0.00-21.11) ng/mg Cr; p,0.001]. In 96% of unconcentrated urine samples from control and 94% from transplant subjects, ACE2 was identified by western analysis as a protein doublet of 120 kDa and 90 kDa. As with urinary ACE2 activity and ELISA measurements, western analyses revealed a significant increase in urinary ACE2 protein in transplant patients compared to controls ( Figure 1C : p = 0.001). Within the group of 100 transplant patients, multiple linear regression using primary renal diseases as explanatory variables revealed a significant association between the diagnosis of diabetic nephropathy (n = 21) and urinary ACE2 protein by ELISA or western analyses (p,0.001 for both). In contrast, there was no significant association between urinary ACE2 levels and other primary causes of renal disease in these patients. Further studies were performed on the association between diabetes mellitus and urinary ACE2 in the transplant patients. Of the 100 transplant patients, 38 had a diagnosis of diabetes mellitus listed on their hospital chart. The majority of these patients were taking insulin (31 out of 38, 81.6%), 6 (15.8%) were taking oral hypoglycemic agents, and only 1 patient (2.6%) was on dietary therapy alone. Transplant patients with diabetes (n = 38) had significantly higher levels of urinary ACE2 activity, compared to non-diabetic (n = 62) subjects [ Figure 2A : diabetics; 8.75 (IQR, 3.09-15.54) vs non-diabetics; 2.32 (IQR, 0.63-5.37) ng-eq/mg Cr610 2 ; p,0.001]. Similarly, ACE2 protein levels by ELISA and western analysis were significantly increased in diabetic subjects, compared to non-diabetics ( Figure 2B , C; p,0.001 vs non-diabetics for both). In transplant patients with diabetes, both the 120 kDa and the 90 kDa ACE2 protein bands were significantly increased on westerns, compared to non-diabetics (p = 0.002 vs non-diabetics for the 120 kDa band, and p,0.001 vs non-diabetics for the 90 kDa band). Moreover, urinary ACE2 protein levels by ELISA were significantly increased in subjects with diabetes pre-transplant (n = 21), compared to subjects who developed diabetes after transplant (n = 17; p = 0.024). The state of glycosylation of the urinary ACE2 proteins at 120 kDa and 90 kDa was studied by treating urine samples with the deglycosylase enzyme PNGase F. Urinary ACE2 fragment sizes were reduced to ,85 kDa and ,65 kDa by PNGase F treatment, indicating that both fragments were originally glycosylated ( Figure 2D ). RNA was extracted from urinary pellets and subjected to realtime PCR for ACE2. ACE2 mRNA was detected in 45% of urine samples of transplant patients, with no significant difference between diabetic [0.669 (IQR, 0.00-6.15) pg mRNA/mg Cr610 25 ] and non-diabetic subjects [0.00 (IQR, 0.00-2.06) pg mRNA/mg Cr610 25 ; p = 0.091]. After adjusting for potential confounding factors [age, gender, diabetes, eGFR, albuminuria, hypertension, and use of calcineurin inhibitors] only diabetes was significantly associated with urinary ACE2 activity (p = 0.003) and protein levels (p,0.001, Table 2 ) in transplant patients. An increase in urinary ACR was associated with urinary ACE2 protein by ELISA (p = 0.026), but not with ACE2 activity or ACE2 protein by western. Female gender was associated with higher urinary ACE2 mRNA levels in transplant patients (p = 0.004, Table 2 ). Since the numbers of patients taking calcineurin inhibitors (95/ 100 patients) were highly skewed (Table 1) , multivariate analysis was also performed with exclusion of this variable. This did not alter the significance levels for any parameters reported in Table 2 . Urinary ACE Activity, Protein, and mRNA Urinary ACE activity was detected in 65% of transplant patients [median for all 100 patients: 1.17 (IQR, 0.00-3.43) ng-eq/mg Cr610 2 ]. By western analysis, ACE was detected in 94% of urine samples of transplant patients as a protein band of approximately 190 kDa, consistent with previous studies [18] (Figure 3 ). Urinary ACE activity and protein levels were significantly increased in diabetic transplant subjects, compared to non-diabetic patients ( Figure 3A After adjusting for potential confounding variables, increased urinary albumin/Cr was associated with urinary ACE protein by western analysis in transplant patients (p,0.001, Table 2 ). However, diabetes did not associate with increased urinary ACE activity or ACE protein. Female gender and urinary albumin/Cr were significantly associated with increased urinary ACE mRNA levels in transplant patients ( Table 2 ). Transplant patient with diabetes had significantly higher levels of urinary Ang II, compared to non-diabetics ( Figure 4A : p = 0.027). Urinary Ang-(1-7) levels did not differ between the two groups ( Figure 4B : p = 0.13). However, a significant positive linear correlation was found between urinary ACE2 activity and urinary Ang-(1-7) (r = 0.260, p = 0.009). In the multivariate analysis, diabetes remained significantly associated with urinary Ang II levels (p = 0.042, Table 2 ). Neither diabetes, nor any other clinical factor (eGFR, urinary albumin/Cr, etc.) was associated with urinary Ang-(1-7) levels in the multiple linear regression model. The major finding of this study is that urinary ACE2 activity and ACE2 protein are increased in kidney transplant recipients, compared to healthy control subjects, and the presence of diabetes strongly associates with urinary ACE2 levels in the patient population, by multivariate analysis. Female gender associates significantly with urinary ACE2 mRNA and ACE mRNA levels, which are detected in 45% and 65% of transplant recipients, respectively. ACE2 is expressed at relatively high levels in the kidney, particularly in the proximal tubule, and is thought to be renoprotective via its ability to reduce Ang II levels. ACE2 also increases production of Ang-(1-7), which can oppose the delete- rious non-hemodynamic actions of Ang II in tubular cells [19] . Kidney expression of ACE2 is reduced in experimental models, including the remnant kidney model of CKD [16, 20, 21] . In humans with diabetic nephropathy, expression of ACE2 is decreased in glomeruli and proximal tubules [22] , suggesting a predisposition to Ang II-mediated injury. Our results support the hypothesis that transplant patients with diabetes exhibit increased urinary excretion of ACE2 protein. This was independent of eGFR as determined by multivariate analysis. Both soluble ACE2 and ACE have been detected in human urine [10] , and in humans with CKD, Mizuiri et al. reported increased urinary levels of ACE2 protein by ELISA, compared to healthy subjects [14] . Patients with diabetic nephropathy had higher urinary levels of ACE2 than those without diabetic nephropathy, and use of ACE inhibitor and/or angiotensin receptor blockers did not affect urinary ACE2 levels [14] . In our study, no subjects were taking inhibitors of the RAS. Using three independent assays for urinary ACE2 (enzyme activity, ELISA, and immunoblot), multiple linear regression revealed that diabetes was the only variable consistently predictive of higher urinary ACE2 levels. Age and gender had no influence on urinary ACE2 activity or protein levels. Similarly, in patients with CKD Mizuiri et al. found no difference in urinary ACE2 levels between males and females [14] . Although gender did not influence urinary ACE2 activity or protein, in the present studies females had significantly higher urinary ACE2 mRNA levels. The gene for ACE2 is located on the X chromosome [2] , and in the rat renal wrap model of hypertension, intrarenal ACE2 activity and expression are upregulated by estrogens [23] , raising the possibility that female transplant recipients may express ACE2 mRNA at higher levels in the kidney compared to males. Female gender was also associated with higher urinary ACE mRNA levels, as was albuminuria, the latter being in agreement with studies in patients with diabetic nephropathy [24] . The cellular origin of urinary mRNA for ACE2 or ACE is unknown, although it has been speculated that proximal tubular cells may be the source, since they express all components of the RAS, and are found in the urine sediment [24] . However, the physiological significance of urinary ACE2 or ACE mRNA levels is unclear. In this regard, diabetes, albuminuria and eGFR did not influence urinary ACE2 mRNA levels in our transplant recipients, a result that differs from diabetic nephropathy patients, where proteinuria was found to correlate positively with urinary ACE2 mRNA, and eGFR to correlate negatively with ACE2 mRNA [24] . Factors related to transplant could therefore have an independent influence on urinary ACE2 mRNA. As one possibility, the use of immunosuppressive drugs in transplant subjects might regulate shedding of cells expressing ACE2 mRNA into the urine. Urinary ACE2 protein could originate at least partly from plasma (via filtration at the glomerulus), or it could be derived via excretion from renal cells. Although Mizuiri et al. found higher urinary ACE2 levels in CKD patients, there was no difference in plasma ACE2 protein between CKD patients and healthy subjects [14] . However, the CKD patients in their study had significant albuminuria, suggesting that ACE2 may have leaked into the urine across the glomerular barrier. In the present study, we found no association between albuminuria and either urinary ACE2 activity or protein levels by western analysis. The data suggest that urinary ACE2 protein likely derives via shedding from cells along the nephron, and not from filtration from plasma. In support of this hypothesis, soluble ACE2 is shed from the plasma membrane in cell culture systems, via cleavage at its ectodomain by the protease enzyme ADAM-17 [11] [12] [13] . Interestingly, renal ADAM-17 is upregulated in Ang II-induced kidney injury [25] , and de novo expression occurs in human kidney disease, in proximal tubule, podocytes, and mesangial cells [26] . Whether ADAM-17 is activated in the transplant kidney in the presence of diabetes, and mediates shedding of soluble ACE2 into the urine requires further study. Two bands for ACE2 were detected by immunoblot of urine samples in the current study. One band was found at 120 kDA, consistent with the full-length protein, and the other at 90 kDa, which is likely a cleaved ACE2 fragment. Thus, the 90 kDa fragment does not simply represent a deglycosylated form of fulllength ACE2, since incubation with PNGase F resulted in a further reduction in its size, as shown in Figure 2D . To determine if this 90 kDa fragment is generated by ADAM-17-mediated cleavage of ACE2, further characterization is required, including use of antibodies directed against the cytoplasmic carboxyterminus of ACE2 (which may be lacking in this fragment), and/or direct sequence analysis of the polypeptide. Furthermore, although shedding of ACE2 fragments may occur via ADAM-17 in transplant subjects, proteomic analysis has identified ACE2 as one of the 1132 proteins present in urinary exosomes isolated from human urine [27] , suggesting that membrane-bound full-length ACE2 may also be a source of urinary ACE2. By multivariate analysis, diabetes was significantly associated with urinary Ang II levels in transplant subjects, consistent with intrarenal RAS activation in diabetes [28] . In this regard, although diabetic patients had significantly higher urinary ACE activity and ACE protein levels by immunoblot, this association was not confirmed in the multivariate analysis. Finally, although diabetics had higher urinary ACE2 activity and protein levels, there was no significant association between diabetes and urinary levels of Ang-(1-7) by multivariate analysis, suggesting that urinary Ang-(1-7) may be influenced by other factors. A potential limitation of the urinary angiotensin peptide measurements is that urine specimens were not treated with protease inhibitors, which could have affected the sensitivity for detecting differences. Interestingly, however, we did observe a significant correlation between urinary ACE2 activity and Ang-(1-7) levels. Our study has a number of strengths, including the reliability and reproducibility of the assays, the inclusion of data on mRNA, protein, and ACE2/ACE products, and the use of multiple linear regression to adjust for confounding variables. Limitations include the relatively small number of subjects, the absence of plasma ACE2 or ACE measurements, and the single time point for urinary ACE2, ACE and peptide assays. Furthermore, it is possible that diabetes alone may contribute to increased urinary ACE2 levels, independent of transplant status. Studies in patients with diabetes and normal kidney function are required to answer this question. Larger studies are also needed to determine if urinary ACE2 or ACE are biomarkers of transplant function or if they may predict responsiveness to blockade of the RAS. In summary, in renal transplant recipients diabetes is a strong independent predictor of increased urinary levels of ACE2 activity and protein. Our data further suggest that ACE2 may be shed into the urine in transplant recipients, and could represent a marker to assess the role of the kidney RAS in these patients. Text S1 Detailed methods for enzyme activity assays, immunoblots, real-time PCR assays, and measurements of Ang II and Ang-(1-7). (DOC)

PepMapper: A Collaborative Web Tool for Mapping Epitopes from Affinity-Selected Peptides

Epitope mapping from affinity-selected peptides has become popular in epitope prediction, and correspondingly many Web-based tools have been developed in recent years. However, the performance of these tools varies in different circumstances. To address this problem, we employed an ensemble approach to incorporate two popular Web tools, MimoPro and Pep-3D-Search, together for taking advantages offered by both methods so as to give users more options for their specific purposes of epitope-peptide mapping. The combined operation of Union finds as many associated peptides as possible from both methods, which increases sensitivity in finding potential epitopic regions on a given antigen surface. The combined operation of Intersection achieves to some extent the mutual verification by the two methods and hence increases the likelihood of locating the genuine epitopic region on a given antigen in relation to the interacting peptides. The Consistency between Intersection and Union is an indirect sufficient condition to assess the likelihood of successful peptide-epitope mapping. On average from 27 tests, the combined operations of PepMapper outperformed either MimoPro or Pep-3D-Search alone. Therefore, PepMapper is another multipurpose mapping tool for epitope prediction from affinity-selected peptides. The Web server can be freely accessed at: http://informatics.nenu.edu.cn/PepMapper/

Epitope mapping from affinity-selected peptides has been proven to be a useful approach in identifying native epitopes for immunological applications in recent years [1, 2, 3, 4, 5, 6, 7, 8] . Affinity-selected peptides which are derived from phage-display experiments, also known as mimotopes, are assumed to have similar components with the native epitope [9] . Various ways have been proposed to map the mimotopes back to the genuine epitope. These methods were reviewed and compared in some recent literature [10, 11] . In general, they can be categorized as sequence based [12] , motif based [7, 13] , physicochemical properties based [14] , and graph search based [4, 15, 16] methods. Graph search methods are among the most efficient ways in epitope mapping demonstrated in many recent publications [4, 15, 16] because they take advantages of more information provided by using both the 3D structure of a protein than using the traditional amino acid sequence and the information from mimotope set. The essential idea of graph search methods is to find a group of simple paths on a graph generated from the residues on the surface of a protein and find out some paths from the graph best matched to the query peptides derived from in vitro screening against a target antibody [6] . Searching in Pep-3D-Search [4] is achieved through an algorithm based on ant colony optimization (ACO) whereas PepSurf [6] realizes the mapping using a dynamic programming based stochastic color-coding algorithm. However, finding a simple path on a graph is computationally intractable for any large scale of searching problem. For example, PepSurf takes a few hours to get the result for a peptide of 14 or 15 amino acids. MimoPro [16] has brought improvement on processing speed and sensitivity over both PepSurf and Pep-3D-Search. It uses an adaptable distance threshold (ADT) regulated by an appropriate compactness factor to define a graph from a small patch on the surface of a protein. Such a regulated graph contains a certain number of edges, which can guarantee that searching through the graph is more efficient. On average, MimoPro achieved the best performance among the three, but individual cases produced mixed outcomes. This indicates that no one dominates over others in all circumstances but each has its advantage in dealing with particular cases. Perhaps the best strategy is to combine two or more methods together to deal with various cases of epitope-peptide mapping in practice. Pepitope [17] combined both PepSurf and Mapitope [1] together as a Web tool for epitope-peptide mapping so as to complement with each other. The algorithm used in PepSurf maps the affinity-selected peptides directly back to the protein surface. The most significant alignments are then clustered into a patch, from which the epitope location is inferred. In Mapitope, each peptide is first deconvoluted to amino acid pairs, and those pairs of residues that are significantly overrepresented in the panel of peptides are then identified. Epitopic regions are finally predicted through searching for a cluster of those enriched pairs on the 3D structure of the antigen. Although significant progress has been made in epitope prediction through mimotope mapping, we must acknowledge that the performance of any algorithm devised and any tool developed so far was evaluated based on the outcomes of very limited test cases in which the epitopic region must be known and both the structure of the antigen and the peptide set derived from high-throughput screening must be available. If a single method is applied to a case in which the epitopic region is unknown, the mapping simply returns a candidate epitope (or none) with aligned paths formed by the antigen surface residues (or none). Such candidate epitope will become the focus of further investigation through other means. If no any single experimentally derived peptide is related to any region on the antigen, it only indicates that this method is not applicable for the case through the mapping. However, it does not mean that no interacting epitope exists on the antigen, which may be detected by other methods. In this regard, the likelihood of finding a genuine epitopic region on an antigen should be higher if more associated peptides can be detected through the mapping. Furthermore, if more mapping methods can be combined together for exploring as many associated peptides as possible through the mapping, the likelihood of finding a genuine epitopic region on the antigen should be enhanced. On the other hand, users of a mapping tool would prefer to know some sort of certainty about the candidate epitope determined by the associated peptides through the mapping in relation to the likelihood of being a genuine epitope. In other words, some kind of verification on the candidate epitope, if not a confirmation, will be much helpful for the users to make an initial assessment on the quality of the candidate. A single method cannot achieve this goal by self verification, but a combined approach of two or more independent methods would be able to provide mutual verification on the candidate of the same case. Web tools realizing such collaborative concept have not been tried so far. In this paper, we report our effort on combining both MimoPro and a modified version of Pep-3D-Search together to realize such a collaborative Web tool for supporting users in peptide-epitope mapping. In addition to the process of either MimoPro or Pep-3D-Search alone, the combined operation of Union captures the concept of exploring as many associated peptides as possible from both methods. The concept of mutual verification is realized by the combined operation of Intersection from both methods. In the next section, we introduce the processes of MimoPro, Pep-3D-Search, and the combined approach of PepMapper. Their online implementations are then briefly outlined. Construction of test cases and assessment of mapping are incorporated with discussions of the experimental results. Conclusions are finally drawn. The process of Pep-3D-Search [4] is illustrated in Figure 1a . Given a 3D structure of an antigen, Pep-3D-Search identifies all the surface residues and creates a surface graph using those residues. An ACO algorithm is then used to search the matched paths on the antigen surface with respect to the query peptides or motif. Each matched path is then rated by its P-value score [4] . A set of highly rated paths are selected to create a weighted graph of resultant paths. The Depth-First Search (DFS) algorithm is finally used to screen and cluster this weighted graph to define the candidate epitopes. The process of Pep-3D-Search has two unique features. Firstly, Pep-3D-Search is able to deal with both mimotope searching and motif mapping on the residue surface graph. Secondly, the adoption of ACO algorithm allows longer mimotopes or motifs to be processed with reasonable efficiency. The performance of Pep-3D-Search assessed by a few comparative studies [16] seems to be above the average level. The process of MimoPro [16] is illustrated in Figure 1b . Initially, the surface of a protein is divided into some overlapping patches and each patch is centered at atom C b of a surface residue with a radius of 15 Å . This radius allows most epitopes to be encompassed in such a patch [16] . Secondly, each surface patch is further transformed to a graph bounded by neighbor amino acids that are determined using a parameter called adaptive distance threshold (ADT). Afterwards a patch-based complete graph search algorithm is utilized to find the best alignment for each mimotope sequence in each graph. During this iteration, similarity between a path and the corresponding mimotope is rated. Finally the patch with the highest score is selected as a potential candidate for the native epitope. This approach has some new features different from other similar methods. Firstly, the ADT that is changeable in different regions of a protein is introduced in generating a graph from a surface patch. Such a distance threshold is adjustable so that a longer distance is used in loose regions of an antigen to include more useful connections whereas a shorter distance is adopted in dense regions to preclude some insignificant connections. Secondly, a compactness factor is introduced to make sure that all resultant graphs share a uniform compactness so that searching over any regulated graph is simpler and faster compared with previous methods. Thirdly, the adopted algorithm not only employs dynamic programming (DP) to reduce repeating searches and prune some insignificant paths encountered in the traditional search algorithm, but also introduces the branch and bound method to optimize the candidate set of rated paths during the DP process. The performance of MimoPro assessed by a few comparative studies [16, 18, 19] shows that MimoPro seems to be the most sensitive tool on average among the compared tools. PepMapper provides users with a united platform to conduct peptide-epitope mapping through either MimoPro or Pep-3D-Search or both for different purposes. The processes of PepMapper are illustrated in Figure 2 . If a user selects either MimoPro or Pep-3D-Search, PepMapper works almost exactly as either does individually, except some possible minor variations in results of this modified version of Pep-3D-Search from its original version [4] . If a user selects the Both option, PepMapper executes both MimoPro and Pep-3D-Search concurrently without mutual interference. The user will get a complete report of processed results through the emailed link. The user can access the normal result of either MimoPro or Pep-3D-Search as each works alone. To view the results of the Both option, the user has to press Compare on the left side of the result Webpage, which will produce a new Webpage showing the text results of both Intersection and Union of the two methods ( Figure 3 ). By clicking Jmol button on this Webpage, the 3D image of the result from either Intersection (by default) or Union can be displayed ( Figure 4 ). The combined operations of Union and Intersection are defined as where A and B are two sets of epitopic amino acids predicted by MimoPro and Pep-3D-Search respectively. Union captures the concept of exploring as many associated peptides as possible from both methods by constructing a new set that consists of not only the common epitopic amino acids in both A and B, but also all the distinctive epitopic amino acids in either A or B. Therefore, Union should be more sensitive than either method in epitope detection. Intersection realizes the concept of mutual verification from both methods by creating a new set that consists of only the common epitopic amino acids in both A and B. Hence, Intersection should be more reliable than either method in epitope detection if its outcome is positive. Without confirmation from real experiment results, mutual verification from artificial prediction methods can only provide an indication of where the true epitopic region is likely located on the antigen surface. Ideally if both methods produce exactly the same epitopic amino acids, both Intersection and Union should return the same set of epitopic amino acids. Hence MimoPro and Pep-3D-Search share a consistency of 100% to each other on the case However, this failure in mutual verification only means that the two methods cannot support each other on the case under study, but it does not mean that the predicted epitopic regions by either method are not related to the genuine epitope. Other approaches It can be clearly seen on these images that Intersection provides more confined prediction as most of the residues lie on the interface whereas Union outlines a larger area that may cover (part of) the potential epitopic region. doi:10.1371/journal.pone.0037869.g004 are needed to verify the epitopic regions predicted by either method. Commonly, consistency of epitope prediction from the two methods falls between 0 and 1. To present this indication numerically, we define the Consistency of the two methods in epitope prediction as The higher the Consistency, the larger the overlapped area of predicted epitopic regions by both methods; hence it is an indirect indication that a genuine epitope is more likely to be found around the overlapped area on the antigen surface under study. PepMapper has been implemented using C++ as a Web-based tool located at http://informatics.nenu.edu.cn/PepMapper. It is currently deployed on Linux using tomcat server 6.0 and has been tested using many popular Web browsers, such as IE7-9, Firefox, and Opera. Three options, MimoPro, Pep-3D-Search, and their combination, are available for the users to choose for the purpose of their applications. Note that the original Web tool of Pep-3D-Search was implemented using VB.NET. Pep-3D-Search in PepMapper is re-implemented using C++ and a modified ACO algorithm is adopted for more efficient searching (See Table S1 for details). If a user has multiple requests and needs the results to be returned fairly quickly, it is suggested to choose MimoPro for meeting such purpose because MimoPro is arguably the fastest in processing [20] . If the user wants to verify the results, it is suggested to choose the combination mode. When accessing PepMapper online, Mapping is the default interface displayed. The input to PepMapper is the structure of a chosen antigen and the peptide library screened from the corresponding antibody. The user needs to specify both the identifier of an antigen in the PDB database through its PDB_ID and the identifier of the interacting chain through Chain No. The user then needs to specify at least one peptide in the box labeled as Mimotopes. The peptides should be grouped in the FASTA format or just in separated lines of sequences. At last, the user needs to provide a valid email address in the text box. By clicking Query, PepMapper begins processing and the results will be sent to the user through the email provided. The result from PepMapper is a candidate epitope along with the alignment for each peptide sequence. Users can see the result in three ways: text/table, 3D graphics, and Rasmol scripts. In text/table format, texts are used to list all potential amino acids. The resultant alignments for individual peptide sequences are tabulated with corresponding P-values. In 3D graphics through Jmol, the candidate epitope is shown in filled balls and the other amino acids are shown as backbones by default (Figure 4) . Results can also be presented in Rasmol script that can be downloaded by clicking the link provided. This is useful when the network connection is poor. A new function Compare is also provided to make mutual verification easier between the results of the two methods. By clicking Compare on the left in the result Webpage, the peptides constituting the candidate epitope are displayed in two boxes corresponding to both methods. Clicking Compare under the left box will return a new Webpage that shows the results of both Intersection and Union from both methods (Figure 3) , which can also be viewed as 3D images by clicking Jmol button on this Webpage (Figure 4 ). The task of epitope prediction based on the peptide set is to map it back to the epitopic region on an antigen that interacts with the target molecule during in vitro screening. Although there may be other epitopes on the antigen surface, we only consider the active epitope in the designated context and regard the rest part of the antigen as nonepitope. For the test cases that correspond to the same epitope and same reference antigen structure in PDB database [18, 21] but different mimotope sets, we retain only one representative to avoid the possible bias caused by the duplication. Those cases with antigen smaller than 80 amino acids are excluded because they are too small to reflect the performance. Based on these rules, the final dataset was constructed by 27 test cases (Table 1 ). In order to analyze the performances of Pep-3D-Search, MimoPro and PepMapper, the outcome is assessed by a number of measurements, including sensitivity (Se), specificity (Sp), and precision (Pr) defined as follows: Pr~T P TPzFP : In these expressions, TP is the number of predicted epitopic amino acids proven to be the true epitopic amino acids. FP is the number of predicted epitopic amino acids proven not to be the true epitopic amino acids. TN is the predicted non-epitopic amino acids proven not to be the true epitopic amino acids. FN is the number of predicted non-epitopic amino acids proven to be the true epitopic amino acids. We use PE to denote the number of all predicted epitopic amino acids (the sum of TP and FP). To demonstrate the improved performance of PepMapper over either MimoPro or Pep-3D-Search alone, we first present the results from MimoPro and Pep-3D-Search run separately and then the results from the combined operations. Table 2 presents the evaluation results from the two methods run separately. The performances of prediction from MimoPro and Pep-3D-Search varied in different test cases. MimoPro provided some good results on 2ADF_A, 3EZE_B, 1JRH_I, 1BJ1_H, 1N8Z_C and 1ZTX_E with sensitivity exceeding 0.8 and specificity higher than 0.6; meanwhile the worst results were observed in 1YY9_A, 2NY7_G, 2GRX_A, 1D4V_B, 3BT1_A and 1HX1_A with sensitivity approaching to 0. For the rest cases, the sensitivity of prediction was between 0.25 and 0.6 and the specificity is consistently higher than 0.8. Comparatively, Pep-3D-Search gave better results in 1HX1_A and 1D4V_B, in which MimoPro failed to predict any epitopic amino acids. However, Pep-3D-Search failed in 2ADF_A, 1EER_A and 1MQ8_B whereas MimoPro produced useful results. On average, MimoPro gives better results in terms of sensitivity (0.446) and precision (0.267), but slightly worse than Pep-3D-Search in specificity. Both MimoPro and Pep-3D-Search failed in 1YY9_A, 2GRX_A, 1EER_A, and 3EZE_B. Consequently, PepMapper failed as well. We think that this failure could be attributed to a number of factors, including the quality of the experimental data and the complexity of the predicting tasks. For instance, we found that the mimotope set used for predicting the epitopic region of 2GRX_A is screened against the whole complementary protein rather than the restricted region of the two interacting proteins. Therefore it is reasonable to suppose that there may be multiple regions on the surface of the target antigen to which the mimotope can bind. As a result, the mimotopes may bind to the regions that are different from the preferable region. Additionally, the limited number of mimotopes (1D4V_B, 1EER_A) and surface amino acids may also complicate the matter, since the small number of mimotopes contains little information for locating the epitopic region, especially where too many surface amino acids exist. Furthermore, even though the dataset for our experiments has been the largest ever reported publicly, a few bad results can still greatly influence the statistical results. The Intersection operation of PepMapper captures the idea of mutual verification of epitope prediction. Intuitively, the more the commonly shared peptides in the same area are, the more likely the area to be a part of an epitope is. On average, this operation has the highest specificity of 0.930 and a high precision of 0.256 compared to that of the Union, MimoPro, and Pep-3D-Search (Tables 2 & 3) . However, its sensitivity is the lowest because some epitopic amino acids predicted by either method but not in common are left out in the calculation. This also reveals the weakness of the Intersection operation of PepMapper, i.e., in case of no overlapping between the two methods, it does not mean that no epitopic sites may be predicted by either MimoPro or Pep-3D-Search alone. 1MQ8 is such a case without common peptides, but MimoPro still predicts some positive epitopic sites. Fortunately, the union operation of PepMapper complements the weakness of Intersection operation by joining the results from the two methods together to increase the size of potential epitopic sites. The Union operation produced the best performance in sensitivity but the worst in precision and specificity compared to that of the Intersection, MimoPro, and Pep-3D-Search (Tables 2 & 3) . This is because the increased size of potential epitopic sites brought in by the Union operation also contains more false positives in the candidates. Using Consistency defined in Equation (3) as an indirect sufficient condition to judge the likelihood of successful prediction of epitope by combining both Intersection and Union, our tests tend to support its usefulness in indicating the likelihood of successful prediction (Table 3) . Although the number of tests is still insufficient for us to draw any exclusive conclusion on its implication on epitope prediction, our initial analysis leads to the following indications: PepMapper, a combination of both MimoPro and a modified version of Pep-3D-Search together, sets a collaborative Web platform, on which users can conveniently conduct peptideepitope mappings. In addition to the normal process of either MimoPro or Pep-3D-Search alone, the combined operation of Union captures the concept of exploring as many associated peptides as possible from both methods and thus increases sensitivity in finding potential epitopic regions on a given antigen surface. The Intersection operation of PepMapper realizes largely the concept of mutual verification by the two methods and hence increases the likelihood of locating the genuine epitopic region on a given antigen with respect to the interacting peptides. The Consistency between Intersection and Union can be used as an indirect sufficient condition to assess the likelihood of successful peptide-epitope mapping. In the future, we will consider to ensemble more methods in more rationalized ways to minimize the occurrence of nil Consistency, which should enhance the effectiveness of PepMapper in peptide-epitope mapping. Effort should also be made on refining the indication of Consistency in epitope prediction by conducting more tests for various conditions. We will try to improve the efficiency of the server through utilizing distributed and/or cloud computing as well. Availability. We introduced a new server, PepMapper, to incorporate both MimoPro and Pep-3D-Search which is implemented in C++ and deployed at http://informatics.nenu.edu.cn/ PepMapper. It is free for the science community and academic research. However, for commercial purposes, permission must be granted by the owner of the Web tool. Table S1 Adaption from former Pep-3D-Search. To improve the time efficiency of Pep-3D-Search, we made few adaptations from the former one. These includes a quicker approach in the generating a random background distribution for scoring the best aligned paths from graph search as well as the adjustment of the key parameters. As is shown in the Table S1 , the performance improved on 3IU3_I, 1D4V_B in the adapted Pep-3D-Search on which the former Pep-3D-Search failed to predict any epitopic amino acids. On average, the new Pep-3D-Search has similar sensitivity and specificity, but higher precision. (DOC)

Autophagy: More Than a Nonselective Pathway

Autophagy is a catabolic pathway conserved among eukaryotes that allows cells to rapidly eliminate large unwanted structures such as aberrant protein aggregates, superfluous or damaged organelles, and invading pathogens. The hallmark of this transport pathway is the sequestration of the cargoes that have to be degraded in the lysosomes by double-membrane vesicles called autophagosomes. The key actors mediating the biogenesis of these carriers are the autophagy-related genes (ATGs). For a long time, it was assumed that autophagy is a bulk process. Recent studies, however, have highlighted the capacity of this pathway to exclusively eliminate specific structures and thus better fulfil the catabolic necessities of the cell. We are just starting to unveil the regulation and mechanism of these selective types of autophagy, but what it is already clearly emerging is that structures targeted to destruction are accurately enwrapped by autophagosomes through the action of specific receptors and adaptors. In this paper, we will briefly discuss the impact that the selective types of autophagy have had on our understanding of autophagy.

Three different pathways can deliver cytoplasmic components into the lumen of the lysosome for degradation. They are commonly referred to as autophagy (cell "self-eating") and include chaperone-mediated autophagy (CMA), microautophagy, and macroautophagy. CMA involves the direct translocation of specific proteins containing the KFERQ pentapeptide sequence across the lysosome membrane [1, 2] . Microautophagy, on the other hand, entails the invagination and pinching off of the lysosomal limiting membrane, which allows the sequestration and elimination of cytoplasmic components. The molecular mechanism underlying this pathway remains largely unknown. The only cellular function that so far has been indisputably assigned to microautophagy is the turnover of peroxisomes under specific conditions in fungi [3] . Recently, it has been reported the existence of a microautophagy-like process at the late endosomes, where proteins are selectively incorporated into the vesicles that bud inward at the limiting membrane of these organelles during the multivesicular bodies biogenesis [4] . In contrast to CMA and microautophagy, macroautophagy (hereafter referred to as autophagy) entails the formation of a new organelle, the autophagosome, which allows the delivery of a large number of different cargo molecules into the lysosome. Autophagy is a primordial and highly conserved intracellular process that occurs in most eukaryotic cells and participates in stress management. This pathway involves the de novo formation of vesicles called autophagosomes, which can engulf entire regions of the cytoplasm, individual organelles, protein aggregates, and invading pathogens ( Figure 1 ). The autophagosomes fuse with endosomal compartments to form amphisomes prior to fusion with the lysosome, where their contents are degraded and the resulting metabolites are recycled back to the cytoplasm (Figure 1 ). Unique features of the pathway include the double-membrane structure of the autophagosomes, which were originally characterized over 50 years ago from detailed electron microscopy studies [5] . Starting in the 1990s yeast mutational studies began the genetic and molecular characterization of the key components required to initiate and build an autophagosome In response to inactivation of mTORC1 (but also other cellular and environmental cues), the ULK1 complex is activated and translocates in proximity of the endoplasmic reticulum (ER). Thereafter, the ULK1 complex regulates the class III PI3K complex. Atg9L, a multimembrane spanning protein, is also involved in an early stage of autophagosome formation by probably supplying part of the membranes necessary for the formation and/or expansion. Local formation of PI3P at sites called omegasomes promotes the formation of the phagophore, from which autophagosomes appear to be generated. The PI3P-binding WIPI proteins (yeast Atg18 homolog), as well as the Atg12-Atg5-Atg16L1 complex and the LC3-phosphatidylethanolamine (PE) conjugate play important roles in the elongation and closure of the isolation membrane. Finally, the complete autophagosome fuses with endosomes or endosome-derived vesicles forming the amphisome, which subsequently fuses with lysosomes to form autolysosomes. In the lysosomes, the cytoplasmic materials engulfed by the autophagosomes are degraded by resident hydrolases. The resulting amino acids and other basic cellular constituents are reused by the cell; when in high levels they also reactivate mTORC1 and then suppress autophagy. [6] . Subsequently, genetic and transgenic studies in plants, worms, fruit flies, mice, and humans have underscored the pathway's conservation and have begun to unveil the intricate vital role that autophagy plays in the physiology of cells and multicellular organisms. For a long time, autophagy was considered a nonselective pathway induced as a survival mechanism in response to cellular stresses. Over the past several years, however, it has become increasingly evident that autophagy also is a highly selective process involved in clearance of excess or dysfunctional organelles, protein aggregates and intracellular pathogens. In this introductory piece, we will briefly discuss the molecular mechanisms of selective types of autophagy and their emerging importance as a quality control to maintain cellular and organismal health, aspects that will be presented in deep in the reviews of this special issue of the International Journal of Cell Biology and highlighted by the research papers. 2.1. The Function of the Atg Proteins. Autophagosomes are formed by expansion and sealing of a small cistern known as the phagophore or isolation membrane ( Figure 1 ). Once complete, they deliver their cargo into the hydrolytic lumen of lysosomes for degradation. A diverse set of components are involved in the biogenesis of autophagosomes, which primarily includes the proteins encoded by the autophagyrelated genes (ATG). Most ATG genes have initially been identified and characterized in yeast. Subsequent studies in higher eukaryotes have revealed that these key factors are highly conserved. To date, 36 Atg proteins have been identified and 16 are part of the core Atg machinery essential for all autophagy-related pathways [7] . Upon autophagy induction, these proteins associate following a hierarchical order [8, 9] to first mediate the formation of the phagophore and then to expand it into an autophagosome [10, 11] . While their molecular functions and their precise contribution during the biogenesis of double-membrane vesicles remain largely unknown, they have been classified in 4 functional groups of genes: (1) the Atg1/ULK complex, (2) the phosphatidylinositol 3-kinase (PI3K) complex, (3) the Atg9 trafficking system, and (4) the two parallel ubiquitinlike conjugation systems ( Figure 1 ). The Atg1/ULK complex consists of Atg1, Atg13, and Atg17 in yeast, and ULK1/2, Atg13, FIP200 and Atg101 in mammals [12] [13] [14] [15] . This complex is central in mediating the induction of autophagosome biogenesis and as a result it is the terminal target of various signaling cascades regulating autophagy, such as the TOR, insulin, PKA, and AMPK pathways [16] (Figure 1 ). Increased activity of the Atg1/ULK kinase is the primary event that determines the acute induction and upregulation of autophagy. It is important to note that ULK1 is part of a protein family and two other members, ULK2 and ULK3, have been shown play a role in autophagy induction as well [14, 17] . The expansion of this gene family may reflect the complex regulation and requirements of the pathway in multicellular long-lived organisms. Stimulation of the ULK kinases is achieved through an intricate network of phosphorylation and dephosphorylation modifications of the various subunits of the Atg1/ULK complex. For example, Atg13 is directly phosphorylated by TOR and the phosphorylation state of Atg13 modulates its binding to Atg1 and Atg17. Inactivation of TOR leads to a rapid dephosphorylation of Atg13, which increases Atg1-Atg13-Atg17 complex formation, stimulates the Atg1 kinase activity and induces autophagy [18, 19] . The mAtg13 is also essential for autophagy, but seems to directly interact with ULK1, ULK2 and FIP200 independently of its phosphorylation state [13, 14] . In addition, there are several phosphorylation events within this complex as well, including phosphorylation of mAtg13 by ULK1, ULK2, and TOR; phosphorylation of FIP200 by ULK1 and ULK2; phosphorylation of ULK1 and ULK2 by TOR [13, 14] . Additional studies are required to fully characterize the functional significance of these posttranslational modifications. Autophagy is also regulated by the activity of PI3K complexes. Yeast contains a single PI3K, Vps34, which is present in two different tetrameric complexes that share 3 common subunits, Vps34, Vps15, and Atg6 [20] . Complex I is required for the induction of autophagy and through its fourth component, Atg14, associates to the autophagosomal membranes where the lipid kinase activity of Vps34 is essential for generating the phosphatidylinositol-3-phosphate (PI3P) that permits the recruitment of other Atg proteins [9, 21] (Figure 1 ). Complex II contains Vps38 as the fourth subunit and it is involved in endosomal trafficking and vacuole biogenesis [20] . There are three types of PI3K in mammals: class I, II, and III. The functions of class II PI3K remains largely unknown, but both classes I and III PI3Ks are involved in autophagy. While class I PI3K is principally implicated in the modulation of signalling cascades, class III PI3K complexes regulate organelle biogenesis and, like yeast, contain three common components: hVps34, p150 (Vps15 ortholog), and Beclin 1 (Atg6 ortholog). The counterparts of Atg14 and Vps38 are called Atg14L/Barkor and UVRAG, respectively [22] [23] [24] . The Atg14L-containing complex plays a central role in autophagy and functions very similarly as the yeast complex I by directing the class III PI3K complex I to the phagophore to produce PI3P and initiate the recruitment of the Atg machinery ( Figure 1 ). Atg14L is thought to be present on the ER irrespective of autophagy induction [25] . Upon starvation, Atg14L localizes to autophagosomal membranes [8] . Importantly, depletion of Atg14L reduces PI3P production, impairs the formation of autophagosomal precursor structures, and inhibits autophagy [8, 24, 26, 27] . The UVRAG-containing class III PI3K complex also regulates autophagy but it appears to act at the intersection between autophagy and the endosomal transport pathways. UVRAG initially associates with the BAR-domain protein Bif-1, which may regulate mAtg9 trafficking from the trans-Golgi network (TGN) [28, 29] . UVRAG then interacts with the class C Vps/HOPS protein complex, promoting the fusion of autophagosomes with late endosomes and/or lysosomes [30] . Finally, the UVRAG-containing class III protein complex binds to Rubicon, a late endosomal and lysosomal protein that suppresses autophagosome maturation by reducing hVps34 activity [26, 31] . Importantly, both the Atg14L-and UVRAG-containing complexes interact through Beclin 1 with Ambra1, which in turn tethers these protein complexes to the cytoskeleton via an interaction with dynein [32, 33] . Following the induction of autophagy, ULK1 phosphorylates Ambra1 thus releasing the class III PI3K complexes from dynein and their subsequent relocalization triggers autophagosome formation. Therefore, Ambra1 constitutes a direct regulatory link between the Atg1/ULK1 and the PI3K complexes [32] . Together with the Atg1/ULK and the PI3K complexes, Atg9 is one of the first factors localizing to the preautophagosomal structure or phagophore assembly site (PAS), the structure believed to be the precursor of the phagophore [9, 34] (Figure 1 ). Atg9 is the only conserved transmembrane protein that is essential for autophagy. It is distributed to the PAS and multiple additional cytoplasmic tubulovesicular compartments derived from the Golgi [35] [36] [37] . Atg9 cycles between these two locations and consequently it is thought to serve as a membrane carrier providing the lipid building blocks for the expanding phagophore [37] . One of the established functions of Atg9 is that it leads to the formation of the yeast PAS when at least one of the cytoplasmic tubulovesicular compartments translocates near the vacuole [34] . Atg9 is also essential to recruit the PI3K Complex I to the PAS [9] . Retrieval transport of yeast Atg9 from the PAS and/or complete autophagosome is mediated by the 4 International Journal of Cell Biology Atg2-Atg18 complex [38] and appears to be regulated by the Atg1/ULK and PI3K complexes [37] . Mammalian Atg9 (mAtg9) has similar characteristics to its yeast counterpart. mAtg9 localizes to the TGN and late endosomes and redistributes to autophagosomal structures upon the induction of autophagy (Figure 1 ) [39] , further promoting pathway activity [29, [40] [41] [42] . As in yeast, cycling of mAtg9 between locations also requires the Atg1/ULK complex and kinase activity hVps34 [39, 43] . The core Atg machinery also entails two ubiquitin-like proteins, Atg12 and Atg8/microtubule-associated protein 1 (MAP1)-light chain 3 (LC3), and their respective, partially overlapping, conjugation systems [44] [45] [46] (Figure 1 ). Atg12 is conjugated to Atg5 through the activity of the Atg7 (E1like) and the Atg10 (E2-like) enzymes. The Atg12-Atg5 conjugate then interacts with Atg16, which oligomerizes to form a large multimeric complex. Atg8/LC3 is cleaved at its C terminus by the Atg4 protease to generate the cytosolic LC3-I with a C-terminal glycine residue, which is then conjugated to phosphatidylethanolamine (PE) in a reaction that requires Atg7 and the E2-like enzyme Atg3. This lipidated form of LC3 (LC3-II) is attached to both faces of the phagophore membrane. Once the autophagosome is completed, Atg4 removes LC3-II from the outer autophagosome surface. These two ubiquitination-like systems appear to be closely interconnected. On one hand, the multimeric Atg12-Atg5-Atg16 complex localizes to the phagophore and acts as an E3-like enzyme, determining the site of Atg8/LC3 lipidation [47, 48] . On the other hand, the Atg8/LC3 conjugation machinery seems to be essential for the optimal functioning of the Atg12 conjugation system. In Atg3-deficient mice, Atg12-Atg5 conjugation is markedly reduced, and normal dissociation of the Atg12-Atg5-Atg16 complex from the phagophore is delayed [49] . Some evidences suggest that these two conjugation systems also function together during the expansion and closure of the phagophore. For example, overexpression of an inactive mutant of Atg4 inhibits the lipidation of LC3 and leads to the accumulation of a number of nearly complete autophagosomes [47] . While controversial [50] , it has been postulated that Atg8/LC3 also possesses fusogenic properties, thus mediating the assembly of the autophagic membrane [51, 52] . It has to be noted that mammals possess at least 7 genes coding for LC3/Atg8 proteins that can be grouped into three subfamilies: (1) the LC3 subfamily containing LC3A, LC3B, LC3B2 and LC3C; (2) the gammaaminobutyrate receptor-associated protein (GABARAP) subfamily comprising GABARAP and GABARAPL1 (also called GEC-1); (3) the Golgi-associated ATPase enhancer of 16 kDa (GATE-16) protein (also called GABARAP-L2/GEF2) [53] . Although in vivo studies show that they are all conjugated to PE, they appear to have evolved complex nonredundant functions [54] . Membranes. The origin of the membranes composing autophagosomes is a long-standing mystery in the field of autophagy. A major difficulty in addressing this question has been that phagophores as well as autophagosomes do not contain marker proteins of other subcellular compartments [55, 56] . A series of new studies has implicated several cellular organelles as the possible source for the autophagosomal lipid bilayers. The plasma membrane and elements of the trafficking machinery to the cell surface have been linked to the formation of an early autophagosomal intermediate, perhaps the phagophore [57] [58] [59] [60] [61] . It is possible that early endosomal-and/or Golgiderived membranes are also key factors in the initial steps of autophagy [34, 36, 39] . The Golgi, moreover, appears also important for autophagy by supplying at least in part the extra lipids required for the phagophore expansion [29, [62] [63] [64] [65] . The endoplasmic reticulum (ER) is also central in this latter event. While the relevance of the ER in autophagosome biogenesis was already pointed out a long time ago [5, 55, 66, 67] , recently two electron tomography studies have demonstrated the existence of a physical connection between the ER and the forming autophagosomes [68, 69] . These analyses have revealed that the ER is connected to the outer as well as the inner membrane of the phagophore through points of contact, supporting the notion that lipids could be supplied via direct transfer at the sites of membrane contact. In line with this view, it has been found that Atg14L is associated to the ER and PI3P is generated on specific subdomains of this organelle from where autophagosomes emerge under autophagy-inducing conditions [25, 70] (Figure 1 ). It has also been proposed that the outer membrane of the mitochondria is the main source of the autophagosomal lipid bilayers, but while the experimental evidences appear to show that mitochondria are essential for the phagophore expansion, it remains unclear whether these organelles play a key role in the phagophore biogenesis [71] . The discrepancy between the conclusions of the various studies has not allowed yet drawing a model about the membrane dynamics during autophagosome biogenesis. The different results could be due to the different experimental conditions and model systems used by the various laboratories. Alternatively, the lipids forming the autophagosomes could have different sources depending on the cell and the conditions inducing autophagy [72, 73] . A third possibility is that the source of phagophore membrane could depend on the nature of the double-membrane vesicle cargo. Additional investigations are required to shed light on these issues. Despite the potential of curing, quite a substantial range of specific pathological conditions by inducting autophagy, there are currently no small molecules that allow to exclusively stimulate this pathway [74] . Nevertheless, there is a variety of chemicals that by acting on signaling cascades that also regulate autophagy permit to trigger this degradative process. These agents fall into two distinct categories based on the mechanism of action; whether they work through an mTORdependent (Rapamycin or Torin) or mTOR-independent pathway (e.g., lithium or resveratrol) [74] . In addition to these compounds, there are biological molecules such as interferon γ (IFNγ) and vitamin D that can be used to stimulate autophagy especially in experimental setups [75, 76] . International Journal of Cell Biology 5 Inhibition of autophagy can also be beneficial in specific diseases but as for the inducers there are no compounds that exclusively block this pathway without affecting other cellular processes. The small molecules inhibiting autophagy include wortmannin and 3-methyladenine, which hamper the activity of the PI3K; Bafilomycin A and chloroquine, which impair the degradative activity of lysosomes [77] . They are currently solely used in the basic research on autophagy. It is becoming increasingly evident that autophagy is a highly selective quality control mechanism whose basal levels are important to maintain cellular homeostasis (see below). A number of organelles have been found to be selectively turned over by autophagy and cargo-specific names have been given to distinguish the various selective pathways, including the ER (reticulophagy or ERphagy), peroxisomes (pexophagy), mitochondria (mitophagy), lipid droplets (lipophagy), secretory granules (zymophagy), and even parts of the nucleus (nucleophagy). Moreover, pathogens (xenophagy), ribosomes (ribophagy), and aggregate-prone proteins (aggrephagy) are specifically targeted for degradation by autophagy [78] . Selective types of autophagy perform a cellular quality control function and therefore they must be able to distinguish their substrates, such as protein aggregates or dysfunctional mitochondria, from their functional counterparts. The molecular mechanisms underlying cargo selection and regulation of selective types of autophagy are still largely unknown. This has been an area of intense research during the last years and our understanding of the various selective types of autophagy is starting to unravel. A recent genomewide small interfering RNA screen aimed at identifying mammalian genes required for selective autophagy found 141 candidate genes to be required for viral autophagy and 96 of those genes were also required for Parkin-mediated mitophagy [79] . In general, these pathways appear to rely upon specific cargo-recognizing autophagy receptors, which connect the cargo to the autophagic membranes. The autophagy receptors might also interact with specificity adaptors, which function as scaffolding proteins that bring the cargo-receptor complex in contact with the core Atg machinery to allow the specific sequestration of the substrate. The selective types of autophagy appear to rely on the same molecular core machinery as non-selective (starvation-induced) bulk autophagy. In contrast, the autophagy receptors and specificity adapters do not seem to be required for nonselective autophagy. Autophagy receptors are defined as proteins being able to interact directly with both the autophagosome cargo and the Atg8/LC3 family members through a specific (WxxL) sequence [80] , commonly referred to as the LC3-interacting region (LIR) motif [81] or the LC3 recognition sequences (LRS) [82] . Based on comparison of LIR domains from more than 20 autophagy receptors it was found that the LIR consensus motif is an eight amino acids long sequence that can be written D/E-D/E-D/E-W/F/Y-X-X-L/I/V. Although not an absolute requirement, usually there is at least one acidic residue upstream of the W-site. The terminal L-site is occupied by a hydrophobic residue, either L, I, or V [83] . The LIR motifs of several autophagy receptors have been found to interact both with LC3 and GABARAP family members in vitro, but whether this reflects a physiological interaction remains to be clarified in most cases. It should be pointed out that not all LIR-containing proteins are autophagy cargo receptors. Some LIR-containing proteins, like Atg3 and Atg4B, are recruited to autophagic membranes to perform their function in autophagosome formation [84, 85] , whereas others like FYVE and coiled-coil domain-containing protein 1 (FYCO1) interact with LC3 to facilitate autophagosome transport and maturation [86] . Others might use an LIR motif to become degraded, like Dishevelled, an adaptor protein in the Wnt signalling pathway [87] . The adaptor proteins are less well-described, but seem to interact with autophagy receptors and work as scaffold proteins recruiting and assembling the Atg machinery required to generate autophagosomes around the cargo targeted to degradation. Examples of autophagy adaptors are Atg11 and ALFY [88, 89] . The list of specific autophagy receptors is rapidly growing and the role of several of them in different types of selective autophagy will be described in detail in the reviews of this special issue. Here we will briefly discuss the best studied form of selective autophagy, the yeast cytosol to vacuole targeting (Cvt) pathway, as well as the best studied mammalian autophagy receptor, p62/sequestosome 1 (SQSTM1) (Figure 2 ). The Cvt pathway is a biosynthetic process mediating the transport of the three vacuolar hydrolases, aminopeptidase 1 (Ape1), aminopeptidase 4 (Ape4) and α-mannosidase (Ams1), and the Ty1 transposome into the vacuole [90, 91] . Ape1 is synthesized as a cytosolic precursor (prApe1), which multimerizes into the higher order Ape1 oligomer, to which Ape4, Ams1, and Ty1 associate to form the socalled Cvt complex, prior to being sequestered into a small autophagosome-like Cvt vesicle. Sequestration of the Cvt complex into Cvt vesicles is a multistep process, which requires the autophagy receptor Atg19, which facilitates binding to Atg8 at the PAS, as well as the adaptor protein Atg11 (Figure 2(a) ) [92] . Atg11 acts as a scaffold protein by directing the Cvt complex and Atg9 reservoirs translocation to the PAS in an actin-dependent way and then recruiting the Atg1/ULK complex [40, 93] . The PI3P-binding proteins Atg20, Atg21, and Atg24 are also required for the Cvt pathway [94, 95] , but their precise function remains to be elucidated. Interestingly, Atg11 overexpression was found to recruit more Atg8 and Atg9 to the PAS resulting in more Cvt vesicles. This observation indicates that Atg11 levels could regulate the rate of selective autophagy, and maybe also the size of the cargo-containing autophagosomes in yeast [90, 96] . Indeed, a series of studies has revealed that Atg11 is also involved in other types of selective autophagy such as mitophagy and pexophagy. However, the autophagy receptors involved in the different Atg11-dependent types Figure 2 : Representative selective autophagy. (a) The cytoplasm-to-vacuole targeting (Cvt) pathway. Ape1 is synthesized as a cytoplasmic precursor protein with a propeptide and rapidly oligomerizes into dodecamers that subsequently associate with each other to form a higher order complex. The autophagy receptor Atg19 directly binds to the complex and mediates the recruitment of another Cvt pathway cargo, Ams1, leading to the formation of the so-called Cvt complex. Atg19 also interacts with the autophagy adaptor Atg11 and this protein allows the transport of the Cvt complex to the site where the double-membrane vesicle will be generated. At this location, Atg11 tethers the Atg proteins essential for the Cvt vesicle formation and the direct binding of Atg19 to Atg8 permits the exclusive sequestration of the Cvt complex into the vesicle. (b) A model for p62 and NBR1 as autophagy receptors for ubiquitinated cargos. p62 and NBR1 bind with ubiquitinated cargos via their ubiquitin-associated (UBA) domain and this interaction triggers the aggregate formation through the oligomerization of p62 via its Phox and Bem1p (PB1) domain. Furthermore, p62 interacts with both autophagy-linked FYVE protein (ALFY), which serves to recruit Atg5 and to bind PI3P, and directly with LC3. This latter event appears to organize and activate the Atg machinery in close proximity of the ubiquitinated cargos, which allows to selectively sequester them in the autophagosomes in analogous to the Cvt pathway. of selective autophagy are different as Atg32 is required for mitophagy [97, 98] , whereas Atg30 is essential for pexophagy [99] . Like Atg19, these two proteins have an Atg8-binding LIR motif and directly interact with Atg11. Mammalian cells appear to not possess an Atg11 homologue, and further studies are necessary to delineate the molecular machinery involved in sequestration and targeting of different cargoes for degradation by autophagy in higher eukaryotes. The mechanism of the Cvt pathway is reminiscent of the selective form of mammalian autophagy called aggrephagy, which involves degradation of misfolded and unwanted proteins by packing them into ubiquitinated aggregates. In both cases aggregation of the substrate (prApe1 or misfolded proteins) is required prior to sequestration into Cvt vesicles or autophagosomes, respectively [100] [101] [102] . Similar to Cvt vesicles, aggregate-containing autophagosomes appear to be largely devoid of cytosolic components suggesting that the vesicle membrane expands tightly around its cargo [88] . Aggrephagy also depends on proteins with exclusive functions in substrate selection and targeting [81, 88, 100, 103] . The autophagy receptors p62 and neighbour of BRCA1 gene (NBR1) bind both ubiquitinated protein aggregates through an ubiquitin-associated (UBA) domain and to LC3 via their LIR motifs and, thereby, promote the specific autophagic degradation of ubiquitinated proteins (Figure 2(b) ) [81, 82, 100, 103, 104] . NBR1 and p62 also contain an Nterminal Phox and Bem1p (PB1) domain through which they can oligomerize, or interact with other PB1-containing binding partners [83] . In addition to being a cargo receptor for protein aggregates, p62 has been implicated in autophagic degradation of other ubiquitinated substrates such as intracellular bacteria [105] , viral capsid proteins [106] , the midbody remnant formed after cytokinesis [107] , peroxisomes [108, 109] , damaged mitochondria [110, 111] , and bacteriocidal precursor proteins [112] . The PB1 domain was recently found to be required for p62 to localize to the autophagosome formation site adjacent to the ER [113] , suggesting that it could target ubiquitinated cargo to the site of autophagosome formation or alternatively promote the assembly of the Atg machinery at this location. International Journal of Cell Biology 7 The large scaffolding protein autophagy-linked FYVE (ALFY) appears to have a similar function as the specificity adaptor Atg11. ALFY is recruited to aggregate-prone proteins through its interaction with p62 [101] and through a direct interaction with Atg5 and PI3P it serves to recruit the core Atg machinery and allow formation of autophagic membranes around the protein aggregate [88] (Figure 2(b) ). Interestingly, ALFY is recruited from the nucleus to cytoplasmic ubiquitin-positive structures upon cell stress suggesting that it might regulate the level of aggrephagy [114] . In line with this, it was found that overexpression of ALFY in mouse and fly models of Huntington's disease reduced the number of protein inclusions [88] . It will be interesting to determine whether ALFY, as p62, is involved in other selective types of autophagy such as the one eliminating midbody ring structures or mitochondria. It is well known that posttranslational modifications like phosphorylation and ubiquitination are involved in the regulation of the activity of proteins involved in autophagy and degradation of autophagic cargo proteins, respectively. However, little is known about how these modifications may regulate selective autophagy. The fact that the core Atg machinery is required for both nonselective and selective types of autophagy gives raise to the question of whether these two types of autophagy may compete for the same molecular machinery. Such a competition could be detrimental for the cells undergoing starvation and to avoid this, there might be a tight regulation of the expression level and/or activity of the proteins specifically involved the selective autophagy. It has recently been proposed that phosphorylation of autophagy receptors might be a general mechanism for the regulation of selective autophagy. Dikic and coworkers noted that several autophagy receptors contain conserved serine residues adjacent to their LIR motifs and indeed, the TANK binding kinase 1 (TBK1) was found to phosphorylate a serine residue close to the LIR motif of the autophagy receptor optineurin. This modification enhances the LC3 binding affinity of optineurin and promotes selective autophagy of ubiquitinated cytosolic Salmonella enterica [115] . In yeast, phosphorylation of Atg32, the autophagy receptor for mitophagy, by mitogen-activated protein kinases was found to be required for mitophagy [116, 117] . The Atg8/LC3 proteins themselves have also been found to become phosphorylated and recent works have identified specific phosphorylation sites for protein kinase A (PKA) [118] and protein kinase C (PKC) [119] in the Nterminal region of LC3. Interestingly, the N-terminal of LC3 is involved in the binding of LC3 to LIR-containing proteins [120] . It is therefore tempting to speculate that phosphorylation of the PKA and PKC sites might facilitate or prevent the interaction of LC3 with LIR-containing proteins such as p62. It has been found that phosphorylation of the PKA site, which is conserved in all mammalian LC3 isoforms, but not in GABARAP, inhibits recruitment of LC3 into autophagosomes [118] . The role of ubiquitin in autophagy has so far been ascribed as a signal for cargo degradation. Ubiquitination of aggregate prone proteins, as well as bacteria and mitochondria, has been found to serve as a signal for recognition by autophagy receptors like p62 and NBR1, which are themselves also degraded together with the cargo that they associate with [83] . The in vivo specificity of p62 and NBR1 toward ubiquitin signals remains to be established under the different physiological conditions. Interestingly, it was recently found that casein kinase 2-(CK2-) mediated phosphorylation of the p62 UBA domain increases the binding affinity of this motif for polyubiquitin chains leading to more efficient targeting of polyubiquitinated proteins to autophagy [121] . CK2 overexpression or phosphatase inhibition reduced the formation of aggregates containing the polyglutamine-expanded huntingtin exon1 fragments in a p62-dependent manner. The E3 ligases involved in ubiquitination of different autophagic cargo largely remains to be identified. However, it is known that the E3 ligases Parkin and RNF185 both regulate mitophagy [122, 123] . SMURF1 (SMAD-specific E3 ubiquitin protein ligase 1) was recently also implicated in mitophagy, as well as in autophagic targeting of viral particles [79] . Interestingly, the role of SMURF1 in selective autophagy seems to be independent of its E3 ligase activity, but it rather depends on its membrane-targeting C2 domain, although the exact mechanism involved remains to be elucidated. It is also not clear whether ubiquitination could serve as a signal to regulate the activity or binding selectivity of proteins directly involved in autophagy, and whether this in some way could regulate selective autophagy. The role of ubiquitinlike proteins as SUMO and Nedd in autophagy is also unexplored. Acetylation is another posttranslational modification that only recently has been implicated in selective autophagy. The histone de-acetylase 6 (HDAC6), initially found to mediate transport of misfolded proteins to the aggresome [124] , was lately implicated in maturation of ubiquitinpositive autophagosomes [125] . The fact that HDAC6 overproduction in fly eyes expressing expanded polyQ proteins is neuroprotective further indicates that HDAC6 activity stimulates aggrephagy [126] . Furthermore, the acetylation of an aggrephagy cargo protein, muntant huntingtin, the protein causing Huntington's disease, is important for its degradation by autophagy [127] . HDAC6 has been also implicated in Parkin-mediated clearance of damaged mitochondria [128] . The acetyl transferase(s) involved in these forms of selective autophagy is currently unknown, but understanding the role of acetylation in relation to various aspects of autophagy is an emerging field and it will very likely provide more mechanistic insights into these pathways. Basal autophagy acts as the quality control pathway for cytoplasmic components and it is crucial to maintain the homeostasis of various postmitotic cells [129] . While this quality control could be partially achieved by nonselective autophagy, growing lines of evidence have demonstrated 8 International Journal of Cell Biology that specific proteins, organelles, and invading bacteria are specifically degraded by autophagy (Figure 3 ). Mice deficient in autophagy die either in utero (e.g., Beclin 1 and Fip200 knockout mice) [130] [131] [132] or within 24 hours after birth due, at least in part, to a deficiency in the mobilization of amino acids from various tissues (e.g., Atg3, Atg5, Atg7, Atg9, and Atg16L knockout mice) [49, [133] [134] [135] [136] . As a result, to investigate the physiological roles of autophagy, conditional knockout mice for Atg5, Atg7, or FIP200 and various tissue-specific Atg knockout mice have been established and analyzed [133, 137, 138] . For example, the liver-specific Atg7-deficient mouse displayed severe hepatomegaly accompanied by hepatocyte hypertrophy, resulting in severe liver injuries [133] . Mice lacking Atg5, Atg7, or FIP200 in the central nervous system exhibited behavioral deficits, such as abnormal limb-clasping reflexes and reduction of coordinated movement as well as massive neuronal loss in the cerebral and cerebellar cortices [137] [138] [139] . Loss of Atg5 in cardiac muscle caused cardiac hypertrophy, left ventricular dilatation, and systolic dysfunction [140] . Skeletal muscle-specific Atg5 or Atg7 knockout mice showed age-dependent muscle atrophy [141, 142] . Pancreatic β cell-specific Atg7 knockout animals exhibited degeneration of islets and impaired glucose tolerance with reduced insulin secretion [143, 144] . Podocytespecific deletion of Atg5 caused glomerulosclerosis in aging mice and these animals displayed increased susceptibility to proteinuric diseases caused by puromycin aminonucleoside and adriamycin [145] . Proximal tubule-specific Atg5 knockout mice were susceptible to ischemia-reperfusion injury [146] . Finally, deletion of Atg7 in bronchial epithelial cells resulted in hyperresponsiveness to cholinergic stimuli [147] . All together, these results undoubtedly indicate that basal autophagy prevents numerous life-threatening diseases. How does impairment of autophagy lead to diseases? Ultrastructural analyses of the mutant mice revealed a marked accumulation of swollen and deformed mitochondria in the mutant hepatocytes [133] , pancreatic β cells [143, 144] , cardiac and skeletal myocytes [140, 141] and neurons [138] , but also the appearance of concentric membranous structures consisting of ER or sarcoplasmic reticulum in hepatocytes [133] , neuronal axons [137, 139] and skeletal myocytes [141] , as well as an increased number of peroxisomes and lipid droplets in hepatocytes [133, 148] . In addition to the accumulation of aberrant organelles, histological analyses of tissues with defective autophagy showed the amassment of polyubiquitylated proteins in almost all tissues (although the level varied from one region to another) forming inclusion bodies whose size and number increased with aging [149] . Consequently, basal autophagy also acts as the quality control machinery for cytoplasmic organelles (Figure 3(a) ). Although this could be partially achieved by bulk autophagy, these observations point to the existence of selective types of autophagy, a notion that is now supported by experimental data. p62/SQSTM1 is the best-characterized disease-related autophagy receptor and a ubiquitously expressed cellular protein conserved among metazoan but not in plants and fungi [83] . Besides a role of p62 as the receptor, this protein itself is specific substrate for autophagy. Suppression of autophagy is usually accompanied by an accumulation of p62 mostly in large aggregates also positive for ubiquitin (Figure 3(a) ) [104, 150] . Ubiquitin and p62-positive inclusion bodies have been detected in numerous neurodegenerative diseases (i.e., Alzheimer's disease, Parkinson's disease, and amyotrophic lateral sclerosis), liver disorders (i.e., alcoholic hepatitis and steatohepatitis), and cancers (i.e., malignant glioma and hepatocellular carcinoma) [151] . Very interestingly, the p62positive aggregates observed in hepatocytes and neurons of liver-and brain-specific Atg7 deficient mice, respectively, as well as in human hepatocellular carcinoma cells, are completely dispersed by the additional loss of p62 strongly implicating involvement of p62 in the formation of diseaserelated inclusion bodies [104, 152] . Through its self-oligomerization, p62 is involved in several signal transduction pathways. For example, this protein functions as a signaling hub that may determine whether cells survive by activating the TRAF6-NF-κB pathway, or die by facilitating the aggregation of caspase 8 and the downstream effector caspases [153, 154] . On the other hand, p62 interacts with the Nrf2-binding site on Keap1, a component of the Cullin 3-type ubiquitin ligase for Nrf2, resulting in stabilization of Nrf2 and transcriptional activation of Nrf2 target genes including a battery of antioxidant proteins [155] [156] [157] [158] [159] . It is thus plausible that excess accumulation or mutation of p62 leads to hyperactivation of these signaling pathways, resulting in a disease onset (Figure 3(b) ). Paget's disease of bone is a chronic and metabolic bone disorder that is characterized by an increased bone turnover within discrete lesions throughout the skeleton. Mutations in the p62 gene, in particular in its UBA domain, can cause this illness [160] . A proposed model explaining how p62 mutations lead to the Paget's disease of bone is the following: mutations of the UBA domain cause an impairment in the interaction between p62 and ubiquitinated TRAF6 and/or CYLD, an enzyme deubiquitinating TRAF6, which in turn enhances the activation of the NF-κB signaling pathway and the resulting increased osteoclastogenesis (Figure 3(b) ) [160] . If proven, this molecular scenario could open the possibility of using autophagy enhancers as a therapy to cure Paget's disease of bone. It is established that autophagy has a tumor-suppressor role and several autophagy gene products including Beclin1 and UVRAG are known to function as tumor suppressor proteins [161] . The tumor-suppressor role of autophagy appears to be important particularly in the liver. Spontaneous tumorigenesis is observed in the livers of mice with either a systemic mosaic deletion of Atg5 or a hepatocytespecific Atg7 deletion [152, 162] . Importantly, no tumors are formed in other organs in Atg5 mosaically deleted mice. Enlarged mitochondria, whose functions are at least partially impaired, accumulate in Atg5-or Atg7-deficient hepatocytes [152, 162] . This observation is in line with the previous data obtained in iBMK cell lines showing that both the oxidative stress and genomic damage responses are activated by loss of autophagy [163, 164] . Again, it is clear that accumulation of p62, at least partially, contributes to tumor growth because the size of the Atg7 −/− liver tumors is reduced by the additional deletion of p62 [162] , which may cause a dysregulation of NF-κB signaling [165] and/or a persistent activation of Nrf2 [166] . Almost all tissues with defective autophagy are usually displaying an accumulation of polyubiquitinated proteins [149] . Loss of autophagy is considered to lead to a delay in the global turnover of cytoplasmic components [137] and/or to an impaired degradation of substrates destined for the proteasome [167] . Both observations could partially explain the accumulation of misfolded and/or unfolded proteins that is followed by the formation of inclusion bodies. As discussed above, p62 and NBR1 act as autophagy receptors for ubiquitinated cargos such as protein aggregates, mitochondria, midbody rings, bacteria, ribosomal proteins and virus capsids [83, 168] (Figure 3 ). Although these studies suggest the role of p62 as an ubiquitin receptor, it remains to be established whether soluble ubiquitinated proteins are also degraded one-by-one by p62 and possibly NBR1. A mass spectrometric analysis has clearly demonstrated the accumulation of all detectable topologies of ubiquitin chain in Atg deficient livers and brains, indicating that specific polyubiquitin chain linkage is not the decisive signal for autophagic degradation [169] . Because the increase in ubiquitin conjugates in the Atg7 deficient liver and brain is completely suppressed by additional knockout of either p62 or Nrf2 [169] , accumulation of ubiquitinated proteins in tissues defective in autophagy might be attributed to p62mediated activation of Nrf2, resulting in global transcriptional changes to ubiquitin-associated genes. Further studies are needed to precisely elucidate the degradation mechanism of soluble ubiquitinated proteins by autophagy. Concomitant with the energy production through oxidative phosphorylation, mitochondria also generate reactive oxygen species (ROS), which cause damage through the oxidation of proteins, lipids and DNA often inducing cell death. Therefore, the quality control of mitochondria is essential to maintain cellular homeostasis and this process appears to be achieved via autophagy. It has been postulated that mitophagy contributes to differentiation and development by participating to the intracellular remodelling that occurs for example during haematopoiesis and adipogenesis. In mammalian red blood cells, the expulsion of the nucleus followed by the removal of other organelles, such as mitochondria, are necessary differentiation steps. Nix/Bnip3L, an autophagy receptor whose structure resembles that of Atg32, is also an outer mitochondrial membrane protein that interacts with GABARAP [170, 171] and plays an important role in mitophagy during erythroid differentiation [172, 173] (Figure 3(c) ). Although autophagosome formation probably still occurs in Nix/Bnip3L deficient reticulocytes, mitochondrial elimination is severely impaired. Consequently, mutant reticulocytes are exposed to increased levels of ROS and die, and Nix/Bnip3L knockout mice suffer severe anemia. Depolarization of the mitochondrial membrane potential of mutant reticulocytes by treatment with an uncoupling agent results in restoration of mitophagy [172] , emphasizing the importance of Nix/Binp3L for the mitochondrial depolarization and implying that mitophagy targets uncoupled mitochondria. Haematopoietic-specific Atg7 knockout mice also exhibited severe anaemia as well as lymphopenia, and the mutant erythrocytes markedly accumulated degenerated mitochondria but not other organelles [174] . The mitochondrial content is regulated during the development of the T cells as well; that is, the high mitochondrial content in thymocytes is shifted to a low mitochondrial content in mature T cells. Atg5 or Atg7 deleted T cells fail to reduce their mitochondrial content resulting in increased ROS production as well as an imbalance in pro-and antiapoptotic protein expression [175] [176] [177] . All together, these evidences demonstrate the essential role of mitophagy in haematopoiesis. Recent studies have described the molecular mechanism by which damaged mitochondria are selectively targeted for autophagy, and have suggested that the defect is implicated in the familial Parkinson's disease (PD) [178] (Figure 3(c) ). PINK1, a mitochondrial kinase, and Parkin, an E3 ubiquitin ligase, have been genetically linked to both PD and a pathway that prevents progressive mitochondrial damage and dysfunction. When mitochondria are damaged and depolarized, PINK1 becomes stabilized and recruits Parkin to the damaged mitochondria [122, [179] [180] [181] . Various mitochondrial outer membrane proteins are ubiquitinated by Parkin and mitophagy is then induced. Of note, PD-related mutations in PINK1 and Parkin impair mitophagy [122, [179] [180] [181] , suggesting that there is a link between defective mitophagy and PD. How these ubiquitinated mitochondria are recognized by the autophagosome remains unknown. Although p62 has been implicated in the recognition of ubiquitinated mitochondria, elimination of the mitochondria occurs normally in p62-deficient cells [182, 183] . When specific bacteria invade host cells through endocytosis/phagocytosis, a selective type of autophagy termed xenophagy, engulfs them to restrict their growth [184] (Figure 3(d) ). Although neither the target proteins nor the E3 ligases have yet been identified, invading bacteria such as Salmonella enterica, Listeria monocytogenes, or Shigella flexneri become positive for ubiquitin when they access the cytosol by rupturing the endosome/phagosome limiting membrane [185, 186] . These findings raise the possibility that ubiquitin also serves as a tag during xenophagy. In fact, to date, three proteins, p62 [105, 185, 187] , NDP52 [188] , and optineurin [115] have been proposed to be autophagy receptors linking ubiquitinated bacteria and LC3. An ubiquitin-independent mechanism has recently been revealed; recognition of a Shigella mutant that lacks the icsB gene requires the tectonin domaincontaining protein 1 (Tecpr1), which appears to be a new type of autophagy adaptor targeting Shigella to Atg5-and WIPI-2-positive membranes [189] . Interestingly, the Shigella icsB normally prevents autophagic sequestration of this bacterium by inhibiting the interaction of Shigella VirG with Atg5 indicating that some bacteria have developed mechanism to inhibit or subvert autophagy to their advantage [190] . This latter category of pathogens also includes viruses such as Herpes simplex virus-1 (HSV-1), which express an inhibitor (ICP34.5) of Atg6/Beclin1 [106] . However, it was recently shown that a mutant HSV-1 strain lacking ICP34.5 becomes degraded by selective autophagy in a SMURF1dependent manner [79] , suggesting that selective autophagy plays an important role in our immune system. Recently, a different antimicrobial function has been assigned to autophagy and this function appears to be selective. During infection, ribosomal protein precursors are transported by autophagy in a p62-dependent manner into lysosomes [112] . These ribosomal protein precursors are subsequently processed by lysosomal protease into small antimicrobial peptides. Importantly, it has been shown that induction of autophagy during a Mycobacterium tuberculosis infection leads to the fusion between phagosomes containing this bacterium and autophagosomes, and the production of the antimicrobial peptides in this compartment kills M. tuberculosis [112] . While the molecular mechanism is largely unknown, autophagy contributes at least partially to the supply of free fatty acids in response to fasting (Figure 3(e) ). Fasting provokes the increase of the levels of free fatty acids circulating in the blood, which are mobilized from adipose tissues. These free fatty acids are rapidly captured by various organs including hepatocytes and then transformed into triglycerides by esterification within lipid droplets. These lipid droplets appear to be turned over by a selective type of autophagy that has been named lipophagy in order to provide endogenous free fatty acids for energy production through β-oxidation [148] . Indeed, liver-specific Atg7 deficient mice display massive accumulation of triglycerides and cholesterol in the form of lipid droplets [191] . Agoutirelated peptide-(AgRP-) expressing neurons also respond to increased circulating levels of free fatty acids after fasting and then induce autophagy to degrade the lipid droplets [192] . Similar to the case in hepatocytes, autophagy in the neurons supplies endogenous free fatty acids for energy production and seems to be necessary for gene expression of AgPR, which is a neuropeptide that increases appetite and decreases metabolism and energy expenditure [192] . Originally, it was assumed that autophagy was exclusively a bulk process. Recent experimental evidences have demonstrated that through the use of autophagy receptors and adaptors, this pathway can be selective by exclusively degrading specific cellular constituents. The list of physiological and pathological situations where autophagy is selective is constantly growing and this fact challenges the earliest concept whether autophagy can be nonselective. It is believe that under starvation, cytoplasmic structures are randomly engulfed by autophagosomes and delivered into the lysosome to be degraded and thus generate an internal pool of nutrients. In yeast Saccharomyces cerevisiae, however, the degradation of ribosomes, for example, ribophagy, as well as mitophagy and pexophagy, and the transport of the prApe1 oligomer into the vacuole under the same conditions requires the presence of autophagy receptors [97, [193] [194] [195] . As a result, these observations suggest that autophagy could potentially always operate selectively. This is a conceivable hypothesis because this process allows the cell to survive stress conditions and the casual elimination of cytoplasmic structure in the same scenario could lead to the lethal depletion of an organelle crucial for cell survival. Future studies will certainly provide more molecular insights into the regulation and mechanism of the selective types of autophagy, and this information will also be important to determine if indeed bulk autophagy exists. AgRP: Agouti-related peptide AMPK: AMP-activated protein kinase ALFY: Autophagy-linked FYVE protein Ams1: α-mannosidase 1 Ape1: Aminopeptidase 1 Ape4: Aminopeptidase 4 Atg: Autophagy-related gene Bnip3L: B-cell leukemia/lymphoma 2 (BCL-2)/adenovirus E1B interacting protein 3 CK2: Casein kinase 2 CMA: Chaperone-mediated autophagy Cvt: Cytoplasm to vacuole targeting ER: Endoplasmic reticulum FIP200: Focal adhesion kinase family interacting protein of 200 kD FYCO1: FYVE and coiled-coil domain-containing protein 1 GABARAP: Gamma-aminobutyrate receptor-associated protein GATE-16: Golgi-associated ATPase enhancer of 16 kDa HDAC6: Histone de-acetylase 6 HOPS: Homotypic fusion and protein sorting HSV-1: Herpes simplex virus-1 Keap1: Kelch-like ECH-associated protein 1 LC3: Microtubule-associated protein 1 (MAP1)-light chain 3 LIR: LC3-interacting region LRS: LC3 recognition sequences NBR1: Neighbour of BRCA1 gene NDP52: Nuclear dot protein (NDP) 52 NF-κB: Nuclear factor κB NIX: Nip-like protein X Nrf2: NF-E2 related factor 2 PAS: Phagophore assembly site PB1: Phox and Bem1p International Journal of Cell Biology PE: Phosphatidylethanolamine PD: Parkinson's disease PI3K: Phosphatidylinositol 3-kinase PI3P: Phosphatidylinositol 3-phosphate PKA: Protein kinase A PKC: Protein kinase C ROS: Reactive oxygen species Rubicon: RUN domain and cysteine-rich domain containing Beclin 1-interacting protein SMURF1: SMAD-specific E3 ubiquitin protein ligase 1 SUMO: Small ubiquitin-like modifier SQSTM1: p62/sequestosome 1 TBK1: TANK binding kinase 1 Tecpr1: Tectonin domain-containing protein 1 TRAF6: Tumour necrosis factor receptor-associated factor 6 TOR: Target of Rapamycin TGN: Trans-Golgi network UBA: Ubiquitin associated ULK1: Unc-51-like kinase 1 UVRAG: UV-resistance associated gen Vps: Vacuolar protein sorting.

The Sigma Class Glutathione Transferase from the Liver Fluke Fasciola hepatica

BACKGROUND: Liver fluke infection of livestock causes economic losses of over US$ 3 billion worldwide per annum. The disease is increasing in livestock worldwide and is a re-emerging human disease. There are currently no commercial vaccines, and only one drug with significant efficacy against adult worms and juveniles. A liver fluke vaccine is deemed essential as short-lived chemotherapy, which is prone to resistance, is an unsustainable option in both developed and developing countries. Protein superfamilies have provided a number of leading liver fluke vaccine candidates. A new form of glutathione transferase (GST) family, Sigma class GST, closely related to a leading Schistosome vaccine candidate (Sm28), has previously been revealed by proteomics in the liver fluke but not functionally characterised. METHODOLOGY/PRINCIPAL FINDINGS: In this manuscript we show that a purified recombinant form of the F. hepatica Sigma class GST possesses prostaglandin synthase activity and influences activity of host immune cells. Immunocytochemistry and western blotting have shown the protein is present near the surface of the fluke and expressed in eggs and newly excysted juveniles, and present in the excretory/secretory fraction of adults. We have assessed the potential to use F. hepatica Sigma class GST as a vaccine in a goat-based vaccine trial. No significant reduction of worm burden was found but we show significant reduction in the pathology normally associated with liver fluke infection. CONCLUSIONS/SIGNIFICANCE: We have shown that F. hepatica Sigma class GST has likely multi-functional roles in the host-parasite interaction from general detoxification and bile acid sequestration to PGD synthase activity.

The liver flukes, Fasciola hepatica and Fasciola gigantica are the causative agents of fasciolosis, a foodborne zoonotic disease affecting grazing animals and humans worldwide [1] . Liver fluke causes economic losses of over US$ 3 billion worldwide per annum to livestock via mortality, reduction in host fecundity, susceptibility to other infections, decrease in meat, milk and wool production and condemnation of livers [1] . The disease is increasing in livestock worldwide with contributing factors such as climate change (warmer winters and wetter summers supporting larger intermediate mud snail host populations); fragmented disease management (only treating sheep not cattle and limiting veterinary interaction); encouragement of wet-lands; livestock movement; and/or failure/ resistance of chemical control treatments in the absence of commercial vaccines [1, 2] . Fasciolosis is also a re-emerging human disease with estimates of between 2.4 and 17 million people infected worldwide [3] . In response, the World Health Organisation have added fasciolosis to the preventative chemotherapy concept [4] . There are currently no commercial vaccines and triclabendazole (TCBZ) is the most important fasciolicide, as the only drug with significant efficacy against adult worms and juveniles [5] . Evidence from developed countries where TCBZ has been used widely exposes the reliance on this drug as an Achilles heel of liver fluke chemotherapeutic control, with well-established evidence of drug-resistance [5] . Therefore, TCBZ does not offer a long-term sustainable option for livestock farmers worldwide. The need for a liver fluke vaccine is further underscored by the fact that the costs associated with anthelmintic intervention for fluke control make short-lived chemotherapy an unsustainable option in developing countries. Protein superfamily studies in liver fluke have provided a number of leading vaccine candidates. High quality one-gene based vaccine discovery research has identified several vaccine candidates from protein superfamilies that provide significant, but often variable protection rates in challenge animal trials against liver fluke. For example, Mu class Glutathione transferase (GSTs) have been widely investigated as vaccine candidates for fasciolosis [6] [7] [8] [9] . The Mu class GSTs have established roles in general Phase II detoxification of xenobiotic and endogenously derived toxins in F. hepatica within the host bile environment [10] . The general detoxification role is supported by GSTs contributing to 4% of the total soluble protein in F. hepatica, with a widespread tissue distribution. Proteomics and EST sequencing approaches have now delineated what members of the GST family are expressed in F. hepatica and two new classes of GST, Sigma and Omega, have been uncovered [11] . In the related trematode, Schistosoma mansoni, the Sigma class GST (Sm28) has generally shown more robust protection in vaccine trials against schistosome infection [12] , than the F. hepatica Mu GSTs against F. hepatica infection. Sigma class GSTs, unlike Mu Class GSTs, have been characterized as GSH-dependent hematopoietic prostaglandin synthases responsible for the production of prostaglandins in both mammals and parasitic worms [13] [14] [15] [16] [17] [18] . Prostaglandins have been extensively studied in mammals and are shown to be involved in a range of physiological and pathological responses [19] [20] [21] [22] [23] . Parasite-produced prostaglandins may be involved in parasite development and reproduction as well as the modulation of host immunity, allergy and inflammation during establishment and maintenance of a host infection [16, [24] [25] [26] [27] [28] ]. The host protection success of Sigma GST based vaccinations in schistosomiasis may therefore be related to neutralising specific functions in host-parasite interplay, such as prostaglandin synthase activity. In this manuscript we follow four work pathways to functionally characterise the newly identified Sigma GST from F. hepatica. 1) We confirm its designation as a Sigma class GST using substrate profiling, 2) we assess prostaglandin synthase activity and its effect on host immune cells, 3) we localise the Sigma GST within adult fluke and between ontogenic stages and 4) assess its potential as a vaccine candidate. GST proteins representative of recognised GST superfamily classes were obtained from European Bioinformatics Institute Interpro database (http://www.ebi.ac.uk/ interpro/), and from non-redundant databases at NCBI (http://www.ncbi.nlm.nih. gov/). A mammalian and a helminth or invertebrate GST sequence were selected for each GST class where available. Sequences were aligned via ClustalW program [29] in BioEdit Sequence Alignment Editor Version 7.0.5.2. [30] and sequence identity matrices produced from multiple alignments. Phylogenetic bootstrap neighbour-joining trees were produced as PHYLIP output files in ClustalX Version 1.83 [31] according to the neighbour-joining method of Saitou and Nei [32] . ClustalX default settings for alignments were accepted using the GONNET protein weight matrices with PHYLIP tree format files viewed within TREEVIEW [33] . Recombinant Fasciola hepatica glutathione transferase Sigma class (rFhGST-S1) production Full-length cDNA for FhGST-S1 was available in the form of an expressed sequence tag (EST) clone Fhep24h03, details of which can be obtained from the previously published Sigma class GST [11] and is identical to the submitted GenBank accession No. DQ974116.1 (NCBI http://www.ncbi.nlm.nih.gov/). FhGST-S1 was amplified via PCR using the following primer pair: rFhGST-S1 forward primer, 59 GGAATTCCATATGGA-CAAACAGCATTTCAAGTT 39;rFhGST-S1 reverse primer, 59 ATAAGAATGCGGCCGCCTAGAATGGAGTTTTTGCAC-GTTTTTT 39. Restriction enzyme sites (in bold type and underlined) for NdeI (forward primer) and NotI (reverse primer) were included so that the entire ORF could be directionally cloned into the pET23a (Novagen) vector. Recombinant protein was produced in Escherichia coli BL21(DE3) cells (Novagen). Protein purification of rFhGST-S1 and native F. hepatica GSTs rFhGST-S1 protein was purified according to the glutathione affinity chromatography method of Simons and Vander Jagt [34] from transformed E. coli cytosol following protein expression. Native GSTs were purified from F. hepatica soluble cytosolic supernatants as previously described [11] . Purity of rFhGST-S1 was assessed by electrospray ionisation (ESI) mass spectrometry, sodium dodecyl sulphate polyacrylamide gel electrophoresis (SDS-PAGE) and 2DE according to LaCourse et al. [35] . A range of model and natural substrates (see Table 1 for details) were used to profile the Sigma GST. A number of ligands were also assessed for their ability to inhibit GST activity with 1-chloro-2, 4-dinitrobenzene (CDNB) as the second substrate [36] . Values were reported as the concentration of inhibitor required to bring GST specific activity to 50% of its original activity (IC50). At least six different inhibitor concentrations were used in each IC50 determination in triplicate. Inhibitors were pre-incubated for 5 minutes prior to starting reactions. IC50 values were estimated graphically [37] . Prostaglandin synthase activity was assessed via an adapted method based upon those of Sommer et al. [26] and Meyer et al. [16, 38] , with extraction modifications based upon Schmidt et al. [39] . In brief, reactions were performed in glass vials in 2 mM sodium phosphate buffer, pH 7.4, containing 10 mM glutathione, Combating neglected parasitic diseases is of paramount importance to improve the health of human populations and/or their domestic animals. Uncovering key roles in host-parasite interactions may support the vaccine potential portfolio of a parasite protein. Fasciola hepatica causes global disease in humans and their livestock but no commercial vaccines are available. Members of the Sigma class glutathione transferase (GST) family have long been highlighted as vaccine candidates towards parasitic flatworms. To this end, a Sigma class GST is currently undergoing phase II clinical trials to protect against infection from the schistosomes. In this study we characterise the protein from F. hepatica following four work pathways that 1) confirm its designation as a Sigma class GST using substrate profiling, 2) assess prostaglandin synthase activity and its effect on host immune cells, 3) localise the Sigma GST within adult fluke and between ontogenic stages and 4) measure its potential as a vaccine candidate. The work presented here shows F. hepatica Sigma class GST to have key host-parasite roles and we suggest, warrants further investigation for inclusion into vaccine formulations. Recombinant FhGST-S1 shows activity towards a broad range of model and natural GST substrates with a similar enzymatic profile to the Schistosomiasis vaccine trialist (Sm28GST -P09792). rFhGST-S1 also displays high glutathione-dependent lipid peroxidase activity compared to both Sm28GST and Sj26GST (Q26513) [47] . Reasonably high GSH-dependent lipid peroxidase activity has also been seen in a 'weak affinity' fraction following chromatofocusing of GSH transferase activity that failed to bind GSH-sepharose [10] . ND -Not determined. **Data taken from Walker et al. [47] . doi:10.1371/journal.pntd.0001666.t001 Sigma Class Glutathione Transferase of F. hepatica www.plosntds.org 50 mM NaCl, 0.5 mM tryptophan, 1 mM hematin, 1 U COX-1 enzyme, 100 mM arachidonic acid (All Sigma, UK. COX-1 [C0733]) and rFhGST-S1 at final concentration ranges of 0.1-100.0 mg/ml. Negative control reactions lacking either GST or COX-1 were also prepared. Reactions were incubated for 5-10 min in a water bath at 37uC. This was followed by 4 minutes incubation at 25uC in a shaking water bath. Prostaglandins were extracted by adding 860 mL of ice-cold ethyl acetate. Reactions were vortexed for 30 s then centrifuged briefly at 10,0006g at 4uC for 2 min. The upper ethyl acetate layer was retained and solvent was evaporated under a nitrogen stream at 45uC. The remaining residue was reconstituted in 50 ml of methanol/water/fomic acid (25:75:0,1) mix at pH 2.8 and stored at 280uC until ready for mass spectrometry analysis. Standards of prostaglandins D2, E2 and F2a (Cayman, Ltd) were also prepared in methanol/water/ formic acid mix for analysis. The nano LC-MS analyses were performed using a Waters Q-Tof micro mass spectrometer (Waters) coupled to a LC-Packings Ultimate nano LC system (Dionex). The pre-column used was a LC Packings C18 PepMap 100 and the nano LC column used was a LC Packings 15 cm PepMap 100 C18 (both Dionex). Samples were loaded on the pre-column with mobile phase A (25% methanol with 0.1% formic acid added). Loading flow rate was 0.03 ml/min for 6 min. The samples were eluted on to the nano LC column using mobile phases B (60% acetonitrile) and C (100% methanol). A typical gradient profile was 100% B to 100% C in 10 min (flow rate of 0.2 ml/min) with the column held at 100% C for 1 hour. The mass spectrometer was operated in the negative ion nano electrospray mode with a source temperature of 80uC and capillary voltage 2.8 kV. The scan range was 40 to 400 Da for 1.5 s. Liver fluke extract and excretory/secretory (ES) product preparation F. hepatica adults were collected, cultured in vitro for 4 h and the ES products collected and prepared as previously described [40] . Newly excysted juveniles (NEJ) were excysted from metacercariea in vitro and cultured in Fasciola saline for 4 h post excystment as previously described [41] . F. hepatica (adult and NEJ) soluble fractions were obtained by homogenisation of frozen fluke at 4uC in a glass grinder in lysis buffer (20 mM KHPO 4 , pH 7.0, 0.1% Triton-X100 and a cocktail of protease inhibitors [Roche, Complete-Mini, EDTA-free]). Homogenates were centrifuged at 100,0006 g for 1 h at 4uC. Supernatants were considered as the soluble cytosolic fraction. Cytosolic protein extracts were treated and resolved by 2DE as described previously [11] . F. hepatica eggs were isolated, cultured and protein extracted as previously described [42] . Recombinant F. hepatica Sigma GST (rFhGST-S1), and native F. hepatica S-hexylGSH-affinity purified GST samples (and human/rat recombinant PGD-synthase) were subjected to standard SDS-PAGE and 2DE, electro-transferred to membranes [43, 44] and western blotted with a polyclonal antibody (1:20,000 dilution) raised in rabbits to the recombinant F. hepatica Sigma GST by Lampire Biological Laboratories, USA. Membranes were also probed with Mu class GST antibody (represented by the anti-Schistosoma japonicum GST26 Mu class antibody [1:1,000 dilution] and an anti-rat PGD-synthase antibody [1:1,000 dilution], Pharmacia-Biotech 27-4577). F. hepatica eggs, NEJs (somatic and ES preparations) and adults (somatic and ES preparations) were subjected to SDS-PAGE and also electro-transferred as described above and probed with the polyclonal antibody raised in rabbits to the recombinant F. hepatica Sigma GST. All western blots were developed as described previously [11] . Immunolocalisation studies F. hepatica Sigma class GST (FhGST-S1) was detected by immunohistology in tissue sections of whole adult F. hepatica extracted from bile ducts of sheep liver and also in situ from sections of liver. Staining for FhGST-S1 was performed on formalin-fixed and paraffin-embedded tissue sections according to the method described previously [45] . Sections were washed in Tris-buffered saline (TBS; 0.1 M Tris-HCl with 0.9% NaCl [pH 7.2]), treated with 0.05% (w/v) protease (type XXIV, bacterial: Sigma) in TBS for 5 min at 37uC for antigen retrieval, before three further 5 min washes in ice-cold TBS. Following TBS washes, sections were incubated for 10 min in 50% (v/v) swine serum in TBS followed by incubation for 15-18 h at 4uC in rFhGST-S1 polyclonal antibody (diluted at 1:500 in 20% swine serum in TBS). Sections were again washed in TBS before further incubation at ambient temperature (approximately 20uC+/23uC) with anti-rabbit peroxidise anti-peroxidase (PAP; diluted at 1:100 in 20% swine serum in TBS). Following washes with TBS, sections were incubated, with stirring, for 10 min, with 3,3-diaminobenzidine tetrahydrochloride (DAB; Fluka, Buchs, Switzerland) with 0.01% v/v hydrogen peroxide in 0.1 M imidazole buffer pH 7.1, before counterstaining with Papanicolaou's hematoxylin for 30 s. Sections were then rinsed, dehydrated in alcohol, cleared in xylene, and mounted. Consecutive sections from each tissue were used as negative controls in which the rFhGST-S1 polyclonal antibody was replaced by TBS. Animals. C57BL/6 mice were purchased from Harlan Ltd (UK) and TLR4KO (on a C57BL/6 background) bone marrow cells were a gift from Professor Padraic Fallon (Trinity College Dublin, Ireland). All mice were maintained according to the Irish Department of Children and Health. Cell culturing and cytokine analysis. Bone marrowderived immature dendritic cells (DCs) were prepared by culturing bone marrow cells isolated from the femurs and tibia of C57BL/6j and TLR4 2/2 mice in complete RPMI 1640 (cRPMI; 5% [v/v] heat inactivated Fetal Calf Serum [FCS] [30 mins at 60uC], 100 U/ml penicillin, 100 mg/ml streptomycin, 2 mM L-glutamine and 50 mM 2-mercaptoethanol) with recombinant mouse GM-CSF (20 ng/ml; R&D Systems), at 37uC. On days 3 and 6 of culture, fresh medium with GM-CSF (20 ng/ml) was added to the cells. On day 8, cells were harvested, counted and stained with CD11c (Caltag Laboratories) for analysis by flow cytometry to determine purity (.90%). J774 cells and RAW264.7, murine macrophage cell lines were cultured in cRPMI 1640 medium containing 10% (v/v) FCS. All cells were used to conduct experiments when they reached ,90% confluence. For all experiments, cells were seeded into 24-well plates (Nunc) at 10 6 /ml in complete RPMI 1640 except for DCs where GM-CSF (5 ng/ml) was also added. Cells were treated with medium only, rFhGST-S1 (10 mg) or LPS (Alexa; 100 ng/ml) for 18 h. Levels of total prostaglandin (PG), prostaglandin E2 (PGE2) and prostaglandin D2 (PGD2) were measured using the Cayman competitive EIA. The values were calculated using free data analysis software available at www.caymanchem.com/analysis/ eia. Data are presented as the mean 6 SEM following subtraction Sigma Class Glutathione Transferase of F. hepatica www.plosntds.org of medium controls and are representative of two separate experiments. Prior to experimentation, rFhGST-S1 was assessed for endotoxin (i.e. LPS) contamination using the Pyrogene endotoxin detection system (Cambrex). Experimental design. Nineteen 5-month old, Malagueña breed goats were used for a vaccine trial. The animals were free of parasitic and infectious diseases as indicated by fecal analysis and absence of clinical signs. Group 1 (n = 10) were immunized with two subcutaneous injections of 100 mg of rFhGST-S1 in 1 ml of Quil A in 1 ml of PBS each injection separated by 4 weeks. Group 2 (n = 9) served as an infected control and was immunized at the same time with 1 ml of Quil A in 1 ml of PBS. Twelve weeks after the first immunization, animals were orally infected with 100 F. hepatica metacercariae of bovine origin. Three animals from each group were killed at 7, 8 and 9 days post-infection to study hepatic changes and host response during the early stages post-infection; the remaining animals (7 and 6 goats per group) were killed 15 weeks after infection to study fluke burdens, fecal egg counts and hepatic lesions. All goats were sacrificed by intravenous injection of thiobarbital. The experiment was approved by the Bioethical Committee of the University of Cordoba (N. 7119) and it was carried out according to European (86/609/CEE) and Spanish (RD 223/1988) directives for animal experimentation. Fluke burdens and morphometrics. At necropsy, gallbladders and bile ducts were opened and flukes recovered. Llivers were cut (,1 cm pieces) and washed in hot water to collect the remaining flukes. Flukes were counted, measured and weighed. Fecal egg counts. Sedimentation techniques at 12 and 13 weeks after infection using four grams of feces were conducted to give eggs per gram (EPG). Pathological assessment. At necropsy livers were photographed by the visceral and diaphragmatic aspects for gross pathology evaluation as described previously [46] . Gross hepatic lesions were scored as: absent [2] ; mild [+] (less than 10% of hepatic surface affected); moderate [++] (10-25% of hepatic surface affected); severe [+++] (25-50% of hepatic surface affected) and very severe [++++] (more than 50% of hepatic surface affected). Tissue samples were collected from the left (6 samples) and right (2 samples) hepatic lobes, fixed in 10% buffered formalin and embedded in paraffin wax. Tissue sections (4 mm) were stained with haematoxylin and eosin (HE) for the histopathological study. Specific IgG response. Specific IgG anti-rFhGST-S1were measured using the ELISA method as described previously [46] . A total of 10 mg/ml of rFhGST-S1 was used to coat microtitre plates, 100 ml/well of goat sera diluted in blocking buffer and rabbit antigoat IgG peroxidase conjugated (whole molecule -Sigma) diluted in blocking buffer at 1:10000. Serum pools were from ten experimentally infected goats and ten uninfected goats as positive and negative controls, respectively. All samples were analysed in duplicate. Results were expressed as antibody titre (Log10). Aligning Sigma class GSTs of trematodes shows the extent of identity and similarity across this class of GSTs ( Figure S1 ). An amino acid sequence comparison of FhGST-S1 with other trematode GSTs places FhGST-S1 into the Sigma class of GSTs, with identities averaging approximately 45%. Comparison with the most closely matching mammalian GSTs shows sequence identities averaging only approximately 28% (Table S1 ). Despite phylogenetic neighbour-joining trees place mammalian and trematode GSTs within the same broad Sigma class ( Figure S1 ) there remains a distinct separation of the trematode and mammalian clusters. Full sequence length recombinant F. hepatica Sigma Class GST (rFhGST-S1) was shown to be purified to a high level from transformed E. coli cytosol following expression yielding 57.3 mg of rFhGST-S1 from a 1 litre culture of BL21 (DE3) cells. Purity was judged by the presence of a single band upon SDS-PAGE at the estimated size and a dominating single peak via ESI MS at the precise calculated theoretical mass for the complete protein sequence (Figure 1 ). Analysing this fraction by 2D SDS-PAGE revealed a single protein resolving into 3 protein spots. Western blotting of the 2DE profile with anti-rFhGST-S1 antibody confirmed all 3 resolved protein spots as rFhGST-S1 (2DE and western blot data not shown). No recognition was seen probing the 3 spots with an anti-Mu class antibody. rFhGST-S1 was produced as an active protein, displaying significant enzymic activity towards the model GST substrate 1chloro-2,4-dinitrobenzene (CDNB) and a range of substrates commonly used to characterise GSTs (Table 1) . F. hepatica GST is very similar in terms of its enzymatic profile to the GST of S. Sigma Class Glutathione Transferase of F. hepatica www.plosntds.org japonicum currently undergoing clinical vaccine trials. FhGST-S1 also displays higher glutathione-dependent lipid peroxidase activity compared to both Sm28GST and Sj26GST [47] . Interestingly, ligand inhibition studies on rFhGST-S1 showed the enzymic activity of rFhGST-S1 with CDNB was inhibited by the major pro-active form of the main liver fluke drug Triclabendazole. The sulphoxide derivative (TCBZ SO) gave an IC50 (50% enzyme inhibition) of 5765 mM (5 replicates). Bile acids, potentially natural ligands for liver fluke tegumental associated proteins in the host bile environment, were also assessed for activity inhibition. The rFhGST-S1 interacted with all three bile acids tested using five replicate assays: Cholic acid (IC50 302673 mM); Deoxycholic acid (IC50 223621 mM) and Chenodeoxycholic acid (IC50 6469 mM) . Previous studies on the Sigma class GSTs from both mammals and helminth parasites have revealed a capacity to synthesise Prostaglandin D2 (PGD2) and PGE2. Since prostaglandin synthase activity may be a conserved role of Sigma class GSTs, we also tested the ability of rFhGST-S1 to synthesise prostaglandin eicosanoids using a coupled assay with COX-1. COX-1 catalyses the conversion of arachidonic acid to the H2 form before the prostaglandin isomer is converted to either the D or E form. Nano-LC/MS analysis enabled us to detect the presence of both PGD2 and PGE2 in the assay mixture with the PGD2 form being the more abundant of the two prostanoids ( Figure 2 ). While some PGE2 in the mixture could have arisen from rapid degradation of the unstable PGH2, nano-LC-MS was unable to detect either PGD2 or PGE2 in negative control reactions lacking either COX-1 or GST. The rFhGST-S1 catalyses PGD2 formation in a concentration-dependent manner as previously described for rOvGST-1 [26] . PGD2 was also detected in coupled assays with rFhGST-S1 and COX-1 using an Enzyme Immno Assay (EIA) detection kit (Cayman) and showed similar results (results not shown). FhGST-S1 was first identified in adult liver fluke in S-hexyl-GSH affinity isolated fractions of cytosol [11] . Western blots confirmed the presence of FhGST-S1 in NEJs and adult flukes and further enabled us to identify the Sigma GST in relative abundance in egg extracts, suggesting that it may play a metabolic role in embryogenesis/reproduction (Figure 3) . Western blot analyses demonstrate that FhGST-S1 is consistently expressed during the course of in vitro parasite embryonation (days 1-9, only data for days 2, 7 and 9 shown in Figure 3 ). In contrast, immunoblot analysis of freshly voided (day 0) eggs reveals that expression of the Sigma class GST is greatly reduced at the time of voiding from the host (Figure 3) . However, immunolocalisation studies of adult parasites revealed an abundance of FhGST-S1 in the vitelline cells and eggs, emphasising the likely importance of this enzyme in egg formation and development. Some staining was also found in the parasite parenchyma and tegument, also suggesting a role at the hostparasite interface ( Figure 4) . Indeed, FhGST-S1 was detected in ES products of adult fluke cultured in vitro ( Figure 3 ) suggesting that the protein could, in principle, come into contact with the host immune system as it is released from the tegument during tegumental turnover and sloughing of the fluke body surface. Figure 2 . Detection of prostaglandin synthase activity of rFhGST-S1 via a mass spectrometry approach. A coupled assay with rFhGST-S1 and COX-1 catalyses the conversion of arachidonic acid to the H2 form before the prostaglandin isomer is converted to either the D or E form. Nano-LC/MS analysis allowed detection of both PGD2 (A) and PGE2 (B) in the assay mixture with the PGD2 form being the more abundant of the two prostanoids (C). Boxed figures above peaks show the fragmentation ions specific to detection of PGD2 (a) and PGE2 (b) according to the method of Schmidt et al. [39] . doi:10.1371/journal.pntd.0001666.g002 Sigma Class Glutathione Transferase of F. hepatica www.plosntds.org Influence of rFhGST-S1 on prostaglandin synthesis in host immune cells rFhGST-S1 exhibited prostaglandin synthase activity producing PGE2 and PGD2. In addition, it has been shown previously that rFhGST-S1 activates DCs in vitro [48] . Therefore, an attempt to determine if rFhGST-S1 could induce the secretion of total prostaglandin, PGE2 and PGD2 from DCs was performed. Prior to experimentation, endotoxin levels in rFhGST-S1 were assessed and were similar to that of the media alone. Both of which were below the lower limit of detection (,0.01 EU/ml). When examining prostaglandin induction DCs stimulated with rFhGST-S1 secreted total prostaglandin and PGE2 (DC (WT); Figure 5 ) but not PGD2 (data not shown). Since it has been previously determined that the activation of DCs by rFhGST-S1 was dependent upon TLR4 [48] we repeated the experiment in DCs from TLR4KO mice and in keeping with previous findings demonstrated that the secretion of total prostaglandin and PGE2 by rFhGST-S1 was significantly reduced in the absence of the TLR4 receptor (DC (TLR4KO); Figure 5 ). rFhGST-S1 was then further assessed for its potential to induce prostaglandin secretion from macrophages by exposing two macrophage cell lines with rFhGST-S1. After 18 hours the levels of total prostaglandin, PGE2 and PGD2 were measured. In this assay, both macrophage cells lines stimulated with rFhGST-S1 secreted total prostaglandin, PGE2 and PGD2 ( Figure 6 ). However, the levels secreted by J744 cell line were higher when compared to the amount secreted by RAW264.7 cell line. In these experiments we included medium only as a negative control and LPS as a positive control. In all experiments the levels of prostaglandin in response to rFhGST-S1 was comparable to the levels secreted in responses to LPS. Assessment of goat vaccinations with rFhGST-S1 challenged with F. hepatica Following the completion of the vaccine trial, liver fluke were recovered and the livers scored. The resulting data is summarised in Table 2 . When assessing fluke burdens, length, weight and fecal egg counts, no significant differences between rFhGST-S1 immunised and Quil A immunised groups were observed. Despite this lack of significance, at 7-9 days post-infection (dpi) the number of gross hepatic lesions appeared reduced in rFhGST-S1 immunised groups compared to the Quil A control group. At 15 weeks post-infection (wpi), a similar outcome is observed. Liver hepatic lesion scoring appeared to show reductions in the severity of damage occurred in the rFhGST-S1 immunised group compared to the Quil A only group, despite no significant differences in the aforementioned morphometric data. Microscopically, at 7-9 dpi animals from the Quil A group showed tortuous necrotic tracts surrounded by a scarce inflammatory infiltration with occasional eosinophils ( Figure 7A ). Older necrotic areas were surrounded by macrophages, epithelioid cells and multinucleate giant cells and lymphocytes. Some migrating larvae were found in the liver parenchyma without inflammatory infiltrate associated to them. In goats immunised with rFhGST-S1 smaller necrotic areas associated to a heavy infiltration of eosinophils ( Figure 7B ) were seen. Unlike the Quil A immunised group, all migrating larvae found were surrounded by a heavy infiltration of eosinophils. A significant increase of IgG anti-rFhGST-S1 was observed two weeks after vaccination with a strong increase after the second injection at week 4 in immunised animals ( Figure 8 ). The Quil A control group did not show any specific IgG response until 2 weeks after infection. Specific IgG titres increased during infection in both groups, but they were consistently higher in the immunised group throughout the duration of the the experiment. Previous studies have highlighted the importance of parasite GSTs, including Sigma class GSTs, in host-parasite interactions and as potential vaccination candidates. With this in mind, we have studied the relatively newly identified Sigma class GST from F. hepatica to both enhance our understanding of this important enzyme in Fasciola and the Sigma class of GSTs as a whole. Alignments and phylogenetics classified FhGST-S1 alongside trematode and mammalian Sigma class GSTs, yet there remains a Sigma Class Glutathione Transferase of F. hepatica www.plosntds.org distinct divide between the parasites and their hosts, a phenomenon also observed for the recently reclassified 'Nu' class of GSTs from nematodes [49] . Therefore, it may be that trematode GSTs are sufficiently distinct to support a sub-classification within the broad Sigma class. The distinction of FhGST-S1 from fasciolosis host Sigma class GSTs enhances its potential as a therapeutic target. Substrate activity profiling of rFhGST-S1 using model substrates showed the enzyme to have comparable activity to other trematode Sigma class GSTs such as Sm28GST [47] . However, rFhGST-S1 exhibits relatively high GSH-conjugating activity towards the potentially natural reactive aldehyde, 4-hydroxy-nonenal (4-HNE) toxin and high GSH-dependent peroxidase activity towards the tested lipid peroxides which includes the endogenous substrate linoleic acid hydroperoxide. 4-HNE is the major aldehydic end-product of lipid peroxidation that is involved in signalling of host immune cells leading to apoptosis of T-and Bcells [50] . Assessing the inhibition of rFhGST-S1 activity with CDNB revealed that both bile acids and the flukicide TCBZ appear to bind to the enzyme. In particular, the interaction of the bile acid cholate with rFhGST-S1 is approximately ten fold higher than GSTs from the sheep intestinal cestode Moniezia expansa [51] . Host bile acids are known as triggers of physiological processes in trematodes including Fasciola sp. [52, 53] . Therefore, molecular interaction of bile acids with FhGST-S1 warrants further investigation especially, given that FhGST-S1 is localised to near the body surface of the fluke, where it could potentially bind cholate and other free bile acids found in abundance in host bile (cholate is found at approximately 100 mM in sheep bile) [54] . The hydroxy-TCBZ SO levels in the bile have been shown to be in excess of 100 mM [55] thus, the IC50 of 5765 mM for TCBZ SO suggests the abundant FhGST-S1 could be involved in TCBZ response in phase III sequestration based detoxification. This finding warrants further investigation to understand the role of FhGST-S1 in TCBZ action or detoxification. Sigma class GSTs from both parasites and mammals have been known to exhibit prostaglandin synthase activity. To this end, the Sigma GST from F. hepatica shares a high sequence identity with recognised Sigma class GSTs with prostaglandin synthase activity, including rOvGST-1 from the filarial parasite, Onchocerca volvulus. Using a coupled assay with COX-1 we have shown that rFhGST-S1 is capable of synthesizing both PGD 2 and PGE 2 , with PGD 2 being the predominant prostanoid. Parasite-derived eicosanoids, including prostaglandins, are known to be important in the establishment of parasitic infection and the survival and proliferation within the host. Therefore, eicosanoids produced by parasitic helminths may play a role in pathophysiological changes during helminth infections. For example, chronic fasciolosis is associated with fever and changes in liver biochemistry, both of which could be associated with parasite-derived eicosanoids thromboxane B2 (TXB2), PGI 2 , PGE 2 and leukotriene B4 (LTB 4 ), detected in the ES products and homogenates of adult F. hepatica worms [56] . In addition, the migration of host epidermal Langerhans cells, which play a key part in immune defence mechanisms, has been shown to be inhibited by parasite-derived PGD 2 in the Schistosoma mansonimouse model of human infection, thus allowing schistosomes to manipulate the host immune system [57] . Earlier studies have revealed the presence of eicosanoids produced by S. mansoni cercariae which could also play a role in establishment of infections through loss of the cercarial tail following penetration of the skin [58] . It therefore seems likely that prostaglandins synthesised via FhGST-S1 will have a role in establishing the infection within the host. In general, prostaglandins and eicosanoids have potent biological activities in reproduction. For example in the zebrafish egg, high levels of PGE 2 were seen post fertilisation coupled with high PGD 2 synthase transcript levels during the early stages of egg Figure 4 . Images of FhGST-S1 localisation within F. hepatica tissue. A) Anti-F. hepatica FhGST-S1 immunohistochemical stain of a fluke in cross section within the host sheep liver bile duct. Heavily stained eggs (E) are shown released from the fluke into the bile duct in the top left-hand corner. Brown stained areas show the presence of FhGST-S1 proteins. The lack of staining in the host liver (L) highlights the specificity of the antibody. Composite picture. B) Enlarged region of A showing the intense anti-F. hepatica FhGST-S1staining in the voided eggs (E). The spines (S) present in the tegument (T) can be clearly distinguished by their lack of FhGST-S1 presence. C-E) Cross sections of a F. hepatica adult highlighting staining of FhGST-S1 in the parenchyma (P), musculature (M),the tegument (T), basal membrane (Bm) and most intensely in the vitelline cells (V) and developing eggs (DE) . No staining can be seen in the tegumental spines (S), testes (T) or the intestinal caecum (IC). doi:10.1371/journal.pntd.0001666.g004 Figure 5 . rFhGST-SI stimulates the production total prostaglandin and PGE2 from dendritic cells (DCs) in a TLR4 dependent manner. DCs derived from the bone marrow from C57BL/ 6j mice were cultured in vitro with medium, rFhGST-S1 (10 mg/ml) or LPS (100 ng/ml) for 18 hours, and the production of total prostaglandin, PGE2 and PGD2 (data for PGD2 not shown) released into supernatants determined by competitive EIA. Data are presented as the mean 6 SEM following subtraction of medium controls and are representative of two experiments. WT -wild type; TLR4KO -Toll like receptor 4 knock out. doi:10.1371/journal.pntd.0001666.g005 Sigma Class Glutathione Transferase of F. hepatica www.plosntds.org development concomitant with an exponential decrease of PGD 2 levels over the next 120 h post fertilisation [59] . However, in F. hepatica, eggs in gravid adults are released in an immature state in the bile duct, where they pass to the external environment via the host's excretory system and complete embryogenesis ex-host. Therefore, FhGST-S1 may have a secondary, or indeed primary, function in egg development and embyrogenesis. A role in egg development is further supported by proteomic studies of F. hepatica ontogenic stages which reveal the presence of FhGST-S1 in eggs ( [42] and the current study). FhGST-S1 appears to be highly abundant in eggs with western blotting showing FhGST-S1 to be constitutively expressed, despite its association with a large spot consisting of multiple co-migrating proteins unresolved via 2DE (for association see [42] ). Immunolocalisation studies revealed that FhGST-S1 is closely associated with vitelline cells of mature adult worms. Given the importance of PGs in reproduction, we hypothesize that PG synthase activity exhibited by rFhGST-S1 contributes to developmental cues during egg formation. Interestingly, no FhGST-S1 was seen in day 0, unembryonated, eggs by western blotting yet in situ immunlocalisation showed freshly voided eggs, equivalent to day 0 eggs, to contain copious amounts of FhGST-S1. While it is most likely that FhGST-S1 is present in day 0 eggs, albeit at a reduced expression, the discrepancy seen between the two techniques is probably related to the antibody dilutions used for each method; in total a 40-fold difference in favour of immunolocalisation. FhGST-S1 was also identified in both NEJs and adult worms using western blotting. This finding emphasises the multifunctionality of FhGST-S1, where in NEJs egg productions is not yet in process, suggesting its main function is in PG synthesis for host modulation or as a detoxification enzyme. In the adult worm, FhGST-S1 could also be localised, to a smaller extent, in the parenchyma and tegument. Given the high activity of FhGST-S1 towards the toxic 4-HNE and to lipid hydroperoxides this suggests a detoxification role at the host-parasite interface. With near surface expression of FhGST-S1, in the parenchyma and tegument, there is the potential for this enzyme to be readily released into the host environment. Indeed, we have identified FhGST-S1 in the ES products of adult worms. With this in mind, previous studies have highlighted the importance of parasite Sigma class GSTs in immunomodulation of the host immune response. This includes our recent study implicating rFhGST-S1 in chronic inflammation through the activation of dendritic cells (DCs) [48] . While active rFhGST-S1 was able to induce levels of IL-12p40 and IL-6 cytokines in DCs in a dose-dependent manner, the previously described F. hepatica Mu-class GSTs failed to induce any cytokine secretion. Since denatured rFhGST-S1 also failed to induce any cytokines in DCs, activation of DCs is likely related to the structure and activity of the enzyme. However, inhibition of nitric oxide production, involved in driving a Th2 immune response, may also be a contributing factor in skewing the host response to fasciolosis [60] . F. hepatica infections are associated with a T-helper-cell type 2 (Th2) immune response dominating during the chronic phases of infection [61] , but pro-inflammatory responses are suppressed [62] . Suppression of allergic responses during chronic parasitic worm infections has a mutually beneficial effect on the parasites' proliferation and the hosts' survival. Prostanoids, including PGD 2 , are important in mediating these allergic inflammatory responses. While generally regarded as pro-inflammatory molecules, these important lipid molecules are also involved in mediating antiinflammatory responses [63] . Helminth-derived molecules are thought to be involved in driving the Th2 response stereotypical of parasitic worm infections. DC and macrophage cell cultures Figure 6 . rFhGST-SI stimulates the production PGE2 and PGD2 from the macrophage cell lines J774 and RAW264.7. J744 and RAW264.7 macrophage cell lines were cultured in vitro with medium, rFhGST-S1 (10 mg/ml) or LPS (100 ng/ml) for 18 hours, and the production of total prostaglandin, PGE2 and PGD2 released into supernatants determined by competitive EIA. Data are presented as the mean 6 SEM following subtraction of medium controls and are representative of two experiments. doi:10.1371/journal.pntd.0001666.g006 Sigma Class Glutathione Transferase of F. hepatica www.plosntds.org exposed to rFhGST-S1 showed elevated levels of Th2 cytokines after 24 h [48] . In this study, the effects of rFhGST-S1 exposure onprostanoid synthesis in host immune cells was investigated. The results of which show the stimulation of PGD 2 and PGE 2 in both DCs and macrophage cell lines suggesting FhGST-S1 is one such helminth derived molecule capable of driving the Th2 response. As we have shown FhGST-S1 to have key roles in F. hepatica, both in NEJs and adult worms, coupled with the near surface expression and release of the enzyme via the ES products, we assessed the potential of FhGST-S1 to be used as a vaccine candidate. This was especially poignant given that the S. mansoni Sigma GST homologue (Sm28) is in phase II clinical trials [12] . Unfortunately, the current goat based vaccine trial did not show any significant differences in fluke burdens between the rFhGST-S1 immunised and Quil A control group. However, a high individual variability was recorded, particularly in the vaccinated group also reported in previous trials using goats vaccinated with alternative candidates such as cathepsin L1 [64] and Sm14 [65] . The vaccine trial shown here using a target species with an acceptable adjuvant may have been adversely affected by the strain of F. hepatica used to challenge goats. Here we have shown an unusually high infectivity rate with the strain of F. hepatica used; which we have reported in a previous trial using goats [64] . Using an alternative strain of F. hepatica for experimental infections in this species has given normal infectivity rates ranging from 14% to 26.5% [65] . In the present trial it appeared that goats immunised with rFhGST-S1, despite no variations in fluke burdens or morphometrics, showed reduced gross hepatic lesions during early infection, up to day 9 post infection, which continued to week 15 post infection where liver scores for hepatic lesions appeared reduced for rFhGST-S1 immunised animals. These results suggest that animals from the immunised group produced an early response to migrating larvae that has induced some partial protection from liver damage. The early and consistent specific IgG response found in the present work also agrees with the results obtained in a previous trial using naïve FhGST [46] . However, in both studies high levels of specific IgG did not induced a protective response reducing worm burdens. A promising aspect of producing anti-helminth vaccines is developing multivalent vaccines. In many cases the greatest protection from challenge is by vaccinating with a combination of Fasciola antigens [66, 67] . Therefore, based on the immunisation with FhGST-S1 showing an early response reducing hepatica damage, could be considered for inclusion into a multivalent vaccine against Fasciolosis. In addition, in light of our findings showing FhGST-S1 to be highly prominent in egg production and the egg itself, as with previous vaccination trials [67] , it will be important to investigate the ability of eggs voided from vaccinated animals to embryonate. The potential to reduce pasture contamination by inhibiting egg embryonation, combined with the www.plosntds.org demonstrated reduction in liver damage, warrants further exploration using rFhGST-S1 as a vaccine candidate. In summary, we have further promoted the concept that FhGST-S1 clearly demonstrates key host-parasite roles in synthesising PGs and stimulating PG release from host innate immune cells. In addition we have shown FhGST-S1 to be a key protein for detoxification, which may well be involved in TCBZ response. In line with current vaccine development theory we have shown FhGST-S1 to have multi-functional roles in the liver fluke physiology. Furthermore, we have shown FhGST-S1 to be expressed across ontogenic stages, localised to the fluke surface, and to the egg, both characteristics vital for vaccine development and success. Whilst no protection from fluke burden was seen in trials, the inclusion of rFhGST-S1 as a multivalent vaccine component should be investigated. However, it is important to fully characterise the host immune response during the early stages post-infection to better understand the mechanism mediating an effective host response. This will be essential to improve any future vaccine formulation. [36, 51, [68] [69] [70] [71] Table 2 Refs. Figure S1 Multiple sequence alignment and neighbourjoining phylogenetic tree across seven species-independent classes of GSTs. A) Alignment of the sigma class GSTs of trematodes shows the extent of identity and similarity across this class of GSTs. Boxed residues indicate complete identity between all sequences. Residues shaded in grey indicate conserved residues. B) Neighbour-joining tree placing mammalian and trematode GSTs within the same broad Sigma class. A distinct separation of clusters within this Sigma class is observed as with the recently reclassified 'Nu' class of GSTs from nematodes [49] . Sequences were aligned via the ClustalW program [29] in BioEdit Sequence Alignment Editor version 7.0.5.2. [30] . Phylogenetic neighbourjoining bootstrap trees were produced and viewed within TREE-VIEW [33] . Key to sequences in 1a and 1b. Author Contributions

Discriminating Active from Latent Tuberculosis in Patients Presenting to Community Clinics

BACKGROUND: Because of the high global prevalence of latent TB infection (LTBI), a key challenge in endemic settings is distinguishing patients with active TB from patients with overlapping clinical symptoms without active TB but with co-existing LTBI. Current methods are insufficiently accurate. Plasma proteomic fingerprinting can resolve this difficulty by providing a molecular snapshot defining disease state that can be used to develop point-of-care diagnostics. METHODS: Plasma and clinical data were obtained prospectively from patients attending community TB clinics in Peru and from household contacts. Plasma was subjected to high-throughput proteomic profiling by mass spectrometry. Statistical pattern recognition methods were used to define mass spectral patterns that distinguished patients with active TB from symptomatic controls with or without LTBI. RESULTS: 156 patients with active TB and 110 symptomatic controls (patients with respiratory symptoms without active TB) were investigated. Active TB patients were distinguishable from undifferentiated symptomatic controls with accuracy of 87% (sensitivity 84%, specificity 90%), from symptomatic controls with LTBI (accuracy of 87%, sensitivity 89%, specificity 82%) and from symptomatic controls without LTBI (accuracy 90%, sensitivity 90%, specificity 92%). CONCLUSIONS: We show that active TB can be distinguished accurately from LTBI in symptomatic clinic attenders using a plasma proteomic fingerprint. Translation of biomarkers derived from this study into a robust and affordable point-of-care format will have significant implications for recognition and control of active TB in high prevalence settings.

Tuberculosis is the leading bacterial cause of death worldwide, with an estimated 8.8 million new cases of active disease and 1.6 million deaths per year [1] . Much of the burden of disease lies in the developing world, where annual incidence can reach 700 per 100,000 in certain regions [1] . New and unrecognised cases drive the epidemic, with transmission usually occurring before the index case is diagnosed. Multi-drug resistant cases and HIV co-infection further complicate control efforts [2] . Pulmonary TB is the most frequent clinical and transmissible manifestation of active disease. Rapid diagnosis and treatment are critical in the prevention of transmission. The global burden of active TB occurs on a background of quiescent or latent TB infection (LTBI), affecting one third of the world's population and a higher proportion of the population of TB-endemic areas [3] . Respiratory and constitutional symptoms overlapping with those of pulmonary TB are very common in communities where TB is endemic [4] . In this scenario the challenge is to distinguish symptomatic patients with active TB from those with latent disease but whose presenting symptomatology is attributable to some other infectious or inflammatory process. In terms of rapid diagnosis, sputum microscopy will only identify approximately 50% of patients with active pulmonary TB. Conversely, while the interferon gamma release assays (IGRAs) represent a major advance in the detection of latent TB, they cannot distinguish active TB from symptomatic patients with latent infection in this context [5, 6] . This overlap between LTBI, active TB and non-specific clinical manifestations presents a formidable obstacle to the rapid recognition of active TB and the timely and appropriate targeting of anti-TB chemotherapy or chemoprophylaxis. In practice this difficulty may give rise to 2 types of therapeutic error. In the first instance, erroneous diagnosis of active TB in a symptomatic patient with LTBI may result in inappropriate administration of full course TB treatment. Conversely, offering chemoprophylaxis to a patient with supposed LTBI in whom active TB has not been recognized, will drive emergence of drug resistance. Pulmonary TB is characterised by granuloma formation, caseation and ultimately cavitation, reflecting a complex interplay between distinctive components of the innate and acquired immune response and the pathogen [5] . Traditional serological analysis of single circulating proteins is notoriously unreliable for TB diagnosis [7] . In contrast, patterns of circulating proteins could provide an accessible readout of pathophysiological status. Discovery of such discriminatory biomarkers could open the way for the development of new point-of-care tests based on a lateral flow format such as dipsticks. Proteomic analysis using Surface Enhanced Laser Desorption Ionisation Time of Flight (SELDI-ToF) mass spectrometry is a high throughput profiling methodology, which enables rapid comparison of protein patterns from large numbers of patients. The conceptual approach employed in the present study is termed proteomic fingerprinting. It is based on the principle that distinctive combinations of circulating proteins characterize different disease states. This strategy has been applied to the discovery of discriminatory proteomic patterns for a range of diseases including cancer [8] , vascular disease [9] [10] [11] and infectious diseases [12] [13] [14] [15] . Previously, we have demonstrated that proteomic patterns based on such profiles can distinguish active TB from healthy and symptomatic controls [12] . In the present study we hypothesized that plasma proteomic differences would also distinguish patients with active TB from those without active TB but with overlapping clinical symptoms, irrespective of the co-existence of LTBI. Here we show that using this approach, we can indeed discriminate accurately between such patient groups. All participants gave written informed consent and the research was approved by internationally accredited ethics committees including Universidad Peruana Cayetano Heredia (Lima, Peru) and Imperial College London (London, United Kingdom). The study involved adults from 15 years of age. Informed consent was obtained from the next of kin, carers or guardians on the behalf of the young adults involved in the study. Participants were recruited over a period of two years from adults over the age of 15 years attending 16 community TB clinics serving a population of ,400,000 in the shantytown of Ventanilla on the outskirts of Lima, Peru ( Figure 1 ). All patients underwent the local standardized clinical workup for TB. This included up to 4 consecutive sputum samples for microscopy and culture. Participation in the study did not change patients' routine clinical management. The local incidence of TB in this population is ,130 per 100,000/year [16] and 95-97% of TB cases are HIV negative [17] . We recruited patients with active TB and individuals, termed symptomatic controls, presenting with respiratory symptoms suspicious of TB in whom TB was subsequently excluded. Active TB cases were recruited on the basis of positive sputum microscopy with subsequent confirmation by culture. Mycobacterial culture was by automated liquid culture (BACTEC MGIT 960 TM , BD) as well as the Microscopic Observation Drug Susceptibility (MODS) assay which we have previously established as a standard local laboratory protocol [18] and which has since been adopted as the standard operating procedure by the national TB programme in Peru. Symptomatic controls, those patients with respiratory symptoms without active TB, were recruited if they had a persistent cough and one or more of the following clinical features: fever, weight loss, decreased appetite or haemoptysis. Symptomatic controls had 1-4 sputum smears and cultures to exclude active TB and were followed for 6 months to confirm that cultures had not become positive or were re-classified accordingly. Additional TB cases and symptomatic controls were identified through tracing household contacts, from whom sputum smears and cultures were obtained if symptomatic. An IFN-c Release Assay (IGRA) (QuantiFERON-TB Gold In-TubeH) was performed on all participants. Latent TB was defined as a positive QuantiFERONH assay in the absence of clinical or microbiological evidence for active TB. The Tuberculin Skin Test (TST) has limited value in the diagnosis of active TB and it was not carried out in our active TB patient group. We carried TST in the symptomatic controls group. A 4 ml blood sample was obtained from each participant in an EDTA blood collection tube for subsequent plasma separation. Three additional aliquots were obtained at the same time for the QuantiFERONH-TB Gold in tube assay. Plasma was obtained before initiating TB treatment; otherwise plasma was taken within 1-2 days of treatment. Blood samples were transferred to the central laboratory on ice. Plasma was separated (3500 rpm, 10 minutes), aliquoted and frozen at 270uC at 6 hours following collection. QuantiFERONH -TB Gold in Tube Assay This was performed according the manufacturer's instructions (Cellestis Plc, Sydney, Australia). Plasma was profiled using Surface Enhanced Laser Desorption/ Ionisation-Time Of Flight (SELDI-TOF) mass spectrometry. All samples underwent a single freeze-thaw cycle prior to analysis. Samples were coded, blinded and randomised before application onto weak cation exchange (CM10) ProteinChipH arrays (Bio-Rad) in duplicate, as previously described [12] . Each ProteinCh-ipH included 1 quality control standard derived from a single healthy individual, placed at random. Liquid handling steps were automated using a Biomek 3000 Laboratory Automation Workstation (Beckman Coulter) and a 96 well BioprocessorH (Bio-Rad). Mass spectra were generated on an automated System 4000 Bio-Rad ProteinChipH reader. Mass spectra data were collected and analysed using the ProteinChipH Data Manager Client 3.5 software (BioRad Inc.). Spectra were generated at both high (3,000 nJ) and low (1,600 nJ) laser energies with mass focus set to 40,000 Da and 6,000 Da respectively. Spectra were normalised by total ion current starting with a minimum mass/charge (m/z) of 2,500. Spectra with normalisation factor outside mean 62 standard deviations were removed. The remaining spectra were re-normalised by total ion current. Spectral peaks corresponding to mass/charge (m/z) clusters were detected and clustered using the ProteinChipH Data Manager Client 3.5 software (BioRad Inc.) by auto-detecting peaks to clusters in two steps. For the first step a signal to noise ratio of 5 and valley depth of 3 were used, with a minimum peak threshold of 20% of all spectra. For the second step a signal to noise ratio of 3 and valley depth of 1 were chosen. The cluster window was set at 1.0 peak width and expression difference mapping performed over m/z range of 2,500 to 200,000. Instrument calibration was performed using All-in-1 Peptide and Protein calibrants (Bio-Rad). Reproducibility was determined by measuring the inter-ProteinChipH coefficient of variation (CV) for the quality control spectra, based on all peaks in the spectrum with intensity .1 mA. Overall interchip CV for the quality control sample was 20%, consistent with similar studies. Because highly abundant proteins/peptides suppress signal from lower abundance analytes in complex mixtures such as crude plasma, SELDI-ToF spectra were generated from both crude and pre-fractionated plasma to determine whether accessing the 'deeper' proteome yielded additional diagnostic information. Anion-ex- To visualize the covariance within the mass spectral profiles we used Principal Component Analysis (PCA). PCA encapsulates the covariance within a set of variables by extracting a ranked set of independent factors or principal components. The first 3 components encompass a high proportion (,95%) of the informational content of a multivariate dataset. We plotted each patient with respect to the first 3 components, in 3-dimensional space, color-coding according to patient group. Although PCA is useful for visualizing data it cannot provide a classification rule for discriminating between patient categories. To find such discriminatory proteomic patterns, we adopted a supervised learning approach in which patient categories are used to train an algorithm to derive a classification rule. We used a Support Vector Machine (SVM) method [19] . Briefly, we used 10-fold cross validation to select parameters for the SVM. For the final model parameters, we selected those that gave the overall highest accuracy across the whole 10 fold cross validation. We next selected a subset of the most relevant mass clusters using the Recursive Feature Elimination (RFE) algorithm [20] which ranks variables based on their contribution to the classifier. To obtain accuracy estimates for the classifier, we took 1000 random resamplings of the original data, using 90% for training and 10% for testing. We selected as a final classifier the one that produced the highest accuracy while requiring the smallest number of m/z clusters. Results were expressed as sensitivity, specificity and accuracy (proportion of correct classifications) and as Receiver Operator Characteristic (ROC) curves. We assessed the different performances of classifiers derived from crude and pre-fractionated plasma by comparing mean values for sensitivity, specificity and accuracy using unpaired 2-tailed t tests. Comparisons of categorical data were by Fisher's exact test. 151 patients with active TB and 110 symptomatic controls were recruited (Figure 1 ). Of patients with active TB, 139 were both smear and culture positive, with the remainder either smear or culture positive. 48% of symptomatic controls had LTBI on the basis of a positive QuantiferonGold assay. Symptomatic controls had clinical features overlapping those of active TB patients, including cough, haemoptysis, fever, night sweats and weight loss, although symptom duration was generally longer among TB patients. Similar proportions of TB patients and symptomatic controls reported a previous history of TB (22% vs. 18%). The proportion reporting a history of TB was higher among controls with LTBI than among those without but did not reach statistical significance. Patients with active TB had lower BMIs at the time of recruitment compared with symptomatic controls (21.6 vs. 24.1 p,0.001). As expected, a higher proportion of patients with LTBI based on a positive IGRA had positive TSTs (.10 mm) compared with those without LTBI (62% vs. 30%, p,0.001). There was a higher proportion of female patients among the symptomatic controls than among the TB group. The effects of this potential bias are discussed below. Other key clinical features of the participant groups are given in Table 1 . We plotted crude plasma global protein expression profiles in a heat map (Figure 2 ) that shows spectra patterns from active TB patients and unhealthy controls. The most striking area of upregulation in TB patients is seen in the 11 kDa region where a series of protein peaks are seen in red amongst TB patients ( Figure 2) . A parallel area of up-regulation is seen at 5 kDa and a third smaller area seen at the 21 kDa region (Figure 2 ). Inspecting in more detail the spectra in the 5.8 and 11.5 kDa regions ( Figure 3 ) reveals a complex of peaks at both these regions, which is more abundant in patients with active TB. We assessed overall separability of patient groups by PCA of mass spectra from crude and pre-fractionated plasma (Figure 4 ab) . In figure 4 , each patient sample is plotted in a 3-dimensional space defined by the first 3 principal components. The spectra from patients with active TB (purple spheres) cluster relatively tightly together and are well separated from symptomatic control patients (blue and green spheres) regardless of LTBI. This analysis, however, does not clearly separate symptomatic controls with or without LTBI (blue and green spheres, respectively). The SVM classifiers distinguished active TB from both classes of symptomatic controls. The ROC curves in Figure 5 (a-f) summarize the performance of the classifiers, in terms of the tradeoff between sensitivity and specificity, for each of the different comparisons. In each case, the area under the curve (AUC) exceeded 0.9, irrespective of whether crude or pre-fractionated plasma was analyzed, indicating a high level of discrimination. Tables 2 and 3 and Tables S1 and S2 summarize the performance of the classifiers in discriminating active from latent tuberculosis in symptomatic patients using the number of selected relevant m/z clusters ( Table 3 in brackets). It was possible to distinguish patients with active TB from undifferentiated symptomatic controls with partially overlapping respiratory and constitutional symptoms with an overall accuracy of 85% using crude spectra with 98 relevant m/z clusters (Table 2, Table 3 , Figure 5a) . A higher specificity for active TB (90% vs. 84%, p,0.001) was achieved using prefractionated plasma with a total of 54 relevant m/z clusters ( Table 2, Table 3 , Figure 5b) . Notably, these levels of discrimination were achieved despite nearly half of the symptomatic controls having LTBI (Table 1) . To further investigate the influence of background LTBI on classifier performance, separate comparisons were made between active TB and symptomatic controls either with or without LTBI. In both comparisons, active TB could be distinguished from symptomatic controls with overall classifier accuracies of at least 87% (Table 2, Table 3 , Figure 5 c-f, Tables S1 and 2). Active TB was readily distinguishable from symptomatic controls without LTBI using both crude and fractionated plasma, with overall accuracies, sensitivities and specificities of at least 90% (Table 2, Table 3 , Figure 5 e,f and Table S1 and S2). The main influence of LTBI among the symptomatic controls was to reduce classifier specificity, reflected in a higher proportion of false positives. Strikingly, plasma pre-fractionation improved specificity from 75% to 82% only using four m/z clusters (Table 2, Table 3 , Figure 5 c,d, p,0.001) . To address the issue of the gender bias in cases and controls we reanalysed the data to determine whether a classifier based on the proteomic profile could reliably discriminate males from females. This was found not to be the case, suggesting that gender is not a major confounder in our analysis. As a further test, a new classifier was trained on male patients alone, to discriminate active TB from symptomatic controls. When we applied the trained classifier to the female subjects, this classifier was nevertheless still capable of classifying TB to an accuracy of approximately 80%. We also confirmed the presence of differential expression of the Serum Amyloid A (SAA, 11.5-11.8 kDa) and transthyretin (13.7-13.8 kDa ) peak complexes which emerged in our previous study [12] as important informative markers for active TB. SAA was identified by specific immunodepletion (data not shown). In this study we have shown that a distinctive pattern of plasma proteins distinguishes patients with active TB from non-TB patients with overlapping clinical features, even in the presence of LTBI. This both reinforces and substantially extends our previous findings where we first showed that proteomic patterns could be used as a diagnostic approach for active TB [12] . We have now shown that the proteomic pattern does not merely reflect the presence of TB infection per se. Rather, it can be used to identify active TB even in a highly TB-endemic setting with high prevalence of both respiratory symptoms and background LTBI. (a,b) active TB vs. all symptomatic controls using crude or pre-fractionated plasma respectively; (c,d) active TB vs. symptomatic controls with latent TB using crude or pre-fractionated plasma respectively; (e,f) active TB vs. symptomatic controls without latent TB using crude or pre-fractionated plasma respectively. The ROCs are derived from 1000 random train/test resamplings of the data. Error bars show standard deviations. The Area Under the Curve (AUC) is shown in the centre of each plot. doi:10.1371/journal.pone.0038080.g005 Table 2 . Discrimination of active from latent tuberculosis in symptomatic patients. The ability to discriminate rapidly in a symptomatic patient between active TB and non-tuberculous disease has profound implications for both individual clinical management and TB control programs [21] . For example, current diagnostic limitations frequently result in many patients in resource-poor settings being treated empirically for community acquired pneumonia before eventual diagnosis of active TB. This may lead to on-going transmission during the interval preceding diagnosis as well as greater individual morbidity. The alternative strategy of empirical anti-TB chemotherapy is sometimes employed, but cost, toxicity and logistics often preclude this. Adjuncts to conventional microbiology for diagnosis of active TB in widespread use include the TST and IGRAs. The use of TSTs in the diagnosis of active TB in high prevalence settings is greatly limited by its poor specificity for active TB as reactivity is also seen in LTBI, previous BCG vaccination and exposure to environmental mycobacteria. Nor has the recent introduction of IGRAs into clinical practice resolved this key diagnostic issue. This is because of their inability to distinguish active TB from LTBI [6] and frequent false negative results in acute active TB [22] , limitations which are especially problematic in high prevalence settings [23] . Thus a diagnostic that overcomes these limitations is urgently required and would be a major advance in the management of the global TB pandemic. Recently it has been reported that a TNF-alpha + TB-specific CD4+response can be used to differentiate latent infection from active TB but the sensitivity was just 67% [24] . Moreover, that study relied on polychromatic flow cytometry limiting the feasibility of being translated in high prevalence settings. In contrast, our approach provides improved accuracy, 87%, by detecting relevant protein biomarkers in plasma. Despite the discovery-phase of our approach using sophisticated proteomic methodologies, the identification of relevant plasma proteins leads to a clear translational path for antibody-based point-of-care devices that can be used to measure these plasma proteins in the future. There is increasing interest in the identification of novel biomarkers for TB -in the contexts of diagnosis, treatment response monitoring, prediction of relapse or re-activation and as surrogates for vaccine protection. Most studies have focused on individual markers such as secreted M. tuberculosis antigens, serological responses, microbiological indices and host inflammatory markers, with mixed results [7, 25] . There is growing recognition of the advantages of using combinatorial biomarker panels or 'omics'-based methods to achieve sufficient levels of accuracy [25] . However, relatively few studies have utilized such strategies. Proteomic fingerprinting for biomarker discovery has been applied in the past decade to a variety of disease states, particularly in the sphere of cancer diagnostics [26, 27] . The power of this approach is reflected by the recent granting of FDA approval of a novel blood test derived from a SELDI-based fingerprinting method, for distinguishing malignant from benign ovarian tumours [27, 28] . In many infectious diseases, there are clinically important distinctions to be made between different manifestations associated with the same underlying pathogen. For example, distinguishing colonization or latent disease from active infection has obvious clinical and therapeutic implications. TB is a clear case in point. Proteomic fingerprinting has enormous potential for defining and distinguishing these disease states but has only recently received attention in this area [12] [13] [14] [29] [30] [31] . Because the circulation samples deep tissues throughout the body, local proteomic changes in organs such as the lungs can be reflected in the plasma proteome. Moreover, host modulation by the pathogen is likely to generate changing patterns of protein expression associated with different clinical manifestations. Thus the plasma proteomic response is a plausible index of disease state. Proteomic patterns are highly dynamic and it may be possible to define those that reflect stages in progression from latency to active disease. However, the complexity of the plasma proteome with its enormous dynamic range of solute concentrations means that detection of informative lower abundance proteins is particularly challenging. It is possible that differences between active TB and LTBI in symptomatic patients are reflected better by such lower abundance proteins not easily detectable in crude plasma. This may explain the higher specificity for active TB obtained from prefractionated as compared to the crude plasma spectra. The gold standards used for defining patient groups in this study are notoriously imperfect. For example, while active TB was defined by positive microbiology, it is possible that some patients designated symptomatic controls may actually have had smear and culture negative TB. This might have resulted in an underestimate of the specificity of our diagnostic pattern for active TB, although our 6 months follow-up and appropriate re-labelling should have identified most of these. The lack of an adequate gold standard for defining LTBI must also be considered. While IGRAs show greater specificity than TSTs, sensitivity may be compromised especially in early active TB [22] . Thus some patients with unrecognized smear and culture negative TB may have been mislabeled as symptomatic controls without LTBI. We did not perform routine HIV testing in our patient cohort and it is possible that over-representation of HIV seropositivity in our active TB group may have had a confounding effect. We believe this is unlikely in view of the low prevalence of HIV coinfection among TB patients in Peru (,5%) found in previous studies [17] . Important areas of future study will be to establish the applicability of this approach in the contexts of TB-HIV coinfection and smear-negative TB. Our present findings confirm the utility of defining the host proteomic response in distinguishing clinically overlapping patient (5) Total number of mass/charge (m/z) clusters obtained from SELDI-ToF profiling of crude and pre-fractionated plasma. In brackets number of relevant discriminatory m/z clusters selected by the RFE algorithm. F1 = fraction 1 at pH 9; F2 = fraction 2 at pH 7; F3 = fraction 3 at pH 5; F4 = fraction 4 at pH 4; F5 = fraction 5 at pH 3; F6 = fraction 6 organic phase. doi:10.1371/journal.pone.0038080.t003 groups in a TB clinic setting. Moreover, this study shows that active TB can be identified by a blood test in a population of community TB clinic attenders, on a background of non-TB attributable symptoms, despite the coexistence of LTBI. Ultimately, a significant impact on control of TB in high prevalence settings will depend on the ability to translate these findings into a robust, affordable point-of-care format. Incorporation of a panel of biomarkers derived from this study into a lateral flow device or similar platform is the logical next step. Finally, the utility of defining proteomic patterns in TB may extend beyond diagnostics to provide new methods for monitoring treatment response and disease stage. Table S1 Selected relevant m/z clusters from crude plasma. (XLS)

Novel Evidence of HBV Recombination in Family Cluster Infections in Western China

Two hepatitis B virus (HBV) C/D recombinants were isolated from western China. No direct evidence indicates that these new viruses arose as a result of recombination between genotype C and D or a result of convergence. In this study, we search for evidence of intra-individual recombination in the family cluster cases with co-circulation of genotype C, D and C/D recombinants. We studied 68 individuals from 15 families with HBV infections in 2006, identified individuals with mixed HBV genotype co-infections by restriction fragment length polymorphism and proceeded with cloning and DNA sequencing. Recombination signals were detected by RDP3 software and confirmed by split phylogenetic trees. Families with mixed HBV genotype co-infections were resampled in 2007. Three of 15 families had individuals with different HBV genotype co-infections in 2006. One individual (Y2) had a triple infection of HBV genotype C, D and C/D recombinant in 2006, but only genotype D in 2007. Further clonal analysis of this patient indicated that the C/D recombinant was not identical to previously isolated CD1 or CD2, but many novel recombinants with C2, D1 and CD1 were simultaneously found. All parental strains could recombine with each other to form new recombinant in this patient. This indicates that the detectable mixed infection and recombination have a limited time window. Also, as the recombinant nature of HBV precludes the possibility of a simple phylogenetic taxonomy, a new standard may be required for classifying HBV sequences.

Not all viruses are equally prone to recombination. Recombination has not been detected in several viruses despite repeated searches [1] . Whether recombination does or does not exist is important for understanding the evolution and replication mechanism of a specific kind of virus. Hepatitis B virus (HBV), a major human pathogen, has been classified into 10 genotypes and several sub-genotypes [2, 3] . Many sub-genotypes were identified by polygenetic analysis as recombinants. But there is no direct evidence to indicate that these subgenotypes arose as a result of recombination or perhaps a result of convergence. Coinfection with different HBV genotype strains is a prerequisite for recombination. As more than one genotype is predominant in most of the geographic regions, coinfection between the predominating HBV genotypes is not a rare finding, especially for B and C, or A and D. The prevalence of mixed HBV genotype infections has been reported using varied genotyping methods [4, 5, 6] . Our previous study found two kinds of HBV C/D recombinants in northwest China [7] . In a further study of ethnic groups of five provinces, we confirmed the geographic and ethnic distribution of the HBV C/D recombinant in northwest China [8] , and found that family-cluster HBV infections were common in these endemic areas. We hypothesize that infected members of HBV family clusters would gain exposure to various genotypes through marriage, while at the same time; competent strains would be selected through vertical transmission. It would be useful to observe the mixed infection in family-cluster cases, especially in patients infected with C/D recombinants. The aim of this study was to evaluate the possibility of recombination between two HBV genotypes within an individual by finding cluster-infected families in which individual members were infected with different HBV genotypes. We would then look for individuals within these families with multiple-genotypes that were likely to have been obtained from other family members as a result of vertical or horizontal transmission. Novel viral genomes within an individual with a multiple genotype infection that were mosaics of the known viral genotypes in the family, but not present in any of the other family members, would be consistent with the hypothesis that they arose within the individual with multiple genotype infections. We enrolled 68 patients with a chronic HBV infection from 15 families. All the families were from a district located at the boundary of Gansu and Qinghai provinces, where the prevalence of genotype C2, D1 and C/D recombinant HBV were known to be high [8] . The families were initially identified with cluster HBV infection in an epidemiological survey in 2002. Sixty-eight individuals were sampled in June 2006 and December 2007 for the purpose of assigning HBV genotypes to chronically infected individuals and finding individuals with multiple HBV genotype co-infections. None of the patients received anti-viral therapy or immunosuppressant drugs. A written, informed consent was obtained from each family, and the study protocol was approved by the Southern Medical University Ethics Committee. HBV DNA was extracted from 400 mL of serum by QIAamp UltraSens Virus Kit (Qiagen GmbH, Germany), then resuspended in 50 mL water and stored at 220uC until analysis. HBV genotypes, including C/D recombinant, were initially assigned using the PCR based restriction fragment length polymorphism (RFLP) methods described previously [9] , [8] . For samples with mixed genotype infections, PCR cover HBV S gene (nt136-1110) was performed using the primers and thermocycling conditions descirbed by Sugauchi et al [10] . For samples needing further recombination analysis, PCR was performed using the primers and thermocycling conditions described by Günther to obtain full-length HBV genome [11] . Alternatively, a nested PCR was used to produce two overlapping fragments in subjects with low HBV DNA levels as described by Sugauchi et al [12] . The spanning of fragment A cover nucleotides 2813 to 1824, and fragment B included nucleotides 1821 to 237. LA-Taq (TAKARA, Japan) and high-fidelity polymerase COD-FX (TOYOBO, Japan) were used to produce amplimers for cloning and direct sequencing respectively. Finally, Fragment C (HBV nt56-nt1824) was obtained from a PCR amplification of Y2 HBV-DNA to which an aliquot of genotype B HBV-DNA had been added. The purpose of this experiment with in-tube control of genotype B was to determine if the recombinant clones were being generated during the PCR amplification. PCR products were gel-purified and cloned into the PMD19-T vector (TAKARA, Japan) according to the manufacturer's instructions, and used to transform JM109 competent cells (TAKARA, Japan). A minimum of 15 clones were sequenced from subjects with a mixed-strain infection and three clones were sequenced from family members with a single-strain infection. All sequencing of clones and PCR products was performed by Invitrogen Ltd. (Shanghai, China). Genotypes of clones were determined by phylogenetic tree analysis and recombination analysis. The sequences were assembled using SeqMan II software (DNAStar Inc.). Sequence alignments were performed using ClustalW and confirmed by visual inspection. Phylogenetic trees were constructed by the neighbour-joining (NJ) method (Saitou & Nei, 1987) . To confirm the reliability of the phylogenetic tree analysis, bootstrap resampling and reconstruction were carried out 1000 times. A phylogenetic tree analysis of HBV strains isolated from the mixed infection family was compared with reference strains from GenBank. Accession numbers are indicated on the tree. Bootstrap values are shown along each main branch. The lengths of the horizontal bars indicate the number of nucleotide substitutions per site. The regions included in the analysis were the same with fragment A, B and C or a little shorter. Phylogenetic and molecular evolutionary analyses were conducted using MEGA version 5 (Tamura, Peterson, Stecher, Nei, and Kumar 2011). Recombination signals were initially detected by RDP3.b.4 software [13, 14] . Bootscan, Geneconv and Siscan were used. The highest acceptable P-value was 0.05. Bootscan and Siscan window sizes were 300 bp, step size was 30, replicates for 100 times. A genotype F sequence (GenBank accession numbers is X75658 and X75663) was used as external reference. The precise map of recombination was determined by split phylogenetic tree and alignment. Split phylogenetic trees were constructed by the method same as above. In alignment, each clone was compared to reference C2, D1 and CD1 consensus sequences. We then inspected the alignments to determine the identical crossover sequences around the breakpoint within which the recombination occurred. GenBank accession number of reference sequences of HBV genotype C2, D1, CD1 and CD2 are indicated in phylogenetic tree. Accession Numbers of Y2 clones are JX036326-JX036359. Different HBV genotypes were found in three families among 15 families. The flow of participants in the study and family trees of families with mixed genotypes/subgenotypes of HBV infection are shown (Figure 1 ). Family V had infected members across two generations and two genotypes: In 2006, the mother (V1W) and daughter (V2F) were infected with subgenotype D1 while the son (V2M) had a CD1 recombinant. In 2007, the daughter (V2F) had subgenotype D1 while other family members had HBV DNA levels below the detection limit of the nested PCR assay. Family Q had infected members across three generations and two genotypes/subgenotypes. In 2006, the grandmother (Q1W) and grandson (Q3M) were infected with CD1 recombinant while father (Q2) and granddaughter (Q3F) had mixed infections of genotype C2 and CD1 recombinants. In 2007, the same genotypes were detected in all family members except that the granddaughter (Q3F) had an HBV DNA level below the detection limit of the nested PCR assay. Family Y had affected members across three generations and three genotypes/subgenotypes. In 2006, the grandfather of family Y (Y1) was infected with genotype C2 while grandmother (Y1W) had mixed infections of CD1 and C2. Mother (Y2W) and granddaughter (Y3F) were infected with the CD1 recombinant. Father (Y2) had triplicate infections of genotype C2, D1 and CD recombinant. Grandson's (Y3M) serum was unavailable. In 2007, the grandfather (Y1) and mother (Y2W) had HBV DNA levels below the detection limit while the grandmother (Y1W) and granddaughter (Y3F) had genotype CD1. Father (Y2) and grandson (Y3M) had genotype D1. A phylogenetic tree constructed from HBV nt 36-1110 from the clones of family Y is given (Figure 2A ). The clones (dotted) of family Y exhibits three clusters on genotype C2, D1 and CD1. The phylogenetic tree construct from HBV nt136-1110 from the clones of families Q and V is given ( Figure 2B ). The clones of family Q (indicated by black dots) exhibit two clusters of subgenotypes C2, and CD1. The clones (indicated by black triangles) from family V exhibit two clusters of subgenotypes D1 and CD1. A phylogenetic tree constructed from HBV nt 36-1110 of novel recombinants clones of Y2 is given in Figure 2C . The dotted clones are from Y2. The topology of phylogenetic tree with recombinants is totally different from typical trees. Recombinant sequences blurred the typical branch,in other words, blurred the typical genotype. Results of recombination analysis of Y2 clones are as bellow: Three kinds of analytical methods certificated the same recombination map. The initial pictures of the three methods were all provided as supplemental figures. Recombination events detected by RDP software are shown in Figure S1 , S2, and S3. Split phylogenetic trees constructed by MEGA software are shown in Figure S4 , S5, and S6, (clone number and fragment used to construct tree are indicated beside each tree). Sequence alignments are shown in Figure S7 , S8, and S9. The region where recombination breakpoints had the highest probabilities was recognized as crossover region, which is a region that one parental genotype switches to another. Upstream sequence of crossover region will have specific mutation of one genotype but with no specific mutation of another, downstream just opposite. At the same time, these two genotypes should share same sequence at crossover region. We indicated the crossover region in direct alignment by black bars in Figure S3 initially and marked it in recombination map by colorful bars in Figure 3A and black bars in Figure 3B . The clonal sequences of 2006 showed 17 unique crossover regions in fragments A, B and C. We could not identify any common motif within these sequences that might suggest a common mechanism for crossovers in the HBV. The size of switch region share the same sequence are different in different strains, from 6-174 bp (6 bp for Y2M-2 clone in Figure S7 and 174 bp for Y2M-29 clone in Figure S8 ). To illustrate the recombination map in a simple way. An abbreviated alignment of fragment A, B and C are shown in Figure 3B . Green and pink bars indicated the genotype C2 and D1 respectively. Black bars showed the crossover region. The aligned sequences provide a snapshot of the recombinant HBV strains. Genotype C2, D1 and CD1 recombinant clones of Y2 were all used as parental sequences to recombine with each other to form new recombinants. A series of novel recombinants were found in three fragments. In 15 clones of fragment A, there were five genotype C (Y2-6,9,13,14,15,); two genotype D (Y2-11,12); one CD1 (Y2-10) and seven novel different C/D recombination (Y2- 1, 2, 4, 7, 8, 3, 5) . In Of the 56 clones of fragment C(in which genotype B HBVDNA were added as an in-tube control to exclude the recombination by PCR procedure), there were 32 pure genotype B clones; nine genotype C clones(Y2-B10,B5,B8,B9,B13,B16.B17.B18,B24); five genotype D clones(Y2-B22,B3,B4,B21,B23), two CD1 clones (Y2-B1,B11) and eight novel C/D recombinants (Y2-B6,B7,B14,B15,B19,B2,B12,B20). No recombinants of genotype B were found. Recombination is one of the major mechanisms contributing to the evolution of retroviruses [15] . Since the HBV has a reverse transcription step in its life cycle, it is conceivable that recombination also contributes to diversity in HBV genomes. Although just four cases were observed with mixed genotype infections, we obtained a snapshot of naturally occurring HBV recombinants generated in the absence of selection and after selection. Our result showed direct evidence of HBV recombination, with new information of recombining crossovers compared with similar studies [16, 17, 18, 19] . The recombination analysis of Y2 quasi-species showed variable types of recombinant between genotype C2, D1 and CD1 in 2006. Some studies show that hotspots of recombination most on the boundary of ORFs [12, 20] . Our results showed that two or more strains of HBV can recombine with each other at any region along the genome. Crossover regions can be hundreds or just several base pairs, The length of crossover region is depends on the location of it on HBV genome. If it is located in a conserved HBV region, for another word, where many different genotypes share the same sequence, the length of crossover region may be long. If it is located in a non-conserved region, it may be very short. At the same time, we found that the crossover region distributed totally at random on HBV genome. Consistent with our results, in vitro evidence showed the initial recombination events in a laboratory system of MHV were almost entirely randomly distributed along the sequence [21] . It was only after passage through cell culture, with the opportunity for selection to remove less fit variants, that crossover sites became ''localized'' to just a small area of the region examined. Crucially, they also suggested initial products of recombination may go undetected because of the action of strong purifying selection which will remove new, deleterious combinations of mutations. The conclusion is therefore an interpretation for the genotype change of Y2. The Y2 presented multiple strain infections of C2/D1/CD1 and many new recombinants with no obvious dominant genotype strain in 2006. After 18 months, however, all the type C2 and CD recombinant strains disappeared while the D strain became dominant. A similar case of mixed HBV genotype infection in which one genotype was lost and another prevailed was previously described in patients with HBeAg seroconversion [4, 22] . Epidemiologically, HBV genotype CD1 and C2 are the most common strains in ethnic minorities of northwest China with CD2 and D1 as minor strains. Precise mapping of recombination suggests C2 and D1 are parental sequences of CD1 and CD2 recombinants. Virological differences among HBV genotypes were demonstrated in vitroand in CHiM mice, with genotype C having a higher replication capacity than D [23] . Why does the replication-deficient genotype D virus predominate over replication-competent genotype C? As mixed HBV infections together with recombination are rare, we have little knowledge about i this situation. On the one hand, we know little about host impact on different genotypes and recombinants. On the other hand, we know little about interference and competition in the quasispecies of mixed infection. In vitro results showed the replication capacity of individual clone, exclude the influence of host and other strains of quasi-species. An example from a ChiM mice study showed that monoinfection of HBV/G in ChiM mice display a very slow replication while coinfection with HBV/A remarkably enhanced the replication of HBV/G. The replication of HBV/G is heavily dependent on coinfection with other genotypes. When HBV/G superinfected on other genotypes, a rapidly takes over of HBV/G from original genotype were observed, though they are indispensable [24] . This study confirms that in a mixed infection system of different genotypes, the replication capacity of a genotype may be different from that of monoinfection. At the same time, replication capacity is not the only factor to influence which strain will become dominant. Variable recombinants found in our study may be mechanistically capable of genetic exchange, but strong selection guaranteed the elimination of hybrid genomes. The mechanism of selection in mixed infection also needs more investigation. We found mixed HBV genotypes infection with many novel recombinants at one point in time, but just one genotype was found 18 months later. This may indicate that the detectable mixed infection and recombination has a limited time window due to the sensitivity of detection or strong selection power of the host. That's why in most studies, we can identify a major genotype in one patient. Even so, evolutionarily visible and invisible recombination of HBV could occur and play an important role by generating genetic variation or reducing mutational load. However, this study had limitation, because recombination signals were detected by RDP3 software and confirmed by split phylogenetic tree and alignment, indicating the recombinant or recombinantlike form should depend on the software. If we use another software, the results might be different. Studies of HBV in endemic areas throughout the world have resulted in large numbers of full genome sequences available for phylogenetic analysis enabling the identification of novel, mosaic HBV genomes that appear to be the result of recombination between previously known sequences [7, 25, 26] . One of the most comprehensive analyses of putative HBV inter-genotype recombinants showed the existence of 24 phylogenetically independent HBV genomes involving all known human genotypes [27] . Some of these recombinants are unique to individual subjects, but some undergo expansion in specific populations and become recognized as new genotypes or subgenotypes [12, 28, 29, 30] . Four stages in the process of generating popular HBV recombinant genomes should be recognized. The first stage is the co-circulation of different HBV strains or genotypes in the same geographic area. The second is the existence of individuals who have been infected with more than one strain of HBV. The third is the generation of a novel recombinant strain(s) within an individual. The fourth is the selection of a recombined strain with the ability to replicate and be transmitted. Our data show the natural process of the formation and selection of recombination though the recombinant strains of Y2 that appeared in 2006 that were all removed from samples in 2007. By using phylogenetic trees and homology calculations, HBV variants infecting humans are currently classified into ten genotypes that differ from each other in nucleotide sequence by 7.5 to 13% [2, 3] . There are some characteristic length differences between the genotypes that facilitates their detection and discrimination. However, as shown in Figure 2 , existence of a recombinant makes the topology of the phylogenetic tree totally different from one with no recombinant. Recombinant strains obscured the definition of genotypes. Based on the algorithm creating a phylogenetic tree, sequences with high homologues cluster together. With the same logic, recombinants always clustered with the backbone parental sequence, in other words, with which they have high similarity with the larger proportion of the recombination region. Therefore, recombinants always seem to be a subgenotype of their backbone parental sequence. Similar to Y2-8 clone in Figure 2C , for recombinants with similar proportion of both parental genotypes, the sequence shows a divergent trend different from both parental genotypes. Based on phylogenetic topology changes of different regions of HBV, it was hypothesized that some of the genotypes that are conventionally regarded as ''pure,'' actually were recombinant. Genotype E strains show evidence of recombination with genotype D at 1950-2500. new reported genotype ''I'' actually belongs to genotype C. Furthermore, Subgenotype Ba possesses the recombination with genotype C at 1740 to 2485 [31, 32, 33] . Recombinants comprising regions with different histories have important implications for the way we think about HBV evolution. It means that there is no single phylogenetic tree that can describe the evolutionary relationships between genotypes. In conclusion, mixed HBV genotypes infection with many novel recombinants at one point in time ended up with just one genotype 18 months later in this study. This may indicate that the detectable mixed infection and recombination have a limited time window due to the sensitivity of detection or strong selection power of the host. Also, as the recombinant or recombinant-like nature of HBV precludes the possibility of a ''true'' phylogenetic taxonomy, a new standard may be required for classifying HBV sequences.

Identifying Live Bird Markets with the Potential to Act as Reservoirs of Avian Influenza A (H5N1) Virus: A Survey in Northern Viet Nam and Cambodia

Wet markets are common in many parts of the world and may promote the emergence, spread and maintenance of livestock pathogens, including zoonoses. A survey was conducted in order to assess the potential of Vietnamese and Cambodian live bird markets (LBMs) to sustain circulation of highly pathogenic avian influenza virus subtype H5N1 (HPAIV H5N1). Thirty Vietnamese and 8 Cambodian LBMs were visited, and structured interviews were conducted with the market managers and 561 Vietnamese and 84 Cambodian traders. Multivariate and cluster analysis were used to construct a typology of traders based on their poultry management practices. As a result of those practices and large poultry surplus (unsold poultry reoffered for sale the following day), some poultry traders were shown to promote conditions favorable for perpetuating HPAIV H5N1 in LBMs. More than 80% of these traders operated in LBMs located in the most densely populated areas, Ha Noi and Phnom Penh. The profiles of sellers operating at a given LBM could be reliably predicted using basic information about the location and type of market. Consequently, LBMs with the largest combination of risk factors for becoming virus reservoirs could be easily identified, potentially allowing control strategies to be appropriately targeted. These findings are of particular relevance to resource-scarce settings with extensively developed LBM systems, commonly found in South-East Asia.

First detected in 1996 [1] , highly pathogenic avian influenza virus subtype H5N1 (HPAIV H5N1) has spread across 3 continents and is now considered to be endemic in several South-East Asian countries and Egypt. Due to its potential to recombine with human influenza strains to produce highly virulent reassortants [2] , ongoing circulation of HPAIV H5N1 continues to be a major public health concern. Live bird markets (LBMs), in which the virus has been frequently detected in both disease-endemic and epidemic regions [3, 4, 5] , are suspected to play a major role in the epidemiology of HPAIV H5N1 [6, 7] . The LBM system provides consumers with freshly slaughtered birds. It is a dead-end for poultry, but not necessarily for viruses. LBMs have been shown to contribute to the spread of, and the possible maintenance of, HPAIV H5N1 within the poultry sector [8] . Zoonotic transfer to humans has also been documented in LBMs [9, 10] . Birds are introduced into LBMs daily and stocked at a high density. If these birds remain in the market for a sufficient time to become infected and transmit virus to susceptible birds, LBMs would offer optimal conditions for amplifying and sustaining virus circulation and could, thus, become viral reservoirs themselves [11] . Given the range of species present in typical LBMs and the variety of detected influenza viruses [12, 13, 14, 15, 16, 17, 18, 19] , LBMs could potentially act as drivers of viral evolution, promoting the emergence of new variants. Identifying those LBMs that could act as viral reservoirs is, therefore, crucial for improving surveillance and control. Although the potential for an LBM to become a viral reservoir is determined by the management practices of poultry traders [11] , an understanding of these practices in high risk areas is lacking. In response to this need, a cross-sectional survey was conducted in northern Viet Nam and Cambodia to assess whether traders of live poultry engaged in practices that could sustain virus circulation in LBMs. Viet Nam is one of the most severely affected countries in the current HPAI H5N1 pandemic [20] , with the disease considered to be endemic in both northern and southern Viet Nam. Only sporadic outbreaks have been reported in Cambodia [20] . However, the reporting of isolated human cases without prior notification of poultry outbreaks [21] suggests that widespread, undetected virus circulation and persistence of HPAIV H5N1 in the Cambodian poultry population cannot be ruled out. In both countries, LBMs are common and potentially involved in virus spread. An LBM for this study was defined as an open space with 2 or more traders selling live poultry at least once per week and with official government authorization to do so. ''Live poultry'' referred to finished birds, meaning birds intended to be slaughtered and eaten by the end-user. Although LBMs are numerous in Viet Nam and Cambodia, the live poultry trade is only a small, irregular activity in most markets, necessitating a purposive sampling strategy. In the selected areas, only the largest LBMs in terms of the number of poultry sold were eligible. The selection of Cambodian LBMs was based on a previous cross-sectional survey on commercial poultry movements conducted in 2006-2007 [22] . Eight Cambodian LBMs with the highest volumes of poultry sales were recruited. Data on the frequency and volume of live poultry sales in northern Vietnamese LBMs were not available. Study provinces were selected based on demographic features rather than outbreak reports, given that most disease events were presumed to be undetected [23, 24, 25] : Ha Noi, the most densely populated province in northern Viet Nam [26] , and Bac Giang, a rural province with a large poultry population [27] . A snow-ball sampling approach was adopted. A first set of major LBMs were identified in each study area through meetings with trade and veterinary service officers. In these LBMs, traders were asked to identify the other LBMs of which they were aware and to rank them according to the number of live poultry sellers. These identified LBMs were then integrated into the survey and their traders were in turn asked to name the LBMs that they considered to be the biggest. As there are 4 times as many people in Ha Noi as in Bac Giang [26] , the number of LBMs was also expected to be much higher, with traders likely to be aware only of markets located in the immediate vicinity of those from which they trade. For this reason, the snowball approach described above was applied only at the province level in Bac Giang, but also at the lower administrative level, the district level, in Ha Noi. The present day province of Ha Noi is the result of the recent merging of an urban centre (former Ha Noi province) and a rural area (former Ha Tay province and a district from Vinh Phuc province). After exclusion of districts where live poultry marketing was prohibited, 4 of 5 districts in the urban centre and 2 of 13 districts in the rural area were randomly selected. Meetings held with Ha Noi veterinary services also identified the 4 main LBMs supplying the province, which were then integrated into the survey. In total, 30 markets were recruited in northern Viet Nam. Assuming that each commune (administrative division of district) had 1 or 2 LBMs, the LBM sampling rate per district ranged between 3% and 25%. Study areas are shown in Fig. 1 . To participate in the survey, traders in the markets had to sell or purchase birds in an LBM at least 1 day per month. Live poultry traders consisted of sellers and middlemen. A seller was defined as selling live poultry mostly to the end-user (e.g. consumers, restaurants) or to another trader who would then sell the poultry at another location. A middleman was defined as selling live poultry to mostly sellers, or purchasing live poultry from sellers and then re-selling them at another location (e.g. market, restaurants). All traders that were present during the visit of the selected LBMs were interviewed, except in the largest LBM in Viet Nam where half were interviewed. Of note, in all other LBMs, the numbers of interviewed sellers corresponded well to numbers indicated by market managers. The study periods, April-May 2009 in Viet Nam and June-July 2009 in Cambodia, did not include any seasonal festivals which could potentially influence live poultry sale patterns. Three standardized questionnaires were designed for interviewing market managers, sellers, and middlemen. The questionnaires were translated into Vietnamese and Khmer and administered by trained interviewers. All questionnaires were piloted in Vietnamese LBMs that were not included in the survey. Informed oral consent was sought prior to interviewing. At the conclusion of each interview, the completed questionnaire was reviewed by the author for missing, unclear, or inconsistent answers. Questions for which the accuracy of the answer was doubtful were posed a second time. Market observations, such as counting the number of birds offered for sale and the number of birds left unsold at the end of the market day, were made during the visits. The interviewees' answers were consistent with these observations. LBMs were described based on demographic features provided by market managers: days of the month and time of the day during which LBMs were open, average number of sellers, and seasonal variations. Sellers were classified as retailers or wholesalers if they reported selling most birds to consumers or traders, respectively. Markets were classified as either retail or wholesale markets if more than two thirds of sellers were retailers or wholesalers. Markets that did not fall into one of these categories were classified as mixed. Traders' practices likely to influence the sustainability of virus circulation in LBMs were recorded; essentially, those that could impact the length of time during which birds remained in the market chain and the contact rate between birds. These factors included the number of days during which traders were active, and the length of time they spent at market in a day. The number of poultry sold within a day, and the type of poultry were also considered. Indeed, the susceptibility of chickens and ducks (muscovy or mallard duck derived breeds) to infection is known to differ [28] . Moreover, the supply management, and frequency and quantity of the surplus (unsold poultry reoffered for sale the following day) could impact on the length of time that poultry spent at market. The surplus frequency referred to the proportion of days traders reported having had surplus and the surplus volume to the proportion of birds left unsold when having a surplus. The surplus frequency was captured in 2 ways: the usual surplus frequency, recorded as a categorical variable (categories: never, sometimes, half of the time, often, always), and the surplus frequency in the last week, defined as the proportion of days with a surplus out of the number of trading days during the week preceding the interview. Since both variables were highly correlated (correlation ratio = 0.84), only the surplus frequency in the past week was kept in the analysis. The management of the supply described the frequency at which poultry were purchased, and whether poultry were purchased the day before being offered for sale and kept overnight at traders' homes. Changes in trade practices during festivals were also described. Information regarding the origin of poultry and the number and type of visited farms and LBMs was also collected, as these contacts could influence the likelihood of spreading infection into and out of LBMs. Questionnaire data were entered in Microsoft Access 2007H (Microsoft Corp., Redmond, WA, USA) database. The accuracy of the data entry was verified by cross-checking each questionnaire with the recorded entry. Numerical variables were summarized as medians with interquartile ranges (IQR); binary and categorical variables as frequencies and percentages. Multivariate analysis was performed to describe trader profiles, which were based on poultry management practices that could increase the risk of sustained virus circulation in LBMs ( Table 1) . As variables were both numerical and categorical, factor analysis for mixed data (FAMD) [29, 30] was used. This method allows a reduction in the dimensions of multivariate data, creating a smaller number of synthetic (and uncorrelated) factors accounting for most data variability. Further details are provided in Text S1. Hierarchical cluster analysis (HCA) [31] was then used to group traders into clusters according to their level of similarity in the factors created by the FAMD. The Manhattan distance was used to assess the level of dissimilarity between 2 traders. The algorithm was agglomerative, and the Ward's criteria for linkage was adopted. Finally, a consolidation using the k-means algorithm was performed. FAMD and HCA were implemented in R 2.12.0 [32] , using the package FactoMineR 1.15 [33] . There were 340 sellers and 221 middlemen interviewed in 30 Vietnamese LBMs, and 54 sellers and 30 middlemen in 8 Cambodian LBMs. The refusal rate was 8% among Vietnamese traders, with the principal reason being that they were too busy to participate. In Cambodia, only 2 (2%) people declined interviews. LBM features are described in Table 2 . LBMs in Bac Giang province in Viet Nam were either open every day (n = 4) or periodically (n = 5). Periodic markets were open 6 or 12 days per month. LBMs in Ha Noi province were grouped as either retail and mixed markets (n = 17), hosting from 2 to 13 sellers, or wholesale markets (n = 4), which were the largest markets, hosting from 30 to 80 sellers. Cambodian LBMs were grouped as either urban markets (n = 4), located in Phnom Penh or its close proximity, or peri-urban markets (n = 4), in other provinces. Urban markets were open throughout the day for 12-15 hours, whereas peri-urban markets were only open in the morning for 6 hours or less. Most Vietnamese (80%, n = 340) and Cambodian (70%, n = 54) sellers reported having a surplus, at least occasionally. In contrast, most middlemen reported never having a surplus (Viet Nam: 76%, n = 221; Cambodia: 97%, n = 30). As they generally spent less than 1 hour in a LBM, middlemen kept their poultry in LBMs for only a very short time and were therefore excluded from the multivariate analysis. Following the FAMD and the HCA, Vietnamese sellers were divided into 4 clusters and Cambodian sellers into 2 clusters. Tables 3 and 4 present the distribution of poultry management and contact features for each seller profile in Viet Nam and Cambodia, respectively. A description of the factors is provided in Text S1. Vietnamese sellers in Cluster V.1 were farmers or occasional sellers. Most farmers' flocks consisted of 50-500 birds and were located in the market vicinity. They were characterized by an infrequent and short presence at market and a very low number of sales. Encompassing 48% (n = 340) of Vietnamese sellers, Cluster V.2 was composed of sellers trading larger volumes and more frequently than Cluster V.1 sellers. However, they spent little time in LBMs, with 60% (n = 162) only trading 4 hours or less per day. Their surpluses were also low, with the median frequency of having a surplus being 14% and the median proportion of unsold chickens and ducks being 10% and 6%, respectively. Cluster V.3 was also composed of regular sellers. Although their number of sales was slightly lower than Cluster V.2 sellers, they reported higher surpluses than traders in other clusters. The proportion of traders reporting a surplus every day was 51% (n = 71), and the median proportion of unsold chickens and ducks was 32% and 17%, respectively. Moreover, 45% (n = 71) of Cluster V.3 sellers were not supplied every day and 66% (n = 71) purchased birds the day before offering them for sale, keeping them overnight at home. These proportions were higher than in other clusters. For traders who were not supplied every day when operating at a market, part of the newly purchased birds were stored at home for 1 to several days before bringing them to market. Cluster V.4 included sellers spending more time at market, with 64% (n = 64) trading at least 10 hours per day, and selling substantially more poultry than other clusters. However, the median surplus frequency of 29% and the median proportion of unsold birds, less than 10% for chickens and ducks, were much lower than those reported by Cluster V.3 sellers. Except for Cluster V.1, most Vietnamese sellers were supplied by farms, the majority of which were small commercial farms (50-500 birds). Cluster V.2 and V.3 sellers were also supplied by backyard farms (,50 birds), whilst Cluster V.4 sellers were also supplied by large farms (.500 birds). The number of sellers visiting several markets to purchase or sell poultry was higher in Cluster V.2 (48%, n = 162) than in Clusters V.3 (38%, n = 71) and V.4 (3%, n = 64). Whilst the proportion of Cluster V.2 sellers was high in all market groups, 74% (n = 43) Cluster V.1 sellers were found in Bac Giang markets, 80% (n = 71) Cluster V.3 sellers were in Ha Noi retail and mixed markets and 95% (n = 64) Cluster V.4 sellers were in Ha Noi wholesale markets. All Bac Giang markets were strictly or predominantly populated by Cluster V.1 or V.2 sellers, or both (Fig. 2) . Cluster V.3 was the only, or the predominant, seller profile in 13 of the 17 Ha Noi retail and mixed markets. This seller profile was, however, absent in 2 markets located in peri-urban areas, far from the main urban centres. Contrary to other market groups, Ha Noi wholesale markets were highly heterogeneous in terms of seller composition. However, when considering the proportion of poultry traded by each seller profile in each market (Fig. S1 ), largescale sellers (Cluster V.4) were predominant in all Ha Noi wholesale markets but 1. The market location was, therefore, a good predictor of the seller composition. Cambodian sellers were classified into 2 clusters. Cluster C.1 sellers spent little time in LBMs and either rarely, or never, had a surplus. In contrast, most Cluster C.2 sellers spent all day in LBMs and reported high surpluses. The median length of time spent in LBMs was 4 hours for Cluster C.1, and 11 hours for Cluster C.2 sellers. Most Cluster C.1 sellers (58%, n = 26) never had any unsold poultry at the end of the market day, whereas all Cluster C.2 sellers but 1 reported surpluses. Whilst most Cluster C.1 sellers were supplied by farmers, most Cluster C.2 sellers were supplied by traders. Contrary to Viet Nam, all supplying farms were backyard farms (,50 birds), and none of the Cambodian sellers visited other markets to buy or sell poultry. Cambodian seller profiles were also associated with market groups, with 85% (n = 26) Cluster C.1 sellers operating in periurban markets and 86% (n = 28) Cluster C.2 sellers operating in urban markets. Three of 4 peri-urban markets were exclusively populated by Cluster C.1 sellers, and Cluster C.2 predominated in 3 of 4 urban markets. Chicken sales peaked in Viet Nam and Cambodia during the Tet and the Chinese New Year (late January or early February), respectively, with 72% (n = 340) and 96% (n = 54) sellers reporting an increase in chicken sales by 100% on average. Likewise, the number of sellers operating at markets increased. Sellers did not report any other changes in their practices during these periods. Despite considerable variation in poultry management practices between sellers, some patterns were evident such that seller profiles of epidemiological importance could be identified. The high surplus frequency and volume reported by Clusters V.3 sellers (medium-scale sellers with high surplus) increased the time spent by birds in the LBM system. Moreover, their low supply frequency and the practice of purchasing birds the day before offering them for sale extended the time spent by birds in the sellers' flocks. This created opportunities for newly purchased birds to mix with unsold birds brought back from LBMs. These practices would make infection of susceptible birds more likely. This seller profile would thus be at high risk for contributing to the maintenance of HPAIV H5N1 in LBMs. In contrast, surplus and supply features exhibited by Cluster V.2 (medium-scale sellers with low surplus) and V.4 sellers (large-scale sellers) meant that their birds spent little time in LBMs, limiting their potential to sustain virus circulation. However, these seller profiles may still play an active role in virus spread. A substantial proportion of Cluster V.2 sellers were mobile, visiting several markets to sell and/or purchase poultry. This could enable them to spread infection between LBMs. If infectious birds are Although these newly infected birds are unlikely to remain in LBMs for a sufficient period of time to infect others, their sale to traders operating in other LBMs could lead to virus spread between LBMs. Indeed, most large-scale sellers (V.4) operated in Ha Noi wholesale markets, from which a substantial part of the poultry population was then sold to retail market sellers (data not shown). In contrast, Cluster V.1 sellers (farmers and irregular sellers) were unlikely to play a major role in spatial virus spread as they only traded in markets located in their vicinity. The distribution of seller profiles was associated with market groups. Therefore, knowing the type and location of a given LBM provides a good indicator of its seller profile composition, the proportion of poultry sold by each seller profile and, thus, the market's risk of sustaining HPAIV H5N1 circulation. LBMs located in the rural province of Bac Giang probably play a limited role in virus perpetuation. By contrast, most Ha Noi retail and mixed markets were dominated by Cluster V.3 sellers, which could permit virus maintenance. However, the low number of sellers and the low volume of sales might increase the risk of stochastic extinction of viruses, even with frequent surpluses. Two LBMs of this group were predominantly or exclusively populated by Cluster V.2 sellers and were located in districts which shared key features with Bac Giang province: rural areas with a large number of poultry farms. In contrast, other Ha Noi retail and mixed markets were located in, or close to, urban centres. This urban tropism of sellers at higher risk of maintaining virus circulation was also observed in Cambodia. The markets located in Phnom Penh or its outskirts were almost exclusively populated by sellers with high surpluses (C.2), similar to Vietnamese Cluster V.3 sellers, whilst Cambodian Cluster C.1 sellers, unlikely to allow virus maintenance, operated in provinces other than Phnom Penh. Basic information on market type and location could be easily collected from each LBM to aid the identification of markets at high risk of virus maintenance, where risk mitigation strategies should be implemented. This information could be directly collected from market managers and would not require a laborintensive survey. The implementation of strategies aiming at breaking virus amplification cycles in all markets is unnecessary, and also impractical given that LBMs are ubiquitous. Targeting control measures to a few selected markets would not only reduce the overall cost, but would also allow closer monitoring and proper implementation. Simple hygiene measures and culling of unsold birds may be very effective in breaking the virus amplification cycle [11, 34, 35] . Successfully implemented in Hong Kong, such measures would, however, need to be adapted to each local setting in order to minimize their negative impact on trade. HPAIV H5N1 has been isolated in northern Vietnamese LBMs [36, 37] , and poultry trade is suspected to spread the infection in Cambodia [38] . HPAIV H5N1 is likely to circulate in the study population. However, only the potential of LBMs to become virus reservoirs has been assessed in this survey. Complementing the findings with virological sampling of markets would be necessary to conclude that these high risk LBMs are truly virus reservoirs. Moreover, since the sampling frame is not representative of the population as a whole but of the largest markets in specific regions, inferences about the study population are necessarily limited. When asked to name the most populated markets, traders were more likely to name markets where they regularly operated, or of which they had personal knowledge, for example by being in their vicinity. However, as traders interviewed in Bac Giang province often named the same markets, regardless of the district where the interview took place, it is likely that most of the largest markets were identified. Three markets located in Bac Giang city were among the most commonly named markets, but authorization to visit was not possible as poultry sales at these sites had been officially prohibited. In Ha Noi province, some large retail and mixed markets may have been missed, as only a few districts were visited. However, the retail and mixed markets identified in Ha Noi were likely to be similar to those not identified, given similar population densities and housing. In conclusion, this study was able to identify specific profiles of live bird sellers in Viet Nam and Cambodia which could play a key role in virus perpetuation. Moreover, the type and the location of an LBM could be a good predictor of its seller profile composition and, thus, of its potential for sustaining virus circulation. Therefore, results suggest that control strategies aiming at preventing HPAIV H5N1 maintenance in LBMs could potentially be targeted towards specific high risk LBM groups. This is of particular importance in resource-scarce countries with extensively developed LBM systems. Figure S1 Distribution of the number of poultry traded by seller clusters across markets and market groups. For each market group, a barplot (on the left) shows the proportion of the poultry flow (number of poultry sold) traded by each seller cluster in the market group, and a plot (on the right) shows the distribution of its markets according to the proportion of the poultry flow traded by each seller cluster in each market. Where the number of markets in a group is greater than 5, box plots are shown; otherwise each market (circle) and the median (line) are presented. Text S1 Supplementary information includes further details on the approach used to construct a typology of traders, and on the results of the multivariate analysis and hierarchical cluster analysis. (DOC)

Porcine Noroviruses Related to Human Noroviruses

Detection of genogroup II (GII) norovirus (NoV) RNA from adult pigs in Japan and Europe and GII NoV antibodies in US swine raises public health concerns about zoonotic transmission of porcine NoVs to humans, although no NoVs have been detected in US swine. To detect porcine NoVs and to investigate their genetic diversity and relatedness to human NoVs, 275 fecal samples from normal US adult swine were screened by reverse transcription–polymerase chain reaction with calicivirus universal primers. Six samples were positive for NoV. Based on sequence analysis of 3 kb on the 3´ end of 5 porcine NoVs, 3 genotypes in GII and a potential recombinant were identified. One genotype of porcine NoVs was genetically and antigenically related to human NoVs and replicated in gnotobiotic pigs. These results raise concerns of whether subclinically infected adult swine may be reservoirs of new human NoVs or if porcine/human GII recombinants could emerge.

Detection of genogroup II (GII) norovirus (NoV) RNA from adult pigs in Japan and Europe and GII NoV antibodies in US swine raises public health concerns about zoonotic transmission of porcine NoVs to humans, although no NoVs have been detected in US swine. To detect porcine NoVs and to investigate their genetic diversity and relatedness to human NoVs, 275 fecal samples from normal US adult swine were screened by reverse transcription-polymerase chain reaction with calicivirus universal primers. Six samples were positive for NoV. Based on sequence analysis of 3 kb on the 3′ end of 5 porcine NoVs, 3 genotypes in GII and a potential recombinant were identified. One genotype of porcine NoVs was genetically and antigenically related to human NoVs and replicated in gnotobiotic pigs. These results raise concerns of whether subclinically infected adult swine may be reservoirs of new human NoVs or if porcine/human GII recombinants could emerge. Norovirus) cause diarrhea in humans and animals (1) (2) (3) . The NoV genome is 7.3-7.7 kb long with 3 open reading frames (ORFs) encoding a polyprotein that undergoes protease processing to produce several nonstructural proteins, including an RNA-dependent RNA polymerase (RdRp), a major capsid protein (VP1, capsid), and a minor capsid protein (VP2) (1, 4, 5) . The capsid protein contains a conserved shell (S) and hypervariable protruding (P) domains (6) . Noroviruses are genetically diverse and make up 27 genotypes within 5 genogroups, GI/1-8, GII/1-17, GIII/1-2, GIV, and GV, based on the capsid genes of 164 strains (7) . Human NoVs cause an estimated 23 million cases of illness annually in the United States (8) and >90% of nonbacterial epidemic gastroenteritis worldwide (1) . The low infectious dose, environmental resistance, strain diversity, shedding from asymptomatic persons, and varied transmission vehicles render human NoVs highly contagious. Norovirus RNA was detected by reverse transcription-polymerase chain reaction (RT-PCR) in 4 of 1,017 normal slaughtered pigs in Japan (9) and in 2 of 100 pooled pig fecal samples in the Netherlands (10) . These porcine NoVs (Sw43/97/JP, Sw918/97/JP, and 34/98/NET) are genetically similar and are classified into GII (9,10), like most epidemic human NoVs (11) (12) (13) . Also, the viruslike particles (VLPs) of Sw918 strain cross-react with antibodies against human GII but not GI NoVs (14) . The close genetic and antigenic relationships between human and porcine NoVs raise public health concerns regarding their potential for zoonotic transmission and as reservoirs for emergence of new epidemic human strains. Farkas et al. (14) reported that US swine sera react with Po/NoV/GII/Sw918 strain, but no direct detection of NoV from US swine has been reported. To detect porcine NoVs and assess their genetic diversity and relatedness to human NoVs, we screened 275 pig fecal samples from US swine by RT-PCR with a calicivirus universal primer pair p290/110 targeting the RdRp region (15, 16) , followed by sequencing the 3 kb on the 3′ end of the genome for 5 NoV strains. Gnotobiotic pigs were inoculated with porcine NoVs to examine their infectivity and to produce convalescent-phase antiserum for antigenic analysis. Fecal RT-PCR was performed separately by using primer pair p290 (5′-GATTACTCCAAGTGGGACTCCAC-3′) (15) and p110 (5′-ACDATYTCATCATCACCATA-3′) (16) as previously described (15) but at 48°C for annealing (317 bp for NoV or 329 bp for sapovirus). To amplify the 3-kb 3′ end fragment, cDNA was synthesized by SuperScript III First-Strand cDNA synthesis kit (Invitrogen) with primer VN 3 T 20 (5′-GAGTGACCGCGGCCGCT 20 -3′). PCR was then performed with TaKaRa Ex Taq polymerase (TaKaRa Mirus Bio, Madison, WI, USA) with primers p290 and VN 3 T 20 . Quantitative (endpoint titration) RT-PCR (17) was performed with primer pair PNV7 (5′-AGGTGGTGGCC-GAGGAYCTCCT-3′) and PNV8 (5′-TCACCATAGAAG-GARAAGCA-3′) targeting the RdRp (211 bp) of QW101 strain. RT-PCR products were purified with the QIAquick Gel Extraction kit (Qiagen) before cloning into pCR2.1-TOPO (T/A) or PCR XL cloning kit (Invitrogen). Five clones of each sample were sequenced. DNA sequencing was performed with BigDye Terminator Cycle and 3730 DNA Analyzer (Applied Biosystems, Foster City, CA, USA). Sequence editing was performed by Lasergene software package (v5, DNASTAR Inc., Madison, WI, USA). The Basic Local Alignment Search Tool (BLAST, http://www.ncbi.nlm.nih.gov/BLAST) was used to find homologous hits. Multiple sequence alignment was performed with ClustalW (v1.83) at DNA Data Bank of Japan (http://www.ddbj.nig.ac.jp). Phylogenetic and bootstrap (1,000 replicates) analyses were conducted by using MEGA (v2.1) (18) . Identification of recombinants was performed by using the Recombinant Identification Program (RIP, http://hivweb.lanl.gov/RIP/RIPsubmit.html) (19) . The classification and GenBank accession numbers of NoVs are listed in Table 1 . Four gnotobiotic pigs were maintained and euthanized as previously described (25, 26) . The inoculate was a 20% fecal filtrate (0.2 µm) in EMEM of the QW126 or QW144 (QW101-like, GII-18) strains or EMEM only (2 negative control pigs). One pig was inoculated with QW126 orally and intranasally at 9 days of age, and convalescent-phase antiserum LL616 was collected at postinoculation day (PID) 26. A second pig was inoculated with QW144 orally at 35 days of age and euthanized at PID 5. Immune electron microscopy (IEM) was performed as described previously (27) . For enzyme-linked immunosorbent assay (ELISA), the recombinant baculovirusexpressed human NoV VLPs and rotavirus VP2 and VP6 (2/6)-VLPs (negative control) (28) were CsCl-gradients purified. We coated 96-well microplates with VLPs (200 ng/well) in carbonate buffer (pH 9.6) and blocked with 5% nonfat dry milk in phosphate-buffered saline (PBS)-Tween 20 (0.05%). Serially diluted serum samples that included positive and negative controls were added to duplicate positive-and negative-coated wells, and the plates were incubated. After washing, horseradish peroxidase (HRP)labeled goat anti-pig immunoglobulin G (IgG) (H + L) for pig sera or goat anti-human IgG + IgA + IgM (H + L) (KPL, Gaithersburg, MD, USA) for human serum was added. After incubation and washing, the substrate 3,3′,5,5′-tetramethylbenzidine was added. The cutoff value was the mean absorbance of the negative coatings multiplied by 2. Western blot was performed as described previously (29) . Nitrocellulose membranes were incubated with pig convalescent-phase antiserum LL616 against porcine GII-18 NoV or negative control serum in PBS containing 4% nonfat dry milk followed by goat anti-pig IgG (H + L)-HRP conjugate. Porcine NoVs were classified into 3 genotypes within GII based on the complete capsid sequences: 1 genotype with prototype Japanese strains Sw43 and Sw918 and 2 new genotypes. A total of 19 of 275 samples showed a potential positive band after agarose gel electrophoresis of the RT-PCR products of primer pair p290/110. Fourteen samples representative of each potentially positive farm or the slaughterhouse were sequenced. After performing BLAST search, we identified 6 NoVs (QW48, Michigan farm A; QW101, QW125, and QW126, Ohio farm B; and QW170 and QW218, Ohio slaughterhouse), 3 sapoviruses, and 5 sequences that had no significant hit in the database. Because the QW126 shared 99% nucleotide (nt) identity with the QW101 and QW125 strains in the 274-nt RdRp region, it was not sequenced further. We sequenced the 3-kb 3′ end of the genome containing the partial RdRp, VP1 and VP2 genes, and the 3′ untranslated region of the 5 strains. The porcine NoVs represented 3 distinct clusters: 1) Sw43, Sw918, and QW48; 2) QW101 and QW125; and 3) QW170 and QW218, on the basis of the size of each gene and the ORF1-ORF2 overlap region (Table 2 ). Across the 3 kb, the QW101 and QW125 strains and the QW170 and QW218 strains shared 99% nt identity. The amino acid identity of the predicted complete and S and P domains of the capsid protein of the 5 porcine NoVs, the previously reported porcine NoVs (Sw43 and Sw918), and representative human, bovine, and murine NoV strains is summarized in Table 3 . In the complete capsid, the QW48 strain was most closely related to the porcine NoV prototype Sw43 strain (98% amino acid identity); the QW170 and QW218 strains shared the highest amino acid identities (81%) to porcine Sw43 and Sw918 strains; the QW101 and QW125 strains showed the highest amino acid identity to human GII-3/Mexico (71.4%), then to human GII-6/Baltimore (71.0%), porcine QW218 (71.0%), and porcine Sw43 (70.6%) strains. The S and P domains of these NoVs showed similar relationships. A neighbor-joining phylogenetic tree based on the amino acid sequences of the complete capsids ( Figure 1) showed that QW48 grouped with Sw43 and Sw918 strains into GII-11 and that QW170 and QW218 formed a new genotype (GII-19), which was closer to porcine than to human strains. However, QW101 and 125 formed a new genotype (GII-18) between human and porcine GII NoVs. Further analysis of the predicted C-terminal ≈260 amino acids of the RdRp region ( Figure 2 ) showed similar grouping results for QW48, QW101, and QW125 strains but different for QW170 and QW218 strains, which were in the same cluster (GII-11) as Sw43, Sw918, and QW48 in the RdRp region. This finding suggested that a recombination event occurred between QW170/218-like and Sw43-like NoVs. The complete VP2 sequences of representative strains were also analyzed (data not shown). Results were similar to those of the capsid sequence classification. A potential recombination event occurred between QW170/218-like and Sw43-like strains. To examine where the recombination occurred, we performed RIP analysis by placing the 3′-end RdRp and the capsid sequence of QW170 or QW218 as a query sequence and the corresponding sequences of Sw43 and QW101 as background sequences. The resulting diagram ( Figure 3A) showed that QW170 had high similarity to Sw43 in the RdRp but not in the capsid region. This abrupt change happened in the RdRp-capsid junction region. Therefore, we performed sequence alignments of the RdRp-capsid junction of NoVs, including the calicivirus genomic-subgenomic conserved 18-nt motif (20) (Figure 3B ). Between Sw43, QW170, and QW218, all 18 nt were identical, but identities decreased downstream of this motif. QW170 and QW218 grouped with Sw43 with a high bootstrap value of 95 in the RdRp tree (Figure 2 ), whereas they segregated from Sw43 with the highest bootstrap value of 100 in the capsid tree (Figure 1 ). We could not clarify which was the parent or progeny strain. The porcine NoVs replicated in gnotobiotic pigs. Two pigs were inoculated with QW101-like GII-18 porcine NoVs (QW126 and QW144 strains) to verify their replication in pigs as confirmed by quantitative RT-PCR and IEM and to produce convalescent-phase serum to examine antigenic reactivity with human NoVs. These 2 strains were confirmed as QW101-like porcine NoVs in both the RdRp (169-nt) and the capsid S domain (363-nt) regions by sequence analysis of the RT-PCR products (Q.H. Wang and L.J. Saif, unpub. data). They shared 99% and 100% amino acid identities to the QW101 strain in the 2 regions, respectively. Porcine NoV shedding, assessed by quantitative RT-PCR with primer pair PNV7/8, was detected at PID 3-5 (euthanized) after QW144 exposure, coincident with mild diarrhea. The RT-PCR-detectable units of the rectal swab RNA increased from negative at PID <2, 10 3 at PID 3-4, and 10 4 at PID 5 (large intestinal contents). Norovirus shedding was detected only at PID 5 without diarrhea after QW126 exposure. Examination of the intestinal contents of the pig inoculated with QW144 by IEM with pig convalescent-phase antiserum LL616 showed clumps of ≈32-nm NoV particles (Figure 4) . The 2 control pigs had no virus shedding or diarrhea. Detailed studies of the pathogenesis of porcine NoVs in gnotobiotic pigs are in progress (S. Cheetham and L.J. Saif, unpub. data). Antisera to QW101-like (QW126) porcine NoVs crossreacted with VLPs of human GII NoVs in ELISA and Western blot. In ELISA (Table 4) , the pig convalescentphase antiserum (LL616) to QW101-like porcine NoV QW126 strain showed higher titers (1:400-1:800) to GII-3/Toronto, GII-4/MD145, GII-4/HS66, and GII-6/Florida strains; a lower titer (1:100) to GII-1/Hawaii strain; and lowest titer (1:10) to GI-3/Desert Shield strain. In Western blot ( Figure 5 ), the capsid proteins (59-60 kDa) of Toronto, MD145, HS66, and Florida strains, but not the Hawaii and Desert Shield strains, were detected by pig antiserum LL616 but not the negative control serum (data not shown). Thus, 1-way antigenic cross-reactivity exists between human NoV antigens and porcine NoV (GII-18) antiserum, with moderate cross-reactivity to human NoVs GII-3, 4, and 6; low cross-reactivity to GII-1; and very low cross-reactivity to GI-3. All porcine NoVs were detected from pigs without clinical signs (9, 10) . Subclinically infected pigs may be natural reservoirs for NoVs, and because porcine GII NoVs are genetically and antigenically related to human NoVs, concerns exist about their zoonotic potential. Whether human NoV strains similar to the QW101-like porcine NoVs circulate among people with occupational exposure to pigs is unknown, but such studies could provide information on the zoonotic potential of these porcine NoVs. The RdRp-capsid junction region of NoVs contains a highly conserved 18-nt motif in genomic and subgenomic RNA that is believed to be a transcription start signal (1, 20) . All 18 nt were identical within each genogroup except for the Hu/GII/J23, Po/GII/QW101, and Po/GII/QW125 strains ( Figure 3B , sequence alignments on other GI and GIII strains are not shown). This finding suggests that homologous recombination may occur within this motif between NoVs of different genotypes within the same genogroup. Recombinant human GII NoVs have been reported previously (20) (21) (22) (23) (24) . To our knowledge, this study is the first identification of a potential recombinant between pig NoVs. At present, NoV recombinants have been detected exclusively between viruses within the same genogroup and within the same host species, but few animal NoVs have been sequenced (RdRp and capsid) for comparative analysis, especially those from animals in developing countries, where humans and animals may be in close contact. The QW101-like porcine NoVs replicated in gnotobiotic pigs with fecal shedding, documented by quantitative RT-PCR and IEM. No cell culture system or animal disease models are available for human NoVs, which impedes the study of their pathogenesis, replication strategies, host immune responses, and preventive approaches. The infection of pigs with porcine NoVs may provide a new infection or disease model to study NoV infections. In this study, 1-way antigenic cross-reactivity occurred between antiserum to QW101-like porcine NoVs and the capsid proteins of human NoVs, with highest cross-reactivity to GII-3, 4, and 6 NoVs. This finding coincides with the finding that the QW101 strain shares high amino acid identity with GII-3 (71%), GII-6 (71%), and GII-4 (63%) NoVs. In summary, 3 genotypes of porcine NoVs were detected in US swine. One genotype (QW101-like, GII-18) was genetically and antigenically most closely related to human GII NoVs. Potential recombinant porcine NoV strains were identified. The QW101-like NoVs infected gnotobiotic pigs, and NoV particles were evident in intestinal contents. These results raise questions of whether pigs may be reservoirs for emergence of new human NoVs or if porcine/human GII recombinants could emerge. Figure 5 . Antigenic cross-reactivity between human genogroup (G) II norovirus (NoV) capsid proteins and a pig convalescent-phase antiserum (LL616) against porcine QW101-like (GII-18) NoV was determined by Western blot. The CsCl-gradient purified viruslike particles (1,250 ng) were separated by sodium dodecyl sulfate 10% polyacrylamide gel electrophoresis, blotted onto nitrocellulose membranes, and tested with LL616. The sucrose-cushion (40%, wt/vol) purified Sf9 insect cell proteins acted as a negative control (lane 8). Lane 1, molecular weight marker (kDa); lanes 2-7, Hu/GI-3/Desert Shield, Hu/GII-1/Hawaii, Hu/GII-3/Toronto, Hu/GII-4/MD145, Hu/GII-4/HS66, and Hu/GII-6/Florida, respectively.

Clinical Relevance and Discriminatory Value of Elevated Liver Aminotransferase Levels for Dengue Severity

BACKGROUND: Elevation of aspartate aminotransferase (AST) and alanine aminotransferase (ALT) is prominent in acute dengue illness. The World Health Organization (WHO) 2009 dengue guidelines defined AST or ALT≥1000 units/liter (U/L) as a criterion for severe dengue. We aimed to assess the clinical relevance and discriminatory value of AST or ALT for dengue hemorrhagic fever (DHF) and severe dengue. METHODOLOGY/PRINCIPAL FINDINGS: We retrospectively studied and classified polymerase chain reaction positive dengue patients from 2006 to 2008 treated at Tan Tock Seng Hospital, Singapore according to WHO 1997 and 2009 criteria for dengue severity. Of 690 dengue patients, 31% had DHF and 24% severe dengue. Elevated AST and ALT occurred in 86% and 46%, respectively. Seven had AST or ALT≥1000 U/L. None had acute liver failure but one patient died. Median AST and ALT values were significantly higher with increasing dengue severity by both WHO 1997 and 2009 criteria. However, they were poorly discriminatory between non-severe and severe dengue (e.g., AST area under the receiver operating characteristic [ROC] curve = 0.62; 95% confidence interval [CI]: 0.57–0.67) and between dengue fever (DF) and DHF (AST area under the ROC curve = 0.56; 95% CI: 0.52–0.61). There was significant overlap in AST and ALT values among patients with dengue with or without warning signs and severe dengue, and between those with DF and DHF. CONCLUSIONS: Although aminotransferase levels increased in conjunction with dengue severity, AST or ALT values did not discriminate between DF and DHF or non-severe and severe dengue.

Dengue is a mosquito-borne arboviral infection endemic to most tropical and subtropical countries [1] . Elevation of the liver enzymes aspartate aminotransferase (AST) and alanine aminotransferase (ALT) is common in acute dengue illness, occurring in 65-97% [2, 3, 4, 5] of dengue patients, peaking during the convalescent period of illness (days 7-10) [2, 4, 6] . In dengue-endemic countries, dengue is an important cause of acute viral hepatitis [7] . Elevated AST and ALT levels have been associated with bleeding [2, 4, 6] and dengue hemorrhagic fever (DHF) [3, 8] . Liver failure has been recognized as a complication and unusual manifestation of dengue [9, 10] but occurred infrequently in 3 of 270 patients in Taiwan [6] and 5 of 644 patients in Vietnam [4] . In Malaysia, 8 of 20 pediatric DHF patients developed liver failure, 1 died, and the rest recovered completely [11] . In Singapore, AST or ALT levels were not independent predictors of DHF in 1973 adult dengue patients [12] . In 2009, the World Health Organization (WHO) revised its dengue guidelines and proposed severe organ impairment as one category of severe dengue in addition to severe plasma leakage and severe bleeding [1] . Severe liver involvement was defined as AST or ALT$1000 units/liter (U/L). In Taiwan, AST.10 times the upper limit of normal (ULN) occurred in 11% of dengue patients [6] , while in Brazil this occurred in 4% of their cohort [3] . In this study, we aimed to evaluate the clinical relevance of elevated AST and ALT levels and correlate liver aminotransferase levels with dengue severity according to WHO 1997 and 2009 classifications. All laboratory-confirmed dengue patients identified from our hospital microbiology database and treated using a standardized dengue clinical care path at the Department of Infectious Diseases, Tan Tock Seng Hospital (TTSH), Singapore from 2006 to 2008 were retrospectively reviewed for demographic, serial clinical and laboratory, radiological, treatment, and outcome data. These cases were positive by real-time polymerase chain reaction (PCR) [13] . We included patients with only positive dengue serology in only subgroup analyses, as we did not have paired sera, and other etiologies for elevated AST and ALT could not be excluded without more extensive evaluation. Cases were categorized using serial clinical and laboratory data from the entire clinical course as dengue fever (DF), DHF, or dengue shock syndrome (DSS) using WHO 1997 classifications [9] . Dengue fever classification requires fever and at least two of the following: headache, eye pain, myalgia, arthralgia, rash, bleeding, and leukopenia. Dengue hemorrhagic fever requires all of the following: fever, platelet count #100610 9 /liter, bleeding, and plasma leakage [9] . Dengue shock syndrome is a case of DHF with either tachycardia and pulse pressure ,20 mmHg or systolic blood pressure ,90 mmHg [9] . Cases were also categorized as dengue without warning signs (WS), dengue with WS, or severe dengue using WHO 2009 classifications [1] . Dengue (WHO 2009) requires fever and two of the following: nausea, vomiting, rash, aches and pains, leukopenia, or any warning sign [1] . Warning signs include abdominal pain or tenderness, persistent vomiting, clinical fluid accumulation, mucosal bleeding, lethargy or restlessness, hepatomegaly, or hematocrit rise ($20%) with rapid drop in platelet count (,50,000/liter) [1, 14] . We modified the WHO 2009 warning sign of rise in hematocrit concurrent with rapid drop in platelet count by quantifying it as hematocrit $20% concurrent with platelet count ,50,000/liter, as this was shown to correlate significantly with dengue death in our adult dengue death study [14] . Severe dengue includes severe plasma leakage, severe bleeding, and severe organ impairment [1] . We performed a subgroup analysis for median maximum AST and ALT values stratified by febrile (days 1-3 of illness), critical (days 4-6), and convalescent (days 7-10) phases as defined by WHO 2009 [1] and compared across dengue severity classification according to WHO 1997 [9] and 2009 [1] . We excluded severe dengue due to isolated elevation of AST or ALT$1000 U/L from our definition of severe dengue outcome, as this would be a confounder in assessing the relevance of AST or ALT levels in defining dengue severity. Patients had AST/ALT taken at presentation and then throughout hospitalization at the physician's discretion. Maximum AST and ALT values recorded at a median of 4 days of illness (interquartile range [IQR]: 3-5 days) were used in this analysis. Those with pre-existing liver diseases were excluded. At TTSH, the ULN for AST is 41 U/L; for ALT, it is 63 U/L for males and 54 U/L for females. We assessed the clinical relevance of elevated AST or ALT levels using four liver failure criteria-two for acute liver failure, and two that determine prognosis from chronic liver disease. The American Association for the Study of Liver Diseases (AASLD) recommends defining acute liver failure in a patient as: international normalized ratio (INR)$1.5, any degree of altered mental status, and illness ,26 weeks in duration without preexisting cirrhosis [15] . The King's College criteria assess prognoses in those with acute liver failure; the criteria are: prothrombin time .100 seconds or 3 of the following: age .40 years, prothrombin time .50 seconds, serum bilirubin .18 mg/dL, time from jaundice to encephalopathy .7 days [16] . The model for endstage liver disease (MELD) determines three-month mortality based on the following formula: 3.86(log serum bilirubin [mg/ dL])+11.26(log INR)+9.66(log serum creatinine [mg/dL])+6.4 [17] . The Child-Pugh criteria include assessment of degree of ascites, serum bilirubin and albumin, prothrombin time, and encephalopathy to determine one-and two-year survival [18] . The Mann-Whitney U and Kruskal-Wallis tests were used to determine statistical significance for continuous variables, and chisquare or Fisher's exact test for categorical variables. Statistical tests were conducted at the 5% level of significance. Receiver operating characteristic (ROC) curves showing the area under the curve (AUC) were generated to determine the discriminatory performance of aminotransferase values. All statistical analyses were performed using Stata 10 (Stata Corp., College Station, TX). This was a retrospective study involving data collection from medical records. All patient data were anonymized during analysis. This study was approved by the Institutional Review Board, National Healthcare Group, Singapore [DSRB E/08/ 567]. From 2006 to 2008, 690 dengue PCR positive cases were reviewed. Males comprised 493 (71%) of the cases, and the median age of the cohort was 35 years (IQR: 27-43 years). A Charlson comorbidity index $3, which predicts increased one-year mortality [19] , was noted in 5 (0.7%) patients. With WHO 1997 classification, 62% had DF, 31% DHF, and 7% DSS. With WHO 2009 classification, 14% had dengue, 62% had dengue with warning signs, and 24% had severe dengue. Hence, by WHO 1997 classification, 38% of patients with DHF/DSS needed close monitoring, while by WHO 2009 classification, 86% of patients with warning signs or severe dengue required close monitoring. Median length of illness from onset to hospital presentation was 4 days (IQR: 3-5 days), while median length of hospital stay was 5 days (IQR: 4-6 days). Intravenous fluid was administered to 641 (93%) and platelet transfusion to 86 (12%). Intensive care unit (ICU) admission was required in 3 patients, and death occurred in 1 patient due to pneumonia. (1) Elevation of AST/ALT and risk of liver failure Overall, 595 (86%) had AST above the ULN, and 316 (46%) had ALT above the ULN. Seven patients (1.0%) had severe dengue according to WHO 2009 criteria concurrent with AST or ALT$1000 U/L while three additional patients had severe dengue due to AST or ALT$1000 U/L only. Of the former seven patients, 86% had severe plasma leakage, 29% had severe bleeding, and none had severe organ impairment other than isolated AST or ALT$1000 U/L. Among the 3 patients admitted to the ICU, AST or ALT values were above the ULN but below 1000 U/L. No patients in our cohort developed acute liver failure under AASLD or King's College criteria. With Child-Pugh scoring, 2 (0.3%) belonged to Child-Pugh class C. With MELD scoring, predicted three-month mortality of 6% were identified in 68 (10%) Dengue is a global public health problem, as the incidence of the disease has reached hyperendemic proportions in recent decades. Infection with dengue can cause acute, febrile illness or severe disease, which can lead to plasma leakage, bleeding, and organ impairment. One of the most prominent clinical characteristics of dengue patients is increased aspartate and alanine aminotransferase liver enzyme levels. The significance of this is uncertain, as it is transient in the majority of cases, and most patients recover uneventfully without liver damage. In this study, we characterized this phenomenon in the context of dengue severity and found that, although liver enzyme levels increased concurrently with dengue severity, they could not sufficiently discriminate between dengue fever and dengue hemorrhagic fever or between non-severe and severe dengue. Therefore clinicians may need to use other parameters to distinguish dengue severity in patients during early illness. Association between Transaminase Levels and Dengue www.plosntds.org patients in our cohort and 19.6% in 2 (0.3%) patients. The same two patients who were Child-Pugh class C also had a predicted 19.6% three-month mortality using MELD scoring; they both had DSS and severe dengue. (2) Dengue severity and AST/ALT values Median AST and ALT values for dengue without warning signs, dengue with warning signs, and severe dengue (Table 1) In other hemorrhagic fevers, higher AST:ALT ratios correlated with disease fatality [20] . In our PCR-positive cohort, median AST:ALT ratios for DF, DHF, and DSS were 1.68, 1.68, and 1.88 (p = 0.29) and for dengue without WS, dengue with WS, and severe dengue, they were 1.60, 1.68, and 1.78 (p = 0.10), respectively. The majority of our patients' maximum AST and ALT values were recorded during febrile (n = 258) and critical (n = 377) phases of acute dengue illness. By WHO 2009 dengue severity classification, the median AST and ALT values were significantly higher for severe dengue compared to dengue with and without warning signs during both the febrile and critical phases but not the convalescent phase (Table 3) . By WHO 1997 classification, the median AST and ALT values were significantly higher for DHF versus DF and DSS in the febrile phase only but not critical and convalescent phases (Table 4 ). (4) Does a threshold AST or ALT value defining severe dengue exist? In order to determine the reliability of AST and ALT values in defining dengue severity, ROC curves for AST and ALT against severe dengue excluding isolated transaminitis were determined (Figure 1 ). The AUC for AST was 0.62 (95% confidence interval [CI]: 0.57-0.67) and for ALT, 0.60 (95% CI: 0.54-0.64). This demonstrates that AST or ALT levels are insufficient to differentiate among the WHO 2009 dengue classifications. They were also poorly discriminatory between DF and DHF, as the areas under the curve (AUC) for AST and ALT were 0.56 (95% CI: 0.52-0.61) and 0.55 (95% CI: 0.51-0.59), respectively ( Figure 2 ). In our serology-positive cohort, the AUC values for AST and ALT were 0.56 and 0.54 for differentiating between DF and DHF. The AUC values for severe and non-severe dengue were 0.64 and 0.60 for AST and ALT, respectively. The box plots in Figure 3 for the distributions of AST values show considerable overlap among the liver enzyme values for those with dengue with and without warning signs, and severe dengue. Because there were extreme outliers in our cohort, only those with AST below 1000 U/L were included in these plots. Figure 4 shows overlapping AST values among those with DF and DHF. Similarly, considerable overlap was observed in ALT values for patients with dengue with and without warning signs, and severe dengue, as well as for DF versus DHF (data not shown). Our analysis showed that liver aminotransferase levels were associated with but did not adequately differentiate between dengue severity. Although median AST and ALT values were significantly higher in those with DHF/DSS versus DF, and severe dengue versus non-severe dengue, very few (1.0%) had AST or ALT$1000 U/L. Notably, none developed liver failure, and death occurred in only 1 patient (0.1%). The majority of patients recovered uneventfully. The lack of acute liver failure in our study was not unusual, as the incidence of acute liver failure in dengue patients was 1.1% in studies by Trung and Kuo [4, 6] . The largest study to date reported no acute fulminant hepatitis [3] . In contrast to these adult studies, it is noteworthy that in dengue-endemic countries, dengue may be an important cause of acute liver failure in children [21, 22] . While some studies have shown that AST and ALT values differ between DF and DHF [3, 4, 8] , few studies support AST or ALT as an independent predictor of DHF [23] . Two studies in Singapore found liver aminotransferase levels to be significantly elevated among DF and DHF patients [12] and survivors and nonsurvivors of dengue [24] on univariate analysis, but this association was lost after adjusting for confounders on multivariate analysis. Trung et al. showed significant differences comparing other febrile illness, dengue without plasma leakage, and dengue with plasma leakage with and without shock during critical and convalescent phases for AST but during critical phase for ALT only [4] . We made the novel finding that liver aminotransferase Association between Transaminase Levels and Dengue www.plosntds.org levels may significantly vary according to dengue severity during the febrile phase. For DHF by WHO 1997 classification, both AST and ALT were significantly higher during the febrile phase compared to DF or DSS, and for severe dengue by WHO 2009, AST and ALT were significantly higher during the febrile and critical phases. The impact of co-infection with hepatitis viruses or concomitant hepatotoxic drugs was not assessed in our retrospective study, although we did exclude those with known liver comorbidities. Kuo et al. found that hepatitis B or C did not increase the extent of liver aminotransferase elevation in a retrospective adult dengue study in Taiwan [6] . In contrast, Trung et al. found that hepatitis B co-infection modestly increased ALT levels without significant clinical impact in a prospective adult dengue study in Vietnam [4] . Tang et al. showed that dengue and hepatitis B co-infected patients showed an aberrant cytokine secretion profile compared with those with dengue alone. [25] . In Singapore, seroprevalence for hepatitis B was 2.8% [26] and hepatitis C 0.37% [27] . The etiology of elevated aminotransferase levels during acute dengue illness is unclear since AST is expressed in the heart, skeletal muscle, red blood cells, kidneys, brain, and liver, while ALT is secreted primarily by the liver [28, 29] . Because dengue infection can cause acute damage to these non-hepatic tissue types that express AST, raised aminotransferase levels may not be entirely due to severe liver involvement. It is therefore possible that the patients with high AST levels were also more likely to be classified as severe dengue under the 2009 criteria due to the common pathways to non-hepatic tissue damage, even though there is no association with poorer outcome. Our retrospective study has some limitations. Aspartate and alanine aminotransferase values were tracked according to clinical judgment rather than at regular intervals during illness. We did not have dengue serotype data for each patient, but in 2006, DENV-1 was predominant in Singapore with a switch to DENV-2 in 2007-2008 [30] . Serology-positive cases were not included in primary analyses because our clinical laboratory used a rapid diagnostic test with potential for false positive results [31] , we did not have paired sera to confirm dengue diagnosis [9] , and not every patient with elevated AST or ALT was comprehensively evaluated for other etiologies of viral and non-viral hepatitis. Although serology-positive cases presented later during illness, we saw no difference in outcome. Five serology-positive patients (0.34%) required ICU admission versus 0.43% of PCR-positive cases, while four patients (0.27%) died in the serology-positive cohort, versus 1 patient (0.14%) among PCR-positive cases. However, relative data accuracy in our retrospective study was made possible by using a standardized dengue clinical care path. Another limitation of this study is the relatively few cases with substantially elevated liver aminotransferase levels. At the same time, since our cohort comprised primarily adults, additional studies in pediatric populations will be useful to confirm our findings. In patients with DHF/DSS or severe dengue, early diagnosis and proper management may improve outcome in most patients without comorbidities. However, in resource-limited countries, patients with severe disease may present late to the hospital with shock, with or without organ impairment at the time of admission. Our study highlights that early diagnosis and proper management of dengue patients may lead to excellent prognosis without organ injury. In conclusion, elevated aminotransferase levels were associated with DHF/DSS and severe dengue in our cohort of adult patients with confirmed dengue. However, no threshold values discriminated between DF and DHF or between severe dengue and nonsevere dengue.

Directed Fusion of Mesenchymal Stem Cells with Cardiomyocytes via VSV-G Facilitates Stem Cell Programming

Mesenchymal stem cells (MSCs) spontaneously fuse with somatic cells in vivo, albeit rarely, and the fusion products are capable of tissue-specific function (mature trait) or proliferation (immature trait), depending on the microenvironment. That stem cells can be programmed, or somatic cells reprogrammed, in this fashion suggests that stem cell fusion holds promise as a therapeutic approach for the repair of damaged tissues, especially tissues not readily capable of functional regeneration, such as the myocardium. In an attempt to increase the frequency of stem cell fusion and, in so doing, increase the potential for cardiac tissue repair, we expressed the fusogen of the vesicular stomatitis virus (VSV-G) in human MSCs. We found VSV-G expressing MSCs (vMSCs) fused with cardiomyocytes (CMs) and these fusion products adopted a CM-like phenotype and morphology in vitro. In vivo, vMSCs delivered to damaged mouse myocardium via a collagen patch were able to home to the myocardium and fuse to cells within the infarct and peri-infarct region of the myocardium. This study provides a basis for the investigation of the biological impact of fusion of stem cells with CMs in vivo and illustrates how viral fusion proteins might better enable such studies.

Mesenchymal stem cells (MSCs) show promise for therapeutic recovery of function of damaged myocardium [1] [2] [3] [4] [5] . MSCs home to injured tissues [6, 7] and contribute to the structure or functional recovery of the myocardium via (1) secretion of paracrine factors that can inhibit immune responses [8] and/or facilitate angiogenesis [7, 9, 10] , (2) transdifferentiation/metaplasia [11, 12] , and (3) nuclear reprogramming through fusion with resident cardiomyocytes (CMs) [13] . The latter has been largely dismissed since the frequency at which fusion is detected is low relative to the number of transplanted MSCs. However, recent studies by us [14] and others [15] [16] [17] suggest that despite the low frequency cell fusion still may exert a dramatic impact on stem cell programming or reprogramming in the heart. Cell fate determination was once thought to be unidirectional [18] , that is, as progenitor cells differentiate there is a progressive and permanent inactivation of specific genes that allow for their potency. However, technological advances suggest this is not strictly the case. Pioneering experiments of nuclear reprogramming utilized cell fusion to demonstrate that cytoplasmic elements of one fusion partner can impact nuclear transcription factors of the other fusion partner, inducing programming or reprogramming [19] [20] [21] . Later studies pinpointed specific transcription factors that, when activated exogenously, can fully reprogram somatic cells to an embryonic-like state [22] [23] [24] [25] [26] . Though successful reprogramming has been realized with this tailored in vitro approach, programming may require greater temporal control. Spontaneous physiologic cell-cell fusion is a temporally and spatially regulated process essential for programming or 2 Stem Cells International differentiation of certain cell types [27, 28] . Thus cell fusion may also confer a regulated transfer of transcriptional control necessary to drive stem cell or progenitor cell differentiation for repair of tissues in mature animals. Cell-cell fusion occurs when the plasma membranes of neighboring cells fuse to form a multinucleated cell. To fuse, lipid bilayers of cell membranes must come into very close contact, in the range of several angstroms. To achieve this degree of close proximity, the two surfaces must become at least partially dehydrated as water bound to the membrane enhances polar repulsion of membranes. Next, one or both bilayers must be destabilized in some way, inducing a localized rearrangement of the bilayers. If both bilayers are destabilized, an aqueous bridge is formed and the cytoplasmic contents of both cells mix. Destabilization of membranes can occur as the result of physical stress (e.g., electrofusion) or chemical interference (e.g., polyethylene glycol). Electrofusion utilizes short pulses of electricity to mechanically disrupt the lipid bilayer of a cell to form pores and if two disrupted membranes come into contact, cell fusion may occur [29] . Unfortunately, this process is toxic and the cells must be in contact with one another at the time the electric field is administered. Laser trapping prior to electrofusion has been used to more effectively position fusion partners, however the process is low throughput and cytotoxic [30, 31] . A less toxic, but also less effective and less reproducible approach uses polyethylene glycol (PEG) [32, 33] . The exact mechanism of PEG-induced fusion is unknown but is theorized to be due to either local dehydration leading to unfavorable molecular packing of the bilayer or to dehydration of the "water shell" near the lipid bilayer, causing the water molecules between cells to be displaced, thereby forcing the two membranes together and subsequently fusing the cells [34] . This technique has proven useful, but fusion only occurs during the time of administration of PEG, thus cell delivery with PEG would induce fusion immediately and nonselectively. A mechanism that would better regulate fusion either to specific cells or specific regions within tissues is necessary to study fusion in vivo. In nature, destabilization of cell membranes and subsequent membrane fusion utilizes the activation of specific integral membrane proteins, termed fusogens. The primary source of information about fusogen architecture, receptor binding, and activation are from viruses. The most extensively characterized fusogens are influenza hemagglutinin (HA) and human immunodeficiency virus type 1 envelope protein (HIV-1 Env). Both fusion peptides are hydrophobic and require proteolytic cleavage, but HA is activated under acidic pH during endocytosis, while HIV-1 Env fuses at neutral pH (reviewed in [35] [36] [37] ). Less well-studied are the fusogens required for eukaryotic cell fusion such as the fusion of osteoclasts, myoblasts, and trophoblasts. The greatest challenge has been establishing which proteins are true fusogens and which proteins facilitate fusion by placing cells in close proximity. Many putative fusogens have been shown to be supporting proteins (i.e., essential for adhesion or migration). The identification of true fusogens is so difficult that groups have proposed ranking schemes to clarify the nature and function of these proteins [28] . Because putative fusogens for spontaneous stem cell fusion have not been identified, developing alternative strategies to direct stem cell fusion could augment our understanding of the biological impact of such fusion. Here we utilize viral machinery from vesicular stomatitis virus (the glycoprotein, VSV-G), of the Rhabdoviridae family, to induce heterotypic fusion between human MSCs and mouse CMs in vitro and in an in vivo mouse model of myocardial infarction. Following MSC-CM fusion, we tracked the phenotype and morphology of fusion products for one week in vitro and 3 weeks in vivo. VSV-G was selected because it does not require proteolytic cleavage, is the sole mediator of receptor binding and fusion, and is pH dependent [38, 39] . In particular, VSV-G does not require facilitating proteins to either dock to the host membrane prior to fusion, or enzymes to prompt the activation of the fusogen. Furthermore, the pH dependence of VSV-G is advantageous as the local heart pH after acute ischemic injury [40] [41] [42] is within the acidic range needed to initiate a conformational change in VSV-G [38, 39] . In this way, selective activation of VSV-G on transfected MSCs (vMSCs) at the site of myocardial injury should induce local fusion in situ, thereby increasing donor cell engraftment and integration within the tissue and potentially facilitate cardiac differentiation. MSCs derived from human embryonic stem cells (MSCs from WA-09, a gift of Dr. Peiman Hematti) and HL-1 cardiomyocytes (a gift of Dr. William Claycomb) were expanded and cultured as previously described [43, 44] . Briefly, MSCs were cultured on a 0.1% gelatin (Sigma-Aldrich, St. Louis, MO, USA) pretreated flask containing α-minimum essential medium-(MEM-) complete. Alpha-MEM-complete consisted of α-MEM (Invitrogen, Carlsbad CA, USA), 10% fetal bovine serum (Hyclone, Logan UT), 0.1 mM nonessential amino acids (Invitrogen), and 2 mM Lglutamine (Invitrogen). MSC cultures were allowed to grow to 60-70% confluency and were replated at a concentration of 1,500 cells/cm 2 . CMs were cultured on fibronectin/gelatin (1.25 mg fibronectin/100 mL 0.02% gelatin) (Sigma-Aldrich) pretreated flasks containing Claycomb-complete. Claycombcomplete medium was comprised of Claycomb medium (SAFC Biosciences, St. Louis, MO, USA), 10% fetal bovine serum qualified for CMs (SAFC Biosciences), 100 U/mL: 100 μg/mL penicillin-streptomycin (Lonza, Walkersville, MD, USA), 0.1 mM norepinephrine (Sigma-Aldrich), and 2 mM L-glutamine (Invitrogen). CMs were passaged at 100% confluence and split 1 : 2. Experiments were performed using passages 7-10 and 60-110 for MSCs and CMs, respectively. All cultures were maintained at 37 • C in 5% CO 2 . MSCs were transfected with a pCVSV-G-1 plasmid [45] that encodes VSV-G under a CAG promoter using the Neon Transfection System (Invitrogen), according to manufacturer's protocol. Briefly, 5 × 10 5 cells were transfected with 2 μg of plasmid with one 1,300 V pulse for 20 msec and plated in 6-well plates. To determine transfection efficiency, electroporated cells were cultured for 24 h and immunocytochemistry (ICC) was performed to detect VSV-G protein expression. Briefly, cells were washed with two rinses and two incubations of 1X PBS. Cell fixation was performed with 4% PFA, followed by another set of washes, and probed with the 1 : 50 dilution of FITCconjugated anti-VSV-G antibody (GeneTex, San Antonio TX, USA) in 3% BSA for 60 min. Cultures were washed a final time and mounted in DABCO/DAPI mounting medium (2. The Quantum MESF kit consists of 5 populations of microspheres with increasing surface-labeled fluorochrome, which have been standardized to specific concentrations of pure fluorophore per microsphere. Each population was analyzed via flow cytometry and a standard curve was generated by plotting population (i.e., concentration of fluorophore per microsphere) versus intensity. QuickCal software was used to verify the linearity of the standard curve. Next, vMSCs and corresponding control populations were labeled with an anti-VSV-G-FITC antibody and analyzed via flow cytometry. Using the standard curve and the measured intensity value for vMSC populations and corresponding controls, the number of fluorophores per cell was determined. This value was divided by the average number of fluorophores (4.2) that bind to a single antigen to determine the number of proteins expressed per cell. Ten thousand cells and three replicates were analyzed per population. Populations included vMSCs with anti-VSV-G antibody, MSCs with anti-VSV-G antibody and vMSCs without antibody. To determine if vMSCs fuse more readily with cardiomyocytes than untreated MSCs, vMSCs and MSC controls were cocultured with CMs and analyzed for incidence of fusion. To distinguish cell types in cocultures, MSCs and CMs were stained with 1 μm Cell-Tracker Green CMFDA and 20 μm Red CMTPX (Molecular Probes Eugene, OR, USA), respectively, according to the manufacturer's protocol. Following labeling, 5 × 10 5 CMs were plated and cultured for 4 h followed by the addition of 1.5 × 10 5 MSCs or vMSCs per well in 6-well plates (BD Biosciences). After 14 h of coculture, suspensions were washed with 1X PBS and then bathed for 2 min in fusion media [46] of varying pH (i.e., pH 5.5, 6.5, or 7.5 that correspond to active and inactive forms of the VSV-G fusion protein) adjusted with HCl. For long-term characterization of fusion products, medium was changed 1 day and 4 days after coculture. Cocultures of CMs with MSCs or vMSCs were maintained in culture medium for 4 h after incubation with fusion medium, followed by imaging and flow cytometry. Images were acquired with a 20X UPlanFluor objective (NA = 0.5), FITC and Texas Red filters, on an IX71 inverted deconvolution fluorescence microscope (Olympus Center Valley, PA, USA) and analyzed with Slidebook software (Intelligent Imaging Innovations Denver, CO, USA) and ImageJ (Fiji; open source software, http://pacific.mpi-cbg.de/wiki/index.php/Fiji). Images were normalized using unstained controls. Cells were analyzed at the UWCCC Flow Cytometry Facility on a FACSCalibur flow cytometer (BD Biosciences). Events were live/dead gated with forward scatter and side scatter plots. Fusion products were quantified by gating the region positive for FL1 and FL2 channels, corresponding to CellTracker Green CMFDA and Red CMTPX, respectively. MSCs or CMs in monolayer were stained for proteins characteristic of MSCs (CD73, CD90, and CD105), as well as proteins characteristic of CMs (sarcomeric myosin (MF20)). Cell cultures were fixed with 4% paraformaldehyde for 10 min, followed by two washes with phosphate buffered saline (Fisher Scientific). Cells were probed with goat anti-CD73 (V-20, Santa Cruz Biotech, Santa Cruz, CA, USA), rabbit anti-CD90 (RB3970, Abgent, San Diego, CA, USA), goat anti-CD105 (GKY02, R&D Systems, Minneapolis, MN, USA), and mouse anti-MF20 (IgG2b, Developmental Studies Hybridoma Bank, Iowa City, IA) diluted 1 : 25, 1 : 50, 1 : 50, and no dilution, respectively, in diluting buffer (5% BSA (Fisher Scientific), 0.02% NaN3-(Acros Organics) in phosphate buffered saline (Fisher Scientific)) and incubated for 30 min at room temperature or overnight at 4 • C, followed by incubation with fluorescent secondary antibodies: donkey anti-goat Alexa Fluor (AF488, Invitrogen), goat anti-rabbit Alexa Fluor (AF647, Invitrogen), and donkey anti-mouse (AF546, Invitrogen) at a 1 : 200 dilution in preadsorption solution (90% diluting buffer, 5% human serum (Pelfreez, Brown Deer, WI, USA), and 5% mouse serum (Equitech-Bio, Inc, Kerrville, TX, USA)) for 45 min at room temperature. Samples were counterstained with DABCO/DAPI mounting solution. Fluorescence emission was detected on an IX71 inverted deconvolution fluorescence microscope (Olympus). Images were acquired with a 20X UPlanFluor objective (NA = 0.5), and analyzed using MSC-CM or vMSC-CM cocultures were probed with antibodies against CM marker (MF20) and MSC marker (CD105) to evaluate the morphology of fusion products and the phenotype of cells within coculture. Positive events for fusion were calculated as the percentage of CD105 and MF20 positive cells containing a nucleus divided by total number of nuclei obtained from analysis of at least eight optical fields per sample. Fields (3-10 fields) were selected based on cell number (minimum of 3 cells) and position within the wells (center of wells) n = 2. Myocardial infarction was induced in C57BL/6 mice (Jackson Laboratory, Bar Harbor, ME, USA) by left coronary artery ligation as previously described [47, 48] and as is routinely performed in the University of Wisconsin Cardiovascular Physiology Core Facility. All animal procedures were performed in accordance with the guidelines of the American Association for Laboratory Animal Science and the University of Wisconsin-Madison Animal Care and Use Committee. to the Murine Myocardium. TissueMend (TEI Biosciences) was prepared and cells were seeded as previously described [48] . Briefly, TissueMend matrices (2 mm × 2 mm × 0.8 mm) were placed in wells of 24-well plates containing α-MEM-complete culture medium. Following electroporation, vMSCs were seeded on the TissueMend sections at a concentration 3 times greater than MSCs (cell control due to 30% cell viability after electroporation, yielding 1 × 10 6 cells/mL. Medium was changed at 24 and 48 h, at which point the TissueMend matrix containing ∼2.3 × 10 4 MSCs, vMSCs, or unseeded (matrix control) was tacked to the myocardium at each corner of the matrix. The matrix was placed such that it was in contact with both the infarct and the peri-infarct regions of the myocardium [48] . Murine hearts were harvested three weeks after matrix implantation to assess the occurrence and, if detected, the frequency of in vivo fusion. Hearts were bisected longitudinally through the matrix. The tissues were immediately placed into 10% buffered formalin (pH = 7.2; Fisher Scientific) for 24 h followed by 24 h of fresh 10% buffered formalin, and a final 24 h of 70% ethanol. Samples were further processed for paraffin embedding and sectioning as previously described [49] . Fluorescent in situ hybridization (FISH) tissue digestion kit with all human centromere probe (red) and all mouse centromere probe (green) (Kreatech, Amsterdam, the Netherlands) was performed on sections to detect fusion events. Samples were processed by the Cytogenetics Laboratory (WiCell Research Institute, Madison, WI, USA) according to manufacturer's protocol. Briefly, slides with paraffin embedded sections were baked for 4 h at 56 • C. Specimens were incubated with pepsin for 70 min for tissue digestion prior to sequential hybridization of the human probe followed by mouse probe. Images were acquired with a 60X UPlanSApo (NA = 1.35 Oil), DAPI, Green, and Orange filters, on an Olympus BX41 Upright Fluorescence Microscope (Olympus Valley, PA, USA), and analyzed with FISHView Version 5.5 software (Applied Spectral Imaging, Vista CA). Fusion events were defined as nuclei with positive staining for both human centromeres (red) and mouse centromeres (green). The frequency of fusion events was reported as the number of fusion events per total nuclei for a given region of the heart tissue: myocardium (myo), myocardial infarct (MI), border region (border), within Tissuemend patch (TM), and in the healthy myocardium (healthy). Five to twelve fields were selected for each location and the number of hearts analyzed were, n = 1 for TM only and n = 3 for TM + vMSC. For comparison of VSV-G expression, fusion frequency, and fusion product morphology versus controls, a normal distribution was assumed and oneway analyses of variance and Student's t-test were used. Data were analyzed with Microsoft Excel (Microsoft, Redmond, WA, USA). MSCs were induced to express VSV-G via transfection by electroporation. Low transfection efficiency would limit the ability of VSV-G to promote fusion and so VSV-G expression on MSCs was determined following electroporation. Twenty-four hours after transfection control MSCs and MSCs transfected with VSV-G (vMSCs) were probed with an anti-VSV-G antibody conjugated to fluorescein isothiocyanate (FITC) and visualized with fluorescence microscopy to determine the percentage of cells expressing VSV-G. The average transfection efficiency was 32% ± 5% (n = at least 6 optical fields per sample per trial for 3 trials, Figure 1(a) ). Since vMSCs will be harvested for in vivo studies, we also assessed VSV-G expression via flow cytometry after removal from culture plates with trypsin. We found expression of VSV-G plummeted to 5% ± 2% (n = 1 replicate per sample per trial for 3 trials, Figure 1(b) ). This is perhaps not surprising as others have reported decreased stability of VSV-G with trypsin treatment [50, 51] . Trypsin is a serine protease that cleaves carboxyl groups on the cell surface to remove cells from a culture surface. VSV-G is a cell surface protein that would be exposed to the dissociation reagent [52] . The disruption to VSV-G by trypsin was corroborated by evaluating the number of VSV-G proteins per cell. The administration of trypsin significantly reduced the number of VSV-G proteins on the cell surface of vMSCs (Figure 1(d) ). Thus we replaced trypsin with Accutase, a mixture of proteases and collagenases that has been shown to improve cell viability compared to trypsin [53] . With Accutase treatment, the average number of cells expressing VSV-G after cell harvest was 21% ± 7%, a significant improvement over treatment with trypsin and at a level high enough to discern whether expression of VSV-G can impact MSC-CM fusion (n = 1 replicate per sample per trial for 3 trials). Fusogens such as VSV-G are most potent on the cell surface when there are adequate amounts of protein to facilitate fusion, but a low enough amount to avoid immune responses. Thus VSV-G protein expression per cell was determined using a Quantum MESF kit (Bangs Laboratories, Inc.), on the FACSCalibur. Using this method, the average number of VSV-G proteins per cell was 8 × 10 4 ± 2 × 10 4 with trypsin and 1 × 10 6 ± 8 × 10 3 with Accutase (n = 1 replicate per sample per trial for 3 trials, Figure 1(c) ). Thus, all further experiments were performed using Accutase as the dissociation reagent to prevent VSV-G cleavage. To determine whether MSCs expressing VSV-G are better equipped to fuse with CMs than unmanipulated counterparts, vMSCs or MSCs were cocultured with CMs. To distinguish cell types in cocultures, MSCs and vMSCs were stained with Red Cell-Tracker, while CMs were stained with Green CellTracker fluorescent probes prior to being combined. Since VSV-G undergoes a conformational change from its inactive form to its active form at pH < 6.2 [54, 55] , cocultures were briefly incubated (2 min) with fusion medium of pH = 5.5. Image analysis of vMSC-CM cocultures treated with fusion medium of pH = 5.5 revealed cells with colocalization of green and red fluorescence, indicating fusion events, while MSC-CM cocultures under the same pH condition exhibited limited colocalization (Figure 2(a) ). To accurately assess the amount of cell fusion, cocultures were harvested 24 hours after seeding and analyzed via flow cytometry (double positive cells correspond to fusion events). vMSCs treated with acidic medium (pH 5.5) had significantly higher rates of fusion with CMs (4.7% ± 1.1%) than MSCs treated in the same way (i.e., spontaneous fusion, 1.4% ± 0.2%) (P < 0.05) (Figure 2(b) ). Further, the percentage of fusion products identified in vMSC-CM cocultures exposed to fusion medium of pH = 6.5 or 7.5 (maintaining VSV-G in the inactive form) did not differ from MSC-CM cultures (n = 3 replicates per sample per trial for 3 trials, Figure 2 (c)). Many studies have demonstrated that stem cell programming is influenced by the microenvironment [56] [57] [58] . To determine whether the phenotypic fate of vMSC-CM fusion products could be regulated by the microenvironment, following treatment with fusion media, we cultured vMSC-CM fusion products under either MSCspecific or CM-specific culture conditions and examined the incidence of fusion and morphology of MSC-CM fusion products. At days 5 and 7 following the induction of cell fusion, cocultures were probed with CM and MSC specific antibodies (anti-MF20 and anti-CD105, respectively, n = 1 replicate per sample per trial for 3 trials). At day 5, vMSC-CM cocultures contained a relatively high number of cells that expressed both MF20 and CD105 and the percentage of MF20 + /CD105 + cells relative to the total cell number was significantly greater than that of MSC-CM cocultures for both culture conditions (P < 0.005) (Figure 3(a) ). Of note, the percentage of MF20 + /CD105 + cells was much higher than the percentage of double positive cells detected using CellTracker dyes and flow cytometry (Figure 2(b) ). This could reflect the loss of VSV-G sustained by cell harvest, the different analytical approach (i.e., flow cytometry versus image analysis) and/or the behavior of fusion products between day 1 and day 5 (i.e., proliferation). By day 7, the percentage of MF20 + /CD105 + cells decreased to levels not statistically different from controls for both culture conditions. At the same time, the number of cells expressing MF20 alone increased substantially for both culture conditions. The change in percentage of MF20 + /CD105 + cells from day 5 to day 7 could reflect death of fusion products, or programming of the MSC fusion partner to a cardiomyocyte phenotype or both. If death of fusion products occurred, one would expect unfused CMs and MSCs to proliferate to fill the voids of the culture space. Interestingly, only the CM population increased from day 5 to day 7 and at rates significantly higher than that of control cultures, suggesting at least a portion of fusion products were maintained, and ultimately adopted a cardiomyocyte-like phenotype. This result was observed independent of the culture conditions. Of note, this experimental approach does not exclude the possibility that metaplasia rather than fusion occurred, that is, MSCs differentiate into CMs as a consequence of soluble factors in the coculture medium and maintain (at least transiently) expression of each cell type. However, MF20 + /CD105 + cells were rare in MSC-CM cocultures, suggesting metaplasia alone cannot account for coexpression of MF20 + /CD105 + or subsequent loss of MF20 + /CD105 + cells. In addition, MF20 + /CD105 + cells exhibited two distinct morphologies; some were long and spread, displaying MSC-like morphology (MSC medium = 16.59% ± 6.32%; CM medium = 14.03% ± 1.59%) while the majority (P < 0.05) were round and cobblestone-like, indicative of CM-like morphology (MSC medium = 80.49%±10.45%; CM medium = 85.97%± 1.60%) (Figures 3(b) and 3(c), Supplementary Figure 1B) . These results further support the possibility that CM nuclear material and cytoplasmic elements direct programming of MSC-CM fusion products independent of culture conditions. To determine whether MSCs expressing VSV-G could fuse with cardiac cell types in vivo, vMSCs were delivered to the damaged myocardium via a TissueMend patch. We have previously demonstrated that MSCs delivered in this way are maintained in the patch and in the tissue between the patch and the myocardium up to 3 weeks after delivery at higher percentages than with conventional delivery modalities [48] . Furthermore, Laflamme et al. have found one of the major factors for cell loss during transplantation is anoikis [59] , and thus providing anchorage support to transplanted cells increases viability and retention. In this study, we sought to determine whether VSV-G expressing MSCs (donor) would be able to migrate to the damaged myocardium and fuse with recipient cardiac cell types. Thus, one day following induction of infarction via ligation of the left anterior descending artery, a patch containing vMSCs was applied to the heart in contact with healthy and damaged tissue. Three weeks after cell transplantation, heart excision, and histology were performed on left ventricular tissue as previously reported [48] . Histological sections were probed using FISH for human-specific and mouse-specific centromeres and all nuclei containing both probes were considered fusion products. Human cells were found in the TissueMend patch and in the "border region" (the area between the patch and the myocardium). Donor-host cell fusion was evident in the TissueMend patch, the border zone, and in the infarcted myocardium of hearts receiving TissueMend with vMSCs. No human cells or fusion products were found in the healthy cardiac tissue of hearts receiving TissueMend with vMSCs. In addition, no human cells or fusion products were found in the TissueMend patch, border zone or infarcted myocardium of hearts receiving TissueMend only. In regions of hearts receiving TissueMend with vMSCs and selected for high density of fusion events, the frequency of cell fusion relative to the total number of nuclei in a given region was 22% ± 16%, TissueMend patch (n = 3 hearts, 12 fields); 14%±10%, border zone (n = 3 hearts, 5 fields); 19% ± 10%, infarcted myocardium (n = 3 hearts, 8 fields). Though these levels represent the maximum amount of fusion per region, they are substantially higher than those previously reported for Merge CMs Fusion of transplanted stem cells with recipient cardiomyocytes has been observed in murine [13, 60] and porcine model systems [49] . But since these first observations, few have sought to unravel the mechanisms that govern stem cell fusion or to study the implications of cell fusion for stem cell programming. Lack of study reflects the overwhelming opinion that cell fusion occurs too infrequently to be of relevance for stem cell programming and, by corollary, for tissue repair. However, this opinion fails to appreciate the possibility that (1) detection methodologies may be insufficient to accurately gauge the contribution of cell fusion following stem cell transplantation and/or (2) that we might be able to control or increase the frequency of cell fusion to more effectively induce programming of stem cells following transplantation. We have begun to explore this second possibility by co-opting the well-described fusion machinery of viruses. We find that mesenchymal stem cells modified to express viral fusogen VSV-G are more apt to fuse with cardiomyocytes in a pH-dependent manner. vMSC-CM fusion products formed in this way are prone to adopt cardiomyocyte phenotype and morphology. In addition, vMSCs delivered to the myocardium of mice following infarction can fuse with resident cardiac cell types at rates much higher than previous reports of spontaneous fusion [13, 61] and are more apt to fuse at the site of infarction than in the healthy myocardium. Increasing the frequency of MSC-CM cell fusion will aid in the study of cell fusion in vitro and may improve the therapeutic benefit of MSCs in vivo. One way that the therapeutic benefit may be improved is via induction of programming of MSCs to a cardiomyocyte fate. Differentiation of MSCs into CMs can be initiated in vitro via soluble factors including 5-azacytidine [62] [63] [64] or with exposure to insoluble factors including laminin [65] . However, functional differentiation of MSCs to cardiomyocytes has only been accomplished to date via cell fusion with mature cardiomyocytes. This result has been demonstrated in vitro [66] and in vivo wherein MSC-CM fusion products take on a cardiomyocyte morphology, express cardiomyocyte markers, and couple to adjacent cardiomyocytes [60] . Here we find that when MSC-CM fusion is induced with viral fusogens, the CM fusion partner is dominant in that the majority of fusion products (regardless of medium type) adopt a CM-like morphology and maintain expression of MF20 and lose CD105. These data further support the exciting possibility that induction of fusion with viral fusogens could enhance MSC programming to a CM fate in vivo. Of note, the CMs utilized here are HL-1 CMs. This cell line was used to enable largescale and long-term studies. However, the heterogeneity and immortal nature of these cells may account for the seeming dominance of the CM phenotype and future studies will utilize primary fetal cardiomyocytes or induced pluripotent stem cell-derived cardiomyocytes. Our results suggest that the differentiation of MSCs to a CM fate can be promoted by cell-cell fusion. However, in certain circumstances in vitro, MSC-CM fusion products can reenter the cell cycle and proliferate suggesting cell-cell fusion can also promote reprogramming of the CM [67] [68] [69] . Proliferation of fusion products may be as advantageous for cardiac tissue repair as differentiation of functional cell types since more cells could be produced to replace lost cells. In addition, recent evidence has demonstrated that MSC-CM fusion includes mitochondrial exchange, which is essential for somatic reprogramming [69] . Understanding cell-cell fusion in conjunction with mitochondrial preservation may provide alternate, simple, and direct mechanisms to rescue cells following ischemia-induced damage. There is evidence indicating that the fusion product's proliferative capacity is regulated by the stem cell while the developmental direction is dictated by the somatic cell [70] [71] [72] , and the combination of both outcomes presented herein are means to repopulate the myocardium for functional improvement. While we have utilized vMSCs to both understand and exploit the physiological role of MSC-CM fusion, induction of fusion of another stem cell, progenitor, or even mature cell types may augment our ability to repopulate and repair the damaged myocardium [59, [73] [74] [75] [76] [77] [78] [79] . In the case of mature or progenitor cell transplantation, the induction of fusion may be less beneficial from a differentiation standpoint and more beneficial from an engraftment or retention standpoint. One of the primary challenges for stem cell delivery is the ∼90% cell loss after transplantation [80] [81] [82] that has prompted the development of new methods to deliver and maintain cells in the heart [48, 83, 84] . This is particularly problematic for cardiac therapy as the heart is mechanically active, rapidly flushing cells from the intended target region. If stem cells transiently express a viral fusogen, they might rapidly adhere and so be maintained long term in the heart. The added advantage of pH sensitive fusogens, such as VSV-G, is the ability to control activity such that cells only fuse at pH lower than 6.5. This has major implications for inducing temporally (the window during ischemia) and spatially (the ischemic region) regulated fusion in vivo. In fact, vMSCs delivered to the heart were found in the patch and in damaged myocardium fused with mouse cells. The ability for VSV-G to induce fusion in the patch may be due to close proximity to the ischemic region, causing the environment to be more acidic or by the remodeling of the collagen patch [48] . Collagen remodeling has been shown to occur via MSC secretion of matrix metalloproteinases (matrixins), serine proteases, and cysteine proteases [85] . While matrixins are active at neutral pH, serine and cysteine proteases are active at acidic pH, indirectly demonstrating cells are able to make the microenvironment acidic [86] . Taken together, the induction of cell fusion in the heart could exert functional benefit via multiple mechanisms. A primary limitation of this approach is introducing viral machinery to an already damaged recipient. The entire virion, VSV, is known to be immunogenic and, at high enough concentrations, is lethal to mice [87] . Purified VSV-G or VSV-G reconstituted in lipid bilayers administered to in vitro cell culture is mitogenic (>0.8 μg/mL) [88] . Interestingly, if the lipid concentrations were increased, while VSV-G concentration was held constant, the mitogenicity decreased, suggesting that the spacing of VSV-G in the membrane plays a role. Confirming the importance of VSV-G arrangement, Ochsenbein et al. demonstrated that 1,000 times more antibody is produced by C57BL/6 mice against highly organized VSV-G on the nucleocapsid of intact VSV versus poorly organized VSV-G in micelles [89] . The amount of viral proteins we delivered (based on the mass of the protein [39] , the proteins expressed per cell combined with the number of cells delivered) is 7 orders of magnitude below the reported amount to elicit an immune response [88] and we express only the fusogen and not the entire virion. Even if methods were developed to increase expression levels per cell and/or in combination with high numbers of cells, spacing could be evaluated to avoid immune responses. However, based on the reported concentration required to elicit a response, delivery of vMSCs as prepared in this study would not trigger a response. While vMSCs may not be immunogenic, transfection itself may cause adverse genetic effects. For instance, stable transfection with most viral systems causes integration of the gene at a random site in the genome [90] [91] [92] . When mutagenesis occurs, integration may occur at a site that interferes with cells ability to regulate itself, resulting in deregulation of proliferation and tumorigenesis [93, 94] . In addition to experimental evidence of malignancy, this has been seen clinically ( [95, 96] , reviewed in [97] ). Here, transfection is largely transient and only rarely integrates into the genome. Clinical use would require further safeguards, perhaps including liposomal delivery of the protein. The data presented support the utility of VSV-G-mediated fusion to study the effects of stem cell fusion on cell reprogramming and functional improvement of tissues including the heart. Future studies may also employ VSV-G to rescue damaged cells of other ischemic tissues in the body, or even selectively target cells for destruction. For example, the microenvironment of tumors and the overactive osteoclasts in Paget's disease are below the pH threshold necessary to activate the conformational change in VSV-G. Local administration of VSV-G in liposomes containing toxic factors or highly acidic pH to this microenvironment may fuse with these poorly regulated cells and dampen their detrimental effect. This paper was supported by funding from the National Institutes of Health (HL089679).

Design Novel Dual Agonists for Treating Type-2 Diabetes by Targeting Peroxisome Proliferator-Activated Receptors with Core Hopping Approach

Owing to their unique functions in regulating glucose, lipid and cholesterol metabolism, PPARs (peroxisome proliferator-activated receptors) have drawn special attention for developing drugs to treat type-2 diabetes. By combining the lipid benefit of PPAR-alpha agonists (such as fibrates) with the glycemic advantages of the PPAR-gamma agonists (such as thiazolidinediones), the dual PPAR agonists approach can both improve the metabolic effects and minimize the side effects caused by either agent alone, and hence has become a promising strategy for designing effective drugs against type-2 diabetes. In this study, by means of the powerful “core hopping” and “glide docking” techniques, a novel class of PPAR dual agonists was discovered based on the compound GW409544, a well-known dual agonist for both PPAR-alpha and PPAR-gamma modified from the farglitazar structure. It was observed by molecular dynamics simulations that these novel agonists not only possessed the same function as GW409544 did in activating PPAR-alpha and PPAR-gamma, but also had more favorable conformation for binding to the two receptors. It was further validated by the outcomes of their ADME (absorption, distribution, metabolism, and excretion) predictions that the new agonists hold high potential to become drug candidates. Or at the very least, the findings reported here may stimulate new strategy or provide useful insights for discovering more effective dual agonists for treating type-2 diabetes. Since the “core hopping” technique allows for rapidly screening novel cores to help overcome unwanted properties by generating new lead compounds with improved core properties, it has not escaped our notice that the current strategy along with the corresponding computational procedures can also be utilized to find novel and more effective drugs for treating other illnesses.

Diabetes mellitus is a group of metabolic diseases that has been classified as a disease of glucose overproduction by tissues without enough insulin production, or a disease resulting from cells not responding to the insulin in human body [1] . Type-2 diabetes is the most common type among all the diabetes mellitus forms. The risk of developing type-2 diabetes (T2D) increases with age, obesity, cardiovascular disease (hypertension, dyslipidaemia), lack of physical activity, and family history of diabetes. Increasing dramatically in the US and worldwide, type-2 diabetes has reached epidemic scale. There are nearly 50 million individuals (US) and 314 million individuals (worldwide) with the metabolic syndrome [2] . People suffering from overweight or obesity are of huge risk for developing T2D. Peroxisome Proliferator-Activated Receptor (PPAR) has drawn increased attention as a drug discovery target by regulating glucose and lipid metabolism [3] . PPAR, and its subtypes PPARa and PPARc, belong to the superfamily of nuclear receptors that function as transcription factors activated by several ligands. PPARs played a vitally important role in treating obesity, atherogenic dyslipidemia, hypertension, and insulin resistance as main therapeutic targets [4] . The primary function of PPARa is to act as regulator responding to transport and degradation of free fatty acids as well as reverse cholesterol transport by peroxisomal and beta-oxidation pathways [5] . A class of lipid-lowering drugs, such as fenofibrate and gemfibrozil, specially activate PPARa [6, 7, 8] . PPARc played a significant role in transcriptionally regulating lots of physiological pathways, including adipocyte differentiation and glucose homeostasis [9] . Thiazolidinediones (TZDs) are a class of the antidiabetic drugs, which act by activating the special PPARc [10] . If used alone, although each of the antidiabetic drugs could enhance the insulin sensitivity and hence lower glucose or fatty acid levels in type-2 diabetic patients [9] , some side effects would be caused, such as weight gain, fluid accumulation, and pulmonary edema [11] . Recently, new dual agonists have received considerable attention for developing powerful drugs against diabetes. The strategy of dual agonists was aimed to treat both insulin resistance and dyslipidemia [12] . A critical challenge for developing dual agonists is how to identify the receptor subtype selectivity ratio [13] . Many studies have indicated that computational approaches, such as structural bioinformatics [14, 15] , molecular docking [16, 17] , pharmacophore modelling [18] , QSAR techniques [19, 20, 21, 22, 23, 24] , as well as a series of user-friendly web-server predictors developed recently, such as GPCR-2L [25] for identifying G protein-coupled receptors and their types, En-zClassPred [26] for predicting enzyme class, iLoc-Euk [27] and iLoc-Hum [28] for predicting subcellular localization of eukaryotic and human proteins, NR-2L [29] and iNR-PhysChem [30] for identifying nuclear receptors and their subfamilies, and HIVcleave [31] for predicting HIV protease cleavage sites in proteins [32, 33] , can timely provide very useful information and insights for drug development. The software of ''Core Hopping'' [34] is another very powerful and cutting-edge computational technique that is particularly useful for de novel drug design [35] . Encouraged by the aforementioned researches in successfully utilizing various computational approaches for drug development, the present study was initiated in an attempt to screen the fragment database for finding new PPAR dual agonists for treating type-2 diabetes. To realize this, the techniques of ''core hopping'' with glide docking [34, 36] as well as molecular dynamic simulation were utilized to analyze the binding interactions between the agonist and PPARs in hoping that the findings thus obtained may provide useful insights for developing new and powerful agonists against diabetes mellitus. The L-tyrosine analogue GW409544 was obtained by modifying the structure of farglitazar, a dual agonist for both PPARa and PPARc [37] . The main difference between GW409544 and farglitazar is that the former contains a vinylogous amide as the Ltyrosine N-substituent [37] . That is why we chose GW409544 as a starting template structure for designing the new PPAR dual agonists. The representative complex crystal structures of PPARa (PDB ID 1k7l) and PPARc (PDB ID 1k74) with the same ligand GW409544 [37] were download from the PDB Bank [38] , and were to be used for the molecular modelling studies. All the calculations were carried out on Dell Precision TM T5500 computer with Schrodinger software package [34, 36] and Desmond 2.4 [39] . The proteins with PDB codes 1k71 and 1k74 were chosen for modeling. In addition to the available knowledge of their 3D (dimensional) structures, the reasons of selecting the two proteins as receptors are as follows. (1) The two proteins contain the same ligand GW409544 as PPARa and PPARc do, and their binding affinities with the ligand are also quite similar; however, the former selectivity is about 10-fold weaker than the latter [37] . (2) The source organism of both PPARa and PPARc was from human. In the process of preparing receptors for modelling, the protein preparation facility [40] was used that included the operations of assigning bond orders, adding hydrogen, treating metals, treating disulfides, deleting waters and alleviating potential steric clashes, adjusting bond order, building missing heavy atom and formal charges, as well as minimizing energy with the OPLS2005 force field [41] and refining the protein by imposing the 0.3 Å RMSD limit as the constraint. The protein binding-site was identified by the SiteMap tool embedded in Schrodinger Suite 2009 (www.schrodinger.com) as described in [42, 43, 44] . The binding-site encompassed the ligand GW409544, which was observed in the crystal structures of both PPARa (1k71) and PPARc (1k74) as a ligand. The information of the binding pocket of a receptor for its ligand is very important for drug design, particularly for conducting mutagenesis studies [14] . In the literatures, the binding pocket of a protein receptor to a ligand is usually defined by those residues that have at least one heavy atom (i.e., an atom other than hydrogen) with a distance #5 Å from a heavy atom of the ligand. Such a criterion was originally used to define the binding pocket of ATP in the Cdk5-Nck5a* complex [45] that has later proved quite useful in identifying functional domains and stimulating the relevant truncation experiments [46] . The similar approach has also been used to define the binding pockets of many other receptor-ligand interactions important for drug design [15, 17, 47, 48] . In this study, we also used the same criterion [45] to define the binding pockets of proteins 1k7l and 1k74 for the ligand GW409544. A close-up view for the protein-ligand interaction at the binding pocket thus defined is shown in Fig. 1 , where panel A is for the interaction between PPARa (1k71) and GW409544, while panel B for the interaction between PPARc (1k74) and GW409544. Because the natural ligands of PPARs are fatty acids, the binding site of PPARs is almost hydrophobic. Several hydrophobic interactions with three arms of the Y-shaped ligand binding to the site are taken as the key point for designing the new PPARs agonist [49] . The PPAR binding site is composed of three arms, namely Arm I, Arm II, and Arm III, as explicitly marked in Fig. 1 . The first arm has mainly polar character including the AF2 (transcriptional activation function 2) helix indicated by red ribbon. The hydrophilic head group of the PPAR ligands forms a network of hydrogen bonds with AF2 of Arm-I; while the hydrophobic tail of PPAR agonist is either interacts with Arm II or Arm III. The network hydrogen bonds forms an important conformation for AF2-helix to generate a charge clamp, thus reducing the mobility of AF2 via binding a ligand and hence make it able to regulate the gene expression [4] . The drug-like database and the fragment database derived from ZINC [50] were used for virtual screening and core hopping searching, respectively. Many useful insights for drug design could be gained via molecular docking studies (see, e.g., [14, 17, 51, 52] ). To acquire even more useful information for drug design, a new docking algorithm called ''Core Hopping'' [36] was adopted in this study that is featured by having the functions to perform both the fragment-based replacing and molecular docking. Core Hopping [36] is a very powerful and cutting-edge technique for de novel drug design because it can significantly improve the binding affinity of the receptor with its ligands, e.g., GW409544 (Fig. 2) in the current study. The binding conformation thus obtained will be taken as an initial structure for further optimization by searching the fragment database to find the optimal cores that are attached to other parts of the template. During the process of core hopping, the 1st step is to define the possible points at which the cores are attached. It is performed in the ''Define Combinations'' step from the Combinatorial Screening panel in Schrodinger2009 (www.schrodinger.com). The 2nd step is to define the ''receptor grid file'', which was done in the ''Receptor Preparation'' panel. The 3rd step is the cores preparation that was operated with the ''Protocore Preparation'' module to find the cores attaching to the scaffold using the fragment database derived from ZINC [50] . The 4th step is to align and dock the entire molecular structure built up by the core and scaffold. The cores are sorted and filtered by goodness of alignment and then redocked into the receptor after attaching the scaffold, followed by using the docking scores to sort the final molecules. As the products of the core hopping operation, a total of 500 chemical compounds were prepared with the LigPre module [53] , which consists of the procedures of generating possible states by ionization at target pH 7.062.0, desalting, retaining chiralities from 3D structure and geometry minimization with the OPLS2005 force field [41] . When the above steps were accomplished, all investigated compounds were docked into the receptor pocket through the rigid protein docking model with the Stand-Precision (SP) scoring function [54, 55] to estimate the binding affinities. Many marvelous biological functions in proteins and DNA as well as their profound dynamic mechanisms, such as switch between active and inactive states [56, 57] , cooperative effects [58] , allosteric transition [59, 60] , intercalation of drugs into DNA [61] , and assembly of microtubules [62] , can be revealed by studying their internal motions [63] . Likewise, to really understand the interaction mechanism of a receptor with its ligand, investigations should be aimed not only at their static structures but also at the dynamic process obtained by simulating their internal motions. Here, the ''Desmond 2.4 Package'' [39] was adopted to study the internal motions of the receptor-ligand system. According to the software, the OPLS 2005 force fields [64, 65] was used to build aqueous biological systems, and the TIP3P model [66] was used to simulate water molecules. The orthorhombic periodic boundary conditions were set up to specify the shape and size of the repeating unit. In order to get an electrically neutral system, the minimum number of sodium and chloride ions needed to balance the system charge was placed randomly in the solvated system, and 0.15 mol/L sodium and chloride were then added to mimic the osmotic effect of water. Molecular dynamics simulations were carried out with the periodic boundary conditions in the NPT ensemble. The temperature and pressure were kept at 300 K and 1 atmospheric pressure using Nose-Hoover temperature coupling and isotropic scaling [67] . After all restrains were removed via the 3ns (nanosecond or 10 29 of a second) system minimization and relaxation, the operation was followed by running the 10 ns NPT production simulation and saving the configurations thus obtained in 2ps (picosecond or 10 212 of a second) intervals. All the molecular dynamics simulations were performed with a time step of 2fs (femtosecond or 10 215 of a second). The QikProp [68, 69] is a program for predicting the ADME (absorption, distribution, metabolism, and excretion) properties of the compounds. With the QikProp software, a total of 44 properties of compounds can be predicted, including the principal descriptors and physiochemical properties. All the compounds investigated need not the treatment for neutralization before using QikProp because it will be automatically done in QikProp. The normal mode was applied in the program. The property analyses for the partition coefficient (QP logP o/w), van der Waals surface area of polar nitrogen and oxygen atoms (PSA), predicted aqueous solubility (QP logS b ), and apparent MDCK permeability (QPP MDCK c ) [70] , were considered in the QikPro to evaluate the acceptability of the compounds. The process of core hopping and the final agonists' structures thus selected are illustrated in Fig. 2 , from which we can see the following. The structure of GW409544, which is conceived as an agonist model to develop novel therapeutic agents for treating metabolic disorder, may be divided into three parts, Core A, Core B, and Core C, as marked by dash lines. Considering the great importance of the acidic head in Core A for activating PPARs receptors, let us retain the Core A part during the core hopping calculation as described below. The 1st core hopping operation was aimed at the Core C part (see red part of Fig. 2) , generating five cores, Core C1, Core C2, Core C3, Core C4, and Core C5, respectively, to replace Core C. The 2nd core hopping operation was aimed at the Core B part (see (1k74) . The binding pocket is defined by those residues that have at least one heavy atom with a distance of 5 Å from the ligand [45] . The ligand GW409544 (in grey color) was extracted from the crystal structure while the ligand Comp#1 (rendered by three colors: grey for Core A; red for Core B; and blue for Core C) was generated by the ''core-hopping'' method. The hydrophobic surface of the receptor is colored in green. blue part of Fig. 2) , also respectively generating five cores, Core B1, Core B2, Core B3, Core B4, and Core B5 to replace Core B. Consequently, we have a total of 1|5|5~25 different combinations for the GW409544 derivatives thus generated. Subsequently, each of the 25 derivative compounds was docked into the two receptors PPARa (1k71) and PPARc (1k74), respectively. Listed in Table 1 are the 25 derivative compounds ranked roughly according to their docking scores to the receptors PPARa and PPARc, respectively. The top ten compounds highlighted with boldface type in Table 1 are those derivatives that are stronger than the original GW409544 in binding affinity with the two receptors. Of the top ten derivatives, the Comp#1, i.e., ''Core A-Core B1-Core C1'', has the strongest binding affinity with both PPARa (1k71) and PPARc (1k74), and hence it was singled out for further investigation. Shown in Fig. 1 is the docked conformation of Comp#1 when aligned with GW409544 extracted from (A) the crystal complex in PPARa (1k7l) and (B) the crystal complex in PPARc (1k74), respectively. As described in [37, 71] , the conversed H-bonding network formed by the polar acidic head of Core A in both GW409544 and Comp#1 to the four key residues of PPARa (or PPARc), such as Ser280 (or Ser289), Tyr314 (or His323), Tyr464 (or Tyr473) and His440 (or His449), were observed in our docking study. This Hbonding network played the role in stabilizing the conformation of the AF2-helix in arm I (red helix in Fig. 1) , which is vitally important for receptor-binding and activation [37, 72, 73, 74, 75, 76, 77] . The hydrophobic tail of both Core A and Core C of agonists are buried well in the hydrophobic arm I and arm II that are formed by hydrophobic residues as shown by the green surface in Fig. 1 . Compared with GW409544 (shown with grey color in Fig. 1) , the compound of Comp#1 (purple color in Fig. 1 ) has more bulky molecular volume owing to the large hydrophobic Core C1, which is more fitted to the hydrophobic arm II, resulting in the much better binding affinity than GW409544 (cf. Table 1 ). Molecular dynamics can provide useful information for characterizing the internal motions of biomacromolecules with time. For a comparison study, the 10 ns molecular dynamics simulations were performed, respectively, for the crystal structures of PPARa (1k7l), PPARc (1k74), as well as their complexes with GW409544 and Comp#1, i.e., PPARa-GW409544, PPARc-GW409544, PPARa-Comp#1, and PPARc-Comp#1. As we can see from Fig. 3 , all the characters concerned reached the simulation equilibrium within the 5ns (see panels A and C). Meanwhile, the corresponding root mean square deviation (RMSD) value curves of the protein backbone for PPARa, PPARc, PPARa-GW409544, PPARc-GW409544, PPARa-Comp#1, and PPARc-Comp#1 were also computed, respective-ly. It is interesting to see that the RMSD curves for PPARa-Comp#1 and PPARc-Comp#1 are remarkably more stable than those of PPARa-GW409544 and PPARc-GW409544, particularly for the case of PPARa (1k7l) system, where only a fluctuation of around 0.3 nm was observed when the complex system reached the plateau. The detailed fluctuations of the aforementioned six different structures, as well as the root mean square fluctuations (RMSF) of their side-chain atoms, were also computed within 10 ns molecular dynamics simulations (see panels B and D of Fig. 3) . It is instructive to point out that the RMSF curve of PPARa-Comp#1 or PPARc-Comp#1 is highly similar to that of PPARa-GW409544 or PPARc-GW409544, respectively. This is especially remarkable in the binding site of AF2 helix region with the residues 459-465 for PPARa-Comp#1 and residues 469-477 for PPARc-Comp#1 (see the grey frames in Fig. 3B and D) , indicating that the new designed compound, Comp#1, is very likely to have the same function for activating the AF2 helix as done by GW409544. As a negative control, the similar molecular dynamics simulation was also performed for Comp#8 (A-B2-C3), which is ranked number 25 according to the strength of binding affinity with PPARa and PPARc (cf. Table 1 ). The corresponding simulation results thus obtained are shown in the Fig. 4 , from which we can see that the fluctuating magnitudes of molecular dynamics for PPARa-Comp#8 and PPARc-Comp#8, including the RMSD and RMSF, are much larger than those of PPARa-Comp#1 and PPARc-Comp#1, especially for the binding site of AF2 helix region (see the gray frames in Fig. 4B and D) . These phenomena indicate that Comp#8 is not as good as Comp#1 in stably binding to PPARa and PPARc, and hence Comp#8 might not have the same function for activating the AF2 helix as GW409544 had. Some pharmaceutically relevant properties of the new designed agonist derivatives as well as the original GW409544 compound, such as the ''partition coefficient'' (logPo/w), ''van der Waals surface area of polar nitrogen and oxygen atoms'' (PSA), ''aqueous solubility'' (logS), and ''apparent MDCK permeability'' (PMDCK), were predicted by means of the QP program embedded in the ''Schrodinger2009 Software Package''. The results thus obtained are also listed in the Table 1 , respectively. Since PPARa and PPARc have a more spacious pocket (,1400 Å 3 ) than any other nuclear hormone receptors [37, 72] , it is quite natural that the agonist derivatives designed based on the two receptors by combining their three cores would have relatively large molecular weight (MW.500) and bulky volume, a trend quite similar to case in designing the inhibitors against the protein tyrosine phosphates (PTPase) [78] . As shown in Table 1 , the values calculated by the QP program, such as PAS, logPo/w, logS, and PMDCK for the newly designed agonists are all within the reasonable ranges. Although the higher logPo/w value of a compound, the stronger its affinity to PPARs is, it is not a good idea to excessively enhance logPo/w because this would induce bad distribution of the compound on fat and body fluid [70] . It should be pointed out that, rather than the experiential values within the range between 26.5 and 0.50, most of the log S values for the new agonists are quite close to that of GW409544. Such a phenomenon might result from the core A part which was kept unchanged during the process of designing the newly compounds as mentioned above. If the core A part was modified as well, the log S value would be further improved accordingly. Also, as mentioned above, the values for the four ADME properties listed in Table 1 are all within the acceptable range for human beings, indicating that most of the 25 compounds, particularly the top 10 derivatives found in this study as highlighted in Table 1 , can be utilized as candidates for the purpose of developing new drugs. The goal of this study was to find new and more powerful dual agonists for PPARa and PPARc. The new technique of ''core hopping'' adopted in this study allows for the rapid screening of novel cores to help overcome unwanted properties by generating new lead compounds with improved core properties. A set of 10 novel compounds were found in this regard. Compared with the existing dual agonist, the new agonists not only had the similar function in activating PPARa and PPARc, but also assumed the conformation more favorable in binding to PPARa and PPARc. It is anticipated that the new agonists may become potential drug candidates. Or at the very least, they may stimulate new strategy for developing novel dual agonists against type-2 diabetes.