Datasets:

Daniel-P-Gonzalez

/

OCD

like 0

# Introduction Bats and toothed whales have independently evolved a sophisticated biosonar system, allowing both clades to diversify and occupy many different niches. Toothed whales constitute a morphologically and ecologically diverse group of predators, inhabiting every ocean and several large, freshwater river systems. Some species forage on deep-sea squid at mesopelagic depths (e.g. sperm whales), others prey on large schools of fish sparsely distributed in oceanic habitats (e.g. dusky dolphins) or on individual shrimp and fish encountered in shallow river systems inhabited by several species of river dolphins, including Irrawaddy and Ganges river dolphins. While the biosonar signals of many marine toothed whales have been studied in detail, we know little about the polyphyletic assembly of true river dolphins and how the biosonar of these old lineages have evolved to their freshwater habitat. Toothed whale biosonar signals have been studied in captivity over the last 60 years and increasingly also in the wild. Studies of captive animals have contributed greatly towards our understanding of the biosonar performance including dynamic biosonar control. Studies of free-ranging animals complement laboratory studies by revealing how animals use echolocation in the wild, where the natural habitat may have physical characteristics very different from captive settings. Four different types of odontocete biosonar signals have been identified: Sperm whales produce highly directional echolocation signals characterized by low centroid frequency and very high peak-to-peak source level (SL) exceeding 235 dB_pp re 1 µPa @1 m, which enables them to echolocate deep-sea squid or other prey at relatively long range. Whistling delphinids use very short, broadband clicks with centroid frequencies above 60–80 kHz, and peak-to-peak SL of 210–228 dB. Beaked whales produce frequency- modulated clicks centered around 45 kHz. Peak-to-peak source levels are slightly lower than delphinid clicks, but due to their much longer duration, they contain comparable amounts of energy. Lastly, a polyphyletic assemblage of porpoises, six non-whistling delphinids of the Cephalorhynchus and Lagenorhynchus families, pygmy sperm whales (*Kogia sp.*), and the Franciscana dolphin (*Pontoporia franciscana*) all use Narrow Band High Frequency (NBHF) clicks where energy is concentrated in a narrow frequency band around 130 kHz. These animals seem to produce nearly as directional biosonar signals as delphinids, but at lower source levels. Despite the many studies quantifying sonar parameters for free-living, marine toothed whales, much less variation in signal type or biosonar parameters has been found compared to bats, especially among delphinids. However, most of the delphinids studied to date forage in habitats that may differ less acoustically than is the case for the different bat guilds. Instead it seems that an inverse scaling of frequency with body mass to achieve a similar directionality may be a major driving force across the toothed whale suborder. However, it is unclear how these selective pressures for high amplitude, high source level biosonar signals can be extrapolated to the acoustically complex, relatively shallow and turbid environments inhabited by river dolphins. To address this question, we studied two species of toothed whales that co-occur in waterways of the Sundarbans mangrove forest of Bangladesh. Irrawaddy dolphins (*Orcaella brevirostris*) are freshwater cetaceans living in shallow coastal waters, generally associated with freshwater inputs, as well as far upstream in three large, Indo-Pacific river systems. The extent of their inland range in the Sundarbans varies with seasonal freshwater regimes and may be influenced by the distribution of Ganges river dolphins. Ganges river dolphins (*Platanista gangetica gangetica*) are obligate freshwater dolphins found in the Ganges, Brahmaputra and Karnaphuli river systems where they exhibit a peculiar, side-swimming form of locomotion. The extent of their downstream range in the Sundarbans is also determined by seasonally dynamic freshwater flows, with the Ganges river dolphin favouring low salinity, high turbidity and moderate depth. Both Irrawaddy dolphins and Ganges river dolphins have relatively small bodies comparable to small marine delphinids and porpoises. In the Sundarbans, they inhabit geomorphically complex areas with extremely variable depth, salinity and turbidity in contrast to the more stable characteristics of marine environments. Given the complex acoustic environment and high amount of clutter and reverberation, it may be hypothesized that Irrawaddy dolphins and Ganges river dolphins employ echolocation signals characterized by low-amplitude, high frequency sonar signals emitted at high repetition rates like small bat species hunting in cluttered habitats. In this study, we quantify the biosonar source parameters of Ganges river dolphins and Irrawaddy dolphins to test this hypothesis. We show that these animals use consistently lower source levels and higher repetition rates than oceanic delphinids, possibly limited by high amounts of clutter and reverberation. We demonstrate that Ganges river dolphins have a slightly broader beamwidth than other toothed whales due to their very low centroid frequency but that they achieve a higher directionality than expected from a direct scaling with centroid frequency and size, possibly by using a novel set of bony plates in the forehead. We conclude this study by discussing means to use acoustics to help better understand the conservation needs of these highly endangered freshwater toothed whales. # Materials and Methods ## Study Area Recordings were obtained in the waterways of the Bangladesh part of the Sundarban mangrove forest where recording depths varied from 6.5 to 23 m, (mean 12.94 m). Recordings took place during daylight hours between the 4^th and16^th of February 2010 from a 12 m long, wooden research boat. All research was conducted under a research permit issued to the Bangladesh Cetacean Diversity Project of the Wildlife Conservation Society by the Ministry of Environment and Forest, Government of Bangladesh. ## Recording Equipment A vertical array of four Reson TC4034 spherical hydrophones (Reson A/S, Slangerup, Denmark) was formed by mounting hydrophones in a Perspex rod (4 cm diameter, hollow) with 0.75 m spacing. The first hydrophone was positioned at 2 m depth while the last hydrophone was at 4.25 m depth. A buoy was attached to the top of the array, and a 4 kg weight was fixed to the bottom to help maintain the array vertical in the water. Signals were amplified 60 dB by a custom-made amplifier and filter box (1 kHz 1-pole high-pass and 200 kHz 4-pole low-pass filter), then digitized by two synchronized National Instruments USB-6251 A/D converters (National Instruments, Texas, USA) at a sampling rate of 500 kHz per channel and a resolution of 16 bits. The calibrated clip level of the recording chain was 174 dB re µPa (peak), and the frequency response of the recording chain was flat (±2 dB) from 2–180 kHz. ## Data Collection Ganges river dolphins were recorded while foraging or resting at the convergences of channels. Irrawaddy dolphins were recorded during different behaviors (travelling, foraging, and socializing). The boat engine was turned off and the array was lowered into the water once the animals were within about 100 m of the vessel. Data acquisition was initiated and terminated manually and files were stored approximately every minute. Start and end time, position and depth were recorded for every recording event, as well as group composition and behavior. ## Click Analysis Signal analysis was carried out with custom-written routines in Matlab 7.5 (The Mathworks, Inc., Natwick, MA, USA). Each click series (also referred to in the literature as a click train) was examined visually and discarded if more than one animal was present to avoid underestimating interclick intervals. Echolocation clicks were then located on the third hydrophone using an automated click detector with a variable detection threshold chosen during visual inspection of waveforms to exceed the background noise level and detect individual click series. Each click was further analyzed only if detected on all four channels. ## Acoustic Localization Source location relative to the hydrophones was obtained through acoustic localization techniques based on time-of-arrival differences of the same click on the four receivers. To find the time of arrival differences, the signal recorded on the top hydrophone was cross-correlated with the signals recorded on the other hydrophones, excluding surface reflections. A sound speed of 1500 m/s was measured in each recording habitat by emitting pulses with a portable echosounder (Speedtech, Virginia, USA) at the position of the top hydrophone and cross-correlating to find the time-of-arrival at the remaining hydrophones at known distances. For each pair of hydrophones, the time-of-arrival difference can be explained by the equation for a single hyperbola in the two-dimensional plane of the array. Using four receivers, equations for three independent hyperbolas can be generated, and the position of the sound source found by solving the three equations with a least-squares method. Acoustic localization with this array was calibrated in Aarhus Harbour, Denmark, using artificial clicks (2 cycles at 70 kHz) generated by an omnidirectional HS70 hydrophone (Sonar Products) connected to a waveform generator (model 33220A, Agilent Technologies, California, USA). Pulses were emitted from a depth of 2 m and at distances from 5 m to 40 m from the array. Speed of sound during this calibration was calculated using the Leroy equation from measured temperature and salinity values. ## Source Parameter Estimation The interclick interval (ICI) was defined as the time between each click and the previous. Received levels were calculated as peak-peak (pp) and root-mean-square (rms) sound pressure levels within a time window given by the −10 dB end points relative to the peak of the amplitude envelope. The temporal duration of clicks was defined as the length of the −10 dB time window. The energy flux density was calculated for each click as the sum of squared sound pressure values within the −10 dB analysis window. Subsequently, the click power spectrum was calculated as the squared Fast Fourier Transform of a 32-point window centred on the peak envelope of each signal. The power spectrum was then normalized and interpolated with a factor of 100 using a low-pass interpolation. Peak frequency, centroid frequency (defined as the frequency separating the power spectrum into two halves of equal energy) and signal bandwidth (−3 dB power and −10 dB power) was calculated from this power spectrum. Source levels (SL) were defined as the back-calculated sound pressure level 1 m from the source on the acoustic axis, and calculated from received levels by compensating for the transmission loss (dB re. 1 m), estimated as the combination of spherical spreading and frequency- dependent absorption (taken at the centroid frequency of the received click) over the range from the source coordinates to the receiver. ## On-axis Criteria Off-axis signals are subjected to distortion. This means that it is essential to quantify the signal on or as close as possible to the acoustic axis when investigating source parameters of highly directional biosonar signals. With a linear array, the vertical angle of incidence can be estimated, but the horizontal angle of incidence is unknown. To maximize the likelihood of analyzing on-axis clicks, we selected only the highest-amplitude click in a longer click sequences (scans) with clicks of increasing and decreasing amplitude. These scans are most likely associated with the acoustic beam of the animal passing across the axis of the array. Assuming the animal maintains the same source level and directionality, the click with the highest amplitude has the highest likelihood of being on-axis in the horizontal plane. The criteria used to determine if the click was on axis is similar to that described in previous studies with similar arrays, : (1) the click could be localized; (2) the click had the highest received level in a scan (and thus assumed to be on- axis in the horizontal plane); and (3) the highest received level was recorded on one of the two central hydrophones, allowing for estimation of the angle of incidence in the vertical plane. ## Implications for Passive Acoustic Monitoring To evaluate the use of sound source parameters for passive acoustic monitoring studies without the potential for identifying on-axis clicks, a set of click series with only one clicking animal was identified. Each of these click series was passed through an automatic click detector (described above) to find accurate inter-click intervals for the two species. Subsequently, the power spectrum of each click was analyzed to find the centroid frequency. # Results Irrawaddy dolphins were recorded on 16 different occasions during a total of 9 hours, 58 minutes of recordings. The median group size encountered during recordings of Irrawaddy dolphins was 3 animals. During recordings, this species was observed while foraging and travelling. Ganges river dolphins (median group size 4 animals) were recorded in two different occasions and a total of 57 minutes of recordings were obtained from these encounters. In both recording occasions, the Ganges river dolphins were located in channel convergences. The hydrophone localization calibration indicated that clicks within 40 m were localized with a resulting error in the transmission loss estimates of less than 3 dB, which was deemed acceptable in accordance with previous studies. Consequently, only clicks recorded within a 40 m range of the hydrophone array were used for the analysis of the source parameters. A total of 15 Irrawaddy dolphin and 29 Ganges river dolphin clicks met the on- axis criteria and were recorded within the localization range of 40 meters. Only one click from each scan was used for analysis, and all recording areas were well separated to prevent recording the same groups of animals repeatedly. Clicks for both species were broadband transients similar to those of marine, whistling delphinids. Mean click duration ± SD was 13.4±3.0 µs for Irrawaddy dolphins and 21.7±2.2 µs for Ganges river dolphins, and Q ratios (defined as the ratio of centroid frequency to RMS bandwidth) was 3.2±0.3 (mean±SD) for Irrawaddy dolphins and 3.1±0.3 for Ganges river dolphins. Ganges river dolphin click source levels were significantly lower than the source levels of Irrawaddy dolphin clicks (Kruskal-Wallis: p\<0.0001). Peak-to- peak source levels (mean±SD) were 194.5±3.6 dB re 1 µPa at 1 m for Irrawaddy dolphins and 183.3±3.4 dB re 1 µPa at 1 m for Ganges river dolphins. For both species, these source levels are significantly lower (Kruskal-Wallis: p\<0.0001) than source levels produced by a marine delphinid, the Indopacific Bottlenose dolphin (*Tursiops aduncus*) recorded in a 5–8 m shallow bay (mean peak-to-peak source levels ± SD of 205±7 dB re 1 µPa at 1 m) and lower than published source levels from most other free-ranging toothed whales with the exception of some species producing narrow-band high-frequency clicks. Similarly, the mean source energy flux density was 136.3 dB re 1 µPa²\*s at 1 m for Irrawaddy dolphins and 126.6 dB re 1 µPa²\*s at 1 m for Ganges river dolphins. There was no significant relationship between the recording range and the source levels for either species (Kruskal-Wallis: p = 0.46 for Ganges river dolphins and p = 0.45 for Irrawaddy dolphins). The centroid frequency (mean±SD) for Irrawaddy dolphins was 94.6±9.7 kHz, with −3 dB bandwidth of 64.4±15.8 kHz. Ganges river dolphins had a significantly lower centroid frequency (mean±SD) of 61.4±4.9 kHz (Kruskal-Wallis: p\<0.001) and correspondingly also a significantly lower −3 dB bandwidth of 43.8±7.1 dB (Kruskal-Wallis: p\<0.001). Interclick intervals were measured for both species for all on-axis clicks. The ICI values for on-axis clicks were higher than ICI values measured across entire click series. Interclick intervals (mean±SD) for Irrawaddy dolphin on-axis clicks was 44.8±24.6 ms and for Ganges river dolphin on-axis clicks it was 35.0±18.4 ms. In addition, the ICI was measured for entire click series with good signal-to-noise-ratio (SNR) and only one clicking animal at a time. A total of 923 clicks across 41 click series were analyzed for the ICI values of Irrawaddy dolphins and 614 clicks across 25 click series for Ganges river dolphins. For the entire click series, ICI (mean±SD) for Irrawaddy dolphins was 33.5±13.5 ms, and for Ganges river dolphins it was 29.9±9.0 ms. To test the potential for species discrimination in passive acoustic monitoring, probability density functions for Ganges river dolphin and Irrawaddy dolphin centroid frequencies were calculated using means and standard deviations from this paper, and assuming a normal distribution. In addition, a normalized probability density function for the Yangtze finless porpoise species (*Neophocaena phocaenoides asiaeorientialis*) was calculated using peak frequency (comparable to centroid frequency for narrowband high frequency species) and standard deviations from Li et al.. An estimated best separation criterion of 72.5 kHz provided a theoretical 98.7% correct classification of Ganges river dolphin clicks and 98.9% correct classification of Irrawaddy dolphin clicks, whereas an estimated best separation criterion of 112.35 kHz provided 97.2% correct classification of Irrawaddy dolphins and 96.7% correct classification of finless porpoises. For off-axis clicks, spectral distortion increases low-frequency energy so centroid frequency estimates decrease. This meant that the classification of Irrawaddy dolphins decreased to 72.7% (N = 971) with the remainder being misclassified as Ganges river dolphins. Ganges river dolphins, in contrast, were successfully classified 99.2% of the time (N = 641). # Discussion The study of toothed whale biosonar signals has developed rapidly during the last decade. Most studies have focused on marine delphinids and have revealed consistent high amplitude, highly directional echolocation signals from these species. Here, we recorded two small toothed whale species inhabiting areas that are more acoustically complex compared to the open ocean environments of many delphinids to better understand the evolutionary factors shaping different biosonar parameters of echolocating toothed whales. Both species produce broadband echolocation clicks characterized by a short duration and a low Q ratio of centroid frequency to RMS bandwidth of around 3. A short, broadband echolocation click is characteristic of all whistling delphinids, as well as sperm whales. The family platanistidae is an ancient evolutionary lineage that diverged not long after physeteridae. Its use of short, broadband clicks corroborates the hypothesis that the echolocation signal evolved by the shared ancestor of toothed whales was a short, broadband click that gradually evolved towards higher frequencies as greater high-frequency hearing sensitivity co-evolved with the capacity for high-frequency sound production. Echolocating toothed whales normally wait until the echo from a potential target has been received before producing a new click, meaning that the interclick interval between clicks exceeds the two-way travel time plus a processing lag time. When animals are searching, the interclick interval may also reflect the limits of their environment, such as the back wall of a pool or for a deep- diving animal, the altitude above the sea floor where the animal is operating. The interclick interval is therefore often taken as a maximum estimate of the acoustic search range of an echolocating animal. The two animals studied here both had higher click repetition rates compared to Indo-Pacific bottlenose dolphins (*Tursiops aduncus*) and even higher click repetition rates than coastal harbor porpoises \[mean ICI: 80.5 ms, 47\] and riverine Yangtze finless porpoises \[mean ICI: 60.4 ms, 47\]. This indicates that both Irrawaddy dolphins and Ganges river dolphins were searching for prey within a shorter range than most other studied odontocetes. Concurrent with the higher repetition rates, the two species also produced echolocation signals with much lower source level compared to similar sized marine delphinids. Irrawaddy dolphins (mean source levels ± SD of 194.7±4 dB re 1 µPa pp at 1 m) and Ganges river dolphins (183.6±3.5 dB re 1 µPa pp at 1 m) echolocate at more than 10 dB to 20 dB (respectively) lower source levels than other small, oceanic delphinids such as free-ranging pygmy killer whales (*Feresa attenuata*), bottlenose dolphins (*Tursiops sp.*), white-beaked dolphins (*Lagenorhynchus albirostris*), spinner (*Stenella longirostris*) and spotted dolphins (*Stenella attenuata*), and dusky dolphins (*Lagenorhynchus obscurus*, max 210 dB pp). Common to these species is that they often forage in an environment where background noise is the limiting factor that determines how far away the faint echoes from prey organisms can be detected. In a noise- limited echolocation scenario, the echo-to-noise ratio increases proportionally with the source level so that a greater detection range can be achieved by increasing the amplitude of the outgoing signals. For many of these exclusively marine species, the detection range of sparse, patchily distributed prey is a crucial parameter for survival. Selection for a long detection range would therefore promote the evolution of high-amplitude echolocation signals within the constraints provided by the size of the animal, principally the dimensions, composition and biomechanics of the sound-generating nasal structures. The overall body size of many oceanic delphinids is larger than the animals studied here, and it is possible that this size difference could account for the lower source levels of our animals. Indeed, large echolocating animals tend to produce echolocation clicks at high source levels and scaling of source level with body size might explain the low source levels produced by small species such as dusky dolphins. However, Ganges river dolphins are about the same size as dusky dolphins and spinner dolphins and produce similar biosonar clicks (as characterized by short duration and low Q) but with a maximum measured source level of 191 dB re 1 uPa (pp), about 20 dB lower than the maximum measured source levels for the dusky dolphins. Irrawaddy dolphins are larger than both dusky dolphins and Ganges river dolphins yet produce source levels on average nearly 10 dB lower than dusky dolphins. Porpoises and other NBHF species have also been thought particularly adapted to coastal environments, and these species are mostly similar in size or smaller than the Ganges river dolphin. The longer duration of NBHF signals compared to broadband delphinid signals means that it is most appropriate to compare the click energy flux density between species. Source levels of porpoises are comparable to the two species recorded here, with source energy flux density (SL_EFD)for harbor porpoises (*Phocoena phocoena*) (mean SL_EFD: 137 dB re 1 µPa²\*s) similar to the source energy flux density of Irrawaddy dolphin clicks; Peale’s dolphins (*Lagenorhynchus australis*) with somewhat intermediate source levels (mean SL_EFD: 133 dB re 1 µPa²\*s); and Commerson’s dolphins (*Cephalorhynchus commersonii*) with source levels as low as Ganges river dolphin (mean SL_EFD: 125 dB re 1 µPa²\*s). However, while porpoises and other NBHF species resemble the two study species here both in size and source level, they echolocate at much higher peak and centroid frequencies around 130 kHz. These species have seemingly undergone evolutionary selection for a high-pass filtered biosonar signal, possibly to avoid predation from other toothed whales such as killer whales (*Orcinus orca*). Ganges river dolphins diverged out early in the evolution of *odontoceti*, and it is unlikely that these animals ever risked predation by killer whales. However, the NBHF signal type is a subsequently derived biosonar signal that comes at the cost of a smaller bandwidth and thereby presumably less information about the acoustic environment and it does not help explain why the two species in this study produce source levels below those of oceanic delphinids. One important challenge that these animals face is the task of locating and catching food in an acoustic habitat with high reverberation and clutter levels. Several studies have shown how close proximity to clutter or to the bottom may interfere with the detection of targets. Reverberation from the bottom will necessarily depend on signal frequency, grazing angle, bottom sediment type, and especially depth. The two species here both forage for sparse prey through relatively shallow environments (10–15 m in the Sundarbans). While it is difficult to quantify both underwater clutter and reverberation, it is reasonable to assume that a shallow, restricted river habitat provides more challenging acoustic conditions than the open ocean. Unlike a noise-limited situation, higher source levels do not help detect targets in either reverberation or clutter limited conditions, as the backscattered echo from clutter or bottom will be just as much greater as the echo from potential targets. In addition, forward masking of the outgoing click may play an increasingly important role for toothed whales echolocating at very close range. Consequently, we argue that the acoustic properties of the shallow-water habitat might have favored the use of clicks with relatively low source levels in Irrawaddy and Ganges river dolphins. If reverberation can play an important role in shaping the source levels of echolocating toothed whales, this might also explain the lower source levels found for the Indo-pacific bottlenose dolphins (*Tursiops aduncus*) in a shallow coastal habitat, compared to deep-water common bottlenose dolphins (*Tursiops truncatus*). While common bottlenose dolphins are capable of detecting a metal target on a sandy bottom at up to 70 m range despite the clutter caused by the environment, the typical prey of Irrawaddy dolphins and Ganges river dolphins constitute small fish and shrimp. The low target strength and varied bottom composition in shallow water may prove to be a more complex discrimination task for the animals than detecting high target strength, metal objects. While quantitative measurements of prey target strength and reverberation in different river habitats are needed to support this, we hypothesize that both Irrawaddy dolphins and Ganges river dolphins gain an advantage by using low source level clicks for detecting and discriminating small prey items in shallow-water, cluttered environments. This is not unknown among echolocating animals. Brinkløv et al. demonstrated that the long-legged bat (*Macrophyllum macrophyllum*) gradually decreased the source levels of its echolocation calls when operating in three increasingly cluttered environments. Clutter-imposed constraints from such habitats may have resulted in microchiropteran bats having specialized into guilds inhabiting different foraging niches, with longer detection range seemingly favored for open space foragers compared to bats hunting within dense vegetation. This situation may be paralleled for source levels of toothed whales: Oceanic delphinids use high source levels to find prey at long range in open areas; Irrawaddy dolphins utilize coastal habitats and venture upriver while using intermediate source levels for echolocation; and Ganges river dolphins, which diverged early from the remaining toothed whales and evolved in a spatially restricted freshwater habitat, received little advantage from long- range echolocation and use the lowest measured source levels best suited for echolocating prey at short range. It therefore seems that the selective pressures that have favored the evolution of high frequency, high source level biosonar signals in marine toothed whales cannot be extrapolated to the complex acoustic habitats of freshwater cetaceans. A central component in the high source levels of toothed whales is the production of a narrow echolocation beam through partial collimation of the acoustic energy. Evolution appears to have favored toothed whales with a high directionality index that seems to be remarkably similar across species, with horizontal −3 dB (half-power) beamwidths reported between 13.1 degrees for a harbor porpoise to 6.5 degrees for a beluga (*Delphinapterus leucas*) and 6.2 degrees for a false killer whale (*Pseudorca crassidens*). Large odontocetes (such as sperm whales or beaked whales) can achieve a certain directionality with lower frequencies than smaller whales (such as porpoises or small delphinids) , and this might explain the overall negative correlation between biosonar frequency and body size in toothed whales. From this relationship between body size and frequency, we would predict a relatively high centroid frequency of around 80–100 kHz for the moderately sized Irrawaddy dolphins and a higher centroid frequency of around 80–120 kHz for the small Ganges river dolphins. While Irrawaddy dolphins produced clicks with a relatively high centroid frequency (mean of 92 kHz), the Ganges river dolphins produced clicks with a surprisingly low centroid frequency (a mean±SD of 61.4±4.9 kHz) compared to their body size. Other toothed whales of similar size use biosonar centroid frequencies of around 70–85 kHz (Pygmy killer whales), 80 kHz (Hawaiian spotted dolphins and spinner dolphins), 90–100 kHz (Dusky dolphins) and around 130 kHz for the many NBHF species. The measured centroid frequency and the small size of the Ganges river dolphin would predict approximately half the directionality (6 dB smaller DI) and consequently a much broader beamwidth compared to delphinids and porpoises. Using equations derived from Au et al. and Madsen and Wahlberg (2007), the Ganges river dolphin should have a symmetric −3 dB beamwidth of some 20 degrees and a directionality index (DI) of some 19 dB. This prediction conflicts with findings reported in the only paper investigating the directionality of Ganges river dolphins: Bahl et al. reported that the −3 dB beamwidths of the Ganges river dolphins were in the order of 10 degrees in the horizontal plane and 14 degrees in the vertical plane. We find a similar, but slightly higher value, when fitting the data from Bahl et al. (2007) with a piston that best describes the variation in the data. The data indicate a single-lobed sound beam like all other toothed whales studied so far rather than the peculiar, double-lobed sound beam reported in the early literature. The best-fitting piston model provides a composite beamwidth of 14.5 degrees in the horizontal plane. Such a half power beamwidth corresponds to a DI of 22 dB which is comparable to or slightly lower than the half power beamwidth of harbor porpoises, but around 3 dB (50%) better directionality index than predicted from the low frequency clicks and the small head size of the Ganges river dolphin. Thus, somehow Ganges river dolphins seem to generate a beam directionality that, albeit slightly lower than most toothed whales, is comparable to that of similar sized toothed whales operating almost an octave higher in frequency. The reason for this apparent discrepancy might well lie in the unusual head anatomy of this species: Ganges river dolphins possess two unusual bony maxillary crests that project anteriorly over the facial region and virtually encircle the melon. They are asymmetrical and skewed to the left, and their ventral surfaces are dominated by a thin network of air sacs that seem to have grown dorsally from the pterygoid air sinus system. Purves and Pilleri and Pilleri and colleagues proposed that the crests might function in directing the sound from the melon. It is thus possible that these air-filled bony crests could help provide a better directionality than expected from scaling, and hence explain why Ganges river dolphins can produce clicks at centroid frequencies about an octave below what should be predicted from their size and still achieve a sufficient directionality. These findings support the notion that one of the evolutionary drivers for the echolocation click frequency in toothed whales is indeed directionality. The estimated beamwidth of Ganges river dolphins is still in the broad end of measured toothed whale biosonar beams. While this might be considered a more primitive condition, a slightly wider beam combined with the greater short-range maneuverability of these animals (a consequence of having completely free cervical vertebrae), may facilitate the capture of highly maneuverable prey items at close range throughout a shallow, cluttered rivers habitat. The significant difference in frequency content for these two species might be useful for acoustic species recognition such as seen in songbirds and other animals, and arguably also for some sympatric delphinids. Passive acoustic monitoring efforts may exploit such differences to locate critical species- specific hotspots for these endangered species. The three toothed whale species typically found in the coastal and river areas of the Sundarban National Forest include *Platanista gangetica gangetica*, *Orcaella brevirostris* and *Neophocaena phocaenoides*. The on-axis biosonar centroid frequencies of these species are well separated, and spectral parameters may be a promising way of both detecting and discriminating these animals acoustically. However, because biosonar signals are somewhat distorted when recorded off the acoustic axis, signals recorded away from the acoustic axis will have a lower frequency emphasis ( B and C). Applying the centroid frequency criteria that best separates on-axis clicks to a long series of clicks that would resemble what a passive acoustic monitor could record, results in clicks from Ganges river dolphins classified correctly nearly all the time (99.2% correct classification) whereas clicks from Irrawaddy dolphins were classified less successfully (72.7% correct classification). This results in some Irrawaddy dolphin clicks being incorrectly classified as Ganges river dolphins. The same degree of spectral distortion does not happen with NBHF clicks, whereby passive acoustic monitoring would be able to detect the presence of both finless porpoises and Irrawaddy dolphins reliably. Other criteria would be necessary to reliably classify Ganges river dolphins and discriminate such detections from off-axis Irrawaddy dolphins. One way of doing this would be to shift the separation criteria slightly upwards, and to use only the maximum centroid frequency for a series of clicks. For this dataset, reliable discrimination would be achieved based on the maximum frequency of 11–15 clicks and evaluated using a separation criterion of 74 kHz. In addition to spectral species discrimination, source levels presented here would be essential for estimating the detection function of an acoustic monitoring system, providing the basis for quantifying abundance of these threatened freshwater species. Acoustic monitoring has proven to be a powerful method for determining range, seasonality, and abundance of animals, and may prove essential for understanding the population parameters of cryptic, aquatic animals such as beaked whales, or finless porpoises. Freshwater dolphins all face significant extinction risks, primarily due to habitat loss and fisheries interactions, which led to the recent functional extinction of the Baiji (*Lipotes vexillifer*). Robust acoustic discrimination mechanisms that allows for monitoring of Irrawaddy dolphins and Ganges river dolphins could be especially helpful for managing protected areas such as the three new wildlife sanctuaries that were established by the Government of Bangladesh in the Sundarbans for the conservation of both species and provide better information that can help prevent a continued decline or extinction of these two threatened freshwater species. ## Conclusion Irrawaddy dolphins and Ganges river dolphins within the river systems of the Sundarban mangrove forest use high repetition rate, low source level echolocation clicks compared to marine species of similar size. Whereas obligate marine delphinids use high source level echolocation signals, Irrawaddy dolphins, inhabiting coastal and upriver habitats, produce lower source levels, with mean source levels of 194.7 dB (max 203 dB) re 1 µPa_pp and Ganges river dolphins, living exclusively in a shallow river habitat, produce even lower source levels of 183.6 dB (max 191) re 1 µPa_pp. The ultimate cause of these low source levels may be a relaxed selection for long- range echolocation inhabiting restricted, shallow, geomorphically complex river systems, with limits on echolocation range imposed by reverberation and clutter. Interestingly, the centroid frequency of the clicks used by Ganges river dolphins is almost an octave lower than expected from their size. The unusual, air-filled bony maxillary crests found in this species may compensate in part for this lower frequency by providing a larger effective baffle and hence a more directional sound beam than the biosonar frequency and head size would predict. The beamwidth of Ganges river dolphins is still wider than most other toothed whales, and it is possible that this may facilitate capture of highly maneuverable prey items in shallow water. Acoustic discrimination between freshwater odontocetes may facilitate acoustic monitoring efforts and may help prevent a continued decline of these two threatened freshwater species. This study was made possible through the logistical and field support of the Bangladesh Cetacean Diversity Project of the Wildlife Conservation Society. The study was conducted under a research permit issued by the Ministry of Environment and Forest, Government of Bangladesh. E. Fordyce, A. Galatius, J. Ososky, J. Mead, and C. Potter kindly provided pictures and helpful discussions of internal skull morphology. J. S. Jensen and N. U. Kristiansen provided technical support and K. Beedholm assisted with invaluable discussions and technical help. Finally, three anonymous reviewers offered helpful comments and constructive feedback to improve on the manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: FHJ AR PTM. Performed the experiments: FHJ AR RMM BDS. Analyzed the data: FHJ AR. Contributed reagents/materials/analysis tools: PTM VMJ. Wrote the paper: FHJ AR RMM BDS VMJ PTM.

# Introduction *P. marneffei* is considered an indicator pathogen for AIDS. It mainly exits endemically in area of South East Asia that causes fever, lymphadenopathy, hepatosplenomegaly and cutaneous lesions. *P. marneffei* has the unique feature among the species of *Penicillium* of being thermally dimorphic for diagnosis, it grows as a mycelium at 25°C, and a soluble red pigment is produced, whereas, at 37°C, it grows as a yeast form. The clinical features of *P. marneffei* in AIDS patients have been well described, and like other opportunistic pathogens, the infection of *P. marneffei* would exacerbate deterioration of the immune response and accelerate AIDS disease progression, while the mechanism remains elusive. Dendritic cells (DCs) play pivotal roles in host defense by initiating innate immunity and bridging adaptive immunity. DCs are widely distributed in the sub- mucosa, yet have been believed to be involved in HIV-1 infection and transmission. The migration property of DCs has been hijacked by HIV-1 for viral dissemination to CD4⁺ T cells by a process that is known as *trans*-infection. The formation of DC-T-cell conjunction, or so-called virological synapses, at which numerous intact viral particles and viral receptors can be recruited, appears to be required for efficient viral transfer. Upon activation by stimuli, such as the bacterial product LPS (lipopolysaccharide), DCs could uptake much more viruses and recruit significantly amounts of viral particles on the virological synapses for enhancement of HIV-1 *trans*-infection. DCs express HIV-1 receptors and can serve as targets for productive HIV-1 replication. Persistent infection of HIV-1 may generate the potential long-lived viral reservoirs in DCs. DCs appear to take vital roles in HIV-1 infection and viral pathogenesis, and a better understanding of the interactions between HIV-1 and DCs would facilitate the elucidation of AIDS pathogenesis. We hence isolated *P. marneffei* from the cutaneous lesions of AIDS patients and analyzed its effects on HIV-1-dendritic cells interaction. We found that MDDCs could be activated by both dimorphic forms of *P. marneffei* for significantly promoting HIV-1 *trans*-infection of CD4⁺ T cells, and the *Candida albicans* (*C. albicans*), which has been proved to possess the similar capacity, was used as control. Increased expression of intercellular adhesion molecule 1 (ICAM-1) was observed on fungus-activated MDDCs, and the tighter DC- T-cell conjunction recruited significant amounts of virus particles for viral transfer. We also found that *P. marneffei*-activated MDDCs efficiently activated resting CD4⁺ T cells through cell-cell contact and thereby could result in more susceptible targets for viral infection. Our findings demonstrate that DC function and its interaction with HIV-1 have been modulated by opportunistic pathogens such as *P. marneffei* for viral dissemination and infection amplification, highlighting the importance of understanding DC-HIV-1 interaction for viral immunopathogenesis elucidation. # Results ## P. marneffei stimulation promotes the activation of MDDCs In current study, the *C. albican* which has been described previously for induce DC activation was used as a control. *P. marneffe* and *C. albicans* were isolated separately from the skin lesions or the tongues of AIDS patients and cultured in Sabouraud agar plates. *P. marneffei* has the unique feature among the species of *Penicillium* of being thermally dimorphic, it grows as a mycelium at 25°C, similar to *Aspergillus* spp, and a soluble red pigment is produced (PMm-i and-ii), whereas, at 37°C, it grows as a yeast form (PMy). *C. albicans* was identified with sub-inoculation in CHROMagar Candida (, CA-ii), in Corn Tween agar (CA-iii) and with the API 20C AUX yeast identification system. These fungi were sub-cultured for amplification and harvested for heat inactivation. To investigate the potential activation of DCs by fungi, MDDCs were incubated separately with heat-inactivated *C. albicans*, PMy and PMm at a ratio of 1∶10 of DCs to fungi. The phenotypes of MDDCs were examined by immunostaining of cell surface markers; MDDCs showed high level of CD11c expression and were measured for CD83, CD86 and HLA-DR expression levels. Fungal stimulation significantly increased CD83, CD86 and HLA-DR expression compared with medium-treated cells, indicating of fungus-induced MDDCs activation, whereas surface expression of a C-type lectin, DC-SIGN (DC-specific intercellular adhesion molecule 3-grabbing nonintegrin), was decreased in fungus-treated MDDCs. These data suggest that stimulation of *P. marneffei* promotes DCs activation. ## HIV-1 infection of P. marneffei-activated MDDCs is blocked To examine the effects of the stimulation of *P. marneffei* on HIV-1 infection of MDDCs, fungus-treated MDDCs were inoculated with single-cycle, luciferase reporter HIV-Luc/JRFL (CCR5-tropic), and HIV-1 infection was measured by detecting the luciferase activity in cell lysates at 3-9 days post-infection. HIV-1 infection was fully blocked in all fungus-treated MDDCs compared with medium-treated controls, and the detected luciferase activities decreased by at least 75% at 5, 7 and 9 days post-infection. The *C. albicans*, which has been shown to inhibit HIV-1 replication in DCs, was used as a control. These data suggest that HIV-1 infection in MDDCs is blocked after the stimulation of *P. marneffei*. ## P. marneffei stimulation promotes DC-mediated HIV-1 transmission to CD4⁺ T cells To determine the capacity for DC-mediated HIV-1 transmission after *P. marneffei* stimulation, MDDCs were treated with heat-inactivated fungi as above and GFP-containing HIV-1 VLPs were used to measure viral transmission efficiency using flow cytometry. HIV-1 VLP-loaded MDDCs were co-cultured with human CD4⁺ T-cell line Hut/CCR5 for 30 min, Hut/CCR5 cells with the CD11c^- staining were gated, and Gag-GFP-associated cells were quantified. The level of GFP association on Hut/CCR5 cells increased from 11% in medium-treated controls to 27–30% in fungus-activated MDDCs, the MFI values showed an over 2-fold increase. Thus, the fungus, including *P. marneffei* and the *C. albicans* control, can promote MDDC-mediated HIV-1 transfer to CD4⁺ T cells. Viral *trans* infection was also quantified by using the DC-T-cell co-culture system as described previously. Pseudotyped single-cycle, luciferase reporter HIV-1 was used. Hut/CCR5 and activated autologous primary CD4 ⁺ T cells were used as the target cells. Differently treated MDDCs loaded HIV- luc/JRFL or HIV-luc/NL4-3 were co-cultured with target cells for 3 days, and HIV-1 infection was monitored by measuring luciferase activity. Fungus- stimulated MDDCs significantly enhanced the capacity to mediate HIV-luc/JRFL or HIV-luc/NL4-3 *trans* infection of HutCCR5 cells, there was a 4.8- to 6.5-fold increase in luciferase activity, when activated primary CD4⁺ T cells were used as target, fungus-stimulated MDDCs enhanced HIV-luc/NL4-3 *trans-*infection of primary CD4⁺ T cells approximately 2-3- fold. *C. albicans*, having been shown previously to promote DC-mediated HIV-1 transmission, was used as a positive control. Together, these results indicate that MDDCs stimulated by PMm and PMy enhanced their capacity to mediate HIV-1 *trans* infection of CD4⁺ T cells. ## Enhanced endocytosis and altered intracellular trafficking of HIV-1 in fungus-activated MDDCs LPS-activated DCs potently enhance HIV-1 *trans* infection and the endocytosis of large amounts of viruses, and the harboring of intact viruses in non- classical multiple vesicular bodies might provide viruses with a means to escape from intracellular proteolysis. To investigate whether fungal stimulation similarly affect on viral endocytosis, fungus-stimulated MDDCs were pulsed with HIV-1 VLPs for 2 h. Trypsin was used to strip cell-surface-bound virus particles, and the internalized viral particles were quantified by detection of Gag-GFP by flow cytometry. Numerous viruses were internalized by fungus- stimulated MDDCs. Gag-GFP was demonstrated in 53.5–57.3% of MDDCs stimulated by *C. albicans*, PMy or PMm, compared with 15.9% in medium-treated immature MDDCs, and the calculated MFI of GFP also exhibited a two-fold increase relative to that in medium-treated controls. The majority of MDDCs-associated viruses could not be removed by trypsin digestion. Blocking with anti-DC-SIGN antibodies before VLP pulsing did not inhibit viral uptake, suggesting a DC-SIGN- independent endocytosis process. By contrast, as shown previously, immature MDDCs bind HIV-1 mainly through surface-expressed DC-SIGN, which can be easily removed by trypsin treatment. To better understand intracellular trafficking in fungus-stimulated DCs, HIV-1 VLP-pulsed MDDCs were visualized by confocal microscopy with immunostaining. Fewer viruses were evenly distributed on the surface or were internalized in medium-treated immature MDDCs, whereas many viral particles were endocytosed and concentrated into the CD81⁺ DC-SIGN^- compartments in fungus-treated MDDCs. These results are consistent with previous observations that LPS-stimulated MDDCs sequester intact HIV-1 in a specialized and tetraspanin CD81⁺ compartments. These data suggest that the stimulation by PMy, PMm or *C. albicans* largely promotes HIV-1 endocytosis and sequestration within the tetraspanin CD81⁺ compartments of fungus- activated DCs. ## Fungus-activated MDDCs increase ICAM-1 expression, and facilitate DC-T cell contact formation and viral concentration in virological synapses The virological synapses have been demonstrated to provide the most efficient route for HIV-1 transfers between contacting cells. The ICAM-1-LFA-1 interaction has been proved to be involved in the formation of DC-T-cell conjunction and contribute to efficient HIV-1 transfer. To investigate further the mechanism by which fungal treatment enhances viral transfer, ICAM-1 expression on the cell surface was measured. Stimulation with heat-killed PMy, PMm and *C. albicans* enhanced ICAM-1 expression on MDDCs by 2.4- to 2.6-fold, which indicates the potential for tighter cell conjunction formation. The formation of virological synapses was visualized by confocal microscopy, and Hut/CCR5 or activated primary CD4⁺ T cells were used as target cells. Many viral particles were efficiently concentrated at the fungus-stimulated DC-T cell contact sites to form virological synapses. In more detailed analysis of the staining, the tetraspanin molecule of CD81 was recruited to the contact sites, which suggests a potential role of CD81 in HIV-1 trafficking. Taken together, these data demonstrated that the enhanced ICAM-1 expression and virological synapses account for increased HIV-1 *trans*-infection mediated by fungus-stimulated MDDCs. ## Fungi facilitate DC-induced activation of resting CD4 ⁺ T cells and promote viral infection by providing more permissive cell targets DCs can efficiently active naïve T cell, and the activated T cells can provide more permissive targets for robust viral infection. To examine the potential effects of fungi on DC-induced CD4⁺ T cell activation and HIV-1 infection of T cells, MDDCs were pulsed with heat-killed fungi or control medium for 2 h. After washing, MDDCs were co-cultured with allogeneic resting CD4⁺ T cells for an additional 48 h. T-cell activation was monitored by detection of CD69 expression in gated CD3⁺ cells. Overall, fungus- pulsed MDDCs facilitated T-cell activation. In the presence of fungi-pulsed MDDCs, CD69 was expressed on the surface of around 12% of T cells, compared with 4.8% of T cells cultured with MDDCs without fungi. In order to demonstrate the DC-T cell direct contact is requirement for efficient T activation, the transwell plates with an insert membrane size of 0.4 µm were used to separate the MDDCs from T cells. As expected, much less or non- activation of resting T cells was observed. Direct stimulation with heat-killed fungi alone induced very little, transient expression of CD69 on resting T cells, or no expression at all, which demonstrated the need for DCs for activation of resting T cells. We investigated whether fungus-loaded MDDCs can facilitate T-cell susceptibility to HIV-1 infection. T cells co-cultured with fungi-loaded MDDCs were purified and plused with pseudotyped sing-cycle HIV-Luc/NL4-3 reporter virus, and viral infection was detected 5 days later by measuring luciferase activity in cell lysates. HIV-1 infection was significantly enhanced in primary CD4⁺ T cells activated by fungus-pulsed MDDCs compared with control MDDCs without fungi. Moreover, direct cell-cell contact was required for initiating T-cell susceptibility to viral infection. Direct treatment of heat-killed fungi did not induce susceptibility of resting T cells to HIV-1 infection. PHA-activated CD4⁺ T cells were used as a positive control, which displayed around 30% cells expressed CD69 expression and supported efficient HIV-1 infection in treated T cells. These data suggest that fungus-pulsed DCs facilitate the activation of resting T cells and activate more permissive T cells targets for robust HIV-1 replication. # Discussion Microbial translocation has been proposed as the cause of systemic immune activation in chronic HIV-1 infection ; however, it has not been extensively studied how these co-pathogens speed up deterioration of the immune response. DCs appear to be the common targets for HIV-1 invasion and translocation of other opportunistic pathogens at the mucosa. The functional compromise of DCs by HIV-1 infection is associated with immunosuppression and lack of control of microbial translocation. Given the pivotal roles of DCs in host immunity and viral pathogenesis, the interactions of DCs with HIV-1 have been preferentially targeted for exploiting the potential effects of opportunistic pathogens. DCs treated with opportunistic pathogens, such as *Malaria hemozoin*, *Mycobacterium tuberculosis*, and *C. albicans*, impairs degradative processing and MHC-II presentation of HIV-1 antigens to CD4⁺ T cells, and alters cytokine secretion, the enhanced DC-mediated viral *trans* infection was also observed during certain opportunistic infections. In those published studies, the synthetic *hemozoin* products, the *M. tuberculosis* cell wall, or the heal- killed *M. tuberculosis* or *C. albicans* laboratory strains was used. Here, the effects of *P. marneffei* on HIV-1-DC interactions were investigated. The difference is that the used *P. marneffei and the C. albicans* were directly isolated from AIDS patients. Our results demonstrated that both thermally dimorphic forms of *P. marneffei* activated DCs and promoted DC-mediated HIV-1 *trans*-infection of CD4⁺ T cells. Moreover, *P. marneffei* -stimulated DCs could further activate resting CD4⁺ T cells to induce more susceptible targets for HIV-1 infection. Our results have also shed light on the detailed mechanisms for the enhancement of viral spread. We demonstrated that heat-killed *P. marneffei*, along with *C. albicans*, promote viral uptake in MDDCs, altered viral intracellular sequestration, and importantly, facilitated MDDC-T cell contact by increasing ICAM-1 expression and efficiently concentrating HIV-1 particles in virological synapses. DC activation and altered viral intracellular trafficking are associated with enhanced viral spread. Upregulation of HLA-DR, costimulatory molecules CD83 and CD86, and intercellular molecules on fungus-activated DCs, in general, would encourage DC-T cell conjugate formation. We and other groups have previously reported that increased ICAM-1 expression on DCs correlates with promoted viral transfer, due to stronger DC-T cell interactions through ICAM-1 binding to T-cell-expressed LFA-1. Fungus-stimulated DCs accelerate viral uptake and sequestrate intact viral particles in non-conventional, non-lysosomal tetraspanin CD81⁺ compartments. The harboring of intact virus into the non-classical multiple vesicular bodies might provide virus a means to escape from the cellular proteolysis. Upon encountering with CD4⁺ T cells, more viruses were recruited on the DC-T cell contacted sites. High levels of endocytosis and altered intracellular trafficking of HIV-1 appear to account for enhanced viral transmission mediated by fungus-activated DCs. *P. marneffe*-stimulated DCs were less permissive for productive infection, which is consistent with previous reports of LPS and *malaria hemozoin* treatment. However, it remains to be clarified which fungal component(s) is responsible for HIV-1 restriction and the underlying mechanisms. LPS-matured DCs show dis-association of the susceptibility for HIV-1 infection with the capacity for mediating HIV-1 *trans* infection. Post-entry restriction of HIV-1 infection in LPS-induced mature DCs has been noted, and inhibition on the levels of reverse transcription and post-integration have been further identified by using real time PCR quantification of viral DNA and integration. Reduced gene expression, such as for co-receptor CCR5, has been reported to be responsible for impaired productive infection of HIV-1 in *malaria-hemozoin-*treated DCs. Higher levels of APOBEC3G and APOBEC3F (for “apolipoprotein B mRNA-editing enzyme, catalytic polypeptide-like 3G and 3F”) also have been shown to mediate the post-entry block of HIV replication in DCs and LPS can upregulate the expression of APOBEC3G/F,. The antiretroviral protein, namely SAMHD1 (SAM domain HD domain-containing protein 1), has been recently identified to inhibit the early step of HIV-1 replication in dendritic- and myeloid cells. It might be possible that fungus-treated DCs increased the expression of these HIV-1 restriction factors and therefore become more resistant to HIV-1 infection. DCs activate resting T cells and can provide more permissive targets for HIV-1 infection. We found that the stimulation of *P. marneffe* significantly accelerated DC-induced activation of resting CD4⁺ T cells, which indicates the pivotal importance of DC-driven T-cell activation for the high level of viremia and exacerbation of T-cell depletion in the late stage of HIV-1 infection. It would be interesting to confirm these *in vitro* observations in HIV-1-infected individuals. Our findings revealed that DC function and its interaction with HIV-1 have been modulated by opportunistic pathogens for viral dissemination. Enhanced HIV-1 spread by DCs can target activated CD4⁺ T cells, which could further accelerate T-cell depletion and immunosuppression, leading to the lack of control of both viral and fungal pathogens. Our results highlight the importance of studying DC-HIV-1 interactions for understanding viral pathogenesis, and might provide a new insight into the interventions against HIV-1 infection and spread. # Materials and Methods ## Ethics statement and fungi isolation and identification This study was reviewed and approved by the Medical Ethics Review Committee of Yunnan Province, Kunming, China. Written informed consent was provided by study participants and/or their legal guardians. Fungi were isolated from skin lesions or tongue of AIDS patients and cultured on Sabouraud agar plates. Fungus species were identified, sub-cultured for amplification, then harvested and killed by boiling for 1 hr. Fungal cell counts were determined under a light microscope and diluted at 1×10⁸/ml in PBS. MDDCs were stimulated with fungi for 48 hrs at a 1∶10 ratio of cells. ## Cell culture Human peripheral blood mononuclear cells (PBMCs) from healthy donors were provided by the Blood Center of Shanghai, Shanghai, China. CD14⁺ monocytes and resting CD4⁺ T cells were purified from PBMCs using magnetic beads (BD Biosciences) as described before. CD14⁺ monocytes were cultured with granulocyte-macrophage colony-stimulating factor and interleukin (IL)-4 for 6 days to generate the immature DCs. Resting CD4⁺ T cells were activated with 5 µg/ml of phytohemagglutinin-P (PHA-P) (Sigma-Aldrich) for 48 h in the presence of 20 IU/ml of recombinant IL-2 (R&D). The human embryonic kidney cell lines HEK293T and the CD4⁺ T-cell line Hut/CCR5 (kind gifts from Dr. Vineet KewalRamani, National Cancer Institute, USA) have been described previously. ## HIV-1 or virus-like particle stocks Pseudotyped single-cycle HIV-1 stocks were generated by using calcium phosphate- mediated co-transfection of HEK293T cells with the plasmid pLAI-Δ-env-Luc and expression plasmids of either JRFL (R5-tropic) or NL4-3 (X4-tropic) envelope glycoproteins as described previously. Virus like particles (VLPs), HIV-1-Gag- GFP/JRFL, were generated by cotransfection of HEK293T cells with a plasmid encoding HIV-Gag-GFP and with an expression plasmid of JRFL (kind gifts from Dr. Vineet KewalRamani, National Cancer Institute, USA). Harvested supernatants of transfected cells that contained HIV-1 particles were filtered and titrated with p24^gag capture ELISA. ## Flow cytometry Cells were stained with specific monoclonal antibodies (McAbs) or isotype- matched IgG controls. McAbs against the following human molecules were used for staining (clone numbers and resources are given in parentheses), Phycoerythrin (PE)-conjugated CD3 (UCHT1; eBioscience), PerCP-cy5.5-CD3 (OKT3, eBioscience), PE-CD11c (3.9; eBioscience), APC-Alexa Fluor750-CD11c (B-ly6; BD Pharmingen), PE-ICAM-1(CD54) (HA58;eBioscience), PE-CD69 (FN50; eBioscience), PE-CD83 (HB15e;eBioscience), PE-CD86 (IT2.2; eBioscience), PE-DC-SIGN (eB-h209; eBioscience), and APC-cy7-HLA-DR (LN3; eBioscience). Stained cells were detected with an LSRII flow cytometer (BD Pharmingen) and analyzed with FlowJo 7.6.1 software. ## HIV infection and transmission assays The luciferase reporter system was adopted for assay of HIV-1 infection and transmission as previously described. In brief, MDDCs were pulsed with 5 ng p24^gag amounts of pseudotyped HIV-luc/JRFL or HIV-luc/HXB2 for 2 h, and cells were washed for culture or for co-culture with CD4⁺ T cells. Hut/CCR5 or PHA-P-activated primary CD4⁺ T cells were used as targets. Cells were harvested after 3 days post-infection, and the cell lysates were measured for luciferase activity with a commercially available kit (Promega). DC-mediated HIV-1 transmission also was detected by flow cytometry. HIV-1 VLP (HIV-Gag-GFP/JRFL) was used, and after 1-h co-culture of virus-loaded MDDCs with Hut/CCR5 cells, the T cells were distinguished based on CD11c^- staining from the mixed cell culture, and the numbers of Gag-GFP associated with T cells were measured. ## HIV-1 binding and internalization assay HIV-1 binding and internalization were quantified by flow cytometry using the VLPs (HIV-Gag-GFP/JRFL). MDDCs were incubated with 40 ng p24^gag amounts of viruses for 2 h at 37°C and washed. The numbers of Gag-GFP associated with MDDCs were quantified by flow cytometry, and mean fluorescence intensity (MFI) was calculated. When indicated, anti- human DC-SIGN specific McAbs (10 µg/ml) (120507; Abcam) were used for pre-incubation with MDDCs before viral pulsing, or 5 min treatment with 0.25% trypsin (without EDTA) was used after viral-loading to remove surface-bound HIV-1. ## T-cells activation assay and viral infection Various fungus-stimulated MDDCs or heat-killed fungi species were used to coculture with or treat allergenic resting CD4⁺ T cells for 48 h at the same ratio of cells. The T cells were gated based on CD3-positive populations, and the activation was monitored by detecting the transient expression of CD69 by flow cytometry. For viral infection, the activated T cells from DC-T cell co-cultures were purified by magnetic beads and then challenged with 5 ng p24^gag amounts of HIV-1/NL-43 for 2 h. After washing, the cells were further cultured for 5 days, and HIV-1 infection was detected with luciferase activity assay as mentioned above. Transwell culture plates with a membrane size of 0.4 µm were used to separate MDDCs and T cells. ## Confocal microscopy and image analysis HIV-1 intracellular trafficking and the formation of virological synapses were observed by confocal microscopy. MDDCs were pulsed with HIV-1 VLPs, or virus- loaded MDDCs were co-cultured with CD4⁺ T cells for 30 min. Cells were seeded on the poly-L-lysine coated microscope slides and fixed with 4% paraformaldehyde (Sigma-Aldrich) for 1 h at 4°C. Cells were immunostained with anti-CD81 McAbs (M38; Abcam), anti-DC-SIGN McAbs (120507; R&D system), or anti- βactin (AC-15, Sigma), followed by staining with Alexa-Fluor 555-labeled goat anti-mouse IgG (Invitrogen). Nuclei were stained indicated with DAPI. Slides were mounted with Fluorescent Mounting Medium (Dako) and observed using a laser scanning confocal microscope (Leica SP5). ## Statistical analysis Statistical analysis was performed using paired *t* test with the SigmaStat program. We thank Dr. Vineet KewalRamani for generous reagents. [^1]: Conceived and designed the experiments: YQ KL JW. Performed the experiments: YQ YL WL RT QG. Analyzed the data: YQ YZ KL JW. Contributed reagents/materials/analysis tools: YL SL HL DZ LW. Wrote the paper: LW JW. [^2]: The authors have declared that no competing interests exist.

# Introduction ## Background The SARS-CoV-2 (COVID-19) pandemic is posing major challenges for health care systems across the world. Throughout the pandemic, the primary goal has been to protect the population from infection and provide medical care for infected persons. In the first peak of infections in spring 2020, the German federal government, in consultation with the federal states, enacted containment measures for the general public, including social distancing, the isolation of positive or suspected cases, a ban on admissions to nursing homes and a ban on visitation in hospitals, nursing homes and hospices. Although these were helpful strategies to reduce infection and mortality, by March 2021, more than 70,000 people had died from or with COVID-19 in Germany. Throughout the pandemic, people have continued to require end-of-life care for cancer and other advanced chronic diseases. However, in Germany as all over the world, adequate palliative care for the severely ill and dying (with and without COVID-19), and their loved ones, has not been available at all times and in all settings during the pandemic. This has caused psychological, social and spiritual distress for patients, thereby compromising their quality of life. Palliative care aims at maintaining patients’ quality of life. It can be provided on at least two levels: general and specialist. When administering palliative care, GPs maintain close contact with patients and their relatives. Frequently, they form significant and long-term relationships with patients, initiating and coordinating care and further treatment with other health care providers. Thus, in Germany, GPs represent a key provider of general palliative care, and the COVID-19 pandemic has served to underline their significance in this respect. ## Study aim This paper aims at describing GPs’ experiences, challenges and perspectives relating to general palliative care during the COVID-19 pandemic in Germany. # Materials and methods The present study, based on an online survey with GPs in Germany, is part of the German collaborative project "National Strategy for Palliative Care of Severely Ill and Dying People and their Relatives in Pandemics (PallPan) in Germany," led by the National Research Network of University Medicine on COVID-19. PallPan aims at developing and achieving consensus on a national strategy for the care of seriously ill, dying and deceased adults (with and without COVID-19) and their relatives during a pandemic. ## Pre-study Recent subjective field experiences were explored through two informal conversations with resident GPs in July and August 2020. The topics that emerged in these conversations were further discussed and explored in September 2020 within an online focus group involving three GPs, as well as telephone interviews with two GPs. The focus group and interviews were audio-recorded and transcribed verbatim. Main reported experiences and challenges were for instance limited home visits, restricted physical closeness to patients and less visits from relatives, less body-related therapies, and an increased isolation of patients. ## Survey development and pre-test Between September and November 2020, a standardized questionnaire was developed using the synthesized findings from our pre-study. The questionnaire was pre- tested by six GPs using the online survey tool Unipark, with special attention paid to the questionnaire’s structure and coherence, comprehensibility, technical aspects and duration. Information was collected on the following sections: 1. sociodemographic data on the study population; 2. patient contact; 3. telephone contact; 4. video consultation; 5. cooperation with other health care providers; 6. psychosocial aspects; and 7. needs and suggestions for managing end-of-life care in the context of a pandemic. The questionnaire used 4- and 5-point verbal rating scales (i.e. *totally agree*, *rather agree*, *rather disagree*, *fully disagree*) to determine the extent of (dis)agreement with the presented statements, which reflected subjective experiences, challenges and perspectives pertaining to end-of-life care during COVID-19. Free-text options were also provided to allow for respondents’ comments on their individual provided statements. ## Recruitment of the study population In November and December 2020, the information and invitation letter, including a direct, non-personalized link to the GP survey in Unipark, was sent to: 1. nine university institutes for general practice in seven federal states (Mecklenburg-Western Pomerania, Berlin, Lower Saxony, Hessen, North Rhine- Westphalia, Baden-Wuerttemberg, Bavaria), for distribution to their teaching and research networks; 2. three GP Associations in Lower Saxony and Bremen; 3. the German College of General Practitioners and Family Physicians; 4. the German Association for Palliative Medicine; and 5. the Competence and GP Training Center of Lower Saxony, for distribution to their members. The research team had no direct access to the distribution lists of the abovementioned parties and cannot quantify the number of GPs who were contacted. However, we estimate that the survey was distributed to at least 3,000 GPs. In the invitation letter, participants were asked to forward the letter to other interested parties, thereby triggering a snowball effect to maximize the study population. Each participant was asked to complete the questionnaire only once. Participation was completely anonymous. The survey was open from November 23 to December 18, 2020. ## Data analysis The SPSS 26 statistical software package was used to calculate descriptive statistics (mean value, standard deviation, minimum, maximum) and the absolute and percentage frequencies of the questionnaire data. Outliers were treated with the full dataset. Missing data are reported explicitly. The qualitative analysis of the pre-study data and free-text comments was based on content analysis (according to Kuckartz), using MAXQDA version 18. The main categories of the qualitative interview guide were used as the basis for the questionnaire content domains. ## Ethical requirements A written positive ethics vote (No. 9232_BO_K\_2020 of 24.07.2020) for the project was issued by the Ethics Committee of the Hannover Medical School. Prior to a subject’s participation in the online questionnaire, the participants had to confirm a check box that they have read and understood the written informed consent form concerning ethics and data protection and accept the regulations. Without this confirmation the participation was not possible. # Results ## Sociodemographic data on the study population The survey was completed by 410 GPs, comprising an approximately equal number of women and men. Their average age was 54 years (range 31–73 years), and they represented all 16 federal states in Germany. Approximately half of the GPs (51.5%) had completed additional training in palliative care. On average, they required 23 minutes to complete the questionnaire. Almost all of the GPs were experienced palliative care providers and reported that they had seen patients with COVID-19 in their practice. ## Experiences and challenges during the pandemic The following results for patient contact, telephone contact, video consultation, cooperation with other health care providers, psychosocial aspects, and needs and suggestions for end-of-life care in the context of a pandemic refer exclusively to GPs’ experiences caring for severely ill and dying patients during the first peak of the pandemic, in spring 2020. ### Patient contact The majority of respondents assessed the quality of their patients’ end-of-life care as consistent (61.5%), while 36.8% reported a decrease in quality relative to the pre-pandemic period. Of the GPs who made private home visits to severely ill and dying patients, 61.4% reported a consistent number of visits, 28.5% fewer home visits, and 9.1% more home visits (1.1% missing data) compared to the pre-pandemic situation. Similar results were found with respect to nursing home visits. The most frequently cited reason for reduced visits was restricted access. ### Telephone contact In total, 62.7% of the GPs reported increased telephone contact to replace personal contact with severely ill and dying patients. Of these, 36.2% indicated that the quality of end-of-life care worsened due to the lack of personal contact, because there were no physical examinations, communication was challenged and they were less able to provide emotional support. Similar problems were reported for telephone contact with relatives. ### Video consultation In total, 36.1% of the GPs offered video consultation in lieu of face-to-face contact with severely ill and dying patients and their relatives, which was generally only realized in individual cases by primary care physicians at private homes and nursing homes. Many GPs stated that video consultation was not used or requested by severely ill and dying patients. Their cited reasons for this included technical difficulties, lack of user competence on the part of the patient and poor quality of care. Video consultation was, however, used to support relatives of severely ill and dying patients. ### Cooperation with other health care providers The respondents rated their cooperation with other GPs, community nursing services and nursing homes as good. In contrast, they evaluated their cooperation with physio, occupational and other therapeutic professions, medical specialists, hospitals and health authorities as satisfactory, and thus slightly worse. Local health authorities, described as overburdened, were criticized primarily for their lack of accessibility. The GPs described nursing homes’ hygiene concepts as inconsistent. In the free- text fields, some GPs (n = 15) reported that they were challenged in their attempts to access nursing homes. The main reason for this was that nursing home stakeholders were uncertain of how to interpret and apply hygiene-related contact restrictions. In addition, the GPs also reported problems admitting severely ill patients to nursing homes. According to the GPs evaluation, many relatives could have been restricted (48.5%) or prohibited from visiting (33.4%) patients in nursing homes. The GPs perceived deterioration in the physical and mental health of patients in private homes and nursing homes as a consequence of this restricted contact. The GPs also perceived that relatives saying goodbye to their loved ones was only possible to a very limited extent (91.7%) or not at all (56.1%). ### Psychosocial aspects of severely ill and dying patients and their relatives The GPs observed an increased fear of loneliness among severely ill and dying patients in nursing homes (91.9%), private homes (87.3%) and hospitals (86.1%). With regard to the psychosocial burden on relatives, the majority of the GPs reported increased distress due to relatives’ reduced receipt of information about the patient (85.9%) and inability to support them with their physical presence (99.3%). ### Needs and suggestions for end-of-life care in the context of a pandemic The GPs identified social contact with relatives and face-to-face contact with physicians as the most important aspects of patient care. In total, 92.4% of the GPs (fully/rather) agreed that GPs should be involved in local crisis teams, and 79.5% (fully/rather) agreed that palliative care physicians should also be involved. These local crisis teams were imagined to improve the exchange between outpatient and inpatient care providers and facilitate efficient, decentralized coordination and decision making at the local level. # Discussion The present study administered a nationwide online survey to collect GPs’ experiences, challenges and perspectives with respect to caring for severely ill and dying patients and their relatives during the COVID-19 pandemic. Almost all of the participating GPs had treated patients with COVID-19 in their practice. Throughout the pandemic, despite many efforts to adapt their individual practice management, the GPs felt challenged in their ability to administer high quality palliative care. Of note, the GPs reported deterioration in patients’ physical and mental health in both private and nursing homes, due to contact restriction. This concerning trend has also been observed by the German Association for Palliative Medicine and other professional organizations. In the present study, the GPs reported an increased fear of loneliness in their patients, as well as greater psychological distress in patients’ relatives, due to an inability to support their loved ones in person or to say goodbye. Girum et al., in a systematic review of 22 studies, demonstrated that quarantine and isolation measures have been effective in controlling the spread of COVID-19. Thus, protective measures (e.g. social distancing) are recommended, especially for those at greater risk of infection. However, while these measures may be important for managing the wider spread of the pandemic, the present study and other research has highlighted their serious physical and psychological consequences for severely ill and dying patients. To prevent these negative consequences, the GPs in our study recommended that social contact be maintained for patients receiving palliative care. The long-term effects of contact restriction and isolation on vulnerable groups must be investigated in future studies. Germany’s Federal Government Commissioner for Long-Term Care and the Federal Minister of Health addressed this challenge in December 2020, proposing regulations for visitation to care facilities. They emphasized the central role of social relationships for residents and listed basic measures to enable relatives to safely visit during the pandemic, with as few restrictions as possible. In contrast, the German College of General Practitioners and Family Physicians advised, in its “Action Recommendation on the New Coronavirus,” reduced face-to- face contact between GPs and nursing home residents, where possible. A reduction in home visits across both private and nursing homes was confirmed by approximately one-third of our GP respondents. In the event of reduced home visits, the German College of General Practitioners and Family Physicians recommended that GPs should reduce personal patient contacts, but engage in telephone and video consultation. They also recommended that these methods be widely applied in GP practices, to ease pressure. Our survey showed that this occurred in almost two-thirds of our respondents’ practices, with the GPs increasing the frequency of their telephone consultations with patients relative to the pre-pandemic period. Saint-Lary et al. showed a similar trend for increased use of telephone communication with patients in their observational survey with French GPs. In the present study, video consultation was offered by slightly more than one- third of the GP practices, for individual treatment. In contrast, another study found that 81% of GPs in Norway offered video consultation and found it suitable for maintaining patients’ continuity of care. In our study, the GPs connected their minimal use of video consultation to their reservations about technical implementation, user competence and reduced quality of care (due to a lack of physical examination). Similar attitudes were found by Randhawa et al., in their qualitative study of 12 GPs in London. Hawley et al. and Lieneck et al. reported further barriers to the use of this technology, including uncertainty among patients, concerns over data protection and lack of access to mobile devices among older patients. While reservations towards video consultation should not exclude the expansion of digital communication in health care, they reveal the need for training, broader implementation in nursing homes and clarity around data protection. In its draft “Digital Care and Nursing Modernization Act” of January 20, 2021, the Federal Cabinet addressed some of the abovementioned technical and infrastructure issues. As further guidelines on digital communication are developed, they must consider the views and experiences of health care providers, patients and their relatives. Based on their research in Atlanta, Kuntz et al. see great potential for video communication with relatives of palliative care patients. Drawing on data from their online survey with 67 caregivers and 10 semi-structured telephone interviews, the authors evaluate digitally-mediated family meetings as feasible and efficient. In addition, they conclude that video meetings might allow relatives to understand both the health and the care of the patient and to express their thoughts and feelings. ## Limitations As we did not have access to the distribution lists used for our survey, we cannot comment on the response rate or non-responder characteristics, and we cannot fully exclude the possibility that some individual GPs participated in the survey more than once. Furthermore, due to the cross-sectional study design, we can only provide data related to changes over time during the pandemic. Since this survey was based on the GPs’ recall of their experiences, there is the potential for recall and confirmation bias. Finally, it can be assumed that, among the study participants, GPs with a particular interest in palliative care were disproportionately represented. Therefore these findings may not be fully representative of primary palliative care and end of life care provision by GPs across Germany. # Conclusion The present work provides insights into the nationwide pandemic management of a representative group of GPs in Germany. The findings may support the development of a national strategy for palliative care during a pandemic. We conclude that, during a pandemic, the preservation of face-to-face visits by relatives and the development of feasible and safe video communication should be prioritized. Finally, to address end-of-life care issues appropriately, GPs and palliative care specialists should be involved in COVID-19 task forces on the micro, meso, and macro levels of health care. We thank all of the GPs who participated in the survey. We also acknowledge Valerie Appleby’s excellent editorial scrutiny of the present research article. [^1]: The authors have declared that no competing interests exist.

# Introduction Diabetes mellitus is one of the most common chronic diseases worldwide, with an increasing incidence in most countries. There were more than 382 million people with diabetes mellitus in 2013 and this is forecasted to reach 592 million people by 2035.\[–\] Most of the deaths among patients with diabetes are not due to diabetes itself but due to the complications associated with it. This includes cardiovascular disease which can cause about 50% of the deaths among patients with diabetes. In Saudi Arabia, diabetes mellitus has become an overwhelming health problem with an overall prevalence among adults of approximately 23.7%. The complications that are usually associated with diabetes mellitus are due to macrovascular or microvascular disease.\[–\] The prevalence of these complications in Saudi Arabia is quite high. In one study done in that country, the prevalence of complications were: myocardial infarction (14.3%), retinopathy (16.7%), acute coronary syndrome (23.1%) and nephropathy (32.1%). Erectile dysfunction (ED) is a common association of diabetes and is caused by a neuropathy or vasculopathy.\[–\] ED is defined as “the persistent inability to attain and maintain an erection that is sufficient to permit satisfactory sexual performance”. There is some evidence that suggests that low testosterone might be involved in both the development of type 2 diabetes and the subsequent complication of ED. Several epidemiological studies have shown that both type 1 and type 2 diabetes are associated with higher risks of ED. Also, it has been recognized that ED can be found even in preclinical or newly diagnosed diabetes. The prevalence of ED among men with diabetes ranges from 35% to 90% depending on the method used to identify it. In Saudi Arabia, studies have shown that the overall prevalence of ED in men with diabetes range from 63.5% to 83%.\[–\] There is a threefold increased risk of ED in men with diabetes compared to men without diabetes. Furthermore, even after adjusting the risk of ED for age in men with diabetes, the risk is still double compared to those without the disease. Moreover, ED in men with diabetes occurs 10–15 years earlier than in men without diabetes. It has been shown that quality of life is reduced in men with diabetes who are suffering from ED. In addition, ED is considered a predictor for cardiovascular events and can be associated with silent myocardial ischemia among men with type 2 diabetes mellitus.\[–\] In addition to that, ED in people with diabetes can be the first sign of future cardiovascular events. The existence of ED in men with diabetes is a reason to screen for other diabetic complications caused by microangiopathy in target organs. Doctors can diagnose ED by several methods. A detailed medical history and physical examination can give a good idea about its causes or degree of severity. Obviously, the most important step to start with is to simply ask men with diabetes about this problem during a routine clinical review. The UK NICE guidelines for diabetes mellitus recommend that “Review the issue of erectile dysfunction with men annually”. Surprisingly, few doctors ask men with diabetes about ED and this problem is frequently overlooked. For example, in one study done in England only 9% of men with diabetes were asked by their physicians about ED. In another study done in the United states, physicians initiated the discussion about ED with only 18% of their patients with diabetes. On the other hand, very few studies have addressed the issue of the discussion of ED from the perspective of the patient with diabetes, for example, whether they are willing to be asked about ED by their physicians. For example, a study done in Taiwan showed that 56.6% of patients with diabetes wished to discuss ED with their physicians. The literature indicates that barriers that might prevent health professionals from asking about sexual problems such as ED include lack of time or knowledge, lack of training among physicians, false beliefs about sexuality, thinking that this is a job for another physician, patients not being ready to discuss these issues, believing it is not an important subject, fear of increasing patient anxiety and patients being too ill or too old to be asked. In the Middle East, the literature is lacking in studies on the proportion of men with diabetes who have been asked about ED or their willingness to discuss ED with their physicians. Also, in searching the literature, no studies were found that described the barriers faced by patients with type 2 diabetes to discuss ED with their physicians. This study was aimed primarily to find out the proportion of Saudi men with type 2 diabetes who have been asked about ED in the last year by their physicians in hospital-based primary care clinics in Riyadh. We also aimed to determine the willingness of Saudi men with type 2 diabetes to discuss ED with their physicians and the factors that either increase or reduce their willingness to discuss this issue. # Methods and materials ## Study design This study employed a cross-sectional survey using a quantitative self- administered questionnaire investigating the proportion of Saudi men with type 2 diabetes who have been asked by their physicians about ED in the last year, their willingness to discuss ED with their physicians, and the factors that may be related to their willingness to discuss ED with their physicians. ## Study site The study was conducted in hospital-based primary care clinics at King Khalid University Hospital, Riyadh, Saudi Arabia. These primary care clinics consist of 8–10 clinic sessions each day, led by approximately 40 staff specialized in family medicine. Each clinic has about 30 patients daily with the majority of the patients having diabetes. ## Eligibility criteria The participants were included in this project if they were married, adult (i.e. \> 18 years), diagnosed with type 2 diabetes mellitus, having at least one year of follow-up in the clinics, and could read and write Arabic. We excluded any participant with anatomical penile deformities, past history of spinal cord injury or past history of prostate diseases or prostate surgery. ## Patient enrolment Patients were approached at the reception desk after they completed their review with their physician and asked about the inclusion and exclusion criteria. Those who met our eligibility criteria were included in the study. The participants were informed about the study’s objectives. They were asked to enter the study and for those who accepted this request, written consent was obtained. Confidentiality of their information was assured. The data were collected by a family physician (the principle investigator) from July to September 2015 ## Instrument development The questionnaire consisted of 26 items divided into 4 sections. The first section of the questionnaire collected information about the socio-demographic background of the patients. The second section contained two questions, the first one regarding the proportion of patients with type 2 diabetes who have been asked about ED. The responses were either yes, no, or I can’t remember. The second section measured the degree of willingness to discuss ED by the patients with their physician (i.e. unwilling, slightly willing, moderately willing, and very willing). The third section consisted of self-reported statements about 11 barriers preventing patients from discussing ED with their physicians. The responses to these statements were recorded on a five point Likert scale ranging from strongly agree to strongly disagree. These statements were developed after reviewing the literature. The last section of the questionnaire comprised a brief survey to assess the presence of ED in respondents by using a validated Arabic translation of the Index of Erectile Function (IIEF-5) questionnaire. IIEF-5 is a well-known tool to screen for ED and it has been used extensively in previous studies. It comprises only five questions. Also, it categorizes ED according to its severity as follow: severe ED (1–7), moderate ED (8–11), mild to moderate ED (12–16), mild ED (17–21), and no ED (22–25). Apart from the last section of the questionnaire which was adopted from the previously validated Arabic version, the majority of the questionnaire was developed in English, translated by an accredited translator in Arabic, and then back-translated by another accredited translator into English. The mismatches between the two English versions, the original and back-translated versions, were discussed and resolved by the primary author and the translators. The questionnaire was pretested and piloted on 30 monolingual patients with type 2 diabetes to ensure the comprehensibility and readability of the final Arabic version. The participants in the pilot study were recruited from medical out- patient clinics to prevent contamination with the main sample for the current study. ## Sample size calculation A sample size calculation was based on a pilot study with 30 participants which showed that 15% of patients with type 2 diabetes had been asked about ED in the last year. So, a sample of 306 patients with type 2 diabetes was required to obtain a 95% confidence interval of +/- 4% around the prevalence estimate of 15%. Assuming 10% of questionnaires in the pilot study were incomplete or not returned, a total of 336 questionnaires was required. ## Statistical analysis Descriptive statistics were used to describe the study sample characteristics and the participants’ identified barriers to discussing ED with their physicians. To test the association between patients who were willing to discuss (i.e. very willing, moderately willing, and slightly willing) and unwilling to discuss ED with the physicians, and the participants’ socio-demographic and clinical characteristics, chi-square tests were used for categorical variables. In one variable (current occupation), we collapsed some groups together to meet the conditions of the Chi-square test. For continuous variables, the normality of the data was checked by the Kolmogorov-Smirnov test. For normally distributed data, an independent sample t test was used to compare means. For skewed data, the Mann-Whitney U test was used to compare medians. Chi-square tests were used to compare the association between the willingness to discuss ED with the physician and the different participants’ barriers, after removing the neutral response variable. The Fisher exact test was used to test the association between participants having ED and their willingness to discuss but were not yet asked by their physicians in the last year, as the conditions of the Chi-square test were not met. Multivariable logistic regression analyses were performed to predict the willingness to discuss ED with the physicians by using the participants’ barriers, socio-demographic and clinical characteristics as covariates. For the logistic regression analysis, participants’ willingness to discuss ED was categorized as a binary variable, comparing those who reported any degree of willingness (very willing, moderately willing, and slightly willing) with those who were unwilling. The data were analysed using the statistical software package IBM SPSS Statistics for Windows, Version 22.0 (IBM Crop., Armonk, NY, USA). A p-value of less than 0.05 was considered to be statistically significant for all analyses. ## Ethics The study protocol was reviewed and approved by the Monash University Human Research Ethics Committee and the Institutional Review Board of King Khalid University Hospital, Riyadh, Saudi Arabia, where the data were collected. The participants were informed about the study’s objectives and their permission to enter the study was requested. Written consent was obtained from the participants. Confidentiality of their information was assured. # Results ## Participants’ socio-demographic and clinical characteristics Out of the 336 distributed questionnaires, 309 were completed and returned. The response rate was therefore 92%. The median age of the respondents was 60 years and the median duration of diabetes among the respondents was 10 years, with over half (59.2%) on tablets alone as treatment for this condition. Few (9.7%) had been asked by their physicians about ED in the last year although most (84.8%) were willing to discuss this problem with them. The presence of ED among the respondents was 89% with one third of them (28.2%) suffering from severe ED. The remaining socio-demographic and clinical characteristics are shown in. ## Prevalence of identified participants’ barriers to discussing ED with their doctors shows the distribution of participants’ barriers to discussing ED with their physicians. The most prevalent barriers among these respondents were having sex is not important to me (49.5%) and the treatment is too expensive (24.6%). ## Willingness to discuss erectile dysfunction (ED) and participants’ socio-demographic and clinical characteristics shows the association between the willingness of respondents to discuss ED with their physicians, and the respondents” socio-demographic and clinical characteristics. The participants who were willing to discuss ED with their physicians were younger with the mean age being 59.3 compared to the mean age of 65 in unwilling participants (P\< 0.001). Participants with low monthly incomes (i.e. \<5000 SR) (53.2%) were unwilling to discuss ED with their physicians (P = 0.03). Also, among participants who have ED, those who were complaining of severe ED (63.1) were unwilling to discuss it with their physicians. There were no significant associations between a willingness to discuss ED with the physicians and the highest education level, location of residency, current occupation, smoking status, duration of diabetes, type of diabetes treatment, and presence of ED. Multivariable logistic regression analysis was used to predict the participants’ willingness to discuss ED by their socio-demographic and clinical characteristics. After adjusting for the educational level, location of residency, monthly income, current occupation, smoking status, duration of diabetes, type of diabetes treatment, presence of ED, two participants’ characteristics were associated with willingness to discuss ED with the physicians. These characteristics were age above 60 (OR = 0.25, 95% CI: 0.11–0.55), and having severe ED (OR = 0.26, 95% CI: 0.08–0.85). No significant association has been found between the participants’ willingness to discuss ED and the other socio-demographic and clinical characteristics. ## Participants’ willingness to discuss ED and identified barriers shows the comparison between participants’ willingness to discuss ED with their physicians and the identified barriers. Comparing the ‘unwilling’ participants to the ‘willing’ ones revealed that the barriers which provide the main obstacles to discussing ED with the doctors are: embarrassing my doctor (63.9%, P \< 0.001), ED is a personal issue (60.6%, P \< 0.001), too old now (59.4%, P \< 0.001), feeling embarrassed to talk about it (57.1%, P \< 0.001), too sick now to address ED issues (55.9%, P \< 0.001), no effective treatment is available (54.8%, P \< 0.001), and my doctor is too young to discuss my ED with him (54.8%, P \< 0.001). ## Predicting participants’ barriers to their willingness to discuss ED shows the multivariable logistic regression analysis which was used to predict the participants’ willingness to discuss ED by their identified barriers. After adjusting for the age and severity of ED as possible confounders, two participants’ barriers were associated with willingness to discuss ED with the physicians. These barriers were “it may embarrass my doctor” (OR = 0.04, 95% CI: 0.01–0.2), and “It is a personal issue” (OR = 0.05, 95% CI: 0.01–0.28). ## Participants who have not been asked about ED and their willingness to discuss it shows that among the respondents who have not yet been asked about ED in the last year by their physicians, 91% of them have ED and would be willing to discuss it with their physicians (P = 0.02). Even if they do not have ED, twice as many are willing to discuss this matter as unwilling. # Discussion This survey has shown that few (9.7%) patients with type 2 diabetes mellitus have been asked about ED in the past year by their physicians, in spite of the majority (84.8%) being willing to discuss it. Further, the presence of ED was high (89%) among these patients, with one third of them (28.2%) suffering from severe ED. In spite of guidelines recommending physicians to enquire about ED in patients with diabetes, this does not take place in most cases. Grant et al found that only 9% of the patients with diabetes have been asked about ED in their last diabetes review consultation. In addition, Perttula found that physicians discussed ED with just 18% of their patients who have diabetes. This is in spite of the prevalence of ED being high in these patients. Also, the low rate of asking about ED in patients who have diabetes by physicians who work in primary care settings is similar to what happens in specialty practices where patients are at high risk for ED, such as those seen by cardiologists. Nicolai et al found that only 16% of cardiologists admitted to discussing sexual function regularly with their patients. So, there is a wide gap between recommendations and what takes place in practice. A study done in Bulgaria has shown that this gap reflects, in part, physicians’ beliefs that patients with ED rarely share this problem with their physicians, Overall, a large majority of patients (84.8%) were willing to discuss this topic. Unfortunately, to date few studies have examined ED-related issues from the patient perspective. Jiann et al showed that 56.6% of patients with type 2 diabetes wished to discuss ED with their physicians while Lo et al found that 76.1% of patients with type 2 diabetes would want to receive treatment for ED from their physicians. However, most patients think that the discussion should be initiated by the physicians. At the same time, most physicians seem to assume that patients do not like to be asked about sexual problems. The study’s findings suggest that two main factors were associated with a willingness to discuss ED: age and severity of ED. The patients above 60 years were 70% less willing to discuss ED with their physicians compared to the patients less than 60 years old. In addition, the patients who do have severe ED were 75% less willing to discuss ED with their physicians compared to the patients who have mild ED. As mentioned above, we found that elderly people were less willing to discuss ED with their physicians compared to younger patients in spite of the majority of the elderly population remaining interested in sexual activity. This group needs to be given more attention by their physicians as they have a very high prevalence of ED. In addition to that, ED is often underreported and underdiagnosed in the older male population. It has been shown that physicians are not proactive in discussing and managing the sexual health of elderly people. In a study done by Harding and Manry in the United States among health care providers, it was found that only 28% of the surveyed health care providers would usually asses the sexual health of elderly patients. Also, a negative attitude has been found in a study done to examine American psychologists’ willingness to assess the sexual health of older adults. In addition to the age of patients with diabetes as a predictor for willingness to discuss ED, the level of ED severity plays a major role, with this study showing that patients who have diabetes with severe ED are less willing to discuss this with their physicians compared to those with mild ED. This is particularly important as the literature suggests that diabetes is associated with more severe forms of ED. Men with diabetes also require more aggressive therapy to treat ED. In a study done by Walsh et al. it was found that men with diabetes were likely to need more aggressive therapy, and most went on to second line therapy (i.e. penile prosthesis surgery) for ED as these patients were less responsive to first line therapy (oral agents). It is important to identify the group who have diabetes and suffer from severe ED to optimise diabetes control and treat the ED as best one can. This should lead to improvement in both their sexual function and depressive symptoms, as shown by the SUBITO-DE study, an Italian multicentre study. The main barriers contributing to an unwillingness to discuss ED were: embarrassing the doctor, ED is a personal issue, too old, too sick to address ED issues now, no effective treatment available, and the doctor is too young to discuss ED with. Jiann BP et al found that patients’ embarrassment and false beliefs about ED treatment being either ineffective or harmful accounted for three quarters of the reasons why patients with diabetes will not discuss ED with their physicians. Embarrassment was the key factor preventing this discussion according to Rutte et al. as also shown by Gott M and Hichliff S. Other studies have revealed the importance of other barriers including differences in patient characteristics, i.e. their age, lack of knowledge, and difference in their culture. These differences in patients’ barriers found by various studies can be explained by difference in customs, traditions, culture, and health systems. The study findings also suggest that most (91%) of the patients who have not yet been asked about ED in the last year actually have ED and are willing to discuss it. This is contrary to what has been reported by Smith et al. who found that sexually active men are more likely to discuss sex with their physicians. These discrepancies in findings might be related to differences in sociocultural factors including social norms and attitudes. The current study has several implications for clinical practice. Firstly, ED is a major problem among patients with type 2 diabetes and this is frequently ignored by physicians even though a majority of these patients are willing to discuss this problem. Physicians who are involved in treating these patients should initiate the discussion. Secondly, patients with diabetes who are older and suffer from severe ED are less likely to discuss ED with their physicians. Targeting this sub-group of patients through education and the building of better relationships between physicians and their patients should help. Thirdly, there are multiple barriers that prevent patients with type 2 diabetes discussing ED with their physicians which could be reduced by better patient education and the addressing of psychological factors. There are several limitations to this study. The sample was taken from one hospital and may therefore not be generalizable. Also, the patients were taken from primary care clinics affiliated to a teaching hospital so that they might have more severe diabetes, and be older than patients in other primary care clinics. In addition to that, and due to the nature of the study design, the results revealed associations and not necessarily causal relationships between a range of factors and willingness to discuss ED. However, we believe that our findings shed an important light on this very sensitive issue among patients with type 2 diabetes. Also, no comparable work has been done in this country, and so it is of importance within this health care system. # Conclusions ED is a highly prevalent condition among patients who have type 2 diabetes. Most of these patients are not asked about ED within the last year of attending a clinic, even though most are willing to discuss it with their physicians. Many patients’ barriers to discussing ED have been identified, including being older and suffering from more severe ED, with these patients being less willing to discuss this with their physicians. Further research is needed to explore the barriers which prevent physicians from discussing ED with their patients who have diabetes. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Tuberculosis (TB) remains a global threat with an estimated 1.8 million deaths in 2015; approximately 14 percent attributed to multi-drug resistant (MDR-TB) and rifampicin resistant (RR-TB) tuberculosis. Advances in case detection and treatment regimens have dramatically reduced incidence and mortality; yet lengthy and complex treatment regimens continue to take a toll on treatment completion and success rates. Patients who default on TB treatment continue to contribute to the infectious disease burden of the community, increase their risk of developing MDR-TB, and are at an increased risk for TB-related mortality. According to the WHO 2016 report, Ukraine is among the top 20 highest drug- resistant TB burden countries in the world. National survey data estimate 25% of new incident cases and 58% of previously treated cases are MDR/RR-TB cases, resulting in approximately 22,000 cases per year. Driving these increasing MDR/RR-TB rates is an unchecked treatment default rate. Among the 2014 national TB cohort, treatment success was only 72% compared to 83% globally. Patient predictors for treatment default vary across Europe and Central Asia. In Spain, Cayla and colleagues found that immigrants, patients living alone, patients previously treated, and injection drug users (IDU) were at higher risk for default; whereas in Russia, the homeless, unemployed and alcoholics were at highest risk; in Moldova patients who were homeless, living alone, less educated, or living for extended periods outside the country were at risk; and in Estonia alcohol abuse, unemployment, MDR-TB, urban residence and previous incarceration increased risk. Timing of default was found to be heterogeneous across a large meta-analysis by Kruk and colleagues. In Uzbekistan, the first two-month intensive phase led to high rates of default, in Moldova the risky period was between intensive and continuation treatment, while data overall suggest that lengthy treatment during the continuation phase is the most likely period for defaulting. Strategies to improve treatment adherence and success center on directly observed therapy short-course (DOTS) with adaptations to different clinical service and social environments. DOTS has been implemented in clinical settings, through home visits by medical personnel, use of community volunteers, and DOTS by family members. Patient incentives are sometimes included to improve adherence, most commonly periodic food packages, transportation vouchers, and cash payments. However, evaluation of the impact of these strategies on treatment success remains inconclusive. The standard TB treatment for smear-positive patients in Ukraine covers 2–4 months of intensive inpatient therapy at the central TB dispensary. Once the patient tests smear-negative, (s)he is referred to a TB cabinet or polyclinic closest to their home for 2–5 months of outpatient continuation therapy. Directly observed therapy is the standard of care, requiring direct contact between patients and providers to administer the TB medication. According to the National TB Program (NTP), adherence to therapy is insufficiently controlled, with a 7.6% national default rate in 2010. In 2010, the Ukraine Red Cross Society (URCS), funded by the United States Agency for International Development (USAID), piloted a community-based social support program designed to improve TB treatment adherence during outpatient continuation therapy. URCS provided DOTS to a limited number of patients in their homes. Additionally, incentive food packages, psychological and career counseling, and/or vouchers for transportation or other necessities were provided periodically based on client needs. TB physicians managing patients’ continuation treatment at the outpatient facility made case-by-case referrals to URCS based on identification of the patient as high risk for defaulting on treatment according to established criteria for program inclusion. In 2011, the URCS program was suspended due to insufficient funds. By 2012, the program was again active and expanded to cover 10 oblasts in eastern and southern Ukraine with reported default rates ranging from 6.1–12.7%. This study measures the effect of the URCS social support program on the rate of treatment default among those at risk for defaulting during continuation therapy. Given the high rate of treatment default and the growing problem of treatment-resistant TB strains, identifying effective strategies for treatment adherence is critical. # Methods ## Study settings and sample population Three oblasts in Ukraine were purposively chosen for this study due to their high TB caseloads and high treatment default rates. In 2010, Dnipropetrovsk reported 1,077 TB cases and 12.4% default rate; Kharkiv reported 738 cases with 11.1% default; and Odessa reported 789 cases and 9.4% default rate. The study population was composed of patients receiving TB continuation treatment in these three oblasts in 2011 and 2012. Patients were classified as high-risk for TB treatment default per program criteria covering eleven self-reported risk factors: HIV positive, alcoholism, injection drug use (IDU), a contact to a TB case, a co-morbidity, homeless, unemployed, a health care worker, a migrant, a refugee or immigrant, an ex-prisoner, and room to record other risk factors that a provider might deem noteworthy. Low-risk patients did not report these risk factors with the exception of unemployment. Facility staff revealed that TB patients routinely report being unemployed. This is to avoid stigma in the workplace, as workplace TB screening is routinely undertaken after a case is diagnosed. Consequently, we considered a patient as low risk during sample selection if the only reported risk factor was unemployment. During data cleaning, minimal corrections were made for misclassification of risk status at time of data entry. All study patients completed TB intensive treatment, initiated TB continuation treatment, and had a TB treatment outcome recorded in their medical record. Five patient cohorts were sampled: high-risk (HR) patients enrolled in the URCS social support program January 1 –May 31, 2012 (henceforth 2012 HR- Intervention); high-risk patients not enrolled in the social support program in January 1 –May 31, 2012 (henceforth 2012 HR-Comparison); low-risk (LR) patients not enrolled in the social support program in January 1 –May 31, 2012 (henceforth 2012 LR-Comparison); high-risk patients not enrolled in the social support program in January 1 –May 31, 2011 (henceforth 2011 HR-Comparison); and low-risk patients not enrolled in the social support program in January 1 –May 31, 2011 (henceforth 2011 LR-Comparison). A cohort of 2012 HR-Intervention patients (n = 409) was randomly sampled from a complete listing of URCS enrollees, stratified and proportionate in size to the TB patient population by oblast. These patients served as the index cases. Each TB facility where an index case was receiving continuation therapy served as the facility match point to ensure controls experienced similar service environments to the randomly selected cases. In order to obtain a group of controls not exposed to the program, four patients from these facilities’ TB registries were matched to the index patient: one 2012 HR-Comparison patient; one 2012 LR-Comparison patient; one 2011 HR-Comparison patient; and one 2011 LR-Comparison patient. The primary comparison group for this analysis was the 2012 HR-Comparison cohort; however, additional cohorts were sampled to explore selection. The second matching variable was the start date for TB continuation treatment for the index case in order to control for seasonality of TB and services. Additional matching on sex and age was done if more than one match was eligible. Data from 1630 TB patients across the five cohorts were collected. Sampling weights were generated to account for sampling proportionate to the varying size of TB caseload per oblast and non-response. During data cleaning, minimal corrections were made for misclassification of risk status at time of data entry. Data entry misclassifications included: dropping cases who received the intervention in 2012 but had no reported risk factors (n = 12); reclassifying those whose only risk factor was unemployment from high risk to low risk (n = 16); and reclassifying another 33 cases as high risk based on a risk factor reported. A survey of 50 TB polyclinics and cabinets that provided continuation therapy to the study population was also completed to provide details on the referral and treatment practices at these facilities (see report for details). ## Data collection and definitions For each study patient, retrospective data were abstracted from TB medical records (national form TB01). The data abstracted included basic demographics, sex, age, employment status, urban or rural residence; and TB diagnosis, treatment, interruptions to intensive treatment and treatment outcomes. Standard WHO definitions were used for TB classification (e.g., first diagnosis, re- initiated treatment, treatment failure, relapse, and referral); for clinical TB (e.g., pulmonary or extra-pulmonary); for WHO diagnostic categories to indicate treatment regimens; and for treatment outcomes (e.g., success, death, treatment failure, treatment interrupted, and transferred). A patient’s outcome was successful if the full course of prescribed treatment was completed or follow-up testing indicated patient was cured. Treatment default included anyone who missed treatment for more than 60 consecutive days per WHO standards. Additional data from the TB records were abstracted from form TB01-01, the risk screening form used by providers to identify a patient’s risk for defaulting on treatment. For the 2012 HR-Intervention group, data on social support services received such as home visits, food packages, clothing, transport vouchers, monetary incentives, and counseling were abstracted from the URCS records and merged with the patient TB record. ## Data analysis Descriptive statistics were generated to compare demographics, TB disease characteristics, and reported risk factors for treatment default across the five cohorts. Logistic regression models with average marginal effects (AME) were estimated to test the study questions. All analyses used data weighted for sample selection; reported standard errors are clustered at the facility level. To validate risk factors predictive of treatment default, the social support program criteria risk factors were regressed on treatment default among patients receiving continuation therapy. Dichotomous variables for seven individual risk factors (HIV positive, alcoholic, IDU, contact to a case, co-morbidity, homeless, and unemployed) were included. Very few patients reported being a health care worker, migrant, refugee, or ex-prisoner; hence, these were combined with the unspecified risk factor as “other”. A dichotomous variable indicating the presence of more than one risk factor was also added. All risk factors were run simultaneously first (Model A). Next, we controlled for basic demographics and four dichotomous disease and treatment characteristics due to their hypothesized role in TB treatment adherence and outcome: first time TB diagnosis, pulmonary TB, WHO Category I and more than 2 interruptions in care during intensive therapy (Model B). To identify the salient risk factors for default in the absence of an intervention, this analysis was restricted to data from 2011 when the URCS program was not operating. To evaluate the impact of the social support program on treatment default, the second regression analysis was limited to the 2012 HR-Intervention and the 2012 HR-Comparison cohorts. Prior to estimating impact, balance between the intervention and comparison groups was examined. The final model estimated the impact of the social support program on treatment default, controlling for risk, disease status, and demographics. Analyses were produced using Stata SE version 13 (College Station, TX). This study was approved by the Office of Human Research Ethics at the University of North Carolina at Chapel Hill and the ethical review board of the F.H. Yanovskyi Institute of Phthisiology and Pulmonology, Academy of Medical Sciences of Ukraine. Both review committees waived the requirement for informed patient consent. Data collection was performed by the IFAK Institut in Ukraine. Data collectors had access to patient names in order to track patients from registry entries to patient records; however, names were not recorded on data collection tools nor reported to the researchers. # Results ## Study population The final dataset included 1,618 records from TB patients across the five cohorts: 2011 LR-Comparison (n = 308), 2011 HR-Comparison (n = 340), 2012 LR- Comparison (n = 262), 2012 HR-Comparison (n = 311), and 2012 HR-Intervention (n = 397). The study populations shared similar demographic profiles across risk cohorts and years. Approximately two-thirds of the patients were male in every risk group, just over three-quarters were under fifty years of age, and a large majority lived in urban areas. Over half of all patient cohorts were unemployed, ranging from 55–72%. Half of the study population received TB continuation treatment in Dnipropetrovsk (50.0%), with the remainder evenly divided between Kharkiv (25.4%) and Odessa (24.6%). Among the three HR cohorts, unemployment was the most common reported risk factor (48–62 percent), followed by alcoholism (34–44 percent), co-morbidities (33–36 percent) and being HIV-positive (21–47 percent). A majority (63–72 percent) reported between 2 and 3 factors putting them at risk for treatment default, while 3–7 percent reported four or more. Notably, the proportion of HR patients who reported injection drug use in their medical records was small, ranging from 5–12 percent. In discussions with facility staff, it was noted that information on IDU status and treatment is not routinely recorded in the TB charts nor shared across cabinets due to concerns of confidentiality. As expected over half of the LR patients reported no risk factors for treatment default, while the remaining reported unemployment. Overall, 81.1 percent of the TB patients were undergoing treatment for a first diagnosis, although among the HR cohorts, a higher percentage re-initiated treatment after earlier failure or relapse compared to the LR cohorts (7–12 percent versus 3–5 percent). Ninety-three percent of all cases were pulmonary TB, a majority was classified as WHO Category I (63.8 percent), and 81.3 percent reported only one or fewer interruptions in intensive treatment. TB treatment outcomes in 2011 were significantly different between the LR and HR cohorts on treatment adherence. Treatment default among the 2011 LR-Comparison cohort was 4.2 percent compared to 13.3 percent in the 2011 HR-Comparison cohort (p\<0.000); while 90.6 percent of the LR-Comparison cohort reported treatment success compared to only 74.3 percent of the HR cohort (p\<0.000). Similar differences were measured in 2012 when comparing the LR-Comparison and HR- Comparison on default, 4.6 percent and 10.6 percent (p\<0.006), and success, 87.0 percent and 69.8 percent (p\<0.000) respectively. The 2012 HR-Intervention cohort fared better than the 2012 HR-Comparison on default (1.3 and 10.6 respectively, p\<0.000) and on success (88.4 and 69.8 respectively, p\<0.000). However, comparisons between the 2012 LR-Comparison and the 2012 HR-Intervention on default (4.6 and 1.3 percent respectively) and success (87.0 and 88.4 percent respectively) found no statistical differences. Lastly, statistical differences were found between the 2011 HR-Comparison and the 2012 HR-Intervention cohorts for both treatment default and treatment success (p\<0.000); while no statistical differences were found between the 2011 HR-Comparison and 2012 HR- Comparison groups. These comparisons across cohorts highlight that the difference in outcomes between the 2012 HR and LR comparison cohorts is similar to the differences between the 2011 HR and LR cohorts. This supports the impact identification strategy employed in our evaluation of the program. ## Predicting treatment default The URCS social support program was designed to target those at highest risk of treatment default and provide support to improve treatment adherence. The official eligibility criteria for program support cover eleven risk factors. Among patients from 2011, only those who reported being an alcoholic (p = 0.002) or an IDU (p = 0.043) were more likely to default on TB continuation treatment, while a patient reporting a co-morbidity was less likely to default (p = 0.028). Additionally, those patients enrolled in continuation care who had two or more interruptions recorded during intensive treatment were more likely to default during outpatient treatment (p = 0.028). Estimated marginal effects predict that an individual’s probability of default increased by 0.08 (p = 0.017) if an alcohol abuser and by 0.05 (p = 0.043) if prior treatment interruptions were noted; yet one’s probability of default decreased by 0.06 (p = 0.043) if reporting a co-morbidity. ## Evaluating program impact Primary impact analyses were limited to the 2012 HR-Intervention and the 2012 HR-Comparison cohorts. Mean differences between the two groups were tested for 18 variables; seven (38 percent) were unbalanced at standard statistical levels (p\<0.05). Looking at the intervention group, a higher proportion of patients were alcoholics or unemployed, and were 18–29 years of age, while the comparison group had a higher proportion of persons with HIV, homeless, undergoing WHO treatment category 1, and who were male. All risk factors, disease characteristics, patient demographics and treatment oblast were controlled for in the final impact model. Measuring impact, results indicate that the HR patients receiving the social support program decreased their probability of treatment default by 0.101 (p\<0.000) compared to the comparison cohort. A second analysis, comparing outcomes for the 2012 HR-Intervention cohort to the 2011 HR-Comparison cohort, produced similar results. Five of 18 variables (28 percent) were not balanced between the cohorts. Controlling for all variables, the 2012 HR patients receiving the social support intervention were significantly less likely to default (p\<0.000) compared to the 2011 HR- Comparison cohort, and the probability of default decreased by 0.120 (p\<0.000) (data not shown). However, alcoholism remained a significant risk factor with the probability of default 0.069 (p\<0.014) higher among alcoholics compared to non-alcoholics. Additionally, the probability of default among those with more than two treatment interruptions during intensive care was higher at 0.047 (p = 0.017). # Discussion In 2012, TB patients receiving social support provided by URCS in Ukraine reduced their probability of defaulting on continuation treatment by 10 percentage points compared to high-risk patients who did not receive social support in 2012 or 2011. Treatment success rates for the high-risk patients receiving social support were comparable to the low-risk cohorts and significantly improved over the high-risk comparison cohorts. This result was found despite the heterogeneity of the patient population and the services provided. Although treatment oblast was not predictive of success or default in 2012, routine implementation of DOTS and social support varied by study oblast. According to reported practices in 2014, 89 percent of surveyed facilities in Dnipropetrovsk provided facility-based DOTS and a majority of these facilities required daily DOTS visits (83 percent). Among the sites offering home-based DOTS, half provided weekly or bi-weekly visits. In contrast, 88 percent of the facilities in Odessa provided home-based DOTS and the majority of home visits were daily. This increase in home-based services may reflect a growing recognition in Odessa that facility DOTS is insufficient to assure compliance. Variation across oblasts may reflect patient population needs or facility capacity. Further investigation into best practices for DOTS and social support in Ukraine is warranted. URCS was the only provider of social support in Kharkiv and Odessa in 2012 and the primary provider in Dnipropetrovsk. In 2011, only 23 percent of the facilities referred patients for social support, increasing to 94 percent by 2012. In all oblasts the primary point of referral was the city or raion TB physician. This is in keeping with URCS’ policy to only provide social support to smear-negative patients who successfully completed intensive TB treatment and initiated continuation treatment. This focus on continuation patients ignored patients who defaulted during intensive treatment, which could be substantial. In Russia, Jakubowiak et al. found that 44 percent of TB treatment defaulters exited treatment during the intensive regimen. In Moldova, the highest default rates were recorded during the first month of inpatient intensive treatment. Our data did not include patients who defaulted during inpatient treatment, however in 2011, patients with more than two treatment interruptions during inpatient care increased their probability of defaulting during outpatient care by 5 percentage points. This is similar to findings by Jakubowiak in Russia and Santha in India, where gaps in intensive treatment were associated with future treatment default. Prioritization for support services may benefit those who had difficulty during the intensive phase. Alcoholism was the one risk criteria predictive of defaulting among the 2011 cohorts, increasing the probability of default by 5 percentage points. Neither positive HIV status nor reported injection drug use were statistically associated with higher default rates. Under-reporting of these two risk factors may be one explanation for their lack of significance. In Ukraine, sharing of confidential patient information between service delivery clinics is limited. The risk factor information documented on a patient’s TB form is all self- reported and possibly under-reported due to fear of stigmatization. For example, a patient seeking HIV-related services may not report their status to the TB physician. Unless an infectious disease specialist is overseeing services for both TB and HIV patients, this case of co-infection may go undetected by the individual clinics, despite best practices of routine HIV screening among TB patients. According to the facility survey, only 32% of facilities providing DOTS also provided ART for persons living with HIV. For IDUs the challenge may be two-fold. First, the availability of drug-substitution therapy for IDUs in Ukraine is scarce; only 16% of the outpatient TB facilities reported offering this service. Without adequate substitution therapy, many IDUs may drop out of service during the intensive, inpatient TB treatment phase, excluding them from our sample. Second, the stigma for drug addiction may discourage IDUs from revealing this risk to their TB physician. In either scenario, the risk of default among IDUs may not be adequately reflected in our data. Patients with reported co-morbidities reduced their probability of defaulting by almost 6 percentage points in 2011, possibly due to additional support received from providers caring for the co-morbidities. For all other risk factors, no statistical associations were found with default. Whether this is due to the small numbers of patients with these other risks or because these factors do not increase one’s risk of default is undetermined in this study. Interestingly, almost 20% of the high-risk cohort receiving the intervention had an undetermined or “other” risk factor recorded. Provider interviews suggested that compliant patients in our study sites may have been referred to URCS as a reward for their adherence. If widespread, this preferential referral of adherent patients could create selection bias, affecting results. This is one of the limitations of retrospective data analysis, it is difficult to measure the fidelity of program implementation retrospectively. However, if there was widespread selective referral one would expect the 2012 HR-Comparison group to have reported a higher default rate than the 2011 HR-Comparison group. The comparability of the default rates among these two cohorts suggests that very little selection bias exists. In an era of declining health resources and increasing drug-resistant TB, refining and standardizing the referral criteria for additional social support may reduce the national default rate, but not without a cost. This study shows that social support is effective in reducing default rates but whether or not that means it should or can be scaled-up depends on the cost of wider implementation and the cost relative to other potential interventions that might also reduce default rates. # Conclusions This study demonstrates the positive impact of providing social support to those at-risk for treatment default. Targeting services to those who will most benefit is critical to reduce continuing TB transmission. Further research is recommended to differentiate the costs and benefits from home-based DOTS versus additional services offered through social support programs. Prospective cohort studies could refine targeting of programming, evaluate social support program fidelity, identify which populations respond best to select services, and what barriers might still exist to achieving better adherence. With that information, tailoring programs to most effectively reach and serve clients in a patient- centered approach may reap substantial rewards for Ukraine. Prioritizing support services for clients who struggle with alcohol or drug addictions or struggle with adherence to intensive inpatient treatment regimens, may improve treatment success. Identifying approaches to assure intensive treatment completion and flagging those upon completion for additional follow-up during continuation treatment, has the potential to further reduce program defaults and improve outcomes. MEASURE Evaluation is funded by the U.S. Agency for International Development (USAID) under Cooperative Agreement GHA-A-00-08-00003-00 and is implemented by the Carolina Population Center at the University of North Carolina at Chapel Hill, in association with The Palladium Group, ICF International; John Snow, Inc.; Management Sciences for Health, and Tulane University. We are grateful for the Carolina Population Center (R24 HD050924) for general support. We also thank the F.H. Yanovskyi Institute of Phthisiology and Pulmonology, Academy of Medical Sciences of Ukraine for their support in conducting this study in Ukraine. [^1]: The authors have read the journal's policy and have the following conflicts: Stephanie Mullen is employed by John Snow Inc. This affiliation does not alter our adherence to PLOS ONE policies on sharing data and materials. [^2]: Current address: Office of Family and Community Health Improvement, Washington Department of Health, Olympia, Washington, United States of America

# Introduction Ferulic acid (FA) is a phenolic substance and an important active ingredient that is common in various plants. It occurs at high concentrations in food ingredients such as coffee, grain hulls, vanilla beans, wheat bran, and rice bran. The FA molecule is a 4-hydroxy-3-methoxy cinnamic acid (C₁₀ H₁₀ O₄), and FA exhibits various biological activities and physiological functions and has low toxicity. Importantly, FA has been shown to exhibit antioxidant and anti-inflammatory properties and is an effective modulator of multidrug resistance in cancer. FA can prevent acute liver injury due to sepsis by attenuating the inflammatory response, and it improves corticosterone-induced liver damage. Moreover, FA also improves corticosterone- induced depressive behaviour and oxidative stress in mice, and enhances the antibacterial activity of quinolone antibiotics. Geese is a nutritious and healthy food resource. the short reproductive periods, low hatchability, and high embryo mortality of geese. it is reported that the content of protein and trace elements in the meat of goose is higher than in other poultry products. So, geese breeding industry has broad prospects. Intestine plays a key supporting role in the growth of animals. and early growth and development of the gastrointestinal tract are critical to optimizing the growth of poultry. In the early stages of goose growth and development, a large amount of nutritional support is required. After the intestinal structure and function are damaged, it will affect the individual’s early nutritional absorption, causing irreversible damage to the body’s growth and development in the early stages. Lipopolysaccharide (LPS) is a pathogenic compound that occurs in the outer membrane of the cell wall of all gram-negative bacteria, and it can elicit multiple signaling events in the cell. LPS can act as a strong inflammatory mediator and is widely used in animal studies to simulate bacterial infection. Peng et al. studied inflammation in bovine endometrial epithelial cells caused by LPS and found that FA exerted an anti-inflammatory effect by inhibiting the release of cytokines. Chen et al. reported that 30 mg/kg FA increased the antioxidant capacity of the liver and repaired liver damage, reducing hepatocyte death due to LPS in blunt-nosed sea bream. Further, FA plays a positive role in reducing renal injury in HFD/STZ-induced DN mice by enhancing autophagy and inhibiting inflammation. However, to date, the effects of FA treatment in geese after triggering oxidative stress have not been studied, nor has the oxidative stress damage encountered during the development of the geese industry been addressed, which would lead to mortality in geese, and could adversely affect the development of the geese industry. In the present study, FA was added to the diets of Jilin white geese at different concentrations, and LPS was injected intraperitoneally at 14 and 21 days of age to investigate the effects of FA on the growth performance and intestinal antioxidant capacity of LPS-stressed Jilin white geese. Our findings show that FA can protect animals under oxidative stress conditions and provide a scientific basis for the implementation of FA as an antioxidant agent. # Materials and methods ## Experimental design This study was carried out in strict accordance with the recommendations of the Guide to Nursing and Use of Experimental Animals of Jilin Agricultural University. The study was verbal approval by the Experimental Ethics Committee of Jilin Agricultural University. All operations were performed under pentobarbital sodium anesthesia, and every effort was made to reduce pain.120 male Jilin white geese of similar weight at 7 d of age were used. There was a pre-feeding period of 7 d, and the trial period was 21 d. After the pre-feeding period, the 120 Jilin white geese were randomly divided into six groups with five replicates (four birds in each group). Groups F1 (60 mg/kg feed FA), F2 (120 mg/kg feed FA), F3 (180 mg/kg feed FA), F4 (240 mg/kg feed FA), and L were given intraperitoneal injections of LPS (500 μg/kg BW) on days 14, 17, and 20. The doses and routes of LPS administration referred to the previous studies. Group C was given intraperitoneal injections of normal saline (0.5 mg/kg BW). The test was carried out by establishing an oxidative stress model. shows the composition of rations. During the test period, all geese had access to feed and water ad libitum throughout the trial. Water was provided in a half-open plastic cylindrical water tank, and the feed was provided in feeders on one side of each pen. The geese were reared indoors conditions (temperature: 26.0°C ± 3.0°C; relative humidity (RH): 60.5 ± 5.0%; lighting period: 16 h; space allocation: 0.49 m2/gander), and the feed intake and body weight were recorded daily. On day 21, 10 animals in each group were randomly selected for slaughter, and tissue samples of the heart, liver, spleen, kidney, bursa of fabricius, and thymus organs, duodenum, jejunum, and ileum were collected. 1\) The premix provided the following per kg of diets: VA 2500 IU, VD₃ 1000 IU, VE 3100 mg, VK₃ 200 mg, VB₁ 100 mg, VB₂ 1 200 mg, VB₆ 200 mg, VB₁₂ 2 mg, Nicotinic acid 600 mg. Pantothenic acid 1 700 mg, Folic acid 200 mg, Biotin 20 mg, Fe (as ferrous sulfate) 6 000 mg, Cu (as copper sulfate) 300 mg, Mn (as manganese sulfate) 15 000 mg. Zn (as zinc sulfate) 8 500 mg, I (as potassium iodide) 10 mg, Se (as sodium selenite) 30 mg. ## Test materials ### Test animals and reagents The test animals were purchased from Jilin Yuhong Ecological Agriculture Technology Co. We also used *E*. *coli* lipopolysaccharide (Sigma Chemical Co., St. Louis, MO, USA), FA (97%) (Shanghai Maclean Biochemical Technology Co., Ltd), and Malondialdehyde assay (MDA)kit, Total Antioxidant Dismutase Assay (SOD) Kit, Glutathione peroxidase assay (GSH-PX) kit, Hydrogen peroxidase assay (CAT) kit, Total Antioxidant Capacity Assay(T-AOC) Kit (Nanjing Jiancheng Bioengineering Institute, Nanjing, China). ## Sample collection and indicator determination Growth performance. The initial weight, day 14 weight, and final weight of the geese were recorded. The average daily gain (ADG), average daily feed intake (ADFI), and feed conversion ratio (F/G) in the stages of days 1–14, 15–21, and 1–21 in the trial were calculated. ### Visceral index All surgeries are performed under pentobarbital sodium anesthesia, and every effort was made to reduce pain. Slaughter was carried out on day 21 of the trial. The organs of the geese (heart, liver, spleen, bursa, and thymus) were removed after slaughter and were weighed after removing excess fat. Measurement of intestinal oxidative stress indicators. Determination of MDA, SOD, GSH-PX, CAT and T-AOC in the intestinal tract of geese. ## Data analysis The raw data were processed using Microsoft Excel. A one-way ANOVA and Duncan’s multiple comparison test were performed using SPSS 26.0 software. Results are presented as mean ± standard error (SE), with *P \<* 0.05 indicating a significant difference. # Results ## Growth performance The LPS damage model was developed to observe the change in growth performance of each group before and after oxidative stress. Between days 1 and 14, final body weight (FB) and ADG were significantly higher in group F3 than in group L (*P \<* 0.05), and ADFI was significantly higher in groups F2 and F3 than in groups C and L (*P* \< 0.05). The FCR was significantly lower in group F3 than in group L (*P* \< 0.05). On days 15–21, FB was significantly higher in the groups with FA added than in group L (*P* \< 0.05). ADG was significantly higher in the groups with FA added than in group L and significantly lower than in group C (*P* \< 0.05). ADFI was significantly higher in the groups with F1 and F4 than in group L and significantly lower in the groups with FA added and L than in group C (*P* \< 0.05). FCR was significantly lower in the groups with FA added than in group L and significantly higher in groups F1, F2, F3, and L than in group C (*P \<* 0.05). From days 1 to 21, ADG was significantly higher in the groups with FA added than in group L and significantly lower in groups F1 and L than in group C (*P \<* 0.05). ADFI was significantly higher in groups C, F2, and F3 than in group L (*P \<* 0.05). FCR was significantly higher in group L than in the other groups (*P \<* 0.05). ## Organs indices To reflect the changes in organ weight in geese after oxidative stress, samples were taken at 21 days and the organs indices was measured. The thymic indices was significantly higher in group F4 than in groups C (*P \<* 0.05). ## Antioxidant activity in the duodenum We measured the changes in antioxidant activity in the duodenum of geese after oxidative stress to reflect the antioxidant activity of the duodenum. Groups F1, F3, F4, and C showed a significant decrease in MDA compared to group L (*P \<* 0.05). Group F1 showed a significant decrease in SOD compared to group C and Group F4 showed a significant reduce in GSH-Px compared to group C (*P \<* 0.05). Group C showed a significant increase in SOD and GSH-Px compared to group L (*P \<* 0.05). Groups L, F1, F2, and F4 showed a significant decrease in CAT compared to group C (*P \<* 0.05). ## Antioxidant activity in the jejunum We measured the changes in antioxidant activity in the jejunum of geese after oxidative stress to reflect the antioxidant activity in the duodenum. MDA was significantly lower in groups F3 and F4 than in group L and in group F4 than in group F1 (*P \<* 0.05). SOD was significantly higher in group C than in groups L and F1 (*P \<* 0.05). GSH-Px was significantly higher in group F3 than in groups L, C, and F1 (*P \<* 0.05). The CAT content of group L was significantly lower than that of all other groups (*P \<* 0.05). ## Antioxidant activity in the ileum We measured the changes in antioxidant activity in the jejunum of geese after oxidative stress to reflect the antioxidant activity in the duodenum. MDA was significantly lower in groups F2, and C than in group L, (*P \<* 0.05). GSH-Px was significantly higher in group C than in groups L and F1, F2, F4 and significantly higher in group F2, F3, F4 than in group L and significantly higher in group F3 than in group F1 (*P* \< 0.05). # Discussion During the process of goose breeding, oxidative stress can lead to a decrease in gse growth performance and a loss of economic benefits. LPS, a major component of the cell wall of gram-nega-tive bacteria, is a pathogenic compound. In this study, we obtained equivalent experimental results: LPS injection at 0.5 mg/kg body weight significantly reduced both the body weight gain of geese at d 15 to 21 and the feed intake during, and, but increased the FCR of geese on d 15 to 21. These initial results indicated that the oxidative stress model was established successfully and was suitable for subsequent investigation of the effects of FA on geese health. The addition of 6 mg/kg of FA to the feed increased beef the tenderness, juiciness, flavor intensity, and amount of some fatty acids. Supplementation with 180 mg/kg of FA in the present study enhanced growth performance and reversed the decline in growth performance brought about by oxidative stress, indicating that ferulic acid can alleviate the damage caused by oxidative stress. The liver, as the main metabolic and detoxification organ in the body, is closely related to various physiological functions and circulation, and the effects of immune organs such as the liver on oxidative stress occur through the recruitment of immune cells. In the present study, the addition of 240 mg/kg FA increased the thymic index, probably because FA mitigates the effects of oxidative stress from LPS by affecting the weight of the thymus and increasing body immunity. Reported that FA has cytoprotective capacity and may promote gastrointestinal health and microbial protein synthesis. In mice with high oxidative stress levels, insulin synthesis was improved in pancreatic β-cells. In the present study, the addition of 180 mg/kg of FA significantly alleviated the intestinal damage caused by oxidative stress, may be related to the enhancement of intestinal epithelial cell activity by ferulic acid, enhances cell activity to resist LPS induced damage. FA also inhibited the activation of the ROCK/NF-κB signaling pathway, thereby improving the dysregulation of oxidative stress and inflammation and exerting an effective hepatoprotective effect. The above growth performance and intestinal oxidative indexes were due to excessive reactive oxygen species production from LPS-induced oxidative stress, which indirectly affected the changes in physiological indexes and enzyme activities. This may be related to antioxidant signaling pathways, such as Nrf2/HO-1 and ROCK/NF-κB. Administering FA after LPS-induced oxidative stress can alleviate the effects of oxidative stress on growth performance and reduce intestinal antioxidant activity. # Conclusion 1\. Adding 180 mg/kg of FA can increase the body weight of geese and promote their growth. Adding 60 mg/kg FA can improve the thymus index, alleviate the damage to immune function caused by stress, and reduce the negative effects of stress. 2\. Adding 180 mg/kg FA can alleviate oxidative stress damage in the duodenum, ileum, and jejunum, reduce cell membrane damage, maintain the homeostasis of the membrane lipid bilayer, and protect cells from oxygen ions. In conclusion, adding 180 mg/kg of FA promoted the growth of geese and alleviated the effects of oxidative stress and the damage caused by oxidative stress in the duodenum, jejunum, and ileum. [^1]: The authors have declared that no competing interests exist.

# Introduction Computational fluid dynamic (CFD) simulations of biological valves have steadily improved over the years; however, procedures accounting for the formation of actual solid aggregates, such as calcifications or clots, have not been implemented yet. At the same time, researchers have also devised mathematical models for clot formation and growth; however, these models have been developed independently and are not usually associated to the dynamics of the valve. We propose a particle-based method that, by taking advantage of its mesh-free nature, can compute the fluid dynamics, together with valve deformation and formation of solid aggregates. In general, the simulation of biological valves, where a solid structure interacts with the surrounding flow, constitutes a fluid-structure interaction (FSI) problem. The algorithms to solve FSI problems may be broadly classified into two categories: conforming mesh methods and non-conforming mesh methods. Conforming mesh methods divide the computational domain in two parts: (i) a part occupied by the liquid where the Navier-Stokes equations are solved, and (ii) a part occupied by the structure where the stress-deformation equations are solved. Since the structure moves and/or deforms with time, re-meshing is needed as the solution advances. Non-conforming mesh methods, most notably the Immersed boundary methods (IBM), treat the interface between the fluid and the structure as a constraint and the force exerted by the structure to the fluid becomes a source term in the momentum equation. As a result, the fluid and solid equations are solved independently and re-meshing is not necessary. Both methods, however, have difficulties handling phenomena such as calcification and clotting that involve some sort of transition where part of the liquid transforms into a solid. In general, attempts to account for the formation of solid aggregates in CFD/FSI studies are based on ‘numerical artifices’ such as fluids with higher viscosities to mimic clotting, or membranes with higher stiffness to mimic calcification. For a different approach see. On the other hand, modelling of clot formation and growth had followed an independent path that, in some cases, has brought to particle-based techniques such as Lattice-Boltzmann or Coarse grained Molecular Dynamics. In general, however, these models assume simple hydrodynamic conditions and/or refer to straight blood vessels with not moving/deforming parts. Additionally, coupling fluid dynamics and solid deformation can be implemented with the Smoothed Particle Hydrodynamics method. But, it can’t easily account for other phenomena such as contact mechanics and agglomeration. In order to account for the fluid dynamics, the valve deformation, and the formation of solid aggregates at the same time, we propose *discrete multi- physics*: a mesh-free approach, where computational particles are employed for both the flow and the structure. With this method, the distinction between liquid and solid depends exclusively on the types of forces that act on each particle: pressure and viscous forces characterize liquid particles, while elastic forces characterize solid particles. By changing the type of forces on specific groups of particles, we can change their status form liquid to solid and vice versa. In, we used this idea to model melting and solidifying flows; in this paper, we extend it to the formation of agglomerates in biological valves. This article is organized as follows. Initially, we discuss the basic ideas behind our discrete multi-physics technique and describe the geometry used in the simulations. Next, we validate the model against both traditional modelling techniques and experimental data. Finally, we introduce the formation of solid aggregates at the membrane surface and in the flow. The objective of this paper is to apply discrete multi-physics to biological valves in general. For this reason, we do not focus on a specific type of valve at this stage. However, in order to test our model in the most challenging scenario, we chose dimensions and velocities similar to those occurring in aortic valves. We consider these conditions to be the most challenging scenario because (i) they involve higher velocities, which generate complex recirculation patterns, and (ii) they involve higher stresses, which generate larger membrane deformations. From this point of view, the fact that we simulate bicuspid valves, while the aortic valve has three leaflets, is not a limitation. Given the same mechanical stress, in fact, deformations are higher in bicuspid valves than in tricuspid valves. Therefore, by forcing velocities that are typical of aortic valves in bicuspid valves, we test our model under conditions that are even more critical (i.e. produce higher deformations) than those occurring in aortic valves. # Modelling ## Discrete multiphysics Our discrete multi-physics approach is based on the so-called discrete multi- hybrid system (DMHS). This technique combines various mathematical models to achieve a representation of fluid-structure interactions and solid-liquid systems that is not attainable with each model separately. Elsewhere, we showed that the linkage of different models is mathematically complex and computationally time consuming. In order to facilitate this, the DMHS combines models that share a common discrete (particle-based) paradigm, such as SPH (Smoothed Particle Hydrodynamics), CGMD (Coarse-grained Molecular Dynamics), DEM (Discrete Element Method) or BD (Brownian Dynamics). In this study, models for solid contact/collision (i.e. DEM), or for fluctuating hydrodynamics (i.e. BD) are not necessary; consequently, the coupling is limited to SPH (liquid phase) and CGMD (solid phase). For the numerical solution, the model was implemented in LAMMPS. A mathematical introduction to SPH and CGMD and the approach used to couple the models is given in. Specific details of the DMHS and other mesh-free hybrid techniques can be found in. In previous DMHS publications, there is an interchangeable use of the terms CGMD and Mass-Spring Model (MSM). This depends on the fact that these articles cover different scales. Articles dealing with microscopic scales use CGMD, whereas articles dealing with macroscopic scales use MSM. Mathematically, however, the two techniques are equivalent. In the main text, we prefer MSM, which is more consistent with the scale under investigation. In, we use CGMD, which is more consistent with the original DMHS formulation. ## Geometry We use a 2D simplified geometry for modelling a generic bicuspid valve as illustrated in. The channel half-thickness is *Z* = 0.0125 m, the length of the membrane is *L* = 0.016m and the radius of the circular area is *R* = 0.0215 m. As mentioned above, the DMHS combines various particle-based modelling techniques. In this study, the simulations are based on two models and three types of particles: SPH particles for the fluid, fixed SPH particles for the walls and MSM particles for the flexible leaflets (membrane). Periodic boundary conditions are used at the inlet/outlet. To model the Young modulus *E* and the flexural rigidity *F* of the membrane, the MSM particles are joined together by numerical ‘springs’ and ‘hinges’, as described in. The relation between the spring (*k*_*b*) and hinge (*k*_*a*) constants and the actual Young modulus and the flexural rigidity is given in. ## Pulsatile flow In order to test the model in the most critical conditions, we target high velocities typical of cardiac valves such as the aortic valve. The flow is pulsatile and corresponds to a normal cardiac output of 5.5 L min^-1, a beat rate of 72 bpm and an aortic pressure of 100 mmHg. This frequency gives a peak velocity of around 0.9 m s^-1. In the simulation, we force the flow by means of a sinusoidal pressure *P* gradient $$\frac{dP}{dx} = A\text{sin}(\frac{2\pi}{T}t),$$ where *A* is the force amplitude and *T* the period. To obtain this pressure gradient in the simulations, we impose to each liquid particle the acceleration $$g = g_{0}\text{sin}(2\pi ft),$$ with *g*_*0* = 500 m s^-2 and oscillation frequency *f* = 1/*T* (*T* = 1 s). Under these conditions, we reach high velocities but the flow remains laminar. We focus on the laminar regime for two reasons: (i) blood flow under normal conditions is laminar, and (ii) we want to test, at this stage, the accuracy of our model without dealing with the additional complexity of turbulence. ## Dimensionless analysis In Section *Membrane deformation*, we compare our simulations with experimental data. The comparison is based on specific dimensional groups that are defined in this section. Dimensional analysis bring to three fundamental groups Re (Reynolds Number), N_f (dimensionless frequency) and Λ (geometric ratio), defined as $$\text{Re} = \frac{\rho UZ}{\mu},$$ $$N_{\text{f}} = \frac{\rho f^{2}d^{5}}{F/L},$$ and $$\Lambda = \frac{Z}{L},$$ where *ρ* is the density of the fluid, *U* is a reference velocity (here we use the max velocity in the channel), *Z* the half-thickness of the channel, *μ* the fluid viscosity, *f* the oscillation frequency, *d* the membrane thickness, *F* the flexural rigidity, and *L* the length of the membrane. The computational particles used in our simulations are point particles; strictly speaking, they do not have an actual thickness. Their thickness is the result of the repulsive forces acting on the particles to impose no-penetration boundary conditions. The value of *d* in, therefore, is calculated from *k*_*a* and *k*_*b* as discussed in. The value of the Young modulus *E* does not compare explicitly in any of the dimensionless numbers above; this is due to the fact that *F* and *E* are interchangeable as discussed in. In theory, we should also account for another dimensionless group based on *R* (radius of the convex area). In practice, however, this group is not necessary as discussed in Section *Membrane deformation*. Each dimensionless number provides specific information about the geometric constants and the physical forces acting in the system. Re indicates the extent of the inertial forces with respect to the viscous forces. N_f indicates the membrane resistance with respect to the stress generated by the oscillating flow. Λ indicates the geometric ratio between the channel thickness and the membrane length. In comparing the simulations with the experimental data (see Section *Membrane deformation*), we found that the group $$\text{N}_{\text{R}} = \text{Re}.\text{N}_{\text{f}} = \frac{\rho^{2}f^{2}d^{5}UZ}{\mu F/L}$$ is particular relevant. This dimensionless number compares the effect of the forces that tend to deform the membrane (numerator) with those that tend to oppose the deformation (denominator). In Section *Membrane deformation*, we show that different geometries and flow conditions generate the same type of membrane deformation if N_R is the same. # Results There are two types of parameters required for the simulations: model parameters and simulation parameters. The first group consists of internal parameters used by the SPH and MSM solvers; the second refers to the operative conditions. This Section focuses on the second group (i.e. *Z*, *L*, *R*, μ, *ρ* and *F*); the internal parameters (e.g. *k*_*a*, *k*_*b*, number of particles, time step, smoothing length, etc.) can be found in. The geometric parameters *L*, *Z*, *R* are given in Section *Geometry* (see also ). All the simulations assume blood as liquid medium. Blood is a viscoelastic fluid, but in flow simulations is often considered Newtonian. In our calculations, we also use the Newtonian approximation with *ρ* = 1056 kg m^-3 and *μ* = 0.0035 Pa s. We consider membranes with different flexural rigidities, the specific value of *F*, for each case, is given in Section *Membrane deformation*. This section is divided in three parts. The first part is dedicated to the flow, and we validate our results against traditional CFD simulations. The second part is dedicated to the membrane, and we validate our results against experimental data. The third part focus on the formation of solid aggregates, and highlights the main advantages of the DMHS in modelling biological valves. ## Hydrodynamics We compare results obtained with our model with traditional CFD simulations performed with Abaqus 6.14^® with the same geometry and under similar flow conditions. In these simulations, the membrane is fixed in order to focus solely on the hydrodynamics. This is done on purpose: if at this stage we had considered both the fluid and the membrane together, we could not have distinguished whether potential errors originated from the fluid dynamics or the membrane mechanics. Calculations are run at two constant inlet velocities, 0.2 m s^-1 and 0.9 m s^-1. Because of the different nature of the two modelling techniques, the inlet/outlet boundary conditions (b.c.) are not the same. The DMHS uses periodic inlet/outlet b.c., while in the CFD simulation the inlet has constant velocity and the outlet fixed pressure. shows the CFD results; shows the DMHS results. Comparison between Figs and shows a good agreement between the CFD and the DMHS calculations. Both models, in particular, capture the recirculation zones in the circular chamber in the centre. For the velocity, minor differences (2–5%) can be found at the tip of the valve. These differences depend on the different inlet conditions between the DMHS and the CFD model and the nature of the discretization method (particles vs. mesh). Another important variable often reported in literature is shear stress (Figs). But also in this case CFD and DMHS results are similar. Both models, in particular, identify a region of high stress near, but not exactly at, the end of the leaflets. ## Membrane deformation In this section, we account for the flexibility of the leaflets and calculate both the flow and the membrane dynamics. For validation purposes, we compare the membrane deformation observed during the simulations with those obtained experimentally by. In both simulations and experiments the valve has two leaflets and the flow is pulsatile. The geometric conditions, however, are not exactly the same: in, in fact, *L* = 0.0263 m, *Z* = 0.015m, and the channel is straight without the circular chamber in the centre (this is why, in Section *Dimensionless analysis*, we did not introduce a forth dimensionless number). Additionally, our simulations are 2D and based on blood, while employ water in a rectangular channel with depth *w* = 0.05 m. gathers all the parameters used in the simulations and in the experiments. The values of *k*_*a* and *k*_*b* corresponding to a specific *F* in the simulations are given in. The values of *d* for the simulations are calculated according to the procedure described in. After running a large number of simulations and observing how the membrane deforms under various flow conditions, we realized that the fundamental group that affects the membrane dynamics is N_R. Therefore, we chose specific values of *f*, *g*₀ (which gives *U*), *k*_*a* and *k*_*b* (which give *F* and *d*) to obtain in our simulations the same N_R of the experiments. We consider three cases that we call, soft membrane, intermediate membrane, and hard membrane. We start with the case of the intermediate membrane: the ‘normal’ case, which subsequently is compared to the soft and the hard membrane. shows the comparison between simulations and experiments in the case of the intermediate membrane. The overall dynamics is very similar, but there are same noticeable differences. In the simulation, the maximum opening of the membrane is wider. This is due to the absence of the central chamber in the experimental set-up. Another minor difference occurs at the end of the cycle when the backpressure closes the valve completely. When closed, the experimental valve has a more elongated shape since the leaflets are longer. This is a consequence of the fact that, besides the main group N_R, also Λ has a (minor) effect on the membrane. Experiments and simulation have the same N_R, but not exactly the same Λ; some (minor) differences, therefore, are expected. shows the comparison between simulations and experiments in the case of the soft membrane. The soft membrane can be considered defective since its position is completely reversed by the backflow. As in the previous case, there are some minor differences, but overall the membrane behaviour is well captured by the model. Similar deformation profiles have also been by other studies. shows the comparison between simulations and experiments in the case of the hard membrane. Also the hard membrane can be considered defective since it does not completely opens. In this case, the comparison focuses on the membrane’s tip. At this location, the two leaflets slide one over the other and symmetry is lost. This phenomenon is captured in both the simulations and the experiments. The loss of symmetry suggests that the simulations should account for the whole geometry and not only half of it (considering only one leaflet). Besides validating our model, this section also highlights the importance of N_R. The values of Re, Λ and N_f between the simulations and in the experiments are different, but, since N_R is the same, the membrane behaviour in both cases is similar (with the little caveat about Λ as discussed above). ## Formation of solid aggregates This section introduces solid aggregation in the DMHS. We consider two cases: solid deposits at the membrane surface, and formation of aggregates in the main flow. We generally indicate the first case as ‘calcification’ and the second as ‘clotting’. Our focus, however, is not to the formation and the evolution of actual calcifications and clots. These are very complicated biochemical phenomena and their full dynamics is beyond the scope of this article. The goal is here to illustrate how, given a criterion for aggregation, this can be implemented in our model. Once this has been achieved, more complicated agglomeration models can be implemented. Both calcification and clotting imply the formation of solid aggregates developing from the liquid. In the DMHS framework, this can be achieved by changing the forces acting on certain particles from SPH to MSM. describes the algorithm used in the simulations. The procedure starts from an agglomeration seed. In our simulations, the seed is chosen arbitrarily, but it can depend on a specific criterion; for example, when local shear stress exceeds a threshold value, the particle at that location becomes a seed. Once the position of the seed is known, the algorithm propagates the agglomerate. Every *N* time-steps, it identifies all the particles within a distance *R*_MAX from the seed, and, with a certain probability, transforms some of the liquid-particles in solid agglomerate-particles by (i) changing the forces acting on the particles from SPH to MSM, (ii) and creating a bond between the seed and the newly created agglomerate-particle. The strength of the new bond determines the material properties of the agglomerate. In our simulations, the probability of transforming a liquid-particle in an agglomerate-particle has been related to a fixed value, but, as mention before, it can be associated to a specific criterion (e.g. shear stress threshold). The algorithm repeats the above procedure iteratively to propagate the agglomerate further and new agglomerate particles create bonds only to other fluid particles and not to existing agglomerate particles. At the next time step, the previously generated agglomerate-particles become seeds; these seeds create new agglomerate-particles and so on as illustrated in. We can affect the final shape of the agglomerate by changing, as time progresses, the behaviour of the seeds. If the seeds are active all the time, they continue to create new agglomerate-particles around them until they are fully surrounded by the agglomerate. In this case, the overall shape of the agglomerate tends to be circular. If the seeds remain active only for one time step, the agglomerate propagates in one preferential direction and tends to assume a filiform (thread-like) shape. shows three types of simulations where the algorithm is applied to three different configurations. The simulation parameters (*N*, *R*_MAX, agglomeration probability, etc.) for all three cases are gathered in. With the goal of obtaining a sizable aggregate in a few cycles, we accelerated the agglomerate formation by using higher aggregation probabilities. This is a typical technique used to study phenomena with very different timescales as those occurring in pipelines erosion. In the first case (called ‘calcification’), the initial seed is located in the region between the membrane and the wall and the deposit propagates following the circular agglomeration algorithm illustrated in. An interesting feature of the simulation is that, as time progresses, the agglomerate makes increasingly difficult the movement of the lower leaflet until it stops almost completely. In the second case (called ‘free clot’), the initial seed is located in the flow and the agglomerate propagates following the circular algorithm. The presence of a solid aggregate alters the hydrodynamics as indicated in. Once a liquid particle transform into a solid particle, the fluid streamlines must change direction to account for the new solid-liquid boundaries. This feature would not be possible with mesh-based algorithms and highlights one of the advantages of discrete multi-physics. The third case (called ‘filiform clot’) is similar to the previous case. This time, however, the initial seed is located at the tip of the leaflet and the agglomerate propagates according to the filiform algorithm. As the filiform aggregate grows, it moves alternately on the right and on the left of the membrane due to the oscillating flow. We can emphasize another advantage of using a particle-based technique by introducing fragmentation. The drag between the fluid and the agglomerate creates internal stresses in the solid. These stresses tend to pull apart the agglomerate-particles that respond with a stronger binding force (according to equation J). At this point, we can slightly modify the algorithm and introduce a criterion for break-up: if the force between two agglomerate-particles exceeds a certain value, the bond breaks. In, we used a threshold force of 1.3·10⁻⁷ N. At a certain point of the simulation, the threshold force is exceeded and the agglomerate breaks in two parts. One part remains attached to the leaflet; the other becomes free and moves unrestricted with the flow. # Conclusions Mesh-free methods are usually considered viable alternatives to traditional modelling, but have never enjoyed the same popularity of mesh-based techniques. Many mesh-free methods have been developed only in relatively recent years and offer, to the potential user, less available information, experience and software. On the other hand, a specific sub-set of mesh-free algorithm (e.g. SPH, DEM, CGMD, BD etc.) share a common particle-based framework that makes particularly easy their linkage in multi-physics problems. We call this approach *discrete multi-physics* and, in this paper, we show that, in certain circumstances, it is more than a mere alternative to traditional modelling. Discrete multi-physics can tackle, with relatively little effort, problems that are considered very challenging with mesh-based multi-physics. Elsewhere, we focused on solid-liquid flows where the dispersed phase is made of deformable, breakable, dissolving, melting or solidifying particles. Here, we apply the same approach to biological valves including the formation of solid aggregates in the flow and at the membrane surface. To the best of our knowledge, this is the first study to directly account for the hydrodynamics, the membrane deformation and the formation of solid aggregates at the same time and, as such, it has the potential to open a new prospective to the modelling of biological valves. # Supporting information [^1]: The authors from LivaNova didn't provide any financial support to these works. Commercial affiliation of authors from LivaNova does not alter our adherence to PLOS ONE policies on sharing data and materials. [^2]: **Conceptualization:** MA M. Bussone FG AA. **Data curation:** MA AA. **Formal analysis:** MA MHA M. Bussone FG FB M. Barigou AA. **Funding acquisition:** M. Barigou AA. **Investigation:** MA MHA AA. **Methodology:** MA AA. **Project administration:** M. Barigou AA. **Resources:** M. Bussone FG FB M. Barigou AA. **Software:** MA MHA. **Supervision:** AA. **Validation:** AA. **Visualization:** MA AA. **Writing – original draft:** MA AA. **Writing – review & editing:** MA AA. [^3]: ‡ These authors also contributed equally to this work.

# Introduction According to the World Health Organization (WHO), more than five million deaths per year are the result of direct use of tobacco, whereas more than 600 000 second hand smokers also perish from cigarette exposure (World Health Organization, 2014). According to the US Centers for Disease Control and Prevention, cigarette smoking is the dominant risk factor for the development of chronic obstructive pulmonary disease (COPD) and emphysema. A wide range of dangerous agents are found in cigarette smoke and, aside from solid particles, it includes more than 4000 chemicals, of which at least 250 are known to be highly harmful and more than 50 are tumorigenic. Altogether, these factors may induce airway inflammation, cellular recruitment, lung fibrosis, mucus hypersecretion and also cancer. Among the features of cigarette smoke-induced pathology, parenchymal fibrosis and emphysema may be considered the most characteristic and also the most deleterious, resulting in a significant organic impairment and restricted life quality of the patients. Lung inflammation induced by cigarette smoke is characterized by an initial phase of inflammatory cells recruitment to the parenchymal space, matrix metalloproteinase (MMPs) activation, extracellular matrix degradation and tissue damage. This is followed by the intense release of pro-inflammatory cytokines, such as IL-1 and IL-6, chemokines as CCL2 and CXCL1 and also lipid mediators, as PGI and LTB_4, as reviewed. These molecules are able to recruit more inflammatory cells, as neutrophils, macrophages and also CD4 and CD8 T lymphocytes of the Th1 and Tc1 IFN-γ-secreting subtypes, respectively, perpetuating lung tissue damage. Chronic exposure to cigarette noxious agents leads to the perpetuation of the inflammatory and fibrotic processes, with more and more infiltrating cells recruited, progressive elastin and collagen degradation, followed by parenchymal destruction and finally the establishment of lung emphysema. Thus, campaigns to prevent cigarette smoking and to alert the population about its morbidity are extremely important. However the need for effective and also less expensive therapeutic approaches for the treatment of cigarette smoke-induced emphysematous patients is unquestionable. In this context, mesenchymal stromal cells (MSCs) therapy seem very promising, not only due to its regenerative capacity and immunomodulatory features, but also due to its low cost and relatively easy manipulation, as reviewed. Adult MSCs are typically defined as undifferentiated multipotent cells endowed with the capacity for self-renewal and the potential to differentiate into several distinct cell lineages, as recently reviewed. These cells may be obtained from different organs and tissues, as bone marrow, skeletal muscle, adipose tissue, dental pulp, umbilical cord, fallopian tubes and other tissues. It is noteworthy that MSCs are well tolerated *in vivo*, either under syngeneic or xenogenic conditions. Concerning its immunomodulatory function, MSCs are known to: i) secrete anti-inflammatory cytokines; ii) express reduced levels of MHC and costimulatory molecules; iii) express the tryptophan-depleting enzyme indoleamine-2,3-dioxigenase iv) induce T regulatory cells and many others. Altogether, these mechanisms may be useful tools for the treatment of several chronic inflammatory diseases such as multiple sclerosis, arthritis, diabetes, lupus, and also COPD and emphysema. Moreover, the possibility to use associated approaches able to boost the overall response must be investigated. In this sense, Low Level Laser Therapy (LLLT) seem to be an interesting approach, specially due to its non-invasiveness, lack of secondary effects, low cost and ability to promote stem-cell proliferation *in vitro*. LLLT has already shown interesting results for different human diseases, as oral mucositis, coronary disease and also in experimental models, as for muscle dystrophy, asthma and articular inflammation in mice. Actually, our group has recently demonstrated that LLLT is able to reduce lung inflammation both during the asthma model as well as secondary to gut ischemia Among the possible mechanisms proposed, we have recently demonstrated that alveolar macrophages irradiated with LLL had augmented AMPc synthesis, and this is able to reduce NF-κB activation and thus IL-6 and IL-1 secretion. In this context, we decided to evaluate the effectiveness of associating human tubal mesenchymal stromal cells (htMSCs) and LLLT using the murine model of cigarette smoke-induced COPD. It is worthy to mention that we have previously reported that htMSCs display stemness properties and also miogenic and adipogenic capacity to differentiate. Here, our results demonstrate that the co-therapy with htMSCs and LLLT are very effective in lowering important pulmonary inflammatory parameters, as cytokine secretion, cellular infiltrate of leukocytes, mucus secretion, collagen deposition and the activation of important transcription factors, as NF-κB and NF-AT. Altogether, we show that the association of stem cells therapy with LLL irradiation are clinically beneficial, and thus, may be considered as further interesting therapeutic approach. # Materials and Methods ## Cigarette Smoke–Induced Emphysema and Treatment Female C57BL/6 mice (6–8 weeks) were exposed to cigarette smoke during 75 days by whole body exposure using an adapted protocol from Biselli et all. All experiments were performed on day 76. Briefly, animals were exposed to 14 commercially available cigarettes (containing Alcatrão 13mg, nicotine 1,10 mg and carbon monoxide 10 mg each) divided on 7 cigarettes in the morning and 7 cigarettes in the afternoon and each session lasted for 30 minutes. LLL irradiation with diode laser (30mW/3J at 660 nm) was performed twice a day from day 60 to day 75 (one hour after each cigarette exposure) during 180 seconds on the skin over the right upper bronchus with a spot size of 0.785 cm². htMSCs (1x10⁶ cells) were infused on days 60 and 67. Experimental groups were divided in 7 groups, as follow: 1) Basal 2) COPD 3) COPD + LLL 4) COPD + MSCs (i.p) 5) COPD + MSCs (i.n) 6) COPD + MSCs (i.p.) + LLL 7) COPD + MSCs (i.n.) + LLL. We declare that all experiments were approved by the UNINOVE animal research committee (Comitê de Ètica no uso de Animais—CEUA—#AN005/2013) and human research committee (Comitê de Ética em Pesquisa com Seres Humanos (CEPSH) of University of Sao Paulo (# CEP-IBUSP-106/2010). ## htMSCs Isolation and Culture Human tubal tissue was obtained from women (age 35–53 years) submitted to hysterectomy or tubal ligation/resection surgery. Samples were collected during the proliferative phase from fertile women who had not undergone exogenous hormonal treatment within the last three months. htMSCs experiments were approved by both the ethics committee of the Biosciences Institute of the University of São Paulo and the UNINOVE–Universidade Nove de Julho. Each sample was collected in HEPES-buffered Dulbecco Modified Eagle Medium / Hams F-12 (DMEM/F-12; Invitrogen, Carlsbad, CA) with 100 IU/mL penicillin (Invitrogen) and 100 IU/mL streptomycin (Invitrogen, Carlsbad, CA), maintained at 4°C and processed within 24 hours period. All samples were washed twice in phosphate saline buffer (PBS, Gibco, Invitrogen, Carlsbad, CA), finely minced with a scalpel and transferred to a 50 mL tube, and dissolved with collagenase IV (Sigma-Aldrich) at 0,1% diluted in PBS (Invitrogen) for 15 minutes, at 37°C, in a water bath. Further, samples were washed with 10 mL of DMEM/F-12 and added with 15 ml of pure TripLE Express, (Invitrogen, Carlsbad, CA) for 15 minutes at 37°C in a water bath. Subsequently, TripLe was inactivated with DMEM/F-12 supplemented with 20% FBS, 100 IU/mL penicillin and 100 IU/mL streptomycin, and pelleted by centrifugation at 400g for five minutes at room temperature. Supernatant was removed with a sterile Pasteur pipette. Cells were then plated in DMEM/F-12 (5mL) supplemented with 20% FBS, 100 IU/mL penicillin and 100 IU/mL streptomycin in plastic flasks (25cm²), and maintained in incubator with controlled humidified atmosphere of 5% CO₂ at 37°C. The medium used for expansion was initially changed every 72 hours and routinely replaced twice a week thereafter. Data concerning osteogenic, chondrogenic and adipogenic differentiation, as well as cellular phenotype were previously published by our group demonstrating the stem cell potential of this population. Human tubal tissue was obtained after written and signed informed consent approved by Comitê de Ética em Pesquisa com Seres Humanos (CEPSH) of University of Sao Paulo (# CEP-IBUSP-106/2010) and maintained at the Division of Human Genome Research Center, Biosciences Institute, University of São Paulo, São Paulo–SP—Brazil. ## Bronchoalveolar Lavage Fluid Cells and Cytospin After euthanasia, bronchoalveolar lavage fluid was obtained as previously described. Briefly, 1.5 mLs of PBS as intra-tracheally injected in the lungs and 1 mL was re-collected and centrifuged at 450 g 4°C during 5 minutes. Supernatants were discarged and the pellet suspended in the desired amount of PBS 2% FBS. Total cell count was performed in Newbauer chambers and differential cell count after cytospin protocol. For cytospin, aliquots of 100 μL were centrifuged at 300 g for 5 minutes. Samples were stained by May-Grunwald-Giemsa method and 300 cells per sample were counted on a blind fashion. ## Lung Mononuclear Cells and Flow Cytometry Lungs were obtained after right heart perfusion of 10 mLs of cold PBS. The tissue was minced with scissors and incubated for 45 minutes at 37°C with 2,5 mg/mL of Colagenase D (Roche) in HBSS. To stop collagenase, cells were suspended in Ca²⁺/Mg²⁺ free HBSS and centrifuged at 450 g and 4°C for 5 minutes. Cellular pellet was suspended in 5–6 mLs of 37% Percoll (GE) in HBSS and gently laid over 5–6mLs of 70% Percoll in HBSS using 15 mLs conical falcon tubes. Samples were centrifuged at 950 g and 4°C for 30 minutes without breaks. Cell-containing ring was collected from the percoll gradient interface and suspended in PBS 2% FBS and centrifuged again at 450 g and 4°C for 5 minutes. Further, cells were suspended in PBS 2% FBS, counted and used as desired. For flow cytometry analysis, 5x10⁵ cells were first incubated with 0,25 μg of anti-CD16/32 at 4°C for 20 minutes to avoid inespecific binding. Cells were then stained with 0,25 μg of anti-CD4 APC and 0,25 μg of anti-CD8 PE in 25 μL of PBS 2% FBS for 20 minutes at 4°C. In the next step, cells were washed twice with 200 μL of PBS 2% and finally suspended in paraformaldehyde 1%. Cells were acquired in the flow cytometer Accuri C6 (BD Biosciences). 5x10³ T CD4⁺ events were collected after elimination of cell doublets by FSH-height x FSC-area plot analysis. ## Cytokine Secretion BAL lavage fluid was centrifuged at 450 g and 4°C during 5 minutes. Supernatants were used to evaluate the presence of IL-1β, IL-6, IL-10, TNF-α, IFN-γ and KC by the ELISA method according to manufacturer instruction (R&D System). ## Histomorphometric Analysis of Inflammation, Collagen and Mucus After perfusion, lung samples were maintained in formaldehyde 4% for inclusion in paraffin. Slices were stained with Siriud Red for collagen detection or Periodic Acid Schiff (PAS) for the detection of mucus. Were analyzed 15 airways per each animal under a 400X magnification. H&E staining was used for the analysis of peribronchial infiltrate and mean linear intercept measurement. Morphological features were analyzed using the Image Pro Plus 4.5 (Media Cybernetics, Rockville, MD, USA). ## Peribronchial Inflammation The area between the airway basal membrane and the airway adventitia was quantified using the software Image Pro-Plus and the number of mononuclear and polymorphonuclear cells was quantified in this area according to the morphological criteria. The results were expressed as the number of mononuclear and polymorphonuclear cells per square millimeter. ## Mucus Production and Collagen Fibers Deposition For the analysis of mucus production, the epithelium area of the whole airway, 15 airways per mouse were quantified. The positive stained area in the epithelium area was quantified and the results were expressed as the percentage of positive epithelial area. For the analysis of collagen fibers deposition in the airways wall, the area between the outlayer of epithelium until airway adventitia were measured and the amount of positive stained area was calculated. The results were expressed as percentage aof collagen fibers in the airway wall. ## Immunohistochemistry For immunohistochemistry analysis, all samples were previously submitted to deparaffinization. Endogenous peroxidase activity was blocked with H₂O₂ 3% three times during 10 minutes. Samples were then washed and blocked with bovine serum albumin 10% during 1 hour and then incubated with primary rabbit anti-mouse IgG antibodies: i) anti-NFκB (Santa Cruz, CA) at 1:500; ii) anti-NF-AT 1:500 (Santa Cruz, CA) and iii) anti-IL-10 (Santa Cruz, CA) at 1:500 during 2 hours at room temperature. Samples were washed twice with TBS- BSA 10% and then incubated with secondary antibody goat anti-Rabbit IgG at 1:1000 for 1 hour. After washing samples twice with BSA 2% diaminobenzidine (DAB) was added. Finally samples were washed again and counter stained with H&E. Slides were analyzed under light microscope on a blind fashion. ## Statistical Analysis All analysis were performed using the GraphPad Prism software (GraphPad Software, Inc). For parametric analysis we used one-way ANOVA followed by Tukey post-test whereas for non-parametric analysis, we used Kruskal-Wallis followed by Dunn´s post-test. All groups were compared to COPD control group. Differences were considered significant when p\<0,05. # Results ## htMSCs and LLLT Reduce Bronchoalveolar Lavage Fluid Cellularity in the Lungs of COPD Animals To evaluate whether htMSCs and/or LLLT were effective in reducing lung inflammation after cigarette smoke-induced COPD, we first analyzed BAL fluid cellularity from all experimental groups for total and differential cell counts. Interestingly, a significant reduction in total cell counts, macrophages and neutrophils was observed only when htMSCs were associated with LLL therapy, although a trend is observed in all other groups, even for LLL alone. However, we had a discrepant result for lymphocyte counts, as we observed that only htMSCs by i.p. delivery was able to reduce its amount. Association with LLLT did not changed the result. It is worthy to mention that the lowest amount of total cells, macrophages and neutrophils was observed in the group of htMSCs associated with LLLT, either i.p or i.n, demonstrating that association with LLLT is more important for these features than the route of htMSCs delivery. It is also noteworthy that such reduction in BAL cellularity reached levels no different from naïve controls. Moreover, intraperitoneal delivery of htMSCs seemed more effective than intranasal delivery in reducing neutrophil and lymphocyte infiltration. Next, we performed flow cytometry analysis in the BAL fluid from all experimental groups to evaluate changes in the frequency of CD4⁺ and CD8⁺ T cells in the lungs. This could be indicative, either of a reduced T cell activation in the lymph nodes, or an impairment in cell migration to the target organ. In fact, corroborating findings from for polymorphonuclear and total cells, the frequency of both T CD4⁺ and T CD8⁺ cells were significantly reduced irrespective of the therapeutic approach used, i.e, htMSCs alone or in association with LLLT. However, no differences were observed, when lung tissue was analyzed for perivascular infiltrate by immunohistochemistry. ## htMSCs and/or LLLT Reduce Inflammatory Cytokines in the BAL Fluid of COPD Animals We next analyzed the amount of cytokines as IL-1β, IL-6, TNF-α, IFN-γ and the chemockine KC, to evaluate whether reduced cellularity in the lungs of our experimental culminated also in reduction of inflammatory mediators. Our results show that IL-1β was reduced when htMSCs either ip. or i.n. were associated with LLLT, although htMSCs i.n alone also reached significance. The same pattern was observed for TNF-α. KC, an important chemokine for the recruitment of neurotrophils, was reduced in all groups irrespective of the treatment. Interestingly, IL-6 was reduced only when htMSCs were associated with LLL whereas no difference was observed for IFN-γ Very interestingly, the anti- inflammatory cytokine IL-10 was up-regulated only with LLL irradiation alone. ## htMSCs and/or LLLT Reduced Mucus Secretion and Alveolar Enlargement in the Lungs of COPD Animals Another important feature in the pathophysiology of COPD is the intense secretion of mucus, resulting in reduced airway lumen, airflow and breathlessness. Thus, we next evaluated whether either htMSCs or LLL therapies, associated or not, were able to reduce the amount of mucus observed in the airways. In fact, our data demonstrated significantly decreased amount of mucus in the bronchi of all experimental animals under therapeutic protocols, with the exception of the LLLT alone. Significant differences reached from 10 to 15-fold reduction when experimental groups were compared to COPD control. On the other hand, when associated with htMSCs, either nasally or intraperitoneally, LLL therapy restored mucus secretion to basal levels, with no difference from naïve animals. We next decided to verify some possible effects of htMSCs and LLLT in the enlargement of lung parenchyma, which reflects the destruction of the alveolar septa. As indicated by, mean linear intercept measures were significantly reduced in the COPD + MSC(ip)+LLLT group when compared to control animals, indicating reduced lung damage. ## Collagen Deposition Was Reduced Only When htMSCs Were Associated with LLL Aside from mucus secretion, collagen deposition is also considered an important marker for COPD. Therefore, we also evaluated the amount of peribronchial collagen deposition through Sirius Red methodology. Very surprisingly, a significant collagen decrease of around 4-fold was observed only when LLLT was associated with htMSCs intranasally. Although there was a trend for the htMSCs (i.p) + LLLT, none of the other groups reached statistical significance. ## htMSCs and/or LLLT Reduces NF-κB and NF-AT Transcription Factors Expression in the Lungs of COPD Animals Most pro-inflammatory cytokines evaluated, such as IL-1, IL-6 and TNF-α, have their transcription under the control of the transcription factor NF-κB. Moreover, these cytokines may also signal through NF-κB after engagement with its receptors cognate recptors. NF-κB is found in a great variety of cell types, as macrophages and neutrophils, but also T lymphocytes. Thus, with the aim to correlate the overall reduction in BAL cellularity and pro-inflammatory cytokine secretion with a reduction in NF-κB expression, we sought to evaluate its expression by immunohistochemistry of the perialveolar space of COPD animals. Corroborating the reduction in pro-inflammatory cytokine secretion, all groups displayed significantly reduced NF-κB staining in the target tissue when compared to the COPD control group. Differences vary from 2 to 3-fold depending on the treatment. Another transcription factor also important for the activation of the immune cells is the NF-AT, which is mostly expressed on T cells, both CD4⁺, CD8⁺ and NK cells, and also in the lung tissue, as shown by the literature. NF-AT is the main transcription factor involved in the synthesis of IL-2 and its active form is the non-phosphorilated state, as reviewed. It was very surprising to notice that LLLT by itself was already able to significantly reduce NF-AT activation. This reduction was no different from that observed when LLL was associated with htMSCs i.n.. ## LLL Therapy Boosts IL-10 Secretion in Lung Tissue In order to confirm the data obtained by the ELISA method, and concerning its importance as an anti-inflammatory factor, we also evaluated by immunohistochemistry the amount of IL-10 in lung tissue of the experimental groups. Interestingly, LLL irradiation *per se* was able to increase the amount of epithelial IL-10 whereas htMSCs i.p. was not able to so. Consistently, this was also observed only when htMSCs were associated with LLL therapy, although htMSCs i.n. alone could also up-regulate this cytokine. # Discussion In the present research we show that the treatment with htMSCs associated with LLL irradiation significantly reduced lung inflammation in mice with cigarette smoke-induced COPD. Many of the evaluated parameters, such as BAL cellularity, pro-inflammatory cytokine secretion, perivascular infiltrate and the presence of inflammatory transcription factors, as NF-κB and NF-AT were significantly reduced. This was associated with a better maintenance of tissue integrity when compared to COPD untreated controls, evidenced by reduced mucus secretion, collagen deposition and tissue damage. It is worthy to mention that these features are greatly responsible for tecidual destruction and further decline in patient´s life quality, as it greatly reduces airflow, lung complacence and lowering pO₂. Moreover, our findings also demonstrate that the route of administration of htMSCs, i.e. intraperitoneal vs. intranasal, were both effective in suppressing the disease, although with some peculiarities. For instance, intranasal delivery was more effective in reducing the presence of NF- AT as well as collagen deposition when associated with LLL therapy, which was not observed after intra-peritoneal injection. This might be explained by a local regenerative/suppressive mechanism, as the cells were directly delivered to the lungs. For intra-peritoneal route, however, immune modulation on lymphoid organs, specially mediastinal lymph nodes is more likely. Cigarette smoking-associated diseases, which may result in severe decrease of life quality represent an important social and economical burden, as billions of dollars are spent each year in the treatment of emphysematous patients worldwide. Despite the constant campaigns against tobacco, it is still the fourth leading cause of death in the United States. Therefore, the need for a more efficient and yet cheaper therapeutic approach in the management of COPD patients is unquestionable. Cell therapy with MSCs has shown clinical benefits by us and other groups in conditions such as experimental autoimmune encephalomyelitis (EAE), multiple sclerosis, chronic renal failure, ischemic cerebral stroke, myocardial stress and even for COPD. In fact, several of our findings were corroborated by previous research, in which lung inflammation, weight loss and lung integrity were ameliorated after adipose-tissue MSCs treatment. On the other hand, a recent clinical trial on MSCs have not shown promising results, as COPD patients treated with MSCs (Prochymal) had not shown improvement of lung function, as shown for forced expiratory volume (FEV₁) and forced volume capacity (FVC). Interestingly however, it seemed that, corroborating our findings, lung inflammation was decreased, inferred by the lower level of C-reactive protein. Our report however, although using an experimental model, is the first to associate htMSCs with LLL irradiation, considered a promising approach. COPD is a chronic and obstructive disease of the lungs, result of the chronic exposure of the airways to the noxious agents found in cigarette smoke and droplets. It has been established that lung destruction directly correlates with the amount of total particulate matter found in each cigar. Initial exposure leads to cellular infiltrate of neutrophils and blood-derived monocytes secreting pro-inflammatory and pro-fibrotic cytokines, as IL-1, IL-6, IL-12, TNF-α and TGF-β, along with chemokines, lipid mediators and several other molecules. This is greatly accompanied by extracellular matrix degradation by MMP-1 (matrixmetalloproteinase-1) secreted by alveolar macrophages causing tissue destruction and emphysema. Chronic exposure leads to intense T CD4 and T CD8 lymphocyte infiltration, and IFN-γ is probably the most important T cell- derived cytokine. It is relevant that IFN-γ-secreting T CD8 cells are also very important in the late phase of the disease. Due to its pro-inflammatory function, and in association with other cytokines, as TNF-α, TGF-β, IFN-γ induces important activation of both immune and parenchymal cells of the lung and finally leading to fibrosis and parenchymal destruction. In this context, therapeutic approaches with the capacity to dampen such activation could greatly contribute to the maintenance of lung integrity and both MSCs and LLL therapy independently had already proven it. However, we did not observe differences for IFN-γ. A growing body of evidence have shown several different immunosuppressive mechanisms of MSCS, such as: lowering the expression of MHC and costimulatory molecules; up-regulation of indoleamine-2,3-dioxigenase and FAS-L expression; expansion of Tregs; block of IFN-γ and IL-17 secretion, reduces tecidual caspase-3 activation and many others, as reviewed. Altogether, these mechanisms greatly impair immune cells activation, further avoiding or reducing tissue inflammation and destruction. However, although we may only speculate the mechanisms used by the htMSCs + LLL irradiation to suppress COPD, it is suitable to think that many of the aforementioned mechanisms may be taking place in our system. Corroborating our findings, several other groups have already addressed the capacity of MSCs in modulating chronic diseases, including COPD. Bone marrow- derived MSCs were shown to significantly reduce tissue destruction in COPD mice by a mechanisms dependent on VEGF. Consistently, the group observed among many other cytokines, reduction in IL-1β and IL-6, in accordance with our data. More relevant was the fact that lung function, including inspiratory capacity and vital capacity were significantly higher when compared to control animals, which is again discrepant from data obtained in humans. Adipose tissue-derived MSCs (ASCs) transplantation was also shown to be effective in improving COPD in mice reducing lung infiltration of inflammatory cells, as neutrophils and macrophages reduced associated with parenchymal destruction, as evaluated by active caspase-3. In line with our findings, the group also demonstrated reduced lung infiltration of inflammatory cells, as total cells, neutrophils and macrophages. Associated with its immunosuppressive activity, it is also possible that MSCs exert some regenerative/reparative function *in situ*. For instance, intra- tracheal delivery of bone marrow-derived stem cells after hyperventilation- induced injury in rats greatly improved lung recovery, reducing cellular infiltrate of neutrophils and lymphocytes, pro-inflammatory cytokine secretion, associated with normal alveolar space. In accordance, we have observed that intra-nasally treated animals have also displayed a significant reduction in several inflammatory markers, as cellular infiltrate and cytokine secretion. Moreover, we also observed amelioration in alveoli enlargement in animals treated with htMSCs(ip) + LLL, evidenced by mean linear intercept, but unfortunately lung function was not assessed by us. However, it is noteworthy that collagen deposition, a very significant factor for tissue fibrosis and lung emphysema, was significantly reduced when htMSCs were associated with LLLT. This may indicate that LLLT boosts htMSCs regerative capacity, however we may not describe the mechanisms so far. LLLT started to be studied in the late 60\`s, when reports had shown its use in improving hair growing in rats. Several other reports soon followed using different models, as wound healing and muscle regeneration. On the other hand, the effect of laser therapy is poorly understood. We have previously observed that the *in vitro* proliferative capacity of dental pulp derived mesenchymal stem-cells increased with low intensity laser application. Interestingly, recent reports have shown that LLL therapy exert its function by increasing intracellular AMPc and thus suppressing important inflammatory transcription factors as NF-κB, which is consistent with our findings. In fact, the intriguing capacity of LLL therapy in suppressing the immune response has been previously shown by us and others, as mentioned. LLLT irradiation had the capacity to reduce joint inflammation in rats treated locally with papain. Consistently, irradiated groups demonstrated reduced cellular infiltration and also reduction in IL-1β and IL-6, corroborating our findings. In the context of lung disease, we have previously published that LLLT reduces lung inflammation using OVA- induced lung inflammatory disease. In this case, several parameters, as cellular infiltrate of the lungs, cytokine secretion, mucus secretion and collagen deposition were significantly reduced after irradiation and antigen challenge. As expected, many of these results were reproduced in the present research, and also it indicated that htMSCs and LLLT may act by an additive manner. Aside from overall reduction in lung inflammation, another interesting observation was the reduction of the transcription factors NF-κB and NF-AT which could be secondary to the reduced infiltration of macrophages / neutrophils and lymphocytes in lung tissue, respectively. Moreover, we have also detected lower levels of IL-1β in BAL fluid, which is known to signal through IL-1r and thus induce NF-κB activation. However, we may not exclude the possibility that LLLT could also directly abrogate NF-AT activation, as observed when it was used alone. It is plausible to think that LLLT somehow changes intracellular Ca²⁺ levels, consequently modulating NF-AT activation by calcineurin / calmodulin. It is relevant that NF-AT, which may also be found in lung arteries, is responsible for up-regulating the expression of smooth muscle α-actin and myosin heavy chain after hypoxia, a very important mechanism for chronic hypoxia-induced pulmonary vascular remodeling. In summary, our results clearly indicate that, aside from lack of toxicity and other complications, both htMSCs and LLLT were shown to be safe for the treatment of COPD in an experimental model. More important, however, is the fact that although htMSCs and LLLT could act independently, some pathological parameters were effectively reduced when both therapies were associated. It may indicate an important additive effect that may be responsible for the overall reduction in lung inflammation and tissue destruction in cigarette smokers. Our results of the combined use of htMSCs and LLLT were effective in reducing inflammatory immune response and further overall destruction of lung parenchyma in experimental COPD. Thus, we encourage other groups to keep focus on the potential of MSCs in COPD, and also to consider the relevance of associating it with LLLT. This is not only to reinforce our data, but mainly to refine the therapeutic scheme with the aim to reach a more translational approach. # Conclusions Our research highlights the important suppressive capacity of htMSCs in reducing overall lung inflammation during COPD in mice. Besides, we also observed a beneficial additive effect when htMSCs and LLLT are associated. Reduced lung cellularity and cytokine secretion, mucus production, collagen deposition and transcription factors activation are among the downregulated parameters. Interestingly, our data also show that this phenomenom is very consistent, irrespective of the MSCs administration via, i.e. intraperitoneal or intranasal. In summary, our data highlights the possibility of using this approach for the treatment of chronic inflammatory lung diseases. # Supporting Information We would like to thank Vanessa Roza da Silva for her great support with maintaining the experimental animals and Prof. Regiane Albertini for kindly sharing the LLL irradiator. We would like to thank also Eliane Gomes for technical support. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JPSP PP CEC APLdO. Performed the experiments: MM LE SH MGN PP CEC JPSP AAdB MP ECdS WNB LBV FRG MCO-J. Analyzed the data: JPSP NOSC FA RdPV MZ APLdO. Contributed reagents/materials/analysis tools: JPSP NOSC FA RdPV MZ APLdO. Wrote the paper: JPSP MZ APLdO.

# Introduction Reef-forming oysters are habitat-structuring species in coastal and estuarine areas providing essential ecosystem goods and services to human society. Both their reef structures and suspension-feeding behaviour exert large ecosystem influences. Conservation and restoration of reef-forming oyster is therefore important to maintain ecosystem health and provide multiple ecosystem services including: (1) shoreline stabilization; (2) water quality regulation; (3) ecosystem succession; and (4) fisheries production. This implies also a sustainable management of these aquatic resources. To restore or create healthy oyster reefs, it is necessary to know the habitat requirements of the target species. The intertidal rock oyster, *Saccostrea cucullata* is the dominant oyster species living along the south-eastern coast of Bangladesh, but the natural population is under great threat for habitat deterioration caused by recent developmental activities (e.g., Matarbari power plant project, LNG import terminal in Moheshkhali Island). At the same time, oyster reef development is considered to enhance coastal resilience in Bangladesh. Successful and sustainable oyster reef development largely depends on the selection of suitable sites that support long-term growth and survival of oysters. In fact, site selection for such approach is very challenging for the coastal zone of Bangladesh. The area is very dynamic and influenced by the annual monsoonal climate. To enhance survival and growth, one requires an understanding of the complex interactions between oysters and their environment. Based on these complex relationships, a model was developed to determine suitable locations for oyster reef creation. In the present paper, we developed a habitat suitability index (HSI) model for *S*. *cucullata* in Bangladesh that can be a useful tool for coastal resource managers. A HSI model provides spatially explicit information on the relative potential of a given area to support a particular species of interest. Over 150 HSI models for wildlife species were published prior to 1990, with many others developed since then. For oysters, the HSI efforts focus on their: (1) aquaculture; (2) fishery production; and (3) restoration. To determine the reliability and utility of an HSI model, a four-step process is used consisting of development, calibration, verification, and validation. A comprehensive field monitoring program was initiated to quantify the forcing functions of the model covering all seasons along the entire south-east coast of Bangladesh. Then, based on experimental physiological data along with the data from literature, environmental factors were calibrated in order to develop the habitat suitability functions for each environmental parameter considered. Finally, model results were verified using an independent and spatially explicit population dataset. The aim of this study was to develop and test a spatially explicit HSI model for *Saccostrea cucullata* as a function of selected site characteristics that can be used to identify areas for oyster restoration and reef development. # Materials and methods ## Study area The present study was located in the south-eastern coastal waters of Bangladesh covering about 1,050 km of coastline including tidal river banks, from the Big Feni River in the west to the mouth of the Naaf River in the east (See). The area consists of rivers, streams/tributaries, estuaries, channels, coastal waters and nearshore islands. No specific permissions were required for these locations/activities, as the field studies did not involve endangered or protected species. The northern part of the study area is a regular, unbroken stretch of coastline having intertidal mudflats and submerged sand banks. More to the south, a continuous sandy beach runs from Cox’s Bazar to the southern tip of the Teknaf peninsula. The coastal areas are characterised by a subtropical maritime climate. There are four seasonal weather patterns: (1) the dry winter season (December-February); (2) pre-monsoon (March-May); (3) monsoon (June- September); and (4) post-monsoon (October-November), which are principally governed by the southwest and northeast monsoon winds. Among these four seasons, monsoon months are distinct from the non-monsoon months (see as an example). About 80–90% of the annual rainfall is confined to the monsoon months, which makes the coastal environment very dynamic, with a lot of fluctuations in biotic and abiotic conditions. During the season winter, the climate is mild and dry, with minimum air temperatures from 7–13°C and maximum temperatures from 24–31°C. The winds are predominantly north-easterly at the beginning of the winter and north-westerly at the end. May is generally the hottest month with air temperatures potentially reaching 40°C. The heavy southwest monsoon rains begin in early June and continue into mid-October. During the monsoon period, floodwaters from extended rainfall pushes the freshwater to near the coast, while salinity variations in other seasons are relatively small. The annual average rainfall varies from 1,500–3,500 mm. Semi-diurnal tides are typical in these coastal waters, with a tidal range of approximately 3–6 m during the spring tide season. Coastal water temperatures have distinct bimodal seasonal cycles with two warm and two cool seasons per year. Daily average water temperature is lowest (26.9°C) during winter months (December-January) and highest (29.7°C) during early summer months (April-May). Though the air temperature drops in winter for a short period, it has minor effect on seawater temperatures as it is buffered by the Bay of Bengal with its strong water circulations. ## Assimilation of data sets on environmental variables The most common variables utilised in HSI models for oysters are: temperature, salinity, pH, dissolved oxygen, water flow velocity, particulate inorganic matter (PIM), and chlorophyll-a (as a proxy of food for oysters, reviewed in). For the present study, we collected environmental data from 80 sampling stations representing tributaries, river mouths, estuaries, channels and nearshore waters in south-eastern Bangladesh, covering about 1000 km of coastline. To cover seasonal influences, a total of 7 surveys visiting all 80 locations were conducted during the 12 month investigation period (January 2016 to December 2016). During the non-monsoon period (October–May), sampling was carried out only in representative months (January, May and November) covering three of the four seasons (i.e. winter, pre-monsoon and post monsoon). During the monsoon period (June—September), monthly sampling was carried out to quantify the variability during monsoonal period. Therefore, to reduce the high environmental variation, two s mean datasets (monsoon and rest of the seasons, here-after called non-monsoon) were created for model application. Each survey was conducted during the full moon phase to cover the maximum tidal range and data were collected during flood and ebb tide periods to consider diurnal variations due to tides. Hand-held SCT (salinity, conductivity, temperature) and dissolved oxygen sensors (YSI model 30 and 55 respectively; YSI Inc., USA) were used to record water temperature (°C), salinity (ppt), dissolved oxygen (mg l^-1).Water pH was recorded using a hand held pH meter (Model HI98107, HANNA Instruments, Romania). Water flow velocity (m sec^-1) was measured by deploying a flowmeter (SKU 2030R; General Ocean Inc., USA) in mid flood and ebb tidal periods for 10 minutes. Concentration of total particulate matter (TPM, mg^-l) was determined from water samples as weight of residue remaining on a filter (GF/C Whatman glass microfiber with 1.2μm pore size) after drying at 60°C for 12h. After ignition of TPM filter at 450°C for 5 h, particulate inorganic matter (PIM) concentrations were determined from weight loss. The chlorophyll-a concentration (*μ*g l^-1) in water samples were determined by fluorescence meter (FluoroSense, Turner Designs, USA), calibrated by taking data from chlorophyll extraction into acetone following the procedure of Strickland and Parsons (1972). ## Model description The HSI model is composed of two life stage components: (1) the settling larval stage (at metamorphosis); and (2) the post-settlement life stages (spat and adult). Gametes, eggs and planktonic larval stages were excluded from the model as they have no habitat requirements beyond the water conditions which permits their parents to spawn. illustrates how the HSI is related to the variables and life stages of the oyster. The cycle starts at the metamorphosis, where the eyed-pediveliger larval stage that needs to settle onto a hard substrate. Ambient salinity, and the presence of suitable substrates are considered as key components for successful spatfall while high water flows can limit the settlement of oysters in turbulent waters. Temperature, salinity, pH, dissolved oxygen, PIM, and Chlorophyll-a were considered as important environmental variables for growth and survival of juveniles (= spat) and adult oysters. To calculate component indices for determining HSI, Suitability Index (SI) graphs were used that were obtained from existing literature (see and) except for salinity. Suitability Index (SI) graph for salinity are derived from empirical data from present study. Laboratory experiments were conducted to determine the influence of salinity on adult oyster respiration. Respiration rates were measured at 0, 5, 10, 15, 20, 25, 30, 35 ppt water salinities by keeping individual adult oysters (n = 12; Size = 5±0.2 cm) in closed chambers of 1 l capacity filled with water of 28 ± 0.5°C. Seawater was diluted by adding freshwater to get the desired salinity for the respiration experiments. Suitability scale was standardised from maximum respiration rates at observed salinity levels (i.e. maximum respiration rate = 1). Before running the respiration experiments, oysters were acclimatize for 24 h at the desired salinity condition to avoid stress related to change in physiological response. Respiration rates were measured when the oysters were actively filtering, which can easily be observed with shells open. Hand-held dissolved oxygen sensors (YSI model 55; YSI Inc., USA) were used to record the oxygen consumption rates at time intervals of five minutes, to check for a deviation in the linear decline. Each experimental trial was continued for about 2 hrs. Attention was given to prevent low oxygen concentration (\< 3 mg O₂ l^-1) during trial. RR = -V\*(pO_2,end−pO_2,start)/t, where RR = respiration rate in ml O₂.h^-1; V = volume of the chamber in l; pO_2,start and pO_2,end = oxygen concentration in ml l^-1 at the start and at the end of the measurements; t = time difference in hour between start and end of the experiment. SI is the Suitability Index for the environmental variables indicated in the. To obtain component index (CI) values for the two life stage components of the model, the SI values for appropriate variables were grouped and summarised by their geometric mean, as this is more sensitive to changes in individual variables than the arithmetic mean. It means that if an SI of 0 for any variable results in a CI of 0. Overall CI for settlement and post-settlement stages were estimated by using the following equations. For the larval settlement: $${CI}_{settlement - m} = \left( {SI}_{Ss - m} \times {SI}_{V - m} \right)^{1/2}$$ $${CI}_{settlement - nm} = \left( {SI}_{Ss - nm} \times {SI}_{V - nm} \right)^{1/2}$$ $${CI}_{settlement} = \left( {CI}_{settlement - m} + {CI}_{settlement - nm} \right)/2$$ For the post-settlement: $${CI}_{post - settlement - m} = \left( {SI}_{T - m} \times {SI}_{Sg - m} \times {SI}_{pH - m} \times {SI}_{DO - m} \times {SI}_{PIM - m} \times {SI}_{Chla - m} \right)^{1/6}$$ $${CI}_{post - settlement - nm} = \left( {SI}_{T - nm} \times {SI}_{Sg - nm} \times {SI}_{pH - nm} \times {SI}_{DO - nm} \times {SI}_{PIM - mn} \times {SI}_{Chla - nm} \right)^{1/6}$$ $${CI}_{post - settlement} = \left( {CI}_{post - settlement - m} \right)^{1/3}{\times \left( {CI}_{post - settlement - nm} \right)}^{1/3}$$ In these equations, CI _settlement is the component index for the larval settlement stage, which was considered for two seasonal component indices (i.e. CI _settlement-m, CI _{settlement-nm}) as the conditions for larval settlement can be different during the monsoon and non-monsoon periods. Thus, two different environmental mean data sets were used for the two periods (i.e. m = monsoon, nm = non-monsoon). During monsoon, a site may not be suitable for larval settlement, still it can have successful recruitment in the non-monsoon period. Therefore, arithmetic mean for seasonal larval CI is used instead of geometric mean, to consider the overall seasonal influences on the larval stage. Field observations indicated two seasonal settlement peaks in the investigated areas, thus equal weight coefficients were used for seasonal component indices (i.e., CI _settlement-m and CI _{settlement- nm}). CI _{post-settlement} is the component index for the post- settlement (i.e. spat/adult) stage. Mean environmental data for monsoon and non- monsoonal were used as well to calculate CI_{post-settlement}, as these seasons differ from each other. Based on the length of the seasonal periods (Monsoon = 4 months = 0.33 yr.; Non-monsoon = 8 months = 0.66 yr.), different weight coefficients were used for the seasonal component indices (i.e., CI _{post-settlement-m} and CI _{post-settlement-nm}) in determining the component index for the post-settlement stage. In contrast to the component index for larval settlement, a multiplication function is used for the post-settlement phase because the habitat conditions need to be suitable throughout the entire year. After obtaining the mean environmental data, the suitability indices (SIs) were determined by using suitability graphs and the component indices (CIs) were then calculated using the appropriate life stage equations. From the component indices, the overall HSI was determined following below equations as suggested by Cake: 1. If the component index for the post-settlement stage (CI _{post- settlement}) is the lowest component index (i.e., if CI _{post- settlement} \< CI _settlement), then *HSI* = *CI*_{*post*−*settlement*} 2. If the component index for the post-settlement stage (CI _{post settlement}) is not the lowest component index (i.e., CI _{post- settlement} \> CI _settlement), then *HSI* = (*CI*_{*post*−*settlement*}×*CI*_*settlement*)^1/2 ## Habitat suitability map Habitat suitability indices were calculated for the 80 sampling locations along the south-east coast of Bangladesh using the measured environmental variables. To get a first estimate of the length of coastline that is suitable for oyster restoration, the HSI values of the 80 sampling locations were interpolated over the entire south-east coastline using a nearest neighbour algorithm. For each HSI class, the total length (km) of the coastline was calculated using ArcGIS (version 10.5). ## Oyster data for model verification To verify the model results with field observations, an oyster population survey was conducted after the monsoon. Based on the availability of substrates (jetty pillars, sluice gates, bridge pillars, and boulders), 53 sites among the 80 sampling locations were available for this survey. At the remaining sampling locations no nearby suitable substrates were present and therefore those sites were omitted from the analysis. Population data for model verification can be affected due to long sampling period during survey time, particularly for large scale area of the study. To avoid it, three voluntary teams simultaneously engaged at northern, middle and southern part of the study area and complete the filed survey within a week, covering only 2–3 stations in a day using speed boat. Oyster density, shell height and condition index (percentage of dry shell weight-dry flesh weight ratio) were determined by taking oyster samples at each site. For this, replicated (\>5) quadrats (25 cm×25 cm) were used for sampling oysters from substrates available in the intertidal areas, which were positioned randomly along a 15 m long transect line (parallel to coastline) above mean lowest low water level (MLLW, \~ 0.5m), having similar emersion times for all locations. Quadrat areas without any oysters counted as zero. Quadrat areas with oysters were excavated without damaging the oysters. Living oysters were separated from dead shell remains. Living specimens were cleaned from epibionts and transported to the field laboratory, where individual shell height and fresh weight were measured. The soft tissue of each living oyster was separated from their shells, drained on paper towel and weighted after drying at 60°C for 12h. Geospatial oyster density data for the 53 locations were plotted on the potential HSI map where the size (area) of the circle represents the observed oyster density. ## Data analysis Statistical differences in mean environmental variables for the monsoon vs the non-monsoon seasons were verified, using a simple t-test. Moreover, multiple linear regression models were used in order to relate the response variables (i.e. oyster density, shell height, and condition index,) to a set of independent variables (i.e. temperature, salinity, pH, dissolved oxygen, PIM, Chlorophyll-a, water velocity) recorded for the non-monsoon season. The non- monsoon season had more influence on settlement and growth, as oyster growth is almost stagnant during monsoon. A forward stepwise procedure was followed by linear modelling to determine the environmental variable(s) that most influence oyster density, condition index, and shell height during growth season (i.e. non-monsoon months). Obtained data ranges for independent variables were checked whether they showed linear relationship with the suitability function curves used for the HSI modeling. Variance inflation factors (VIF) were used to check how much amount multicollinearity (correlation between independent variables) existed in a given regression analysis. The models were: $$y_{d} = \beta_{0} + \beta_{1}x_{T} + \beta_{2}x_{S} + \beta_{3}x_{pH} + \beta_{4}x_{DO} + \beta_{5}x_{PIM} + \beta_{6}x_{Chla} + \beta_{7}x_{V} + \varepsilon_{d}$$ $$y_{h} = \beta_{0} + \beta_{1}x_{T} + \beta_{2}x_{S} + \beta_{3}x_{pH} + \beta_{4}x_{DO} + \beta_{5}x_{PIM} + \beta_{6}x_{Chla} + \beta_{7}x_{V} + \varepsilon_{h}$$ $$y_{CIndex} = \beta_{0} + \beta_{1}x_{T} + \beta_{2}x_{S} + \beta_{3}x_{pH} + \beta_{4}x_{DO} + \beta_{5}x_{PIM} + \beta_{6}x_{Chla} + \beta_{7}x_{V} + \varepsilon_{CIndex}$$ Where, *y* is the response variable indicating oyster density (d), shell height (h) and condition index (CIndex). *T* = temperature; *S* = salinity; *pH* = water pH; *DO* = dissolved oxygen; *PIM* = PIM; *Chla* = chlorophyll-a; *V* = water velocity. The parameter *β*₀ is the *y*-intercept, which represents the theoretical expected value of *y* when each *x* is zero. The other parameters (*β*₁, *β*₂, …, *β*₇) in the multiple regression equation are partial slopes. *β*_*j* (here, j = 1....7) representing the expected change in *y* for a given unit increase in *x*_*j* while holding all other *x*s constant, and does not depend on the value of any other *x*. Other assumptions were: *E*(*ε*_*i*) = 0 for all *i*, where *ε* is the residual terms of each model and i = 1….7 assigned for seven environmental parameters (i.e., T, S, pH, DO, PIM, Chla, and V) respectively; Var $\left( \varepsilon_{i} \right) = \delta_{\varepsilon}^{2}$ for all *i*; the *ε*_*i**s* were independent; and *ε*_*i* was normally distributed. Before statistical analysis, the normality of a response and independent variables were tested using the Kolmogorov-Smirnov test and homogeneity of variances using Levene’s test. All analyses were performed using IBM SPSS statistics software (Version 2015) using α = 0.05. # Results ## Environmental variables Environmental conditions showed both spatial and seasonal variations over the study period. A strong seasonal effect was observed, with monsoon months (June–September) differing from non-monsoon period (October–May). Particularly, salinity, PIM, chlorophyll-a concentrations and water flow velocity during monsoon period showed significant (p \< 0.05) differences with the non-monsoon period. Spatially, a clear salinity gradient was observed for both seasonal periods showing an increasing trend from north to south. Feni, Mirsarai, and upper Chittagong coastal areas received strong influences from the nearby river systems and mean salinities ranged from 0.5–7.0 ppt with high mean particulate inorganic matter concentration (360–707 mg l^-1). Mean salinities and suspended concentration in the lower part of Chittagong coast, Kutubdia, Moheshkhali, Cox’s Bazar, and Teknaf ranged from 6.7–29.5 ppt and 66–433 mg l^-1, respectively. Among these areas, a few sites like Sonadia (southern Maheshkhali) and the Teknaf peninsula were strongly dominated by the Bay of Bengal, showing smaller variation even in monsoon months. Chlorophyll-a concentrations varied from 0.8–9.6 μgl^-1 and was relatively high in the southern part of the study area compared to the freshwater dominated and turbid northern part of the study area. Mean water temperatures for all stations showed minimal variation (27–28.6°C) throughout the entire study period. Water pH levels ranged from 7.4–8.5 along the station sampled. pH was relatively high in non-monsoonal months and showed reduced values from north to south, probably influence of river discharge. Saturation level of dissolved oxygen varied from 49–91%. No significant differences (p = 0.70) in monsoon and non-monsoon season were observed for dissolved oxygen concentration, but a decreasing trend was observed from South to North which might be related to the organic loading from the rivers. Water flow velocity became stronger in monsoon periods and higher in exposed vs sheltered sites, ranging from 0.2–2.4 m sec^-1. ## Model estimation By considering the effect of the two seasons, 37 sites, occupying approximately 397 km of coastline were predicted as suitable (HSI score \>0.50) for year round growth of oysters. Most of these sites were located in the area of lower Chittagong (Banskhali, Chanua), Pekua, Kutubdia, Moheshkhali, Sonadia, Cox’s Bazar and Teknaf coastal waters. Among those sites, 114 km of coastline along Sonadia, south-western Maheshkhali channel and southern tip of Teknaf peninsula were predicted as places with the highest HSI score (HSI score \>0.7). In addition, 24 scattered sites in the southern part, representing a coastline of approximately 269 km, were less suitable for oysters (HSI score: 0.3–0.5). 19 sites representing approximately 391 km of coastline showed least prospect (HSI score: 0.0–0.3) for oyster development. Most of these sites belong to the coastline between Sandwip-Feni to mouth Karnaphully River. Few sites in the inner parts of the Moheshkhali channel, Chokoria and Cox’s Bazar coast (Inani and Monkhali) were also not considered as potential sites for oyster development. A habitat suitability map is presented in based on HSI scores. ## Field verification Three population descriptors were used for the assessment: (1) oyster density; (2) shell height; and (3) condition index. These were then correlated with the HSI scores for model verification. HSI values showed a strong positive relationship with the mean oyster densities (r = 0.87). Oysters were not observed at sites which had an HSI score less than 0.27. The highest number of oysters (1064–1596 indiv. m^-2) were observed at sites which showed highest HSI scores (\>0.70). Moreover, mean oyster size (shell height) varied among the sites and showed positive relationship (r = 0.95) with HSI values as well. Variation in shell height was higher for upper HSI values, suggesting that the oysters in high HSI sites have multiple age classes due to multiple recruitments years. The oysters grew bigger in size (\>5 cm shell height) in those sites, where the HSI score exceeded 0.50. Regression results for HSI and condition index also showed a similar trend. Field data showed that the shell- body flesh weight ratio largely varied (4.0–10.9%) among sites. Condition index increased with the increasing HSI scores, showing good correlation (r = 0.98). Condition indices were found relatively high (\>6%), when the HSI score exceeded 0.50. Condition index for lower HSI sites showed more variability, which might be due to larger seasonal variation at these sites. Conversely, sites with high HIS values showed less variability in soft tissues coinciding with smaller seasonal variation. All the population descriptors were also strongly correlated (r \> 0.90) with each other , thus showing good agreement with HSI scores. ## Influencing environmental factors The linear regression model results indicated that salinity, chlorophyll-a, pH, and dissolved oxygen are the main predictors of oyster occurrence and their conditions (for more details see). Water temperature, PIM and water flow velocity were removed from the model during the stepwise procedure, as these factors failed to improve the model outputs. Salinity and Chlorophyll-a were found as common explanatory variables in the model that influenced oyster density, shell height and condition index. Oyster density and also shell height had high values in areas where the oxygen saturation level was relatively high. pH values also contributed to explain observed condition index of oysters. Scatter plots and correlation coefficients among all variables also gave the same results. Collinearity statistics in the linear model showed that variance inflation factors (VIF) were less than 5. It rejected the hypothesis of a multicollinearity relationship among environmental factors, thus explanatory variables used in the linear models were independent. # Discussion Selection of relevant environmental variables for HSI model development is critical. It depends on the magnitude of the environmental factors related to habitat quality, as they vary in time (i.e. seasons) and space and the tolerance range of the oysters. In this study we developed an HSI model for the intertidal rock oyster, *S*. *cucullata* using seven environmental factors. Out of seven environmental factors, four factors viz., salinity, chlorophyll-a, dissolved oxygen and pH were found to be predictors of oyster density, condition index, and shell height. Water temperature, PIM concentrations and water flow velocity were not consider as predictors in the linear models used in this study. More than 70% of the variation in dependent descriptors (i.e. oyster density, condition index, and shell height) was explained by adding the variables: salinity, Chlorophyll-a, dissolved oxygen and pH (See). A large number of sites (n = 42) investigated in this study showed a decrease in salinity (\> 5ppt) during the monsoon period. Most of these sites were located in upper south-eastern coast of Bangladesh, where large number of newly settled oysters die during the monsoon period. HSI scores were correspondingly low in these low salinity areas. Indeed, fluctuations in salinity regulate metabolic activities in oysters living in shallow marine and estuarine areas. The reproductive capacities, spat settlement and growth of oysters are typically impaired by low salinities. Low salinities can also cause mass mortalities of tropical oysters during the monsoon season, if the exposure to low salinities last too long. In contrast, oysters flourished at sites of the investigated area where the salinity remained more than 10 ppt. Chlorophyll-*a* concentrations varied both spatially and seasonally within the study area, and were lower during the monsoon period as compared to the non- monsoon period. Chlorophyll-*a* concentration increased with decreased suspended sediment load which might be due to better light penetration enhancing primary productivity. Higher HSI scores were found at locations where chlorophyll-a concentrations were high. In a study by Sasikumar et al. (2007), oyster growth was positively correlation with Chlorophyll-*a* concentrations. Dissolved oxygen levels appeared not to be critical in the investigated areas (all sites have values \>50% saturation level). pH values were relatively high in the non-monsoon season as compared to the monsoon season. However, both parameters showed an increasing trend towards the south, which might be the influence of strong water circulation from the Bay of Bengal. Oyster densities and related condition indices were relatively high in the area, where oxygen saturation levels (\>70%) and pH (\>7.9) also were high. In the Indian coasts, dissolved oxygen and water pH showed positive correlation with oyster spat settling rate. Physiological activities slowed down with decreasing pH (\<7.75). Water temperatures did not vary much along the coastline and did therefore not show any significant correlation with any dependent variables. PIM concentration (i.e. suspended sediment) in the study area varied between 21 and 1044 mg l^-1 depending on the distance from river mouth. It showed a clear seasonal pattern at all sites with a large increase during the monsoon period, when 80 percent (\~ 1850 mm) of the total rainfall occurs along with huge amounts of suspended sediments carried to the coast via rivers. Oysters can feed in turbid environments, but are less efficient and produce copious amount of pseudofaeces, which affects gill sorting. However, geo-spatial field PIM data did not show any significant correlation with the population descriptors used for model validation. Field observations also confirmed that *S*. *cucullata* population can thrive under high turbid (\<700 mg l^-1) conditions, if other environmental variables are optimal. Water flow velocity also did not show a good correlation with any of the dependent descriptors as the oysters were present in both high and low energy coasts, where other environmental factors are favourable. So, water temperature, PIM concentrations and water flow velocity can be neglected for determining the oyster habitat suitability in Bangladesh coast. In this regard, a sensitivity test was performed through simple model application, where these three factors were not included. It provided similar results (R² = 0.96) in categorizing the site characteristics in terms of HSI scores. Means of annual survey data are often used as input variables into HSI models. It may not provide appropriate HSI scores to evaluate a site, if strong seasonal influences exist as in monsoon dominated areas. Some habitat factors can be constant over time (like in our case water temperature), while other factors viz., salinity, pH, PIM, dissolved oxygen, Chlorophyll-a may show strong seasonal differences. The use of extreme values for dynamic environmental variables in a HSI model can predict presence or absence of target species. However this approach might underestimate habitat quality, if the extreme values are not lethal to target species. Annual means without a seasonal considerations and also extreme environmental variable values were applied to evaluate the consequences in our model outputs. These provided high and low number of suitable sites respectively, which did not reflect field situations. Extreme values (i.e. observed lower ranges) only need to use, when they reach at or near the lethal levels and limit the survival. Otherwise mean seasonal data should consider to determine the component index of each environmental factor. Moreover, tolerance ranges could vary with different life history stages. Though adult oysters can tolerate extreme low salinities for extended periods, small periods of low salinity have a pronounced effect on settlement rate. Spat settlement was generally unsuccessful during the monsoon period at many sites, but this phenomenon may not determine the quality of a habitat over the entire year. Spatfall after the monsoon is also important to maintain the population in dynamic coast. Particularly, the oysters that survive the non-monsoon period can then also survive the next monsoon months as they grow and tolerate low salinities. These seasonal and longer life stage considerations improved the model outputs with respect to previous unreported version of the model, and HSI scores showed strong correlations (r \>0.87) with the oyster population descriptors (i.e. density, shell height and condition index). This study also assumed that all model inputs and functions were independent. However, one environmental factor can be influenced by others. For example, water pH can be regulated by salinity conditions as both the factors are influenced by the Bay of Bengal. This type of relationship should be considered to further improve the model. Field verification is a critical part to determine the accuracy of HSI models, but is often lacking in oyster HSI models. Adult oyster density is commonly used to validate an HSI model; however, this may not explain the complete picture. Here we not only considered adult oyster density, but also size (shell height) variation and condition index from 53 sites. All these descriptors demonstrate strong positive relationships with the HSI scores. Mean oyster size (\> 5 cm) and condition index were to be high at those sites, where the HSI scores exceeded 0.50. In Bangladesh coastal waters, it usually takes more than a year to reach 5cm size (shell height). It confirms that oysters survive longer than a single year and that oyster populations probably can become self-sustaining with multiple year (size) classes when HSI scores are greater than 0.50. Population data were collected from a single survey after the monsoon period; this may differ with other seasons, yet it confirmed that oyster survival occurred after the monsoon. Still, demographic information for various seasons may improve the validation. Incorporation of the HSI scores into a GIS interface provides a visual aids in the format of maps for coastal resource managers and policy makers. An attempt was made to develop an HSI geospatial map for oysters, where validation data that reflect the population survey were used. We investigated about 1050 km of coastline using 80 representative sampling sites, and the strong spatial patterning in the map of HSI scores shows regions with good habitat suitability within the study area. This gives a spatially explicit visualization of potential oyster habitats along the south-east Bangladesh coast. A simple nearest neighbour algorithm was used as an interpolation technique in a GIS interface to categorize the length of coastline using HSI scores. This forms a good basis for site selections, thus can be further expanded upon by increasing the number of sampling sites and the extent of the temporal environmental samples. The present study has attempted to include all the available information to identify suitable sites for oysters through HSI model development. Nevertheless, the approach could be refined further with additional information. Such as, the amount of substrate available is important as it would also both contribute to the HSI and also could affect oyster density. There are some areas that showed potential for oyster development, but oysters were absent due to lack of substrate. Artificial hard substrate can be added there to test the model results. In this regard, bottom characteristics and wave energy conditions of coastal sites could be useful for determining the substrate types, which is inevitable for oyster reef formation. Yet, the verification of the model with field surveys shows a good fit for this oyster HSI model. # Conclusion This study developed an HSI model for the intertidal rock oyster, *Saccostrea cucullata* and applied it to the entire south-eastern coast of Bangladesh. Salinity, Chlorophyll-a, dissolved oxygen and pH were identified as driving factors that determine the habitat quality of oyster populations along this region. The results clearly show that freshwater dominated low saline estuaries and nearby coastal areas with high suspended sediments are least suitable for oyster settlement and growth. In contrast, the bay dominated areas with relative high salinity, Chlorophyll-a, dissolved oxygen and pH were found to be suitable for oyster settlement and growth. Seasons (i.e. monsoon and non-monsoon) and life stage (i.e. settlement and post settlement) considerations are found effective and suggested as integral part in habitat suitability model formulation for subtropical dynamic coastal systems. In this study, the HSI model results match the current distribution of oysters throughout the investigated area. The good correspondence with the field data enhances the reliability of the presented HSI model as a quantitative tool for planning oyster restoration and managing oyster resources along the south-eastern coast of Bangladesh. # Supporting information We would like to thank all the ShoreLAB volunteers (undergraduate students of IMSF, University of Chittagong), who helped during field data collections. We also thank to Professor Sayedur Rahman Chowdhury at Institute of Marine Sciences (IMS) for his valuable suggestions to formulate the model equations. We are thankful to Professor Dr. Loren Coen (Florida Atlantic University, USA) for review this article and his valuable remarks helped us to improve the manuscript. [^1]: The authors have declared that no competing interests exist.

# Introduction Malignant melanoma is an aggressive and chemoresistant type of skin cancer that originates in melanocytes. Although less than 5% of skin cancers are melanoma, it causes a large majority of skin cancer related deaths. The poor prognosis for late stage melanoma patients is due to the low response rates to conventional chemotherapy treatments with dacarbazine or its derivative temozolomide, that are below 20%. The platinum analog cisplatin is known to be highly effective against various solid tumors. In malignant melanoma, the single-agent cisplatin reaches a response-rate of less than 10%. However, higher response rates have been reported in combinational therapy but without extending survival significantly due to greater toxicity. Cisplatin mediates the activation of mitogen activated protein kinases (MAPK), nuclear factor kappa B (NF-ĸB), p53 and apoptotic pathways by interaction with DNA and forming DNA-adducts. Induction of apoptosis is one of the main purposes of anti-cancer drugs, and therefore resistance to apoptosis is suggested as a possible mechanism leading to unresponsiveness of cancer cells. Enhanced NF-ĸΒ activity was found in many different cancers, including melanoma. NF-ĸΒ activation is induced by different stimuli including growth factors, cytokines, lymphokines, UV, stress, and pharmacological agents such as cisplatin. The importance of NF-ĸB in tumorigenesis has been reasoned to its involvement in the regulation of inflammation, apoptosis, angiogenesis, tumor cell invasion and chemoresistance. Furthermore, there is ample evidence that MAPKs play a significant role in the activation of NF- ĸB, besides their normal role in the activation of the transcription factor activator protein-1 (AP-1). Concerning its oncogenic potential, AP-1 has been described as double edged sword because it may play a role in cell survival and apoptosis in cancer cells. In this context, JNK and ERK have also been reported as both, pro- and anti-apoptotic regulators. α-Catulin is a cytoskeletal linker protein that stimulates the expression of anti-apoptotic genes. Previously, we demonstrated that α-catulin is an IKK- interacting protein that augments activation of NF-ĸB after stimulation with the inflammatory stimuli TNF-α or IL-1 as well as the activation of the Rho signaling pathway in HeLa and HEK293 cells. Recently, we showed that α-catulin is highly expressed in malignant melanoma cells compared to melanocytes and that α-catulin is a key driver of tumor progression, invasion and metastasis via the upregulation of E-cadherin and the downregulation of mesenchymal genes such as N-cadherin, Snail/Slug and the matrix metalloproteinases 2 and 9. Accordingly, Cao et al. (2012) reported that α-catulin was highly expressed in squamous cell carcinoma and its knockdown decreased the migratory and invasive behavior in both tumor cells *in vitro* and in xenotransplants *in vivo*. In addition, it has been shown that α-catulin increased NF-ĸB through an ILK-dependent pathway leading to elevated fibronectin and integrin α_vβ₃ expression and therefore promoted tumor cell migration, invasion and metastasis in lung cancer cells. In this study we show for the first time, that down-regulation of α-catulin diminished NF-ĸB, MAPK and AP-1 activation in malignant melanoma cells. Furthermore, α-catulin knockdown sensitized melanoma cells to treatment with cisplatin and other chemotherapeutic drugs. Cisplatin treatment decreased ERK-, JNK- and c-Jun activity in α-catulin knockdown melanoma cells, which was accompanied by reduced cell proliferation and enhanced apoptosis compared to control cells. # Materials and Methods ## Biochemicals and Antibodies Purified non-labeled mouse/rabbit mono-/polyclonal antibodies were anti-CTNNAL1, Mcl-1, CBP, p38 and p-p38 (Abcam, Cambridge, MA, USA), ERK, p-ERK, JNK, p-JNK, c-Jun, p-cJun, p21^waf/cip1, p53, GAPDH, (Cell Signaling Technology, Inc., Danvers, MA, USA), KI-67 (Santa Cruz, Texas, USA). HRP-conjugated secondary anti-mouse and anti-rabbit antibodies were obtained from Life Technologies and Cell Signaling Technology, Inc., respectively. Anti-α-tubulin- HRP conjugated antibody was purchased from Cell Signaling Technology, Inc. Texas-Red conjugated secondary anti-mouse and anti-rabbit antibodies were purchased by Jackson Immuno (Newmarket, UK). HGF and TNF-α were purchased from PeproTech Inc. (Rocky Hill, NJ). LPS was purchased from Sigma-Aldrich (St. Louis, MO, USA). 5x NFκB-luc-Reporter and AP-1-luc-Reporter plasmids were obtained from Agilent technologies (Santa Clara, USA), pGL3 reporter and control vectors were purchased from Invitrogen, E-cadherin si-RNA was purchased from Santa Cruz Biotechnology, Inc. and Caspase3/7, 8 and 9 assay and CellTiter-Blue cell viability Assay from Promega (Madison, USA). Staurosporine was purchased from eBioscience (CA, USA). Chemotherapeutic agents dacarbazine, cisplatin and paclitaxel were acquired from Sigma-Aldrich. ## Cells and cell cultures Human metastatic melanoma cells from spleen (Mel.7), skin (Mel.17) and lymph node (Mel.15) were isolated and cultivated as described previously. Human melanoma cells A375 were purchased from Sigma-Aldrich and cultivated in DMEM (Invitrogen). Human primary melanocytes were obtained from Provitro and cultivated in melanocyte growth medium (Provitro, Germany) with 10% FCS (PAA Laboratories, Pasching, Austria) and 1% Penicillin/Streptomycin (Sigma-Aldrich). In all experiments cell culture medium was supplemented with 10% FCS and 1% Pen/Strep unless indicated otherwise in the figure description. Spheroids were generated by hanging drop method with 100 cells per drop and after four days of incubation spheroids were used for viability assay and microscopy. ## Quantitative Real-time PCR Total RNA was extracted using the RNeasy Mini Kit (Qiagen, USA). RNA was reverse-transcribed with the First strand cDNA synthesis Kit (Roche Diagnostics, Germany) according to manufacturer’s instruction. Real Time PCR was performed with TaqMan Gene expression Master Mix, Assay, Primer and Probes (HS00972094 CTNNAL1, HS99999905 GAPDH; Applied Biosystems, USA). Reactions were run on the Light Cycler 480, Roche. Threshold cycle (Ct) values of the target genes were converted to arbitrary expression values by extrapolation from the standard curve and finally normalized to the internal control. ## Western blotting For Western blotting, proteins were extracted from 10⁵ cells from each cell line. Total protein extracts were separated by 4–20% SDS-PAGE (Peqlab, Erlangen, Germany) and transferred with Trans-Blot Turbo, Transfer System to a Trans-Blot Turbo TransferPack Mini Format, 0,2 μm PVDF membrane (BioRad, California, USA). Membranes were blocked with 5% nonfat milk in Tris-buffered saline pH7,4 (TBS, BioRad), and immunodetection was carried out using specific antibodies via chemiluminescence with ChemiDoc MP Imaging System, Universal Hood III (BioRad). ## Expression vectors and lentiviral infection Lentiviral vector constructs pLEX-MCS, pLEX-JRED and pGIPZ shRNAmir CTNNAL1 (Table 1) and pGIPZ nonsilencing control were purchased from Thermo Scientific. Myc-catulin vector was described previously. For stable transfection, 5 x 10⁶ HEK293T cells were used for the production of lentiviral stocks with a translentiviral packaging system (Thermo scientific). Melanoma cells (2 x 10⁵) were infected with lentiviral stocks and selected with puromycin. Sequences of shRNAmir CTNNAL1 RNAs (Open Biosystems) were as follows: V3LHS_356693 here called sh-catu1, sense strand `5´-AGCTCAAAGCAAGAAAACA-3´;` antisense `5´-GTTTTCTTGCTTTGAGCT-3;` sh-catu2 (V3LHS_356695), sense strand `5´-AGCTTGTTGAGACCTGTCG-3;` antisense: `5´-CGACAGGTCTCAACAAGCT-3´`. ## Cell transfection 4x10⁵ melanocytes or melanoma cells per well were transiently transfected by the lipofectamin-2000 method using 2μg DNA and 7μl lipofectamin (Life Technologies). After 6 hours transfected cells were rinsed and incubated for additional 16–48 hours before they were used for the experiments. ## Luciferase Reporter gene assay A 5x NF-κB-Luc reporter gene was transfected into melanocytes or stable transfected melanoma cells (sh-catu2 and n.s.). Cells were either co-transfected with IKK-β or p65 and/or stimulated with TNF-α, LPS, HGF or 10% FCS. Luciferase levels were normalized for a co-transfected JRED control and luciferase values were determined and normalized for co-transfected RFP values. An AP-1-luc reporter gene was transfected into stable infected melanoma cells (sh-catu2 and n.s.) Cells were non-stimulated or stimulated with TNF-α or LPS luciferase levels were detected using Infinite F200 PRO multiplate reader (TECAN, Männedorf, Switzerland) and normalized for a co-transfected JRED control. ## Flow Cytometry analysis For cell cycle distribution analysis Melanoma 7 cells (n.s., sh-catu2) were treated with cisplatin for 48 hours, detached, washed with PBS and fixed with 70% ice cold Ethanol. Then cells were washed again with PBS and stained with propidium-iodide solution (0.1% Triton-X-100, 2 mg DNAse free RNAse A and 500 μg/ml propidium-iodide) for 15 min at 37°C. Thereafter flow cytometry analysis was performed using Accuri Flow Cytometer and CFlow plus software (BD Biosciences). For apoptosis-necrosis detection Melanoma 7 cells (n.s., sh-caut1 and 2) were treated with cisplatin for 48 h. Then cells were detached, centrifuged and washed with ice cold PBS. Thereafter cells were stained with APC-Annexin V (BD- Pharmingten) in 1x binding buffer (BD-Pharmingten) for 30 min at 4°C. Then Propidium Iodide Solution (Sigma-Aldrich) was added and Flow cytometry analysis was performed using Accuri Flow Cytometer and CFlow plus software (BD Biosciences). Cytochrome c release assay was performed with the InnoCyte Flow Cytometric cytochrome c release Kit from Millipore (Billerica, MA, USA). Melanoma 7 cells (sh-catu2 and n.s.) were treated with different concentrations of cisplatin for 6 hours. Permeabilization, fixation and staining of the cells were performed according to the manufacturer’s instructions. Cells were analyzed using Accuri Flow Cytometer and CFlow plus software (BD Biosciences). ## Cell viability assays and apoptosis Stable infected Mel.7, Mel.17 and Mel.15 cells (n.s., sh-catu2) were treated with different concentrations of cisplatin, dacarbazine, paclitaxel or staurosporin for 48h, cell survival normalized to untreated cells (pos. contr.) and analyzed by CellTiter-Blue cell viability assay substrate. After 2 hours of incubation the cells were analyzed with Infinite F200 PRO multiplate reader (TECAN). Caspase-Glo 3/7, 8 and 9 Assay from Promega was used to determine cells in apoptosis. Cells were treated as in the cell viability assay and luciferase substrate was added after 24 h. Luciferase was detected with Infinite F200 PRO multiplate reader (TECAN) and normalized for GFP values (stable infected). JC-1—Mitochondrial Membrane Potential Assay Kit (Abcam) was used to investigate the level of cells in apoptosis. Stable infected Mel.7 cells (n.s., sh-catu2) were treated with different concentrations of cisplatin for 6 h and stained with JC-1 solution for 10 min. Relative fluorescence level was detected at a wavelength of 535 nm with Infinite F200 PRO multiplate reader (TECAN) and normalized for GFP values (488 nm). ## Cell proliferation assay Stable infected Mel.7 cells (n.s., sh-catu2) were seeded in 96 well plates at an amount of 2500 cells per well. After 24 hours cells were treated with 20, 10, or 5 μg/ml cisplatin or were left untreated and after 48 hours CytoSelect BrdU Cell Proliferation ELISA Kit (Cell Biolabs, Inc.) was used to determine cell proliferation. Therefore, cells were treated with BrdU solution, fixed and stained with antibodies against BrdU (provided in the kit) and detected with the multiplate reader at 450 nm wavelength. BrdU values were normalized to gfp values measured before treatment with BrdU solution. ## Statistical analysis The Student's paired *t*-test was used. Reported *P* values are three-tailed and *P* \< 0.05 (\*) was considered statistically significant, P\<0.01 (\*\*) and P\<0.001 (\*\*\*) was considered as statistically highly significant. # Results ## α-Catulin increases NF-κB activation in malignant melanoma cells NF-ĸB activity has been shown to be upregulated in melanoma cells. Previously, we demonstrated that α-catulin is an IKK-interacting protein that augments activation of NF-ĸB after stimulation with the inflammatory stimuli TNF-α or IL-1 as well as the activation of the Rho signaling pathway in HeLa and HEK293 cells. To examine whether α-catulin can enhance NF-ĸB activity in melanocytes, we transfected normal human melanocytes (NHM) cells with a NF-ĸB reporter gene (5x NF-ĸB-Luc) plus different concentrations of α-catulin or mock, and either together with IKKβ or p65 expression vectors (.), or stimulated the cells with TNFα or LPS. We found that α-catulin increased already the basal NF-ĸB activity in melanocytes in a concentration dependent manner, but further augmented NF-ĸB activation in a highly significant manner when cotransfected with IKKβ or stimulated with TNFα or LPS for 8 h. In contrast, p65-mediated activity of the reporter was not enhanced by α-catulin. To further substantiate our findings, we knocked down the expression of α-catulin using lentiviral vectors in three different freshly isolated melanoma cells. Down-regulation of α-catulin significantly decreased the level of the NF-ĸB-luciferase reporter in all three melanoma cell lines (n.s. versus sh-catu1 and sh-catu2). α-Catulin knockdown also decreased NF-ĸB activity after TNFα-, LPS-, HGF- and Serum (10% FCS) activation in Melanoma 7 cells, which is consistent with our previous finding that α-catulin is central for mediating NF-ĸB activation. Previously, it was demonstrated that α-catulin regulates E-cadherin and that E-cadherin regulates NF-κB and AP-1/c-JUN. To examine whether α-catulin and E-cadherin have synergistic effects melanoma cells were transfected with a NF-ĸB reporter gene (5x NF-ĸB-Luc) plus sh-catu (downregulation) or myc-α-catulin (ectopic expression) together with or without si-E-cadherin RNA. Interestingly, E-cadherin silencing significantly increased the low (sh-catu-mediated) NF-ĸB- luciferase reporter level whereas the elevated NF-κB activity caused by myc-α- catulin was further enhanced after E-cadherin silencing, suggesting that α-catulin increases NF-κB expression via downregulation of E-cadherin. ## Down-regulation of α-catulin diminishes AP-1 activity and ERK phosphorylation in malignant melanoma cells Together with NF-κB, AP-1 is known as an important transcription factor that regulates the expression of genes involved in inflammation, cell growth, survival and death. Having demonstrated that α-catulin plays a pivotal role in the activation of NF-ĸB, the next step was to determine the influence of α-catulin expression on AP-1 activity. Therefore, Mel.7 cells (n.s. and sh- catu1/2) were transfected with an AP-1 luciferase reporter gene and the cells were non-stimulated or stimulated with LPS or TNF-α. Down-regulation of α-catulin significantly decreased the basal AP-1-activity level and additionally reduced the AP-1 activity after stimulation with TNF-α or LPS in Mel.7 cells. It is known that MAPKs are involved in the regulation of NF-κB and AP-1 activation and that they play a pivotal role in tumor cell growth, proliferation, apoptosis and survival. Hence, we analyzed the effect of α-catulin knockdown on the activation of the MAPK family members ERK, JNK and p-38 using Western blot analysis. ERK phosphorylation was significantly reduced in Mel. 7 cells with α-catulin knockdown compared to control cells whereas JNK- and p-38 phosphorylation was not altered. The reduced phosphorylation and expression of c-Jun in α-catulin knockdown cells correlates with decreased AP-1 activity in the reporter assay. Together these data indicate that the increased expression of α-catulin in malignant melanoma cells amplifies NFκB and AP-1 activity and the level of ERK phosphorylation. ## α-Catulin knockdown reduces ERK-, JNK- and c-Jun phosphorylation in cisplatin treated melanoma cells It is known that cisplatin treatment induces several signaling pathways in melanoma cells, including NF-ĸB, MAPKs and apoptosis. Having demonstrated that α-catulin increased NF-ĸB, AP-1 activity and ERK phosphorylation in malignant melanoma cells (Figs.), we next examined how α-catulin influences NF-ĸB and MAPK signaling in cisplatin-treated melanoma cells. Mel.7 cells were treated (n.s. vs. sh-catu2) with different concentrations of cisplatin for 24 h and analysed for expression of p-ERK, p-JNK, p-cJun and the target genes Mcl-1 and CBP (CREB binding protein). Interestingly, the phosphorylation of the kinases ERK, JNK and c-Jun but also the protein level of Mcl-1 and CBP decreased dramatically in a dose dependent manner in α-catulin knockdown cells compared to control cells. ## α-Catulin knockdown sensitizes Melanoma cells to treatment with cisplatin Since the previous results have shown that NFκB, AP-1 and ERK activity are increased in α-catulin expressing melanoma cells, we next sought to examine the effect of α-catulin knockdown on cell survival. Therefore, Mel.7, Mel.17 and Mel.15 cells (n.s. vs. sh-catu1/2) were treated with different concentrations of the chemotherapeutic drug cisplatin and cell survival was determined normalized to untreated control cells. In all tested cell lines α-catulin knockdown cells were 2–8.5-fold more sensitive to cisplatin treatment than cells expressing α-catulin. In particular, in Mel.7, Mel.15 and Mel.17 cells the IC-50 values were 50 μg/ml, 12.5 μg/ml and 26 μg/ml in control cells and 6 μg/ml, 2.4 μg/ml and 12.5 μg/ml in α-catulin knockdown (sh-catu2) cells after 48h treatment, respectively. These observations were additionally affirmed in Melanoma 7 cells transfected with sh-catu1 and in primary melanocytes (NHMs) lentiviral transfected with myc-α-catulin and mock control. In order to confirm these observations in a more physiological assay we cultured Mel.7 cells as hanging drops to form spheroids and treated them for 48 hours with cisplatin. The determined IC-50 values were 75 μg/ml in control cells and 40 μg/ml in α-catulin silenced cells. To visualize this effect we treated the spheroids with an overdose of 200 μg/ml cisplatin for 96 hours and determined the size of the spheroids before and after treatment. Before treatment median size was slightly higher in α-catulin silenced cells compared to control cells. After treatment the size of α-catulin silenced spheroids was decreased dramatically, whereas non-silenced spheroids were only slightly smaller than before treatment. To further elucidate the effect of α-catulin knockdown on other therapeutic agents, Mel.7 cells were treated with dacarbazine, paclitaxcel and the potent apoptosis inducer staurosporine for 48h. α-Catulin knockdown reduced cell survival also significantly for these chemotherapeutic drugs (.). ## α-Catulin knockdown decreases cell proliferation in cisplatin-treated melanoma cells AP-1, NFκB and ERK are also important regulators of cell proliferation. To elucidate the influence of α-catulin and cisplatin treatment on cell proliferation we treated Mel. 7 cells with different concentrations of cisplatin and analyzed them for expression of the proliferation marker Ki-67 using Western blot. Ki-67 expression was significantly reduced in α-catulin knockdown cells treated with 10 or 20 μg/ml cisplatin compared to control cells. To further investigate this observation we treated the cells with 5, 10 or 20 μg/ml cisplatin or left them untreated (ctrl) and BrdU Assay was performed as described in. BrdU incorporation was slightly reduced in cisplatin treated Mel. 7 cells compared to the untreated cells. Additionally, the impaired proliferation in α-catulin knockdown cells treated with 20 μg/ml cisplatin compared to non-silenced control cells was confirmed suggesting that knockdown of α-catulin reduces cell proliferation in cisplatin-treated melanoma cells. To further confirm these observations we analysed the cell cycle distribution in Melanoma 7 cells treated with 0 or 10 μg/ml cisplatin for 48 hours. Compared to the untreated controls, the relative amount of cells in G1 was enhanced, whereas the percentage of cells in G2 and S-phase was significantly reduced in both n.s. and sh-catu2 cells after cisplatin treatment. Further, we analysed the influence of α-catulin knockdown on the cell cycle inhibitors p21^waf/cip1 and p53 in cisplatin treated Mel. 7 cells. clearly demonstrate a dose dependent expression of p21^waf/cip1 and p53 in sh-catu2 melanoma cells after cisplatin treatment. ## α-Catulin knockdown enhances apoptosis in cisplatin-treated melanoma cells Measuring cell metabolism, we clearly demonstrated that α-catulin knockdown melanoma cells are more susceptible to cisplatin treatment than control cells. To further investigate the mechanism of cell death, we evaluated the AnnexinV- and propidium iodide (PI)-positive cells. As depicted in, apoptosis (Annexin V staining) was significantly enhanced in α-catulin knockdown cells (sh-catu1: 56.2% and sh-catu2: 52.6%) compared to control cells (n.s.: 41%) after cisplatin treatment for 48 h. No striking differences were observed in necrotic cell (PI- positiv). It has been reported that levels of cytochrome c in the cytoplasm were enhanced in cisplatin treated cells. Cytochrome c release from mitochondria is an early event when cells undergo apoptosis. To investigate the contribution of α-catulin on cell death after treatment with cisplatin we performed a cytochrome c release assay and compared α-catulin knockdown melanoma cells with control cells. Mel. 7 cells were untreated or treated for 6 hours with 2.5 μg/ml cisplatin. Then the cells were fixed, permeabilized and stained with a specific antibody against cytochrome c as described in, followed by flow cytometry analysis. Release of cytochrome c from mitochondria in the course of apoptosis results in a wash-out from the permeabilized cells and reduction of cytochrome c-related fluorescence. In the absence of cisplatin, the percentage of apoptotic cells was almost equal for control and α-Catulin knockdown cells with 11.9% and 12.1%, respectively. However, α-Catulin knockdown significantly increased the percentage of apoptotic cells after cisplatin treatment with 30.5% compared to control cells (n.s.) with 17.6% apoptotic cells. Along with cytochrome c release the mitochondrial membrane is depolarized in cisplatin induced apoptosis. To investigate the effect of α-catulin knockdown on mitochondrial membrane potential in cisplatin treated melanoma cells, we used the fluorescent JC-1 dye. The relative mitochondrial potential depolarized in a dose-dependent manner in both α-catulin knockdown and controls, however, was significantly lower in α-catulin knockdown cells. In the cytoplasm cytochrome c binds to APAF-1 and then activates caspase-9 and further caspase-3/7. To analyse whether these steps of the intrinsic apoptosis cascade are also influenced by α-catulin we treated the cells with different concentrations of cisplatin and detected caspase-9 and-3/7 using a caspase luminescence assay. demonstrates that caspase-9 and caspase-3/7 activities were significantly enhanced in α-catulin-knockdown cells treated with 2.5, 5 or 10 μg/ml cisplatin after 6 hours. As described previously, cisplatin can also induce apoptosis via extrinsic pathways which means that cisplatin-resistance can only be achieved when both the intrinsic and the extrinsic apoptotic pathway are inhibited. We therefore performed a caspase-8 luminescence assay and found that α-catulin-knockdown also increased caspase-8 activity. Taken together, these results indicate that gene suppression of α-catulin in malignant melanoma cells enhances cisplatin-induced apoptosis in both the intrinsic and the extrinsic pathways. These findings suggest that the high level of α-catulin in melanoma cells contributes to drug resistance against cisplatin treatment. # Discussion Cisplatin was described as one of the most potent antitumor agents with clinical activity against many different cancers. The covalent binding of cisplatin to chromosomal DNA leads to the formation of DNA adducts and activates several signal transduction pathways contributing to apoptosis. Resistance of tumor cells to chemotherapeutic agents is a limiting factor of chemotherapy and therefore a better understanding of the mechanisms of chemoresistance is of great interest. In this study, we showed that high expression of α-catulin played a critical role in resistance of melanoma cells to cisplatin. First, we demonstrated that melanoma cells exhibit a significantly higher expression of α-catulin and a concomitant elevated activity of the transcription factor NF-ĸB. In opposition to malignant melanoma cells, normal human melanocytes (NHM) showed no or only low NF-ĸB activity and a low α-catulin expression level. Transfection of α-catulin in melanocytes augmented NF-ĸB activity in a dose dependent manner, whereas α-catulin knockdown reduced it. Significant NF-ĸB activity was found in α-catulin expressing cells after stimulation with TNF-α, LPS, HGF or Serum. Co- transfection of α-catulin with IKKβ demonstrated a significant NF-ĸB activation in α-catulin expressing cells whereas co-transfection of α-catulin with p65 had only a very low effect, confirming that α-catulin operates upstream of p65 in the NF-ĸB signaling pathway. Furthermore, the result proved that α-catulin activate NF-ĸB by repressing E-cadherin, a fundamental event in EMT. There is ample evidence that α-catulin plays an important role in tumorigenesis, and that activation of NF-ĸB pathway is the major signal for the induction of EMT, a developmental mechanism that is characterized by loss of cell-cell adhesion and polarity followed by a disruption of cytoskeletal organization toward a more mesenchymal phenotype. EMT has been closely linked to progression and invasion of different tumors and has been described to be regulated by MAPK-ERK signaling pathways by activation of ZEB1/2 via Fra1, which is an important member of the AP-1 transcription factor complex. Additionally, activation of ZEB1/2 has been connected to NF-ĸB signaling in breast cancer cells, suggesting an involvement of the MAPK, NF-ĸB and AP-1 pathways in tumorigenesis through activation of EMT. Furthermore, MAPK, NF-ĸB and AP-1 are known to play an important role in cisplatin-mediated apoptosis. Therefore, we sought to examine the effect of α-catulin knockdown on the activation of AP-1 and the phosphorylation of ERK, JNK and p38. depicts that α-catulin knockdown reduced AP-1 activation and ERK phosphorylation. However, there are controversial reports about the contribution of MAPK to apoptosis or cell survival depending on the cell type and extent of DNA damage. Dent and Grant (2001) reported that activation of ERK and JNK MAPK cascades by cisplatin counteract apoptosis. Agreeing with this, Persons et al. (1999) showed that ERK inhibition increased the cisplatin-sensitivity of ovarian cancer by accumulating p53. In contrast, Wang et al. (2000) showed that ERK activation is essential for cisplatin-mediated apoptosis in HeLa and lung carcinoma cells. Therefore, we were interested how α-catulin influences different signaling pathways in cisplatin-treated melanoma cells and whether α-catulin knockdown leads to increased sensitivity to chemotherapeutics. Hence, the cisplatin treated cells were analyzed for NF-ĸB, MAPK and AP-1 signaling. In cisplatin treated melanoma cells, α-catulin knockdown reduced NF-ĸB activity, ERK-, JNK- and c-Jun phosphorylation and the protein level of the target genes Mcl-1 and CBP. Consistent with our findings Mirmohammadsadegh et al. (2007) reported that increased phosphorylation of ERK1/2 in melanoma cells promoted tumor progression and partially prevented cisplatin-mediated apoptosis. As a consequence, we treated the melanoma cells with the chemotherapeutic drugs cisplatin, dacarbazine and paclitaxcel to elucidate the effect of α-catulin knockdown along with reduced AP-1 activatioin and ERK and JNK phosphorylation in melanoma cells. α-Catulin knockdown enhanced the susceptibility of melanoma cells to the chemotherapeutic drugs cisplatin, dacarbazine and paclitaxcel (.). This result suggests that NF-ĸB and AP-1 activation and ERK phosphorylation leads to reduced cell death in melanoma cells after treatment with chemotherapeutic agents. Accordingly, Mandic et al. (2001) reported that ERK acted as a pro-survival protein in melanoma cells. Additionally, we treated the cells with staurosporine, a potent inducer of apoptosis, which also showed increased sensitivity of α-catulin knockdown cells (.). The increased expression of NF-ĸB, p-ERK and AP-1 due to cisplatin treatment raised the question, whether cisplatin treatment leads to enhanced proliferation in melanoma cells. Therefore, we analyzed Ki67 expression and found that cell proliferation was not altered due to cisplatin treatment in melanoma cells expressing α-catulin. However, Ki67 expression was significantly reduced in α-catulin knockdown cells treated with higher concentrations of cisplatin. Same results could be confirmed with the BrdU Assay suggesting that α-catulin knockdown reduces cell proliferation in cisplatin-treated melanoma cells. Furthermore, the expression of the cell cycle inhibitors p21^waf/cip1 and p53 demonstrated a dose dependent increase in α-catulin knockdown cells after cisplatin treatment. Next, we investigated the role of α-catulin in cisplatin induced apoptosis. α-Catulin knockdown significantly enhanced apoptosis after cisplatin treatment. Cytochrome c release from mitochondria was described as a major step in the induction of apoptosis mediated by cisplatin. Therefore, we clearly demonstrated that cytochrome c release was increased in α-catulin knockdown cells after cisplatin treatment and that this effect was accompanied by mitochondrial membrane depolarization, caspase-9, -8 and-3/7 activation. # Conclusions In summary we show in this study, that the high level of α-catulin in melanoma is responsible for NF-κB, AP-1 activation and ERK phosphorylation and contributes to a reduction in cisplatin-mediated apoptosis. α-catulin knockdown reduced cisplatin-mediated cell proliferation and enhanced apoptosis in melanoma cells compared to the control cells. Together with our previous findings, these results demonstrate that α-catulin plays an important role in tumor progression, metastasis and chemoresistance suggesting α-catulin as a novel promising target for melanoma therapy. # Supporting Information [^1]: There exists no competing interests and there is no financial intent to market the results of this manuscript somehow. Most of the results were produced at the University of Applied Sciences in Krems, Austria where the first author (B. Kreiseder) was employed before she changed to SeaLife Pharma GmbH. The last author (C. Wiesner) is employed at the University in Krems and at SeaLife Pharma GmbH. Just very few results were produced at SeaLife Pharma mainly to finalize the manuscript. The research of SeaLife Pharma GmbH focuses on the production (isolation) of anti-infective molecules such as Anthraquinone-Derivatives for the treatment of MRSA. There is no interest in cancer research. Up to now, SeaLife Pharma has no products on the market. On behalf of all authors, I declare that there are no competing interests for the purposes of transparency. We confirm that this does not alter our adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: CW AE HH JAS RdM. Performed the experiments: BK YMHS BM NJ CW. Analyzed the data: CW AE BK. Contributed reagents/materials/analysis tools: AP RdM HH. Wrote the paper: CW BK JAS RdM.

# Introduction Considerable evidence links stress with the onset and symptom exacerbation in irritable bowel syndrome (IBS). To better understand the underlying mechanisms underlying this stress sensitivity, and to identify novel targets for drug development, stress-based animal models for IBS have been established and extensively studied, using as stressors electric foot shock, maternal separation, social defeat and overcrowding, as well as repeated water avoidance stress (WAS). Visceromotor responses measured as abdominal electromyographic signals evoked by colorectal distension (CRD), are most commonly used to assess stress-induced visceral hyperalgesia, modeling a cardinal symptom of IBS. However, given the multidimensional nature of pain, the visceromotor response in rodents likely reflects only a portion of the complex human visceral pain experience. In recent years, functional brain imaging technology has emerged as a powerful tool to bridge the measurement gap between preclinical and clinical pain research, providing an objective measurement of pain in humans and laboratory animals alike. Comparing alterations in CRD-evoked brain responses in stress-induced visceral hyperalgesic rodents and that reported in IBS patients by human brain imaging studies can provide important validation for the stress- based animal models for human IBS. A better understanding of such stress-induced alterations in brain nociceptive responses is critical to delineating the underlying mechanisms. There are few published reports on functional brain mapping studies in stress- induced visceral hyperalgesia animal models. Stam et al. examined in rats the effect of foot shock on CRD-evoked expression of c-Fos, a gene marker of neuronal activity. In the central amygdala, as well as prelimbic, infralimbic, insular, and cingulate cortices, previously shocked rats showed *reduced* c-Fos expression following CRD compared with no-shock controls. Wouters et al. used H₂¹⁵O microPET to map CRD-evoked functional brain activation in maternal-separated rats before and 1 day after 1 hour of WAS. Following WAS, rats showed CRD-evoked activation in new areas, including the somatosensory cortex and hippocampus, and greater deactivation in the frontal cortex. While these studies provided important evidence that stress-induced visceral hyperalgesia is associated with alterations in brain responses to CRD, due to the use of anesthesia in both studies, it is difficult to compare the results directly with human brain imaging findings. We have recently adapted an autoradiographic cerebral blood flow (CBF) perfusion mapping method to the rat CRD model. In contrast to the requirement of sedation or restraint in fMRI and microPET studies, the perfusion method allows functional brain mapping in awake and nonrestrained rats. This is particularly important when studying brain mechanisms related to stress and affect related pain modulation, as brain networks involved in nociception, stress and affect significantly overlap, and are subject to influences by anesthetics. Using this method, we have shown that patterns of brain activation in response to acute CRD and in expectation of CRD in the rat are in general agreement with that reported in the human brain imaging literature. Here, we applied perfusion mapping to characterize the effect of repeated WAS, which we have previously shown to induce long-lasting visceral hyperalgesia, on CRD-evoked functional brain activation. Repeated, daily WAS (7–10 days) induces a chronic visceral hyperalgesia in the rodent model. This hyperalgesia persists for periods as long as one month after cessation of the stress, something not seen after single day stress exposure. These observations suggest that chronic/subchronic stress results in a functional reorganization of the nociceptive response. It has been proposed that an important brain mechanism underlying stress-induced visceral hyperalgesia involves chronic stress induced impairment of prefrontal-cortico-limbic pain modulation. The prefrontal cortex (PFC) has been implicated in this corticolimbic regulatory circuit. To examine this hypothesis, we applied seed- region correlational analysis to assess changes in CRD-evoked functional connectivity (FC) of the PFC in the rat following stress. The prelimbic cortex (PrL) of the PFC was chosen as the seed based on our previously reported findings of robust activation of this region both during acute CRD and in expectation of CRD, and the observation that activity of PrL and amygdala were anticorrelated (e.g. showed a negative correlation) during expectation of CRD. One major limitation of FC analysis is that correlation based analysis does not address causality. Furthermore, due to multiple comparisons, false positive findings are inevitable when a simple significance threshold is applied. Constraining FC analysis with structural connectivity (SC) information can reduce the number of false positive reports, as well as provide directionality for the otherwise non-directional FC networks. The concept of anatomically constrained FC analysis has been implemented in effective connectivity analysis of human and animal brain imaging data, but has been largely limited to small- scale networks. Recent studies have suggested a direct association between FC and SC in the human brain by combining resting-state fMRI with structural diffusion tensor imaging (DTI) measurement. With the recent surge in efforts to construct connectome databases for human and rodent brain, it has become possible to combine and compare SC and FC at the whole brain level. Here, we constrained FC analysis with *complete* SC information of the PrL based on reports from tract tracing experiments manually collated in the Brain Architecture Management System (BAMS, <http://brancusi.usc.edu/>). The resulting *structurally linked functional connectivity* (SL-) network keeps only functional connections over direct structural projections. An SLFC network inherits directionality information from the SC network and the sign of functional interaction (positive or negative) from the FC network, constituting a substantive step toward understanding the causality in brain circuits. # Materials and Methods ## Animals Adult male Wistar rats (2–2.5 month old) were purchased from Harlan Sprague Dawley (Indianapolis, IN, USA) and were individually housed in the vivarium on a 12-hour light/12-hour dark cycle with free access to water and rodent chow. All experiments were conducted under a protocol approved by the Institutional Animal Care and Use Committee of the University of Southern California, an institution accredited by the Association for Assessment and Accreditation of Laboratory Animal Care, International. All work was in accordance with the guidelines of the Committee for Research and Ethical Issues of the International Association or the Study of Pain. The numbers of animals in each group were as follows: Sham stress/0-mmHg CRD, n = 9; Sham stress/60-mmHg CRD, n = 10; WAS/0-mmHg CRD, n = 10; WAS/60-mmHg CRD, n = 10. ## Surgical procedures One week before the start of WAS treatment, animals were anesthetized (isoflurane 2% in 70% oxygen and 30% nitrous oxide). The right external jugular vein was cannulated with a 5 French silastic catheter (Dow Corning Corp., Midland, MI, USA), advanced into the superior vena cava. The port at the distal end of the catheter was tunneled subcutaneously and externalized dorsally in the region rostral to the scapula. Subsequently, a telemetry transmitter was implanted to measure abdominal EMG. Such implants can be turned on and off with an external magnet and send a radiofrequency signal of EMG acitvity to a receiver platform placed underneath the rat's cage. The body of the transmitter (TA11-CTA-F40, Data Sciences Intl., St. Paul, MN, USA) was implanted subcutaneously on the dorsum of the animal caudal to the scapula. A skin incision was made on the abdomen and electrodes of the transmitter were tunneled subutaneously to the abdominal incision. Tips of the eletrodes were bared, placed in parallel (0.5 cm apart), and stitched into the left external oblique musculature, just superior to the inguinal ligament. The receiver platform was linked via a data exchange matrix to a PC computer. All animals were allowed to recover for seven days. The catheter was flushed every other day postoperatively to ensure patency (0.3 mL of sterile 0.9% saline, followed by 0.1 mL taurolidine-citrate catheter lock solution, Access Technologies, Skokie, IL, USA). ## Assessment and quantification of the VMR to CRD VMR to CRD was assessed as described before. Briefly, under light isoflurance anesthesia (1.5% isoflurance × 3 min), a flexible latex balloon (length = 6 cm) was inserted intra-anally such that its caudal end was 1 cm proximal to the anus. The silicon tubing connecting the balloon and the barostat (Distender Series II, G&J Eletronics Inc., Toronto, Canada) was fixed to the base of the tail with adhesive tape and covered by a stainless steel spring for protection against animal biting. Animals were allowed to recover for 30 min in the experiment cage, the floor of which was covered with bedding from the animal's home cage. The CRD procedure consisted of two series of phasic distension to constant pressure of 10, 20, 40, and 60 mmHg with 20-s duration and 4-min interstimulus intervals. The visceromotor response was quantified by measuring EMG activity in the external oblique musculature. EMG signals were recorded telemetrically at a sampling rate of 1 kHz, digitized and stored on a PC computer with the Dataquest ART 3.0 software (Data Sciences Intl., St. Paul, MN, USA). EMG waveforms were lowcut filtered at 20 Hz to eliminate movement interference, and then full-wave rectified. Area under the curve (AUC) was calculated for the 20-s distension period normalized by the 20-s before- distension baseline. The visceromotor response was assessed on the day before (day 0) and the day after (day 11) WAS or sham treatment. EMG AUC was further normalized to response to 60-mmHg CRD on day 0 and expressed as a percentage of this baseline value. EMG signals were not recorded in 1 sham-treated rat due to equipment failure and not included in the visceromotor response analysis. ## Water avoidance stress protocol The protocol was as described before. Briefly, the test apparatus consisted of a plexiglas tank (45 cm length × 25 cm width × 20 cm height) with a block (8 cm length × 8 cm width × 10 cm height) affixed to the center of the floor of the tank. The tank was filled with fresh room temperature water (25°C) to within 1 cm of the the top of the block. The animals were placed on the block for a period of 1 hour daily for 10 consecutive days. Sham treatment consisted of placing the rats in the dry tank for 1 hour daily. ## Cerebral perfusion On day 11, animals were allowed to rest for 15 min following the last distension of the CRD series. A piece of silastic tubing was filled with radiotracer \[¹⁴C\]-iodoantipyrine (125 µCi/kg in 300 µL of 0.9% saline, American Radiolabelled Chemicals, St. Louis, MO, USA). The radiotracer-filled tubing was then connected to the animal's cannula on one end, and to a syringe filled with euthanasia agent (pentobabital 50 mg/mL, 3 M potassium chloride) on the other. The animal was allowed to rest for another 5 min before receiving one episode of 60-mmHg. Thirty-five seconds after the onset of the distension, radiotracer was infused at 2.25 mL/min by a motorized pump, followed immediately by 0.7 mL of euthanasia solution, which resulted in cardiac arrest within ∼10 s, a precipitous fall of arterial blood pressure, termination of brain perfusion, and death. This 10-s time window provided the temporal resolution during which the distribution of regional CBF (rCBF)-related tissue radioactivity was mapped. Half of each treatment group received no distension (0 mmHg) during CBF mapping and served as controls. ## Brain slicing and autoradiography Brains were rapidly removed, flash frozen in methylbutane on dry ice (∼−55°C) and embedded in optimal cutting temperature compound (Sakura Finetek Inc., Torrance, CA,USA). Brains were subsequently sectioned on a cryostat (HM550 series, Microm International GmbH, Walldorf, Germany) at −18°C into 20-µm thick coronal slices, with an inter-slice spacing of 300 µm. Slices were heat-dried on glass slides and exposed to Kodak Biomax MR films (Eastman Kodak, Rochester, NY, USA) for 3 days at room temperature. Images of brain sections were then digitized on an 8-bit gray scale using a voltage stabilized light box (Northern Lights Illuminator, Interfocus Imaging Ltd., Cambridge, UK) and a Retiga 4000R charge-coupled device monochrome camera (Qimaging, Surrey, Canada). Autoradiographic CBF mapping in rodents has a spatial resolution of 100-µm, and hence, can provide information on sub-regional activation, such as in individual amygdaloid nuclei. ## Functional brain mapping data analysis rCBF-related tissue radioactivity was quantified by autoradiography and analyzed on a whole-brain basis using statistical parametric mapping (SPM, version 5, Wellcome Centre for Neuroimaging, University College London, London, UK). Recently, we and others have developed and validated an adaptation of SPM for use in rodent brain autoradiograph. In preparation for the SPM analysis, a 3-dimensional reconstruction of each animal's brain was conducted using 57 serial coronal sections (starting at ∼ bregma +4.5 mm) with a voxel size of 40 µm ×300 µm ×40 µm. Adjacent sections were aligned manually in Photoshop (version 9.0, Adobe Systems Inc., San Jose, CA, USA) and using TurboReg, an automated pixel-based registration algorithm implemented in ImageJ (version 1.35, <http://rsbweb.nih.gov/ij/>). This algorithm registered each section sequentially to the previous section using a nonwarping geometric model that included rotations and translations (rigid-body transformation) and nearest- neighbor interpolation. One “artifact free” brain was selected as reference. All brains were spatially normalized to the reference brain in SPM. Spatial normalization consisted of applying a 12-parameter affine transformation followed by a nonlinear spatial normalization using 3D discrete cosine transforms. All normalized brains were then averaged to create a final rat brain template. Each original 3D-reconstructed brain was then spatially normalized to the template. Normalized brains were smoothed with a Gaussian kernel (FWHM  = 3× voxel dimension in the coronal plane). A nonbiased, voxel-by-voxel analysis of regional brain activation was performed. Voxels for each brain failing to reach a specified threshold in optical density (70% of the mean voxel value) were masked out to eliminate the background and ventricular spaces without masking gray or white matter. We implemented a Student's *t*-test at each voxel. For each treatment (WAS or sham), a *t*-contrast was calculated comparing the 60-mmHg CRD to the 0-mmHg control subgroup. Threshold for significance was set at *P*\<0.05 at the voxel level and an extent threshold of 100 contiguous voxels. This combination reflected a balanced approach to control both type I and type II errors. The minimum cluster criterion was applied to avoid basing our results on significance at a single or small number of suprathreshold voxels. Brain regions were identified according to a rat brain atlas. In addition, we ran a factorial analysis to identify rCBF changes reflecting WAS x CRD interaction. Threshold for significance was set at *P*\<0.05 at the voxel level and an extent threshold of 100 contiguous voxels. Data interpretation was focused on gray matter. ## Structurally linked functional connectivity of the prelimbic cortex To test the hypothesis that WAS may result in altered corticolimbic modulation during noxious visceral stimulation, we applied seed-region correlation analysis to assess differences in the CRD-evoked FC of the PrL of the PFC between treatment groups. The seed region of interest (ROI) was hand drawn in MRIcro (version 1.40, <http://cnl.web.arizona.edu/mricro.htm>) for the right hemisphere over the template brain according the rat brain atlas and intersected with clusters defining regional functional activation in the PrL area based on the SPM analysis. The result was one unilateral seed ROI representing the PrL region for each treatment type showing CRD-evoked functional activation. Mean optical density of the seed ROI was extracted for each animal using the MarsBaR toolbox for SPM (version 0.42, <http://marsbar.sourceforge.net/>). Correlation analysis was performed in SPM for each 60-mmHg CRD subgroup using the seed values as a covariate. Threshold for significance was set at *P*\<0.05 at the voxel level and an extent threshold of 100 contiguous voxels. Regions showing significant correlations (positive or negative) in rCBF with the PrL are considered functionally connected with the PrL. Anatomical (structural) connectivity of the PrL in the rat was extracted from BAMS. BAMS includes a large set of rat structural connections collated from the literature, or directly inserted by neuroanatomists,. The collation methodology of neuroanatomical data employed in BAMS is fully described in Bota et al.. Briefly, the connectivity patterns of gray matter regions are collated as reported in the published references recorded in BAMS, or as inferred by collators. The connectivity information is collated from the textual descriptions, and from the maps associated with references. Each connectivity report is associated with a qualitative strength, and with supporting textual annotations. Connectivity reports inferred by collators are associated with textual annotations that describe the annotation process. Overall, the rat PrL is associated with about 1600 connectivity reports in BAMS, with 106 gray matter regions that receive inputs from it and 177 regions that send outputs to it. The inputs of the rat PrL are associated with 37 references, and its outputs 62 references. The set of qualitative strengths of neuroanatomical connections collated in BAMS includes 10 values. Here, this set was encoded on a linear scale from 1 to 7, with 1 being very strong and 7 being very weak. Regions with ‘very weak’ connection to or from PrL, as well as connectivity reports with the strength ‘fibers of passage’, were removed for simplification. Only ipsilateral connections were included due to incomplete understanding of cross-hemispheric connectivity. Results of the SPM seed correlation analysis were only analyzed for those regions structurally connected with the PrL. This is equivalent to taking an intersection of the structural and functional connectivity network of the PrL, resulting in an SLFC network. # Results ## Water avoidance stress induced visceral hyperalgesia Ten days of WAS induced significant increases in visceromotor response to CRD on day 11 as compared to day 0 baseline (*n* = 20, main effect of ‘Day’ *F*(1, 19) = 18.19, *P*\<0.001, two-way repeated measures ANOVA with ‘Day’ and ‘CRD’ as the within-subjects factors). In comparison, the visceromotor response was only moderately increased in the sham-treated animals primarily due to an increase in response to 60-mmHg CRD (, *n* = 18, main effect of ‘Day’, *F*(1,17) = 4.67, *P* = 0.045, two-way repeated measures ANOVA). Compared to the sham condition, stressed rats showed significantly greater increases in the visceromotor response on day 11 (main effect of ‘Treatment’, *F*(1,36) = 4.47, *P* = 0.042, mixed model ANOVA with CRD as the within-subjects factor and ‘Treatment’ the between-subjects factor). ## Comparison of CRD-evoked functional brain activation in WAS- and sham-treated rats CRD-evoked functional brain activation was assessed for each treatment type by contrasting the subgroup receiving 60-mmHg CRD, and the one receiving no CRD (0-mmHg control) of the same treatment type. Sham-treated rats showed CRD-evoked functional activation (increase in rCBF) in the PrL, primary motor, frontal area 3, primary and secondary somatosensory, anterior and posterior insular, and temporal association cortices, as well as in the dorsal caudate putamen, amygdala (lateral amygdaloid n., central amygdaloid n.), and superior olive. Sham rats also showed CRD-evoked deactivation (decrease in rCBF) in the retrosplenial, entorhinal, piriform, and secondary visual cortices, hippocampus, subiculum, thalamus (habenular n., mediodorsal n., posterior n. group, parafascicular n., ventral posteromedial n., ventral posterolateral n.), cerebellum (cerebellar lobule, cerebellar hemisphere), and areas of the brainstem (superior colliculus, dorsomedial periaqueductal gray, red n.; caudal linear n. of the raphe). In contrast, WAS-treated rats showed CRD-evoked brain activation in a drastically different pattern. Major differences were noted in the magnitude and extent of regional activation. Greater activation in the WAS rats was noted in the anterior and posterior insula, and in the amygdala (central n., lateral n., basolateral n., basomedial n., medial n., bed nucleus of the stria terminalis intraamygdaloid division). The WAS rats also showed significant activation in the hypothalamus (medial preoptic area, medial preoptic n., lateral preoptic area, ventromedial hypothalamic n.), nucleus accumbens, and bed nucleus of the stria terminalis (medial division), which was not seen in the sham rats. Whereas sham rats showed activation in the anterior dorsal aspect of caudate putamen, WAS rats showed activation in the posterior and anterior ventral aspects of caudate putamen. Importantly, *reduced* activation in the PrL was noted in the WAS rats compared to sham. Deactivation in the cingulate cortex area 1 and 2 (Cg1, Cg2) was noted in the WAS but not the sham rats. WAS rats showed a similar pattern of deactivation as that seen in the sham condition, though to a lesser extent. Factorial analysis confirmed significant WAS x CRD interaction in the PrL, posterior insula, retrosplenial cortex, secondary visual cortex, anterior striatum, accumbens nucleus, bed nucleus of the stria terminalis medial division, medial preoptic nucleus, hippocampus, and central nucleus of the amygdala. ## Differences in the structurally linked functional connectivity of the prelimbic cortex Seed correlation-based FC analysis was constrained by SC information of the PrL to generate SLFC of the PrL cortex. In sham rats, PrL/PFC SLFC during noxious visceral stimulation was characterized by negative FC with the amygdala, whose nuclei provide either afferent input (basomedial n., posterior part; medial n., posteroventral part; amygdalohippocampal area, posteromedial part; posterolateral cortical n.; amygdalopiriform transition area) (PrL←Amygdala) or bidirectional connections of the lateral amygaloid nucleus to the PrL (PrL←→La). In addition, sham treated rats demonstrated the following SLFC of PrL cortex during noxious visceral stimulation: 1\. Positive FC with a cluster of cortical regions over largely bidirectional structural connections, including orbital (ventral), secondary motor, cingulate, retrosplenial (dysgranular, granular), anterior insular, and ectorhinal cortices (PrL\[+\]←←Ctx). 2\. Positive FC over projections to the striatum (PrL\[+\]←Striatum) or from lateral orbital cortex (PrL\[+\]←LO). 3\. Negative FC over afferent projections from the hippocampal formation (CA1, dorsolateral entorhinal cortex) (PrL←Hippocampal formation). 4\. Negative FC with the hypothalamus over efferent projections to medial preoptic nucleus, the anterior hypothalamic area (anterior part), and subventricular zone (PrL←Hypothalamus), and afferent projections from anterior hypothalamic (central part) and supramammillary nucleus (PrL←Hypothalamus). 5\. Positive FC with the thalamus over efferent projections to the ventral anterior, ventral lateral, anterior ventral, and reticular nuclei, and medial and lateral habenular nuclei (PrL\[+\]←Thalamus), as well as bidirectional connections with medial dorsal and central nuclei (PrL\[+\]←←Thalamus). 6\. Negative FC over efferent projection to the dorsomedial and dorsolateral periaqueductal gray in the midbrain (PrL←PAG). WAS-induced changes in SLFC were most noticeable in relation to a cluster of cortical regions and to the amygdala. Whereas PrL in the sham animals showed significant, positive FC with a cortical cluster consisting of secondary motor, dorsal posterior cingulate (pCg1), ventral cingulate (Cg2), retrosplenial (dysgranular, granular) cortices, this connectivity turned *negative* in the stressed rats. In the sham rats, negative FC was noted between the PrL and the amygdala whose nuclei provide afferent input (basomedial n., posterior part; medial n., posterior ventral part; amygdalohippocampal area, posterior medial part; posterolateral cortical n., amygdalopiriform transition area) or bidirectional connections (lateral n.) to the PrL. In contrast, in the stressed rats, this negative functional connectivity was absent or turned positive (lateral n., anterior cortical n., posterolateral cortical n.). In addition, WAS-induced alterations in SLFC of the PrL/PFC included the following: 1\. Changes from positive to negative FC with the thalamus (ventral anterior, ventral lateral, anterior ventral, mediodorsal, centromedial, medial and lateral habenular nuclei) (PrL→/←/← →Thalamus), subthalamic nucleus, and zona incerta. 2\. Changes from negative to positive FC with the hypothalamus (medial preoptic n.; anterior hypothalamic area, anterior and central parts; subventricular zone) (PrL\[+\]→/←Hypothalamus). 3\. New (no FC in sham rats) positive FC with infralimbic, posterior insular, and piriform cortices (PrL\[+\]→/← →Ctx), hypothalamus (medial mammillary n., ventromedial n., lateral preoptic area)(PrL\[+\]←/← →Hypothalamus), dorsal subiculum, nucleus accumbens, claustrum, dorsal endopiriform nucleus, ventral pallidum, and nucleus of the diagonal band. 4\. New, negative FC with thalamus (posterior, reuinens, rhomboid, centrolateral nuclei)(PrL→/←/← →Thalamus), and brainstem areas (substantia nigra reticulata and compacta, isthmic and mesencephalic reticular formation, lateral and ventrolateral periaqueductal gray, dorsal raphe n., dorsal and laterodorsal tegmental n., pontine n., central gray, subcoeruleus n. alpha)(PrL→/← →Brainstem). # Discussion Neurobiological sequelae of chronic stress have been the subject of extensive research. For example, it has been well established that chronic stress in rodents can induce both visceral, as well as somatic hypersensitivity. Few studies have examined stress-induced changes in functional activation of brain nociceptive circuits, which can bring new insights into the underlying mechanisms. Using autoradiographic, perfusion-based functional brain mapping, we studied regional and network-level neural correlates of WAS-induced visceral hyperalgesia in awake, nonrestrained rats. Our main findings were that stressed rats showed greater CRD-evoked activation than sham-treated rats in the amygdala, insular cortex, and hypothalamus, but *reduced* activation in the prelimbic area of PFC (PrL/PFC). Profound differences between the stressed and sham rats were noted in the structurally linked functional connectivity of the PrL/PFC with cortical, limbic and brainstem areas. In particular, while negative functional connections were noted between the PrL/PFC and amygdala in the sham condition, these were absent in the stressed rats. These findings, in association with an exaggerated visceromotor response to CRD in the stressed animals, provide direct evidence that stress amplifies sensory and affective responses to nociceptive stimuli, with impairment of PFC-mediated pain modulation as a candidate central mechanism for stress-induced visceral hyperalgesia. We focus the discussion on stress-induced changes. ## Stress-induced changes in activation of amygdala and PFC evoked by noxious CRD In the current study, stressed compared to sham treated rats showed greater CRD- evoked activation broadly across the amygdala, with the central amygdaloid nucleus showing the most significant differences. These findings are consistent with an extensive literature on stress induced sensitization of the amygdala. For example, the amygdala has been implicated in chronic stress-induced sensitization of anxiety- and fear-related responses to an acute stressor. In particular, both chronic stress and chronic pain can increase excitability and responsiveness of subsets of neurons of the central n., a primary output of the amygdala to brain regions involved in autonomic regulation. Chronic restraint stress has also been shown to increase excitability of pyramidal neurons in the lateral amygdaloid nucleus, as well as to increase the number of dendritic spines in the amygdala. Such stress-induced neural plasticity may mediate enhanced responses to noxious visceral stimulus. Further, chronic local application of the stress hormone corticosterone to the amygdala leads to visceral hyperalgesia, suggesting the amygdala as a site where stress hormone can modulate visceral nociception. Activation of the PrL during acute visceral pain in nonstressed rats has previously been reported. In the current study, stressed compared to sham treated rats showed a striking hypoactivation in the PrL in response to CRD. The PrL is a part of the medial PFC in rodents. This brain region is thought to have features of dorsolateral PFC and anterior cingulate cortex of primates. The PFC is a brain area vulnerable to chronic stress. For example, chronic restraint stress – and corticosterone administration have both been shown to induce retraction of dendrites and loss of synaptic connections in the PFC, in line with reduced functional activation as observed in the current study. Chronic psychosocial stress in human subjects impairs PFC-dependent attentional control and disrupts FC within a frontoparietal network, including the PFC. In contrast to results of the current study, Gibney et al. reported exaggerated c-fos expression in the PrL in visceral hypersensitive Wistar-Kyoto rats compared to control Sprague-Dawley rats, albeit in the absence of a chronic stressor. These differences in PrL changes associated with visceral hyperalgesia may be attributable to different animal models of visceral hyperalgesia used, different experimental protocols (anesthetized vs. awake rats, no stress vs. WAS), and different modalities of measurement (c-fos vs. cerebral blood flow). ## Structurally linked functional connectivity analysis and connectome Functional interaction between brain regions can be analyzed through FC analysis of brain imaging data. Here, we focused FC analysis on the PrL and constrained FC results with SC information of the PrL. The integration of whole brain-level FC and SC information reflects a recent trend in human brain imaging field to take advantage of rapid advances in the Human Connectome Project. With the Mouse Connectome Project (<http://www.mouseconnectome.org/>), BAMS, and other rodent structural connectome projects well underway, a similar approach has become feasible for animal research. Based on the complete information of the SC of the PrL collated in BAMS, we were able to thoroughly investigate stress-induced changes in SLFC of the PrL. The first advantage of SLFC over regular FC is the simplification of the connectivity network by the removal of FC connections not substantiated by structural projections. FC connections without SC may be false positive findings, or indirect FC connections. The second advantage of SLFC is that it combines SC directionality with the sign of FC. The resulting SLFC can help generate new hypotheses about the causality in the brain circuits. While both SC and FC data were binarized for simplification in this study, the strength of each SC and FC connection can also be integrated into SLFC analysis. The current study represents one of the first rodent studies to present a detailed integration of structural information into FC analyses. ## Structurally linked functional connectivity of the PFC-amygdala circuit and the effects of stress In rodents, the PrL is reciprocally connected with the lateral and basolateral nuclei of the amgydala, and sends efferent projections to the central nucleus and anterior part of basomedial nucleus, and receives afferent projections from medial, posterior, and cortical nuclei and posterior part of the basomedial nucleus of the amygdala. In sham rats, PrL/PFC SLFC during CRD was characterized in our study by negative FC with the amygdala nuclei over mostly afferent projections from the amygdala (PrL←Amygdala), except bidirectional connection with the lateral nucleus (PrL← →La). These SLFC results suggest that in sham animals, PrL/PFC inhibits the amygdala through its projection to the lateral amygdalar nucleus, whereas the amygdala may in return inhibit PrL/PFC activity though its lateral, basomedial, medial, and cortical nuclei. Bidirectional PFC-amygdala interactions have been extensively studied in humans and in laboratory animals, and changes in these interactions have been implicated in the regulation of negative emotion, mood and pain. The inhibitory interaction between the PrL/PFC and amygdala is likely bidirectional. On the one hand, it has been well documented that PFC regulates amygdala-mediated responses to aversive stimulus. On the other hand, ample evidence exists for amygdala- mediated modulation of PFC, particularly under aversive condition. We have recently applied correlation-based FC and graph theoretical analysis to characterize brain activation at the network level in expectation of visceral pain. Animals previously trained with a step-down passive avoidance paradigm using noxious CRD as the aversive stimulus demonstrated negative FC between the amygdala and areas of the PFC (including PrL and dorsal and ventral cingulate cortex, Cg1, Cg2) when reexposed to the conditioned context—a finding that was interpreted as evidence for inhibitory corticolimbic modulation. In the current study, stressed rats compared to sham rats showed substantial changes in the SLFC of PrL/PFC. The negative FC with the amygdala seen in the sham disappeared, and in its place appeared a few positive connections (with the lateral and corticoamygdaloid nuclei). Our results are consistent with those of Correll et al. who using in-vivo extracellular neural recordings reported that chronic cold stress enhanced the acute footshock-induced response of the central amygdaloid nucleus, and that chronic stress weakened prefrontal inhibitory regulation of this response. Our data suggest similar chronic stress-induced disinhibition of the amygdala by the PFC. ## Stress-induced changes in activation of other brain regions during noxious CRD Previous imaging studies in animal and human subjects \[reviewed by 61\] have also implicated the anterior cingulate cortex in visceral pain processing and regulation, with the majority of visceral distension studies reporting enhanced activation of mid-cingulate subregions. Here, the observed deactivation in the cingulate area in the stressed rats and no activation in the sham rats was unexpected. Previously, we have reported cingulate activation in male, naïve rats (with no prior experience of CRD) receiving acute, noxious CRD. In the current study, rats received a series of CRD eleven days prior, as well as 20 minutes prior to the CBF perfusion procedure. This difference in protocol may have contributed to this different pattern of cingulate activation. For example, CBF level in the cingulate areas in the 0-mmHg control rats may have been elevated from baseline due to prior exposure to repeated, noxious CRD. The exact cause and implication of cingulate deactivation in the stressed rats remains to be further investigated. Whereas both sham and stressed rats showed CRD-evoked activation in the anterior and posterior insula, activation in the stressed rats was much greater in amplitude and extent. We have reported CRD-evoked activation of the anterior and posterior insula in normal (nonstressed), male rats , as well as activation of anterior insula in expectation of CRD. Activation of the insula in response to acute rectal distension is the most consistently reported finding in human brain imaging studies, and alterations of insular functional activation have been reported in IBS patients. The posterior insula in its role as primary interoceptive cortex mediates sensory processing of pain and is part of a sensorimotor network, whereas the anterior insula is involved in a salience network closely linked to emotional arousal. In the rat, anterior insular cortex may modulate pain processing through its projection to the amygdala and periaqueductal gray in the rat. Stressed rats also showed significant CRD-evoked activation in the hypothalamus, bed nucleus of stria terminalis, accumbens nucleus and ventral striatum, but deactivation in the cingulate cortex, which were all absent in the sham rats. The hypothalamus is believed to be modulated by PFC, and in turn regulates activity of descending inhibitory and facilitatory pathways through periaqueductal gray and pontomedullary nuclei,. The ventromedial hypothalamus has also been implicated in the generation of the affective dimension of pain. Increased hypothalamic activation to rectal distension has been observed in IBS patients as compared to healthy controls. The bed nucleus of the stria terminalis, considered extended amygdala, receives heavy projection from the basolateral amygdala and projects to the hypothalamus and brainstem areas, and participates in anxiety and stress responses. Chronic immobilization stress has been shown to induce dendritic remodeling of neurons of the bed nucleus of the stria terminalis. Collectively, augmented activation of the hypothalamus, bed nucleus of the stria terminalis, and amygdala may underlie increased pain responses in the affective dimension, as well as increased descending facilitatory pain modulation. The nucleus accumbens and ventral striatum participate in reward responses and positive emotional states. The accumbens nucleus and ventral striatum are also considered part of the emotional motor system , serving perhaps as a gateway between the limbic system and the motor system. Human brain imaging studies have reported activation of these structures by noxious thermal stimuli, as well as in expectation of aversive somatic stimuli. Interestingly, in a treatment study of IBS patients, Berman et al. showed that Alosetron, a serotonin receptor antagonist, elicited decreases in rCBF in the amygdala, ventral striatum, and dorsal pons, in significant correlation with symptom (abdominal pain) reduction. Their findings suggest a possible role of the ventral striatum in central pain sensitization in IBS. ## Structurally linked functional connectivity of the PFC to other brain regions and the effects of chronic stress During visceral noxious stimulation, sham treated rats demonstrated a positive SLFC of the PrL cortex with the thalamus, anterior insula and other cortical areas. This is consistent with an integrative role of PrL/PFC in the processing of visceral input. In addition, the negative connectivity of PrL/PFC with limbic areas (amygdala, hypothalamus) and periaqueductal gray in the midbrain is consistent with PFC-limbic-periaqueductal gray inhibitory modulation. Water avoidance stress induced substantial changes in the SLFC of the PrL with the thalamus, hypothalamus, and brainstem. Connections with the thalamus which demonstrated significant positive FC in the sham, appeared significantly negative in the WAS rats, whereas FC showed the reverse pattern with the hypothalamus (excluding the subthalamic n., and zona incerta). In the stressed rats, but not the sham, PrL demonstrated significant negative FC with areas of the brainstem, including the substantia nigra, reticular formation, periaqueductal gray, dorsal raphe nucleus, tegmental nucleus, pontine nucleus, central gray, and subcoeruleus nucleus alpha. This complex pattern of alterations in SLFC suggests profound changes in how the PrL/PFC interacts with other cortical and subcortical areas during visceral pain processing as a result of chronic stress. ## Translational implications of rodent brain imaging findings in the study of pain Human functional brain imaging has been extensively applied to investigate central processing and modulation of pain, including visceral pain, and to characterize alterations in central pain responses in functional pain disorders, including IBS. A recent quantitative meta-analysis of imaging data from 19 published studies reported that compared to healthy controls, IBS patients have greater engagement of regions associated with emotional arousal (perigenual anterior cingulate cortex and amygdala) and homeostatic afferent processing (anterior mid-cingulate cortex, medial thalamic regions, mid-insula, areas of the midbrain). In contrast, controls show greater reliable activation in cortical regions involved in modulation of pain and emotion (lateral and medial prefrontal cortex, Brodmann Area 49), which is largely absent in IBS patients. In striking agreement with these human brain imaging findings, the current study showed that stressed rats, compared to sham, had much greater activation in the insula and amygdala, but reduced activation in PrL, a PFC region with features of the dorsolateral PFC, and the anterior cingulate cortex, of primates. This adds to our previous reports of remarkable homology in functional brain activation between the rat and human in response to acute noxious CRD in both males and females, and in expectation of CRD in males. In conclusion, chronic stress induced marked alterations in CRD-evoked functional brain activation characterized by *hypoactivation* of the prelimbic area of PFC and *hyperactivation* of the insular cortex, amygdala, and hypothalamus. Structurally linked functional connectivity analysis further revealed stress-induced disruption of PFC-limbic (amygdala and hypothalamus) inhibitory interaction during CRD. Dysfunction of the PFC, including impairment of the PFC-limbic regulatory circuit, is strongly implicated as a central mechanism contributing to stress-induced visceral hyperalgesia. The findings of hypoactivation in PFC and hyperactivation in limbic/paralimbic structures in the homeostatic afferent processing network and emotional arousal network are in general agreement with human brain imaging findings on altered brain responses to noxious visceral stimulus in IBS patients. These findings provide further support for the face and construct validity of the WAS animal model for human IBS. Functional brain mapping in awake, nonrestrained rodents can be a powerful tool for bridging animal and human visceral pain research, for gaining new mechanistic insights, and for preclinical drug evaluation with presumably greater predictive power. Future work will need to evaluate if activation/deactivation patterns reported in our study would be different and allow for animal-to-human translation if another painful stressor were used (e.g. electric shock; noxious heat). [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: ZW DPH EAM SB. Performed the experiments: ZW MAO RDP DPH. Analyzed the data: ZW MB DPH. Contributed reagents/materials/analysis tools: ZW MB DPH. Wrote the paper: ZW DPH EAM MB.

# Introduction The diet and foraging habits of organisms are critical components of their ecology and often determine where they occur. Understanding the diets of fossil organisms can provide information about long-term trends in vegetation structure and organismal responses to climate change. Multiple paleoecological proxy methods, including dental microwear, dental mesowear, and stable isotope analysis, are used to determine the textural properties and isotopic composition of dietary items (e.g.,). While dental microwear and stable isotope analysis require specialized equipment and a high level of training, dental mesowear is comparatively inexpensive and can be taught to novices in under an hour. Dental mesowear analysis involves examining the cusp shape and relief of teeth to determine the diets of herbivorous mammals (e.g., grazing, mixed feeding, and browsing). Apex cusp shape has traditionally been scored as sharp, round, or blunt, while relief (the difference in elevation between cusp and valley height) is scored as high or low. Subsequent iterations of dental mesowear for horses adopted a scoring system from zero to three, zero to four, and zero to six, that encompassed high relief and sharp cusps (0) to blunt and low relief teeth (3, 4, or 6, respectively). Blunted cusps and low relief across the occlusal surface are thought to be indicative of a phytolith-rich grazing diet (grasses also having a high-silica content and higher fibrousness that results in increased abrasion to the tooth surface during feeding) and/or one that includes a large amount of grit; although grass is a more likely contributor than grit. In contrast, browsers consume less siliceous foliage which causes less abrasion to an organism’s teeth; the occlusal wear on browse-dominant feeders allows for clean cuts and thus tooth-on-tooth attrition during mastication. Mixed feeders may utilize both browse and grass materials and typically have intermediate mesowear characteristics (e.g., rounded cusps and medium relief). Most notably, mesowear has documented dietary evolution of equids in North America from browsers to a cosmopolitan group of mixed-feeders and grazers. While the utility of dental mesowear has been expanded to include both upper and lower teeth and a diversity of ungulates (e.g., antelopes, bovids, bison, camels, and deer) and even marsupials, some aspects of dental mesowear are not well understood (e.g., positive relationships between relief height and carbon isotope values, an unexpected relationship—one would predict reduced relief height with increased carbon isotope values indicative of C₄ grass consumption). Further, there are gaps in our knowledge of how dental mesowear relates to an organism’s local environment. As mesowear is largely documented to infer an animal’s diet, relationships with climate variables may be expected if diet varies with climate (through the lens of vegetation consumed; e.g., eating more grass in drier environments). That being said, increased dust and/or grit in drier environments may also lead to blunter and/or lower relief cusps in mammals inhabiting arid environments. While several studies have investigated the relationship between mesowear and climate (including precipitation and relative humidity), results are equivocal. Mesowear may reflect local-to-regional relative humidity signals in wild African zebras with a high ratio of blunt and low relief cusps occurring in drier regions. When a broad diversity of ungulates were examined, comparing average dental mesowear scores to precipitation at the center of a species range, no relationship was apparent. Similarly, the mesowear of sika deer in Japan is not significantly correlated with precipitation. Further investigation into the relationships between dental mesowear and climatic variables is warranted. In Australia, marsupials are the dominant herbivores, including kangaroos, wallabies, wombats, possums, koalas, and others. To date, only one study of marsupial mesowear has been performed and demonstrated the potential of dental mesowear analysis for marsupial herbivores. However, one of the most ubiquitous and spatially and temporally distributed kangaroos could not be appropriately classified (the eastern gray kangaroo, *Macropus giganteus*). Further, an additional 18 of 43 species could not be classified into appropriate dietary categories based on dental mesowear alone. It is unclear if and why dental mesowear is not highly effective in marsupials and/or if climate is heavily influencing dental mesowear variables. Here, we examine the relationship between Australian marsupial mesowear, diet, and climate in three ubiquitous herbivores. We focus our analysis on extant species whose ranges are broad and dietary interpretations vary from grazing to mixed feeding: *Macropus giganteus* (a grazer to mixed feeder), *Macropus fuliginosus* (a grazer to mixed feeder). We also selected a species with a similarly broad range, which is an obligate browser and eucalyptus specialist, *Phascolarctos cinereus*. We examined specimens from across Australia (including mainland Australia and Tasmania, when present;, Tables A and B) to ensure that taxa from a wide range of environments (that include a broad range of temperatures and evapotranspiration conditions, precipitation/relative humidity) would be sampled. We also examined mesowear across the cheek toothrow to understand how and/or if the inclusion of wear from multiple cusps affected dietary interpretations; however, unerupted/unworn molars were not included in this study. As conditions become more evaporative (i.e., higher temperature and/or lower precipitation/relative humidity), vegetation may become tougher and/or dust on vegetation may increase. Therefore, we test the ability of mesowear to recover such ecological signals through the following hypotheses: 1. Koalas and kangaroos can be distinguished from one another (as browsers and mixed feeders, respectively) via dental mesowear of mandibular teeth. 2. Specimens from more arid regions (lower annual precipitation and/or relative humidity) exhibit more abrasive mesowear (blunter and lower relief teeth) as compared to conspecifics from wetter regions. 3. Specimens from warm regions (higher mean minimum or maximum temperatures) exhibit less abrasive mesowear in koalas (due to increased soft-foliage consumption during a longer growing season), while kangaroos exhibit higher mesowear scores (blunter and lower relief teeth from increased grazing in warmer tropical environments with increased C₄ grass presence) than conspecifics in cooler regions. # Materials and methods The specimens analyzed in this study consisted of 71 *Macropus fuliginosis* specimens, 96 *Macropus giganteus* specimens, and 46 *Phascolarctos cinereus* specimens from across Australia (Tables A-B in) from the Australian Museum (AM, Sydney, NSW, Australia), the Australia National Wildlife Collection (ANWC, Canberra, ACT, Australia), the Museum Victoria (MV, Melbourne, VIC, Australia), the Queensland Museum (QM, South Brisbane, QLD, Australia), the Western Australian Museum (WAM, Perth, WA, Australia), and collections at Flinders University (Adelaide, SA, Australia; individuals sampled for dental microwear by Ref.). Each specimen with a mandible was analyzed from either a photograph (buccal profile) of the toothrow or an epoxy resin cast of the teeth. Mandibles were selected as mandibles and mandibular teeth are typically more common in Pleistocene localities in Australia than maxillas and maxillary teeth. We used the mesowear attributes most commonly used in the literature, scoring both shape and relief and a combined score. Cusp shape was scored as sharp (1), round (2), and blunt (3). Relief was scored on a slightly different scale from high (1), to medium (2), to low (3). While prior studies do not use a medium relief category, we found it helpful when intermediate relief was observed. Additionally, we used a combined score that totaled both scores and subtracted 2 from the resulting score ((Shape + Relief) - 2) = Combined Score), so as to result in a scale of 0–4 (much like Ref., where 0 is equivalent to a tooth with sharp cusps and high relief while 4 is equal to a blunt tooth with low relief). We modified this score slightly from Ref. (and others who have used scores ranging from 0 to 3/4/6) so that teeth with medium relief and sharp cusps and high relief teeth with round cusps were scored the same (with both conditions indicative of similar degrees of mesowear). In addition to scoring the sharpest cusp, per Ref., we also scored all teeth in koalas (lower first through fourth molars, eight cusps on four teeth; see Tables C-D) and two lower molars (four cusps) in kangaroos. In *M*. *giganteus* and *M*. *fuliginosus*, the two teeth (and subsequent four cusps) positioned to process the most vegetation (i.e., the teeth in occlusion and not erupting or in the process of being ejected due to molar progression) were scored for mesowear analysis. Specimens with a lower m1 in occlusion and/or that had already been in occlusion (e.g., potentially worn and/or ejected) were examined; thus, the youngest individuals of all species were excluded. Each specimen was scored by six individuals consisting of undergraduate and graduate students; all had moderate to limited prior experience scoring mesowear cusps (note, Ref. demonstrated the ability of novices to learn mesowear with a brief training session). Median scores were calculated for each specimen, minimizing the effect of the number of mesowear scorers while reducing mesowear observer variability (per Ref., the results of mesowear analysis is improved by incorporating five or more observer scores per specimen). Specimen metadata was gathered from the Atlas of Living Australia, a biodiversity database compiling information about occurrence records and specimens in natural history collections (Atlas of Living Australia). Geographic occurrence data was converted to decimal degrees, if necessary. If geographic data was given via descriptive location data (e.g., 10 km due south of the town of Townsville), it was georeferenced to the finest resolution possible using Google Earth Pro. Mean annual precipitation (MAP, mm) and mean maximum annual temperature (Max. MAT, °C) and mean minimum annual temperature (Min. MAT, °C) data were collected from the Australian Government Bureau of Meteorology weather station, closest to each location that possessed at least a decade of temperature and/or precipitation records (Australian Government Bureau of Meteorology). The maximum distance between a specimen location and its corresponding weather station was less than 125 kilometers; the majority of specimens (87%) were within 50 km of both temperature and precipitation weather stations while 98% were within 100 km of both temperature and precipitation weather stations. When possible, thirty- year averages from 1961–1990 were used. These climate data have the potential to elucidate differences between regions and are more useful to characterizing an animal’s local environment than short term weather events. Relative humidity data were obtained from the NASA Langley Research Center Atmospheric Science Data Center Surface meteorological and Solar Energy (SSE) web portal supported by the United States National Aeronautics and Space Administration LaRC POWER Project. In this dataset, relative humidity is measured at 10 m above the earth’s surface at a 1-degree resolution across the globe. The data are an average of values collected between July 1, 1983 to June 30, 2005 and downloaded on December 10, 2007. Relative humidity data was extracted from the NASA SSE layer using the spatial join tool available in ArcMap 10.4. Non-parametric Spearman’s rank-order correlation coefficients were calculated via XLSTAT to assess if dental mesowear scores were correlated with the climate variables noted above. Further, the koalas and kangaroos were compared using non-parametric Kruskal-Wallis tests to assess if mesowear scores from mandibular teeth showed species differences. # Results ## Diet and dental mesowear Descriptive statistics of dental mesowear scores are noted in and summarized in (all primary data are noted in Tables A-D). Dental mesowear from mandibular teeth of two species of extant kangaroos (*Macropus giganteus* and *Macropus fuliginosus*) have significantly lower relief (higher relief scores), lower combined scores (indicative of blunter and more worn teeth), and blunter shapes than koalas (*Phascolarctos cinereus*) via all mesowear scores analyzed (the sharper cusp and the combined analysis; p\<0.01). *M*. *fuliginosus* has significantly blunter teeth (higher average cusp shape scores using the average of four cusps, see) than *M*. *giganteus* (p = 0.013) despite both being categorized as grazing/mixed-feeding; no other mesowear scores (relief or combined score) are significantly different between these taxa (p\>0.05). ## Climate and dental mesowear Relationships between dental mesowear and climate variables (including correlation coefficients and p-values) are noted in and (all primary data, including information regarding climate stations and corresponding metadata, are noted in Tables A-D). None of these species exhibit significant correlations between any mesowear attribute and precipitation (MAP) or relative humidity (all p-values\>0.1). For koalas, Min. MAT is negatively correlated with the relief and combined scores (both the traditional sharpest cusp method and an alternate method of using the average shape of multiple cusps (4 in kangaroos and 8 in koalas; p\<0.02; ; Tables A-D). Note that for koalas, there are no significant relationships between either mesowear shape score and Min. MAT. Additionally, koala relief scores are negatively correlated with Max. MAT, such that greater Max. MAT results in higher relief (all p\<0.04;). *Macropus giganteus* dental mesowear scores (all six variables) are negatively correlated with Min. MAT (all p\<0.04;), similar to koalas. Further, dental mesowear average cusp shape and the average cusp combined score is negatively correlated with Max. MAT, such that teeth are sharper in regions with higher average maximum temperatures (all p\<0.04). Additionally, no significant relationships were found between any mesowear scoring methods and any of the climate variables here examined (i.e., relative humidity, precipitation, or annual minimum or maximum temperature) in *Macropus fuliginosus* (p\>0.05), despite the specimens examined occurring in states exhibiting significantly different climate variables (based on non-parametric Kruskal-Wallis comparisons of climate variables from specimens of *M*. *fuliginosus* in different states, p≤0.0001; however, the variability in mean annual temperature is less than in the other two taxa). Some relationships between climate variables and *M*. *giganteus* mesowear are stronger using the average of multiple cusps while those in koalas show mixed results (some are slightly weaker while others are slightly stronger, using the average of multiple cusps;). # Discussion Dental mesowear is a valuable method for assessing mammalian diets due to its simplicity and ease of use; however, it is often unclear if and how climatic factors (and subsequent dust/grit loads) affect dental mesowear. Based on the dental mesowear of the mandibular teeth of three ubiquitous taxa with broad geographic ranges that traverse different climatic regimes, one being an obligate browser and the other two eating a mixture of grass and browse, dental mesowear does capture dietary differences. Our first hypothesis, that there is a statistically significant difference between the mandibular mesowear of kangaroos and koalas with disparate diets, is accepted. Our second hypothesis, that specimens from more arid regions exhibited more wear, was rejected; there are no statistically significant relationships between precipitation and/or relative humidity and any of the mesowear scoring methods for any of the three species. For our third hypothesis, koalas living in cooler environments (as defined by lower mean minimum temperatures) had more abrasive wear—as hypothesized. However, *Macropus giganteus* exhibited the same relationship as koalas, contrary to expectations that *M*. *giganteus* consumes more abrasive grasses in warmer regions. *Macropus fuliginosus* had no statistically significant correlations to any of the climate variables, despite significant differences in climate variables for the subset of *M*. *fuliginosus* specimens examined; however, the total temperature range is less than that of the other two taxa. Previous work demonstrated the potential of dental mesowear in marsupial herbivores; however, one of the most ubiquitous kangaroos could not be classified as mixed-feeders based on dental mesowear and were excluded from subsequent analyses with extinct taxa (*M*. *giganteus*). While we do document significant differences between these grazing/mixed-feeding kangaroos and browsing koalas, our data set does not include the broad array of herbivorous marsupials included in the prior study. Further, these differences appear less pronounced as between ungulate taxa which may also be related to the lower time window during which dental mesowear can form on taxa exhibiting molar progression. While Ref. included dozens of species (24–43 species), the sample size of each species was often restricted to specimens from Queensland and/or other states with a state focused collection (and limited to \~10–30 specimens). Here, we trade the diversity of specimens for a more in-depth analysis of fewer ubiquitous taxa, each distributed over a broad range of latitudes and/or longitudes. Fundamentally, diet appears to be recorded via dental mesowear in maxillary teeth, in a subset of their data; while we here demonstrate that dental mesowear is recorded in mandibular teeth in a subset of taxa. Future work should compare maxillary to mandibular teeth. Further, as we here demonstrate, dental mesowear can vary across regions and may demonstrate subtle differences in diet across a species range (although we do not rule out ontogenetic differences, adults were primarily analyzed). Thus, dental mesowear can further be touted as an easy and accessible dietary proxy useful for reconstructing ancient diets of herbivores through time and in response to climate change. With the expansion of digital natural history museum archives, including photographs of both modern and fossil specimens, the use of dental mesowear will likely continue to help clarify diets in a broad range of taxa, including adding marsupial herbivores to the list of focal taxa. None of the mesowear scoring methods for any of the species we examined showed a significant positive or negative relationship with precipitation and/or relative humidity. This is contrary to expectations that regions with lower mean annual precipitation and/or lower relative humidity would have increased grit and/or dust on vegetation resulting in more worn teeth (i.e., lower relief and combined scores)—if grit/dust and not diet are primarily contributing to dental mesowear formation. Instead, there are greater differences in dental mesowear between taxa with different diets than there are between conspecifics occurring in regions experiencing different climatic regimes. These data confirm patterns observed in prior studies, most notably that diet is the primary signal recorded via dental mesowear (i.e., it is grass not grit that is likely being recorded via dental mesowear). Of the climatic variables analyzed, only variations in temperature were shown to affect mesowear. This may be due to changes in dietary resources in *Macropus giganteus* and *Phascolarctos cinereus* in regions experiencing different temperatures. Koala mandibular teeth from warmer regions showed lower mesowear values (higher relief and lower combined scores), consistent with our third hypothesis that mesowear scores in koala mandibular teeth will decrease with increased temperature. As suggested by Ref., this may be due in part to the need for thermoregulation. Koalas exhibit increased panting when temperatures increase to facilitate evaporative cooling, leading to increased water loss. As such, plants with higher water content are preferred in order to regulate body temperature and restore water balance. Assuming leaves with higher water content are also less abrasive, this dietary strategy may lead to reduced wear (and subsequently higher relief and less blunting of the tooth cusps). Similarly, it has also been suggested that when vegetation is abundant, koalas exhibit a preference for new growth, presumably because these leaves are softer, easier to chew, and contain a higher water and/or nutrient content (with dental microwear of koalas also demonstrating the consumption of less abrasive leaves in drier regions. Koalas in cooler regions where growing seasons are shorter, and body sizes and fur depth are larger and deeper, respectively, consume more vegetation due to higher metabolic demands (much like has been documented seasonally, with koalas eating more in the winter than the summer), resulting in increased gross wear—including lower relief (higher values), documented here. Much like koalas but contrary to predictions, *Macropus giganteus* showed significant negative correlations of dental mesowear scores with temperature. This is contrary to what we predicted. Similar to our initial predictions that drier regions contribute to blunter teeth (potentially due to dust/grit), we also expected *M*. *giganteus*, which is known to consume predominantly C₄ grasses in warmer and more tropical (and lower latitude) regions (and per communication with G. Prideaux, the specimens here included from Flinders University consumed primarily C₄ grass based on stomach content analysis and/or stable carbon isotope analysis) to have teeth with lower relief, blunter shapes, and higher combined scores in these regions, in contrast to higher/cooler latitudes. While classified as grazers (≥70% grass) and mixed feeders (\<90% grass) they are both known to vary their diet throughout their geographic range; thus, dietary behavior may be more complex, challenging the use of simplified dietary categories. Further, while both Refs. review the literature and provide consensus views of dietary interpretations, Ref. estimated percent grass consumption based on stable carbon isotopes of bone collagen and tooth enamel in kangaroos. Specifically, they estimated that *M*. *fuliginosus* consumed less than 40% grasses, while *M*. *giganteus* consumed nearly 100% grasses, both based on bone collagen (with enamel estimates of *M*. *fuliginosus* suggesting increased grass consumption, yet still less than 60%). Other studies document highly variable estimates for % grass consumption in *M*. giganteus, ranging from \~50% to between 64–84%. Further, studies of habitat selection in areas where these two species co-occur, suggest that *M*. *giganteus* selects habitats with a larger proportion of grasses and lower mean lateral cover (%), as compared to *M*. *fuliginosus*. Thus, it is rather surprising that *M*. *giganteus* had teeth with sharper cusps and higher relief in areas with a higher incidence of C₄ grass. Further work is needed to better understand how grass consumption affects dental mesowear in kangaroos exhibiting molar progression, building off of the work of Ref.. Specifically, molar progression and rates of tooth wear are decoupled, with gross tooth wear varying in both *M*. *fuliginosus* and *M*. *giganteus* between individuals of the same population at the same time, between individuals within the same region at different times, and between different populations from different regions; yet, little is understood regarding how this affects dental mesowear. Additional work is also needed to assess if the incidence of hard quartz grains varies with latitude along eastern Australia, which may have also affected koalas and kangaroos similarly. Several limitations should be noted in this study. Climate data for each specimen was dependent on distances between the location of specimens and weather stations, some of which included distances of \~100 km (although, this only applied to a subset of specimens, see). Further, while collection data including the year a specimen was collected is often available for museum specimens, many times climate data corresponding with those years is not available (and/or it is unclear under what duration of time the specimen was alive and how many years or months of data should be included with that specimen). For these reasons, we generally characterized a regions climate using 1961–1990 year averages. Further, many of these taxa can have broad home ranges during the life of the individual specimen, and unlike dental microwear which captures the "last supper" (past few days to weeks of an animal’s diet), dental mesowear is cumulative—capturing a multi-season and multi-year dietary average. Experimental studies examining modern populations of kangaroos and/or studies of kangaroos from specific regions that experienced non-drought and pronounced drought conditions could improve our understanding of how diet may vary with climate change, and/or if grit does substantially impact mesowear. Further work is also needed to evaluate if ontogenetic differences influence diet, as inferred from dental mesowear (i.e., molar progression; if mesowear is most telling on teeth with the longest ‘lifespan’). Despite early work which suggests that mesowear may be a function of both diet and grit on the landscape, it appears that diet is the overarching signal recorded in both individual teeth and multiple teeth across the tooth row—at least for the marsupials discussed here. This is consistent with prior work documenting a significant positive relationship between percent grass in an animal’s diet and mesowear score (0–4) while no relationships existed between mean annual precipitation and mesowear scores. While more work is needed to better determine the efficacy and/or benefits of scoring multiple teeth (instead of or in addition to the sharpest cusp), average shape scores across the entire tooth row do differentiate these two mixed feeding kangaroos from one another, which may be eating slightly different foods (either on average or in different regions) from one another—showing that nuanced differences can be garnered in certain cases using the cumulative method. In addition to gaining valuable insights into mesowear variability in different species, we document the absence of any relationship between dental mesowear attributes and aridity (either precipitation or relative humidity). Further, relationships between dental mesowear and temperature are likely related to dietary differences across regions, although further work is needed to experimentally test these relationships. # Supporting information We thank the curators and collection managers affiliated with the following museums for access to and assistance in the collections: the Australian Museum (S. Ingleby), the Australia National Wildlife Collection (A. Drew, L. Joseph, and R. Palmer), the Museum Victoria (L. Cook, K. Roberts, and K. Rowe), the Queensland Museum (H. Janetzki), the Western Australia Museum (K. Travouillon), and collections at Flinders University (G. Prideaux; we are also grateful to G. Prideaux for providing dietary data, based on stable isotopes and/or stomach content analysis, associated with Flinders University specimens here included). We also thank K. Butler, M. Clauss, J. Saarinen, and G. Semprebon for helpful comments on earlier versions of this manuscript. [^1]: The authors have declared that no competing interests exist.

# Introduction 22q11.2 Deletion Syndrome (22q11DS, OMIM \#188400), also commonly known as DiGeorge Syndrome or Velo-Cardio-Facial Syndrome, is a genetic disorder that results from an approximately 1.5-3Mb congenital multigene deletion on the long arm of chromosome 22, which includes the gene for T-Box Transcription factor 1 (*TBX1*). 22q11DS occurs in 1:4000 live births, making it the most common interstitial deletion syndrome and the second most common chromosomal abnormality after Down's Syndrome. Typical physical findings in 22q11DS patients include defects in cardiovascular, thymic, parathyroid and craniofacial structures derived from the pharyngeal arches and pouches. In addition, 22q11DS is associated with high frequencies (80–100%) of neurocognitive disabilities, and it is one of few cytogenetic abnormalities that occurs in tandem with a psychiatric disease. The syndrome is one of the highest known risk factors for schizophrenia, as 25-30% of 22q11DS patients develop schizophrenia during adolescence or adulthood. 22q11DS is also a risk factor for development of otitis media (OM). OM is inflammation of the middle ear cavity (MEC), often presenting with pain and fever. It is the most common disease in young children worldwide, occurring at least once before the age of two in 90% of infants in the developed world. OM is typically classified as either acute or chronic. Acute otitis media is associated with a bacterial infection and often resolves spontaneously within three months. However, in some cases acute OM is followed by otitis media with effusion (OME) that can become chronic. OME is characterized by excessive effusion that accumulates in the MEC, and the absence of obvious signs of acute infection. Excessive effusion often leads to conductive hearing loss, which in severe cases can become permanent due to erosion of the middle ear ossicles. Even in less severe cases, conductive hearing loss due to OME can interfere with speech and language development. The usual treatment and also the most common operation in the United Kingdom is the insertion of grommets into the tympanic membrane to permit ventilation and drainage of exudates from the MEC. Risk factors for OM include infection, altered immune status, exposure to tobacco smoke, and anatomical defects such as cleft palate. In addition, although the pathogenesis of OM is multifactorial, a role for genetic predisposition is increasingly recognized. In 22q11DS, studies have shown that a majority of patients have a history of chronic or recurrent OM. The hemizygous Df1-knockout mouse (*Df1/+*) was genetically engineered to be a model for human 22q11DS; it carries a multigene deletion in a region of mouse chromosome 16 that is orthologous to the 22q11.2 region in humans. Although the region is highly conserved, several ancestral rearrangements have led to changes in gene order, and so the deletion in *Df1/+* mice encompasses 18 of the protein-encoding genes deleted in human 22q11DS. *Df1/+* mice have proven to be an excellent model for major developmental defects in human 22q11DS such as cardiovascular abnormalities and thymic or parathyroid defects, although no gross craniofacial abnormalities such as cleft palate have been reported. Furthermore, both *Df1/+* mice and other mouse models of 22q11DS have been found to show cognitive and behavioural abnormalities associated with human 22q11DS and schizophrenia, including reduced auditory sensorimotor gating. Modern tests of sensorimotor gating depend on the ability to hear, and previous studies have presented some evidence for normal hearing in *Df1/+* mice and similar mouse models. However, mice heterozygous for *Tbx1*, one of the genes involved in the multigene deletion and the most likely candidate gene responsible for the pharyngeal arch-derived defects in 22q11DS, have been shown to suffer frequent middle ear inflammation with associated conductive hearing loss. Here, we aimed to resolve this discrepancy in the literature, using auditory brainstem response (ABR) measurement to assess hearing capability in adult *Df1/+* mice and their WT littermates. To obtain data from a large population of age-matched *Df1/+* and WT mice, we focused on measurement of click-evoked ABR thresholds, a simple and rapid electroencephalographic measure of peripheral and early central auditory activity that could be obtained *in vivo* from each ear for all animals in a litter in a single day. We found that click-evoked ABR thresholds were significantly elevated in 48% of the *Df1/+* animals, often in only one ear. Anatomical and histological analysis of the middle ear revealed a high incidence of OME in *Df1/+* mice, which correlated directly with elevated ABR thresholds. We conclude that *Df1/+* mice, like human 22q11DS patients, are susceptible to otitis media and conductive hearing loss. These results suggest that studies of abnormal auditory sensorimotor gating in *Df1/+* mice need to be revisited using more sensitive assays for hearing loss, and also that *Df1/+* mice are a potentially powerful animal model for studying the genetic and environmental causes of otitis media. # Results ## Elevated ABR thresholds in both male and female *Df1/*+ mice The auditory brainstem response is an electroencephalographic signal arising from sound-evoked activity in neuronal circuits of the ascending central auditory pathway. ABRs evoked by click stimuli were recorded in 44 *Df1/+* mice (24 male, 20 female) and 43 WT littermates (24 male, 19 female), ranging in age from 8 to 40 weeks old. Measurements were taken once in each animal in either one or both ears, under free-field conditions with an ear plug in the opposite ear. Both left and right ears were tested in 31 of the *Df1/+* and 23 of the WT animals, and one ear only in 13 *Df1/+* and 20 WT mice. The ABR database therefore consisted of a total of 75 *Df1/+* and 66 WT ABR recordings. Click ABR thresholds were determined for each recording, and judged to be the lowest click intensity at which characteristic peaks of the ABR waveform could be observed. Click ABR thresholds were significantly higher on average, and also more variable, in both male and female *Df1/+* mice than in their gender-matched WT littermates. Median thresholds (and total ranges) were 35 (25-50) and 37.5 (30-55) dB SPL for male and female WT animals, respectively, but 50 (30-75) and 50 (35-85) for male and female *Df1/+* mice from the same litters. Median thresholds therefore differed significantly between recordings from *Df1/+* and WT mice of the same gender (Wilcoxon Mann-Whitney test, *Df1/+* versus WT: p=6x10^-6 males, p=9x10^-7 females), but not between males and females of the same genotype (Wilcoxon Mann-Whitney test, males versus females: p=0.3 *Df1/+*, p=0.5 WT). Similar results were obtained when recordings from left or right ears were considered separately. ## ABR threshold distribution in *Df1/*+ mice appears bimodal Since there were no significant differences in click ABR thresholds recorded from male and female animals of the same genotype, we pooled data from male and female mice to examine genotype differences in the threshold distributions more closely. The distribution of click ABR thresholds recorded from *Df1/+* mice was significantly different from the distribution recorded from WT mice (Kolmogorov- Smirnov test, p=5x10^-8), even when the two distributions were normalised to align the medians (Kolmogorov-Smirnov test on median-normalised data, p=0.02). In fact, the *Df1/+* threshold distribution appeared bimodal, suggesting that ABR deficits were perhaps restricted to a subset of *Df1/+* animals. To be conservative, we defined a click ABR deficit to be present when the ABR threshold exceeded 55 dB SPL (criterion threshold indicated by dashed lines), since 55 dB SPL was the highest threshold observed in recordings from WT mice. By definition, none of the ABR thresholds recorded in WT mice exceeded this criterion; however, 38% of thresholds recorded from the ears of *Df1/+* mice did. Since elevated ABR thresholds could occur either in only one ear or in both ears, the percentage of affected animals could in principle differ from the percentage of affected ears. We investigated this issue with further analysis of ABR data from the subset of animals for which both left and right ear ABR recordings had been obtained. ## ABR deficit in *Df1/*+ mice can be either monaural or binaural To determine whether click ABR thresholds tended to be similar in the two ears, we compared left and right ABR thresholds for the 31 *Df1/+* and 23 WT animals for which ABR recordings had been collected from both ears. The correlation between left and right ear ABR thresholds was lower in *Df1/+* than WT mice (Pearson's r=0.49, p=0.006 for *Df1/+* mice; r=0.68, p=0.0004 for WT mice), but not significantly so (Fisher transformation test for difference in correlation coefficients). Thus left and right ear ABR thresholds were correlated in both groups of animals. However, among the *Df1/+* mice, 16 (52%) had no significant click ABR deficit in either ear, 9 (29%) had a significant deficit in one ear, and 6 (19%) had deficits in both ears. Thus, 48% of the *Df1/+* mice had a click ABR deficit in at least one of the two ears, and the deficit was monolateral in 60% of those affected animals. This finding is suggestive of a conductive origin for the hearing loss, because most causes of sensorineural hearing loss would be expected to affect both ears. ## ABR deficit in adult *Df1/*+ mice shows no age dependence The C57BL/6 background strain from which *Df1/+* mice and their WT littermates are derived is known to have age-related sensorineural hearing loss, especially at high sound frequencies. Mutations can accelerate age-related hearing loss, so we wondered if ABR deficits in *Df1/+* mice might worsen with age in adulthood. However, we found no significant dependence of click ABR thresholds on age in adulthood, for either *Df1/+* or WT mice. Similar results were obtained whether the analysis was performed on all recorded ABR thresholds as shown in, or on thresholds from left ears or right ears separately. These results demonstrate that click ABR deficits in *Df1/+* animals cannot be explained by aging-related effects, suggesting again that these deficits are likely to be primarily conductive in origin. However, since WT C57BL/6 animals would be expected to have age-related hearing loss for high sound frequencies, the negative results for WT animals also indicate that ABR thresholds for a broadband click stimulus are not a sufficiently sensitive assay to evaluate the possibility of high- frequency sensorineural hearing loss in addition to conductive loss (see Discussion). ## Adult *Df1/*+ mice have a high incidence of OM To determine whether *Df1/+* mice have middle ear problems, we first examined middle ear anatomy and histology in 9 *Df1/+* mice and their WT littermates. The 9 *Df1/+* mice included 6 animals with confirmed ABR deficits and 3 mice that had not undergone ABR testing; 2 of these 3 mice had a negative Preyer reflex. MicroCT scans revealed no gross abnormalities in middle ear anatomy in the *Df1/+* mice compared to their WT littermates, and there were no defects in the ossicular chain. However, histological analysis demonstrated a high incidence of OME in the *Df1/+* animals, with frequent signs of inflammation such as effusion, capillary hyperplasia, a thickened tympanic membrane and thickened MEC mucosa. Affected animals had effusion in one or both ears and the effusion content varied with respect to quantity of infiltrated and inflammatory cells. In some *Df1/+* mice with severe OME, the Eustachian tube (ET) was infiltrated by inflammatory cells. Examination of the mucociliary integrity in *Df1/+* mice with severe OM revealed increased mucus production within the middle ear adjacent to the orifice where the ET enters the MEC, suggesting increased goblet cell density ( compared to C). In *Df1/+* mice displaying mild OM, however, this increased mucus secretion was only observed occasionally (data not shown). ## Morphological changes in the MEC mucosa in *Df1/*+ mice with OM To further investigate OM-associated changes in the pseudostratified mucociliary epithelium lining the MEC, we turned to scanning electron microscopy (SEM). Comparing the density and distribution of cilia next to the opening of the ET, we observed that WT and *Df1/+* mice without infiltrated cells in the MEC displayed a thick lawn of evenly distributed cilia adjacent to a border region of unciliated epithelium. In *Df1/+* mice with OME, however, cilia density was reduced and the cilia were rarefied and shortened; in addition, the MEC epithelium was swollen and partly covered in exudate. ## Elevation of ABR thresholds correlates with the severity of OM in *Df1/*+ mice To determine whether OM could account for click ABR deficits in Df1/+ mice, we performed histological analysis of the MEC on a set of adult littermates (5 Df1/+ and 1 WT, age 8 weeks) for which click ABRs had been recorded in both ears. The WT mouse and 1 Df1/+ mouse had normal ABR thresholds in both ears; 1 Df1/+ mouse had a slightly elevated ABR threshold in one ear and a normal threshold in the other; 2 Df1/+ mice had a significantly elevated ABR threshold in one ear and a normal or only slightly elevated threshold in the other ear; and 1 Df1/+ mouse had significantly elevated thresholds in both ears. Data from these 6 littermates (12 ears) therefore provided a perfect opportunity to test for a correlation between elevation of click ABR thresholds and signs of OM in Df1/+ animals. Two measures of the severity of OM were used: presence of effusion, and increased thickness of the middle ear mucosa. Analysis was performed blind to genotype and was repeated by multiple operators to ensure reliability of classifications. As shown in and, Df1/+ ears for which click ABR thresholds were most elevated (\>70 dB SPL) had effusion with more than 50% of infiltrated cells within the MEC. These mice also displayed a severe thickening of the ME mucosa, which could have produced physical obstruction of movement of the ossicles. In some cases the epithelial thickness was observed to be up to 23 times higher than in WT littermates. Df1/+ ears for which ABR recordings had revealed marginally lower thresholds (60-70 dB SPL) displayed effusion with fewer infiltrated cells and a less severe thickening of the mucosa, indicating a less advanced inflammation. In these animals there was only limited tissue around the ossicles. Df1/+ ears with normal or slightly elevated ABR thresholds (40-45 or 50-55 dB SPL) showed either no effusion or a serous effusion only with no or mild thickening of the mucosa ( or F). Thus both the severity of effusion and the severity of mucosa thickening were significantly correlated with elevation of click ABR thresholds across the Df1/+ ears (Spearman's correlation test: rho=0.88, p=0.0007 for severity of effusion; rho=0.75, p=0.013 for severity of mucosa thickening). The two ears from the WT animal (2\*), for which both ABR thresholds were normal, had no effusion and normal ME mucosa. In addition to these 6 littermates, we also examined 5 more animals which also underwent hearing tests, and we found the same correlation between hearing loss and OM in those additional animals. No evidence of ossicle erosion was observed, but such erosion might only be present in older mice after repeated bouts of OM. ## Further observations ### Bacteriology Bacteriological analysis of middle ear swabs obtained from both ears of 4 *Df1/+* mice revealed, in 4 out of the 8 ears, the presence of commensal bacteria and opportunistic pathogens that do not normally cause infections in a healthy ear. One mouse had scant growth of *Escherichia coli* in both ears; another had moderate growth of *Lactococcus lactis* *spp* *lactis* in one ear; a third was found to have scant growth of *Pantoea* *spp* in one ear. No fungi or yeast were isolated from any of the samples. These results suggest that OM in *Df1/+* mice is unlikely to be caused by unusual susceptibility to a specific pathogen; rather, any bacterial infection of the middle ear in *Df1/+* mice is probably a secondary opportunistic process. ### Hair cell density in the organ of Corti To address the possibility that elevated ABR thresholds in *Df1/+* mice might arise from sensorineural as well as conductive hearing loss, we examined the sensory epithelium of the inner ear in 6 *Df1/+* and 5 WT mice. There was no evidence for significant loss of hair cells sufficient to account for the elevated ABR thresholds in *Df1/+* relative to WT mice. Moreover, the density of hair cells in the cochlea appeared normal in both ears of *Df1/+* animals with monolateral hearing loss. On the basis of our analysis, we cannot entirely rule out the possibility of subtle inner ear abnormalities in *Df1/+* mice, especially given that severe OM can sometimes affect the inner ear. However, we can conclude that the observed click ABR deficit in *Df1/+* mice does not arise from hair cell loss. # Discussion Here we have shown that *Df1/+* mice are susceptible to conductive hearing loss and otitis media, which are also common consequences of the human 22q11.2 deletion that the mice were genetically engineered to model. Mouse models of 22q11DS such as the *Df1/+* mouse have attracted great interest not only as a tool for investigating the origins of various defects and disabilities associated with this relatively common chromosomal disorder, but also as a means of gaining insight into the pathogenesis of schizophrenia, for which 22q11DS is one of the most significant known risk factors. In the context of this previous research, the present study makes three distinct contributions. First, the results suggest a resolution to a discrepancy in the literature between previous studies reporting normal hearing in the *Df1/+* and *Df(16*)*A/+* mouse models of 22q11DS and studies documenting a high incidence of middle ear disease in mice heterozygous for the gene *Tbx1*, which is included in the *Df1/+* and *Df(16*)*A/+* deletion regions. The previous evidence for normal hearing in mouse models of 22q11DS came from supplementary controls in studies of prepulse inhibition (PPI) of acoustic startle. In one study, thresholds for acoustic startle were observed to be slightly lower, and startle amplitudes slightly higher, in *Df1/+* mice than in their WT littermates; in another study, similar results were reported for *Df(16*)*A/+* mice, which have a larger deletion region than *Df1/+* mice. Both studies therefore concluded that there was no evidence for reduction in hearing sensitivity in these mice. However, under some circumstances, mice with partial hearing loss can show reduced startle thresholds and elevated startle amplitudes in acoustic startle testing, perhaps because central auditory adaptation to the reduction in peripheral input leads to hyperacusis for loud sounds. Better evidence for normal hearing in *Df1/+* mice was provided in based on frequency- specific distortion-product otoacoustic emission (DPOAE) testing, which is generally an excellent means of detecting peripheral auditory abnormalities including middle ear problems. However, the DPOAE testing in that study was performed on only 6 *Df1/+* mice, and apparently only on one ear in each animal. Given the intermittent nature of the ABR deficit and OM observed in the present study (48% of *Df1/+* animals, but only 38% of *Df1/+* ears tested), it is possible that this sample size was too small. We therefore suggest that previous evidence for normal hearing in *Df1/+* and *Df(16*)*A/+* mice was inconclusive, and that all mouse models of 22q11DS involving deletion of *Tbx1* may have a high incidence of conductive hearing loss. The second contribution of the present work is to show that previous reports of abnormal auditory sensorimotor gating in mouse models of 22q11DS need to be re- examined to determine whether abnormalities in auditory processing alone might account for the results. Impaired auditory sensorimotor gating, quantified as a reduction in PPI of acoustic startle, is considered an important endophenotype for risk of schizophrenia. Reduced PPI of acoustic startle has been reported both in human 22q11DS patients and in mouse models of 22q11DS. However, as discussed above, the high incidence of conductive hearing loss and otitis media in *Df1/+* mice could itself lead to differences between *Df1/+* and WT animals in PPI of acoustic startle. Moreover, although we found no abnormalities in hair-cell density that could account for the observed elevation of click ABR thresholds in *Df1/+* mice, it is possible that *Df1/+* mice have more subtle sensorineural hearing deficits, especially given that *Tbx1* is involved in inner ear development. The possibility of such abnormalities could be explored further in the future with tone-evoked ABR measurements or distortion-product otoacoustic emission (DPOAE) testing, which are more sensitive assays for frequency-specific hearing loss than click-evoked ABR measurements. The final contribution of this work is to introduce the *Df1/+* mouse as a powerful model system for investigating the pathogenesis of OM in 22q11DS and also more generally. Hearing loss is prevalent among human 22q11DS patients, and is almost always related to middle ear infection. In one study, chronic or recurrent otitis media was reported in 52% of 22q11DS patients; in another 88% had otitis media, 53% had conductive hearing loss, and 39% required surgical implantation of ventilation tubes to drain the middle ear. In these patients the causes of the OM have been proposed to be multifactorial involving immune deficiency, palatal abnormalities and Eustachian tube dysfunction. The high incidence of conductive hearing loss and otitis media in *Df1/+* mice indicates that these animals can be used to tease apart the causes of frequent OM in 22q11DS patients. Moreover, the prevalence of monolateral ABR deficits and OM in *Df1/+* mice creates opportunities for within-animal controls, making the animals a potentially powerful tool for testing hypotheses about the causes of OM. Studies of the mechanism by which genetic predispositions cause OM have been performed in other mouse models, such as the heterozygous *Fbxo11* mouse and *Eya4* knockout mouse. *Fbxo11* is expressed in the lining of the middle ear cavity and has been proposed to affect epithelial inflammatory events in the ear. In contrast, in the *Eya4* mouse the morphology of the Eustachian tube and angle of connection to the middle ear have been shown to be defective, potentially causing ear clearance problems. *Tbx1* has been shown to be expressed in the endoderm of the developing pharyngeal arches and *Tbx1* null mice have a hypoplastic pharynx. As the Eustachian tube arises from the endodermally derived first pharyngeal pouch it is therefore tempting to speculate that a subtle early defect in patterning of the endoderm might be responsible for the high incidence of OM in *Df1/+* and *Tbx1* heterozygous mice. Our ongoing efforts to pinpoint the causes of otitis media in *Df1/+* mice are therefore focusing on the possibility of abnormalities in the morphology of the Eustachian tube and/or the other endodermally derived tissues of the middle ear. In conclusion: *Df1/+* mice, like human 22q11DS patients, are susceptible to otitis media and conductive hearing loss, which affect nearly half the animals but often in only one ear. The findings suggest that abnormal auditory sensorimotor gating previously reported in mouse models of 22q11DS could arise from abnormalities in auditory processing. More broadly, the results indicate that *Df1/+* mice are an important model system for investigating the causes of OM in both 22q11DS patients and the many children worldwide who suffer from chronic middle ear infections. # Materials and Methods ## Ethics statement All *in vivo* experiments were conducted in accordance with the United Kingdom's Animal (Scientific Procedures) Act of 1986, under a project licence approved by the UK Home Office. ## Animals The *Df1/+* mouse line had been maintained on a C57BL/6 background for a minimum of 10 generations prior to the analyses. The *Df1* deletion itself was engineered on a 129S5 SvEvBrd genetic background. ## Hearing tests ### Auditory brainstem response measurement ABR testing was performed in a sound isolation booth (Industrial Acoustics Company, Inc.). Mice were anaesthetised with ketamine and medetomidine. Body temperature was maintained at 37-38°C using a homeothermic blanket (Harvard Apparatus). Subdermal needle electrodes (Rochester Medical) were inserted under the skin with positive electrode at the vertex, negative electrode near the ear being tested, and ground electrode near the opposite ear (which was blocked with a sound-attenuating earplug). For most animals, ABR recordings were obtained from both the left and right ears in turn, with an earplug in the ear opposite to that under test (to ensure monaural stimulus presentation). Auditory stimuli were presented via a free-field speaker (FF1) from Tucker-Davis Technologies (TDT). Speaker output was calibrated before each set of experiments using a Bruel & Kjaer ¼ inch microphone (4939), placed at the location of the ear to be tested. Stimulus generation and data acquisition were accomplished using hardware from TDT (RX6 and RX5 signal processors, RA4LI and RA16SD signal amplifiers, PA5 attenuator, and SA1 speaker amplifier), a custom low-pass filter designed to remove attenuation switching transients, and software from TDT (Brainware) and Mathworks (Matlab). Stimuli were 50 μs monophasic clicks ranging in sound level from 0 to 90 dB SPL, presented at a rate of 20 clicks/sec. ABR recordings typically included 500 repeats of click stimuli presented over a 20-80 dB SPL range at 20 dB intensity increments, followed by 1000 repeats of click stimuli presented over a smaller intensity range at 5 dB intensity increments. The threshold was defined to be the lowest click intensity evoking a clear and characteristic deflection of the ABR wave that was at least as large as the time-dependent standard error in the mean wave at that sound intensity. ### Preyer reflex assessment A few of the mice used for studies of middle ear anatomy and histology did not undergo ABR measurement due to time constraints, and instead were tested for a Preyer reflex. The Preyer reflex is a flick of the pinnae evoked by a transient loud sound. To present such sounds, we used a custom-built click box (MRC Institute of Hearing Research, Nottingham, UK) emitting a brief 18.5 kHz tone burst with intensity 95-105 dB SPL at the distances typically used for testing. Since the Preyer reflex is somewhat unreliable even in animals with normal hearing, we judged the Preyer reflex to be negative only if an animal showed no Preyer reflex across several presentations of the sound. ## Middle ear analyses ### Histology Mouse heads were fixed in 4% paraformaldehyde (PFA) overnight at 4°C and then decalcified in EDTA Solution (67.5% EDTA, 7.5% PBS and 25% PFA (4%)). The tissue was then dehydrated through a methanol series and isopropanol and cleared in tetrahydronaphthalene before embedding in paraffin wax. The 9 μm frontal sections were mounted on Superfrost Plus Slides, dewaxed in Histoclear, rehydrated through IMS, stained with 1% Alcian blue in 3% acetic acid pH2.5, Ehrlich’s haematoxylin, and 0.5% Sirius Red in saturated picric acid, and then mounted in DPX. Slides were imaged on a Nikon Digital Sight Camera. Measurement of mucosa thickness was performed using Image J software. ### Scanning electron microscopy The temporal bones of adult mice were dissected and the middle ear mucosa revealed by removing the outer ear, eardrum, tympanic ring and the malleus and incus. These were then fixed in 2.5% gluteraldehyde in 0.15M cacodylate buffer (pH7.2) overnight at 4°C, and washed and postfixed in 1% osmium tetroxide. Next specimens were dehydrated through an ethanol series and dried using a Polaron E3000 critical point dryer. After mounting and coating with gold (Emitech K550X sputter coater), the surface of the mucosa was examined and images recorded using a Hitachi S-3500N scanning electron microscope (SEM) operated at 10kV in high vacuum mode. ### MicroCT reconstruction Micro computerized tomography (microCT) was used for the three-dimensional analysis of *Df1/+* and WT middle ear morphology. ## Inner ear analysis Cochleae were fixed in 4% PFA in PBS for 2 hrs, then decalcified in 4% EDTA in PBS at pH7.4 for 48 hrs at 4°C. The organ of Corti was extracted in half-turn segments, permeabilised in 0.5% Triton X-100 for 15 min, and incubated with fluorescently conjugated phalloidin at 1 μg/ml for 2 hrs. Phalloidin labels filamentous actin and therefore delineates both hair cell stereocilia and intercellular junctions at the luminal surface of the organ of Corti. Segments were mounted on slides using an anti-photobleaching agent (Vectashield) that also contained DAPI (VectaLabs) to label cell nuclei. Slides were then examined and images taken with a confocal microscope (Zeiss) and viewed for analysis through LSM browser (Zeiss). # Supporting Information We thank Professor Alice Warley and Dr Gema Vizcay-Barrena (Centre for Ultrastructural Imaging, King's College London) for help with the SEM work; Dr Jean-Philippe Mocho (Royal Veterinary College) for assistance with bacteriological analysis; and Ms Jenifer Suntharalingham for help with genotyping. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JFL AST. Performed the experiments: JFL JCF AST FAZ RRT AF. Analyzed the data: JCF JFL AST FAZ RRT AF. Contributed reagents/materials/analysis tools: PJS SI. Wrote the manuscript: JFL JCF FAZ AST.

# Introduction Dogs in two-choice experiments, when selecting between two dishes with snacks placed in front of them, 90° apart, left and right, prefer to turn either clockwise (“right-preferring”) or counterclockwise (“left-preferring”) or randomly in either direction (“irresolute”). This turning preference (or non- preference) is individually consistent in all trials but it is biased in favor of north if they choose between dishes positioned north and east or north and west, a phenomenon we denoted as “pull of the north”. This phenomenon was particularly pronounced in older dogs, females, smaller and medium-sized breeds, dogs exhibiting a turning preference, and especially in the north-east choice. We suggested that “pull of the north” represents a further indication of magnetoreception in dogs, the other being non-random directional alignment during marking, which was, however, significantly changed when exposed to bar magnets, the ability to find a bar magnet, or the existence of the so-called "compass run" exhibited during homing. We are, however, aware that for the ultimate evidence of magnetoreception, experiments in defined manipulated magnetic field and/or under conditions of disturbed magnetoreception are necessary. Moreover, the proximate reason for “pull of the north” remains unclear and should be at least hypothesized. Laterality, i.e. a predictable, non-random preference for using one side of the body (limbs, brain hemisphere, sensory organs) spontaneously or if forced or restricted to choose between two sides, is a known phenomenon in humans and animals. Laterality may be inborn, imprinted, or entrained and has to be taken into account in maze and behavioral two-choice animal experiments. Laterality in dogs has been intensively studied with regard to the motoric (efferent) aspect (paw laterality, Kong-test:; sensory (afferent) aspect; cognitive, and emotional aspects. Interestingly, and contrary to studies in humans, turning (directional, rotational) preference has remained understudied. Most people are right-handed, yet tend to instinctively veer to the left upon entering a new space. Interestingly, the counterclockwise action goes also for most athletic tracks, horse and car races, and for baseball players running the bases. There is even evidence that the chariot races at ancient Rome's Circus Maximus ran counterclockwise, too. So, in sports, where competitors enter the field of play from the outside of a traced circle, a right-directional choice would lead to a counter-clockwise motion. But when entering the field of action from within the circle—walking out of your apartment to take the dog for a walk, and encountering intersections—right directional choices would tend towards tracing a clockwise path. Interestingly, in the countries, where people drive on the left side of the road, retail shoppers tend to turn counterclockwise—when navigating store aisles, while in the countries, where people drive and keep on sidewalks right, veer clockwise. Tendencies of people to turn either direction are known to architects who use them to design shopping galleries to funnel shoppers in the wished direction. While the preference to turn in a certain direction can be explained by individual inborn laterality (handedness) and experience (facilitation), or–e.g. in the context of our experiment of choice between two dishes, which is a visually guided task, through visual laterality—the “pull of north” is expected to have a magnetoreceptive ground. Examination of this phenomenon has a heuristic potential in getting insight into the very seat and mechanism of magnetoreception, which still remain enigmatic. Sensory laterality (or asymmetry) has been described also in the context of spatial orientation in general and magnetoreception in particular. It has been found that homing pigeons rely more on the right olfactory system in processing the olfactory information needed for the operation of the navigational map. An earlier study has shown that the magnetic compass of a migratory bird, the European robin (*Erithacus rubecula*), was lateralized in favour of the right eye/left brain hemisphere. However, it has been later demonstrated that the described lateralization is not present from the beginning, but develops only as the birds grow older. In another study, it was shown that pigeons can perceive and process magnetic compass directions with the right eye and left brain hemisphere as well as the left eye and right brain hemisphere. However, while the right brain hemisphere tended to confuse the learned direction with its opposite (axial response), the left brain hemisphere specifically preferred the correct direction (angular response). The findings thus demonstrated bilateral processing of magnetic information, but also suggested qualitative differences in how the left and the right brain deal with magnetic cues. Based on the hitherto knowledge and the above arguments, 1. We hypothesize that if “pull of the north” is due to magnetoreception (and indeed no other explanation is apparent), it should be demonstrated also in an artificial magnetic field shifted by magnetic coils, i.e. the artificially shifted magnetic North should exert the same effect as the natural geomagnetic North. 2. We expect that, consistently with results of the previous study “pull of the north” is more pronounced in “lateralized” dogs and more in the North- East (N-E) combination than in the North-West (N-W) choice. Furthermore, following questions can be raised (and should be tested) to get insight into the nature of the turning preference: 3. Does the directional preference for turning correlate with motoric laterality (such as paw-laterality, i.e. “handedness”)? 4. Is pull of the north a) symmetrical (bilateral, i.e. of the same strength in the clockwise as in counterclockwise direction), or b) asymmetrical (unilateral, i.e. stronger in one particular direction)? # Material and methods ## Ethics statement The study did not involve any disturbance or discomfort to the study subjects. The Professional Ethics Commission of the Czech University of Life Sciences in Prague has decided that according to the law and national and international rules, this study has not a character of an animal experiment and does not require a special permit. ## Subjects Altogether, 23 domestic dogs *Canis familiaris* (11 M, 12 F) from six breeds with pedigree and an average age of 4.8 (± 2.8) years were used in this study. The dogs were pets living in households. All the dog owners were present with their dogs at trials. ## Experimental equipment The experiment took place in a magnetic coil at the field research station Truba, Kostelec nad Černými lesy, (N 50°0.40480', E 14°50.11145'), a detached workplace of the Faculty of Forestry and Wood Sciences, Czech University of Life Sciences in Prague, Czech Republic. The magnetic coil (a Merritt coil, built according Kirschvink) was 4 x 4 x 4 m and was located in a separate special building. It was shielded from radiofrequency waves. It was controlled from a separate building next to the coil building. The magnetic field in coils was manipulated by a MagFieldG control software through a GMP4 RJ4.01 control unit and three current amplifiers, each for the Bx axis, the By axis and the Bz axis. The generation system for GMP4 3D coil system was used to create a defined direct and slowly changing magnetic field and it served to drive the coil system to create a defined magnetic field. Magnetic induction values in the Cartesian coordinate system (axis Bx = -3225 nT; axis By = 17800 nT; axis Bz = 45448 nT) were set for the experiment, thereby rotating the magnetic field by 90° magnetic North was shifted to the topographic (= geomagnetic) East. The magnetic field strength and inclination were maintained as for geomagnetic values for local geographic conditions. The magnetic coil space was used also for the control experiment to test the dogs under local geomagnetic conditions, while other experimental conditions were preserved identic, i.e. shielding of radiofrequency waves, avoiding other influences (wind, sun, outside sounds). The coil room was equipped with cameras (AXIS P5624-E 50HZ—PTZ IP camera, TD / N, 18x zoom, HD 720p, IP66, PoE +) for video recording of the entire experimental space, network speaker with SIP, PoE support (AXIS C3003-E NETWORK HORN SPEAKER, Double—sided audio) and microphone (AXIS T8353A MICROPHONE 3.5MM) at the control station to secure communication of the leading experimenter in the control workplace with two experimenters in the coil. ## Experimental procedure Dogs were tested indoors, in a room housing the magnetic coils, and should make the choice between two identical dishes. The dishes were placed at a distance of 2.9 m from the point of release of the dog, always a plus and minus 30° from the starting point. Both dishes contained the same treats and dogs were always allowed to empty both. After placing the dishes, the dog was ready for the starting point and waited to obtain a permit to go to a dish. The dogs could not see the placement of the reward dishes. Three experimenters were involved in the experiment; two were present in the magnetic coil (the owner was guarding the dog and prohibited it from seeing the preparation procedure, and the other was preparing the placement of the rewarded dishes), the third experimenter was in the control room using a microphone and headsets to communicate with the two other colleagues, changed the experimental magnetic conditions (switching between control and experimental conditions) according to a randomized schedule and recorded the results (direction of dog first choice). Note that this person was the only one who knew the actual position of the magnetic North inside the coil. Each dog was tested in three to five test series under the control conditions with the magnetic North (mN) being 0°, and in the same number of test series in an artificially shifted magnetic field with mN = 90° (where magnetic north was set on topographic east). The order of the test series (control first, shifted field second or shifted field first, control second) was taken into account. Tests series were performed at different days, at different daytimes, evenly distributed over the whole day. Because a series included four trials in each dish combination alignment (i.e. N-E, E-S, S-W, and W-N), individual dogs experienced either 48 or 80 trials (in 12 or 20 complete series) in which their turning preference (first dish choice) was recorded under control conditions and the same number of records was gathered for experiments in the shifted magnetic field. The difference in the number of series and trials experienced by individual dogs was given by their availability for our study. In addition, the dog's identity, date, time, sequence of trials combinations, and the order of the trials in the respective series were recorded. ## Paw preferences To determine paw preference (motoric laterality of dogs), a modified Kong test \[e.g. 12,16,33\] was used. In this test, it is recorded with which paw (left or right) the dog holds a Kong, a dog toy (KONG Company) when trying to get the food stuffed inside. A plastic yoghurt cup was used instead of Kong. The inner walls and bottom of the cup were covered with a dog's delicacy such as lard, cream cheese. Each dog was tested at home in an open area for 10 minutes while the dog played with the cup and tried to lick it out and the number of touches with either paw was recorded. Simultaneous touches with both paws were also recorded but were not included in the calculation of the index of laterality. The dogs who did not touch the cup during test of paw preference are excluded from the analysis of the Kong test. ## Data analyses From the recorded choices for each dog, in each trial, the left and right turning preferences were summed, for all four combinations (W-N; N-E; E-S; S-W) separately. For data analysis, the **turning preference index** was calculated in tests performed in the control and shifted magnetic field. The formula (R-L / R + L) x 100 was used, where the R = right and L = left sides are the total numbers of the first choice of left or right dishes. The **laterality index** for the paw preference (Kong test) was calculated using the same formula. The value of the index can range from -100 to -25 (= left-pawed dog) to 25–100 (= right-pawed dog). Dogs with index values between -24 and 24 were considered ambilateral. For the turning preference, altogether ten indices (LI) were calculated; one for each dish combination alignment (N-E, E-S, S-W, and W-N), i.e. four altogether in the control conditions and four altogether in the shifted magnetic field conditions. Furthermore, we calculated one mean index for control conditions and one mean index for shifted magnetic field. The dogs were divided in turning preference left-preferring, right-preferring or irresolute (ambilateral) preference according to based on results of the first trials (Initial turning preference). Generalized Linear Model (GLM) contained the interaction between Magnetic field and Turning preference classes. From the recorded choices, preferences for either left or right turn were calculated for all test combinations (N-E, E-S, S-W, W-N) within each trial, and the sum of all trials of each dog. Index of directional preference was then calculated (according to the above formula) for each dog. All data were analyzed using the SAS System (SAS, version 9.4). For calculating Spearman correlation coefficient we used PROC CORR. To analyze the factors affecting the directional preference index (dependent variable) we used a multivariate Generalized Linear Mixed Model (GLM, PROC MIXED). We constructed two GLMs. The models were applied as a fixed-effect models designed for the repeated measures, i.e., in SAS, with REPEATED = order of testing and the SUBJECT = Name of the dog with compound symmetric covariance structures for repeated measures (TYPE = cs). The first GLM was constructed with the predicted fixed factors Magnetic coil in an interaction with the Turning preference classes, and then we added other variables listed in in case they could affect the directional preference index. None of these variables appeared significant and therefore we will not mention them in the text any more. Least squares means (LSMEANs) were calculated for the categorical fixed effects by computing the mean of each treatment and averaging the treatment means. These means of means were then used to compare the factors. The second model was designed to estimate repeatability of the directional preference across experimental conditions. The GLM contained the only fixed factor Magnetic coil. We calculated repeatability as the intraclass correlation coefficient by adding the RCORR option to the REPEATED. Independently, mean directional compass preference based on the frequency of first choices in a given combination in all pooled trials was calculated for each dog using circular statistics with Oriana 4.02 (Kovach Computing). Grand mean vectors were then calculated on the base of those mean dog vectors for all the dogs, and subgroups with respect to turning preference, experimental condition, sex, and age. # Results ## Paw preference (motoric laterality, Kong test) Following the a priori set criterion, out of altogether 17 dogs tested, 3 dogs were classified as left-lateral, 6 as right-lateral, and 8 as irresolute (ambi- lateral). There was no apparent effect of sex, age, breed or owner on this type of laterality. The correlation between the Kong and overall turning preference tests was rather weak (r_s = 0.317, P = 0.22). ## Turning preference under the control (mN = 0°) and experimental (mN = 90°) conditions Following the a priori set criterion, out of altogether 23 dogs tested, 6 dogs were classified as clockwise-preferring (right-lateral), 7 dogs as counterclockwise-preferring (left-lateral), and 10 as irresolute (ambi-lateral). There was no significant difference in turning preferences of individual dogs between control conditions (mN = 0°) and the shifted magnetic field conditions (mN = 90°). There was a variation in the turning preference index according to the magnetic north direction and Turning preference classes (F_{(23, 131)} = 4.59, P\<0.0001, Figs). For the dogs with clockwise turning preference, there was a trend towards increasing the turning preference index from NE, SE, SW and NW. In other words, the clockwise turning dogs exhibited the lowest turning preference index in the combination North-East. However, only the difference between NE vs NW and between NE and SW, and only in the shifted magnetic field, reached the level of significance (P = 0.05) ( left). For the dogs with counterclockwise turning preference, the most intensive counterclockwise preference was shown in SE orientation in comparison with NW and partly NE, while the weakest preference was in shown in the NW combination. Significant differences were achieved in the shifted magnetic field in SE vs NW, and under control conditions in NE vs SE, SE vs NW, SW vs NW (middle). No trend nor differences were detected for dogs showing irresolute turning preference (right). There was significant bias from the overall turning preferences in the eastern hemisphere, expressed as the "pull of the north", in that a dish placed eastwards was more frequently chosen than a dish placed southwards and a dish placed northwards more frequently chosen than a dish placed eastwards, resulting in an average (theoretical) preference for NNE. In a more differentiated view, this result was due to a dominant preference of females and/or clockwise preferring dogs for North (over East) and to an additional weaker pull of the East over South in males and/or counterclockwise preferring dogs. "Pull of the north" in irresolute dogs was indicated but not significant (Figs and). ### Repeatability of turning preference A single factor of Magnetic coil was not significant (F_{1, 22} = 1.16, P = 0.86). On the other hand, Repeatability was high (r = 0.76). # Discussion Turning preference did not correlate with the motoric paw laterality (Kong test). Apparently, both types of preferences are controlled by different proximate mechanisms / pathways. This conclusion is consistent with earlier findings showing that visual (sensory) and paw (motoric) laterality in dogs are independent of each other. None of the dogs had any previous experience with emptying cups (i.e. Kong-type tests). None of the dogs used in this study had a history of being trained "Heel" to come and follow the master at her/his left (or right) side. Consequently, their turning preferences can be considered natural, spontaneous, inborne, and not entrained. Accordingly, there was no significant difference in the turning preference in particular dogs between the first and second experimental series and there was no effect of the respective owner. Interestingly, among the dogs who turned clockwise there were more females, while among the dogs turning counterclockwise there were more males. The sample was, however, too small to allow any general conclusion with regard to the effect of sex on turning preference. In fact, no clear effect of sex on turning preference was found in a previous study (with a different composition of the study sample). Consistently with results, of the previous study in open field, the turning preference was consistent for each particular dog for all combinations of placement of dishes also in an interior with uniform walls, no apparent landmarks, and no sun or wind cues. Concordantly with the results of the previous study, this preference was slightly, yet significantly disturbed (or pronounced) in that the north-placed dishes were more frequently chosen than would be expected according to the average turning preference of each particular dog. Most important in the context of the present study is the finding that, magnetic and not topographic, north affected the mentioned bias. The detailed analysis shows, however, that the "pull of the north" is a more complex phenomenon, involving also "repulsion of the south". These effects are unilateral: the clockwise turning preference in the right-preferring dogs is more pronounced ("accelerated") in the S-W combination, while the counterclockwise turning preference in the left-preferring dogs is "accelerated" in the S-E combination. On the other hand, N-E combination decreases ("decelerates") clockwise turning preference in the right-preferring dogs, while in the N-W combination, the counterclockwise turning preference in the left- preferring dogs will be reduced. In this way, in the total, south-placed dishes are less frequently chosen than would be expected, while the north-placed dishes are apparently more preferred. Since "rotational deceleration" is stronger in N-E than the N-W combination, while the "acceleration" is stronger in the S-E than in the S-W combination, the resulting theoretical mean preference is for Northeast. It may be of relevance and significance in this context that the analysis of published results on magnetic alignment behaviour in a variety of vertebrate species revealed that magnetic alignment typically coincides with the north- south magnetic axis, however, the mean directional preferences of an individual or group of organisms is often rotated clockwise from the north-south axis. The deviation from the magnetic north-south axis could originate at different levels in the sensory hierarchy: it could be related either to asymmetries at the sensor level or to functional brain asymmetries, i.e. central processing. Although the mode of the perception of the magnetic compass direction in animals remains enigmatic, findings from behavioral, histological, neuroanatomical, and electrophysiological studies have led to several physically viable theoretical models that might also apply to dogs. Two mechanisms are most widely discussed in the literature: the magnetite-based mechanism and the radical-pair mechanism. Perhaps the most intuitively appealing mechanism to explain magnetosensitivity in animals is the idea of a small permanent magnet inside the animal that acts like a compass needle. Magnetite-based sensors may be located anywhere in the body, they do not need to be concentrated in (paired) organs and they can be very tiny. Another proposed mechanism for magnetoreception in animals is based on an effect of the magnetic field on the quantum spin states of a photo-excited chemical reaction that forms long-lived, spin correlated radical pair intermediates (radical pair mechanism;. It is believed to occur in the specialized retinal cells. It is assumed that the magnetic field may generate a “visual” pattern of varying light intensity, color, and/or contrast superimposed on the normal visual scene. The model suggests that north or south “patterns” are more clearly recognizable and easier to be followed than east or west “patterns”. Accordingly, and alternatively, the “pull of the north” could be also interpreted as a “deflection / repellence by the east or west”. Given that the choice of a dish in our experiment was visually guided, we may postulate that the turning preference was determined by the dominant eye, so that a dominant right eye resulted in clockwise, and a dominant left eye in counterclockwise turning. Assuming further that magnetoreception in canines is based on the radical-pair mechanism, a "conflict of interests" may be expected, if the dominant eye guides turning away from north, yet the contralateral eye "sees the north", which generally acts attractive, provoking body alignment along the north-south axis. To test this hypothesis, visual dominance (eyedness) in particular dogs should be studied in an independent test, e.g. sensory jump test. Magnetic alignment might have an adaptive function in that it provides a global reference frame that helps to structure and organize spatial behavior and perception over many different spatial scales. For example, one possibility is that magnetic alignment helps to put the animal into register with a known orientation of a mental (cognitive) map, reducing the complexity of local and long-distance navigation, and reduces the demands on spatial memory. This would be analogous to strategies used in human orientation; it is much simpler and intuitive to navigate when the navigators align themselves with a physical map (i.e. the users rotate their body direction to coincide with the alignment of the physical map), rather than to navigate by mentally rotating the physical map to align with the user’s orientation. Therefore, we suggested that the mental map in animals is fixed in alignment with respect to the magnetic field. Indeed, important component(s) of the cognitive map may be derived from the magnetic field (see below) and spontaneous magnetic alignment behavior may help to place the animal into register with this map. This relatively simple alignment strategy would help animals to reliably and accurately ‘read’ their cognitive map and/or extend the range of their maps when exploring unfamiliar environments. Accordingly, animals of different taxa were frequently reported to prefer to head about northwards when feeding (reviewed in). We suggest that the described simple turning test has a high heuristic potential and should be extended for tests of visual laterality and be performed under a wider array of experimental conditions to get more insight into the very mechanism, seat and function of magnetoreception. # Supporting information 10.1371/journal.pone.0245940.r001 Decision Letter 0 Roman Gregg Academic Editor 2021 Gregg Roman This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 20 Oct 2020 PONE-D-20-31110 Turning preference in dogs: north attracts while south repels PLOS ONE Dear Dr. Burda, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ============================== The reviewers believed there was merit to the study, but also that they lacked all the information necessary to make firm conclusions on the result's validity and the conclusions drawn.  Specifically, both reviewers felt the methods sections was not sufficiently clear.  I believe that much more attention to the methods will be needed for publications.   Specifically the clarification of abbreviations, and the number of trials and what where the indices were used.  I felt that the description of how the experimenters were blinded also needed a great deal more clarity - e.g., state very concisely who interacted with the dogs, and who new the location of magnetic north in each experiment.   The results appeared to leave the reviewers with several more questions.  It is unfortunate that Fig4 was missing for the reviewers, however, both reviewer \#1 and \#2 had additional important questions on the interpretations of figs 2 and 3.  Although new experiments may not be needed, Reviewer \#1 had some important questions regarding dog-owner interactions that should be addressable and would hopefully help eliminate some potentially trivial explanations for the North pull.   Additional questions on eye-laterality were brought up in response to the discussion.  I believe these are important questions from reviewer \#1 and please try to answer if possible. If you cannot answer them, these limitations of current knowledge should be at least addressed in the discussion.  There were also questions regarding the role of the eyes in magnetic detection.  Unless you have experimental data that addresses this, I believe this can also be handled in the discussion by referring to data from other vertebrate systems and clearly drawing the inferences.   ============================== Please submit your revised manuscript by Dec 04 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Gregg Roman, PhD Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please upload a copy of Figure 4, to which you refer in your text on page 10. If the figure is no longer to be included as part of the submission please remove all reference to it within the text. 3\. We note that Figure 1 includes an image of participants in the study.  As per the PLOS ONE policy ([http://journals.plos.org/plosone/s/submission- guidelines#loc-human-subjects-research](about:blank#loc-human-subjects- research)) on papers that include identifying, or potentially identifying, information, the individual(s) or parent(s)/guardian(s) must be informed of the terms of the PLOS open-access (CC-BY) license and provide specific permission for publication of these details under the terms of this license. Please download the [Consent Form for Publication in a PLOS Journal](about:blank) ([http://journals.plos.org/plosone/s/file?id=8ce6/plos- consent-form-english.pdf](about:blank)). The signed consent form should not be submitted with the manuscript, but should be securely filed in the individual's case notes. Please amend the methods section and ethics statement of the manuscript to explicitly state that the patient/participant has provided consent for publication: “The individual in this manuscript has given written informed consent (as outlined in PLOS consent form) to publish these case details”. If you are unable to obtain consent from the subjects of the photograph, you will need to remove the figure and any other textual identifying information or case descriptions for these individuals. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: I Don't Know \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: There were a few places where I was a little confused about the method. For example the paragraph beginning on Line 193. I was confused about how many total trials dogs had and what exactly what meant by first choice. The sentence on Lines 220-222 I found similarly confusing. What were the 10 indices? There were not 10 mentioned so I wasn’t sure what was being explored here. Do other studies show similar patterns in laterality? That is, that about half of the dogs do not have a preference? Is laterality preference (or lack thereof) independent of the task? Or could a dog have a preference for one paw on a kong task and a preference for another paw on a different task? It could be helpful to show that the dogs’ paw laterality isn’t just random but is stable within dogs either by citing work that demonstrates that it is so or by giving these dogs a second laterality test and showing that they are consistent. I think the talk of eye dominance is interesting, but the authors’ case would be significantly strengthened by demonstrating that such eye preferences exist. In other words, is there a simple eye dominance task that the authors could do to assess the dogs’ eye dominance? Given that so much of their discussion is based on the assumption about eye dominance it would strengthen the paper significantly to show that such eye dominance exists and tracks with their predictions. Similarly, is there evidence that dogs have magnetic field receptors in their eyes? The authors mention that the dogs do not have a history of coming to heel, but I wonder about other types of owner interactions. Do owner handedness or owner turning preferences track with dogs’ preferences? There are a few awkward sentences (lines 85-88; 132-134) Where is Figure 4? Reviewer \#2: In earlier outdoor experiments, dogs were found to prefer the North direction and avoid South, when choosing between two food dishes placed in front of them. In this paper, the authors repeat these experiments indoors under controlled conditions, and by shifting the north direction of the magnetic field, they demonstrate that this preference is based on the magnetic field. In the introduction, the questions are clearly stated. The description of the experimental procedure appears rather cryptic and suffers from the use of many abbreviations. The result part is hard to read. I have problems to derive from Fig. 2 a “pull of the north”, and also in Fig. 3, it is unclear what the numbers in each quadrant mean. Fig. 4 is missing altogether. The best part of the paper is definitely the discussion. I welcome the authors’ attempt to propose a promising idea for explaining why so many animals show a magnetic alignment. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0245940.r002 Author response to Decision Letter 0 7 Dec 2020 Dear editor, dear reviewers, we appreciate very much your interest in our study and attention given to our manuscript and the very thoughtful and constructive comments aimed to improve our contribution. We have carefully considered all of them, and we response to them point by point and describe how we changed the manuscript (below, in blue print). Thank you very much for giving us the chance to revise and amend the ms and for taking the revised version and our rebuttal into consideration. Best regards Hynek Burda On behalf of all coauthors Comments and summary of the academic editor Editor points out the need for: clarification of abbreviations, and the number of trials and what where the indices were used. This clarification was requested also by the reviewers and we respond specifically below. Also, we have addressed these points in the revised ms. the description of how the experimenters were blinded (e.g., state very concisely who interacted with the dogs, and who new the location of magnetic north in each experiment). We describe this on lines 177-185 of the revised ms. We added further information which points out the way of blinding the experiment. Fig4 was missing This is indeed unfortunate and we apologize. The figure has been uploaded now. Please note that the statistical values in Table 2 have partly changed. This is because we based the figure now on the attribution of dogs to laterality categories according to results in the first trials (and not to mean laterality indices) to ensure comparability of all figures. both reviewer \#1 and \#2 had additional important questions on the interpretations of figs 2 and 3. We react below and in the revised ms. Although new experiments may not be needed, Reviewer \#1 had some important questions regarding dog-owner interactions that should be addressable and would hopefully help eliminate some potentially trivial explanations for the North pull. Additional questions on eye-laterality were brought up in response to the discussion. We address these important and interesting points below and in the revised ms. Reviewer \#1: There were a few places where I was a little confused about the method. For example the paragraph beginning on Line 193. I was confused about how many total trials dogs had and what exactly what meant by first choice. Looking at the sentence with time lag, we see also the problem and agree with the reviewer. We have reworded the sentence and specified the numbers as follows: Because a series included four trials in each dish combination alignment (i.e. N-E, E-S, S-W and W-N), individual dogs experienced either 48 or 80 trials (in 12 or 24 complete series) in which their turning preference (first dish choice) was recorded under control conditions and the same number of records was gathered for experiments in the shifted magnetic field. The difference in the number of series and trials experienced by individual dogs was given by their availability for our study. The sentence on Lines 220-222 I found similarly confusing. What were the 10 indices? There were not 10 mentioned so I wasn’t sure what was being explored here. Again, we agree with the reviewer and apologize. We have changed the text as follows: For the turning preference, altogether ten indices (LI) were calculated; one for each dish combination alignment (N-E, E-S, S-W, and W-N), i.e. four altogether in the control conditions and four altogether in the shifted magnetic field conditions. Furthermore, we calculated one mean index for control conditions and one mean index for shifted magnetic field. Do other studies show similar patterns in laterality? That is, that about half of the dogs do not have a preference? Is laterality preference (or lack thereof) independent of the task? Or could a dog have a preference for one paw on a Kong task and a preference for another paw on a different task? It could be helpful to show that the dogs’ paw laterality isn’t just random but is stable within dogs either by citing work that demonstrates that it is so or by giving these dogs a second laterality test and showing that they are consistent. Concerning general laterality behavior in dogs: These are relevant questions, but it is not really what this study set out to address. There are published papers on laterality in dogs on a variety of behavioral tasks which we mention and cite (reference numbers 6-22), however, what we can say is that laterality, as measured by the Kong test, cannot account for the turning preference exhibited by the dogs evaluated. Indeed, this is a strength of the study rather than a weakness, as it uncouples handedness/lateral dominance (motoric laterality) and magnetic turning preference (presumably a sensory laterality). Similarly, it was shown in an earlier report (Tomkins et al. 2010, a newly added reference) that visual (sensory) and paw (motoric) laterality are independent of each other (see also the point below). I think the talk of eye dominance is interesting, but the authors’ case would be significantly strengthened by demonstrating that such eye preferences exist. In other words, is there a simple eye dominance task that the authors could do to assess the dogs’ eye dominance? Given that so much of their discussion is based on the assumption about eye dominance it would strengthen the paper significantly to show that such eye dominance exists and tracks with their predictions. Unfortunately, at the current stage of research, given the (wo)manpower and current lockdown-like restrictions we're not going to carry out an additional eye dominance experiment to satisfy this request. For sure, and as we state in the text, it is an inspiration and suggestion for follow-up research. We can, nevertheless, bolster the argument of eye dominance with previous studies from vertebrates, with a focus on mammals that show eye dominance/laterality plays a large role in behavioral ecology. There's lots of examples from birds (where different eyes can specialize on different tasks – either for foraging or for surveillance; the eye dominance with reference to magnetoreception in birds was reported in Refs.29-30, cited in our manuscript). Moreover, eye dominance = ocular dominance = eye preference = eyedness is a well known phenomenon to human ophthalmologists (see Wikipedia and basic literature cited there) and there is no reason to assume that dogs would be different from humans in this respect. Indeed, there is one paper published explicitly on this topic, which, unfortunately, was not cited in the first version of the ms, but, fortunately, came to our notice now to be cited in the revised version (Ref. 35 in the revised version). Similarly, is there evidence that dogs have magnetic field receptors in their eyes? There is no direct evidence, but it has been suggested multiple times for canines in previous published studies (Refs. 45-46) and it has been discussed as the putative magnetoreception mechanism in terrestrial vertebrates (Refs. 39-44), with subterranean mammals who evolved under completely different environmental/ecological contexts, being the exception (Ref. 37). Given the robust support for a photoreceptor based mechanism in closely related taxa, it is an interesting hypothesis to propose and is justified based on previous findings from a diverse array of vertebrates, including mammals. The authors mention that the dogs do not have a history of coming to heel, but I wonder about other types of owner interactions. Do owner handedness or owner turning preferences track with dogs’ preferences? Unrelated to the study and the use of different magnetic field alignments clearly shows that the pull of magnetic north mediates these behaviors. It might be interesting look at some of these other factors, however the study design was intended to address the questions outlined in the last paragraph of the introduction. There are a few awkward sentences (lines 85-88; 132-134) In fact the sentence on lines 85-88 is a word-by-word citation from an English book (Ref. 23). We have shortened it and slightly reworded it now and hope that it became more straightforward. Also the second criticized sentence was reworded. Where is Figure 4? As admitted above – this was an unfortunate omission and the figure 4 has been uploaded now. Reviewer \#2: In earlier outdoor experiments, dogs were found to prefer the North direction and avoid South, when choosing between two food dishes placed in front of them. In this paper, the authors repeat these experiments indoors under controlled conditions, and by shifting the north direction of the magnetic field, they demonstrate that this preference is based on the magnetic field. In the introduction, the questions are clearly stated. We are pleased by this assessment. The description of the experimental procedure appears rather cryptic and suffers from the use of many abbreviations. The reviewer might be right but the problem can be solved only on costs of losing some details or lengthening the text and making it even less understandable. Importantly, all abbreviations are either known units (nT = nanotesla), or are commonly used in the literature of this kind and in any case explained when first used (M = male, F = female, N = North, magN = magnetic North, etc.), or explained when used in a formula (R = right, L = left) or they specify marks of the used software or hardware (which information is important for those who would like to assess suitability of our equipment or replicate the experiments and equip their labs. Finally, there are, abbreviations which are of importance and interest only for statisticians who themselves are familiar with statistical Analysis system (SAS), e.g. LSM for Least squares means, GLM (generalized linear model), etc. Again, all these abbreviations are explained when first mentioned. Repeating whole descriptions in each sentence instead of using these abbreviations would not make the text more fluent, readable and understandable. Omitting these and not mentioning these models would, for sure, be criticized by statisticians. The result part is hard to read. We reworded and complemented the text. I have problems to derive from Fig. 2 a “pull of the north” The reviewer is very attentive. There was a mistake in the Figure 2a. The corresponding author uploaded mistakenly an earlier and incorrect version of the figure. Sorry for that error and thank you for finding it. The correct version is uploaded now and the figure is explained in more detail. and also in Fig. 3, it is unclear what the numbers in each quadrant mean. Description of the Fig. 3 is reworded as follows: Fig 3: Numbers in each quadrant (in the respective four compass combinations (N-E, E-S, S-W, W-N) show mean values of turning preference calculated from individual dogs and pooled across all trials (both control and shifted magnetic field alignments). Data were partitioned by turning preference (left figure shows clockwise turning preference, right figure shows counterclockwise turning preference; irresolute dogs were not calculated. The green arrow over the dog's head in the…….. (the further description of the figure remains unchanged). Fig. 4 is missing altogether. As admitted above – we apologize for this unfortunate omission. The figure 4 has been uploaded now. The best part of the paper is definitely the discussion. I welcome the authors’ attempt to propose a promising idea for explaining why so many animals show a magnetic alignment. We appreciate this opinion. 10.1371/journal.pone.0245940.r003 Decision Letter 1 Roman Gregg Academic Editor 2021 Gregg Roman This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 5 Jan 2021 PONE-D-20-31110R1 Turning preference in dogs: north attracts while south repels PLOS ONE Dear Dr. Burda, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ============================== Along with the two reviewers, I believe your manuscript is very nearly ready for publication. I appreciate the detail with which you address the reviewers comments and concerns.  Reviewer \#1 had some last edits that you may wish to consider.  Please consider these edits before returning the manuscript.  I would especially like you to consider a change to the Fig1 legend that would make the meaning of the measures more apparent to someone less familiar with the experimental approach.   Please also read through closely to make sure you catch any other existing typos before resubmission.  Congratulations on a very nice paper.  ============================== Please submit your revised manuscript by Feb 19 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Gregg Roman, PhD Academic Editor PLOS ONE \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: (No Response) Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: I Don't Know Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Some minor things: Lines 153-162 — paragraph should be in past tense Line 219 tense shift Line 244 “according to based on” (redundant) Line 326 “more frequently" need to add "chosen” In the figure caption for Fig 1 please give a sense of what these numbers mean. - ie - left and + right? We don't need the full explanation that comes later in the text, but a sense of what these numbers are telling us would be helpful at this point. I still find the figures confusing, but I’m not an expert on how to present these results with turning preference so I leave it to the other reviewer and the editor's expertise. Reviewer \#2: The text is considerably improved, and it is now easier to follow the argumentation of the authors. I think that the paper can be published now. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0245940.r004 Author response to Decision Letter 1 7 Jan 2021 Dear Prof. Roman, we are pleased about the positive views of the reviewers and appreciate also the careful editing of the manuscript by the reviewer \#1: We accepted all but one suggestions: Lines 153-162 — paragraph should be in past tense Corrected Line 219 tense shift Corrected Line 244 “according to based on” (redundant) We have not changed this sentence, because "based on first trials" is here a complementary information, a fact, which should be pointed out Line 326 “more frequently" need to add "chosen” Corrected In the figure caption for Fig 1 please give a sense of what these numbers mean. - ie - left and + right? We don't need the full explanation that comes later in the text, but a sense of what these numbers are telling us would be helpful at this point. The reviewer means probably Table 1 and/or Figure 3. We complemented the caption in both cases as suggested. I still find the figures confusing, but I’m not an expert on how to present these results with turning preference so I leave it to the other reviewer and the editor's expertise. We assure the reviewer that this way of illustrating the results is common in the literature dealing with spatial orientation. Once again, many thanks for considering our manuscript and ist academic processing. With best regards and wishes for a prosperous and healthy new year Hynek Burda On behalf of all coauthors. 10.1371/journal.pone.0245940.r005 Decision Letter 2 Roman Gregg Academic Editor 2021 Gregg Roman This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 11 Jan 2021 Turning preference in dogs: north attracts while south repels PONE-D-20-31110R2 Dear Dr. Burda, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Gregg Roman, PhD Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0245940.r006 Acceptance letter Roman Gregg Academic Editor 2021 Gregg Roman This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 20 Jan 2021 PONE-D-20-31110R2 Turning preference in dogs: north attracts while south repels Dear Dr. Burda: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr Gregg Roman Academic Editor PLOS ONE [^1]: No authors have competing interests.

# Introduction Single base mutations can be repaired by the introduction of DNA oligonucleotides (ssODN) into a target cell. The frequency of this corrective activity depends on a number of factors including the length of ssODN, the position of the cell in its proliferative cycle – and the presence of double- stranded DNA breaks in the host genome. Studies centered on the effect of S phase transition on gene repair have led to the emergence of a model in which reversal of genotype takes place most often through the incorporation of the ssODN into a newly synthesized DNA strand. The level of gene repair is enhanced dramatically when cells are targeted during S phase and, specifically, when they are slowed in their progression through S phase. Anticancer drugs were tested as agents to stimulate gene editing activity based on the concept that DNA damage (in the form of ds DNA breaks) might activate proteins involved in homology directed repair, slow cell cycle progression and thus stimulate gene correction. Ferara *et al* demonstrated that pre-treatment of cells targeted for gene editing by ssODNs with camptothecin (CPT) enhanced gene editing activity 5–10 fold. But, such treatment of cells leads to nondiscriminate, nonspecific ds breaks which again presents a practical barrier in the development of gene editing for molecular medicine. One solution to this problem appears to lie in the use of enzymes that can produce a “unique” specific ds DNA break in the genome, preferably at or near the position of the mutant base. Although the field is evolving, three major agents are currently used to catalyze specific ds DNA breaks: Zinc-Finger Nucleases (ZFNs), Clustered Regularly Interspersed Short Palindromic Repeats (CRISPR-Cas9) and Transcription Activator-Like Effector Nucleases (TALENs). By using agents that cut specifically, one can reduce the chance of offsite mutations while simultaneously stimulating the frequency of gene editing. These so-called “programmable nucleases” may enable the more efficient use of the ds break as a stimulatory factor in reactions designed to correct single base mutations. We have chosen to utilize TALENs to enhance the frequency of gene editing, directed by ssODNs, and repair a missense mutation in the eGFP gene. A single copy is integrated into the genome of a clonally isolated and expanded HCT116–19 cell line. This well-established model system has unique advantages including the capacity to correlate genotypic and phenotypic changes with functional protein activity. Recently, we showed that the combinatorial action of ssODNs and a TALEN designed to cut at −2/−3 relative to the mutant base (**G→C**) results in a substantial rise in the frequency of gene editing. Importantly, TALENs reduce the level of ssODNs needed for nucleotide exchange, eliminating the onset of the Reduced Proliferation Phenotype (RPP). TALENs and ssODNs had been reported previously to work together to facilitate genome editing, but we have taken a more decidedly reductionist approach to characterize this reaction in somatic cells. In this paper, we build upon that original observation and focus on *(1)* the position of the cut site in the region surrounding the mutant base, and *(2)*, the effect of cell synchronization – on specific TALEN activity to enable gene editing in somatic cells. # Materials and Methods ## Cell Line and Culture Conditions HCT116 cells were acquired from ATCC (American Type Cell Culture, Manassas, VA). HCT116–19 cell line was created by integrating a pEGFP-N3 vector (Clontech, Palo Alto, CA) containing a mutated eGFP gene. The mutated eGFP gene has a nonsense mutation at position +67 resulting in a nonfunctional eGFP protein. For these experiments, HCT116 (−19) cells were cultured in McCoy’s 5A Modified medium (Thermo Scientific, Pittsburgh, PA) supplemented with 10% fetal bovine serum, 2 mM L-Glutamine, and 1% Penicillin/Streptomycin. Cells were maintained at 37°C and 5% CO₂. Custom designed oligonucleotides, 72NT, 40NT and 100NT were synthesized from IDT (Integrated DNA Technologies, Coralville, IA). ## TALEN Design and Construction Nine Left and eleven Right TALEN half-sites were designed to flank the target at a range of −39 to +46 base pairs of the integrated mutant eGFP gene (TA**G** = 0). TALENs were designed according to previously published guidelines to have a Thymine (T) at position 0 of the TALEN binding sequence and a DNA binding domain of 15–20 base pairs (15–20 RVDs). The 20 constructed TALEN half- sites were combined for targeting experiments if the two half-sites produced a spacer between 13 and 29 base pairs. By using all possible combinations of TALEN half-sites, 22 total TALEN combinations were tested that flank the mutant base. Construction was done via the Gold Gate Assembly method originally developed by Cermak et al. and purchased through Addgene (Addgene, Cambridge, MA). The final step of the assembly protocol was modified to include the mammalian expression vector pc-GoldyTALEN, which has been optimized for expression and cutting efficiency in mammalian systems. Following construction, colony PCR and DNA sequencing by Genewiz Incorporated (South Plainfield, NJ) was performed to confirm correct TALEN constructs. The only differences between each TALEN half- site are the order in which the RVDs were arranged, corresponding to the DNA they were designed to target. The RVDs used were HD, NI, NG, and NN only. Following construction, colony PCR and DNA sequencing by Genewiz Incorporated (South Plainfield, NJ) was performed to confirm correct TALEN constructs (for reference, the full sequence of L848–19 TALEN plasmid can be seen). ## Transfection of HCT116–19 Cells and Experimental Approach For experiments utilizing synchronized cells, HCT116–19 cells were seeded at 2.5×10⁶ cells in a 100 mm dish and synchronized with 6 µM aphidicolin for 24 hours. Cells were released for 4 hours prior to trypsinization and transfection by washing with PBS (−/−) and adding complete growth media. Synchronized and unsynchronized HCT116–19 cells were transfected at a concentration of 5×10⁵ cells/100 ul in 4 mm gap cuvette (BioExpress, Kaysville, UT). Single-stranded oligonucleotides and TALEN plasmid constructs were electroporated (250 V, LV, 13 ms pulse length, 2 pulses, 1 s interval) using a Bio-Rad Gene Pulser XCellTM Electroporation System (Bio-Rad Laboratories, Hercules, CA). Cells were then recovered in 6-well plates with complete growth media at 37°C for 48 hours prior to analysis, unless otherwise noted. ## Analysis of Gene Edited Cells Fluorescence (eGFP) was measured by a Guava EasyCyte 5HT Flow Cytometer (Millipore, Temecula, CA). Cells were harvested by trypsinization, washed once with 1x PBS (−/−) and resuspended in buffer (0.5% BSA, 2 mM EDTA, 2 µg/mL Propidium Iodide (PI) in PBS −/−). Propidium iodide was used to measure cell viability as such, viable cells stain negative for PI (uptake). Correction efficiency was calculated as the percentage of the total live eGFP positive cells over the total live cells in each sample. Error bars are produced from two sets of data points generated over two separate experiments using basic calculations of Standard Error. ## TALEN Cleavage Analysis HCT116–19 cells were electroporated at a concentration of 5×10⁵ cells/100 ul in 4 mm gap cuvette (BioExpress, Kaysville, UT) with TALEN pairs −35, −28, −1/+1 and +7/8 at 2 ug and 10 ug. Cells were then recovered in 6-well plates with complete growth media at 37°C for 72 hours. DNA was isolated using the Blood and Tissue DNeasy kit (Qiagen, Hilden, Germany). RFLP analysis was performed on 181 bp amplicons that were created using forward primer, 5′GAGGGCGATGCCACCTACGGC and reverse primer, 5′GGACGTAGCCTTCGGGCATGGC. PCR samples were cleaned up using the QIAquick PCR purification kit (Qiagen, Hilden, Germany) and treated with the indicated restriction enzymes following the manufactures protocol. Digested samples were loaded along with NEB 2-log DNA ladder (NEB, Ipswich, MA) into a 2% TBE agarose gel for analysis. T7 Endonuclease assay was performed on amplicons of 605 bp with forward primer 5′CTGGACGGCGACGTAAACGGC and reverse primer, 5′ACCATGTGATCGCGCTTCTCG. Following PCR cleanup, each TALEN treated sample was placed in a thermocycler for heteroduplex formation. Samples were then treated with 1 ul of T7 endonuclease at 37°C for 1 hour and then subjected to electrophoresis on a 2% TBE agarose gel containing ethidium bromide. Images for both RFLP and T7 assay were collected by the Gel Doc EZ System (BioRad, Hercules, CA). # Results and Discussion ## Results The eGFP gene was mutated near the 5′ end of the coding sequence creating a stop codon (TAG) in place of a tyrosine (TAC). As such, eGFP is produced in truncated form and no green fluorescence is observed in the cell. displays some of the target sequence and the integration vector used to insert a single copy of the gene driven by a CMV promoter in HCT116 cells. Also displayed are three ssODNs that are designed to create a mismatch with the G residue of the TAG codon on the nontranscribed strand (NT) (*or sense or nontemplate strand*), thereby directing gene editing at this base. The three NT-ssODNs are of identical sequence through 40 bases but vary in length from 40 to 72 to 100 nucleotides respectively. This system is well established as a model for analyzing the mechanism of gene editing in human cells. In most applications, 72NT ssODN has been used in optimization studies for delivery and the response of cell and genomic DNA in the gene editing reaction. Once repair of the TA**G→**TA**C** has been facilitated, the population of cells is analyzed by FACS and the percentage of live green fluorescent cells within that population can be presented as the frequency of gene correction. Sorted eGFP⁺ cells are easily quantified and genotype verified by direct DNA sequencing. Thus, genotype and phenotype, expression of a functional protein, is assessed in a valid, simple way, a critical component of reaction optimization or characterization studies. We have shown previously that the gene editing directed by 72-mers (NT) take place at an approximate level of 0.7% unless synchronized and released cells are used; in that case, frequencies approach 2%. In either case, however, in past studies, the level of ssODN required to activate the reaction is so high that the corrected cells cease to proliferate; a phenomenon we termed, Reduced Proliferation Phenotype. By incorporating TALENs into the reaction mixture, we were able to reduce the amount of ssODN in the reaction and corrected cells responded by continuing their normal growth rate. The TALEN pair used in those studies cuts the eGFP gene 5′ (upstream) of the target base. Since TALENs can be “programmed” to cleave at most sites in the DNA, we created an array of TALENs to analyze the impact of the cut site on gene editing of the TA**G** codon. The 20 TALEN plasmids (9 Left and 11 Right:) were designed according to previously published guidelines and recommendations. Briefly, each TALEN half-site binding sequence (left or right plasmid) is preceded by a thymine (T) and contains 15–20 RVDs which bind to the DNA sequence; 15–20 base pairs respectively. Through this strategy, 22 TALEN combinations were created. These combinations allowed for the creation of TALENs with a range of spacer sizes which can dictate or even restrict the position of cleavage sites relative to the mutant base. Cloning of the TALEN constructs was done via the Golden Gate Assembly method originally developed by Cermak *et al* with the final step of the assembly protocol modified to include the mammalian expression vector pc-GoldyTALEN. Thus, the experimental protocol was to introduce ssODN and two plasmids (one expressing the left and one, the right TALEN) into HCT116–19 cells, a clonally expanded line that contains a single copy of the mutant eGFP gene and at various times thereafter, analyze gene editing activity by FACS. The goal is to define the range of genomic cleavage sites that enable gene editing directed by ssODNs bearing different lengths. We had already established that the reaction was dependent on both TALEN arms being present, the presence of a specific ssODN designed to direct the change and an optimized TALEN: ssODN ratio. Three different lengths of ssODN, all complimentary to the sense or non-template strand (NT) were used; 40NT, 72NT and 100NT. The cut sites are predicted to occur at the center of the spacer region for each TALEN pair. In our system, cut sites are designated by their position relative to the target base (G) in the TA**G** codon. For example, where **G** is 0, the 5′ end of **G** is −1, the 3′ end is +1, and so on. When the spacer region of the TALEN pair contains an even number of nucleotides, there is one predicted cut site. When the spacer region contains an odd number of nucleotides, there are two predicted cut sites because with even spacers, the cut site falls directly between two bases in the spacer region while with odd numbered spacer regions, the center falls directly on a unique nucleotide. Consequently, the predicted cleavage can occur on either side of the nucleotide at the center of the spacer region. For example, the TALEN pair L848–19 and R898–19 has a spacer length of 13 nucleotides. The direct center of this spacer region does not fall between two nucleotides but rather on *top* of the 7^th nucleotide. Accordingly, the cleavage is predicted to occur on either side of this nucleotide, which is either −2 or −3 from the TA**G**. Following electroporation, cells were placed in 6-well plates and allowed to recover for 48 hours and fluorescence detected by FACS. The results reveal significant gene editing activity when TALEN activity takes place, −8/−9 bases upstream to +6/+7 bases downstream respectively, relative to the target base (0). In an attempt to expand the range of effective cleavage sites, we utilized a 100-mer (100NT), that would hybridize further upstream and downstream; TALEN pairs that cut at sites −28, −8/−9, −5, −1/+1, +2, +6/+7 and +11 were tested. None of the cut sites enabled higher activity when the 100-mer was used as compared to using 72NT. Past studies had shown low activity with shorter ssODNs, but to complete the analyses, 40NT was also tested with a selection of TALENs that cleave at −8/−9, −5, −3/−4, −2/−3, −2, −1/+1 and +3/+4 respectively. The 40NT displayed nearly undetectable gene editing activity throughout the broad range of target sites consistent with previous observations. ## Cell Synchronization in TALEN/ssODN-directed Gene Editing The frequency of gene editing can be raised if the cells being targeted are progressing through S phase – –. In a previous study, we demonstrated this phenomenon holds true when ssODNs are paired with TALENs to direct gene repair. We extended that protocol and used the most active pairs of TALENs, as defined in, in combination with 72NT. Cells were synchronized for 24 hours then released for 4 hours at which time a pair of TALENs and 72NT were introduced by electroporation. After 48 hours, gene editing activity was assessed and results presented in. In every case, targeting synchronized and released cells produces a higher level of gene editing than targeting an unsynchronized population, albeit to various degrees of stimulation. The differences among the cleavage sites range from essentially within experimental error to approximately threefold. Again, we see that the area immediately around the target base −1/+1 produces the highest levels of gene editing. Synchronization and release of cells destined for gene editing can enable a higher degree of activity. TALENs are designed to catalyze double-stranded DNA breakage and facilitate gene knockout through NHEJ or, as reported herein, facilitate gene editing. In our hands, the level of TALEN activity required for efficient gene editing is lower than what is traditionally used to enable DNA cleavage for gene knockout. Nevertheless, we sought to evaluate a group of TALENs used in this study for DNA breakage; we chose several TALENs that **do not** support gene editing (−35, −28, +7/+8) and the core TALEN (−1/+1) that supports it the best. Our goal is to ensure that both types of TALENs display activity. One assay system utilizes T7 endonuclease to cleave at heteroduplexed DNA–an outcome of reannealing of DNA strands that arise from TALEN cleavage **and** NHEJ. Our initial pass through all 22 TALEN pairs did not yield robust results except in two important cases (see below). Thus, we used an assay that measures loss of a restriction site (RFLP) as evidence of TALEN activity, as employed convincingly by Bedell et al and Qui et al. Within our targeting zone, we have four sites that correspond to restriction enzyme sites, located at −35, −28, −1/+1 and +7/+8 respectively, diagrammed in. Evidence of TALEN activity is the reduction of restriction enzyme cleavage at the designated site. Gene editing was carried out as described in using these four TALENs. The extracted DNA was isolated and amplified across the DNA regions containing sites −35, −28, −1/+1 and +7/+8, then cleaved by BaeGI (−35), BstNI (−28), AvrII (−1/+1) and TspRI (+7/+8) respectively. illustrates the results. In all three cases, TALENs designed for these four sites created a sequence alteration so that a percent of the target DNA is seen to remain resistant to cleavage. The nontreated (NT) lanes display little or no uncut DNA representing the highly efficient activity of the restriction enzyme; a dose dependency (2, 10 µg of TALEN) is also evident. These data demonstrate TALEN activity at sites that do not support gene editing. We also tested the −1/+1 site and the results are displayed in. Again a resistant band appears as a function of TALEN dosage in the reaction. As an extended control to confirm the specificity of this assay, we carried out a restriction enzyme digest with two of the enzymes whose cleavage efficiency and site would not change if TALEN activity was precise at −1/+1. The isolated DNA from treated samples, at −1/+1 was cleaved by BaeGI or BstNI (not AvrII). The results, seen in, show complete cleavage with no residual resistant band. Finally, we were able to obtain robust and reproducible results using the T7 endonuclease assay for sites −1/+1 and +46 (see Figure S1 in). In both cases, TALEN activity was confirmed in a dose- dependent fashion with the predicted band size (459 and 412 bp). Importantly, slightly greater TALEN activity is observed at the +46 site, a site where gene editing activity is undetectable. Hence, by three separate criteria, we find that TALENs supporting or not supporting gene editing exhibit equivalent and significant DNA cleavage activity. # Discussion The correction of an integrated, single copy of a mutated eGFP gene has been achieved by the combinatorial activity of ssODNs and TALENs. The repair of the TA**G** stop codon to TA**C** (tyrosine) converts the protein to wild type with phenotypic fluorescence that can be easily quantitated by FACS. While this target gene lacks the clinical relevance of naturally mutated genes, often resulting in inherited disorders, valid outcomes of gene editing can be easily measured at the genetic and protein *activity* level. Previous data from our lab and others – have indicated that double-stranded breaks introduced by the inclusion of anti-cancer drugs in the reaction, enhance the frequency of this repair. While the mechanism of induction still needs more study, a fairly clear picture of its inner workings has emerged and has been confirmed. With the advent of reliable site-specific nucleases operational in mammalian cells, it is now possible to substitute TALENs, Zinc-Finger-Nucleases or CRISPR- Cas9 reagents for the nonspecific activity of anti-cancer drugs. In this study, we have created 22 pairs of TALENs, designed to cut at various places around the TA**G** codon, and paired each with ssODNs designed to direct the repair of the inherent mutant base. Our data suggest that TALENs that cleave at proximal locations near the target base can enhance the frequency of repair to varying degrees. We had previously shown that TALEN activity at −2/−3 (upstream from the mutant base) stimulated the reaction 100 fold above correction levels observed when only the 72NT ssODN was used in this eGFP⁻ targeting system. This choice was somewhat fortuitous since it is one of the most active sites observed when other cleavage site locations were tested. While each TALEN construct contained a workable, established spacer length, some variation in the region is observed and could result from lower levels of cleavage. The TALEN pair built to cleave on either site of the G-target base −1/+1 produced the highest level of correction driven by 72NT. What is also apparent is that proximal cleavage sites are clearly better targets for gene editing as extending the cleavage sites upstream beyond −8/−9 reduces gene editing activity significantly. Thus, somewhere between −8/−9 and −28 respectively activity falls off precipitously. Another interesting feature of this analysis is that the range of productive TALEN target sites appears to be biased leftward, although some activity is observed at the +6/+7 site, downstream. Such a bias may reflect more about the contribution of the 72NT and its mode of action post cleavage, as opposed to direct impact of TALEN activity. Interestingly, substituting the 100-mer (100NT) for 72NT did not rescue the low activity of upstream or downstream sites. A shorter ssODN, 40NT, produced almost undetectable levels of activity throughout the region. Again, it may be that an optimal length of ssODN is required to participate in the gene editing reaction (see below). A number of other laboratories have clearly demonstrated the productive activity of ssODNs and TALENs to direct genome editing. In many cases, these experiments were carried out in ES cells, iPSCs or model organisms such as Zebrafish while only a few of them deciphering actual mechanism of action in standard, somatic cell lines have been published. Most of the activity surrounding TALENs center on the generation of knockout cell lines or animals. Recently, Yang *et al* published an elegant study focusing on cleavage/target site location and several reaction parameters including ssODN length. These workers found that proximal cleavage within 50–100 bases of the target base produced the highest level of gene editing and that there is an optimal length for the ssODN in driving the reaction. With the advent of tailored nucleases that can cleave at specific sites in the mammalian genome, the pace of development of genome editing toward clinical application has been accelerated. In this work, we employed TALENs in an effort to induce double-stranded DNA breaks, a form of DNA damage that had shown previously to increase the frequency of gene editing directed by ssODNs. While it is prudent to measure cleavage activity, the most commonly used assay, T7 Endonuclease, has been employed by and large to confirm TALEN activity in studies where the objective was to disable, not repair, a gene. And, in many of these cases, the amount of TALEN needed to execute genomic knockout was 2–5 times higher than the optimal level required for ssODN-directed gene editing. In fact, previous data suggest that increasing amounts of TALENs reduce, not elevate, gene editing activity. We chose to monitor TALEN activity using an assay that identifies the introduction of an RFLP, as reported by Bedell *et al* and Qui *et al*. Within the eGFP gene targeting region, there exists several restriction enzymes cleavage sites that correspond with the cleavage sites of some of our TALENs. We found the assay to be more reliable with less manipulation of the target DNA than the T7 Endonuclease assay. For example, if one surveys the literature, reaction conditions for this assay vary widely suggesting that each modified cell line’s genomic DNA must be treated differently to obtain the desired products. While we recognize RFLP changes also have a few drawbacks, in our hands this assay produced more reliable, reproducible and robust data. The assay is obviously limited to TALEN cleavage sites that are coincident with restriction enzyme recognition sequences. Our target is a single copy gene in a mammalian genome, thus “gene repair” by HR or “homology-directed repair” (HDR) serves an unlikely mechanism of action. There is always a tendency to assume that biological reactions must occur by the same pathway, a traditional reductionist view, but because of the differences described above we suggest that combinatorial ssODN and TALEN-directed gene editing follows a different route than TALEN-directed genome modification. However, projections of mechanisms of action have to be made with caution. The cell line HCT-116 is genetically devoid of certain MMR functions such as MLH1, MLH3 and PMS2. Therefore correlations among NHEJ, HR, HDR and even MMR in gene editing cannot be definitively established. The editing frequency of all of the sites tested surrounding the target base are enhanced when synchronized and released cells are used in these experiments. Previously, we had reported this phenomenon with one set of TALENs. These results point again to the importance of DNA replication in the gene editing reaction. Based on the restrictions we observe regarding the need for proximal cut sites, it is likely that these ds breaks provide an entry point for the ssODN to align in homologous register with the target region. Once aligned, it could provide a 3′OH for extension and act as a “quasi Okazaki fragment” as previously suggested. What complicates this simple explanation is the fact that single-stranded annealing (SSA) and extension synthesis is likely to be in competition with NHEJ. Thus, while gene editing may prove to be successful, by design, the resealed break (by SSA and extension) becomes a newly formed target for TALEN activity and perhaps NHEJ. And, it is unclear how many times this cycle can be repeated. The elegant studies of Liu *et al* however may provide some insight. Building on previous work of the Resnick lab, these workers suggest that “SSO-directed information transfer is restricted to the immediate vicinity of the DSB…” This observation predicted the results of the mapping experiments we report in this paper. Liu *et al* also suggest that the SSO may in fact reduce the number of NHEJ events thereby tipping the balance away from the potentially mutagenic activity of NHEJ. In addition, the fact the synchronization and release of these cells enables higher levels of targeting since more cells are traversing through S phase, may shift the balance toward HR or HDR and away from NHEJ. These results also align with the work published by Morozov and Wawrousek in which HR proteins involved in homologous pairing (Rad51 and Rad54) were found to stimulate gene editing while NHEJ proteins KU 70/86 were seen to inhibit the reaction. Shifting the equilibrium toward homology- directed repair (or recombination) may be a fundamental, mechanistic aspect of ssODNs as they direct inheritable changes in the genome. Taking advantage of this biased equilibrium is one area currently under study and is likely to be an important reaction parameter as we define combinatorial methods for utilization in genome modification experiments in mammalian cells. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: NRT BS PB EBK. Performed the experiments: NRT BS PB RAN EBK. Analyzed the data: NRT BS PB RAN EBK. Contributed reagents/materials/analysis tools: NRT BS PB RAN EBK. Wrote the paper: EBK.

# Introduction The ability of block copolymers (BCPs) to self-assemble into periodic structures, with periods ranging from 5 nm to well over 100 nm, has prompted investigation into their potential applications for nanopatterning of integrated circuits,\[–\] bit-patterned storage media,\[–\] optical devices, tissue interfacing,\[–\] and others. Lamellar or cylindrical domains of block copolymers can be used to create linear structures, both large and small, when confined in one dimension as thin films on substrates with appropriate wetting characteristics.\[–\] Such patterns can be used as lithographic masks through etching or as scaffolds to create other nanostructured surfaces and materials. For application in semiconductor fabrication, the International Technology Roadmap for Semiconductors (ITRS) has, in its Directed Self-Assembly Critical Assessment (where the term directed self-assembly is represented by the acronym DSA), identified challenges in 15 metrics, including: Feature sizes of under 10 nm, the ability to “add, exclude or trim individual DSA…features with simple lithography”,²² a low degree of line edge roughness (LER, 3σ) \< 0.6 nm, defect density less than 0.01 cm^-2, and an annealing time of less than one minute. In addition, surfaces require appropriate wetting characteristics and surface energies in order to enable the process of self- assembly in the desired orientation with respect to the surface plane. These metrics have been correctly identified as challenges as they are daunting goals, but they represent very clear, quantified metrics that need to be attained. Lacking, however, is a unified method of accurately determining each parameter ‘in the field’, with actual samples of surfaces patterned *via* block copolymer self-assembly (*vide infra*). Much of the work to optimize BCP DSA has been carried out with a narrow range of polymers, namely polystyrene*-block*-poly(methyl methacrylate) (PS-*b*-PMMA), polystyrene-*block*-polydimethylsiloxane (PS-*b*-PDMS),\[,–\] and polystyrene-*block*-poly(2-vinylpyridine) (PS-*b*-P2VP). Each of these polymers possesses favorable characteristics for nanopatterning, but many other block copolymer systems still remain to be designed, synthesized, and investigated, as the exploration of the space of possible systems, including structural classes and chemical motifs (monomers) is nowhere near complete\[,–\] Dimensions of polymer-space available for exploration include triblock, comb, or other architectures\[–\] and topologies; alternate chemical moieties such as silicon-containing polymers other than PDMS, and oxygen-rich groups such as oligosaccharides and poly(lactic acid); or tailoring polydispersity to modify morphological stability and domain sizes.\[–\] A consequence is that there remains much to be explored synthetically in order to optimize pattern formation, etch selectivity (or resistivity), polymer reactivity, surface energies, Flory-Huggins parameters, LER, and annealing conditions. In particular, there is a persistent analytical barrier that synthetic chemists must overcome in order to readily determine whether their polymeric creations may be applicable to novel DSA applications: they require access to a toolbox capable of analyzing critical features such as the defect density, correlation lengths, and LER of their patterns in order to determine whether their block copolymers have promise. The dearth of accessible tools remains a significant obstacle for the area of directed self-assembly. Defects themselves also warrant a more in-depth investigation, which can only be achieved by studying defects “in the wild”, in the actual nanopatterns as they progress through various stages of annealing. While simulations can find matches to thin film defect structures,\[–\] automated analyses of defects in block copolymer thin films in an experimental setting allow access to statistical data about the frequency and distribution of various defects. Statistical data is generally inaccessible *via* modeling due to computational limits for defects beyond the simplest examples. Furthermore, for cylindrical block copolymer domains, defects do not always have liquid crystal analogues, rendering past defect-detection methods, originally developed for patterns formed in liquid crystal thin films, inappropriate. Hence identification of structures beyond simple counting of disclinations and dislocations would be advantageous. Computerized analyses that are widely used to study images of BCP patterns include to determination of periodicity using azimuthally averaged fast Fourier transform (FFT) images, image filtering and regional analysis of domain orientation using FFTs and defect density measurements. Less commonly measured is LER. Rarely, however, are more than one or two methods packaged together in a published work, which leaves unanswered questions since there is typically a trade-off between factors such as defect density, orientation, nanostructure spacing, line-width, line-width roughness (LWR), LER, and correlation length (i.e., grain size). Initially, in the course of investigating the process of microwave annealing of block copolymer thin films, we developed an in-house algorithm to quantify defects in block copolymer thin films, utilizing particle analysis and skeletonization to identify defect features. Later, while analyzing density doubled cylindrical line patterns, a separate process for measuring the LER was created, which was limited to analyzing nearly-straight segments of nanowire structures. To remedy the lack of a readily available and straightforward analytical tool, we developed an accessible and free-to-download application for analyzing the defects in BCP thin films using a combined particle and skeleton based analysis of the pattern, called Automated Defect Analysis of Block Copolymers, or ADAblock for short (links to the tool provided in). The application was constructed using the ImageJ platform, a free, open-source, Java-based image analysis program, which provides a full and easy-to-understand output. The tool identifies not only the type(s) of defects found in a sample, but also quantifies the density of defects over a range of length scales, accompanied by additional information regarding LER and LWR, as well as an alternate means of accessing the correlation length. In this work, we screened the ADAblock application against a range of nanopatterns prepared *via* block copolymer self- assembly and show the effects of polymer molecular weight on the defect densities of self-assembled BCP films. Additionally, we demonstrate how ADAblock can simultaneously track LER, defects, and correlation lengths. To our knowledge, no previous work analyzing 2D block copolymer line patterns has brought together data on defects, LER, LWR, and correlation lengths into one application or analysis. We believe that this omission is likely due, in part, to the lack of readily available, widely applicable, easy-to-use tools for analysis, and on occasion, may be a result of selection of the ‘makes-it-look- best metric’, rather than a complete description of pattern quality over larger areas of the sample. In this paper, we show how such data can ideally be combined to better describe line patterns derived from BCP assembly in thin films. Images can be deceiving, and we hope that ADAblock will assist researchers in avoiding pitfalls resulting from performing incomplete defect density analyses. # Results and Discussion Examples of four different nanopatterns derived from BCP self-assembled templates are shown in. The patterns have been converted into easily visible platinum lines through a well-described platinization of three different PS-*b*-P2VP BCPs; the Pt nanolines are derived from the P2VP blocks. Although these scanning electron micrographs (SEMs) are similar in appearance, each is subtly different, and thus the question to be posed is how to distinguish one pattern from another and to determine which is more defective. As shown in and, the pattern in 1A has 30% more defect pairs than the patterns in. Moreover, the correlation length of 1B is shorter than any of the others (in part due to the shorter period). Additionally, in terms of the line edge roughness (LER), they all appear at first glance to be quite smooth, but the measured roughness of these lines, as summarized in, would put them out of contention for ITRS targets. With respect to LER, the values of \~ 4 nm are significantly larger than the maximum 0.6 nm suggested, but the feature size here is also \~2x larger than the 10 nm features sought by the ITRS; similarly, the defect density is \~10,000 times higher than ITRS goals. However the present samples lack any features to guide alignment, as is the case with graphoepitaxy, which assists in significantly lowering the observed defect density. ## Outline of the analysis The analysis is briefly outlined in, breaking down the ADAblock sequence into eight broad stages. Details of each stage are provided in detail, *vide infra*. The first stage is simply the representation of the original SEM image, which must be smoothed to reduce noise, while retaining all features of interest. Next, the smoothed image is thresholded, in order to produce a binary image from which data like area, perimeter, and shape can be determined. Next, the period and line-widths are calculated, followed by particle analysis to determine the shapes of the binary objects and to classify line and dot features. The line features identified are then isolated and converted into a skeleton, from which the connectivity can be determined. This resulting structure is groomed and then analyzed for defects. Lastly, the data is recorded and confirmation images are produced for user inspection. All stages noted in the text correspond to the stages represented diagrammatically in. ### Stages 1 and 2: Input of original image and smoothing In order to extract information regarding defect density and LER (above,), a number of factors must be taken into account. Starting with the image itself, basic parameters must be adapted to (1) the image resolution \[for instance, determination of how many nanometres are represented by each pixel (nm/px)\]; (2) the contrast of the image, which can vary considerably image-to-image and instrument-to-instrument; (3) image noise, which creates artifacts not inherent to the actual structure under investigation; and (4) the period and line-width of the block copolymer. Given the nature of block copolymer patterns typically observed, certain presumptions about the structures observed within the images can be made. To begin, predominant structures within a given image are primarily limited to dots, lines, and meshes. Classification into the basic families of structures in turn constrains certain shape characteristics for the features. Additionally, the period can be defined for a relatively narrow range (e.g. 20 to 40 nm for the samples described here, but modifiable for a given system), as block copolymer samples for a given image data set can be manually selected to those having similar period values. As preliminary background data, the period of a pattern can be obtained *via* azimuthal averaging of the image’s fast Fourier transform before application of ADAblock. The first item, the image resolution, is frequently embedded within the image’s metadata and hence can be called by the program or input by the operator. The preponderance of our BCP pattern images were obtained using a Hitachi S4800 scanning electron microscope (SEM), which provided information in a legend at the bottom of the image; the consistency of this feature also provided a means for automated extraction of resolution parameters. Combined with its high resolution and high throughput, SEM can be the ideal imaging tool for BCPs, although it does have some drawbacks: For our work, smoothing was necessary due to random noise, charging effects, and edge effects. In the case of SEM images, edges can possess enhanced brightness, and white noise results in speckling of the image with bright and dark pixels. Without some smoothing, such salt-and-pepper noise can result in unwanted extra features. SEM owes much of its brilliance to edge effects, which result in objects protruding from the surface (such as Pt nanowires on Si) appearing much brighter than surrounding substrate. Typically, smoothing images involves trial-and-error, but linking the smoothing to the period of the pattern and the image resolution gives consistent results: Gaussian and/or median filtering are automatically applied with filter radii calculated in proportion to the period of the pattern to avoid under- and over-smoothing. Median filtering typically is best, as it can preserve and even enhance the structure of the line pattern, as shown in ### Stage 3: Threshold to binary In order to analyze the pattern, the two phases, each corresponding to one of the blocks, must be clearly identified and separated by thresholding. This assigns each feature in the image to one of the two phases, referred to herein as positive (i.e. bright) and negative (dark). Contrast enhancement and thresholding typically requires manual intervention as well. For SEM and atomic force microscopy (AFM) images, which are typically used for block copolymer thin films and patterns, as well as helium-ion microscopy images, a bimodal histogram is either typical or attainable given the nature of the pattern. Such a bimodal histogram can occur either globally (i.e., over the whole image) or locally (over smaller sub-regions); a suitable thresholding filter can be applied on either scale. Several “auto-local” thresholding plugins are available for ImageJ; analysis of our images typically works best utilizing an auto-local threshold which applies Otsu’s clustering method locally across the image, however, other thresholds implemented in ImageJ are available options. When the surface is not uniformly covered by features (e.g. featureless regions), however, automated thresholding can result in additional artifacts, hence subsequent steps are taken to remove noise and incorrectly phased features. ### Stage 4: Initial Line-Width Analysis The first pieces of data that must be determined are the dimensions. While period can be readily and automatically measured from azimuthally averaged FFT patterns, line-widths and spacings cannot be derived directly from the image of the BCP nanopattern. Profile plots can make for easy manual measurement of these features when patterns are regular and aligned, but that is not always the case. Knowledge of the dimensions is useful, even necessary, in contexts where the pattern is poorly ordered. Particle analysis can measure the area and perimeter of each particle accurately. While Feret measurements (See) work for simple particles, the tortuous nature of BCP “lines” calls for more nuanced measurement. Imagine a spaghetti noodle shape confined in 2D; there exists a relationship between the perimeter of the noodle’s edge and the area covered by the noodle. Using these easily measurable geometric quantities—particle area and perimeter—the width of lines can be calculated, straightness and degree of branching notwithstanding. Provided there are enough lines available, the particle area plotted as a function of perimeter is linear as shown in ; the slope of the plot is half of the width of the line. For lines without junctions and only uniform tips, perimeter, *P*, can be broken into $$P = 2P_{t} + 2L$$ where *P*_*t* is the perimeter of each tip region and *L* is the length of the main portion of the line. The area can be calculated similarly $$A = 2A_{t} + wL$$ where *w* is the width and *A*_*t* is the area of the line’s tip region. Area as a function of perimeter can be calculated by substituting $$L = \frac{\left( {P - 2P_{t}} \right)}{2}$$ into the area equation, giving $$A(P) = 2A_{t} + \frac{w}{2}\left( {P - 2P_{t}} \right)$$ $$A(P) = \frac{w}{2}P + \left( {2A_{t} - wP_{t}} \right)$$ As previously stated, the slope is *0*.*5w*. For patterns where junctions rather than terminal points are predominant, *P*_*t* becomes zero, and the intercept is positive (due to an additional area term derived from the junction). In practice the contribution of the ends (thus intercepts) is negligible relative to the segment lengths (*L \>\> w*) for images with junctions. With exclusively semicircular terminal points or triangular junctions, one would expect intercepts of *-0*.*785w*^*2* and *0*.*289w*^*2* respectively. (See SI for calculation.) Intermediate combinations can be avoided by temporarily excluding junctions and breaking down the binary pattern into smaller, junctionless particles, as shown in. By excluding the junctions, along with excessively small particles and sections of particles on the edge, a better fit can be obtained, providing a better estimate of the line-width. (See for a schematic depiction of how junction exclusion achieves this.) The value of the intercept in is given as approximately *-200 nm*^*2*, which is reasonably close to the predicted value of -250 nm², detailed in. Repeating the process for the negative phase, separately, gives a measure of the spacing between lines. Summing the two measurements to approximate the period has, in most cases, been found to come within 5% of the period measured by FFT, usually slightly greater. It appears that this discrepancy may be due, in part, to the particle perimeters being larger than non-discrete analogues, and also due to the approximations made herein. Alternatively, knowledge of the line- width and the period would give the line spacing by difference. ### Stage 5: Particle analysis In the course of annealing spin-coated BCP thin films, the pattern may evolve from a dot pattern, or similarly disconnected collection of features, into an array of lines, with numerous defect-rich intermediate states. The binary image can be analyzed using ImageJ’s built-in particle analysis routine to determine characteristics of particle size and shape descriptors such as circularity and Feret measurements, mean pixel values, and relationships to the image boundary for each feature. Particle analysis is done separately for the positive and negative phases in order to access all the features. Particle analysis data is then used to separate dots (or other objects), which cannot be accurately treated as lines, and identifies them as a specific type of defect. It can also provide information on the evolution of particles in the course of the annealing process (e.g. increases in the average size or length of lines). Moreover, the creation of a binary pattern further enables distinction between noise and misclassified particles. ### Stage 6: Skeletonization The most effective means to analyze the topology of lines and meshes is *via* analysing the connectivity of the pattern by creating a skeleton of it. Skeletonization reduces lines or meshes to binary objects which maintains the connectivity of the original by “thinning” the pattern to create a simplified, single-pixel-wide version of the shape, suitable for pixel-by-pixel analysis. Skeletonization algorithms and skeleton analysis has been widely used in other fields to study the topology of structures, from text recognition algorithms in computer science to numerous subfields in biology, including bone analysis (“bonej”), and studying the structure of neurons, as well as for the recognition of typographic characters. In all cases, skeletonization is used to simplify collections of interconnected shapes and objects into networks to study their properties. Although other algorithms do exist, the default technique is implemented in ImageJ: the skeletonize function in ImageJ uses a lookup table to progressively thin the structure based on each pixel’s 3x3 neighbourhood, leaving a 1-pixel wide topological skeleton. At least two other groups have applied skeletonization as a means to interpret BCP thin film patterns. Rehse and coworkers utilized skeletonization of *one phase* of the polymer pattern to study frame-to-frame correlations between junctions as a measure of BCP dynamics; their work followed that of Scherdel and Vigild who used 3D interpenetrating skeletons to describe gyroidal phases. The skeletonization process itself is quite straightforward, as shown in : to skeletonize a binary image, first dots are removed, leaving only line features, then the image is thinned as described above. It is important to skeletonize both phases of the image, so the original binary image is then inverted, dots from the negative phase are removed, and the inverted image skeletonized. Overlaying the skeletons with the binary images shows that skeletonization indeed preserves the connectivity found in the original image. In order to actually access the defects in striped BCP patterns, it is necessary to investigate each phase separately, thus requiring parallel particle analysis and skeletonization of each phase. It is worth noting that for BCPs, certain kinds of defects will prefer one phase to the other, resulting in a surplus of terminal points or junctions in either phase. This tends to limit the frequency of spatially paired defects. shows some examples of this effect. In the images on the left of, there are ample junctions in the positive phase; in the images on the right, there are almost no junctions in the positive phase, despite having more defects overall. Similarly in, there is a greater proportion of terminals in the negative phase than in the corresponding image on the right side. Such features contribute to the topology of the pattern, which can be affected by the means of annealing. ### Stage 7: Groom and analyze This stage is the most complicated, and it is divided up into separate sections, (a)-(d) based upon the type of analysis. *(a) Locating defects*. The ideal, defect-free line pattern derived from lamellar or cylindrical domains of block copolymers, consists of perfectly parallel straight lines extending across the entire substrate without interruption, as by breaks or junctions in the lines. It is with respect to this ideal that topological defects are defined. The analogy between block copolymers and liquid crystals (nematic and lyotropic phases in particular) inspired previous defect analyses utilizing winding numbers to identify and measure topological defects. While this does work in principle, and many previous analyses have utilized it and other defect-detection methods, these methods typically are published without full working details or code. Kléman suggested in 1983 that defects in two-dimensional line patterns could be simplified to junctions, terminal points, and dots, shown in, rather than the more conventional approach using winding numbers to determine the type of defect. While such methods correctly describe the type of defect, high levels of defects and variability of the patterns, including variations in line-width (or LER) can make it difficult to correctly identify and quantify defects. One author noted an order of magnitude change in the density of defects for one particular image depending on the amount of Gaussian filtering applied; the filtering parameters ultimately selected appeared to be arbitrary. Furthermore, images depicting disclinations and dislocations suggest imprecision in the identification of closely associated dislocations. Moreover, while such methods provide the magnitude of each defect, they do not provide information about the connectivity of defects or orientation of surrounding features. Many frequently encountered defect structures are not isolated dislocations (junction-terminal point pairs immediately adjacent) or disclinations (either terminal points or junctions), but are part of more complex defect structures. These structures can be broken down into component dots, junctions, and terminal points, the elementary components of defects shown in. Dots can be determined best using the particle analysis data, so from the skeleton analysis we locate and characterize the junctions and terminal points. At this point in the analysis, the skeleton is a binary object where lines are represented by a series of 2-connected pixels in any of the 8 directions; terminal points are singly-connected pixels; and junctions occur where more than 2 pixels are neighbouring a given pixel. Dots do not, however, always reduce to single-pixel objects and hence they are treated separately. Furthermore, junctions exist in numerous possible configurations, often with multiple (3+)-connected pixels per junction, hence there will not be a one-to-one correspondence between junction pixels and either the number of junctions or junction types. Identification of defects is done in a manner, which is, in essence, analogous to playing the minesweeper-type games: by counting the number of skeleton pixels adjacent to any given pixel that is part of the skeleton, thus providing a connectivity value for that pixel, as shown in. At its simplest, any connection or disconnection that breaks the 2-connected topology of the skeleton, resulting in a new local topology (or connectedness), is a defect with a corresponding value. For junction points (JP), with each additional branch beyond two (which, on its own, would constitute a line without topological defect), the defect increases in magnitude by ½: $$n_{jp} = - ½\left( {B - 2} \right)$$ where B is the number of branches. Typically junctions come with only 3 branches, but 4-way, or even 5-way, intersections can be found, on occasion, between clusters of dots or other complex features. A 4-way intersection would be *n*_*jp* *= -1*, which can be imagined as being derived from two adjacent junctions, each with *n*_*jp* *= -½*, with a common line-segment, where the intervening line segment’s length decreases to zero; the same approach can be generalized for any number of additional line segments (as shown in). Terminal points (TP) possess only one configuration, hence their value is assigned: $$n_{tp} = + ½$$ Dots can be considered as a line with two terminal points, collapsed to a single point (as depicted), hence their value is twice that of a terminal point: $$n_{dot} = 2n_{tp} = 2\left( {+ ½} \right) = + 1$$ Other more complex structures, such as spirals (containing terminal point dislocations) can be counted *via* their component structures in this regard. Large, solid spots in the bright phase or large regions without any pattern (i.e., large spots in the dark phase) possibly formed due wetting (or other causes) may exist, these regions may be treated as dots with radiating arms, however, we found it was more effective to separate the core of the dot prior to skeletonization. The result treats the dot as a kind of enlarged junction, with defects existing only at the periphery. For all defects to be counted, skeletons for both phases must be generated and connectivity analyzed separately. This raises an important point for defect analysis: that defects exist in a particular phase. The phase dependency of defects has not been sufficiently explored, however hints are seen in the literature, as it directly controls the topology of a system. The dual phase analysis brings about an addition rule for description of the system: on average, the sum of all defects in the pattern should be zero. Alternatively, this can be stated that every defect is “paired”, hence for every junction there is a terminal point; Moreover, for every dot, there will be two junctions: consider this an analogue of a unit cell, as two defect pairs are produced if spontaneously generated and two are required to cancel out through annihilation. Typically there exists a small imbalance between the two measures, which others have observed as well. The pairing of defects does not imply however that the number of defects in the two phases of the block copolymer will be equal. Skeletons also provide a description of the connectivity of defects, which merits further exploration. Defects can also be associated with particles using this method, but perhaps the greatest benefit is derived from the ability to search for and positively identify particular clusters of defects. One such example is an H-junction, which results from a break in the line or a bridging of two adjacent lines, shown in These junctions are supposedly not the result of a defect in the actual thin film structure, but result from (a) incomplete metallization or other means of pattern transfer, (b) image noise, or (c) the smoothing-thresholding process. Hence it may be prudent to recognize them and count them separately or to “correct” such errors in the binary image itself. *(b) Grooming the Skeleton*. Grooming the skeleton consists of trimming away short branches, which may result as artifacts from small “bumps” on the edge of a line. With the dimensions determined in Stage 5, we can create a metric to selectively prune away any branches resulting from variations in line-width or simply from sharp points or edge effects that can influence the skeletonization algorithm. It may be a point for philosophical debate what constitutes a branch, justifying a junction and terminal point, but objectivity can be introduced by basing the grooming procedure on the measured LER. For this purpose, any end point separated from a junction by less than 1.5 × line-width (for a given phase) is considered roughness, rather than an additional defect pair, and is hence pruned, as shown by the example in. Because any image represents a finite sample of a larger structure, defects at image edges must be carefully treated. Depending on the resolution of the image and the domain size of the block copolymer, these can for smaller images, represent a significant fraction of defects; additionally, in otherwise low- defect patterns, features cut off at the edge may appear as additional defects. In particular, three rules must be applied: Any “dot” (or sufficiently small object without junctions) touching 2 edges is not a defect. (See) Lines that run roughly parallel to the edge, touching at all times, are not defects. Lines that terminate at the edge of an image are not defects, as it is not a true terminal point. The third rule requires some manipulation of skeleton points & component terminal points near edges, as ImageJ’s native skeletonize algorithm can produce limited edge artifacts. What these rules do not address is particulate matter; other analysis methods tend to default to manual identification or require equipment unavailable to most researchers. *(c) Line-Edge Roughness and Line-Width Roughness*. One of the chief questions posed for BCP lithography is whether lines can be produced with sufficient uniformity and with smooth edges. LER measures the variation in the position of the edge of a line, which can have different frequency components, leading to undulation of the edge and variation in the width of the line, or LWR. The variation in position is measured as the standard deviation in the position of the edge, and LER is reported as 3σ. Such variations are deleterious for circuit elements: For transistor gate features with widths \< 85 nm, line roughness causes significant variations in the off-current, as well as affecting threshold voltages. For nanometre-scale interconnects, line roughness increases both resistance and capacitance, resulting in degraded transistor performance. The line roughness of block copolymer nanostructures has been considered theoretically and has been shown to depend on χN,\[–\] and polymer polydispersity; results have suggested that the Flory-Huggins χ parameter may need to be increased by a factor of 3 to 4, relative to that of PS-*b*-PMMA, in order to decrease LER sufficiently to accommodate ITRS targets. It has been specifically noted that there are few reports on the topic of LER/LWR in the literature; typically, the actual position of the edge is measured relative to the ideal or average edge position for straight or aligned lithographic patterns. In order to achieve the same measurements for block copolymers, films aligned *via* graphoepitaxy would typically be required in order to have linear lines representing ideal edges. However, we and others have taken the approach of measuring LER for unaligned patterns. One may measure edge positions relative to the centre of the line, rather than with respect to a linear ideal edge position; the standard deviation in the edge position will be the same either way. As lines get narrower, however, the influence of pixel position can begin to slightly increase the measured LER, up to 0.5 nm in our previous work using high resolution (ca. 100,000x) BCP patterns. We mitigate this, in part, by smoothing both the centre line of the skeleton and the outer edge, while constraining the positions of the edge points. Edge-to-skeleton distances are determined for all points on the smoothed line edge, matching with the nearest points (shown) on the smoothed skeleton line which satisfy: $$\left( {x_{edge} - x_{skel}} \right) + slope_{skel}\left( {y_{edge} - y_{skel}} \right) = 0$$ As derived from the dot product of the vector on the edge-to-skeleton distance and the orthogonal vector *(1*, *slope)* of the skeleton at that point, an interpolated point on the skeleton can be obtained (shown in). Line-width measurements can be made in conjunction with edge-to-skeleton measurements by finding a line segment on the opposing edge, which is intersected by the vector made between the edge point and skeleton point of the previous step (shown). The solution exists at a point on the line segment formed by the vector between the edge *(x*_*edge*, *y*_*edge**)* and the skeleton *(x*_*skel*, *y*_*skel**)* is scaled by a factor, *a*, and on the line segment formed by the vector between two consecutive points on the transverse edge *(x*_*trans1*, *y*_*trans1**) & (x*_*trans2*, *y*_*trans2**)*, scaled by a factor, *b* (shown). Provided that the two vectors are not parallel, the equations for the scalars, *a* and *b*, are: $$d = ~(x_{trans2} - x_{trans1})\left( {y_{skel} - y_{edge}} \right) - (x_{skel} - x_{edge})\left( {y_{trans2} - y_{trans1}} \right)$$ $$a = d^{- 1}\left( {\left( {x_{edge} - x_{trans1}} \right)\left( {y_{trans2} - y_{trans1}} \right) - \left( {y_{edge} - y_{trans1}} \right)\left( {x_{trans2} - x_{trans1}} \right)} \right)$$ $$b = d^{- 1}\left( {\left( {x_{edge} - x_{trans1}} \right)\left( {y_{skel} - y_{edge}} \right) - \left( {y_{edge} - y_{trans1}} \right)\left( {x_{skel} - x_{edge}} \right)} \right)$$ An intersection is considered valid when *1 \< a \< 4*, indicating that the side opposite would have a width ranging from 0 to 3 times the width of the first side. The limit, *a \< 4*, prevents identification of points on parallel segments, as with a hairpin, from being identified as valid; typically the period is on the order of 2 times the width of a given line, hence 4 times the half-width of a line. In practice, the values of *a* are in the range *1*.*5 \< a \< 2*.*5*, as can be seen typified in *via* the histograms. The second limit for valid points is that *0 ≤ b ≥ 1*, which ensures that the point of intersection is within the line segment formed by the two consecutive edge points. In order to obtain reasonable measurements of LER and LWR, the blocky structures of binary lines and skeletons need to be smoothed. The smoothing process, which we have utilized here, involves 4 stages: Centring of the skeleton points by adding 0.5 px to each x and y coordinate. This accounts for the slight truncation from the skeletonization process and makes the edge-to-skeleton distances more equidistant on each side. Shifting all edge points to the midpoints between consecutive points. This averaging reduces roughness introduced by the shape of individual pixels. Smoothing the skeleton by iteratively averaging the positions of points, while limiting the displacement to within 0.25 pixels. This provides a smooth, continuous, reasonably centred skeleton line. Smoothing the edges likewise provides a smooth edge while maintaining the shape and deviations in width, from which roughness can be measured. shows the data for a single line as it is modified by each of these four smoothing processes (A, B, C, D). By the fourth stage (D), the data shows considerably less noise. In particular, the histograms of edge widths, depicting which edge is further from the skeleton, for each point on the edge, begins to approach a normal distribution, as one would expect for a line with random variations in width. Visually, the line becomes sufficiently smooth that pixels are no longer apparent, while variations in width are in keeping with the original image, and the sequential widths and edge positions measured do not have large point-to-point changes in displacement. While the skeletonization algorithm is largely effective in finding the centre line, it is imperfect. In particular for lines with pixelated widths less than 7 pixels, the centre will tend to be skewed preferentially depending on the orientation of the line. However this does not affect LWR measurements and smoothing does help to limit the impact on LER. In order for BCPs to be relevant in industrial manufacturing, they must achieve a low frequency LWR (3σ) of 1.1 nm on features 16 nm wide; in order to “significantly exceed” conventional lithography, the patterns would need to be better than 0.6 nm LWR on features 9 nm wide. Presently our best measured samples have a LER (3σ) of \~ 2 to 3 nm, however, no aspect of the process has, as of yet, been explored with respect to minimizing LER or LWR. To avoid the local effects of junctions and to increase the speed of the calculation, the lines are modified, as shown in, to render all lines junction-free. Additionally, points where lines contact image edges are selectively modified, erasing large contacts, to prevent any effects of the image edge. *(d) Correlation Lengths & Order Parameters*. Correlation lengths (or orientational persistence lengths) are typically calculated for large images, often with low resolution (pixels/nm), by subdividing the area into overlapping squares, for which azimuthal angles are derived from two-dimensional FFTs of each region. Lack of clarity for such images sometimes necessitates filtering in order to avoid disordered regions. In this work we implemented an alternative means of determining the 2D correlation function using the skeletonized lines. Skeletons are groomed to remove junctions and loops are broken to provide isolated lines. Orientation along the skeletonized lines can be calculated using a rolling average of each line’s tangent to provide smoothly varying angles along the lines. In a typical image, there can be over 20000 points in the lines; calculating the correlation length using every point is feasible, however for expediency, the set of points can be downsampled or randomly sampled to a smaller set of \~ 4000 points, which provides faster calculation with minimal trade-off in terms of accuracy. From the set of orientation angles, φ(**r**), the correlation function, C(**r**-**r**’), can be calculated. <img src="info:doi/10.1371/journal.pone.0133088.e013" id="pone.0133088.e013g" /> C ( r − r ′ ) = 〈 c o s \[ 2 { φ ( r ) − φ ( r ' ) } \] 〉 Advantages of this method include ease of applicability to higher resolution, smaller-area images and images with disordered regions where, due to defects, line segments are particularly short, and φ(**r**) might not be determinable *via* FFT. This is demonstrated in. The correlation function is fit using an exponential function, $$C\left( {r - r^{\prime}} \right) = exp^{\frac{- r}{\kappa}}$$ where κ is the *correlation length*, a characteristic measure of the degree of ordering in the film, which describes the average distance over which orientational order is preserved. The correlation length should be proportionate to the grain size, as illustrated by the circles in, which are approximately keeping in proportion with the domains visible in the orientationally-colour-mapped pattern image. However the circles are unquestionably smaller than the observed domains. One disadvantage of this method of determining κ *via* skeletonization is that one observes a periodic variation (corresponding to the periodicity of the pattern) in the correlation function, as shown in. This periodic variation is a result of features separated by non-integer line spacings tending toward greater disorder than points separated by integer spacings. This appears to be due to influence by neighbouring defects. The large undulation in the curve can be partially compensated by using both the positive phase and negative phase skeletons (thus reducing the period and amplitude of the variation, however exclusion of non-line areas may be necessary), by binning measurements, as is typically done in FFT-based methods, or by smoothing, as we apply in the algorithm. Herman’s orientational parameter, S, gives a measure of how uniformly oriented the lines within an image frame are. It can also be readily calculated using the set of orientational data: $$S_{2D}\left\{ {0,1} \right\} = 2\left\lbrack {cos\left( \varphi \right)} \right\rbrack^{2} - 1$$ The reference angle can be set as the average orientation for the whole image, thus giving the best orientation parameter for a disordered image. Because it is widely used, we implemented this calculation into our code, however, Herman’s orientation parameter tends to be less useful than the correlation length, as it can be *significantly* influenced by the size of the area sampled. That is to say one can typically choose a sample area small enough to give *S*_*2D* *≅ 1* (perfect net order) or an area large enough to give *S*_*2D* *≅ 0* (no net order). The code may, however, be adapted to set an angle where a particular direction is induced *via* processes such as directional annealing or graphoepitaxy; in such cases, *S*_*2D* *= -0*.*5* is a possibility for samples where the line orientation is orthogonal to the desired orientation. Finally, this skeleton-based approach facilitates generation of pseudo-coloured orientation maps, as in, which also avoid grain-edge averaging problems exhibited with other methods. Such images may assist researchers in qualitatively grasping the orientational ordering in their system. Such visual checks, can provide researchers with an accessible means of confirming numeric results, as it allows for a qualitative, direct measure of grain size on the image. ### Stage 8: Output and confirmation images Finally, as a result of these considerations, we seek to provide self-assurance and quality control by creating confirmation images, wherein features described numerically are mapped onto real images to provide visual feedback of the accuracy of the measurement, as shown in , which shows the defects found alongside the associated SEM images. This step is ultimately the means to determine whether the defects identified are (1) a true representation of the pattern and (2) are in the correct location. Such images of pattern orientation, line roughness, defects, and thresholding provide visual confirmation that all stages of the analysis proceeded correctly. Specifically, one can check simultaneously whether the thresholding, connectivity, grooming, and defect identification have all functioned as expected. Such visual feedback also lets researchers, particularly those presently involved in synthetic work, to tangibly grasp the important aspects of the pattern quality. By encoding the information spatially with colours and shapes rather than relying purely on the abstraction of defect densities and correlation lengths, ADAblock’s visual feedback can function as a guiding indicator for selection of optimum structures and conditions. The data output, both numerical and visual, make it possible to engage in exploratory data analysis to discover new trends, motifs, and outliers in the data available, as demonstrated in and later in Figs and. ## Application of ADAblock In order to demonstrate the utility and versatility of this application, two different scenarios and questions are posed. First, what is the effect of image resolution, and the area sampled, on the measured defect densities, LER, and other parameters for patterns derived from self-assembled BCP thin films? Secondly, what can we learn from investigating the data provided by these samples, by examining the relationships between different features, to identify features that warrant further investigation—and what does this suggest about the resulting properties of a self-assembled BCP film? ### Effect of Resolution and Sampling Area When measuring defect densities, correlation lengths, LER, and LWR, the area sampled and the resolution can potentially affect the measured results. Ideally, for any measurement, the effect of sample size must be analyzed and understood in order to obtain reliable results. To develop a general sense of how this and different polymer sizes are affected in the analysis, we annealed 5 different polymer types, each with approximately ideal thicknesses, for 20 minutes at 200°C and imaged the resulting metallized patterns at different magnifications. The effect of resolution in the LWR measurements in appears to be minimal, although there is a slight downward trend with increasing resolution (smaller image area) for the two smallest polymers, where the LWR (1σ) values decrease from 0.18 to 0.14. The increase in LWR is primarily observed for those samples with the smallest period, which would likely be on account of pixelation of the lines, as suggested by. A confounding effect may also result from the decreased length of line sampled for images of higher resolution. LER data, on the other hand, shows a more consistent trend of decreasing LER with increasing resolution in. LER is likely more affected by pixelation due to the inability of the skeletonization process to precisely locate the line center, in particular when line-widths are a small, even number of pixels. In contrast, the line-width is not strongly constrained by the determination of the line centre. The magnitude of the decrease (-0.03 to -0.05 pixels) here is still small, given that image area changes by a factor of up to 100. Sampling effects can be observed in the measurement of defect pair density at various resolutions in. High resolution images, depending on the distribution of defects, can completely avoid defects or oversample them. Here the smallest BCP (23.6k-b-10.4k) is most affected, due to having a larger grain size. The same effect can be observed for correlation length measurements in, although this affects all of the polymers. In order for the correlation length measurement to be meaningful, the measured value should be significantly shorter than the dimensions of the image. The decrease in the average measured correlation length as a function of the image area suggests that one may be able to estimate the true value based on the size of the image. The plot of 23.6k-*b*-10.4k is particularly telling because it shows the effect of sampling within a single grain or few grains (at low resolution) and the sudden decrease once more grains become involved. The limitation of large grains may be partly avoided by using automated data collection, combined image stitching, which has been demonstrated to be effective for imaging large areas with electron microscopy, however as ordering approaches perfection, grain sizes become infinite, and the correlation function will approach unity. ### Feature Relationships In order to derive lessons from the data, we undertake a form of exploratory data analysis to chart the relationships of different parameters observed. In particular, whether parameters such as LER and LWR are independent of the feature size, and how line-widths, polymers, and periods have a simple relationship. Taking all of the data (across resolutions) for each polymer, we note that as a proportion of the line-widths, the standard deviations in the edge position (LER, 1σ) and line-width (LWR, 1σ) stay constant, about 10% and 16% respectively, indicating that the LER and LWR scale with the line-width dimension of the polymer, as shown in. The set point may be a property of a given BCP’s Flory-Huggins parameter, indicating a higher χ required. However we must caution that other factors, such as the processing, metallization, plasma treatment, and lack of alignment are convoluted with the roughness inherent to the polymer, preventing a direct conclusion. However this method should enable comparison between polymer templates and patterns translated from the BCP *via* etching or other means. The values observed here would however all exceed LWR targets set by the ITRS for LWR (3σ) of less than 6%: 1.1 nm for patterns with 18 nm feature size; or \<0.6 nm for patterns with 10 nm feature size. For aligned patterns, solvent annealed with water as a co-solvent, we have observed significantly better LER and LWR values. We hypothesize that it may be the result of the water selectively partitioning inside of the P2VP block during annealing, resulting in a higher effective χ, leading to a smoother interface than we attain here with thermal annealing. Line-width in shows the expected relationship of being proportionate to the period, although there does appear to be a greater spread in the width of lines than in the FFT-measured periods. This is likely an effect of thresholding, which needs to be done relative to each image. It may be possible for a specific polymer or a series of images to constrain the threshold, as a fraction of area, in order to obtain a narrower distribution of line-widths. ### Limitations of the code As with any programmed analysis, there are drawbacks and trade-offs made in analysis to optimize for speed or accuracy. The approximations we implemented are one reason that necessitates a full sharing of the code. ImageJ’s macro language is interpreted, hence it is slower in processing compared to plugins or other compiled programs. It is, however, easily edited and modified, which enables adaptation where modification may be required. The code was written so that it can be operated in a batch mode to process a folder of images, meaning that a series of images can be processed overnight, or while attending to other tasks. It should be cautioned that in the present state, as ADAblock continues to be developed, the code may produce a reproducible error for \~4% of images at present. Further refinement should reduce this error rate, but at present may limit a series from being completed. With manual intervention, however, the image can be skipped, or the settings modified, and the queue re-continued. Typically an image with dimensions of 1280 x 896 pixels (the default of our SEM, for example) requires \~7 minutes to process when run on the standard personal computers that we used for testing. Higher pixel-resolution (*e*.*g*. 2560 x 1792) images require more time to process, roughly in proportion to the number of pixels. Given the automated nature of the program, it’s possible to run a queue of images overnight, rendering the increased processing time irrelevant. In addition to images showing the locations of defects, the code saves several check images to act as references to help determine whether any errors have taken place or other undesirable operations. Consequently, \~ 16 MB is recorded to the disk for each image processed, as presently conFigd, although non-graphic data only accounts for less than 300 kB. (See for a list of files output by the program.) # Conclusions We have developed a facile, automated, and reliable analysis for striped patterns derived from the self-assembly of BCP thin films, that integrates both conventional and newly developed techniques. This analysis is done in order to quantify defects and their types using a skeletonization-based method; to measure line-edge roughness; and to calculate Herman’s order parameter and the correlation length in a novel fashion, based upon the skeletonized structure. Moreover, the skeletonized structure provides information about the connectivity of patterns. We expect that this will be of use to others carrying out annealing studies and preliminary characterizations of novel self-assembling polymeric materials. Finally, for 5 block copolymers of similar composition, we have found the metallized patterns to have LER and LWR in roughly constant proportion to the line-width. Ultimately, no one measurement provides a “complete description” of pattern quality; typically they are complementary. Hence this work represents an attempt to broaden the scope of analysis and to make tools which may not be readily accessible to all. Additionally having shared protocols, or at least protocols derived from a common origin, we might be able to standardize a broad toolset, providing consistent analysis *via* fully shared code. We hope this aids comparisons between polymers, between papers, and between scientists seeking to understand the characteristics of block copolymers, and in addressing the numerous critical issues associated with block copolymer lithography. # Materials and Methods PS-*b*-P2VP block copolymers were obtained from Polymer Source Inc., QC, in weight-averaged molecular weights of 23.6k-*b*-10.4k, 32.5k-*b*-12k, 34k-*b*-18k, 44k-*b*-18.5k, and 50k-*b*-16.5k and all with polydispersity below 1.1. Toluene was purchased from Fisher Scientific; concentrated H₂SO₄ from Caledon Laboratories; 30% H₂O₂(aq) from Sigma-Aldrich; and Na₂PtCl₄·xH₂O from Strem Chemicals. Silicon wafers were obtained from University Wafer. ## Substrate Preparation 100 mm diameter, single-side polished silicon wafers were diced into squares with dimensions 1.0 cm x 1.0 cm. Prior to cleaning, substrates were scribed, on the unpolished side, with a diamond-tip, to mark the identity of each substrate as part of a set of 10. The samples were then immersed in methanol and sonicated for 15 minutes in glass beakers. Next, after rinsing each substrate square in a series of beakers filled with 18.2 MΩ∙cm water, the substrates were placed polished-side-up in PTFE beakers, and immersed in 6.0 mL of concentrated H₂SO₄, to which was added 2.0 mL of 30% H₂O₂, before placing the beaker to stand in an 80°C hot water bath for 20 minutes. The piranha solution was then decanted to a glass flask to cool prior to neutralization. Following several rinses with water, the substrates were immersed in aqueous 1% NH₄OH solution for 5 minutes to remove any surface sulfonate groups, prior to a final decant and replacement of the solution with 18 MΩ∙cm water. Typically samples were stored immersed in water with the top sealed with paraffin wax. ## Solutions & Spin Coating Immediately prior to spin coating, each wafer was dried under a nitrogen stream. Once dry, the sample was analyzed using fixed-angle, single-wavelength ellipsometry (632.8 nm) to determine the thickness of the thermal oxide at the center; typically 2 nm. Spin coating was carried out under argon or nitrogen gas. Each substrate’s polished side was evenly coated with 10 μL of 10–15 g/L BCP solution; any bubbles were manually removed; then the substrates were spun for up to 15 s, between 3000 rpm and 4000 rpm, with an initial acceleration of 1500 rpm/s. Following this, the film was reanalyzed by ellipsometry, prior to quartering the sample and annealing. ## Annealing Thermal annealing was carried out in ambient atmosphere on a hotplate covered with a thin aluminum sheet. Temperature was monitored directly at the wafer using an OSENSA fiber-optic fluorescence-based temperature probe. For the thickness measurements and for the comparison of the 5 polymers, the substrates were annealed for 20 minutes at 200°C. ## Metallization A solution of 20 mM Na₂PtCl₄ in 0.9 M HCl_(aq) was used for metallizing PS-*b*-P2VP samples. Samples were submerged for at least 2–3 hours prior to removal and rinsing with 18.2 MΩ∙cm water. ## Plasma processing Following metallization, sample sets were placed together in a plasma chamber, and the chamber was evacuated to \< 200 mTorr to remove contaminant gases or adsorbates. Finally, O₂ gas was leaked into the chamber to a pressure of \~ 750 mTorr. The RF coils were then energized and a faint lavender-blue O₂ plasma was maintained for \~ 60 s (depending on the film thickness) to etch the organic materials from the substrate. Finally, samples were imaged using a Hitachi S-4800 scanning electron microscope, sampling regions near the centre of each substrate. ## Computation For image analysis, ImageJ, version 1.49 and above, was used. It is freely available at <http://imagej.nih.gov/ij/>. The code for performing the analyses is available on our institutional repository; updated versions will be available on GitHub. Python scripts used in preparing the data shown here are also available to assist with processing and plotting output from multiple runs. They are available under an MIT license, allowing users to freely copy, redistribute, and modify the code. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JNM. Performed the experiments: JNM. Analyzed the data: JNM KDH JMB. Contributed reagents/materials/analysis tools: JNM JMB. Wrote the paper: JNM JMB KDH. Wrote and developed the software used in analysis: JNM Contributed concepts to planning software: JNM KDH JMB.

# Introduction *Helicobacter pylori* is a spiral-shaped, microaerophilic, gram-negative bacterium that inhabits the human stomach. Depending on the socioeconomic status of the country, the prevalence of infection varies from 40 to over 80% of the population, with higher rates for developing countries. Infection is usually acquired during childhood and always elicits an acute immune response which, in the absence of effective treatment, persists throughout the patient's life, resulting in chronic gastritis. Although this condition may be asymptomatic, some patients develop dyspeptic symptoms, *i.e.*, the so-called non-ulcer dyspepsia (NUD). In about 15% of infected patients, gastritis may progress further to severe gastric diseases, namely peptic ulcer disease (PUD) and gastric cancer (GC). PUD is a multifactorial disease which leads to considerable patient morbidity and mortality. It is considered to be a disease of adulthood being its progression highly dependent on environmental factors, such as the use of non- steroid anti-inflammatory drugs (NSAID), alcohol consumption, diet, smoking and stress. Some viral infections (*e.g*. cytomegalovirus and herpes simplex infections), Crohn's disease and syndromes where an increase in acid secretion occurs (*e.g.*, Zollinger-Ellison syndrome), may also contribute to the onset of PUD. However, a long-term infection with *H. pylori* is considered as the major causative factor for PUD. This is easily demonstrated by the high number of patients infected with *H. pylori* among those who are affected by PUD, and by the fact that, with the successful eradication of the bacteria, non-NSAID- related PUD is healed and rarely recurs. PUD can be subdivided into duodenal ulcer (DU) and gastric ulcer (GU) diseases which are considered divergent. DU is basically a duodenal acid injury that results from acid hypersecretion due to infection of the antrum by *H. pylori*. In contrast, GU is usually associated with acid hyposecretion that results from the gastric atrophy caused by a proximal spread of the infection, and hence, of inflammation. Supporting differences in the pathogenesis of these two clinical outcomes, the prevalence of *H. pylori* infection is higher than 95% in DU cases and around 60% to 80% in GU. The development of PUD in *H. pylori* infected children is a very rare event and usually occurs soon after infection. In Portugal, for example, only about 2% of the estimated 40% of the Portuguese infected children suffer from this disease. Besides genetic susceptibility, the aforementioned environmental factors should have a minor influence on the pathogenesis of PUD in infected children. Therefore, we believe that the virulence factors of the implicated strain must play a crucial role in the onset of PUD in children. We have reported some data on genetic studies of clinical isolates which show the association of some bacterial genes with the development of PUD in infected children. Indeed, in addition to the most well known virulence factors, namely the *cytotoxin- associated gene A* (*cagA*) gene, the “on” genotype (*i.e.*, functional) of the *outer inflammatory protein A* (*oipA*) gene, the s1/m1 allele of the *vacuolating cytotoxin gene* (*vacA*), and the *A2* allele of the *blood group antigen binding adhesin* (*BabA*) gene, we found that *homB* and *jhp562* genes are closely associated with PUD in children and may be useful in determining the clinical outcome of infection. HomB protein is an antigenic outer membrane protein which was shown to be involved in the host inflammatory response, inducing the secretion of interleukin-8 (IL8), and in the adhesion of *H. pylori* to gastric epithelial cells. The *jhp562* gene encodes a glycosyltransferase involved in the synthesis of lipopolysaccharide and may also be implicated in the regulation of Lewis antigen expression. The presence of the triple genotype *cagA*, *jhp562* and *homB* in *H. pylori* strains provides a good discriminatory basis to distinguish PUD and NUD outcomes in infected children. Motivated by these findings, we further characterized five *H. pylori* strains, isolated from Portuguese children with PUD, all positive for *cagA*, *jhp562* and *homB* genes. Their virulence profile was first compared to that of five other clinical isolates collected from children with NUD, by co-culture assays with the NCI-N87 cell line (American Type Culture Collection (ATCC) CRL 5822). We report here the differences that we found in the impact of these two groups of *H. pylori* strains on the viability, phenotype and cytoskeleton organization of the NCI-N87 cells. Moreover, in order to understand the molecular mechanisms that underlie their more pathogenic phenotype, we compared the proteomes of the ulcerogenic *H. pylori* strains with those of the NUD strains. # Results Ten *H. pylori* strains isolated from Portuguese pediatric patients, five suffering from NUD (mean age 10.2 years, range 7 to 14) and five other affected by PUD (mean age 11.8 years, range 10 to 15, having no other etiology for the disease), were carefully selected to be genetically homogeneous within the respective group. The difference between the mean ages of these two groups was not statistically significant. According to their previously reported genotype, all PUD strains were positive for some important virulence factors associated with PUD in children. These include *cagA*, *vacA* s1 (*i.e.*, the toxic allele), *oipA* “on”, *homB* and *jhp562* (with the exception of strain 1846/05) genes. In contrast, all of the NUD strains were *cagA*, *homB* and *jhp562* negative and carried the *oipA* “off” status and the *vacA* s2 (*i.e.*, the non- toxic allele) gene. ## Co-culture assays To clarify whether the virulence of the strains was the determining factor for the development of PUD in children from whom they were collected, we compared their impact on human gastric epithelial cells with that of NUD strains, by *in vitro* infection experiments. These experiments were carried out using the NCI-N87 cell line, known for its unique differentiation status as it is able to form coherent monolayers, expressing the correct cellular distribution of both *E*-cadherin and ZO-1 junction proteins. Confluent NCI-N87 cells monolayers were thus incubated in parallel with two different pools of *H. pylori* strains, one including the five PUD associated strains, and another including the five NUD strains, both at a multiplicity of infection (MOI) of 100. Under our experimental co-culture conditions, the viability of NCI-N87 cells was clearly reduced over time with both pools of *H. pylori* strains. However, this effect was more pronounced in the first hours following infection with the pool of PUD- associated strains. Indeed, cell viability was significantly lower (*p*\<0.05) in NCI-N87 cells infected with PUD-*H. pylori* strains at 1 h (89.90%±6.23) and 12 h post-infection (73.14%±8.96), compared to cells infected with NUD strains (99.80%±3.52 and 89.94%±6.70, respectively). At 24 h post-infection, the viability of the epithelial cells was very low in both cases, 5.40%±4.51 and 10.78%±11.91, for cells infected with NUD and PUD strains, respectively. However, the morphological analysis of the remaining cells showed dramatic differences in the impact of each pool of *H. pylori* strains. Indeed, light microscopy observations revealed much more pronounced cell damage in NCI-N87 cells infected with the pool of PUD-*H. pylori* strains. Moreover, the pool of PUD strains caused the total destabilization of the NCI-N87 cells' microtubule network, as demonstrated by the immunocytochemical assays , which seemed almost unchanged in cells infected with NUD-*H. pylori* strains. In addition, epithelial cells infected with the PUD-*H. pylori* strains showed a drastic reduction in the cytoplasmic Periodic Acid Schiff (PAS) reactivity, which was supported by the destruction of the cell membrane observed with hematoxylin and PAS double staining. We also observed that the nuclear envelope of the NCI-N87 cells infected with the *H. pylori* strains associated with PUD was undisrupted, despite having a very irregular appearance, suggesting cell death by necrosis. In contrast, NCI-N87 cells co-cultured with the NUD-associated *H. pylori* strains showed an accumulation of extracellular vesicles positive for PAS (dark arrow) suggesting the expected exocytosis of intracellular mucus-containing vesicles, also observed in the non-infected control (data not shown). After 24 h of co-culture, NUD strains maintained their bacillary shape and were adherent to the epithelial cell surface, however, PUD strains acquired a coccoidal shape which was already observed at 12 h post infection (data not shown). ## H. pylori proteome profile Total soluble protein extracts from each of the 10 *H. pylori* strains were resolved by 2-dimensional gel electrophoresis (2DE), first in a non-linear pH range of 3–11 and then in a 7–16% (w/v) SDS-PAGE. After Coomassie Brilliant Blue (CBB) staining, the digitalized images of the gels were analyzed using the ImageMaster™ 2D Platinum software. Protein spots were automatically detected and manually corrected and the best resolved 2DE-gel, that of NUD-associated *H. pylori* strain 655/99, was used as the reference map during computer assisted analysis. After automatic matching, followed by a manual correction, gels were separated into two classes, NUD and PUD, according to the pathology with which the strain was associated. Gel-to-gel matching was confirmed by MS identification of matched spots from different 2DE gels. The 2DE maps obtained for each group of strains were found to be highly reproducible and consistent. No statistically difference was observed for the number of detected spots (on average 485±77 *vs.* 443±43 in NUD and PUD 2DE gels, respectively) between these two classes. ## Comparative proteomic analysis In order to search for a protein signature for pediatric *H. pylori* PUD- associated strains, a computer-assisted analysis was performed by statistical evaluation of changes in the % Vol, *i.e.*, the normalized volume, for each protein spot within the two disease-associated classes of 2DE gels. As a result, irrelevant variations between images were eliminated. From this analysis, we observed that the proteome of the *H. pylori* strain 499/02, the unique strain associated with GU, was not fully consistent with the proteome of the other four strains included in the group of PUD related strains isolated from children with DU. Therefore, from this point onward the strain 499/02 was analyzed separately. lists the identification by peptide mass fingerprinting (PMF) analysis and/or by comparison with 2DE databases of 26 protein spots detected as being differentially expressed: 24 of them were statistical significant (*p*\<0.05) between the NUD and DU classes of strains; and one spot detected at a different position within the 2DE-gels. These include 15 proteins which were highly abundant in DU-associated *H. pylori* strains (solid arrows in ; upward arrows), when compared to NUD strains: flagellar hook protein (FlgE, spot 2) and flagellin A (FlaA, spot 3), both components of the bacteria's flagella; the heat shock protein B (HspB, spot 4), a homolog of Hsp60 involved in protein folding; both beta and alpha subunits of urease (UreB, spot 5, and UreA, spot 20), which catalyzes the hydrolysis of urea into ammonia and CO₂; phosphoglycerate kinase (Pgk, spot 12), 3-dehydroquinate dehydratase (AroQ, spot 22) and flavodoxin A (FldA, spot 25), involved in bacteria metabolic cycles; putative aldo-keto reductase (spot 15), a protein of the cell detoxification system; an isoform of UPF0174 protein HP_1588 (spot 17), a protein whose function is unknown; the non-heme iron-containing ferritin (Pfr, spot 23), involved in iron metabolism; the virulence factor neutrophil activating protein (NapA, spot 26); and the HPG27_1480 protein (spot 27), homologous to the HP1542 protein which has a putative function in cell shape. Interestingly, a thirteenth protein, the hypothetical protein HPAG1_1081 (spot 13), was detected exclusively in the DU- associated *H. pylori* strains. This protein is a homolog of HP1143, another protein with a putative function in cell shape. Further experiments are needed to clarify whether the expression of this protein is unique to the DU strains, or it also occurs in NUD and GU strains but at levels below our detection threshold. As previously shown, only the PUD-*H. pylori* strains were *cagA* positive and this was confirmed by the total absence of CagA protein in the 2DE maps of the NUD-strains. This protein corresponds to spot 1 in the PUD-2DE maps. Amongst the group of differentially expressed proteins, we found the following 11 protein spots to be less abundant in DU-associated *H. pylori* strains (dotted arrows in ; downward arrows), when compared with NUD strains: hydantoin utilization protein A (HyuA, spot 6), aspartate ammonia-lyase (AspA, spot 11), succinyl-CoA-transferase (SCOT) subunits A and B (ScoA and ScoB, spots 19 and 21, respectively), pyruvate flavodoxin oxidoreductase, gamma subunit (Porγ, spot 24), proteins involved in bacteria metabolic cycles; ribosomal protein S1 (RpsA, spot 8) and cysteinyl-tRNA synthetase (CysS, spot 10), proteins involved in protein biosynthesis; catalase (KatA, spot 9), involved in the cell detoxification system; HELPY_0944 (spot 16), homologous to HP0958, which is a protein with a putative function in cell motility; and another isoform of UPF0174 protein HP_1588 (spot 18), a protein whose function is unknown. Spot 7, which we were unable to identify, was only detected in NUD strains, with the exception of the *H. pylori* strain 228/99. As before, we cannot say that the expression of this protein is completely absent in DU strains, rather that its expression levels were below our detection threshold. Besides differences in protein expression between these two groups of strains, we detected a difference in the position of the protein spot 14 within the 2DE- gel that was identified as being elongation factor Ts (EF-Ts). Indeed, in the 2DE-maps of the DU strains, this protein spot was shifted to the right relative to its position in the NUD 2DE-maps. Interestingly, we found differences in the abundance of some of these protein spots in GU-associated *H. pylori* strain 499/02, when compared to that of the DU strains. Showing the same tendency of that observed in DU strains compared to NUD strains, we found FlgE (spot 2), FlaA (spot 3), Pgk (spot 12) and aroQ (spot 22). Showing the opposite tendency of that observed for DU strains compared to NUD strains, and with abundance levels similar to those observed in the latter group of strains, we found UreB (spot 5), HPAG1_1081 (spot 13), putative aldo- keto reductase (spot 15), one isoform of the HP_1588 (spot 17), and UreA (spot 20). CagA (spot 1) was less abundant in the GU strain. All of the other protein spots followed the same pattern of expression observed for the *H. pylori* strains associated with DU. Although the MOWSE score in the MS identification did not reach statistical significance (*p*\>0.05) for the CagA (spot 1), CysS (spot 10), AspA (spot 11), ScoA (spot 19), Pfr (spot 23) and HPG27_1480 (spot 27) protein spots, we are confident in our results because the same identification was obtained for equivalent spots from different 2DE gels and it also matched with the identification in the 2DE *H. pylori* database. ## Motility assay As there were proteins related somehow to bacteria motility included in the group of proteins which were expressed differentially between NUD and PUD strains, we further evaluated whether this was translated into differences in motility. Accordingly, the pools of five NUD strains and of five PUD strains under study were inoculated in an agar motility medium and, at 5, 7 and 11 days the diameter of the growth halo was measured. In agreement with the higher abundance of FlgE (spot 2), FlaA (spot 3) in PUD strains (both in DU and GU strains), the *H. pylori* strains associated with PUD in children showed bigger growth halos, indicating that they have higher motility than the strains associated with NUD. These experiments were carried out twice, leading to consistent results. # Discussion The development of PUD in children is a rare consequence of infection by *H. pylori*. Environmental factors should have a minor influence on its pathogenesis since pediatric PUD develops soon after infection. Its dependence on the genetic susceptibility of these children is still poorly understood. The expression of sialyl-Lewis (x) antigens in gastric epithelial cells in association with *H. pylori*-dependent DU in pediatrics was recently reported. However, in contrast to the aberrant expression found in adults patients, the gastric mucosa of these children presents a normal pattern of expression and glycosylation of MUC6 or MUC2. Moreover, the susceptibility associated to male gender has been described and the hypothesis that female hormones, before the onset of the menopause, somehow protect against the development of PUD has been raised. In agreement, we have recently reported that pediatric PUD is significantly more frequent in boys than in girls (63.6% *vs* 36.4%, *p*\<0.025) when analyzing a large cohort of 1,115 Portuguese children. Notwithstanding, the involvement of more pathogenic strains is surely a key factor in pediatric *H. pylori*-dependent PUD, as was clearly demonstrated by our previous reported data. In order – to better characterize the virulence of such strains, we studied a group of five *H. pylori* strains isogenic for important virulence factors associated with pediatric PUD, namely *cagA*, *vacA* s1 allele, *oipA* “on”, *homB* and *jhp562* (with the exception of the strain 1846/05) genes. These were compared to another group of five strains, isolated from Portuguese children presenting NUD, all negative for those virulence factors. In agreement with the more pathogenic behavior of PUD *H. pylori* strains, our co-culture assays showed that, indeed, the five PUD-associated strains, when pooled together, caused a more pronounced reduction in the viability of NCI-N87 cells than the pool of five NUD-associated strains. In contrast to NCI-N87 cells in co-culture with NUD strains, the remaining epithelial cells after 24 h of co- culture with PUD strains presented a dramatic destabilization of the microtubule network and a decrease in the cytoplasmic reactivity to PAS, suggesting lower amounts of mucins. These results are consistent with the decreased levels of mucins observed in gastric biopsies of infected adult patients with PUD and in cell lines in co-culture with *H. pylori* strains positive for both *cagA* and *vacAs1* genes. Double staining with PAS and hematoxylin showed that *H. pylori* PUD strains induce the destruction of the cytoplasmic membrane of some host cells, leading them to release their cytoplasmic content. This is similar to necrosis, a required step in the PUD pathogenesis process. Probably due to changes in environmental conditions caused by massive cell death, PUD strains acquired a coccoidal form. This phenotype was already observed at 12 h of co- culture. Taking advantage of our pediatric strains, we decided to evaluate the proteome of each of the 10 Portuguese *H. pylori* strains. Their proteome profiles were consistent and comparable with proteome profiles in databases. Moreover, 27 proteins were found to be differentially expressed between PUD and NUD strains. It should be stressed that these differences in protein expression profile are intrinsic to the bacteria and were not induced by the environment. In fact, despite having experienced completely different environmental conditions *in vivo*, in this study they were grown under the same controlled conditions which did not fully resemble their natural niche. An important observation was noted for the group of PUD-associated strains: the strain 499/02 presented unique features which required special attention given that it was isolated from a child with GU and the other four were associated with DU. Although these diseases share important molecular mechanisms in their pathogenesis, they are characterized by different patterns of colonization, gastritis and gastric acid secretion. Our results suggest that the ability to induce one of these situations is reflected in the proteome of the implicated *H. pylori* strains and should be investigated further. Thus, the strain 499/02 was considered separately in the subsequent analysis. resumes the following discussion of the differentially expressed proteins grouped according to their known function. We were not able to identify spot 7 but its role should be important role as it was absent from the proteome of all PUD strains. ## Motility For efficient colonization of the gastric niche, *H. pylori* depends on its ability to swim. The results of the five pediatric strains associated with PUD show higher levels of FlgE (spot 2) and FlaA (spot 3) compared to NUD strains, justifying their enhanced motility. We must stress that spot 3 may result from the overlap of FlaA and FlaB (flagellin B), another component of the *H. pylori* flagellum filament, both having equal p*I* and similar molecular weight (53 and 54 kDa, respectively). If that is the case, our results reflect a change in the abundance of both FlaA and FlaB.The abundance of HELPY_0944 protein (spot 16) was decreased in the five PUD strains compared to the NUD strains. Despite its homology with the protein HP0958 (*H. pylori* strain 26695), these results point to a different cellular function for these two proteins. Contrary to our data on the expression of FlgE, FlaA (and FlaB) and HELPY_0944 proteins, *HP0958* mutants result in non-flagellated strains, showing a total abrogation of *flgE* and *flaB* transcription, and very low levels of FlaA. In fact, HP0958 protein is known to induce the transcription of *flgE* and *flgB* through stabilization of the RNA polymerase sigma factor RpoN, and the synthesis of FlaA in an RpoN- independent manner. Besides its polar flagella, *H. pylori*'s spiral morphology is a requirement for its motility. In line with this, HPAG1_1081 (spot 13) was exclusively detected in DU-*H. pylori* strains. HPAG1_1081 is a homolog of HP1143, a coiled-coil protein with a predicted intermediate filament-like function, thus contributing to the morphology maintenance. HPG27_1480 protein (spot 27), another protein found more abundantly in all PUD strains, is homolog to HP1542, a protein involved in spiral shape maintenance. ## Antioxidant system For a sustained infection of the inflamed gastric mucosa, characterized by the presence of large amounts of reactive oxygen species (ROS), *H. pylori* depends on the concerted activity of several proteins for its antioxidant defense. Here we report higher levels of Pfr (spot 23) and NapA (spot 26) proteins and lower levels of KatA (spot 9) for the PUD strains. Pfr is an essential protein for *H. pylori* which serves as a storage of iron in a bioavailable form. Thus, besides protecting the bacterium against oxidative stress caused by the iron-mediated Fenton reaction in the presence of excessive amounts of this ion, Pfr also guarantees the necessary amount of iron for the normal metabolism of the cell under iron-limited conditions. Like Pfr, NapA is capable of sequestering iron under oxidative stress conditions and these two share a protective role of the bacterial DNA. NapA is, however, better known for its role in mediating the adhesion of *H. pylori* to host mucins and its ability to recruit neutrophils and monocytes to the infected gastric mucosa, inducing the secretion of chemokines. According to the literature, abnormally high levels of NapA are expressed in mutant strains lacking KatA. Instrumental in catalyzing the decomposition of H₂O₂ into H₂O and O₂, this enzyme has been described as being essential to the survival of *H. pylori* in a sustained long-term inflammation. We therefore propose that the NUD strains analyzed in this study are prepared to resist in a long-term but less aggressive inflammation. However, the five pediatric strains associated with PUD are able to survive an acute inflammation that results in the early development of PUD. Two isoforms of HP1588 protein (spots 17 and 18) were also differentially expressed between PUD and NUD strains. The exact function of this protein is not known although some studies indicate a potential role in stress resistance, as it is over-expressed under acidic or iron starvation conditions. ## Acid Resistance Urease is a key enzyme in the colonization and persistence of *H. pylori*, playing a central role in the bacteria resistance to gastric acidity by catalyzing the hydrolysis of urea into ammonia and CO₂. Ammonia is, however, cytotoxic to the host cells, causing their necrosis. This multimeric protein of approximately 1100 kDa is composed of twelve small subunits, UreA (27 kDa), and twelve large subunits, UreB (62 kDa). In agreement with our previous findings , urease (UreA plus UreB) was one of the most abundant proteins in the proteome of all of the *H. pylori* strains in the current study. However, the strains collected from children with DU had higher levels of both UreA and UreB, than either the NUD or the GU strains. Higher levels of HspB (spot 4), a protein which functions as an extracellular chaperonin for urease among other key roles, were observed in all PUD strains. According to the literature, excessive urease activity under neutral conditions is lethal to the bacteria. Considering that no difference was detected neither in the growth rate nor in the urease activity among these pediatric strains at pH 7 (results not shown), we hypothesize that the highest amount of urease produced by the DU strains does not reflect a higher activity *per se*. Instead, urease activity must be acid–dependent and thus, once in their natural niche, DU strains are better protected against a sudden drop in pH and/or in Ni²⁺ concentration which explains their ability to survive under abnormally acidic conditions. Moreover, since urease has also been proposed to play a role in adherence to gastric mucins, again in a acid dependent manner, higher levels of urease may facilitate the processes of colonization and infection, contributing to the ulcerogenic phenotype of DU strains. Lastly the putative aldo-keto reductase (spot 15) was also found to be more abundant in DU strains. It was described as an essential enzyme for growth at acidic pH, and is involved in the removal of toxic aldehydes and ketones of the cell. Again, the abundance of this protein in the proteome of the GU-associated strain 499/02 was equivalent to that in NUD strains, further corroborating the idea that only the DU-associated pediatric *H. pylori* strains are prepared to face a hyper-acidic environment. ## Metabolism Among the proteins expressed differentially between NUD and PUD strains, there were 10 proteins involved in the metabolism of the bacterium. A lower abundance of HyuA (spot 6), the enzyme responsible for the conversion of N-methyl hydantoin to N-carbamoyl sarcosine, was observed in all PUD-associated *H. pylori* strains. Not much is known about the role of this protein in *H. pylori* but this was described as a key step in the metabolism of creatine in other bacteria, a final pathway in the urea cycle and amino acid metabolism. Whether this may favor the endogenous production of urea we do not know and it should be investigated. In line with our findings, other authors have indicated HyuA as a biomarker for GC, since positive antibodies against it were found in sera from patients suffering with GC but not with PUD. Interestingly, AspA (spot 11), the enzyme that by catalyzing the deamination of Asp to fumarate ensures the aneuplerotic replenishment of the latter in Krebs cycle, was in low abundance in all PUD strains compared with NUD strains. Asp and fumarate link the Krebs cycle to the urea cycle, therefore, we may suppose that in the former group of strains, Asp is primarily involved in the urea cycle with the concomitant production of urea. However, as *H. pylori* uses amino acids as a primary source of carbon, nitrogen and energy, with Asp being one of the eight most consumed amino acids, we cannot rule out that in PUD strains, Asp is used in the biosynthesis of Asp-derived amino acids. Indeed, the metabolism of amino acids was recently connected to virulence, colonization and stress resistance of *H. pylori*. In *H. pylori*, the Krebs cycle is characterized by the absence of succinate dehydrogenase. Although controversial, the SCOT complex which catalyzes the conversion of succinyl-CoA to succinate in an acetoacetate- dependent manner was proposed to substitute the missing enzyme. Here *H. pylori* strains isolated from children with PUD presented lower levels of both subunits of the SCOT-complex, ScoA and ScoB (spots 19 and 21, respectively) which suggests a down-regulation of its catalytic activity. Since *H. pylori* is a microaerophilic organism, the initiation of the Krebs cycle, *i.e.* the oxidative descarboxylation of pyruvate is catalyzed by an oxygen sensitive enzyme, pyruvate flavodoxin oxidoreductase (POR), instead of the aerobic pyruvate dehydrogenase or the strictly anaerobic pyruvate-formate lyase. Although its stoichiometry is unknown, POR is composed of four subunits, all essential for the *in vitro* viability of *H. pylori*. These are encoded by the *porCDAB* operon. In this reaction, the FldA protein is the electron acceptor, a low potential one, which in turn is re-oxidized by the flavodoxin- quinone reductase (FqrB), generating NADPH. *In vitro* experiments have shown that POR is the rate limiting enzyme in this pathway. The reverse reaction of this POR-FldA-FqrB-NADP reductase complex is believed to be important in the fixation of CO₂, which is essential for replenishing pyruvate consumed by gluconeogenesis. Interestingly, we found significantly lower levels of one isoform of the PORγ subunit (spot 24), encoded by the *porC* gene, in the pediatric PUD-associated *H. pylori* strains compared to the NUD strains. As far as we known, the role of PORγ subunit and its isoforms in the POR complex is still unclear. However, because of the observed higher levels of FldA (spot 25), we interpret our findings as PUD strains having a more functional POR. In this case, we would expect a higher production of NADPH and acetyl-CoA in PUD strains. Further experiments are needed to clarify this. Moreover, the Pgk protein (spot 12), a glycolytic/gluconeogenic enzyme which catalyzes the reversible conversion of 1,3-biphosphoglycerate into 3-phosphoglycerate, was found to be more abundant in the DU-associated strains and even more abundant in the GU related *H. pylori* strain. Chorismate, a precursor of aromatic amino acids, is formed from phosphoenol pyruvate and erythrose 4-phosphate by means of seven enzymes. AroQ (spot 22), an enzyme that catalyzes the conversion of 3-dehydroquinate into 3-dehydroshikimate which is a key step in chorismate synthesis, showed an augmented expression in DU-associated *H. pylori* strains and, again, an even higher expression in the GU strain. This is suggestive of an enhancement in aromatic amino acid biosynthesis in DU strains compared to NUD strains and even more pronounced in the GU strain over all of the other. In *H. pylori*, amino acids in excess of those needed for protein synthesis cannot be stored and are catabolised. Our data point out that this ability is potentiated in the pediatric PUD-*H. pylori* strains. In fact, lower levels of both RpsA (spot 8), one of the 21 ribosomal proteins of the small 30S subunit, and aminoacyl-tRNA synthetase CysS (spot 10) were observed for these strains when compared to NUD strains. These data suggest a down-regulation of translation in general saving amino acids for degradation. If that is the case, the aforementioned higher levels of HspB registered for PUD strains would also play an important role in avoiding protein misfolding. The elongation factor EF- Ts (spot 14), also involved in translation, showed a shift in 2DE gels of DU strains, suggesting a post-translational modification which may influence its activity. In fully agreement with these results, two studies reported that under stress, nutritional or low pH conditions, the stringent response of *H. pylori* included the down-regulation of ribosomal genes and of aminoacyl-tRNAs and, in contrast, the enhancement of amino acid biosynthesis. The beauty of the data presented here is that are not induced by stress conditions but, as mentioned above, are intrinsic to PUD strains, making them much more adapted and, therefore, virulent. ## Virulence factors As already mentioned, all of the *H. pylori* strains associated with PUD in children that we have analyzed carried the genes of some well known virulence factors from which we could only detect the expression of *cagA*. CagA (spot 1 in our 2DE gels) which is known to induce abnormal proliferation, disruption of tight junctions, cytoskeleton rearrangements and IL-8 secretion in the host cells, was less abundant in the GU strain when compared to the other four DU associated strains. It was important to disclose whether the special features of the proteome profile of the PUD-*H. pylori* strains of this study were linked to their *cagA*/*vacA*s1 positive genotype. For that, we checked the abundance of the aforementioned 26 proteins in the proteome of ten other Portuguese *H. pylori* strains isolated from children and adult patients with NUD and from adults patients with PUD and GC (five *cagA/vacA*s1 positive and five others negative for these virulence factors), all currently under study in our lab. Of those proteins, only HspB and Pfr showed a pattern of abundance in the comparison between *cagA* positive and negative strains (data not shown) similar to that presented when comparing pediatric PUD and NUD strains (data from this study). This supports our hypothesis that the pediatric *H. pylori* strains associated with PUD present a specific protein signature which provides them a natural ability of adaptation to the human stomach. This profile combined with the expression of virulence factors (*cagA*, *vacA*s1, *oipA* “on” status, *homB* and *jhp562*) appears to be responsible for their enhanced virulence. Asian isolates which present marked differences in their genetic background compared to European strains, namely being nearly all positive for *cagA*, are an interesting group to study in the future. ## Conclusion The data that we report here clearly show that the pediatric ulcerogenic *H. pylori* strains in our study share a particular proteome profile that, in addition to the well established virulence factors, provides them with higher motility. They also highlight for the strains' better antioxidant defenses and metabolism favoring the biosynthesis of aromatic amino acids, perhaps to be used as source of energy. Additionally, DU strains are apparently better fitted than all the other studied strains to survive under low pH conditions, which may justify their survival following acid hypersecretion which is characteristic of this disease. We believe that the virulence of pediatric ulcerogenic strains is strongly dependent on the synergy of their well established virulence factors and their better adaptation to the natural niche. Despite the relevance of these data, further research is required to determine their biological meaning under stressful conditions (acidic, oxidative and nutrient limited conditions). Moreover, host susceptibility should be evaluated in order to clarify its role in the pathogenesis of pediatric DU. Finally, it will be important to characterize all of the differences in the proteome profile of DU- and GU- associated *H. pylori* strains, to better understand the divergence of these diseases. # Methods ## Bacteria and cell growth conditions A total of 10 *H. pylori* strains isolated from children attending the pediatric gastroenterology units in the Lisbon area (Portugal) for upper diagnostic gastrointestinal symptoms were analyzed. These included five patients presenting NUD and five others with PUD (one GU and four DU). None of the children had received anti-*H. pylori* antibiotic or anti-secretory therapy prior to endoscopy. All of the children suffering from PUD had no other etiology for the disease. These strains belong to the collection of bacterial strains of the Department of Infectious Diseases of the National Institute of Health Dr. Ricardo Jorge, in Lisbon, Portugal. Bacteria were grown in *H. pylori* selective medium (Biogerm, Maia, Portugal) at 37°C in a microaerobic environment (Anoxomat®, MART Microbiology BV, Drachten, The Netherlands) for 24 h. For motility and co-culture assays, a pool of the five NUD strains and a pool of the five PUD strains were prepared by mixing biomasses recovered from 24 h grow plates of each strain (equal amount of each strain). For co-culture assays, the NCI-N87 (ATCC CRL 5822) cell line was grown at 37°C with 5% CO₂ and 99% humidity in Dulbecco's modified Eagle's medium (DMEM/F12) (Invitrogen, Life Technologies, Carlsbad, CA, USA) supplemented with 10% (v/v) of heat inactivated (56°C for 30 min) fetal bovine serum (FBS) (Invitrogen). ## Co-culture assays A pool of the five NUD-associated strains and a pool of the five PUD-associated strains were prepared in NCI-N87 cell growth medium, and diluted to a final concentration of 1×10⁸ CFU/mL. NCI-N87 cells grown on 8-well chamber slides (Nalge Nunc, Roskilde, Denmark) for immunocytochemistry assays or on 24 multi-well plates (Nalge Nunc) for cellular viability determination, until 80 to 90% confluence were rinsed twice with phosphate-buffered saline (PBS) (Invitrogen) and fresh growth medium was added. Bacterial pools were then added at a MOI of 100 and the plates were maintained under NCI-N87 cell growth conditions. Non-infected NCI-N87 cells were used as a control. At 1, 12 and 24 h post-infection, cells were analyzed by light microscopy and stained using PAS (Sigma-Aldrich Co., St. Louis, MO, USA) and hematoxylin (Sigma-Aldrich). Cell viability was determined using the classical trypan blue exclusion test (Sigma- Aldrich). ## Immunocytochemistry analysis After co-culture with the pools of *H. pylori* strains, NCI-N87 cells were rinsed twice with cold PBS and fixed for 30 min at 4°C in a 4% (v/v) formaldehyde (Sigma-Aldrich) and 3.7% (w/v) sucrose (Merck, Darmstadt, Germany) solution, in PBS. After two washes with PBS, cells were permeabilized for 15 min with 0.2% (v/v) Triton X-100 (Sigma-Aldrich) in PBS at room temperature (RT), washed three times more with PBS and blocked with 1% (w/v) bovine serum albumin (BSA) in PBS for 30 min at RT, prior to incubation for 2 h at 37°C with the anti-α-tubulin (clone DM1A) (Sigma-Aldrich) monoclonal antibody (diluted 1∶1000 in 0.5% (w/v) BSA in PBS). Cells were then washed three times with PBS and incubated for 1 h at 37°C with the fluorescein isothiocyanate (FITC)-conjugated anti-mouse IgG (Sigma-Aldrich) diluted 1∶100 and washed three times again. Cell slides were mounted in Vectashield (Vector Laboratories, Burlingame, CA, USA) and their immunofluorescence was observed and recorded on an Axiovert 40CFL fluorescence microscope (Carl Zeiss, Jena, Germany) equipped with an Axiocam MRc5 (Carl Zeiss) camera. Images were processed with the software AxioVision Rel. 4.6.3 (Carl Zeiss). ## 2DE Bacterial total protein extracts and their separation by 2DE was performed as previously described. Briefly, 800–1000 µg of protein in 450 µL re-hydration buffer were loaded onto 18 cm Immobiline DryStrips (GE Healthcare, Uppsala, Sweden) with a non-linear wide range pH gradient (pH 3–11). After active gel strip re-hydration, IEF was run on an Ettan IPGphor 3 unit (GE Healthcare) for a total of 100 kVh during which the voltage was gradually increased up to 5,000 V for a total of 66 h. For SDS-PAGE, the second dimension analysis, the proteins on the gel strips were equilibrated by soaking the strip in equilibration buffer for 15 min at RT and then for another 15 min in blocking buffer. Strips were finally applied onto 7–16% (w/v) gradient polyacrylamide gels and run overnight at 1 W/gel (Ettan DALT*six* System, GE Healthcare). Protein visualization was carried out by CBB staining (Sigma Aldrich). ## 2DE map analysis CBB-stained 2DE gels were scanned in the ImageScanner (GE Healthcare) operated by the software LabScan 5 (GE Healthcare) in transparency mode with a red color filter. Scanning was carried out at 300 dpi and 16 bit grayscale. Images were analyzed using the ImageMaster™ 2D Platinum software (GE Healthcare) as before, taking into account the standardized relative intensity volume of spots (or % Vol, *i.e.* the volume of each spot over the volume of all spots in the gel). Differences in protein abundance among strains were statistically assessed. To assure result reproducibility, all samples were subjected to 2DE twice, making a total of 20 analyzed 2DE gels. ## Sample preparation and mass spectrometry analysis The selected differentially expressed proteins were identified by PMF using an Autoflex III MALDI-TOF/TOF mass spectrometer (Bruker Daltonics, Bremen, Germany), as previously reported, with some minor modifications. Protein spots were excised from the CBB-stained 2DE gels and enzymatically digested in-gel with proteomics grade porcine trypsin (Sigma-Aldrich). Trypsin-digested peptides were loaded in a disposable ready-to-use MALDI target prespotted with α-cyano-4-hydroxycinnaminic acid (Prespotted AnchorChip PAC 384/96, Bruker Daltonics) and were assayed with the mass spectrometer in the positive ion mode. Spectra acquired with the FlexControl (Compass software, version 1.2, Bruker Daltonics) in reflection mode, were processed by using FlexAnalysis (Compass software, version 1.2, Bruker Daltonics). Monoisotopic peptide masses were used to search protein databases (*H. pylori* NCBI nr. 2011.03.04, 13254464 sequences; or Swissprot databases nr. 2010.10, 521016 sequences) using MASCOT software (version 2.3.01, Matrix Science, London, UK). A mass accuracy of 50–100 ppm and 1 missed cleavage were allowed in the searches. In the absence of matches, the mass window was extended up to 200 ppm. Cysteine carbamidomethylation was considered as a fixed modification and protein *N*-terminal acetylation, oxidation of methionine and pyroglutamic-acid on N-terminal glutamic acid were allowed as variable modifications. Proteins with significant MOWSE scores (*p*\<0.05) are reported. In a few special cases, MS identification based on MOWSE scores with no statistical significance were also reported and discussed since the same identification was obtained for the equivalent spot from different 2DE gels and because it matched the identification in the 2DE *H. pylori* database. ## Motility assays 5 µl of each pool of *H. pylori* prepared as mention above, were inoculated with a sterile pipette tip into motility agar plates, consisting of brain-heart infusion (BHI) medium (Oxoid, Hampshire, UK) supplemented with 5% (v/v) of heat- inactivated (56°C for 30 min) FBS, *H. pylori* selective supplement (10 mg/L of vancomycin, 5 mg/L of trimethoprim, 5 mg/L of cefsulodin and 5 mg/L of amphotericin B) (Oxoid) and 0.35% (w/v) agar. Motility was scored by measuring the bacterial halo diameter at 5, 7 and 11 days of incubation of the plates under microaerobic conditions (Campygen CN0025, Oxoid), at 37°C. Differences in motility of the two pools of strains, tested in three independent experiments, were statistically assessed. ## Statistics Whenever necessary, differences were tested by Student's *t* test, being considered as statistically significant when *p*\<0.05. Results are expressed as averages ± standard deviations (SD) of *n* observations. The authors thank Professor Filipa Vale (Faculdade de Engenharia, Universidade Católica Portuguesa) for the scientific revision of the manuscript and Lindsay Mégraud for the careful revision of the English. [^1]: Conceived and designed the experiments: MO MRR. Performed the experiments: IV KDSP ARG AS MO. Analyzed the data: IV AIL MO MRR. Contributed reagents/materials/analysis tools: AIL MO MRR. Wrote the paper: IV MO MRR. [^2]: The authors have declared that no competing interests exist.

# Introduction The SARS-CoV-2 (Coronavirus diseases; COVID-19) has continued to affect many countries, including the United States, since the World Health Organization (WHO) declared a global pandemic in March 2020. Before the declaration, metabolic conditions such as obesity, type 2 diabetes mellitus, and cardiovascular disease have continued to be the leading cause of morbidity and mortality in the U.S. and the world. In 2018, approximately 13% of all U.S. adults had diabetes, with 2.8% of this population being unaware of their status but meeting laboratory criteria for diabetes. Similarly, a national survey from 2017 to 2018 shows that 42.4% of U.S. adults had obesity. These metabolic conditions are associated with severe health risks. Additionally, they predispose people to the risks of death and adverse health outcomes from COVID-19. However, studies on the effects of COVID-19 on these metabolic conditions are scarce. Indeed, the associations between cardiometabolic conditions in U.S. adults and the COVID-19 pandemic and risk factors such as physical inactivity, tobacco use, anxiety/depression, and sociodemographic characteristics remain understudied. Not only did patients with diabetes or obesity have increased mortality due to COVID-19 infection, their overall health was also negatively impacted by the COVID-19 lockdown. Studies demonstrated poor glycemic control and increased body mass index (BMI) for patients with diabetes during the lockdown along with a deterioration in glucose regulation. The timings of lockdown orders in the target populations from these studies differ, as they were based on when the countries declared a lockdown. Other studies have shown that newly diagnosed diabetes is more prevalent in patients following COVID-19 infection, with one in every ten COVID-19 patients diagnosed with new onset diabetes mellitus. In addition to metabolic health outcomes secondary to diabetes, some subgroups, specifically those with cardiometabolic disease, were at an increased risk of poor mental health outcomes, had depressive symptoms, and poor sleep quality. Other negative behavioral activities included reduced physical activity and increased alcohol consumption. While the metabolic conditions such as diabetes and obesity are well documented, there is a paucity of research on the association between conditions such as hypertension and COVID-19. Given the high prevalence of medical conditions such as hypertension and the potential for negative health outcomes secondary to COVID-19, this study explores the relationships between COVID-19 and metabolic conditions before and during the COVID-19 pandemic. We utilize a nationally representative sample of U.S. adults to estimate the prevalence of metabolic conditions (diabetes, hypertension, and obesity) before and during the COVID-19 pandemic declaration. Further, this study examines the association between these metabolic conditions among U.S. adults and the ramifications of the COVID-19 pandemic, including physical inactivity, tobacco use, anxiety/depression, and sociodemographic characteristics. Considering the higher prevalence of these metabolic conditions in the U.S. adult population and throughout the world, it is imperative to understand which populations to target for public health interventions to decrease COVID-19 related morbidities for high-risk populations. # Methods The 2019 and 2020 Health Information National Trends Surveys (HINTSs) de- identified public-use datasets were combined for this study. HINTS is a cross- sectional survey that assesses health-related information (e.g., diabetes, hypertension, and obesity) and behaviors (e.g., tobacco use) among a nationally representative sample of U.S. adults aged ≥18 years. It uses a random sampling technique to select a sample of the U.S. civilian, noninstitutionalized adult population. Details of the methods, questionnaire, and survey administration have been published. The 2019 survey (HINTS 5 Cycle 3) was conducted from January through April 2019, and the 2020 survey (HINTS 5 Cycle 4) was conducted from February through June 2020. These surveys are the recent publicly available HINTS datasets. The combined HINTS 5 Cycles 4 (N = 3,865) and 3 (N = 5,438) datasets consist of a total sample of 9,303 adults. The study was reviewed by the Institutional Review Board (IRB) of East Tennessee State University and exempted as the HINTS datasets are de-identified and publicly available. # Measures and variables ## Dependent variable The dependent variable is "metabolic condition", derived from three distinct questions about diabetes, hypertension, and obesity. For diabetes, the participants were asked, "Has a doctor or other health professional ever told you that you have diabetes or high blood sugar?" (yes/no). Hypertension was assessed with the question, "Has a doctor or other health professional ever told you that you have high blood pressure or hypertension?" (yes/no). Obesity status was determined using body mass index (BMI), and defined as underweight = \<18.5, healthy/normal = 18.5–24.9, overweight = 25.0–29.9, and obese ≥ 30.0. Obese was defined as BMI ≥ 30.0 and not obese as BMI \< 30. Thus, the variable "metabolic condition" in this study was ascertained as participants who had diabetes, hypertension, or were obese. ## Independent variables The main independent variable is the HINTS survey year, which was based on the 2019 and 2020 surveys, given that COVID-19 cases were widespread globally by January 2020. The 2019 HINTS data were used as the pre-COVID-19 pandemic cohort, while the 2020 HINTS data were used as the COVID-19 pandemic cohort for stratified analysis. Other independent variables analyzed in this study included self-reported sociodemographic characteristics, moderate physical activity intensity, cigarette smoking status, e-cigarette use status, and anxiety/depression symptoms. The sociodemographic variables included age (18–25, 26–34, 35–49, 50–64, and 65+), sex (male/female), race/ethnicity (non-Hispanic White, non- Hispanic Black/African American, Hispanic, non-Hispanic Asian, and non-Hispanic others), gender identity (heterosexual/straight or sexual minorities \[homosexual, lesbian, gay, or bisexual\]), marital status (single/never married, married/living as married, divorced/separated, or widowed), level of education completed (less than high school, high school graduate, some college, and college graduate or higher), and total family income (\<\$20,000, \$20,000 to \< \$35,000, \$35,000 to \< \$50,000, \$50,000 to \< \$75,000, or  ≥ \$75,000). General health status was based on self-ratings of overall health as excellent, very good, good, fair, or poor. Due to limited samples, we dichotomized general health status into excellent/very good/good or fair/poor. The number of days per week of moderate intensity physical activity (none and at least one day per week), cigarette smoking status (never/non-smoker, former smoker, and current daily or some days smoker), and e-cigarette use status (never used, former user, and current daily/some days user) were also included. The anxiety/depression symptoms variable was constructed from Patient Health Questionnaire-4 (PHQ-4) in the HINTS 5 survey. The PHQ-4 assesses symptoms/signs of anxiety and depression, with total scores from 0–12 (0–2 = normal/negative, 3–5 = mild, 6–8 = moderate, and 9–12 = severe). Thus, anxiety/depression symptoms were categorized into normal or no anxiety/depression, mild, moderate, and severe. ## Statistical analyses The HINTS sampling weight was applied to the analysis to achieve population estimates and offset nonresponse. We estimated the weighted prevalence of each component of metabolic condition before and during the COVID-19 pandemic. The weighted prevalence and unweighted frequencies of metabolic conditions by the sociodemographic characteristics, moderate-intensity physical activity, cigarette smoking status, e-cigarette use status, and anxiety/depression symptoms were computed to characterize the survey sample before and during the COVID-19 pandemic. Additionally, two logistic regression analyses represented by two models were conducted. While Model 1 assessed the association between the metabolic conditions and independent variables using the data before the pandemic, Model 2 utilized the data during the pandemic. All analyses were weighted using the HINTS sampling weight and replicate weight to offset non- response bias and to achieve nationally representative estimates. The weighted percentages, adjusted odds ratios (AOR), 95% 2-tailed confidence intervals (CI), and statistically significant *p*-value (\< 0.05) have been reported. # Results The prevalence of each metabolic condition (diabetes, hypertension, and obesity) before and during the COVID-19 pandemic is presented in. The results showed that the prevalence of diabetes was higher during the COVID-19 pandemic (18.10%) than before the pandemic (17.28%). However, the prevalence of hypertension (36.36% vs. 36.38%) and obesity (34.68% vs. 34.18%) was similar during and before the pandemic. presents the prevalence of metabolic conditions by independent variables before and during the COVID-19 pandemic. Overall, the prevalence of metabolic conditions (54.96% vs. 56.09%) was higher during the pandemic than before the pandemic. The distribution of the prevalence of metabolic conditions within the sociodemographic groups, moderate intensity physical activity, cigarette smoking status, and e-cigarette use status before and during the COVID-19 pandemic varied. The prevalence of metabolic conditions during the COVID-19 pandemic, compared to before the pandemic increased for individuals aged 35–49 and 50–64 years but decreased for those aged 18–25, 26–34, and ≥65 years. The prevalence also increased for all non-Hispanic racial/ethnic groups but decreased for Hispanic individuals. The prevalence of metabolic condition was higher for individuals who did not engage in moderate-intensity physical activity compared to those who engaged in at least one moderate-intensity physical activity per week. Individuals who were former cigarette smokers or current smokers had an increased prevalence of metabolic conditions. For e-cigarette use groups, the prevalence had increased for those who had never used e-cigarettes and those who currently used e-cigarettes; however, it decreased for former e-cigarette users. shows metabolic conditions and their associated factors before (Model 1) and during (Model 2) the COVID-19 pandemic, respectively. Before the pandemic, compared to age 18–25 years, only two groups of individuals aged 50–64 (AOR = 2.64, 95% CI = 1.20, 5.77) and 65 years or older (AOR = 4.82, 95% CI = 2.25, 10.32) had significantly higher odds of metabolic conditions. During the pandemic, the likelihood of metabolic conditions was significantly higher for four groups: individuals aged 26–34 (AOR = 1.95, 95% CI = 1.04, 3.67), 35–49 (AOR = 4.13, 95% CI = 2.12, 8.04), 50–64 (AOR = 6.19, 95% CI = 3.03, 12.65), and 65 years or older (AOR = 7.82, 95% CI = 3.92, 15.57) compared to age 18–25 years. During the pandemic, males had significantly higher odds of metabolic conditions (AOR = 1.28, 95% CI = 1.01, 1.64) relative to females, whereas the odds were not different before the pandemic. Compared to non-Hispanic White people, the odds were significantly higher for non-Hispanic Black people before (AOR = 2.01, 95% CI = 1.26, 3.22) and during (AOR = 2.09, 95% CI = 1.22, 3.58) the pandemic. Engaging in at least one moderate-intensity physical activity per week was associated with a lower likelihood of metabolic conditions before (AOR = 0.64, 95% CI = 0.46, 0.88) and during (AOR = 0.58, 95% CI = 0.42, 0.79) the pandemic as compared to no physical activity. Having mild (AOR = 1.52, 95% CI = 1.06, 2.19) or severe (AOR = 2.44, 95% CI = 1.27, 4.69) anxiety/depression symptoms, compared to no anxiety/depression symptoms, was associated with higher metabolic conditions before the pandemic. Only mild anxiety/depression symptoms (AOR = 1.55, 95% CI = 1.01, 2.38) were associated with higher metabolic conditions during the pandemic. Compared to people who never smoked cigarettes, former cigarette smokers had significantly higher odds of metabolic conditions before (AOR = 1.38, 95% CI = 1.01, 1.87) and during (AOR = 1.57, 95% CI = 1.10, 2.25) the pandemic, but not current smokers. Before the pandemic, the likelihood of metabolic conditions was significantly lower for current e-cigarette users (AOR = 0.44, 95% CI = 0.23, 0.85) compared to those who had never used e-cigarettes, with no difference observed during the pandemic. # Discussion This study assessed the prevalence of metabolic conditions among U.S. adults and the underlying associated factors before and during the COVID-19 pandemic using the HINTS 2019 and 2020 survey data. To the best of our knowledge, this is the first study to use nationally representative U.S. adult data to highlight associations between metabolic outcomes, sociodemographic factors, and the COVID-19 pandemic. There was an increase in the overall prevalence of metabolic conditions, especially among certain subgroups during the COVID-19 pandemic. This is consistent with a systematic review that assessed the impact of disasters, including pandemics, on metabolic conditions and reported increased incidence and mortality for diabetes and obesity. Our findings indicate that being elderly (aged 50+), non-Hispanic Black person, former smoker, having fair/poor health status, and having mild anxiety significantly increased the likelihood of metabolic conditions pre- and during the pandemic. However, the disparities in these health and sociodemographic factors were greater during the pandemic. Previous studies have established that the COVID-19 pandemic exacerbated and further unmasked existing disparities in metabolic outcomes. For instance, we found that the odds of metabolic outcomes were significantly higher only among the elderly age groups (50–64 and 65+) compared to young adults before the pandemic. However, these increases in odds almost doubled among these age groups during the pandemic. Significantly higher odds were also noted in the middle age groups (26–34 and 35–49), where the odds almost tripled for the 35–49 age group during the pandemic. Consistent with the literature, our results indicated age was the strongest factor associated with an increased likelihood of adverse metabolic conditions during the pandemic with higher age range conferring a higher risk for metabolic conditions. The association between age and metabolic conditions during the pandemic and the risks for adverse health conditions from COVID-19 suggests that health interventions targeted at high-risk groups such as the elderly could optimize outcomes, particularly during disasters such as this global pandemic. Moreover, this study provides additional evidence that individual health behaviors played a critical role in developing metabolic conditions before and during the pandemic. For example, while being a former smoker increased the odds of metabolic conditions, engaging in moderate physical activity decreased the odds. In light of pandemic-related restrictions associated with increased social isolation and psychological distress, people are more likely to smoke and engage in sedentary behaviors such as screen time which limits physical activity and increases the risk of metabolic diseases. A recent systematic review assessing screen-based sedentary behavior among adolescents during the COVID-19 pandemic reported a dose-response association between increased levels of screen time and components of metabolic syndrome. Our findings are consistent with the existing literature on smoking as a risk factor for metabolic conditions and increased physical activity as a protective factor. As such, both smoking cessation and physical activity should be encouraged to reduce the risk of metabolic conditions, especially during this pandemic. ## Implications to practice and research The COVID-19 pandemic has placed extraordinary demand on public health systems and essential services, while individuals with underlying health conditions such as diabetes, hypertension, and cardiovascular diseases are at higher risk of hospitalization and death. Thus, the association between COVID-19 and metabolic conditions alongside the disparities highlighted in this study suggests the need for further research and fair allocation of medical resources to address these conditions during and after the pandemic. Although this study provides additional evidence to the literature on the effects of the COVID-19 pandemic related to metabolic conditions among the generally representative U.S. population, some limitations should be considered. These limitations include self-reporting bias and potential underestimation of chronic health problems, such as metabolic conditions, that develop over time. Given the cross-sectional data and lack of temporal sequence information on the variables, we could not make causal inferences. Longitudinal follow-up should be continued for future research to further validate this study’s findings. Additionally, public health assessment tools specifically validated for chronic diseases, such as metabolic conditions, that could be used for national observatory datasets would allow researchers to more rapidly evaluate data in real time in future public health crises, such as this recent global pandemic. Standardized validated tools would provide more meaningful assessments and results nationally and internationally. Moreover, there are probable effects of confounders not considered in this study such as sedentary behavior, sleep pattern, eating habits, and employment that might give rise to inaccurate estimates of the true association. Furthermore, the HINTS datasets do not contain responses specific to the COVID-19 pandemic. Additionally, because metabolic conditions and lifestyle behavioral risk factors take time to accumulate and change health conditions, the results of this study might be biased and under-estimated given the durations of the data before (January through April 2019) and during (February through June 2020) the COVID-19 pandemic. Therefore, future studies comparing the rates and prevalence of metabolic conditions before and during the pandemic are needed considering the longer of the data. Despite these limitations, these data validate that high-risk groups, such as advanced age, should be targeted for interventions to protect against the negative effects of COVID-19. Another gap in the literature that could be addressed with future research is the health consequences of a public lockdown, which was the mitigation strategy for a global pandemic, compared to the consequences of the infectious disease itself. Development of tools designed to measure outcomes secondary to each of these distinctly different effects would benefit future research and resultant health policy. For example, it would be helpful to know if depressive symptoms were a direct effect of the disease, such as suffering from long-COVID, or from the social isolation secondary to the public lockdown. Overall, the HINTS dataset provides an efficient means to evaluate important public health questions in a rapidly evolving situation such as the COVID-19 pandemic. # Conclusion In this nationally representative sample of U.S. adults, the prevalence of metabolic conditions increased during the COVID-19 pandemic in certain subgroups of individuals. Specifically, there was an increased risk of metabolic outcomes associated with older age. Other groups with signals for increased risk included: non-Hispanic Black people, former smokers, individuals with poor health status, and mild anxiety. Thus, there is a need for proper rationing of resources to address these conditions during the pandemic. 10.1371/journal.pone.0279442.r001 Decision Letter 0 Kim Taeyun Academic Editor 2023 Taeyun Kim This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 15 Sep 2022 PONE-D-22-21818The prevalence of cardiometabolic conditions before and during COVID-19 and its association with health and sociodemographic factorsPLOS ONE Dear Dr. Mamudu, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Oct 30 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Taeyun Kim Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please amend the manuscript submission data (via Edit Submission) to include author Stewart DW,PharmD. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This paper evaluates the population prevalence of diabetes, obesity and hypertension in the US during 2019 and 2020. The paper concludes that the prevalence of these conditions increased during the Covid-19 pandemic. 1\. The title seems to imply that observations were made on Covid-19 status of individuals, which is not the case. Please adjust the wording of the title to make this clear. 2\. In the Abstract, make clear that the study compares 2019 and 2020 and no observations were made on Covid-19 as such. We know that diabetes has been increasing over many years, so this paper does not demonstrate any deviation from the underlying trend, consquently the conclusions that can be darwn are quite limited. 3\. Introduction, lines 68-69, there needs to be more discussion of references 16 and 21 - what do these studies show? Also, how widespread and what were the tiomings of lockdown orders in the target population for this study. 4\. The outcome comprises hypertension, obesity and diabetes. No cardiac conditions are included, so this might be equated with the 'metabolic syndrome'. 5\. In the analysis, explain what weights were employed. 6\. Using a cut-off of P \<0.05 is an out of date approach to interpretation. Please follow the ASA guidelines on P values. <https://www.tandfonline.com/doi/full/10.1080/00031305.2016.1154108> 7\. Figure 1 may be better as a Table. If included it will be preferred to show the results for 2019 and 2020 side by side so these can be more readily compared. 8\. Table 1. More appropriate column headers may be '2019' and '2020'. 9\. Table 2: if the intention is to see whether associations differed in 2019 and 2020, it would be better to include all the data in a one model and test for the interaction of each variable with study year. Most of the associations appear to be quite similar across years. 10\. In the Discussion, what can be concluded is very limited because the study has not evaluated secular trends over time. The difference between 2019 and 2020 could be accounted for by the underlying trend. 11\. In the Limitations section, mention that no caulsa inferences can be drawn. 12\. In the conclusion, where it says 'Disparities in cardiometabolic conditions became more evident after the pandemic', there does not appear to be sufficient evidence from the analysis to support this conclusion. 13\. Where it says ' the prevalence of cardiometabolic conditions increased during the COVID-19 pandemic', only diabetes increased not the other conditions, and we do not know whether the increase exceeded pre pandemic expectations. Please address this text in teh Abstract also. Reviewer \#2: Review of Manuscript Number: PONE-D-22-21818 Introduction: The introduction is lacking the part linking the risk factors especially mental health status during pandemic to cardiometabolic diseases and how COVID-19 pandemic has led to an increased adoption of such behavioral risk factors e.g., physical inactivity and smoking. � Line 62-64 COVID-19 may have an exacerbating effect on glycemic control for patients with diabetes \[14, 20\], and there may be risk of increased body mass index (BMI) as well as a deterioration in glucose regulation due to COVID-19 \[16, 21\]. Reviewer comments: Will you clarify if COVID-19 infection or the implications of COVID-19 related lockdown have led to exacerbation of glycemic control and increased BMI? Methods: � Line 90-91 Briefly, the 2019 survey (HINTS 5 Cycle 3) was conducted from January through April 2019, and the 2020 survey (HINTS 5 Cycle 4) was conducted from February through June 2020. Reviewer comments: COVID-19 was declared pandemic on March 11th, 2020, and lockdown has followed in most of the world regions. Therefore, if you are aiming to compare the rates and prevalence of cardiometabolic diseases, that are mainly linked to lifestyle behavioral risk factors taking time to accumulate and changing health conditions, before and during COVID-19, the results of this study might be biased and under-estimated. � Line 124-125 The number of days per week of moderate intensity physical activity (none and at least one day per week) Reviewer comments: Why have you chosen to ask none or at least one day per week? Physical activity significance should hit the international recommendations of 150 mins/week. Therefore, performing one time/week or so will not add significant information, hence correlation. If you are not using a valid tool to measure your variables, you will be subjected to bias. It is better to question about the days and minuets and calculate the mean. Results Line 147-151 prevalence of diabetes was higher during the COVID-19 pandemic (18.10%) than before the 150 pandemic (17.28%). However, the prevalence of hypertension (36.36% vs. 36.38%) and obesity 151 (34.68% vs. 34.18%) was similar during and before the pandemic. Reviewer comments: Again, the slight variations in the prevalence of cardiometabolic diseases between before and during the pandemic is most likely due to early trials on lifestyle related behavior that work in cumulative effects manner (developing over longer period) developing chronic diseases. � Line 163-166 Individuals who were former cigarette smokers or current smokers had increased prevalence of cardiometabolic conditions. For e cigarette use groups, the prevalence had increased for those who had never used e-cigarettes and those who currently used e-cigarettes however decreased for former e-cigarette users. Reviewer comments: The prevalence was increased for those who never smoked e-cigarettes and those who currently use e cigarette, is bringing so much confusion for the reader and later to decision makers. You need to revisit your analysis or at least explain your odd findings in the discussion section comparing to previous research. Discussion � Line 200-202 Reviewer comments: Discussion needs to be improved. 1\. Correlation of age and risk of cardiometabolic conditions is not well highlighted. First, your study revealed that age group from 26- ≥65 year are at increased risk when compared to 18-25 years. I can notice that the risk is doubled for 26-34 year and tripled for 35-49 year. Therefore, in your recommendations, you need to focus not only on older age group but middle-aged as well. 2\. You need to enrich discussion further to add new implications of your results. Since the findings are not adding additional knowledge to the literature, you need to discuss your odd findings and give explanations. You may discuss how to improve future research in the same area e.g., by adopting validated tools to measure variables, by using more reliable source of data like registry or objective measures as compared to self-reported. Suggestions in how to improve the internal validity of the results are important and show your understanding of pitfalls. Therefore, add a section of “implications to practice & research’. Limitations: Kindly add the limitations discussed above and the probable effect of confounders not considered in this study such as sedentary behavior, sleep, eating patterns, employment and etc., that might give rise to inaccurate estimates of the true association. Overall, the manuscript needs improvement in English writing, linking ideas, relating results to previous findings, and most importantly providing explanation of each finding that disagree with the existing knowledge or literature. Since the study design is cross-sectional, the main finding we are looking for is the prevalence of cardiometabolic conditions before and compare it to during the pandemic. However, your findings are not impressive and not reflecting the actual impact of COVID-19 on NCDs burden due to methodology reasons highlighted above. For this, you have to enrich your paper with additional values such as using the analysis of predictors and justifications of such findings. Add the following article in your referencing (linking COVID-19 to behavioral risk factors which increase the risk of cardiometabolic conditions) COVID-19 and screen-based sedentary behaviour: Systematic review of digital screen time and metabolic syndrome in adolescents \| PLOS ONE Available in PubMed also: COVID-19 and screen-based sedentary behaviour: Systematic review of digital screen time and metabolic syndrome in adolescents - PubMed (nih.gov) \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Martin Gulliford Reviewer \#2: **Yes: **Sarah Rashid Musa \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0279442.r002 Author response to Decision Letter 0 16 Nov 2022 Response to Review of Manuscript Number: PONE-D-22-21818 Reviewer \#1: This paper evaluates the population prevalence of diabetes, obesity and hypertension in the US during 2019 and 2020. The paper concludes that the prevalence of these conditions increased during the Covid-19 pandemic. Comment 1\. The title seems to imply that observations were made on Covid-19 status of individuals, which is not the case. Please adjust the wording of the title to make this clear. Response We appreciate your comment. Our title implies the observations made before and during the COVID-19 pandemic: “The prevalence of cardiometabolic conditions before and during COVID-19 and its association with health and sociodemographic factors.” We have revised the title by adding “pandemic” to further clarify it as: “The prevalence of metabolic conditions before and during the COVID-19 pandemic and its association with health and sociodemographic factors.” Comment 2\. In the Abstract, make clear that the study compares 2019 and 2020 and no observations were made on Covid-19 as such. We know that diabetes has been increasing over many years, so this paper does not demonstrate any deviation from the underlying trend, consequently the conclusions that can be drawn are quite limited. Response We appreciate your comment. We have stated our study’s aim as: “We examined the prevalence and association of cardiometabolic conditions with health and sociodemographic factors before and during the COVID-19 pandemic.” We have revised the methods to further incorporate your suggestion as: “Data were drawn from the 2019 (N= 5,359) and 2020 (N= 3,830) Health Information National Trends Surveys on adults to compare observations before (2019) and during (2020) the COVID-19 pandemic.” Comment 3\. Introduction, lines 68-69, there needs to be more discussion of references 16 and 21 - what do these studies show? Also, how widespread and what were the timings of lockdown orders in the target population for this study. Response Thank you for this observation. We have revised the entire paragraph to incorporate your suggestion. For the lockdown timings, it varies for all studies in that paragraph and was summarized to prevent making that paragraph too long. Below is the explanation of the lockdown timings in the study. The timing of lockdown orders in reference 16 compared was before the lockdown and the 6th month of lockdown. The study population were a research cohort in the Istanbul Research and Education Hospital, Turkey, from March 2019 to October 2020. In reference 20, the study population included all patients with diabetes mellitus who visited the Tohoku Medical and Pharmaceutical University Hospital in Sendai, from January 1, 2019, to August 31, 2020. Japan declared a state of emergency on April 7, 2020, so we presume this is the date their lockdown order started, although they think a state of emergency differs from lockdown orders established by other nationals. In reference 21, the study population included outpatients at the Diabetology Unit of Humanitas Clinical and Research Center, IRCCS in Italy at baseline, between December 15, 2019, and March 1, 2020, and at the resumption of clinical activities, between May 15 and June 30, 2020. In reference 22, the study population included Type 2 diabetes mellitus patients unable to attend clinic follow-up visits due to the lockdown order in Turkey between March 16 and June 1, 2020, but attended follow-ups in July and August 2020 after the restriction had been lifted. Comment 4\. The outcome comprises hypertension, obesity and diabetes. No cardiac conditions are included, so this might be equated with the 'metabolic syndrome'. Response We agree with the reviewer; therefore, we have changed all “cardiometabolic” to “metabolic,” from the title to the conclusion. This is because we do not have data specifically on waist circumference, triglycerides, and HDL, therefore, using the term “metabolic syndrome” is not appropriate. As such, we will use the term “metabolic” instead of “cardiometabolic” as the reviewer suggested. Comment 5\. In the analysis, explain what weights were employed. Response Thank you. We have included weight information as: “The HINTS sampling weight was applied to the analysis to achieve population estimates and offset nonresponse.” (see the first sentence in the statistical analysis section) Comment 6\. Using a cut-off of P \<0.05 is an out of date approach to interpretation. Please follow the ASA guidelines on P values. <https://www.tandfonline.com/doi/full/10.1080/00031305.2016.1154108> Response We appreciate your suggestion. We would like the reviewer to know that our decision on whether a result is statistically significant or not was informed by the p-value and the Confidence Interval (CI). As such, our decision-making approach is consistent with the recommendation by the American Statistical Association \[<https://www.tandfonline.com/doi/full/10.1080/00031305.2016.1154108>\]. Further, as recommended by Leo and Sardanelli \[<https://eurradiolexp.springeropen.com/articles/10.1186/s41747-020-0145-y>\], we have shown the actual p-value so that the reader could determine the extent of the association. Thus, our reporting of the results addresses the statistical issues raised by the reviewer. Comment 7\. Figure 1 may be better as a Table. If included it will be preferred to show the results for 2019 and 2020 side by side so these can be more readily compared. Response Thank you for the suggestion. We believe that a figure could better depict the patterns of the outcomes than a table and, as such, we would like to retain the figure. However, if the reviewer still feels that we should replace the figure with the table, we would be pleased to do. Comment 8\. Table 1. More appropriate column headers may be '2019' and '2020'. Response Thank you. We have included the 2019 and 2020 in the column headers. Comment 9\. Table 2: if the intention is to see whether associations differed in 2019 and 2020, it would be better to include all the data in a one model and test for the interaction of each variable with study year. Most of the associations appear to be quite similar across years. Response We appreciate your comment. Given the differences in the population in 2019 and 2020 based on the outcomes, we could not include all the data in one model. Respectfully, we would like to maintain the table in its current form to avoid any statistical falsification. Comment 10\. In the Discussion, what can be concluded is very limited because the study has not evaluated secular trends over time. The difference between 2019 and 2020 could be accounted for by the underlying trend. Response We agree with you. As such, we have toned down on our interpretations and reasoning by not making causal inferences or making inferences beyond our results. Comment 11\. In the Limitations section, mention that no causal inferences can be drawn. Response Thank you. We have included your suggestion in the limitation section. Comment 12\. In the conclusion, where it says 'Disparities in cardiometabolic conditions became more evident after the pandemic', there does not appear to be sufficient evidence from the analysis to support this conclusion. Response Thank you for pointing this out. We have revised the statement as: “The prevalence of metabolic conditions increased during the COVID-19 pandemic in certain subgroups of individuals. Specifically, there was an increased risk of metabolic conditions associated with older age. Other groups with signals for increased risks include non -Hispanic Black people, former smokers, individuals with poor health status, and mild anxiety.” Comment 13\. Where it says ' the prevalence of cardiometabolic conditions increased during the COVID-19 pandemic', only diabetes increased not the other conditions, and we do not know whether the increase exceeded pre pandemic expectations. Please address this text in the Abstract also. Response We have revised this in the abstract also as suggested: “This study found increased odds of metabolic conditions among certain subgroups of U.S. adults during the pandemic” Reviewer \#2: Introduction: The introduction is lacking the part linking the risk factors especially mental health status during pandemic to cardiometabolic diseases and how COVID-19 pandemic has led to an increased adoption of such behavioral risk factors e.g., physical inactivity and smoking. � Line 62-64 COVID-19 may have an exacerbating effect on glycemic control for patients with diabetes \[14, 20\], and there may be risk of increased body mass index (BMI) as well as a deterioration in glucose regulation due to COVID-19 \[16, 21\]. Response: Thank you for your comment. This comment has been incorporated into the introduction section of the revised manuscript. Reviewer comments: Will you clarify if COVID-19 infection or the implications of COVID-19 related lockdown have led to exacerbation of glycemic control and increased BMI? Response: Thank you for your comment. This statement has been clarified in the introduction of the revised manuscript. Methods: � Line 90-91 Briefly, the 2019 survey (HINTS 5 Cycle 3) was conducted from January through April 2019, and the 2020 survey (HINTS 5 Cycle 4) was conducted from February through June 2020. Reviewer comments: COVID-19 was declared pandemic on March 11th, 2020, and lockdown has followed in most of the world regions. Therefore, if you are aiming to compare the rates and prevalence of cardiometabolic diseases, which are mainly linked to lifestyle behavioral risk factors taking time to accumulate and changing health conditions, before and during COVID-19, the results of this study might be biased and under-estimated. Response: Thank you for pointing out this information. We agree with the reviewer and have added the recommendation as a limitation of the study (see the limitation section). � Line 124-125 The number of days per week of moderate intensity physical activity (none and at least one day per week) Reviewer comments: Why have you chosen to ask none or at least one day per week? Physical activity significance should hit the international recommendations of 150 mins/week. Therefore, performing one time/week or so will not add significant information, hence correlation. If you are not using a valid tool to measure your variables, you will be subjected to bias. It is better to question about the days and minutes and calculate the mean. Response We appreciate your comments. This variable is a standardized measure in HINTS and based on the number of days per week of moderate intensity physical activity. The “150 mins/week” is used to derive the “moderate physical activity” either per day or week, hence the choice for this study. Results Line 147-151 prevalence of diabetes was higher during the COVID-19 pandemic (18.10%) than before the 150 pandemic (17.28%). However, the prevalence of hypertension (36.36% vs. 36.38%) and obesity 151 (34.68% vs. 34.18%) was similar during and before the pandemic. Reviewer comments: Again, the slight variations in the prevalence of cardiometabolic diseases between before and during the pandemic is most likely due to early trials on lifestyle related behavior that work in cumulative effects manner (developing over longer period) developing chronic diseases. Response Thank you for noting this. This suggestion has been included in the limitations of the revised manuscript (see the limitation section). � Line 163-166 Individuals who were former cigarette smokers or current smokers had increased prevalence of cardiometabolic conditions. For e-cigarette use groups, the prevalence had increased for those who had never used e-cigarettes and those who currently used e-cigarettes however decreased for former e-cigarette users. The prevalence was increased for those who never smoked e-cigarettes and those who currently use e-cigarette, is bringing so much confusion for the reader and later to decision makers. You need to revisit your analysis or at least explain your odd findings in the discussion section comparing to previous research. Response Thank you! We revisited the analysis and found the same results. This finding is consistent with the tobacco literature: former tobacco users, including former e-cigarette users, are less likely to engage in unhealthy behaviors; therefore, they are less likely to develop health conditions such as cardiometabolic conditions (<https://nida.nih.gov/publications/research-reports/tobacco- nicotine-e-cigarettes/what-are-physical-health-consequences-tobacco-use;> <https://www.health.ny.gov/prevention/tobacco_control/;> Lowe et al., 2009; U.S. Department of Health and Human Services, 2014). However, the pandemic has altered many lifestyles, including tobacco use behaviors that might have also affected their risks for cardiometabolic conditions. Citations: Lowe, F. J., Gregg, E. O., & McEwan, M. (2009). Evaluation of biomarkers of exposure and potential harm in smokers, former smokers, and never-smokers. Clinical chemistry and laboratory medicine, 47(3), 311-320. U.S. Department of Health and Human Services (2014). The Health Consequences of Smoking—50 Years of Progress: A Report of the Surgeon General. Atlanta, GA: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health. <https://www.hhs.gov/sites/default/files/consequences-smoking-exec-summary.pdf> Discussion � Line 200-202 Reviewer comments: Discussion needs to be improved. 1\. Correlation of age and risk of cardiometabolic conditions is not well highlighted. First, your study revealed that age group from 26- ≥65 year are at increased risk when compared to 18-25 years. I can notice that the risk is doubled for 26-34 year and tripled for 35-49 year. Therefore, in your recommendations, you need to focus not only on older age group but middle-aged as well. Response Thank you! We have included discussions on these middle-aged groups as suggested in the revised manuscript. For instance, we found that the odds of cardiometabolic outcomes were significantly higher only among the elderly age groups (50-64, and 65+) compared to young adults before the pandemic. However, these increases in odds almost doubled among these age groups during the pandemic. Significantly higher odds were also noted in the middle age groups (26-34 and 35-49), where the odds almost tripled for the 35-49 age group during the pandemic. Comment: 2\. You need to enrich discussion further to add new implications of your results. Since the findings are not adding additional knowledge to the literature, you need to discuss your odd findings and give explanations. You may discuss how to improve future research in the same area e.g., by adopting validated tools to measure variables, by using more reliable source of data like registry or objective measures as compared to self-reported. Suggestions in how to improve the internal validity of the results are important and show your understanding of pitfalls. Therefore, add a section of “implications to practice & research.” Response Thank you for this suggestion. We have added a new section of “Implications to Practice and Research,” including limitations of the study, in the discussion section. Regarding the issue of validity and reliability, as in most national surveys, the measures in HINTS have been validated and widely used since 2002/2003. We have, however, noted the limitation of using self-reported measures and recommended using objective measures in future studies. Regarding the internal validity of the results, we followed the analytical recommendations of HINTS in computing accurate estimates. As such, we believe our estimates are accurate and valid. Limitations: Comment Kindly add the limitations discussed above and the probable effect of confounders not considered in this study such as sedentary behavior, sleep, eating patterns, employment and etc., that might give rise to inaccurate estimates of the true association. Response Thank you. We have incorporated your suggestions in the revised manuscript (see the limitation section). Comment Overall, the manuscript needs improvement in English writing, linking ideas, relating results to previous findings, and most importantly providing explanation of each finding that disagree with the existing knowledge or literature. Response We have thoroughly reviewed the entire paper by doing line-by-line editing. Additionally, we asked our colleague not familiar with the study to review the paper for language. Comment Since the study design is cross-sectional, the main finding we are looking for is the prevalence of cardiometabolic conditions before and compare it to during the pandemic. However, your findings are not impressive and not reflecting the actual impact of COVID-19 on NCDs burden due to methodology reasons highlighted above. For this, you have to enrich your paper with additional values such as using the analysis of predictors and justifications of such findings. Response We agree with the reviewer that the cross-sectional data is a limitation of this study. As such, we have included this limitation in the limitations section of the revised manuscript. In future studies, we will consider longitudinal data for the analysis of predictors and justifications of such findings. Additionally, we have enhanced the discussion by relating our study to the extant literature and illuminating its added value to the growing literature on COVID-19. Comment Add the following article in your referencing (linking COVID-19 to behavioral risk factors which increase the risk of cardiometabolic conditions) COVID-19 and screen-based sedentary behaviour: Systematic review of digital screen time and metabolic syndrome in adolescents \| PLOS ONE Available in PubMed also: COVID-19 and screen-based sedentary behaviour: Systematic review of digital screen time and metabolic syndrome in adolescents - PubMed (nih.gov) Response The article has been incorporated in the revised manuscript. Specifically, we have incorporated the following in the discussion section: “Sedentary behaviors such as screen time which limits physical activity \[50, 51\], increases the risk of cardiometabolic diseases. A recent systematic review assessing screen-based sedentary behavior among adolescents during the COVID-19 pandemic reported a dose-response association between increased level of screen time and components of metabolic syndrome \[Musa et al\]” 10.1371/journal.pone.0279442.r003 Decision Letter 1 Kim Taeyun Academic Editor 2023 Taeyun Kim This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 28 Nov 2022 PONE-D-22-21818R1The prevalence of metabolic conditions before and during the COVID-19 pandemic and its association with health and sociodemographic factorsPLOS ONE Dear Dr. Mamudu, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jan 12 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Taeyun Kim Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: 1. In the Abstract, suggest changing 'rationing' to 'allocation'. 2\. In the Abstract, where it refers to ' older adults, non-Hispanic Black people... individuals with poor health status', no data are presented to support this statement in the Abstract. Eitehr include supporting evidence in the Abstract or omit. Reviewer \#2: Thank you for addressing every point and incorporating them into your revised manuscript. The paper sounds much better indeed with feedback of reviewer 1 as well. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: **Yes: **Dr. Sarah Musa \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0279442.r004 Author response to Decision Letter 1 30 Nov 2022 Reviewer \#1: Comment 1\. In the Abstract, suggest changing 'rationing' to 'allocation'. Response We have changed it as suggested. Comment 2\. In the Abstract, where it refers to ' older adults, non-Hispanic Black people... individuals with poor health status', no data are presented to support this statement in the Abstract. Either include supporting evidence in the Abstract or omit. Response We have deleted the emphasis as suggested. Reviewer \#2: Comment Thank you for addressing every point and incorporating them into your revised manuscript. The paper sounds much better indeed with feedback of reviewer 1 as well. Response Thank you for helping to improve the paper substantively and stylistically. 10.1371/journal.pone.0279442.r005 Decision Letter 2 Kim Taeyun Academic Editor 2023 Taeyun Kim This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 7 Dec 2022 The prevalence of metabolic conditions before and during the COVID-19 pandemic and its association with health and sociodemographic factors PONE-D-22-21818R2 Dear Dr. Mamudu, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Taeyun Kim Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0279442.r006 Acceptance letter Kim Taeyun Academic Editor 2023 Taeyun Kim This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 2 Feb 2023 PONE-D-22-21818R2 The prevalence of metabolic conditions before and during the COVID-19 pandemic and its association with health and sociodemographic factors Dear Dr. Mamudu: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Taeyun Kim Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction The entry of HIV-1 into susceptible host cells is mediated by the viral envelope glycoprotein (Env). The functional envelope spike is comprised of three non- covalently associated gp41 transmembrane sub-units and three gp120 surface units. Each gp120 molecule is divided into five constant (C1-C5) and five variable (V1-V5) domains, and contains a binding site for the primary cell- surface receptor CD4. Following CD4-engagement, the envelope undergoes a conformational change that exposes or creates a binding site for its co- receptor, typically CCR5 (R5) or CXCR4 (X4). Binding to the co-receptor triggers conformational changes in gp41 that ultimately result in fusion of the virus and host-cell membranes \[reviewed in. Virtually all HIV-1 infections are established by exclusively R5-using virus, regardless of the presence of R5/X4 or obligate X4 virus in the index case. Furthermore, approximately 1–2% of persons of Northern European descent are homozygous for a 32 base pair deletion in the CCR5 gene (CCR5Δ32) which alters CCR5 expression on the surface of their cells and renders them highly resistant to HIV. Persons heterozygous for CCR5Δ32 have lower cell-surface expression of CCR5, are partially resistant to infection, and tend to progress slower if infected. This rigid constraint on an otherwise fluid and rapidly evolving virus has led to the development and testing of many interventions targeting CCR5 for prevention, treatment, and even cure. Unfortunately, these efforts are at risk of failure should the virus successfully transition to efficient X4 utilization. A better mechanistic understanding of the R5-to-X4 transition will allow scientists and clinicians to better predict, and potentially counter, such an escape strategy. The determinants of co-receptor usage map primarily to the V3 loop; which makes extensive molecular contacts with the co-receptor. It has been well-established that an overall shift towards positive charge, but especially positively charged substitutions at positions 11, 24, and 25 in V3 are predictive of X4 utilization, as is the presence of an isoleucine at position 326 in the V3 stem. Several algorithms have been developed to predict co-receptor usage based on V3 sequence, with accuracies estimated at 70–80%. However, substitutions in other regions including the bridging sheet, C4, V1/V2, and gp41 have also been shown to influence co-receptor usage. For this study, we screened a cohort of treatment-experienced, predominantly subtype-B infected subjects failing their current ARV regimens, using resistance to the CCR5-antagonist maraviroc (MVC) as an initial surrogate marker for efficient X4 utilization. We reasoned that this population would be more likely to harbor the X4 using or transitional variants of interest. Convenience sampling of ten subjects identified three with MVC-resistant virus. From one of these three, we isolated a series of closely related molecular envelope clones with identical V3 sequences but highly variable co-receptor usage. Further *in vitro* characterization revealed that X4 utilization was regulated by polymorphisms in C1 and C2. The C2 polymorphism disrupted a conserved potential N-linked glycosylation site (PNG) important for envelope function but not previously linked to co-receptor selectivity. # Materials and Methods ## Study Population All subjects were treatment experienced and screened for, but unable to enroll in, IMPAACT protocol P1020a. Written informed consent was obtained by study candidate, parent or legal guardian prior to screening and recorded per protocol. The study was approved by the Institutional Review Boards at each investigator site (see listing in acknowledgements and manuscript PMID 25232777) and registered with ClinicalTrials.gov, Identifier NCT00006604. No subjects had prior exposure to entry inhibitors (including Maraviroc). ## Cells and Reagents 293T/17 retroviral packaging cells were obtained from the American Type Culture Collection (ATCC, cat# CRL-11268). TZM-bl cells, a HeLa clone expressing high levels of CD4, CCR5, and CXCR4 as well as ß-galactosidase and firefly luciferase reporter genes under the control of the HIV promoter, were obtained from the NIH AIDS Reagent Program (ARRRP), Division of AIDS, NIAID, NIH from Dr. John C. Kappes, Dr. Xiaoyun Wu, and Tranzyme Inc. (cat# 8129). Parental GHOST cells, as well as GHOST-R5, GHOST-X4, and GHOST-R3/X4/R5 subclones were obtained from the ARRRP (cat#s 3679, 3944, 3685, and 3943) from Dr. Vineet N. Kewal Ramani and Dr. Dan R. Littman. All cell lines were maintained in Dulbecco’s Modified Eagle’s Medium (DMEM; Fisher Scientific, Waltham, MA) supplemented with 10% fetal bovine serum (FBS; Gemini Bio-products, Sacramento, CA), 100 U/mL penicillin- streptomycin (Gibco, Invitrogen, Carlsbad, CA), and 2 mM L-glutamine (Gibco). GHOST cells were supplemented with 500 μg/mL G418 (Invitrogen) and 100 μg/mL hygromycin (Invitrogen), and 1 μg/mL puromycin (GHOST-R5, GHOST-X4, and GHOST-R3/X4/R5; Sigma-Aldrich, St. Louis, MO). MVC (Selzentry) was generously donated by Pfizer. TAK779, AMD3100, and T20/Fuzeon (courtesy of Roche) were obtained from the ARRRP (cat#s 4983, 8128, and 9845). AD-101 (SCH-350581) was generously donated by Schering-Plough. Soluble lectins were purchased from commercial suppliers: Vector Laboratories Inc. \[Burlingame, CA; *Hippeastrum* hybrid (Amaryllis) agglutinin (HHA) and *Galanthus nivalis* agglutinin (GNA), cat \#s L-1240 and L-1380\] and EY Laboratories \[San Mateo, CA; *Urtica dioica* agglutinin (UDA), cat# L-8005-1\]. ## Cloning and Mutagenesis RNA was extracted from plasma using QIAamp Viral RNA Mini Kit (Qiagen, Valencia, CA). Following reverse transcription using the SuperScript III system (Invitrogen) and the primer envB3out (`5′-TTGCTACTTGTGATTGCTCCATGT-3′`), cDNAs were serially diluted until \<30% of reactions were positive as previously described. Full-length gp160s were amplified by nested PCR using high-fidelity polymerase and the primers envB5out (`5′-TAGAGCCCTGGAAGCATCCAGGAAG-3′`)/envB3out for the first round, and envB5in (`5′-TTAGGCATCTCTATGGCAGGAAGAAG-3′`)/envB3in (`5′-GTCTGAGATACTGCTCCCACCC-3′`) for the second. Cycle timings were: first round \[94°C × 2 min, 35 cycles of (94°C × 15 sec, 55°C × 30 sec, 68°C × 4 min), and 68°C × 20 min\] and second round \[94°C x 2 min, 45 cycles of (94°C × 15 sec, 55°C × 30 sec, 68°C × 4 min), and 68°C × 20 min\]. Amplicons were ligated into the pcDNA 3.1 Topo Expression vector (Invitrogen) and transformed into TOP10 E. Coli (Invitrogen) using standard molecular biology techniques. Envelope clones were sequenced bi-directionally, and the contigs assembled and edited using Sequencher (Gene Codes, Ann Arbor, MI). Alignments were constructed using MacVector (MacVector Inc, Cary, NC) and the LANL website tools (<http://www.hiv.lanl.gov/content/sequence/HIV/HIVTools.html>). All amino acid residues are numbered based on the HXB2 reference sequence. Point mutations were introduced by site-directed mutagenesis (QuikChange multi-site directed mutagenesis kit; Stratagene, La Jolla, CA) according to the manufacturer’s instructions and sequenced to confirm integrity. For the remainder of the manuscript, un-mutated envelopes derived from subject plasma will be referred to as ‘parental’. Protease inhibitor (PI), nucleoside reverse transcriptase inhibitor (NRTI), and non-nucleoside reverse transcriptase inhibitor (NNRTI) resistance data were obtained from the IMPAACT P1020 screening records and were determined using the ViroSeq HIV-1 Genotyping System (ABI/Life Technologies, Foster City, CA) and the Stanford HIV drug resistance database (<http://hivdb.stanford.edu/>). The V3 loop sequence of each envelope was analyzed using the web-based tools Geno2pheno (<http://co-receptor.bioinf.mpi-inf.mpg.de/index.phpare>), position- specific scoring matrix (PSSM, <http://indra.mullins.microbiol.washington.edu/webpssm>), and the 11KR/25KR and 11/24/25 net charge rules to predict co-receptor usage \[, , –\]. These methods classify HIV-1 strains as either R5 or X4. The latter designation includes variants that use X4 exclusively as well as those capable of using both R5 and X4 for entry. ## Virus Pseudotyping The env-deficient HIV-1 genome plasmids SG3ΔEnv (cat# 11051) and pNL4-3.Luc.R^-.E^- (carrying a luciferase reporter gene; cat# 3418) were obtained through the AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH from Dr. John C. Kappes and Dr. Xiaoyun Wu, and Dr. Nathanial Landau respectively. The backbone plasmid bearing the green fluorescence protein (GFP) reporter gene, NLENG1-ES-IRES, was kindly provided by Dr. David Levy. ## Inhibition and Entry Assays Pseudoviruses were produced by co-transfection of 293T/17 cells with the patient-derived Env-expressing plasmids, and either SG3Δenv, pNL4-3.Luc.R^-E^- or NLENG1-ES-IRES and titered by X-gal staining of TZM-bl cells as previously described. All cell entry assays (no primary cell data were included) were performed using pseudotype virus in a 96-well plate format as previously described. Briefly, using titers obtained by X-gal staining, \~2000 units of pseudovirus in 16 μg/mL DEAE Dextran were added to each well containing target cells. Luminescence was quantified after a 48-hour incubation using a Promega Luciferase Kit (Promega, Madison, WI) and a FLUOstar luminometer (BMG Labtech, Cary, NC) according to manufacturer instructions. Phenotypic determination of co-receptor utilization was performed by infecting TZM-bl cells in the presence or absence of AMD3100 (X4 antagonist) or TAK-779 (R5 antagonist), and by infection of GHOST cells (expressing CD4 and one or both of the co-receptors) with pNL4-3.Luc.R^-E^-derived pseudoviruses. Drug sensitivity was determined by pre-treatment of TZM-bl cells for 1 hour at 37°C with increasing concentrations of the CCR5 antagonists MVC, TAK-779, AD101, the X4 inhibitor AMD3100, the fusion inhibitor T20, or the lectins HHA, GNA, or UDA prior to addition of pseudovirus. ## Data Analysis Drug sensitivity was quantified by calculating a mean 50% inhibitory concentration (IC₅₀) for each envelope/drug combination with the Prism 4 software (GraphPad, San Diego, CA) using data from at least two experimental replicates. Co-receptor usage, based on TZM or GHOST cell methods was expressed as a percentage normalized to entry in the absence of inhibitors in cells expressing both co-receptors (GHOST-R3/X4/R5 for that platform). # Results ## Clinical Characteristics of Study Subjects The clinical characteristics of our cohort are listed in. All subjects were experiencing virologic failure (median plasma HIV RNA of 106,808 copies/ml) on their current ARV regimen, with at least moderate resistance to two or more drug classes as determined by genotypic analysis using the Stanford HIV drug resistance database. Of the ten subjects included in our initial round of screening, nine were infected with subtype B and one (subject 1002) was infected with subtype C HIV-1. ## Identification of a Subject with both MVC-Resistant and MVC-Sensitive Virus Full-length molecular envelope clones were generated from the plasma of all ten subjects by RT-PCR and fully sequenced. V3 loops were analyzed using the web- based algorithms Geno2Pheno, PSSM, and 11/24/25. Pseudotype viruses were generated from all envelopes and tested for resistance to MVC, which served as a surrogate marker for X4 utilization at the screening stage. Data are presented in. Viral envelopes isolated from seven of the ten patients screened were predicted *in silico* to be exclusively R5-using and were phenotypically sensitive to MVC, thus no further studies were performed on them. Seven clones isolated from subject 1009 were predicted to use X4 by both Geno2Pheno and PSSM and were highly resistant to MVC. Two clones were isolated from subject 1005 with highly divergent V3 loop sequences, one of which was predicted to use X4 by Geno2Pheno and PSSM and was MVC-resistant, while the other was predicted to use X4 by Geno2Pheno but R5 by PSSM, and was sensitive to MVC. Finally, five clones were isolated from subject 1010, all with identical V3 loop sequences that were predicted to use X4 by Geno2Pheno but R5 by PSSM. Two of those clones were sensitive to MVC (IC₅₀ \~2nM) while the other three were highly resistant (IC₅₀ \>200nM). We chose to focus on the isolates from this last subject for further examination. These data are summarized in. ## Identification of Non-V3 Loop Co-receptor Use Determinants in Subject 1010 The five *env* clones from subject 1010, which had identical V3 sequences, differed at nine positions throughout the remainder of gp160 (five residues in gp120 and four in gp41;). Only two positions correlated with MVC sensitivity. The first was the highly conserved bridging sheet residue 123, at which the MVC- sensitive isolates had an unusual isoleucine substitution for the conserved threonine (T123I). Polymorphisms at this site were present in only 6 of 1501 complete subtype B sequences retrieved from the 2012 LANL database (two T123I and four T123A). The second position was the highly conserved serine 264 in the C2 domain, which was mutated to a glycine in the MVC-sensitive isolates (S264G). Polymorphisms at this site were similarly rare, present in only 7 of 1501 subtype B LANL sequences (one S264G, one S264N, three S264T, and two nonsense mutations). For simplicity, envelopes will be referred to by the amino acids at positions 123 and 264 (i.e. I+G or T+S) for the remainder of the manuscript (as shown in). It is important to note here that serine 264 is part of a PNG sequon, \[NX(S/T)\] including the conserved asparagine at 262, which is disrupted by the S264G polymorphism. This glycan has been localized to the outer domain of gp120, where it forms a glycan cluster with N295, N332, and N448 \[most recently in the structure described in\]. To verify that these isolates were indeed using X4 for entry and not simply able to productively utilize MVC-bound CCR5 as described in, they were tested on cells pre-treated with two additional CCR5 inhibitors, TAK779 and AD101, to which they were also resistant. Envelopes were also tested for entry into GHOST cells expressing CD4 and either CCR5 or CXCR4 alone, with results concordant with those obtained from the TZM-based inhibitor assays (summarized in **)**. These results confirmed that we identified clonal envelope variants with highly divergent X4 usage regulated by polymorphisms outside the V3 loop. ## Site-Directed Mutagenesis Reveals a Novel Regulator of X4 Usage We created a panel of mutant envelopes in which the residues at 123 and 264 were exchanged individually and in combination between the MVC-resistant/X4-using (T+S) and MVC-sensitive/R5-using (I+G) variants. This panel was screened using MVC, TAK779, AD101, AMD3100, and assayed for entry into R5- and X4-expressing GHOST cells. Results are presented in. The exchange of both residues (123 and 264) was sufficient to confer MVC- resistance and X4-utilization on the sensitive/R5 envelopes, and vice versa. The intermediate genotype T+G (which lacked a PNG at 262) was relatively sensitive to R5 inhibitors, but demonstrated elevated efficiency of X4-usage compared to the fully MVC-sensitive envelopes (I+G). In contrast, the intermediate genotype I+S (which possessed a PNG at 262) was relatively resistant to R5 inhibitors and had an X4-utilization efficiency much closer to the fully MVC-resistant (T+S) clones. These data indicate that exchange of residues at positions 123 and 264 is sufficient to confer full X4-tropism, and that position 264 is the dominant molecular switch. Only one mutant, 547 I+G, was non-functional for pseudotyping in 293 cells. I+G mutants made using the other parental X4 envelopes (543 I+G and 544 I+G) were fully functional, as was the mutant 547 T+G. The most apparent difference between the parental 547 and 543/544 envelopes is the presence of a conserved PNG at position 392 in clone 547 that is absent in 543/544. Interestingly, this PNG was also absent in the parental R5 variants (542/550), both of which have the I+G genotype. To determine whether a PNG at position 392 was functionally incompatible with the I+G genotype we constructed a second round of mutants, disrupting the PNG sequon with an S394N substitution in the non-functional mutant 547 I+G and adding a PNG site to clone 542 (a parental I+G) at position 392 via the reciprocal N394S substitution. Surprisingly mutant 547 I+G S394N remained non-functional for pseudotyping, while mutant 542 N394S retained function and was similarly sensitive to R5 inhibition as the parental 542 variant (data not shown). The functional and non-functional variants differed at only two additional residues, a glycine-to-aspartic acid at position 395, and a serine-to-glycine at position 668 in the gp41 membrane-proximal external region. Changes in T20 susceptibility, in the absence of gp41 mutations that directly affect binding, have been associated with altered fusion kinetics. T20 binds to the fusion-intermediate conformation of gp41 and prevents final collapse of the gp41 trimer into the 6-helix bundle required for membrane fusion and entry. To determine whether any of our parental or mutant envelopes had gross defects in their entry kinetics, we tested the full panel for susceptibility to T20. We did not observe any significant differences in T20 susceptibility between R5 and X4 variants, nor associated with any of our in vitro mutants. ## Glycosylation of Asparagine 262, Rather Than the Specific Identity of the Residue at Position 264, Controls Co-Receptor Usage and Effects Neutralization by Mannose-, but Not Complex Glycan-, Dependent Soluble Lectins To confirm that the S264G substitution was indeed acting by elimination of the glycan attached to asparagine 262, we constructed a set of mutants designated I+(Q)S and T+(Q)S, in which the PNG sequon was disrupted by an N262Q substitution that left the serine at position 264 intact. The I+(Q)S and T+(Q)S mutants behaved identically to the I+G and T+G variants (all of which lack a PNG at 262). We constructed other mutants, designated I+T and T+T, in which the serine at position 264 was changed to a threonine, preserving the PNG sequon. The I+T and T+T mutants behaved like the I+S and T+S variants (all of which have a PNG at 262). These data confirm that it is the presence of a PNG at position 262, and not the specific identity of the residue at position 264, that regulates co-receptor usage in these clonal variants. We also tested the parental and mutant envelopes with the soluble plant-derived lectins GNA, HHA, and UDA. GNA and HHA both bind high-mannose glycans, while UDA binds complex-type glycans. All three lectins neutralized all tested envelopes. Activity of the two mannose-dependent lectins, GNA and HHA, correlated well with the activity of the R5 antagonists (R² 0.68–0.79): the most lectin- resistant variants had the T+S genotype, the moderately resistant variants the I+S genotype, and the most sensitive had either the I+G or T+G genotypes. Resistance to the complex-glycan-dependent lectin UDA did not vary significantly among any of our isolates, nor correlated with the activity of any of the R5 antagonists (R² 0.01–0.05). It is important to note that previously published biochemical analyses indicate that the glycan at position 262 is predominantly high mannose when produced in 293 cells. Correlation plots are presented in. # Discussion In this study, we analyzed a series of clonal HIV-1 envelope variants with identical V3 loops but varied co-receptor usage. We found that substitutions at conserved residues in the bridging sheet (position 123) and the C2 domain (PNG at position 262) were responsible for regulating X4 usage in these isolates. To our knowledge, neither of these positions has been previously identified as a regulator of co-receptor utilization. It is noteworthy that in this case the atypical (I+G) variants are the more phenotypically ‘normal’ (MVC-sensitive and R5-using), while the genotypically normal (T+S) variants are phenotypically unusual (MVC-resistant with efficient X4-utilization). This limits our ability to draw broad conclusions about X4 evolution in general, but does point to a previously unappreciated role of the N-glycan at 262 in governing the conformation or exposure of the co-receptor binding site. Previous studies have identified this glycan as important for viral infectivity, and it will be of interest to determine what compensatory mutations have emerged in these naturally occurring variants. The lack of effect resulting from exchange of the PNG at 392 between functional and non-functional I+G variants was surprising to us. The conserved N-glycan at position 392 is present in only one of the three X4-using clones and its absence does not affect any measure of co-receptor usage that we tested. Additional studies will be required to determine what functional relationship exists between PNG 392 and positions 395/668, and whether those interactions are strain-specific or generalizable. One limitation of our study is the lack of longitudinal sampling. While it is very likely this subject was originally infected with an R5 virus, we cannot be certain without testing earlier samples. If this were indeed the case, it would be of interest to know whether the transmitted isolate possessed a PNG at 262 and, if not, how the acquisition of that PNG is temporally related to the emergence of a strong X4 phenotype. Moreover, if the transmitted isolate was indeed exclusively R5-using and evolved to utilize X4 while still possessing a PNG at 262, then dissecting the selective forces that favored some variants loosing their ability to use X4, as well as the conserved PNG at 262, may shed more light on the underlying biology of co-receptor selection. It is also of interest that the X4 clones from this subject (both parental and mutant) were relatively sensitive to all three CCR5 antagonists (MVC, TAK779, and AD101), not achieving an IC50 \>200nM until the strain had \>50% predicted X4 usage. Examination of TZM neutralization curves (exemplar shown) indicated that both R5 inhibitor-sensitive and R5 inhibitor-resistant variants had similar slopes but reached plateaus at the approximate maximum-percent-inhibition predicted by the GHOST cell assay (hence the inability to determine an IC₅₀ value for envelopes which used X4 very efficiently). Given that MVC, AD101, and TAK779 bind CCR5 differently and may act through different mechanisms, these data suggest that the envelopes which utilized X4 efficiently maintained a conventional interaction with CCR5, rather than shifting their binding to a different part of the receptor or increasing affinity. Considering that all variants isolated from this subject had identical V3 loops, genotypically predicted to use X4, we speculate that they represent a lineage of V3 sequences that are in fact highly adapted to using both co-receptors, but for which the polymorphism at 123 and the loss of the PNG at position 262 selectively compromise X4 binding. It is also possible, since we used assays with productive infection as their endpoint, that the described polymorphisms do not limit X4-binding *per se*, but instead compromise down-stream conformational changes which ultimately lead to more abortive entry events upon X4 binding, though we consider this alternative hypothesis unlikely. In summary, we have isolated and described a series of closely related, naturally occurring clonal envelope variants with identical V3 loops that can utilize both CCR5 and CXCR4 efficiently. These variants have two molecular switches, a threonine at position 123 and an N-glycan at position 262, that act cooperatively to permit or constrain X4 utilization without affecting CCR5 usage. This study is, to the best of our knowledge, the first published evidence that these two sites have a direct and significant influence on co-receptor utilization by HIV-1 envelope. # Accession Numbers Nucleotide sequences associated with this manuscript have been submitted to GenBank with accession numbers: KP693359–KP693388. # Supporting Information We would like to thank the IMPAACT P1020a team for providing the samples and clinical data that formed the foundation of this study. We are also grateful for the dedication and generosity of the site staff and participants, without whom this study could not have been done. We also thank Elizabeth Withers-Ward, Adrienne Rollie, and Shaun Yang for helpful discussion and critical reading of the manuscript. Participating protocol P1020a sites include Shandukani Research Clinical Research Site (CRS), Johannesburg, South Africa; Soweto IMPAACT CRS, Johannesburg, South Africa; New Jersey Medical School CRS, Newark, NJ; University of California at Los Angeles/Brazil AIDS Consortium CRS, Los Angeles, CA; Texas Children’s Hospital CRS, Houston, TX; Chicago Children’s CRS, Chicago, IL; Columbia IMPAACT CRS, New York, NY; University of Miami Pediatric/Perinatal HIV/AIDS, Miami, FL; University of California at San Diego Mother-Child- Adolescent HIV Program, San Diego, CA; Duke University Medical Center Pediatric CRS, Durham, NC; Metropolitan Hospital NICHD CRS, New York, NY; Children’s Hospital of Boston NICHD CRS, Boston, MA; Boston Medical Center Pediatric HIV Program, Boston, MA; New York University NY NICHD CRS, New York, NY; Jacobi Medical Center NICHD CRS, Bronx, NY; University of Washington/Children’s Hospital Seattle CRS, Seattle, WA; University of South Florida, Tampa, FL; Children’s Hospital of the King’s Daughters, Norfolk, VA; Mount Sinai Medical Center, New York, NY; University of Washington Medicine-Harborview Medical Center, Northwest Family Center, Seattle, WA; University of Washington NICHD CRS, Seattle, WA; San Juan City Hospital, San Juan, Puerto Rico NICHD CRS; State University New York Upstate Medical University, Syracuse, NY; State University New York Stony Brook, Stony Brook, NY; Wayne State University Detroit, Detroit, IL; Howard University Washington DC NICHD CRS, Washington D.C.; University of Southern California Los Angeles NICHD CRS, Alhambra, CA; University of Florida Jacksonville, Jacksonville, FL; University of Colorado Denver NICHD CRS, Aurora, CO; South Florida CDTC Fort Lauderdale NICHD CRS, Fort Lauderdale, FL; Strong Memorial hospital University of Rochester, Rochester, NY; Rush University Cook County Hospital Chicago NICHD CRS, Chicago, IL; University of California at San Francisco NICHD CRS, San Francisco, CA; Johns Hopkins University, Baltimore, Baltimore, MD; Miller Children’s Hospital Long Beach, CA; Tulane University New Orleans NICHD CRS, New Orleans, LA; University of Alabama Birmingham NICHD CRS, Birmingham, AL; St. Jude/University of Tennessee Health Science Center CRS, Memphis, TN; University of Puerto Rico Pediatric HIV/AIDS Research Program CRS, San Juan, PR; The Children’s Hospital of Philadelphia, Philadelphia, PA; Bronx- Lebanon Hospital IMPAACT, Bronx, NY; and WNE Maternal Pediatric Adolescent AIDS CRS, Worcester, MA. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: FL KJN GMA. Performed the experiments: FL KJN TC. Analyzed the data: FL KJN ERS NHT. Wrote the paper: FL KJN GMA NHT TC.

# Introduction In recent years, interest in identifying drug-target interactions has dramatically increased not only for drug development but also for understanding the mechanisms of action of various drugs. However, time and cost requirements associated with experimental verification of drug-target interactions cannot be disregarded. Many drug databases, such as DrugBank, KEGG BRITE, and SuperTarget, contain information about relatively few experimentally identified drug-target interactions. Therefore, other approaches for identifying drug-target interactions are needed to reduce the time and cost of drug development. In this regard, *in silico* methods for predicting drug-target interactions can provide important information for drug development in a reasonable amount of time. Various *in silico* screening methods have been developed to predict drug-target interactions. Among these methods, machine learning-based approaches such as bipartite local model (BLM) and MI-DRAGON which utilize support vector machine (SVM), random forest and artificial neural network (ANN) as part of their prediction model are widely used because of their sufficient performance and the ability to use large-scale drug-target data. For these reasons, many machine learning based prediction tools and web-servers have been developed. Especially, similarity-based machine learning methods which assume that similar drugs are likely to target similar proteins, have shown promising results. Although molecular docking methods also showed very good predictive performance, very few 3D structures of proteins are known, rendering docking methods unsuitable for large-scale screening. As such, a precise similarity-based method must be developed to predict interactions on a large-scale using the low-level features of compounds and proteins. Previous similarity-based methods, such as the bipartite local model (BLM), Gaussian interaction profile (GIP), and kernelized Bayesian matrix factorization with twin kernel (KBMF2K), provide efficient ways to predict drug-target interactions and have shown very good performance. BLM, which uses a supervised learning approach, has recently shown promising results using only similarities from each compound and each protein in the form of a kernel function. In the BLM method, the model for a protein of interest (POI) or compound of interest (COI) is learned from local information, which means that the model uses its own interactions of the COI or POI. This local-approach concept has been used in other methods, such as GIP, BLM-NII and others. Although such methods show very good performance, certain problems remain. Most previously developed methods categorize validated interactions between drugs and target proteins as positive, while unknown interactions are categorized as negative when constructing a predictive model. However, unknown interactions are not truly negative interactions, as they include potential interactions that have not yet been validated as positive interactions. To address this problem, Xia *et al*. developed a semi-supervised learning method (LapRLS) that regards known interactions as positive and unknown interactions as unlabeled data. Chen *et al*. developed an algorithm using a network-based random walk with restart approach (RWRH). However, these methods demonstrate good performance in a limited set of conditions, where the drugs or targets use a drug-target network- based similarity score (NetLapRLS and NRWRH). Because these approaches are limited in predicting the interactions of novel compounds or proteins that do not have any known target or drug information (e.g., newly synthesized compounds or mutated protein sequences), other approaches are needed. In this paper, we propose a drug-target interaction prediction method to predict potential interactions by using a modified BLM method. To classify unknown interactions into negative and unlabeled data, a clustering method was used before the training step. Then, modified bipartite local models, termed self- training bipartite local models (SELF-BLMs), were constructed using a semi- supervised learning approach (self-training SVM) to improve a model’s ability to find potential interactions. shows the overall process of the method. Finally, to train the model, we used a previous dataset for humans involving enzymes, G-protein coupled receptors (GPCRs), ion channels, and nuclear receptors from previous studies. We then constructed another drug-target interaction data set that contained recently updated interaction information for performance validation. As a result, the number of drug-target interactions increased by approximately 60% for each type of protein. Our model showed good performance based on the area under the ROC curve (AUC) and the area under the precision- recall curve (AUPR) values of the updated dataset. In addition, our proposed method found the highest number of potential drug-target interactions compared to other related methods in most cases. # Materials and Methods ## Drug-target interaction dataset for training To train the model and cross-validate its performance, we used four types of drug-target datasets from humans, including enzymes, ion channels, GPCRs and nuclear receptors. The data about the drugs, target proteins and drug-target interactions were derived from the KEGG BRITE, BRENDA, SuperTarget, and DrugBank databases. shows details about the dataset information that was used. ## Drug-target interaction dataset for validation Because the previous dataset was constructed in 2007, many newly identified drug-target interactions have since been discovered. To validate the performance power of predicting potential drug-target interactions, we updated newly identified interactions among drugs and target proteins that belonged to the previous dataset using the DrugBank, KEGG BRITE, and DsigDB databases. The drug- target interactions obtained from DrugBank and KEGG BRITE databases were credible, but the DsigDB database provided manually curated data and text mining data. Because text mining data are massive and not credible, we selectively took manually curated data from the DsigDB database. For this update, the numbers of updated interactions for each interaction type were 4,449, 2,029, 1,268, and 168, respectively. The number of drug-target interactions increased by approximately 60% for each type of protein. Using the updated dataset, we compared the performance and potential identification capability of each method. shows a summary of the previous and updated dataset. ## Similarity metrics The chemical similarities between drugs were calculated with the SIMCOMP method, which computes a global similarity score on the basis of common substructures between drugs using a graph alignment algorithm with the $$S_{d}\left( d,d^{\prime} \right) = \frac{\left| d \cap \right.d^{\prime}|}{\left| d \cup \right.d^{\prime}|}$$ where d and d' are substructures of drugs The structural information for the drugs was taken from the KEGG DRUG and KEGG COMPOUND sections of the KEGG LIGAND database. The similarity between the proteins was calculated using a normalized version of the Smith-Waterman alignment score. The normalized Smith-Waterman score between the proteins *P*_*A* and *P*_*B* was computed by the $$S_{p}\left( P_{A},P_{B} \right) = \frac{SW\left( P_{A},P_{B} \right)}{\sqrt{SW\left( P_{A},P_{A} \right)} \times \sqrt{SW\left( P_{B},P_{B} \right)}}$$ where SW is the Smith-Waterman alignment score The amino acid sequences of the target proteins were derived from the KEGG GENES database. ## Generating negative interactions To categorize unknown interactions as negative or unlabeled interactions, first, for each target protein, if a compound interacted with the target protein, we considered the interaction to be positive. We then clustered drugs and proteins by means of k-medoids clustering. If any of the compounds in a cluster do not interact with the cluster of the target protein, we considered the compounds in the cluster as having a negative interaction with the proteins. The remaining unknown interactions were considered to be unlabeled interactions, which are potentially positive interactions. These unlabeled interactions may later be classified as negative or positive interactions using the semi-supervised learning method. Because we used a k-medoids clustering method, an appropriate and consistent number of clusters was needed to train various datasets. In this study, we allowed to find one or two new positive interactions for each known positive interaction. Therefore, we set the number of unlabeled interactions to be no more than double the number of positive interactions. For example, if a protein has two known positive interactions, we set the maximum number of unlabeled interactions for the protein as four. The reason why we set the stringent limit for the number of unlabeled interactions is that too many unlabeled interactions could generate a decreased number of negative interactions, thereby resulting in a loss of negative data information for model construction. Therefore, we defined the number of clusters of drugs and targets as the resulting number when the overall number was divided by an integer, and we calculated the ratio of unlabeled interactions to positive interactions for the following integers N (one to ten). shows that the ratio was between one and two when the number of clusters was the number of drug and target proteins divided by two for each protein type. Therefore, we finally set k to be the number of drugs and target proteins divided by two. The detail steps of generating negative interactions are described in Algorithm 1 ## Bipartite local model Bleakley *et al*. proposed a method called BLM to predict the interaction between a drug *i* and a target *j*. BLM is described as follows. First, a local model for drug *i* is trained using an interaction profile of drug *i* and a similarity matrix of target proteins. Known interactions are regarded as positive, and unknown interactions are regarded as negative. Next, SVM constructs a classifier that distinguishes known interactions (positive) from unknown interactions (negative) using target similarity as a kernel. The model predicts the probability *p*_*d* (i,j) that a drug *i* and a query target *j* have an interaction by using the similarities between target *j* and the trained targets. Similarly, a local model for target *j* is trained using an interaction profile of target *j* and drug similarity. The model predicts the probability *p*_*t* (i,j) that a target *j* and a query drug *i* will have an interaction using the similarities between drug *i* and training drugs. Finally, we determine the predicted interaction value P(i,j) between drug *i* and target *j* with max(*p*_*d* (i,j), *p*_*t* (i,j)) or 0.5(*p*_*d* (i,j) + *p*_*t* (i,j)). **Algorithm 1:** Generating negative interactions 1 **Generating negative interactions** (*A*, *S*^*d*, *S*^*t*);   **Input :** Drug-target interaction matrix *A*,      Drug similarity matrix *S*^*d*,      Target similarity matrix *S*^*t*   **Output:** Negative labeled drug-target interaction matrix *A*_*n* 2 *k*_*d* ≔ \|*D*\|) / 2;   // D: set of drugs, *k*_*d*: the number of drug cluster 3 *k*_*t* ≔ \|*T*\| / 2;     // T: set of targets, *k*_*t*: the number of target cluster 4 *C*_*d* ≔ k-medodids(*k*_*d*, *S*^*d*);       //the set of drug clusters *C*_*d* 5 *C*_*t* ≔ k-medodids(*k*_*t*, *S*^*t*);       //the set of target clusters *C*_*t* 6 **for** *i* ← 1 **to** \|*D*\| **do** 7   **for** *j* ← 1 **to** \|*T*\| **do** 8    **if** *A*(*i*, *j*) = 1 **then** 9     *A*_*n*(*i*, *j*) ≔ 1;                 //positive 10    **else** 11     *SD*_*d**i* ≔ set of drugs in the cluster containing *d*_*i*; 12     *ST*_*t**j* ≔ set of targets in the cluster containing *t*_*j*; 13     **if** *SC*_*t* *is not related* *SC*_*d* **then** 14      *A*_*n*(*i*, *j*) ≔ -1;                //negative 15     **else** 16      *A*_*n*(*i*, *j*) ≔ 0;               //unlabeled 17     **end** 18    **end** 19   **end** 20 **end** 21 **return** *A*_*n*; ## Self-training support vector machine To classify the unlabeled data, a self-training SVM was used. In a local prediction step, the SVM model was constructed as a BLM using only labeled data. The unlabeled data were then classified by this model. If the unlabeled data passed the threshold, the unlabeled data were classified as positive or negative. The next step was to iterate this process until no unlabeled data failed to pass the threshold. Finally, the model used all labeled data as a local classification model to predict whether a compound targets proteins of interest and whether a protein is targeted by a compound of interest. The detail steps of constructing SELF-BLM models *M*_*t* for prediction of drug are described in Algorithm 2. In similar manner, SELF-BLM models *M*_*d* for prediction of target proteins are constructed. **Algorithm 2:** SELF-BLM 1 **SELF-BLM** (*A*, *S*^*d*, *S*^*t*);   **Input :** Drug-target interaction matrix *A*,       Drug similarity matrix *S*^*d*,       Target similarity matrix *S*^*t*   **Output:** prediction model of target *M*_*t* 2 *A*_*n* ≔ Generating negative interactions ((*A*, *S*^*d*, *S*^*t*)); 3 *I*^*t*(*i*) ≔ *A*_*n*(:, *i*);            //A interaction vector of target *t*_*i* 4 set $I_{L}^{t}(i)$;           //interaction vector of labeled data 5 set $I_{U}^{t}(i)$;          //interaction vector of unlabeled data 6 set $S_{L}^{d}$;            //Similarity matrix of labeled data 7 *M*_*t* ≔ $train\left( S_{L}^{d},I_{L}^{t}(i) \right)$;         //Train a local model for *t*_*i* 8 **do** 9   set ${S_{U}^{d}}_{XL}$;   //Similarity matrix of unlabeled data by labeled data 10   *P*_*U* ≔ $test\left( M_{t},{S_{U}^{d}}_{XL} \right.$);     //Predict unlabeled interactions 11   **if** \|*P*_*U*\| \> *threshhold* **then** 12    change the unlabeled data to labeled data 13    set $I_{L}^{t}(i)$; 14    set $I_{U}^{t}(i)$; 15    set $S_{L}^{d}$; 16    *M*_*t* ≔ $train\left( S_{L}^{d},I_{L}^{t}(i) \right)$ 17   **end** 18 **while** *any unlabeled data is changed*; 19 **return** *M*_*t*; # Results We trained the model using a previous dataset constructed by Yamanishi *et al* and validated the model using a previous dataset and an updated dataset. Because some unknown interactions in the previous dataset turned out to be positive in the updated dataset, we can measure the potential identification capability of models by comparing the performance results. First, we compared the performance of SELF-BLM with that of BLM, BLM-RBF, which includes drug-target network-based similarity using an RBF kernel, such as GIP or BLM-NII, and semi-supervised learning approaches, such as LapRLS, and NetLapRLS, which include network-based similarity. For BLM and BLM-RBF, we used the modified source code that was originally given by the authors. We used the LIBSVM (v.3.21) to use SVM implementation. When implementing SVM, the similarity matrices were used as a kernel without any modification. For parameters of SVM, values of C and gamma were assigned as 1 and 1 over number of features, respectively. For LapRLS and NetLapRLS, we implemented the methods based on the original paper. In the papers reporting these methods, BLM takes the maximum value between a drug-predicted value calculated using drug similarity and target predicted value calculated using target similarity, whereas LapRLS and NetLapRLS take an average value between the drug-predicted value and the target-predicted value; hence, we followed such approaches when we implemented these methods in the present study. SELF-BLM also takes the maximum value between the drug- predicted value and the target-predicted value. Because the all compared models are local models, the models are repeatedly constructed using associated interactions for a given drug or protein. If the methods are evaluated in k-fold cross-validation, positive interactions are frequently not included in the training step. For example, in case of *epinephrine* drug, the drug has three positive interactions with 95 target proteins in the GPCRs dataset. Because of the small number of positive interactions, positive labels are often not included in the training set when the data is segmented into k-sets. Thus, we evaluated the performance of the models using leave-one-out cross-validation (LOOCV). However, in order to confirm robustness of our model, we also evaluated the performance using 10-fold cross-validation. ## Prediction performance We calculated the performance of the interaction prediction in terms of the area under the ROC curve (AUC) value and the area under the precision-recall curve (AUPR) value. The AUC value is a common evaluation approach for binary classification problems. However, the large bias between the negative and positive training data sets often weakens the power of AUC values. Meanwhile, because it is important to classify the positive labels with high accuracy, the AUPR value may be a more appropriate indicator than the AUC value. shows the AUC and AUPR values of the five methods for the four type of proteins in each data set (previous and updated datasets). As the results show, the AUPR values of BLM-RBF were high in most cases when we used the previous dataset for validation. However, with the updated dataset, the AUC and AUPR values of SELF- BLM were the highest for most protein types, except for enzymes. In, it is noticeable that the AUC and AUPR values tend to be decreased in the updated data. The main reason for this result is that some negatively labeled interactions changed into positive interactions when the dataset was updated. Therefore, there are previously predicted a fair number of interactions as negative, the AUC and AUPR values decreased in the updated data. For instance, to predict the interaction between target HTR1E and drug Olanzapine according to type of GPCR, HTR1E was considered similar to HTR2A (0.23) and HTR2C (0.23), which bind to Olanzapine (positive); however, HTR1E is more similar to HTR1B (0.43), HTR1D (0.44), and HTR1F (0.55), which do not bind to Olanzapine (negative) in the training dataset. Thus, BLM does not receive a high indication that HTR1E will bind to Olanzapine. On the other hand, with the SELF-BLM methods, these negative targets were regarded as potential targets, and some targets were considered unlabeled as a result. Thus, SELF-BLM yields high marks using unlabeled data generated by clustering and the self-training SVM method. Moreover, in the case of the previous dataset, because the interaction between HTR1E and Olanzapine is regarded as negative, SELF-BLM seems incorrectly predicting the interaction. This is the main reason why SELF-BLM shows decreased performance in some cases using the previous dataset. However, in the updated dataset, the interaction is now regarded as positive, and the performance of SELF-BLM thus increased. In addition, SELF-BLM could increase the prediction performance with the previous dataset by self-training unknown information. Because potential interactions are regarded as negative in the previous dataset, this approach makes it difficult for a model to be trained accurately. For example, in the case of predicting the positive interaction between target CHRM1 and drug Clozapine among GPCRs, the conditions are as follows. Target CHRM2 binds to Clozapine, and CHRM3, CHRM4, and CHRM5 do not bind to Clozapine (however, these targets actually do bind to Clozapine in the updated dataset). In similarity- based models, the CHRM1 model will choose a similar protein among targets. BLM does not indicate that CHRM1 will bind to Clozapine as CHRM1 is more similar to CHRM3 (0.45), CHRM4 (0.42), and CHRM5 (0.47) than to CHRM2 (0.42). In contrast, because SELF-BLM neither considers CHRM3, CHRM4, and CHRM5 as training data nor changes these targets to positive data beforehand, it predicts that CHRM1 will bind to Clozapine. Therefore, SELF-BLM can yield high performance not only for the updated dataset but also for the previous validation dataset. Furthermore, additional experiment was conducted using up-to-dated drug-target information to show that the results are consistent in other dataset. ## Prediction performance for new interactions Next, we evaluated the performance of models regarding potential interaction identification. We compared the number of potential interactions at each percentage of positive interactions from the top 1% to 100% of the ranked score. For example, the targets of GPCRs have 635 known interactions, so we set the positive as the top six (1%) to 635 (100%) from a total of 21,185 interactions, and the number of potential interactions were compared within the percentage range. As shown in, SELF-BLM finds the most number of potential interactions than other methods for all of the protein types, except for nuclear receptors (see for details). Furthermore, we calculated the potential AUPRs of the four methods for the four types of proteins. In the potential precision-recall curve, positive labels were the potential interactions that were identified in the updated dataset, and negative labels were unknown interactions in the updated dataset. Therefore, we confirmed how the methods found the potential interactions simply by drawing a plot of the potential precision-recall curve. The curves show that SELF-BLM finds many potential interactions with high accuracy. Thus, the AUPR of SELF-BLM was the greatest among the methods for all of the protein types, except for the nuclear receptor type. shows the potential AUPRs of the five methods for the four types of proteins. In our results, BLM-RBF found few potential interactions and had low values of potential AUPR; also, the performance of BLM-RBF showed a greater drop than the other methods in most cases. Because BLM-RBF uses network-based similarities as an important factor for identifying a drug-target interaction, if a COI or POI had few interactions with the training set, the interaction similarities made it difficult to predict potential interactions of a COI or POI. This result shows that network-based similarity helps to find the interaction of a COI or POI that has a large amount of interaction information, but it is unsuitable for finding interactions of compounds or proteins for which little information about interactions is available. Although LapRLS and NetLapRLS are semi-supervised learning methods, we can confirm that these methods do not show good performance or a strong ability to identify potential interactions. # Conclusion In this study, we proposed a modified BLM, termed SELF-BLM, to accurately predict potential drug-target interactions. SELF-BLM uses k-medoids clustering and a self-training SVM algorithm to identify potential interactions among unknown interactions. To validate the performance of the method, we used benchmark datasets and updated recently verified interactions as potential interactions to the dataset using the DrugBank, KEGG, and DsigDB databases. Finally, we demonstrated the capability of SELF-BLM to predict potential interactions between drugs and target proteins. Notably, in most cases, SELF-BLM showed best validation performance with respect to AUC and AUPR for the updated dataset and found more potential interactions with high confidence prediction score compared to other methods. In our study, we used a benchmark dataset for training to compare SELF-BLM with other methods and to validate its capability to identify interactions. However, as the research proceeded, various other similarity methods were developed. Like other similarity based-methods, SELF-BLM majorly depends on drug similarity and target similarity. Therefore, the performance of the model may be improved by using more-effective similarity methods such as kernel fusion method for various data fusion and/or efficient novel similarity features. We emphasize that our SELF-BLM could show the best performance in the field of novel drugs or novel targets identification researches because our method does not require any known drug-target interaction information that is hardly known in novel molecules. Furthermore, in addition to drug-target protein interaction, it is important to deal with data imbalance problems or unlabeled data in many other areas so that, our method as well as the methods used in these areas can help to deal the problems. # Supporting information Authors thank the lab members for their valuable feedback and comments of the study. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** HN JK. **Data curation:** JK. **Formal analysis:** JK. **Funding acquisition:** HN. **Investigation:** JK. **Methodology:** JK. **Project administration:** HN. **Resources:** HN. **Software:** JK. **Supervision:** HN. **Validation:** HN JK. **Visualization:** HN JK. **Writing – original draft:** HN JK. **Writing – review & editing:** HN JK.

# Introduction Musculoskeletal pain (MSP) such as low back, neck or shoulder pain is the most common reason for pain in general and contributes significantly to health related costs in western industrialized countries, second only to cardiovascular diseases. While most patients recover within an expected period of time, in about 10% the condition becomes chronic. Reasons for chronification are multidimensional, as many biopsychosocial mechanisms contribute to the condition in each individual patient. One important dimension of mechanisms are neurophysiological pain mechanisms, and classification of patients to pain mechanism based therapy has been promoted as one promising treatment approach for patients with chronic pain syndromes. Central sensitization is regarded as the most important pain mechanism contributing to chronification of musculoskeletal pain. However, “central sensitization” is an umbrella term comprising a multitude of different mechanisms taking place in the dorsal horn of the spinal cord, ascending and descending pathways in the dorsal column, the brainstem and pain centres in the forebrain, all leading ultimately to amplification of innocuous and painful stimuli and to the extension of receptive fields. For a number of reasons, it seems crucial to identify CS in patients with musculoskeletal pain. Firstly, it offers a plausible physiologic rationale to explain signs and symptoms and resulting disability in absence of relevant and explanatory pathological findings. Secondly, the presence of CS should have practical implications with regard to medical and therapeutic interventions. Patients with CS possibly should avoid further nociceptive input from pain provoking aggressive interventions or too vigorous physical activities. These could potentially aggravate the problem, as more input could lead to further augmentation of the pain system. Instead, interdisciplinary interventions such as graded activity exposure, cognitive behavioural treatment or pharmacological treatment have been recommended allowing tailored and gradual activation and possibly "desensitisation". So far, no gold standard exists to diagnose CS. Even for elaborated but costly and time-consuming procedures such as quantitative sensory testing or laser- evoked potentials there are no generally agreed and diagnostic cut-off values. Thus, clinicians commonly rely on signs, symptoms as well as clinical examination to identify CS. In a recent Delphi survey, pain experts identified the following key clinical features: “Disproportionate, non-mechanical, unpredictable pattern of pain provocation in response to multiple/non-specific aggravating/easing factors”, “pain persisting beyond expected tissue healing/pathology recovery times”, “Pain disproportionate to the nature and extent of injury or pathology”and wide spread pain. Cardinal clinical signs include allodynia, hyperpathia and hyperalgesia. The above mentioned symptoms and signs are based mainly on the perspectives of clinicians and/or researchers and contain multiple subjective components based on clinical judgement. In regards to the broad impact of CS in the genesis and maintenance of chronic musculoskeletal pain, it seems important to gain more detailed information about pain experience from the perspective of individuals with MSP+CS. This information may be valuable to determine patient preferences and values and to help guide the use or development of questionnaires assessing the domain of central sensitisation. The primary objective of this study was therefore to explore pain experience and understanding of individuals with MSP+CS. The secondary objective was to investigate whether pain experiences of patients with MSP+CS differ from those of individuals with neuropathic pain (NP). # 1 Materials and methods ## 1.1 Research team LJ, KW, KR and SN conducted the Delphi rounds. The interviewers were all female and trained physiotherapists (BSc), with a median experience of 8 (IQR 4.5) years in the treatment of musculoskeletal or neurological patients. The interviewers and AS were not known to the participants until the start of the study, FP as head of the pain clinic was known to the participants from consultation and treatment, but was not involved in the data collection process. The conduction of the study was supervised by AS and FP. ## 1.2 Study design A qualitative cross-sectional study with two groups (Group MSP+CS: Patients with musculoskeletal pain + central sensitization; Group NP: Patients with neuropathic pain) was conducted. Data was collected by adapting the Delphi- technique for groups. The Delphi-technique for groups is an explorative approach to determine unknown information and represents a discursive process to gather information about a topic, which is relevant to the participants of the study. The aim of the process is to find consensus within a group of experts. In this study, patients were regarded as experts for their individual pain experience. ## 1.3 Ethical approval Ethical approval for this study was obtained from the Human Research Ethics Committee, Universitätsmedizin Göttingen (Nr. 18/9/11) in accordance with the Declaration of Helsinki on Ethical Principles for Medical Research Involving Human Subjects. All subjects provided written informed consent to participate. ## 1.4 Sample A convenience sample of nine patients with musculoskeletal pain and central sensitization and nine patients with neuropathic pain were recruited from active patients in a multidisciplinary and tertiary pain clinic. All patients considered had undergone thorough clinical and psychological diagnostics at admission and were treated for at least six months at the institution. The diagnostic psychological interview was conducted by a trained psychologist (MP) specialized in treating patients with chronic pain, the clinical assessment by a pain specialist MD. Patients had received different treatments ranging from purely medication interventions to participation in the pain clinic’s multimodal treatment program (five in group MSP+CS and two in NP). Active patients were reviewed by MP and FP and approached if likely to fulfil the inclusion criteria for either group. They were then re-evaluated by FP and MP and recruited, if in- criteria were fulfilled and no exclusion criteria present. Since the aim was to recruit two groups of “representative” patients in terms of the selected criteria and not a representative overall sample no detail on excluded patients were collected (see limitations of the study in the Discussion section). Participants had to be between 18 and 80 years of age and either have a diagnosed musculoskeletal (MSP + CS) disorder or neuropathic pain disorder (NP). Specific exclusion and inclusion criteria for the group MSP+CS are shown in. Inclusion criteria for the group Neuropathic Pain were: - Age between 18 and 80 - Definite neuropathic pain according to Exclusion criteria for both groups were the clinical diagnosis of a somatoform pain disorder based on the initial diagnostic psychological interview on pain clinic admission and re-evaluation. The psychological interviews were conducted by a trained psychologist specialized in treating patients with chronic pain (MP). ## 1.5 Data collection ### 1.5.1 Delphi procedure Central to the Delphi-technique are questions that are used to maintain the focus of the discussion on pain experience and to ensure consistency across groups. To generate questions, a systematic literature search was conducted, searching the databases Medline, Cinahl, Cochrane, PEDro, Sport Discus and the following journals: Pain, physioscience, Manuelle Therapie, Spine and Rheumatologie. Search terms were: 1. (myofascial pain syndromes \[MeSH\] OR “musculoskeletal pain”) AND ("central sensitization" OR "sensory hypersensitivity" OR allodynia \[MeSH\] OR pain threshold \[MeSH\] OR chronic disease \[MeSH\] OR chronic pain OR "widespread pain") 43 relevant articles were screened for pain dimensions and descriptions. The most frequent ones were discussed within the research team to reach consensus. A list of key questions was derived to cover the identified dimensions and descriptions with 13 questions concerning sensory discriminatory (Q 1–12) and associated bodily, mental, emotional and activity related phenomena (Q 13a-d) of central sensitization. The list of questions was pretested with a sample of six chronic pain patients and adapted to enhance comprehensiveness and acceptance. The Delphi procedure took place during a one-day session at a multimodal pain clinic from 9am 5pm, with appropriate breaks between the data collection sessions. Patients with MSP+CS and patients with NP participated separately on two different occasions. To the knowledge of the investigators there was no communication or interaction between patients of the two groups. The research team had no information on the clinical diagnosis of the participating patients beyond their overall group assignment. Data collection followed a structured and standardized procedure. There were two groups with nine patients each, one group consisted of patients with musculoskeletal pain and central sensitization, the other group of patients with neuropathic pain. The procedures were the same for each group. In each group, the nine patients were randomly split into three groups with three participants and one researcher per group to facilitate and moderate the discussion and to take notes of participants statements. The list of questions was explained to the patients. To give the participants an idea about the quality and expected content of the discussion and to facilitate the participants discussion a fictitious case example was read to them. Within this case example, the day of a 52-year old woman with musculoskeletal pain and central sensitisation is described. She exhibits typical features of central sensitization such as hyperalgesia and widespread pain, its impact on social activities (visiting friends) and cognitive function (unable to concentrate at work) are described. Data was collected in two rounds. During the first round (90 Minutes) the questions were discussed. The aim was to collect as many subjective pain- describing items from individual participants as possible. Besides, the participants were asked to draw the location and expansion of their pain into a body chart and to report quantitative aspects of their pain experience such as pain intensity or duration of pain. For the second round of data collection participants were allocated randomly to new groups. Patients were asked to rate the various items from the first round by using a Likert-Scale from 0 to 10 (0 = no relevance; 10 = very high relevance). After the second round of data collection, the ratings from the three groups were collected and a new item list with group ratings was printed and presented to the whole group. Finally, all nine participants could ask questions, make comments and were informed about the further procedure. They received the final item list including the ratings to take home. Participants were asked to reflect on the results and make changes or comments if applicable and to post the commented list back to the research team. ### 1.5.2 Questionnaires To gain further information, patients additionally completed the following questionnaires on another occasion. This data collection was scheduled a few weeks after the survey to not influence the patient’s subjective performance by providing a range of predefined possible answers. - Based on a body diagram the area marked as painful was scored from 0–19 following the regions described in the regional pain scale - Pain Sensitivity Questionnaire (PSQ) measures pain sensitivity by self rating; - Tampa Scale of Kinesiophobia a questionnaire for kinesiophobia; - Pain Perception Scale (SES) measures affective and sensory pain perception; - Depression Anxiety Stress Scales (DASS) screens for depression and anxiety; - Patient Health Questionnaire (PHQ) evaluates severity of somatic symptoms; - Pain Detect Questionnaire screens for neuropathic pain components - Pain Catastrophizing Scale (PCS) a questionnaire to assess catastrophizing in patients with chronic pain - SF-36 measures physical and mental components of quality of life ## 1.6 Data analysis Descriptions of the course of pain were visualised in seven different graphs. Quantitative data (Q1a-c) from round one was summarized by calculating the median and interquartile range (IQR) as well as arithmetic mean and standard deviation (SD). Means were also calculated for the collective ratings of the qualitative data (Q4-13) from round two. These mean ratings were used to reduce the number of items; all items with a mean rating below 3/10 as well as duplicates were removed. To analyse pain descriptions, a qualitative content analysis according to Mayring was conducted. Single items were inductively assigned to main- and subcategories. Hierarchical category trees were then constructed to further categorize the data and to visualize the results. Questionnaire data was compared between groups using non parametric statistics (Mann-Whitney-U-Test), since n was too small to assess normal distribution. For the SES and SF-36 group data was also expressed as mean and SD to allow comparison to normative data typically presented that way. # 2 Results ## 2.1 Participants Data collection took place during two Delphi sessions. In the first session nine patients with MSP+CS participated, consisting of eight females and one male who had the following diagnoses: fibromyalgia (n = 2), spinal pain (n = 3), chronic widespread pain (n = 4). Patients with spinal pain had non-specific pain progressing beyond the initial low back or neck pain (but not radicular pain), patients with chronic widespread pain had no other specific diagnosis, except for mild and non-specific degenerative changes. Median age was 49 years (IQR = 6). In the second session nine patients (five females and four males) with NP participated, median age was 57 (IQR = 19). Patients in the group NP had the following diagnoses: neuralgia (n = 4), polyneuropathy (n = 4) and radiculopathy (n = 1). For details see. Patients with neuralgia had specific and anatomically sound distributions of pain (face V1, cervical postzosteric neuralgia, distal tibial nerve, plantar nerve), polyneuropathy pain was restricted to the distal limbs in variable degrees. ## 2.2 Questionnaires The results of the completed questionnaires are summarized in. Patients with neuropathic pain reported significantly fewer areas as painful compared to MPS+CS. Patients in both groups showed similar PSQ scores indicative of increased pain hypersensitivity with total scores above 4/10 for 5 of 8 patients in the NP pain group and 5 of 7 patients in the MSP+CS group. Likewise, values for the PSQ minor score were not statistically different between the two groups, yet slightly lower in the NP group. Fear of movement was less marked in both groups, patients mean scores were lower than typical data for mean Tampa scores in patients with musculoskeletal pain in other studies ranging from 39.4–42. Scores for subjective sensory and affective pain sensation measured with the pain sensation scale (SES) were lower than typical pain population scores in both groups. The SES measures the emotional burden of pain (SES affective) as well as sensory qualities of pain related to pressure, rhythm or temperature (SES sensory), lower scores reflect less burden and sensory qualities. Patients in group MSP+CS had mean scores that were 1.85 SD (SES sensory) and 2.84 SD (SES affective) below the mean of typical pain population scores, patients in group NP were 1.26 SD (SES sensory) and 2.61 SD (SES affective) below. Mean Depression subscale (DASS) was below the cut off score in both groups, however the proportion of patients exceeding cut off scores for the subscales anxiety and stress was substantial ranging from 43% (n = 3) to 86% (n = 6) within the groups. Screening with the Patient Health Questionnaire (PHQ-15) revealed low somatic symptom severity in group MSP+CS (median = 8, IQR = 5) and medium somatic symptom severity in group NP (median = 12, IQR = 6). The proportion of patients with high somatic symptom severity was 0% (n = 0) in group MSP+CS and 29% (n = 2) in group NP. Six patients of seven in the NP group had Pain Detect (PD) scores \>18 indicating a likely neuropathic pain component, compared to four out of eight in the MSP+CS group. For the rest of the patients a neuropathic pain component was uncertain with PD scores between 12 and 18. For none of the participants a neuropathic pain component could be excluded (PD score \<12). Patients in both groups exhibited impaired general physical health with SF36 physical component summary 1.71 SD (MSP+CS) and 2.05 SD (NP) below the norm. Mental health status was comparable to population normative data with 0.2 SD (MSP+CS) and 0.3 SD below the norm. ## 2.3 Delphi procedure The results of the Delphi procedure are summarized in. ### 2.3.1 Pain localization Patients in the MSP+CS group reported painful areas all over the body. In the body charts they marked eight to twenty painful areas, which often had a widespread extent. Patients in the NP group marked one to three pain regions in the body charts. These reflected the underlying neuropathic disorder. ### 2.3.2 Pain intensity, frequency, duration, course and occurrence (Q1-5) The median highest experienced pain in the last week on an 11-point NRS was 7 (IQR 1) for the MSP+CS group and 8 (IQR 2) for the NP group. The lowest experienced pain of the last week was 3.5 (IQR 1) for group MSP+CS and 4 (IQR 2) for group NP. The patients had median pain intensity during the last week of 5 (IQR 2) in the MSP+CS group and 7 (IQR 2) in the NP group. Participants in both groups typically described constant pain (i.e. “always” or “365 days a year” MSP+CS and NP). Answers concerning pain duration also showed that both NP and MSP patients experience constant pain. While the intensity of pain seemed to be similar (Mann Whitney U Test p\>0.26) and both groups reported a constant component in their pain, the course over time differed between the two groups. Whereas the NP group typically reported constant pain with peaks like “flashes” or “electric shocks”, the MSP+CS group described the pain as a constant pain with waves that seemed to last longer than the peaks of the NP group. This was in line with the use of items that indicate differing variability of pain: “interval” (NP) or “depending on the situation” (MSP+CS), “repeated bouts of 10–30 seconds duration” (NP) or “a few seconds up to four hours” (MSP+CS). ### 2.3.3 Quality of pain (Q 12) In both groups participants described characteristics of their typical pain. Two main categories could be derived: pain quality and affect. For patients in group NP the most relevant subcategories for pain quality were shooting 6.73 (i.e. “fulgurant pain” 7.5), burning (6.3) and dull (6.0). Most relevant descriptions in the category affect (8.0) for patients with NP were “nasty” (9.7) and “nerving” (9.0). Pain descriptions in group MSP+CS were more numerous (33 vs. 17) and could be categorized in more subcategories (11 vs. 8) compared to patients with NP. Most relevant subcategories for patients with MSP+CS were radiating (7.7), burning (6.9) and shooting (6.75). The category affect (7.63) included pain descriptions such as “tormenting” (8.4). ### 2.3.4 Pain trigger, reduction and aggravation (Q 6–8) To investigate patient’s pain experience in regards to influencing factors, three questions were asked: “what triggers the pain?” “what relieves the pain?” and “what aggravates the pain?”. Patients with NP reported pain triggers /Q6) that could be categorized as bodily factors (4.9). Subcategories were motor stimuli (4.9) such as “physical load” (5.5) and sensory stimuli (4.9) such as “unsuitable shoes” (5.0). In group MSP+CS psycho-emotional trigger were most important (6.8), patients mentioned intrinsic factors such as “to put oneself under pressure” (8.2) and extrinsic factors like “lack of time” (8.2) or “excessive demands” (8.2). Patients in the MSP+CS group mentioned a total of 14 different items in the category psycho- emotional pain trigger compared to one item in group NP. In group NP the participants mentioned “medication” (9.3) as the most important pain relieving factor (Q7). Patients in the MSP+CS group stated that physical therapy (8.2), physical stimuli such as “bathtub” (9.5) or “hot springs” (9.7) and relaxation such as “relaxing music” (8.0) were most relevant, also “medication” (6.2) and motor activity i.e. “aqua-fitness” (6.2) were mentioned. Overall, passive activities such as “bathtub” or “hot springs” received the highest rating in regards to pain relieve in group MSP+CS. In contrast, patients in the NP group also expressed active strategies such as “doing joyful things” (4.7) or “change of activity” (3.7) to reduce pain. The category psycho-emotional factors was more highly rated (6.4) in the MSP+CS group compared to the NP group (3.8). This observation was further emphasized when comparing the total number of items mentioned: 27 items for psycho- emotional factors in group MSP+CS and 3 items in group NP. In regards to factors that aggravate pain (Q8), experiences differed between the two groups. For patients in group NP the most relevant category was bodily factors such as physical strain (i.e. “standing for a long time” (5.7). Also relevant in the NP group was the statement “no pain aggravating factors” (4.0). Furthermore, context factors (3.7) were mentioned such as “cold” (3.3). Psycho- emotional factors were less relevant for patients with NP with a mean rating of 3.1. In comparison to group NP, Patients in the MSP+CS group mentioned a greater number of possible factors that might aggravate pain (45 vs. 10 total items). Psycho-emotional factors (6.5) were the most relevant pain aggravating factors for the MSP+CS group with high mean ratings and a large number of items. Examples for descriptions in this category are “emotional distress” (8.0) or “to take things to much to the heart” (7.2). Patients also described personal factors (6.8) such as “working more than one has to” (8.2) and environmental factors (6.1) (i.e. “excessive demands” 7.8) and context factors (6.9) such as “cold” (9.0) or “the job” (4.8). ### 2.3.5 Change of pain since the problem started (Q9) In answer to the question “how did your pain change since the beginning of your problem?” patients in both groups reported improvement as well as worsening. Improvement based on endogenous factors such as “I have learned to deal with my pain (through therapy)” was reported. Worsening of pain was described in terms of a spread of pain without explainable cause in both groups. ### 2.3.6 Allodynia (Q 10) Patients in both groups (NP and MSP+ CS) reported signs of allodynia in relation to stimuli, movement and activities. Mean ratings for descriptions related to allodynia were higher in group MSP+CS compared to group NP. Perception of normally non painful stimuli as painful (MSP+CS 7.5; NP 5.2) such as punctual touch (MSP+CS 7.5) or pressure and strain (NP 6.0) were reported. A second category in both groups was movement related allodynia (MSP+CS: 7.5; NP: 4.9) associated with posture, movement and movement under load. ### 2.3.7 Hyperalgesia (Q 11) Participants in group NP did not describe any signs of hyperalgesia. Patients with MSP+CS mentioned that “to stub against something” (7.7) was more painful than before the start of the problem. Patients in both groups stated that they are less sensitive to painful stimuli compared to the beginning of their problem. ### 2.3.8 Associated phenomena (Q13a-d) Questions about associated phenomena in relation to the body, mental ability, psyche and activities of daily living were asked to gain insight into patient’s experiences on how pain affected different dimensions of function, activities and participation. Patients were asked to report associated phenomena that developed from the beginning of the disease. Associated phenomena related to the body (Q13a) were categorized as sensory phenomena or as symptoms in body systems. These categories differed in relevance between the two groups. For patients in the group MSP+CS the most important pain descriptions were related to sensory phenomena (7.0) such as temperature (7.3) (i.e. “being cold” 9.7), vision (7.2) (i.e. “hypersensitivity to light” 7.2) or hearing (7.0) (“hypersensitivity to noise” 7.0). Other descriptions related to body systems (6.3) such as the skin (“dry skin” 8.3) or nasopharyngeal zone (i.e. “dry tongue” 6.7). Patients in the group NP reported descriptions mainly in the category body systems (mean 5.4); these were related to the musculoskeletal system (i.e. “to adopt a relieving posture during walking” 5.7). Associated phenomena in regards to mental ability (Q13b) had equal relevance for the patients. Both groups reported lack of motivation (MSP+CS 7.2; NP 6.0), with decreased perseverance (MSP+CS 7.7; NP 6.7) and lack of inner drive (MSP+CS 6.7; NP 5.3). Patients experienced a decrease in vigilance (MSP+CS 6.2; NP mean 5.2) as a result of fatigue, lack of concentration and memory deficits. Additionally, both groups reported an impaired ability to communicate (MSP+CS 5.8; NP 4.8) affecting both sending (“I can’t remember certain words” MSP+CS 6.7) and receiving information (“I have difficulties listening” MSP+CS 5.3). Patients in both groups described different phenomena associated to psychological aspects (Q13c). Patients in group MSP+CS provided more descriptions compared to patients in group NP. These were predominantly negative changes in relation to thoughts (6.3) (i.e. “not being able to stop brooding” 6.7), lack of motivation (6.0) (i.e. “this is the bottom line” 6.8) and emotions (5.5) (“being more sensitive” 7.3). Patients in group NP also reported positive changes in regards to their attitudes (4.6) and motivation (3.0) such as “priorities have changed” (6.7) or “I have so much energy that I can manage anything” (3.0). Patients in both groups reported impairments in their activities of daily living (Q13d; MSP+CS 5.6; NP 5.3) such as employment or housework (MSP+CS 6.5; NP 6.8) (i.e. “I had to slow down” MSP+CS 7.7 or “I can’t work as many hours as I would like to” NP 7.3), hobbies (MSP+CS 5.4) and sport activities (MSP+CS 4.6; NP 4.5). Also impairments in regards to social activities with a mean ranking of 5.4 in group of MSP+CS (i.e. “when I am together with friends, I want to go home earlier” 7.3) and 4.3 in group NP (i.e. “I am more cautious to accept invitations” 5.0) were mentioned. # 3 Discussion ## 3.1 Pain experience of patients with MSP+CS Patients with MSP+CS experience constant moderate to severe pain in many body regions. Descriptions of these patients reveal a complex picture of pain manifesting in many dimensions, particularly in regards to emotional and psychological aspects. Exemplary are descriptions of pain trigger, pain aggravating and relieving factors that reveal the importance of psycho-emotional aspects. Pain trigger and pain aggravating factors in group MSP+CS were often external factors such as perceived strenuous work situations. These are factors that cannot be easily influenced by the patients. Consequently, statements in regards to pain reduction in group MSP+CS typically reveal rather passive pain coping strategies. Patients with MSP+CS are affected in many aspects of daily life. In regards to relevant phenomena associated with pain, patients reported impairments in regards to their mental ability, such as reduced motivation, vigilance and resilience as well as impaired ability to communicate. These findings are comparable to the findings of others studies. Turk et al. reported that patients with chronic pain including patients with chronic musculoskeletal pain experience limitations in regards to enjoyment of life, emotional well-being, fatigue, weakness and sleep related problems, also cognitive deficits have been shown. ## 3.2 Differences in pain experience between patients with MSP+CS and patients with NP Our data show similarities and contrasts of pain perception between patients with MSP+CS and NP. Patients in both groups were severely impaired by chronic pain and both report marked restrictions in their activities of daily living. The quantitative aspects of the pain experience such as intensity, frequency, duration, and time of occurrence were comparable between groups. Contrasts could be shown in relation to sensory and affective aspects of pain. In general, structural and somatic factors in relation to pain triggers, reduction, aggravation and associated phenomena were more relevant for patients in group NP, whereas for patients with MSP+CS affective and emotional factors were most relevant. Associated phenomena related to the body were mainly confined to the musculoskeletal system in group NP, whereas patients with MSP+CS reported symptoms in multiple body systems. This pattern also becomes evident in regards to pain triggers. Patients with NP mentioned mainly pain triggers related to the body. These were generally modifiable by behavioural changes such as avoiding physical strain or changing unsuitable shoes. Patients mainly employed active strategies to reduce pain (“doing joyful things”). This implies higher coping abilities or acceptance of pain related disabilities in the group NP. Patients in group MSP+CS weighted psycho-emotional factors higher when describing pain trigger, reduction and aggravation. This depicts contrasts in pain experience and in the understanding of the nature of pain. Patients with NP showed a rather mechanistic and structural pain understanding that may be enhanced by a more localised and “explainable” pain experience, whereas patients with MSP+CS experienced pain as complex and multifactorial, with a higher weight on psycho-emotional factors. This is further supported by the higher mean number of pain descriptions in all categories in group MSP+CS with 5.3 compared to 3.7 in group NP. ## 3.3 Comparison of Delphi and questionnaire data Interestingly, these trends to describe pain were poorly reflected in the actual questionnaire data from the two groups, where patients with neuropathic pain reported more affective disturbances. This may indicate that the relationship between affective symptoms and pain perception could differ between patient groups. On the other hand, the pain detect questionnaire and the extent of pain was in line with Delphi results. When comparing questionnaire data with Delphi results, further differences and discrepancies become obvious. In contrast to Delphi results all SF-36 subscales except bodily pain and PCS were within 1 SD of normative population scores. For example, the SF-36 subscales vitality (-0.32 SD) and mental health status score (MCS) (-0.2 SD) were not different to population normative data, although Delphi results clearly showed that patients felt affected in these dimensions. Also in contrast to Delphi findings, SES scores indicated that the emotional burden of pain was 2.84 SD below the norm for pain populations. When comparing Delphi results to questionnaire data limitations of the used questionnaires to capture the complex picture of patients pain experience and understanding become evident. These differing results may also reflect potential positive effects of therapy in the current patient sample that had received various pain management interventions. Thus Delphi results may represent an overall concept of pain and disability these patients have, while questionnaires may reflect the individual’s current status of health and pain. Importantly, patients with MSP+CS had a more extensive spatial extent of pain, which was possibly biased by selection in this sample. However, it has recently been shown that even in a low back pain cohort, many patients report increased spatial extent of pain, when specifically asked for other pain sites. In another set of studies fibromyalgia like characteristics (including spatial extent) were associated with poorer postoperative outcome following various types of surgery. Recently developed criteria for central sensitization likewise require more generalized pain. This indicates, that although the criteria chosen to select patients with CS in this study were based on and validated in a sample of chronic low back pain, they seem appropriate for other patient groups as well. The Delphi procedure thus has the potential to capture the overall characteristics of the MSP+CS and NP condition in general, irrespective of the current situation of the patients. It stated the relevance of symptom and experiences from the patient’s perspective and not the individual level of symptom intensity as measured by the respective questionnaires. Viewed from this perspective typical clinical perceptions of the two different pain conditions are reproduced and confirmed by the Delphi procedure. On the other hand it becomes obvious how similar the overall burden of pain is for the two groups, which probably is better highlighted by the individual questionnaire data. ## Neuropathic pain versus central sensitization As expected, more patients in the NP group were classified as “probable neuropathic” based on their painDetect scores, however patients in both groups exhibited signs and symptoms indicative for neuropathic pain as no patient was categorized as “unlikely neuropathic pain”. Patients with MSP+CS did not show a neuroanatomically plausible pain distribution and were only included when they failed the algorithm suggested by Treede at al. and when there was no clinical evidence (based on history, examination and available results of diagnotic tests) for a relevant neurological lesion or disease. Thus patients with MSP+CS displayed some of the pain characteristics seen in neuropathic pain, while not having a clinical diagnosis of neuropathic pain. This discrepancy might be related to properties of the painDetect questionaire. The validation describes selection of items based on the literature and not a patient survey. The initial cohort consisted of two patient groups. One included patients with “classic” neuropathic pain conditions, namely postherpetic neuralgia, painful polyneuropathy, nerve trauma, and low back pain with neuropathic origin (presumably radicular pain), while the second group had pain of nociceptive origin, like visceral pain, osteoarthritis, inflammatory arthropathies and mechanical low back pain. Patients with assumed mixed origin, including fibromyalgia and ankylosing spondylitis were excluded. Two specialists had to concur in the diagnosis. Based on these clinically selected groups a score of ≤12 indicated that a neuropathic component is unlikely, a score ≥19 indicates likely neuropathic pain, between the score is uncertain. Sensitivity was reported to be 84% and specificity 84%. When the questionaire was used for patients with fibromyalgia, high frequencies of a positive pain detect score were also seen, which the authors associated with neuropathic pain characteristics indicative of central sensitisation. In an independent validation in a general chronic pain sample, 400 patients with any type of chronic pain diagnosis were included, and 37% of all patients displayed distinct neuropathic characteristics, including patients with musculoskeletal pain. A high level of depression, pain chronicity, and reported pain intensity explained most of the variance in painDetect scores. The authors conclude that any type of chronic pain may have neuropathic characteristics and that the painDetect screening tool was not able to differentiate typical neuropathic entities from other pain syndromes with neuropathic features. The results of our study support this conclusion. This indicates that clinically defined neuropathic pain and central sensitization may share certain pain characteristics, but differ in other clinical features. ## Strengths and weaknesses The Delphi procedure for groups is a valid method for investigating experiences of experts in relation to a particular problem. Experts usually have an overview of many different aspects of a problem, e.g. a clinician with longstanding experience will have seen many patients with a particular health problem. This is not the case for patients, who are only able to relate to their individual experience. Patients are experts for their individual pain, but not for pain in general. As a consequence, the generalizability of the results is limited and should be replicated in a larger sample of patients. A second issue relates to the clinical diagnosis of “central sensitization”. An experienced medical pain specialist (FP) made the clinical diagnosis based on predefined criteria. Although the diagnostic criteria for central sensitization showed reliability and discriminative validity, construct validity may be limited as so far no gold standard exists to diagnose CS and a potential systematic bias in selection cannot be ruled out. Further limitations relate to the sample. Firstly, the two groups were not comparable in regards to sex and age. Patients in group MSP+CS were nine years younger on average and had a higher proportion of females, which may have influenced contrasts in regards to pain experience between the two groups. Furthermore, the selected sample had varying degrees of ongoing treatment for their pain condition when they had entered the study. Groups differed in regards to treatment intensity within the pain clinic. Patients with MSP+CS had on average more multimodal treatment units compared to patients with NP. This might explain differences such as lower kinesiophobia and the generally normal mental component in the SF-36 in group MSP+CS. Patients still had a high level of chronification, yet generalizability to untreated or less chronic patients is limited. Finally, although the sample size of nine per group is within the recommended size for Delphi groups, potential bias like an overrepresentation of patients with widespread pain / fibromyalgia, or a non-representative sample for neuropathic pain in this study (no central pain) limits the generalizability of the findings to a chronic pain population. ## Conclusion The results of the present study illuminate features of a subgroup of patients with musculoskeletal disorders. Firstly, the general impression is, that patients with MSP+CS have a more complex and multifactorial pain experience and understanding, in comparison to a group of patients with NP. Secondly, for patients in group MSP+CS, the affective pain component is most relevant, and pain seems to compromise many different aspects of life and health. Furthermore, patients with MSP+CS are affected by associated phenomena relating to psychological and mental impairments, particularly concentration, vigilance, motivation and communication. These findings are in accordance with other studies and are reflected in a recently developed Central Sensitization Inventory. Patients with MSP+CS seem to present with distinct features that need to be considered in diagnosis and targeted treatment. However, the ability of our current instruments to differentiate between different patient groups (in this study MSP+CS and NP) seems limited. [^1]: The authors have declared that no competing interests exist.

# Introduction The proliferating cell nuclear antigen (*PCNA*) gene product is a nuclear protein that acts as a cofactor for DNA polymerase-δ and participates in DNA synthesis and repair (for reviews see). In addition, by interacting with a wide array of proteins, PCNA serves essential functions in cell cycle progression, epigenetic inheritance, , and gene transcription. *PCNA* gene expression is generally low in quiescent cells, increases with cell proliferation, and is tightly controlled within the cell cycle. In response to proliferative stimuli, *PCNA* mRNA and protein levels both increase during the G1/S transition, commensurate the protein's role in DNA replication. PCNA synthesis is induced by diverse stimuli in a cell-type specific fashion, including: EGF, PDGF, and serum in 3T3 cells, interleukin 2 (IL-2) in T-lymphocytes, and p53 and adenovirus infection in HeLa cells. There appear to be transcriptional and post-transcriptional mechanisms for regulating *PCNA* mRNA levels in 3T3 cells by processes that are not fully characterized. No formal study of *PCNA* gene regulation has been demonstrated in breast cancer cells. Most studies have observed that high *PCNA* gene expression correlates with increased metastatic potential and decreased survival in patients with breast carcinoma. Many breast and uterine cancers depend upon E2 for neoplastic initiation, development, or metastasis, and antiestrogen therapies remain the mainstay of treatment and prevention for ERα-expressing breast cancers. The E2 response in breast cancer cells is predominantly mediated by ERα, a ligand- activated transcription factor. We confirmed that *PCNA* gene expression is enhanced by E2 exposure in MCF7 breast cancer cells which express ERα and proliferate in response to E2. We, and others, have detected two putative estrogen response elements (EREs) in the 5′ region of the *PCNA* gene, one of which is conserved between murine and human species, and both of which may serve as *cis*-regulatory elements for ERα- mediated gene regulation. Recently, PCNA was shown to physically interact with ERα and RARα and to modulate gene transcription regulated by these transcription factors. These observations raise the possibility that E2-stimulated ERα activates *PCNA* gene expression, leading to feedback regulation of ERα transcriptional functions by ERα-bound PCNA. The process of *PCNA* gene induction is likely to be essential to the mitogenic effects of E2 in some ERα- expressing cancers. The *PCNA* promoter is regulated at the transcriptional level by many transcription factors including E1A, ATF1, RFX1, CBP, p107, p53, , and E2F. In some systems, basal transcription is augmented at G1/S by inducible regulatory elements. No role for ERα has been demonstrated in the regulation of *PCNA* gene expression although estrogens act as potent mitogens in both normal and neoplastic breast and uterine tissues. Because eukaryotic *cis*-regulatory elements may reside great genomic distances from target genes,, and because the putative EREs that we identified are located 1,200–10,000 bp from either transcription start site (TSS) demonstrated for *PCNA*, we thought it important to test these elements for functional significance. Our goals were to understand the predictive value of computational ERE detection for an E2-responsive gene and to better define the mechanisms by which estrogen stimulates *PCNA* gene expression in breast cancer cells. Our data indicate that E2 enhances *PCNA* gene expression by an indirect process and that computational detection of EREs, even when evolutionarily conserved and when near E2-responsive genes, requires biochemical validation. # Results ## E2 stimulated *PCNA* mRNA and protein expression in a process that requires *de novo* protein synthesis We recently reported the results of microarray-based gene expression profiling using the MCF7 breast cancer cell line, a model system for E2-dependent breast tumors. MCF7 cells express ERα and proliferate in response to E2 exposure. We observed increased *PCNA* gene expression after 4, 12, and 24 hours of E2 exposure. Notably, two putative EREs were previously detected upstream of *PCNA* by Bourdeau *et al*, who applied large scale computational analyses to the human and mouse genomes for detection of conserved ERE sequences. Our analysis revealed that the both ERE sequences are 100% conserved between Rhesus and human, whereas the 3′-ERE sequence also shares 79% identity with mouse, indicating that the 3′-ERE is more evolutionarily conserved. These ERE sequences were never tested for function. Quantitative reverse transcription polymerase chain reaction (Q-RT-PCR) was applied to MCF7 cell lysates and confirmed greater than 2-fold induction of *PCNA* mRNA after six hours E2 exposure. Known E2-responsive genes also tested include *TFF1, MYC, STC2*, and *DCC1.* Similar changes in PCNA protein levels were observed after E2 treatment of MCF7 cells. The E2-stimulated expression of *PCNA* mRNA was sensitive to co-treatment with the protein synthesis inhibitor cycloheximide (CHX), suggesting a secondary, or indirect, transcriptional effect of E2 exposure. Interestingly, DCC1, a component of the replication factor C (RFC) which loads PCNA onto DNA during DNA replication, demonstrated expression that was similarly E2 responsive and CHX sensitive. These data are consistent with a model in which DNA replication is regulated within the cell cycle, in part, by the regulated synthesis and degradation of the replicative machinery. There exist direct transcriptional targets of ERα that require *de novo* protein synthesis in order to be transcriptionally regulated by the receptor. For example, the ERα gene target *c-fos* is rapidly activated by E2 in uterine tissues, whereas *de novo* protein synthesis is required in order to produce a sustained *c-fos* transcriptional response to E2 in MCF7 cells. Similarly, there exist gene targets that are repressed by ERα only after induction of the corepressor protein NRIP-1. Thus, the fact that the *PCNA* response to E2 was CHX-sensitive did not preclude the gene from being a direct responder to ERα. ## E2-enhanced *PCNA* gene expression was blocked by inhibition of ERα function Blocking an E2-mediated transcriptional response by co-treatment with ICI 182,780, a pure ERα antagonist, indicates that the observed effect is mediated by ERα. In MCF7 cells, the E2-stimulated expression of *PCNA* was blocked by co- treatment with ICI 182,170 and by inhibition of gene transcription using actinomycin D (ActD). Similar results were observed for *MYC* gene expression, which is also E2-responsive and a direct gene target of ERα. These data support the hypothesis that E2-stimulated *PCNA* gene expression requires both the activity of ERα and *de novo* gene transcription. We recently demonstrated that siRNA-mediated knockdown of *PCNA* gene expression greatly inhibited E2-stimulated cell proliferation in MCF7 cells. These data supported the hypothesis that PCNA is an important mediator of E2-stimulated cell proliferation in MCF7 cells. Our analysis of the *PCNA* gene locus confirmed two imperfect ERE's within 10 Kb of the two TSSs described for the gene. Each putative ERE (herein dubbed PCNA-ERE1 and PCNA-ERE2) demonstrates a single nucleotide mismatch from the core 13 bp consensus (or “canonical”) ERE sequence. Notably, the majority of EREs identified and validated in the human genome do not demonstrate perfect consensus sequences, indicating a high degree of heterogeneity for functional ERE sequences. Similarly, the mere presence of an ERE-like sequence is not sufficient to ensure ERα binding or ERα-mediated transcriptional responses in the majority of chromatin contexts. These observations indicate that ERE sequences must combine with additional factors in order to function. Such factors may include regional histone composition, distribution, and post-translational histone modifications, DNA methylation status, and regional DNA sequences (with associated *trans*-factors) that create a transcriptionally permissive environment for activated ERα. ## A predicted ERE sequence near the *PCNA* gene was capable of binding to ERα *in vitro* We performed electrophoretic mobility shift assays (EMSA) using the radiolabeled PCNA-ERE1 sequence with recombinant ERα protein. In order to promote receptor- DNA binding, and to control for the presence or absence of ERα protein, recombinant ERα was combined with cofactors present in nuclear extracts from an ERα-negative, immortalized human endometrial stromal cell (HESC) line. ERα- containing protein complexes were shown to bind to PCNA-ERE1, *in vitro*, as evidenced by supershift with an antibody specific for ERα. No estrogen receptor- containing complex was noted from HESC cell extracts alone (not shown). No supershift was noted using antibodies targeting the transcription factors Sp1, AP-1 or ARP-1/COUP-TF2, although these factors have been demonstrated to bind with ERα at selected promoters. Binding was weaker for PCNA-ERE1 than that observed for a consensus ERE sequence but stronger than that observed with PCNA- ERE2, for which binding was weakly detectable (not shown). ERα-dependent binding to radiolabeled PCNA-ERE1 could be competed away using excess unlabeled (“cold”) PCNA-ERE1 probe (P-ERE1) whereas a similar probe with mutations in the two half-sites of the ERE (P-ERE1mut) did not efficiently compete for labeled PCNA-ERE1-bound receptor. These data suggested that one or both of these putative EREs might represent ERα-responsive *cis*-regulatory elements that modulate *PCNA* gene expression in MCF7 cells. It is recognized that the affinity of receptor binding to an ERE, as measured *in vitro*, need not correlate with the potency of the enhancer function that is observed and that multiple EREs can work cooperatively in order to enhance target gene transcription. ## Two ERE sequences near the *PCNA* gene did not enhance reporter expression in response to E2 treatment We cloned genomic fragments corresponding to each PCNA-ERE into luciferase reporter constructs and tested these for enhancer function, *in vitro*. Surprisingly, neither an 822 bp fragment containing PCNA-ERE1 nor a 551 bp fragment containing PCNA-ERE2 demonstrated E2-inducible enhancer function in MCF7 cells. Similar results were obtained when these constructs were tested with co-expressed ERα in HESC cells (which do not express endogenous ERα but respond to E2 when made to express ERα (not shown)). Thus, changing the cell type in which we tested the reporter constructs, and presumably the cohort of available transcription factor co-regulatory proteins within the cell nucleus, failed to indicate enhancer function for the putative ERE sequences being tested. Luciferase reporter assays that use single copy response elements are sometimes weakly responsive to transcription factors. Further, the function of any cloned enhancer region may be subject to inhibition by neighboring *cis*- and *trans*-regulatory elements, depending upon the length of DNA that is cloned and the inclusion or exclusion of such regulatory elements in the reporter construct. This phenomenon could produce false-negative observations in luciferase reporter assays that would depend, in part, upon the size of the genomic fragment that is employed in the assay. It is, therefore, common to assay multiple tandem copies of *cis*-regulatory elements in order to demonstrate enhancer function, *in vitro*. We engineered luciferase reporter constructs with two tandem copies of the 15 bp PCNA-ERE1 sequence or two copies of the PCNA-ERE2 sequence (2×PCNA-ERE1 and 2×PCNA-ERE2), to test these isolated sequences for enhancer activity. As seen with the larger ERE-containing genomic fragments, the 2× tandem ERE sequences did not demonstrate E2-responsive enhancer function in MCF7 cells. When each tandem ERE sequence was mutated by one nucleotide to conform to a perfect 13 bp consensus ERE (Δ-2×PCNA-ERE1 and Δ-2×PCNA-ERE2), strong E2-responsive enhancer function was observed. These results confirmed that the promoter-reporter construct, pGL2-promoter, is functional in response to E2 in MCF7 cells when harboring *bona fide* ERE enhancer sequences. Taken together, these data did not support the hypothesis that the putative EREs, each computationally detected, are likely to function as ERα-regulated enhancer elements for the *PCNA* gene in MCF7 cells. ## Two ERE sequences near the *PCNA* gene were not bound by ERα *in vivo* In light of the fact that the ERE-like sequences near *PCNA* are nearly consensus EREs, and that *PCNA* gene expression is E2-responsive, we wondered whether the *in vitro* assays that we employed to detect enhancer function could have produced spurious results. It remained possible that the chromatin context, *in vivo,* might dictate enhancer function in ways not observed using plasmid DNA in reporter assays. Imperfect (i.e. non-consensus) EREs have been demonstrated to have function when optimally positioned with regards to target TSSs, wherein they can provide sufficient affinity for ERα binding, permit important DNA bending, and favor specific patterns of coregulator recruitment to the target gene promoter. Although we were unable to detect ERα binding to PCNA-ERE1 or PCNA-ERE2 using ChIP-on-chip in response to E2, we reasoned that this result could reflect established limitations in the sensitivity of this microarray-based approach. When we compared our findings to other reports of genome-wide location analysis for ERα in MCF7 cells, we similarly found no evidence for recruitment of ERα to the PCNA gene locus using ChIP-on-chip or alternative genomic approaches such as ChIP-DSL and ChIP-PET. Notably, the receptor location analysis work in these studies compared ERα target occupancy after chromatin immunoprecipitation using E2-treated cells relative to ERα occupancy of chromatin DNA prepared from cells not subjected to immunoprecipitation (i.e. sheared input). As such, the approaches were not optimally designed to detect changes in target occupancy that depend upon E2 exposure, which can be tested by comparing ChIP of untreated cells (vehicle-treated ChIP) with ChIP from E2-treated cells. In order to test for E2-dependent binding of ERα to PCNA-ERE1 or PCNA-ERE2, we performed ChIP-PCR using primers that span each putative ERE. Using ChIP-PCR, we were unable to demonstrate recruitment of ERα to either putative ERE in response to E2. Similar results were obtained using multiple alternative PCR primer pairs targeting the same genomic regions (not shown). *In toto*, more than 3 primer pairs were used to evaluate ERα occupancy at each putative *cis*-element in response to E2; all demonstrated no evidence for receptor recruitment to the ERE sequences in response to hormone. The ChIP-PCR data shown in included a 45 min exposure to E2, a treatment that has been demonstrated to produce optimal ERα recruitment to target enhancers. We obtained identical results with E2 treatments extended for 3 and 6 hours prior to protein crosslinking (not shown). A genomic locus (chr7:72384008-72385027) that lacks an ERE and that failed to be detected using ChIP-on-chip (dubbed ERE-neg) also failed to be enriched for ERα using ChIP-PCR when comparing control and E2-treated cells. By contrast, ChIP- PCR confirmed E2-dependent recruitment of ERα to several known genomic targets including *TFF1*, *MYC*, *GREB1*, and *CTSD.* In addition, ChIP-PCR confirmed E2-dependent recruitment of ERα to four targets that we previously identified using ChIP-on-chip, dubbed I20, J23, D54, and F15 (see for genomic locations). These data suggest that neither of the putative EREs detected near *PCNA* binds to ERα in an E2-dependent fashion, *in vivo*, in MCF7 cells. Further, the data demonstrate a good correlation between our ChIP-on-chip datasets and ChIP-PCR datasets. # Discussion Our data show that PCNA, an essential participant in DNA replication, epigenetic programming, and a regulator of the cell cycle, is up-regulated by E2 exposure in MCF7 cells. Epigenetic alterations play critical roles in tumorigenesis and cancer progression, and we recently demonstrated that high protein expression of the E2-responsive histone variant H2A.Z in primary breast tumors correlates with decreased patient survival. It is attractive to consider, and previously has been postulated, that the E2-dependent expression of *PCNA* is potentially important for the proliferation of a diversity of tissues and tumors. Nair and colleagues recently reported an ERα-dependent proliferative effect in which estrogen enhanced *PCNA* gene expression in cancers of the cervix but not in normal cervix. E2 exposure is a well-recognized risk factor for cancer of the breast and endometrium, and E2 enhances *PCNA* gene expression in human myometrial and leiomyoma tissues as well. Most-recently, E2-responsive tumor progression has been suggested for epithelial ovarian cancer. Our data indicate that, in MCF7 cells, the E2 effect depends upon new gene transcription and translation and is blocked by the pure ERα-antagonist ICI 182,780. Although excellent computational analysis, and preliminary *in vitro* data (EMSA), suggested that the mechanism of *PCNA* gene regulation might include the function of ERE sequences in the 5′ region of the gene, our testing of the most-likely sequence elements, using several approaches, failed to confirm *cis*-regulatory function in MCF7 cells. That a subset of ERα target genes may require the synthesis of additional cofactor(s) prior to becoming subject to receptor-mediated transcriptional regulation remained a possibility consistent with our preliminary gene expression data. Such a model has been observed for ERα gene targets that are repressed by ERα only after the induction of corepressor protein NRIP-1. Similarly, some genes targeted by ERα require the function of a chromatin modifier, FOXA1, prior to becoming subject to ERα-mediated transactivation. The observation that E2 engenders gene regulatory cascades that can be divided into temporal categories (i.e. immediate, early, and late responses) is well- demonstrated. These data indicate that some genes are poised for immediate regulation, others may represent downstream/indirect targets of ERα function, and still other gene targets may require the synthesis of cofactors to modify chromatin targets in preparation for the arrival of activated ERα. Neither the timing of a transcriptional response (early vs. late) nor the sensitivity of the response to CHX can be taken as formal proof that a given response is direct (i.e. ERα-mediated) or indirect. This report describes two ERE-like sequences upstream of an E2-responsive gene. Both sequences failed to demonstrate transcriptional regulatory function *in vitro* and *in vivo* in MCF7 cells. PCNA-ERE1 resides within a repetitive element (Alu-Sc) while PCNA-ERE2 does not. A single report described a functional Alu-ERE for the BRCA1 gene which, on further inspection, was determined to be non-functional in MCF7 cells. The reported ERE sequence in those studies is not the same as noted for PCNA-ERE1. Our results indicate that *PCNA* is regulated in response to E2 either indirectly, or via a *cis*-acting ERE not detected by several independent genome-wide location analyses for ERα, and only after the synthesis of newly translated protein(s). Importantly, in addition to ChIP-on-chip approaches, two genome-wide location analyses utilized ChIP-PET and the sensitive ChIP-DSL approaches; all demonstrated no evidence for ERα recruitment to *PCNA*. Five members of the E2F family of transcription factors (of which humans have at least eight) are up-regulated by E2 in MCF7 cells although direct transcriptional regulation by ERα remains to be established for these genes. Recent data suggests that E2F family members are capable of binding to identical sequences as homodimers or heterodimers (with DP family members) and may often subserve redundant functions. Increasing data indicate that these factors play important roles in the E2-dependent cellular proliferation of breast cancer cells. E2F1 regulates *PCNA* gene expression in some systems. We tested E2F1 using ChIP-PCR and found no evidence for E2-dependent recruitment of E2F1 to the transcription start sites of the *PCNA* gene (data not shown). Similar negative results were obtained when performing ChIP-PCR using antibodies for Sp1 or AP-1 in order to interrogate both *PCNA* ERE-like sequences and the two TSSs of the *PCNA* gene (not shown). ATF1 and CREB1 are also regulators of *PCNA* gene expression but were not enriched at the *PCNA* promoter in MCF7 cells in genome- wide ChIP-chip analyses (S. Hua and K. White, manuscript in preparation). In order to identify alternative candidates that might mediate the estrogen response of *PCNA* in MCF7 cells, we undertook a computational analysis of predicted transcription factor binding sites in the two promoter regions for *PCNA.* These data were then correlated with gene expression data from our work, and from the work of others, cataloging estrogen-responsive transcription factors in MCF7 cells. The intersection of these datasets provides a list of estrogen-responsive transcription factors with high-confidence binding sites residing within two kilobases of each transcription start site for the *PCNA* gene. In addition to E2F family members that warrant investigation (above), we have identified c/EBPβ, FOXC1, FOXJ2, GATA-3, POU2F1/AP-2γ, RARA, TFAP2C, and TFE3 as estrogen-responsive transcription factors with predicted binding sites in one or both promoter regions of the *PCNA* gene. Taken together, these data reveal good candidates for the mediator of the estrogenic cascade leading to enhanced *PCNA* gene expression in MCF7 cells. These candidates will be pursued in our future studies. Notably, ∼2310 perfect 13 bp consensus EREs (GGTCAnnnTGACC) exist within the human genome. Permitting just one nucleotide mismatch from the consensus sequence reveals nearly 50,000 ERE-like sites throughout the human genome. Our ChIP-on-chip data, supported by biological plausibility, suggest that the overwhelming majority of these sites are not functional in any given cell type. Two groups have used MCF7 cells to perform whole genome ChIP-on-chip for ERα location analysis, revealing between ∼1600–3700 receptor-bound loci in response to E2 (with considerable reproducibility). Analysis of the highest-confidence ChIP sites, the 1017 sites that are common to both datasets, indicates that more than 90% of the ERE-like sequences at these loci are not perfect consensus sequences. Further, these datasets indicate that, most likely, \<5–10% of perfect consensus ERE sites in the genome are ERα-bound in response to E2 in MCF7 cells. The chromatin and cellular determinants of ERα binding to enhancer elements remain to be fully established. This issue raises a cautionary note when drawing conclusions based upon computational analyses of genomic sequence which, even when evolutionarily conserved, will present hypotheses that must be validated by formal molecular biological testing. We conclude that computational detection of *cis*-regulatory elements in the human genome, even when accompanied by appropriate gene expression data, cannot be taken as proof of *cis*-regulatory element function *in vivo*,. Each element must be tested, preferably using *in vivo* assays such as ChIP-PCR and chromatin conformation capture, in order to confirm *cis*-element function in any given cell type and cell context. # Materials and Methods ## Cell Culture MCF7 cells (ATCC) were grown as described. Cells were changed to estrogen- depleted, phenol-free media consisting of MEM alpha (Gibco) with 10% charcoal/dextran-stripped calf serum, insulin (4 µg/ml, Sigma), penicillin G, streptomycin, and L-glutamine (all Gibco), for 72 hours prior to treatments. Where indicated, treatments included vehicle control (100% EtOH), estradiol (10 nM or 100 nM, Sigma), actinomycin D (ActD, 2 µg/mL, Sigma), cycloheximide (CHX, 25 µg/mL, Sigma), and ICI 182,780 (1 µM, Tocris Biosciences). Telomerase- immortalized Human Endometrial Stromal Cells (HESC cells), a generous gift from Dr. Graciela Krikun, were grown in the same media used for the MCF7 cells. HESC cells have normal chromosome numbers and structures. ## Preparation of Nuclear Extracts and EMSA HESC nuclear extracts (NE) were purified using NE-PER Nuclear and Cytoplasmic Extraction Reagents (Pierce) according to the manufacturer's instructions. HESC cells have no demonstrable ERα activity using sensitive luciferase reporter assays and no ERα protein detected by Western blot analysis (data not shown). However, HESC cell nuclei have cofactors that promote the binding of recombinant ERα to target DNA in EMSA and these factors enhance binding when compared to recombinant ERα alone (rERα, Affinity Bioreagents). EMSA experiments were therefore conducted using HESC nuclear extracts combined with rERα. Protein determinations were performed using the Micro BCA assay (Pierce) and 5 µg of nuclear extract (with protease inhibitors, Roche) plus rERα (170 nM) was used in each lane of a 5% acrylamide gel in TBE buffer. Oligonucleotide probes were ³²P-labeled using T4 Polynucleotide Kinase 5′ End Label System (Promega) and purified using Mini Quick Spin Oligo Columns (Roche). Each radiolabeled probe was used at ∼200,000 cpm/lane and binding reactions included Tris-HCl pH 8.0 (25mM), KCl (50 mM), MgCl2 (6.25 mM), Glycerol (10%), DTT (0.5 mM) plus relevant antibody where indicated (400 ng/reaction): anti-ERα Ab-10 (LabVision), anti-cJun(N) sc-45 (Santa Cruz), anti Sp1 H-225 sc-14027 (Santa Cruz) and anti-ARP1/COUP-TFII (Santa Cruz). A complete list of oligonucleotide sequences used as probes for EMSA is presented in the supplementary materials. ## RT-PCR Total RNA was purified from cell lysates using Trizol reagent (Invitrogen). 2 µg of total RNA was used for reverse transcription using Anchored Oligo-(dT)₂₃ (Sigma) as primer for 1st strand synthesis using the RT- AMV kit (Roche). 1:100 dilutions of cDNA were used as template for quantitative PCR using iQ-SYBR Green Master Mix (Biorad) in a Biorad Opticon 2 cycler. Q-RT- PCR values were normalized to *ACTB* mRNA levels for all samples. Primer pairs for RT-PCR of *ACTB, PCNA*, *TFF1*, *MYC*, *STC2*, and *DCC1* are listed in the supplementary materials. ## Western Blotting Cell lysates in 1% SDS lysis buffer were quantified using the Micro BCA Protein Assay Kit (Pierce), and 30 µg of total protein per well was separated by SDS- PAGE, transferred to PVDF membranes, and probed with antibodies against ACTIN (Sigma \#A4700, used at 1:500 dilution) or PCNA (Cell Signaling \#2586, used at 1:1000 dilution). Secondary Goat anti-mouse IgG antibody (conjugated with horseradish peroxidase) was incubated at a dilution of 1:10,000 and blots were developed using Amersham ECL Plus Western Blotting Detection Reagents (GE Healthcare). ## Chromatin Immunopreciptation (ChIP)-PCR ChIP was performed as previously described. Briefly, MCF7 Cells were E2-deprived for 3 days (details above) and then treated with 10 nM E2 or vehicle (45 minutes) at 80% confluence. 45 minutes of E2 exposure has been demonstrated to produce maximal ERα binding to chromatin. ∼5×10⁶ cells per ChIP were cross-linked with 1% formaldehyde for 10 minutes at 37°C then quenched with 125 mM glycine. The cells were washed with cold PBS and scraped into PBS with protease inhibitors (Roche). Cell pellets were resuspended in ChIP lysis buffer (1% SDS, 10 mM EDTA, 50 mM Tris-HCl (pH 8.1) and sonicated (Fisher Sonic Dysmembrinator) to produce sheared chromatin with average length 500 bp. The sheared chromatin was submitted to a clarification spin and the supernatant then used for ChIP or reserved as “Input.” Antibodies used were anti-ERα (Ab-1, Ab-3, and AB-10 from Lab Vision and MC-20 from Santa Cruz). Forward and reverse primer sequences used for ChIP-PCR are listed in the supplementary materials. ## Luciferase Reporter Assays Luciferase reporter assays were performed using the Luciferase Assay System (Promega) according to the manufacturer's instructions. Potential regulatory elements were cloned into pGL2-Promoter (Promega) and transfected into MCF7 cells using the TransIT-LT1 Transfection Reagent (Mirus). Cotransfection with a β-galactosidase expressing plasmid (Promega) enabled normalization of transfection efficiency across samples using a β-galactosidase assay kit (Promega) according to the manufacturer's instructions. ## Cloning and Mutagenesis PCR cloning was performed using PCR amplification of genomic loci from HESC cell genomic DNA which was prepared using the Genomic DNA Extraction kit (Qiagen) according to the manufacturer's instructions. PCR products were ligated with the reporter construct pGL2-promoter (at 5′-KpnI+3′-XhoI sites) for use in Luciferase Reporter assays (Promega). Mutagenized reporter constructs were prepared using the Genetailor Site-Directed Mutagenesis System (Invitrogen) according to the manufacturer's instructions. All clones and subclones were confirmed by DNA sequencing. Primers used for genomic locus amplification and for subcloning are listed in the supplementary materials. ## Statistics Comparisons between two groups were made using a two-tailed Student's t-test with P values indicated. # Supporting Information The authors are grateful to Drs. Graciela Krikun and Charles Lockwood for providing the immortalized human endometrial stromal cell line, and to Drs. Sujun Hua, Kevin P. White, Joshua R. Friedman, Neil Sidell, and Robert N. Taylor for scientific discussions. [^1]: Conceived and designed the experiments: CBK. Performed the experiments: CW JY. Analyzed the data: CW CBK. Contributed reagents/materials/analysis tools: CBK. Wrote the paper: CBK. [^2]: The authors have declared that no competing interests exist.

# Introduction Homeobox-containing transcription factors represent an important class of factors involved in the regulation of embryogenesis and other molecular programs. *HMX1* is a homeobox-containing transcription factor implicated in eye development. In 1992, Stadler et al. described a new homeobox gene called *GH6*. This gene was later renamed *HMX1* and was assigned to the NKX5 family, the reason why *HMX1* is also known as *NKX5-3*. Later, further members were identified: *HMX2* (*NKX5-2*), *HMX3* (*NKX5-1*) and, in chicken, zebrafish and medaka, *SOHo-1*. The NKX5/HMX family of transcription factors contains a unique homeobox region that is phylogenetically conserved. *HMX1*, *HMX2* and *HMX3* contain two other conserved domains called SD1 and SD2, located immediately C-terminally to the homeobox. The function of these domains is still unknown. Whereas Hmx2 and Hmx3 play a role in inner ear development, Hmx1 and SOHo-1 are mainly implicated in eye development. In the mouse eye, *Hmx1* expression can be detected as early as E10.5, and transcripts are more specifically present in the lens and in the antero-medial part of the neural retina. In the developing chicken eye, it is expressed in the dorsal neural retina and lens epithelium as well as in the optic nerve. *HMX1* expression starts 40 hours into development (stage 11) in the surface ectoderm surrounding the optic vesicle. At optic cup invagination (stage 14–15), it is expressed in the anterior/nasal side of the early retina. In zebrafish, *hmx1* is first expressed in the entire eye at 10 somite-of-stage (ss), and is then repressed in the dorsal part at 18 ss. At 24 hours post fertilization (hpf), it is restricted to the nasal retina and, one day later, expression is restricted to the nasal part of the ganglion cell layer (GCL). At four and five days post fertilization, signal is also observed in the nasal part of the inner nuclear layer (INL). In the developing lens, expression is observed from 24 to 72 hpf. We recently reported a family with a 26-bp deletion in exon 1 of *HMX1* leading to the oculo-auricular syndrome of Schorderet-Munier-Franceschetti (OMIM: 612109), characterized by microphthalmia, microcornea, nystagmus, cataract, coloboma, optic nerve dysplasia, RPE abnormalities, rod-cone dystrophy and deformation of the ear lobule. A mouse model containing a mutation in *Hmx1* has been described. It shows laterally protruding ears, subtle changes in cranial bone morphology, perinatal semi-lethality, reduced body mass and microphthalmia with low-grade keratoconjunctivitis sicca and entropion. The eyes show no evidence of microcornea, anterior segment dysgenesis, cataract, coloboma, retinal detachment or retinal dysplasia. Quina et al. observed a significant reduction of geniculate ganglion neurons. *In vitro*, HMX1 binds to a 5′-CAAGTG-3′ sequence, represses transcription from a luciferase reporter containing this binding site and can antagonizeNKX2.5, a cardiac homeo protein, which is activating this same reporter construct. Nkx2.5 is also known to dimerize at its homeodomain and other regions in the C-terminus. In this study, we showed that HMX1 acted as a dimer and that the homeobox and the conserved domain SD1 were needed for dimerization to occur. SD2 was not involved in the dimerization process. We also identified *EPHA6* as a target of HMX1 and showed that HMX1 repressed the *EPHA6* promoter *in vitro*. The inhibitory activity of HMX1 was associated with the presence of the SD1 and homeobox domains. Whereas the *EPHA6* inhibition was lost with mutants of each of these 2 domains, the SD2 mutant showed a small activation of the *EPHA6* promoter. Mutation of the three CAAG(TG) sequences of the promoter attenuated the repression by HMX1. This inhibition was confirmed *in vivo* in zebrafish embryos. # Materials and Methods ## Plasmid Constructions Subcloning was performed according to standard protocols. Mutagenesis was performed using the QuickChange II Site-Directed Mutagenesis Kit (Stratagene, Agilent Technology AG, Basel, Switzerland). The sequence of the primers used in this study is available from the authors. ## Cell Culture and Transfection Human embryonic kidney (HEK) 293T cells were cultured at 37°C and in 5% CO₂ atmosphere, in Dulbecco’s Modified Eagle’s Medium (DMEM) high glucose with stable glutamine (GE-Healthcare, Glattbrugg, Switzerland), supplemented with 10% FBS (Lonza, Basel, Switzerland), 100 U/ml penicillin and 100 µg/ml streptomycin (Invitrogen, Basel, Switzerland). Transfection was performed using the Calcium Phosphate method (ProFection Mammalian Transfection System, Promega, Dubendorf, Switzerland). ## BRET² 200’000 HEK 293T cells in DPBS were distributed into black 96-well microplates for fluorescence quantification. Filter sets were adapted to 485 nm for GFP² excitation and 510 nm for emission. Cells expressing BRET² donor (RLUC) alone were used to determine the fluorescence background. 200’000 cells with comparable fluorescence levels were distributed into white 96-well microplates for luminescence quantification. The luciferase substrate Coelenterazine 400A, DeepBlueC (Chemie Brunschwig, Basel, Switzerland) was added to a final concentration of 5 µM. Filter sets were adapted to 410 nm for luciferase emission and 515 nm for GFP² emission. The emitted fluorescence and luminescence were measured using an Envision 2103 Multilabel Reader (PerkinElmer, Schwerzenbach, Switzerland), and analyzed with the Wallac Envision Manager V1.12 software (PerkinElmer, Schwerzenbach, Switzerland). ## Co-immunoprecipitation 200 µg of proteins were immunoprecipitated overnight at 4°C on a rotating wheel with 2.5 µl anti-Renilla Luciferase antibody (MAB4400, Millipore, Zug, Switzerland). 20 µl of washed protein G plus agarose beads (Santa Cruz, LabForce AG, Nunningen, Switzerland) were added and incubated 2 hrs at 4°C on the rotating wheel. After centrifugation at 4°C, the supernatants were kept as controls. The pellets were resuspended in 25 µl 2x SDS loading buffer and loaded on a 12% SDS-page gel, alongside with 20 µl of supernatant and 40 µg of proteins. ## Western Blot Proteins were extracted from cell cultures using RIPA (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS) and concentrations measured using the Micro BCA Protein Assay Kit (Thermo Fisher Scientific, Reinach, Switzerland) on a Multiplate Reader Synergy HT (Bio-Tek, Luzern, Switzerland) with the KC4 software. The following antibodies were used: HA-Tag (6E2) Mouse mAb \#2367 (Cell Signaling, LabForce AG, Nunningen, Switzerland), GFP N-terminal G1544 (Sigma, Buchs, Switzerland), PARP (46D11) Rabbit mAb \#9532 (Cell Signaling, Labforce AG, Nunningen, Switzerland), p62/SQSTM1 P0067 (Sigma, Buchs, Switzerland), Ub (A-5) sc-166553 (Santa Cruz Biotechnology, LabForce AG, Nunningen, Switzerland) and α-Tubulin Clone B-5-1-2 T5168 (Sigma, Buchs, Switzerland). ## Native Western Blot Cells were lysed in a non-denaturing lysis buffer (20 mM Tris-HCl, 137 mM NaCl, 10% glycerol, 1% Triton X-100, 2 mM EDTA, pH 8.0). The protein concentrations were measured as described above. 5 µg were loaded on a Mini-Protean TGX precast gel 4–15% (BioRad Laboratories AG, Cressier, Switzerland) with non-denaturing loading buffer (300 mM Tris-HCl pH 7.8, 30% glycerol, 0.6% bromophenol blue) and migrated without denaturation in a running buffer without SDS. ## GFP² Fluorescence Imaging and Nuclei Isolation Cells were analyzed 48 hrs post-transfection under a Zeiss Axiovert 200 microscope with filters adapted for excitation and emission at λex = 480 nm and λem = 510 nm, respectively, and the AxioVision 4.2 software. For nuclei isolation, cells were counted and resuspended at 10⁸ cells/ml Nuclei Isolation Buffer (250 mM sucrose, 20 mM Hepes pH 7.8, 10 mM KCl, 1.5 mM MgCl₂, 0.5 mM spermidin). Cells were then homogenized with a Potter and spread on a slide. ## In Silico Search for a Nuclear Localization Signal (NLS) The mouse HMX1 sequence was entered in the NLS-Mapper software that can be found at <http://nls-mapper.iab.keio.ac.jp/cgi-bin/NLS_Mapper_form.cgi>. ## Immunofluorescence Immunofluorescence was performed 24 hrs post transfection. When necessary, 50 µM chloroquine were added for 16 hrs. The primary antibody (LC3B \#2775, Cell Signaling, LabForce AG, Nunningen, Switzerland) was diluted in 1x PBS +2% NGS +0.2% Triton X-100 and incubated overnight at 4°C in a humid chamber. The secondary antibody (Alexa Fluor 594 goat α-rabbit IgG (H+L) (A11012), Molecular Probes, LubioScience, Luzern, Switzerland) was diluted in the same buffer, and incubated 1 hr at RT in a humid chamber in the dark. Nucleic acids were stained with 100 µM DAPI (4,6-diamidino-2-phenyl-indole HCl) (1/1′500 in 1x PBS) for 10 min in a humid chamber in the dark. Cells were then mounted with Citifluor AF1 (Citifluor Ltd, Leicester, UK), and conserved at 4°C. The slides were analyzed under an Olympus BX61 microscope and the Cell^M software (Olympus, Volketswil, Switzerland). ## Hoechst-PI Staining 20 mg/ml bisBenzimide H 33342 trihydrochloride (Sigma, Buchs, Switzerland) and 1 mg/ml Propidium Iodide (Fluka, Buchs, Switzerland) were diluted 1/2′000 into the culture medium. Cells were analyzed under a Zeiss Axiovert 200 microscope and the AxioVision 4.2 software. ## Luciferase Assays 48 hrs post transfection cells were washed with 1x PBS, and 300 µl luciferase assay lysis buffer (100 mM K₂HPO₄ pH 7.8, 0.2% Triton X-100) were added. Cells were scraped at 4°C and centrifuged 3 min at 12′000 rpm at 4°C. 5 µl of supernatant were transferred to a transparent 96-well plate containing 50 µl 2x β-gal buffer (120 mM Na₂HPO₄, 80 mM NaH₂PO₄, 2 mM MgCl₂, 100 mM β-mercaptoethanol). 50 µl of 2x ONPG (1.33 mg/ml 2-nitrophenyl-B-D-galactopyranoside) were added and the plate read at 412 nm of absorbance on a Multiplate Reader Synergy HT (Bio- Tek, Luzern, Switzerland) with KC4 software. If the values were constant in all conditions, 5 µl of supernatants were transferred to a white 384-well plate, 20 µl of Luciferase Assay Reagent (Promega, Dubendorf, Switzerland) were added and luminescence measured on the Multiplate Reader Synergy HT every 3 minutes until the peak of luciferase activity was reached. The obtained values were normalized using a β-gal reporter under the control of a CMV promoter. A mean between the 3 highest values was used for the luciferase/β-gal ratio. Each experiment was performed three times in duplicates. Only transfections with stable β-gal values between the different conditions, indicating similar transfection efficiency, were used. Two-tailed Student’s T-tests with unequal variance were used to determine statistical differences between the conditions. ## Chromatin Immunoprecipitation All experiments involving live animals were authorized by the Veterinary Service of the State of Valais under authorizations N° VS-13 and VS-19. The litter of four was housed with the mother and was anesthetized with isoflurane prior to being euthanized by cervical dislocation. Retinas from four 2-week-old wild-type C57Bl/6J mice were dissected, fixed, and homogenized. Glycine was added to a final concentration of 0.125 M before centrifugation. The pellet was resuspended in nuclei lysis buffer (50 mM Tris-HCl pH 8.0, 10 mM EDTA pH 8.0, 1% SDS). The resulting chromatin was sonicated, snap-frozen in liquid nitrogen, and kept at −80°C. The next day, the tube was centrifuged, and the supernatant transferred to new eppendorf tubes containing 10 µl of Protein A-Agarose beads (Roche, Basel, Switzerland). 100 µl of supernatant were pre-cleared with the 10 µl of beads for 1 hr on a rotating wheel at 4°C. The tubes were centrifuged and the supernatants transferred to new tubes. A control tube containing 10 µl 5% BSA and a test tube with 2 µl Hmx1 antibody (ARP32629_P050, Aviva, LubioScience, Luzern, Switzerland) were prepared. The tubes were incubated overnight at 4°C on a rotating wheel. After 10 µl of Protein A-Agarose beads were added to both tubes and another incubation, the tubes were centrifuged and the supernatants were kept at −80°C ( = TIC). The pellets were washed in IP wash buffer n°1 (0.1% SDS, 1% Triton X-100, 20 mM EDTA pH 8.0, 150 mM NaCl, 20 mM Tris-HCl pH 8.0), 4 times in IP wash buffer n°2 (0.1% SDS, 1% Triton X-100, 2 mM EDTA pH 8.0, 500 mM NaCl, 20 mM Tris-HCl pH 8.0), once in IP wash buffer n°3 (250 mM LiCl, 1% NP-40, 1% deoxycholate, 1 mM EDTA pH 8.0, 10 mM Tris-HCl pH 8.0) and once in TE 10∶1 (10 mM Tris-HCl pH 7.5), 1 mM EDTA pH 8.0). The antibody was eluted from beads by adding 150 µl IP elution buffer (50 mM NaHCO₃, 1% SDS) twice and by shaking 15 min at RT. 12 µl 5M NaCl and 1 µl RNase A (10 mg/ml) (Roche, Basel, Switzerland) were added and the tubes were incubated 5 hrs at 67°C. The TIC samples were thawed and 100 µl transferred to new tubes. 500 µl IP elution buffer, 24 µl 5 M NaCl and 2 µl RNase A (10 mg/ml) were added, and the tubes were incubated 5 hrs at 67°C. After incubation, 2.5 volumes 100% EtOH were added for precipitation overnight at 4°C. The next day, the tubes were centrifuged and the pellets were dissolved in 100 µl TE 10∶1. 25 µl proteinase K buffer for ChIP (50 mM Tris-HCl pH 7.5, 25 mM EDTA pH 8.0, 1.25% SDS) and 1 µl proteinase K (Roche, Basel, Switzerland) were added, and the tubes incubated 2 hrs at 45°C. 175 µl TE 10∶1 and 300 µl Phenol:Chlorophorm:Isoamyl Alcohol 25∶24∶1 (Sigma, Buchs, Switzerland) were added, the tubes shaken and centrifuged. 30 µl of 5 M NaCl, 1 µl of 5 mg/ml glycogen and 750 µl of 100% EtOH were added to the supernatants, mixed and precipitated overnight at 4°C. The next day, the tubes were centrifuged, the supernatants were removed and the pellets resuspended in 30 µl TE 10∶1. PCR analysis was performed on 2 µl of samples. ## Generation of the Zebrafish Hsp70-HMX1 Transgenic Line AB zebrafish were raised and kept under standard laboratory conditions at 28.5°C. Transgenesis was performed by generating Tol2 transposon constructs using the tol2kit. The zebrafish *hmx1* coding sequence was cloned downstream of the hsp70 promoter and the DNA construct together with the transposase mRNA were injected at the one-cell stage. Fish were raised to adulthood and the cardiac GFP expression was used as a marker for germline transmission. Experiments were done on F3 obtained from F2 that were intercrossed in order to increase the number of larvae carrying the transgene. Tg (hsp70: hmx1) and wt were heat shocked at 1 dpf during 30 min at 39°C, euthanized and fixed 4 hrs after. ## Whole-mount *in situ* Hybridization Standard one-color whole-mount *in situ* hybridization was performed at various stages. Hybridization reaction was done at 68°C for 14–18 hrs. Washing steps and antibody incubation were performed in an *in situ* machine (BioLane HTI, Hölle&Hüttner, Tubingen, Germany). Templates used to generate DIG-labeled RNA probes included zebrafish *hmx1* (ID: 797503), *epha4b* (ID: 64270) and *pax6* (ID: 60639). *In vitro* transcription was done with the Roche RNA Labeling Kit (Roche Applied Science, Basel, Switzerland). # Results ## HMX1 dimerizes through the SD1 and homeobox domains The ability to homo- or heterodimerize has been demonstrated for many transcription factors, including the NKX member NKX2-5, a cardiac homeobox gene that dimerizes through its homeodomain (HD). We therefore investigated, using a BRET² approach, whether HMX1 behaved similarly and formed dimers or oligomers. The BRET² technique is based on the energy transfer occurring between the renilla luciferase (RLUC) and the green fluorescent protein (GFP) when they are in close proximity. The principle of the technique is to generate fusion proteins between proteins of interest and the RLUC and the GFP, and measure the energy transfer in culture conditions to determine if the proteins of interest are interacting. When plasmids expressing fusion proteins between RLUC and HMX1, and GFP² and HMX1 were mixed and transfected in HEK 293T cells, a robust increase in the BRET² ratio was observed with increasing concentrations of GFP²-HMX1, indicating that HMX1 dimerized. We confirmed dimerization of HMX1 by co-immunoprecipitation (co-IP) using an RLUC antibody for immunoprecipitation and a GFP antibody for blotting. Out of the six conditions tested, the only condition in which immunoprecipitation occurred was when the two different HMX1 fusion proteins were present. Non-denaturing electrophoresis was also used to further confirm this result. As no western-blot suitable antibody against HMX1 existed, we tagged HMX1 with an HA-tag and used antibodies against HA to visualize the fused HA-HMX1 protein. HA-tagged wild- type HMX1 proteins were loaded on a non-denaturing native electrophoresis gel and sizes were compared to a denaturing gel after western blot analysis with an anti-HA antibody. The size of the band in non-denaturing conditions was twice the size of the band in denaturing conditions, suggesting dimerization. As no bands of higher molecular weight were observed, it is unlikely that trimers or other multimers were formed. In order to determine the dimerization domain of HMX1, we generated deletions of various portions of the protein. *HMX1* is composed of two exons, with three conserved domains in exon two: the homeobox (HD), and two domains called SD1 and SD2, located 3′ to the HD and whose function is presently unknown. We deleted each of these domains separately by site-directed mutagenesis and repeated the BRET² experiments. As shown in, deletions of HD or SD1 led to the loss of dimerization, whereas deletion of SD2 had no effect. This suggested that HD and SD1 were implicated in the dimerization of HMX1. In order to confirm these results and to show that the HMX1 C-terminal region was not involved in dimerization, in contrary to that of NKX2-5, we generated serial deletions of the C-terminal part of the protein. None of these constructs prevented dimerization as shown by BRET². ## The Entire HD is Necessary for Correct Nuclear Localization of HMX1 Fusing HMX1 to the GFP² reporter allowed us to visualize its cellular localization. GFP alone localized to the cytoplasm, whereas GFP-HMX1 localized to the nucleus. All of the generated mutants retained this nuclear localization except one. The HD deletion mutant was expressed in a punctate manner in the nucleus as well as in the cytoplasm. This punctate phenotype could possibly be due to the loss of a nuclear localization signal located in the homeobox. We therefore tested the sequence for potential nuclear localization signals (NLS) using NLS mapper, a bioinformatic tool available online. The analysis of HMX1 revealed the presence of a monopartite NLS -RGGRRKKTRTVF-, with KKTRTVF corresponding to the very beginning of the HD, with a score of 9.5. Deleting this signal could thus explain why the GFP-HMX1 del HD protein lost its nuclear localization. To test this hypothesis, we reinserted the seven-amino acid KKTRTVF into the HMX1 del HD sequence. However, reinserting these amino acids did not modify the punctate expression and localization of this mutant. To verify if the predicted NLS needed additional amino acids to be functional, we generated a new mutant with a deletion of the C-terminal half of HD (30 amino acids). However, this construct was still expressed in a punctate manner similar to the deletion of the entire HD. The homeobox of HMX1 is of helix-turn-helix-loop-helix type. It is likely that removal of any part of this structure prevents the correct folding of the protein and that it activates clearance mechanisms. In an effort to determine the nature of the aggregates generated by the GFP-HMX1 del HD mutant, we tested several hypotheses. First, the shape, size and distribution of the aggregates suggested that they could be autophagosomes induced by the abundant expression of aberrant proteins. We therefore verified if GFP-HMX1 del HD colocalized with LC3B by immunofluorescence, but this was not the case. Even after treating cells with chloroquine to visualize autophagosomes, GFP-HMX1 del HD did not colocalize with autophagosomes. To confirm this result, we also tested whether *p62* and *ubiquitin* expression was increased in the presence of GFP-HMX1 del HD. The role of ubiquitin is to clear abnormal proteins by targeting them for degradation by the 26S proteasome. Poly-ubiquitinated protein aggregates are also sequestered in inclusion bodies containing p62, and the aggregates are cleared via autophagy. In our experiments, we did not observe any increase in expression of these two proteins, indicating that these mechanisms were not activated. To determine whether the cells were suffering from the presence of GFP-HMX1 del HD aggregates, we looked for the presence of increased apoptosis by PARP cleavage assay. Cleavage of PARP by Caspase-3 is a step in the cascade leading to apoptosis. However, we failed to show any such increase (data not shown). Moreover, no increased cell death was observed when performing a Hoechst-PI staining for dying cells (data not shown). The exact nature of these punctae could thus not be determined and we do not know at this time whether they represent pure HMX1 aggregates or a more complex structure. ## HMX1 Binds to the Promoter of *EPHA6/epha4b* and Inhibits its Expression HMX1 and SOHO-1 are defining the EPHA3 expression domain in the developing chick retina. Ephrins act as topographically specific repulsive guidance cues for ganglion cell axons. EPHA3 is expressed in a temporal\>nasal gradient in the developing chick retina and is present on ganglion cell axons during the time of target innervations. HMX1 and SOHO-1 are expressed in an inversed gradient to that of EPHA3 (nasal\>temporal), and when HMX1 and SOHO-1 are expressed ectopically, EPHA3 expression is lost. *EPHA3* thus appeared to be a good candidate as a target for HMX1. However, ephrins do not have the same patterns of expression and do not play the same roles between different species. In the ganglion cell layer, where Hmx1 is expressed on the nasal side, EphA5 and EphA6 (P0 mouse) and EPHA3 (chicken) are only expressed on the temporal side. Chicken EphA5 and EphA6 are uniformly expressed in the chick retina, and EphA3 is not expressed in the mouse retina GCL, –. Therefore mouse EphA5 and EphA6 seem to be functional homologs of chicken EPHA3, which suggests that HMX1 could repress the activity of the EphA5 or EphA6 promoter in mouse. In zebrafish, epha4b is expressed in the same temporal pattern as chicken EPHA3 and mouse EphA5 and EphA6, whereas epha6 is not expressed in the eye (not shown). Amendt et al. showed that HMX1 was preferentially binding to a CAAG(TG) sequence. The *EPHA6* promoter contains three such binding sites, the second being conserved between human and mouse (−39 relative to the ATG), whereas the *EPHA5* promoter does not contain any. *EPHA6* was also identified as a potential target of HMX1 using a predictive promoter model that we recently developed. We therefore analyzed the effect of HMX1 on the human *EPHA6* promoter. The technique we used allowed measuring the activity of the promoter by placing a luciferase reporter under its control. We subcloned a fragment spanning from −150 to +150 nucleotides relative to the *EPHA6* translation initiation codon into a luciferase reporter vector. This fragment represented the minimal *EPHA6* promoter with a 13-fold increased activity compared to pGL3-basic vector. Shorter fragments (−100 to +150 and −50 to +150) displayed reduced promoter activity (five- and three-fold increased activity, respectively, compared to pGL3-basic vector) whereas the +1 to +150 fragment displayed no promoter activity (data not shown). The measured activity values were normalized using a β-gal reporter under the control of a CMV promoter. The subcloned fragment contained three potential binding sites for HMX1: one CAAGTG in the forward direction at position –39, one CAAG in the forward direction (−20) and one CAAG in the reverse direction (−65). We performed luciferase assays with wild-type HMX1 and the mutants deleting the HD, SD1 or SD2. As shown in , HMX1 was inhibiting the *EPHA6* promoter activity by 42%. The physical interaction between HMX1 and the *EphA6* promoter was demonstrated by chromatin immunoprecipitation on retinas isolated from two-week-old C57BL/6J mice, a technique allowing to determine which proteins bind to a DNA fragment by crosslinking them and selecting for the fragments bound to the protein by immunoprecipitation and PCR amplification of the fragment. We also validated this interaction *in vivo* on the zebrafish *epha4b* gene, the functional homolog of *EPHA6*, having two HMX1 binding sites in its promoter. We generated a transgenic fish line expressing a ubiquitous heat-shock activated *hmx1* gene and showed by *in situ* analysis that the aberrant ectopic expression of hmx1 in the temporal retina reduced the expression of *epha4b*, which was not the case for the control gene *pax6*. We then checked whether dimerization was needed for HMX1 repressive activity. Mutant constructs preventing dimerization, i.e. deletions of HD or SD1, had no activity, indicating that only dimerized HMX1 regulates *EPHA6* expression. The mutant with a deleted SD2 domain slightly activated the *EPHA6* promoter, suggesting that this region might represent the binding site of a cofactor needed for the inhibitory activity of HMX1. In order to confirm that the CAAG/CAAGTG sites represented *bona fide* binding sites for HMX1, we mutated them into the sequences shown in red in. Control experiments showed that these mutations did not affect *EPHA6* promoter activity (data not shown). Transfection experiments using these mutated constructs showed a reduction of the inhibitory activity of HMX1 from 42% to 22%. This indicates that the CAAG/CAAGTG sites represent true binding sites for HMX1. However additional sites might exist, as deletion of CAAG/CAAGTG sites failed to completely abrogate the inhibition. # Discussion The interest in the HMX1 transcription factor has surged with the discovery in 2008 that it was causing the oculo-auricular syndrome of Schorderet-Munier- Franceschetti. In addition of being expressed in somatosensory organs, Hmx1 has been shown to retain a neuronal fate in migrating neural crest cells and to modulate the adrenergic/cholinergic program of sympathic neurons. It is also well expressed in sensory spinal and cranial ganglia. In *C. elegans*, the Mls-2 gene, a member of the HMX family, regulates cytoskeletal organization and cell elongation. However, few contributions have been published about its mode of action in the eye. We therefore investigated its role in eye development. We showed that HMX1 exerts an inhibitory effect on *EPHA6* and that dimerization is necessary for this activity. Luciferase assays are known for producing artefactual results. By increasing the number of replicates and analyzing only the experiments where all conditions showed similar transfection efficiencies, we were able to obtain stable results, which were further confirmed by ChIP and *in vivo* experiments in zebrafish. Mutations that removed the dimerization domains of HMX1, i.e. the HD and SD1 domains, abolished its inhibitory efficiency on *EPHA6* promoter. Removing the HD also perturbed the cellular localization of HMX1, which was no longer restricted to the nucleus. All other mutants, including deletion of the SD1 domain, maintained a strict nuclear expression indicating that SD1 is involved in dimerization while the HD is necessary both for dimerization and nuclear localization. In addition to ChIP validation in mouse retina, we also showed that ectopic overexpression of HMX1 in the whole eye in a zebrafish transgenic animal in which expression of HMX1 was under a heat-shock-inducible promoter was accompanied by a reduction of *epha4b* ocular expression, the zebrafish functional homolog of *EPHA6*. The role of the SD2 domain remains unknown. We showed that a deletion mutant, which was dimerizing normally, was not inhibiting the *EPHA6* promoter like the wild type protein, but was slightly activating it, instead. This conserved domain could thus be an interaction site for a cofactor necessary for the inhibition action of HMX1. When deleting the homeobox, we observed that the GFP-HMX1 fusion protein lost its specific nuclear localization, and became expressed in a punctate manner in the nucleus as well as in the cytoplasm. Our first hypothesis was that the protein lacking the homeobox was misfolded, and therefore activated clearance mechanisms, either by autophagy or by the proteasome degradation system. The homeobox of HMX1 has a well defined helix-loop-helix-turn-helix tertiary structure type. It is possible that deleting it entirely or part of it changes the three-dimensional structure enough to activate the clearance mechanisms for misfolded proteins. However, we could not detect any indication that these mechanisms were triggered. One of the main components of autophagosomes is LC3B, and we therefore tested whether it colocalized with GFP-HMX1 del HD. This was, however, not the case even after blocking autophagy using a chloroquine treatment. We did not observe an increase in expression of ubiquitin and p62, confirming that the HMX1 aggregates were not autophagosomes, and that the proteasome was not activated. The HMX1 del HD aggregates did not induce cell death either as we observed no increase in PI-stained cells compared to other transfections (not shown). Moreover, PARP was not cleaved by Caspase 3 in GFP- HMX1 del HD transfections, indicating an absence of apoptosis (not shown). Thus, we do not know at this time what is the exact nature of these GFP-HMX1 del HD aggregates and if they represent pure HMX1 aggregates or a more complex structure. In a previous study, we showed that a morpholino-based knock-down of zebrafish *hmx1* had no effect on retinal patterning, which is in contradiction to the results obtained previously in the chick retina and the results presented here. However, in chicken the relationship between HMX1 and EPHA3 was shown by overexpressing HMX1 on the temporal side of the retina where it is not expressed normally. The same procedure was used in the current work. In our previous study, *hmx1* was knocked-down on the nasal side of the retina, whereas the temporal part was unaffected by this procedure, and *epha4b* was able to play its role in the temporal retina. In summary, we showed that HMX1 exerts its inhibitory activity through a dimer and identified *EPHA6* as a target of HMX1. Identifying other targets will allow us to further understand the role of HMX1. We thank Drs Nathalie Allaman-Pillet, Sandra Cottet, Pascal Escher and Raphaël Roduit from the Institute for Research in Ophthalmology for their help and critics, Mr Cédric Schöpfer, Mrs Tatiana Favez and Céline Agosti for technical help, and Dr Nicole Renner and Mrs Suzan Houghton for editing the manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: FM GB DFS. Performed the experiments: FM GB. Analyzed the data: FM GB DFS. Wrote the paper: FM GB DFS.

# Introduction Choline (2-hydroxy-*N,N,N*-trimethylethanaminium) is an essential dietary nutrient with functions in three areas: as a source of labile one carbon units (CH₃, methyl); as a component of lipids including phosphatidylcholine, sphingomyelin and lipid mediators such as platelet activating factor; and as a component of the neurotransmitter acetylcholine. Recent interest in choline has focused on its role in neural development, with compelling evidence in rodents that maternal dietary choline deficiency in pregnancy alters fetal brain development, with effects that include decreased neural progenitor cells proliferation, increased apoptosis and global histone, DNA and gene specific hypomethylation, culminating in life-long alterations in cognitive and memory functioning. The role of choline as a source of methyl groups is complex and tightly inter- related with the amino acid methionine, as well as folate and vitamin B12, key vitamins which function in methyl group transfer, but require a source of methyl. In these pathways, choline is converted to betaine (*N,N,N*-trimethylglycine) which donates a methyl group to homocysteine to form methionine and dimethylglycine,. Dimethylglycine may be further metabolized to methylglycine (sarcosine) which is then converted to glycine, with each step donating a methyl group that can be used for synthesis of methylene tetrahydrofolate, which in turn can donate a methyl group to homocysteine for synthesis of methionine in a reaction requiring vitamin B12. Methionine is the precursor of *S*-adenosyl methionine (SAM), a crucial methyl donor for numerous cellular methylations including DNA, RNA and histone methylation, conversion of norepinephrine to epinephrine, synthesis of purines and thymidylate (components of DNA and RNA), creatine (energy storage as creatine phosphate), carnitine (fatty acid transport into the mitochondria) and polyamines (cell growth), and inactivation of catecholamines. Methyl transfer also forms a cycle between choline and phosphatidylcholine, since phosphatidylcholine can be synthesized from choline and diacylglycerol, or by sequential transfer of three methyl groups from SAM to phosphatidylethanolamine. Major dietary sources of choline include liver, eggs and milk. Betaine is also present in the diet, with rich sources being beets, spinach, quinoa, other whole grains and some shellfish. A dietary need for betaine in humans has not been established. Before birth, choline is transported across the placenta, with high concentrations of free choline in fetal plasma. Although altered brain development due to deprivation of maternal dietary choline during gestation is well-established in animals, evidence for a similar effect in human pregnancy is lacking. One recent observational study found no association between maternal choline status during pregnancy and childhood intelligence at 5 years of age. However, plasma free choline levels during the first half of gestation among pregnant women in our population overlap with the range of plasma free choline found in adults consuming choline deficient diets. Since this raises the possibility of choline insufficiency, we have examined the potential associations between maternal plasma free choline and related methyl metabolites at 16 weeks of gestation and infant mental and motor skill development at 18 months of age. The primary focus was early pregnancy based on extrapolation of the critical window for long term effects of choline deprivation on brain development in rodents to humans. # Methods ## Subjects This was a prospective study involving 154 healthy mother-infant pairs conducted in Vancouver, Canada. Healthy pregnant women expecting to deliver one infant with no anticipated maternal or fetal complications were enrolled at 16 weeks of gestation. The women were enrolled in a prospective study which involved investigation of the effect of the maternal status of n-3 fatty acid docosahexaenoic acid (DHA) on infant development, including intervention to increase the mothers’ DHA status. Women following a vegan diet, at risk for preterm delivery, or with any known infectious or metabolic disease, were not enrolled. Infant follow-up was done only for single-birth, full-term infants (≥37 weeks gestation) with no complications likely to interfere with growth and development, or feeding. Socio-demographic information was collected. Maternal IQ was assessed using the Test of Nonverbal Intelligence, Third Edition (TONI-3), which is a non-verbal test of intelligence. Usual dietary intakes were assessed using a food frequency questionnaire with the intakes of choline estimated using the United States Department of Agriculture (USDA) database on choline in foods. Infant birth weight, birth length and head circumference were obtained from medical records, and information on infant feeding was recorded monthly. Infant growth was measured to 18 months of age, and weight and length were converted to z-score using the World Health Organization database. The protocol was approved by the Committee for Ethical Review of Research Involving Human Subjects at the University of British Columbia and the British Columbia’s Children’s and Women’s Hospital. All mothers provided written informed consent prior to participation both for themselves and on behalf of their infants. ## Laboratory Methods Blood samples were collected from each woman at 16 and 36 weeks of gestation; the women were requested to refrain from eating after waking until blood collection. Plasma free choline, betaine, dimethylglycine, homocysteine, methionine and cysteine were measured in plasma using isotope dilution liquid chromatography-tandem mass spectrometry (LCMS/MS), as previously described. The intra- and inter- assay CVs for choline, betaine and dimethylglycine in our laboratory are 2.50% and 3.78%, 2.18% and 3.46%, and 2.42% and 3.75%, respectively. Total vitamin B12 (tB12) and holotranscobalamin (holoTC) were measured by microparticle enzyme immunoassay, and plasma folate was quantified by ion capture assay, all using an AxSym Analyzer (Abbott Laboratories, Abbott Park, IL, USA). RBC total lipids were extracted, ethanolamine phospholipids (PE) were separated, and fatty acids were analyzed using gas-liquid chromatography with flame ionization detection for assessment of maternal DHA status. ## Assessments of Infant Neurodevelopmental Outcome Infant neurodevelopment was assessed at 18 months of age using the Bayley Scales of Infant Development, Third Edition (BSID-III). The BSID-III measures infant development across five domains: receptive language, expressive language, cognitive skills, fine motor and gross motor. One point was given for each successfully completed task, and the assessment continued until the infant failed five consecutive items. ## Statistical Analyses Statistical analyses were performed using the SPSS statistical software package for Windows (version 20.0; SPSS Inc., Chicago, IL, USA). Normality of the data was assessed using the Kolmogorov-Smirnov test. Plasma free choline, betaine and dimethylglycine were normalized with single natural log transformations. The potential associations between the measures of maternal methyl status (choline, betaine or dimethylglycine) at 16 or 36 weeks of gestation and the BSID-III infant’s cognitive, language and motor developmental raw test scores were assessed using multivariate regression. Potential confounders included in the model were maternal age, maternal IQ measured with the TONI 3, maternal ethnicity, maternal red blood cell DHA status at 16 and 36 weeks of gestation, infant breast feeding duration and infant sex. Maternal omega-3 fatty acids in pregnancy have been linked to higher scores on test of child development, and dietary intakes of DHA and choline are positively correlated. For this reason, we included the biochemical measure of maternal DHA status at both 16 and 36 weeks of gestation as covariates in the analyses. The infant characteristics of gestation length, single birth or low birth weight were not included as these were controlled for by the inclusion criteria. The present report focuses on results at 16 weeks of gestation since no associations were found between maternal plasma free choline and its metabolites at 36 weeks of gestation and infant development. Spearman correlation coefficients were used to determine the relationships among the maternal plasma free choline, betaine, dimethylglycine, methionine, homocysteine, cysteine, folate, tB12 and holoTC. All p values are based on two-sided tests, with a p\<0.05 considered statistically significant. # Results ## Associations among Methyl Metabolites and B Vitamins at 16 Weeks of Gestation The study population was predominantly white women (72%), with 15% of Asian background. Information on smoking and alcohol consumption were collected by self-report. Only six women reported smoking at any time during pregnancy; one stopped at 6 weeks of gestation and the remaining five women smoked \<1 pack cigarettes/week. Alcohol consumption was reported at some time after conception by 47 of the women, none of whom reported more than one drink per week. All of the women reported that they had taken prenatal vitamin and mineral supplements, and none had a plasma folate \<6.8 nmol/L. Of the 154 women, 14 had a plasma tB12 below the lower limit of normal of 148 pmol/L, and two had a holoTC below the lower limit of normal of 35 pmol/L. HoloTC is the biologically active form of B12 and is the only form of B12 taken up and utilized by cells. All the analyses relating to infant developmental outcome were repeated excluding results for the two infants of mothers with a holoTC suggestive of vitamin B12 insufficiency, and no differences to the outcomes were found. The maternal plasma free choline showed a median of 6.70 µmol/L, with an interquartile range of 5.78 to 8.03 µmol/L at 16 weeks of gestation. The mean ±SD for the estimated intakes of total choline and betaine were 383±98.6 mg/day and 142±70.2 mg/day, respectively, with a median (interquartile range) intake of 378 (307–457) mg/day for choline and 130 (89.9–178) mg/day for betaine. Dietary choline intake was positively correlated with the maternal plasma free choline concentration (r = 0.200, p = 0.013) at 16 weeks of gestation. There was a significant positive correlation between plasma choline and betaine, between choline and dimethylglycine, and between betaine and dimethylglycine (p\<0.001). Plasma methionine was also significantly and positively correlated with plasma betaine, dimethylglycine, homocysteine and cysteine, and inversely associated with tB12 and holoTC. The plasma total B12 and holoTC, but not folate, were also inversely associated with plasma homocysteine. The maternal plasma free choline, betaine and dimethylglycine at 16 weeks of gestation was significantly correlated with the same plasma measure at 36 weeks of gestation, r = 0.322, p\<0.0001, r = 0.433, p\<0.0001 and r = 0.524, p\<0.0001, respectively. There was no significant association between the maternal plasma free choline and measures of DHA at 16 or 36 weeks of gestation, and no difference in the maternal plasma free choline, betaine or dimethylglycine among 36 week gestation women taking or not taking supplemental DHA (maternal plasma free choline: 9.75±2.34 and 9.92±2.20 µmol/L, p = 0.64; betaine: 13.0±2.61 and 13.6±2.87 µmol/L, p = 0.20; dimethylglycine: 1.37±0.57 and 1.38±0.44 µmol/L, p = 0.90 for 36 week gestation women not taking or taking supplemental DHA, respectively). ## Association between Maternal Methyl Status and Infant Developmental Outcome As defined by the inclusion criteria for follow-up, all of the infants were born after full-term gestation and all were single birth infants. At 6 months of age, 72% of the infants were still being breast-fed. The bivariate regression analysis to address the strength of the relationship between the maternal plasma variables and the infants’ development test scores, with no consideration of confounding variables, revealed significant correlations between the maternal plasma free choline (B = 4.589, SE = 1.932, p = 0.019) and betaine (B = 6.366, SE = 1.723, p = 0.0003), and a strong trend for dimethylglycine (B = 2.134, SE = 1.095, p = 0.053) at 16 weeks of gestation and the infants’ cognitive developmental scores. Scatter plots showing the bivariate associations between infant cognitive test scores and the natural log-transformed values for maternal plasma free choline (r = 0.190, p = 0.019), betaine (r = 0.288, p = 0.0003) and dimethylglycine (r = 0.157, p = 0.053) are shown in. The full regression model adjusted for all variables, including the measures of maternal IQ, showed a significant positive association between the maternal plasma free choline (B = 6.054, SE = 2.283, p = 0.009) and betaine (B = 7.350, SE = 1.933, p = 0.0002) at 16 weeks of gestation and infant cognitive developmental score at 18 months of age. Maternal IQ was used as a proxy for family income and maternal education in the final analyses; initial analyses using the latter two variables did not change the results. A trend between maternal dimethylglycine status and infant cognitive score remained (B = 2.169, SE = 1.129, p = 0.078). The adjusted multivariate regression analysis also showed a strong trend between the maternal plasma free choline (B = 2.855, SE = 1.472, p = 0.055) and betaine (B = 2.495, SE = 1.271, p = 0.052) at 16 weeks of gestation and infant gross motor development. Using the adjusted model, each 1 µmol/L increase in maternal plasma free choline, betaine and dimethylglycine at 16 weeks of gestation corresponded to an increase of 2.23, 2.70 and 0.80 in infant cognitive test score, respectively. There was no significant correlation between the maternal plasma homocysteine, methionine, cysteine, folate, tB12 or holoTC at 16 weeks of gestation, and no significant correlations between any of the maternal plasma measures at 36 weeks of gestation and infant developmental outcome. # Discussion This study addressed the importance of choline and its metabolites betaine and dimethylglycine early in the second trimester of pregnancy on measures of child cognitive, language and motor skill development at 18 months of age. The study was confined to mother-child pairs involving only single birth infants born after full-term gestation. The findings provide evidence for an association between the mothers’ methyl status, specifically choline (p = 0.009) and betaine (p = 0.0002) in gestation and child cognitive test scores at 18 months of age, with a strong trend towards a positive association between maternal plasma free choline (p = 0.055) and betaine (p = 0.052) at 16 weeks of gestation and the infants’ gross motor development (adjusted analyses). Studies in rats and mice have emphasized a sensitive window of maternal choline deprivation occurring between gestation days 11 and 17 which lead to morphological and molecular changes in the embryonic brain. This time period in rodents corresponds to gestation beginning about 1.5 to about 3 months of gestation in human pregnancy, with the present study conducted at about 4 months of gestation. These results appear to be consistent with a crucial role of methylation, potentially involving synthesis of important methylated metabolites and intermediates, especially thymidylate and purines, as well as epigenetic mechanisms, such as DNA and histone methylation in early development. This suggestion is supported by the strong significant, positive associations between the maternal plasma free choline and betaine, choline and dimethylglycine, and choline and methionine, but not homocysteine at 16 weeks of gestation, all consistent with the importance of betaine-driven remethylation to maintain methionine for important methylation reactions. However, acetylcholine also plays a crucial role in brain development through its role in the cholinergic system. Large amounts of choline are also needed to support new membrane synthesis associated with cell division and growth. Thus, the importance of choline in early brain development may be multi-factorial. The setting of this study is on background in which the food supply has been fortified with 0.15 mg folic acid per 100 g of cornmeal or flour since 1998. In addition, all of the women in the present study reported taking prenatal multivitamin supplements, which typically contain 400 µg folic acid, variable amounts of vitamin B12, but no choline or betaine. The plasma free choline indicative of deficiency has not been defined. However, studies involving feeding a choline-deficient diet to men and post-menopausal women for 6 weeks showed a decline in plasma free choline from 9.8 to 6.8 µmol/L. A substantial 56% of the women in our study had a plasma free choline concentration below 7.0 µmol/L at 16 weeks of gestation. Furthermore, 74% consumed less than 450 mg/day choline which is recommended as the adequate intake of choline for pregnant women. While dietary choline intakes appear to be low in our population, other factors may influence choline status. These include several single nucleotide polymorphisms (SNP), including SNP in phosphatidylethanolamine *N*-methyltransferase (*PEMT*), choline dehydrogenase (*CHDH*) and methylenetetrahydrofolate dehydrogenase (*MTHFD1*), all of which may influence choline metabolism and increase sensitivity to inadequate dietary choline intakes. Our results contradict the null findings of an association between plasma free choline in pregnancy and child IQ reported by Signore et al. in the U.S. In addition to differences in the maternal and child study populations, setting, and age of child cognitive assessment, the women studied by Signore et al. had a mean plasma free choline of 9.34 µmol/L (interquartile range 7.69−11.50 µmol/L) at 16−18 weeks of gestation, with no change in plasma free choline during gestation. Women in our population appear to have a much lower choline status, with a median plasma free choline of 6.70 µmol/L at 16 weeks of gestation, which increased to a median of 9.40 µmol/L by 36 weeks of gestation. The present study also involved predominately White or Asian women, and assessed only single birth full-term gestation infants. Signore et al., on the other hand, studied women in Alabama of whom 70% were Black, and included both small for gestational age and premature infants. Relatively little, and inconsistent information is available on choline, betaine and dimethylglycine in pregnant women. Some populations have been reported to show a low plasma free choline in early gestation which increases with increasing gestation as in our studies, but others show high plasma free choline in early gestation with no change throughout gestation. Similar to our population, studies in the island of Curaçao (formerly Dutch Antilles) found a mean plasma free choline of 7.32 µmol/L at 16 weeks of gestation which increased to 10.77 µmol/L by 36 weeks of gestation. Studies in Turkey, however, reported much higher plasma free choline concentrations of 14.5 and 16.5 µmol/L at 16 to 20 and 36 to 40 weeks of gestation, respectively. The plasma free choline of 9.34 µmol/L among U.S. women in Alabama falls between the low plasma free choline concentrations in the present study of Canadian women, and women in Curaçao, and the higher plasma choline concentrations of women in Turkey. It would seem important to understand the extent to which diet, genetic or other sources of variability including potential differences among laboratories in methodology to quantify plasma choline may contribute to difference in choline status among pregnant women in different countries. In conclusion, this was an observational study that has shown dietary intakes of total choline are below current recommended intakes levels in a large (74%) proportion of pregnant women. This low intake of choline is accompanied by low plasma concentrations of free choline, lower than reported for pregnant women in other regions of the world. Plasma free choline assessed early in the second trimester of gestation was positively associated with betaine, dimethylglycine and methionine, indicating that choline is further metabolized and contributes methyl groups for regeneration of methionine, the ultimate source of methyl groups for numerous biologically important methylations. The maternal plasma free choline and betaine at 16 weeks of gestation, but not 36 weeks of gestation, were positively associated with infant cognitive development in both unadjusted and adjusted analyses. ## Limitations This study is an observational study and no causative relationships can be drawn from our result. The estimation of dietary choline should be viewed with caution since information on choline in the Canadian food supply is unavailable, and intakes were estimated using the USDA data base on the choline content of foods. This study cohort was mostly white, educated women who breastfed their infants for six months or longer. Results from this study may not be representative of other groups of women in Canada or in other countries. The extent to which genetic variations in pathways relevant to choline metabolism impact maternal plasma free choline or betaine, their transfer to the developing infant, or maternal dietary choline need is also unknown. Although further studies to elucidate the importance of choline or its metabolite betaine in human development are needed, this study provides novel evidence that maternal methyl nutrition may play a role in early human brain development, consistent with evidence for other species. We would like to thank our research staff for their assistance in all aspects of this study. We extend our thanks to Ruth Milner in the Child and Family Research Institute Biostatistical Support Unit for advice and guidance with the statistical analysis. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SMI. Performed the experiments: RAD DJK. Analyzed the data: BTW KJR. Wrote the paper: SMI BTW. Performed the laboratory analyses: RAD DJK. Collected and analyzed the subject data: KJR. Analyzed the dietary data: BTW. Analyzed the data and wrote the paper: BTW SMI.

# Introduction Protein misfolding and aggregation, processes involved in several neurodegenerative diseases, are likely preceded by conformational changes in the proteins involved. The transient nature and the small scale of these conformational changes have made them extremely difficult to study directly. Recent studies have shown that natively unfolded molecules can partially fold and form, in vitro, either toxic oligomeric species or microscopic fibrillar aggregates, which are neurotoxic. How, why, and when misfolding happens in vivo is still unclear. α-Synuclein (aSyn), a small (140 amino acid) neuronal protein of unknown function which is ubiquitously expressed in the brain, displays little secondary structure in vitro and belongs to a group of proteins known as ‘natively unfolded’. Under certain conditions, aSyn can adopt specific conformations in association with model lipids or in the presence of detergents. In PD, there is substantial loss of dopaminergic neurons in the substantia nigra, with the presence of fibrillar inclusions called Lewy bodies (LBs) comprising aSyn as a major constituent. Diseases associated with the accumulation of fibrillar forms of aSyn are commonly known as synucleinopathies. The preferential vulnerability of dopaminergic neurons in PD is unclear, but a link between dopamine biology and aSyn as been hypothesized, since dopamine was shown to form adducts with aSyn in the test tube, appears to stabilize protofibrillar forms of aSyn, and inhibits aSyn fibril formation in vitro. Recently, dopamine-modified aSyn was shown to block chaperone mediated autophagy, but the full spectrum of effects of this dopamine interaction with aSyn in living cells is still obscure. One possibility is that this is part of the normal function of aSyn, but it could also bear a connection with the increased vulnerability of dopaminergic neurons. To explore this question further, we developed a method that specifically detects aSyn conformational alterations within cells, using a highly sensitive and specific assay of molecular proximity called fluorescence lifetime imaging microscopy (FLIM). Here, we applied FLIM to investigate the effect of dopamine and other chemical modulators of neuronal activity on the conformation of aSyn in primary neurons. A deeper understanding of the connection between aSyn and dopamine has implications for current and future PD therapeutic interventions. # Materials and Methods ## Plasmid construction The constructs for human wild type untagged aSyn have been described previously. Briefly, cDNA encoding the genes were cloned into pcDNA3.1 (Invitrogen, CA, USA) expression vectors. N- and/or C-terminal Myc or V5 tags were generated using annealed oligomers coding for Myc or V5, and subcloned into wild type aSyn expressing pcDNA3.1 plasmid. ## Cell culture, transfection, and immunocytochemistry Human H4 neuroglioma cells (HTB-148 - ATCC, Manassas, VA, USA) were maintained in OPTI-MEM (Invitrogen, CA, USA) supplemented with 10% fetal bovine serum. H4 cells were passaged 24 hours prior to transfection and plated in four-well chamber slides for immunocytochemistry (Labtek, Nalgen-Nunc, Naperville, IL, USA). Cells were transfected with equimolar ratios of plasmids using Superfect (Qiagen, Chatsworth, CA, USA) according to the manufacturer's instructions. After 24 hours cells were washed with phosphate buffered saline (PBS), and fixed with 4% paraformaldehyde for 10 min at room temperature (RT). After washing with PBS cells were permeabilized in tris buffered saline (TBS) containing 0.1% Triton X-100 for 20 min at RT. After blocking in 1.5% normal goat serum containing TBS for 1 hour cells were incubated with primary antibody for 2 hours at RT or overnight at 4°C (mouse anti-Myc 1∶1000, Abcam, Cambridge, MA, USA; rabbit anti-V5 1∶3000, AB9116, Abcam, Cambridge, MA, USA) followed by washing with PBS and secondary antibody incubation for 1 hour (goat anti-rat IgG- Alexa488, 1∶300, Molecular Probes, Eugene, OR, USA; goat anti-rabbit IgG-Cy3 1∶500, Jackson Immunoresearch, PA, USA). After a final wash, slides were mounted with aqueous mounting solution (GVA, Zymed, San Francisco, CA, USA) and subjected to fluorescence microscopy and fluorescence lifetime imaging microscopy (FLIM). HEK cells were maintained in DMEM with 10% FBS and handled as described above for the H4 cells. MES23.5 cells, received with permission from Dr. Stanley Appel were maintained as described and transfected using Lipofectamine 2000 according to the manufacturer's instructions. ## SDS-PAGE and immunoblotting 24 hours after transfection, H4 cells were washed with cold PBS, harvested by scraping in cold lysis buffer without detergents (Tris/HCl 50 mM pH 7.4, NaCl 175 mM, EDTA 5 mM pH 8.0, protease inhibitor cocktail, Roche, Basel, CH) and sonicated for 10 seconds. Lysates were cleared from debris by a 9,500 g centrifugation for 10 min at 4°C and were then subjected to SDS-PAGE using 10–20% gradient Tris–Glycine gels (Novex, San Diego, CA, USA) for Western blot analysis. Protein bands on the SDS-PAGE were transferred to Immobilon-P membrane (Millipore, Bedford, MA, USA) and blocked in blocking buffer (Lycor, Lincoln, NE, USA) for 1 hour prior to the addition of the primary antibody at room temperature for 1–2 hours or overnight at 4°C. The blots were washed three times in TBS with 0.2% Tween (TBS-T, pH 7.4), and were incubated at room temperature for 1 hour in fluorescent labeled secondary antibodies (IRDye 800 anti-rabbit or anti-mouse, Rockland Immunochemicals, Gilbertsville, PA, USA 1∶3000 or Alexa-680 anti-rabbit or anti-mouse, Molecular Probes, Eugene, OR, USA 1∶3000). After washing three times in TBS-T immunoblots were analyzed and quantified using the Odyssey infrared imaging system (Lycor, Lincoln, NE, USA). ## Fluorescence lifetime imaging microscopy and calculation FLIM has been recently described as a technique for the analysis of protein proximity. The technique is based on the observation that fluorescence lifetimes of a donor fluorophore shorten if it is in close proximity (\<10 nm) to a FRET acceptor. The decrease in lifetime is proportional to the distance between the fluorophores at R⁶. A mode-locked Ti-sapphire laser (Spectra-Physics, Fremont, California) emits a femtosecond pulse every 12 nanoseconds to excite the fluorophore. A high-speed Hamamatsu (Bridgewater, New Jersey) detector and hardware/software (SPC-830 Becker and Hickl, Berlin, Germany) were used to measure fluorescence lifetimes on a pixel-by-pixel basis. Donor fluorophore (Alexa488) lifetimes were fit to two-exponential decay. One component was fixed to the expected lifetime of Alexa488 without an acceptor (Cy3) in close proximity for energy transfer (negative control – monofit) that was determined by fitting to one-exponential decay curve (mean lifetime monofit). To validate the two component fit procedure, the same cells from the negative control mono- fit were subjected to two exponential decay curve fitting and revealed the experimental value for Alexa488 lifetime that did not differ from the mono-fit lifetime and was used in the experiment as calculated negative control. As a positive control, Alexa 488 lifetime was measured in the presence of a FRET acceptor (Cy3) in close proximity presenting the acceptor with a donkey anti- goat Cy3 labeled Ab, directed against the goat anti-mouse Alexa 488 secondary Ab used to visualize the anti-V5 monoclonal antibody. All combinations of Myc-, V5-, or un-tagged aSyn molecules were stained using the same antibody combination as described above. As a positive control, Alexa488 lifetime was measured in the presence of a FRET acceptor (Cy3) in close proximity presenting the acceptor with a donkey anti-goat Cy3 labeled antibody, directed against the goat anti-mouse Alexa488 secondary antibody used to visualize the anti-V5 monoclonal antibody. Experiments were performed in triplicate using the number of cells indicated. are expressed as mean fluorescence lifetime ± SEM. ## Detergent-Solubility Franctionation and Gel Electrophoresis Detergent solubility was performed by adding Triton X-100 to total cell lysates (final concentration 1%) and incubating for 30 min on ice followed by centrifugation (15,000×*g*, 60 min, 4°C). The supernatant was designated Triton X-100 soluble fraction, and the pellet was redissolved in 2% SDS-containing lysis buffer and sonicated for 10 s (Triton X-100 insoluble fraction). Additional washing of the Triton X-100 insoluble pellet was found to not alter the aSyn content in this fraction (data not shown) and was omitted from the experiments. Protein concentration was determined using a BCA (Pierce, IL, USA) protein assay. 20–40 µg of each cell lysate was loaded onto 4–20 or 10–20% gradient Tris/glycine gels (Invitrogen, CA, USA) for Western blot analysis. SDS- PAGE was performed with SDS containing commercially available standard running and sample loading buffers (Invitrogen, CA, USA). Protein was transferred to Immobilon-P membrane (Millipore, Bedford, MA) and blocked in blocking buffer (Lycor, Lincoln, NE) for 1 hour prior to the addition of primary antibody, anti- aSyn (Syn-1, 1∶1000, BD Transduction Laboratories) at room temperature for 1–2 hours or overnight at 4°C. Following three Tris-buffered saline with Tween 20 washes, infrared fluorescent-labeled secondary antibodies (IRDye 800 anti-rabbit or anti-mouse, Rockland Immunochemicals, Gilbertsville, PA, at 1∶3000 or Alexa-680 anti-rabbit or anti-mouse, Molecular Probes, Eugene, OR at 1∶3000) were incubated at room temperature for 1 hour and immunoblots were processed and quantified using the Odyssey infrared-imaging system (Lycor). Blots were also probed for actinin (anti-actinin, Sigma). ## Primary Cortical Neuronal Cultures CD-1 mice and Sasco Sprague Dawley rats were obtained from Charles River laboratories. A cesarean section was performed and E14 mice or E18 rats were removed. The animals were decapitated and their cortices were removed and placed in Phosphate Buffered Saline (PBS). The tissue was manually triturated in 10% Fetal Bovine Serum (FBS) from, and Neurobasal medium (Invitrogen, CA, USA) and plated onto 4 well chamber slides (Lab tek) or 100 mm tissue culture dishes (Corning). The slides or dishes were coated with Poly-D-Lysine (Sigma) 24 hours prior to the dissection and incubated with Human Placenta laminin (Sigma) and Penicillin Streptomycin (Invitrogen, CA, USA) in Neurobasal Medium overnight at 37°C. Cells were plated in 10% fetal bovine serum (Invitrogen, CA, USA) in Neurobasal medium. One hour later, the media was removed and replaced with B-27 and Neurobasal medium. The cells were maintained and fed in the same media every 5–7 days depending on their density. ## Transfection of Primary Cortical Neurons Cortical neurons were plated on 4 well glass bottom chamber slides (Nunc). Cells were maintained in Neurobasal media (Invitrogen, CA, USA) containing B-27 (Invitrogen, CA, USA) and penicillin streptomycin (Invitrogen, CA, USA). Between 5 and 7 days in vitro (DIV) cells were transiently transfected using Lipofectamine 2000 (Invitrogen). A concentration of 2 µg DNA/ 5 µl of Lipofectamine 2000 was used per each well of the chamber slide. The DNA and Lipofectamine 2000 were added into DMEM (Invitrogen, CA, USA) and incubated separately for 5–15 minutes before being combined. The DNA complex was gently mixed and incubated for 45 minutes at room temperature. The neuronal maintenance media was removed from the cells and they were washed with phosphate buffered saline (PBS) containing no calcium or magnesium. The DNA complex was added to the cells for 2–6 hours at 37°C. The DNA complex was then removed and replaced with neuronal maintenance media. Cells were then fixed and processed for immunocytochemistry. ## Expression and Purification of human wt aSyn The expression and purification procedure of human WT aSyn was a modified version of a previously described method. Briefly, cells of *E. coli* strain BL-21 (GE Healthcare, NJ, USA) were transformed with the appropriate expression vector, and expression was induced by the addition of isopropyl _D-thiogalactopyranoside at a final concentration of 1 mM. Cells were harvested, resuspended in 50 mM Tris (pH 8.5), 50 mM KCl, 5 mM MgAc, 0.1% NaN₃ and 300 µM PMSF, and lysed by three passages through a French cell press. The extract was centrifuged at 18000 g at 4°C for 30 min to eliminate cell debris. The supernatant was saved and boiled for 20 min. The boiled extract was centrifuged at 45000 g at 4°C for 45 min and the supernatant was filtered with a 0.2 µm filter to remove possible pellet contamination. The aSyn containing extract was loaded on to an ion-exchange chromatography Q Sepharose™ (GE Healthcare, NJ, USA) fast flow column equilibrated with 20 mM Tris/HCl (pH 8.0). Proteins were eluted with a linear NaCl gradient (0.12–0.5 M) at a flow rate of 1.5 ml.min⁻¹ and the eluate was monitored at 280 nm. Protein-containing fractions were collected and probed by western blot analysis using Syn-1 anti-aSyn antibody (BD Transduction Laboratories, CA, USA). Fractions containing aSyn were collected, concentrated by centrifugation using Amicon filters (Millipore) and applied to a gel filtration Superdex 75 column (GE Healthcare, NJ, USA), equilibrated with 50 mM Tris/HCl buffer (pH 7.5) containing 150 mM NaCl. Proteins were eluted with the same buffer at a flow rate of 1 ml.min⁻¹. Again, fractions containing aSyn, probed by western blot, were collected and combined for dialysis against water and then lyophilized for future analysis. ## In vitro modification of purified α-syn by dopamine Purified native aSyn (70 µM) was incubated with dopamine (Sigma) at a final concentration of 1, 10, 100 and 1000 µM in 50 mM sodium phosphate buffer (pH 7.4) at 37°C for 1 week in sterile conditions. aSyn concentration was determined spectrophotometrically (ε₂₇₅ = 5974 M⁻¹.cm⁻¹) in a UV-Visible Jasco V-530 spectrometer. ## Far-UV circular dichroism (CD) spectroscopy Secondary structure analysis was performed by far-UV (185–260 nm) CD in a Jasco J810 spectropolarimeter at 37°C (Julabo F25 temperature control unit) with a 0.01 cm path length. CD spectra were deconvoluted using CDSSTR algorithm on Dichroweb (<http://dichroweb.cryst.bbk.ac.uk/html/home.shtml>). All spectra were solvent baseline-corrected. # Results We have previously shown that conformational changes in aSyn can be monitored in immortalized H4 cells using the sensitive fluorescence resonance energy transfer (FRET) based proximity assay, FLIM. To directly study the interaction between the amino-terminus and the carboxyl terminus of aSyn in neurons we overexpressed doubly-tagged Myc-aSyn-V5 in both mouse and rat primary neuronal cultures. Neurons were transfected at 5–7 days in vitro (DIV) using Lipofectamine 2000 and transfection efficiencies in the order of 5% were achieved. Immunostaining using antibodies against Myc and V5 confirmed that both epitope tags were expressed and completely colocalized at the subcellular level. Next, we used FLIM to examine the association between the N- and C-termini of aSyn in neurons. The N-terminus was labeled with the donor fluorophore, Alexa488, and the C-terminus was labeled with the acceptor molecule, Cy3. When we examined the lifetime of the donor fluorophore we detected a striking range of lifetimes throughout the transfected neurons with significantly different lifetime being detected in the nucleus/cell body and throughout the neurites, as demonstrated by the differences in color coding throughout the neurons. These data indicate that aSyn adopts different conformations in specific subcellular environments. Interestingly, the donor fluorophore lifetime was consistently longer in the cell body/nucleus (∼1900 psec) than in the processes (∼1000 psec) suggesting that aSyn adopts a folded conformation in the processes because shortening of the lifetime corresponds to the fluorophores being in closer proximity to one another. Control experiments where the C-terminus of aSyn (V5) was labeled with the donor fluorophore, Alexa488, and the N-terminus (Myc) was labeled with Cy3, yielded similar results (data not shown) and all subsequent experiments were performed using the conditions described above. In order to assess whether the different lifetimes were indicative of conformational changes (intramolecular interactions) or indicative of interactions between distinct aSyn molecules (intermolecular interactions), we co-transfected neurons with the epitope tagged aSyn along with untagged WT aSyn. In this situation, we observed that the lifetimes did not change, when compared to those observed for the tagged aSyn, indicating the majority of the interactions we detected were intramolecular, i.e., due to conformational changes in aSyn. Given the data suggesting that, in vitro, dopamine can impact aSyn conformation, we next asked if exposing aSyn transfected neurons to dopamine would affect aSyn conformation. Primary neurons overexpressing aSyn were treated with 100 µM dopamine (DA) for 10 min. DA significantly decreased the lifetime of the donor fluorophore to ∼650 ps (n = 3 independent experiments, total of 57 cells, p\<0.01) at the dose tested, indicating that DA induces the N- and C-termini of aSyn to be in closer proximity, reflecting a change in conformation. By contrast, various other treatments, including treating cells with 60 mM KCl for 10 minutes led to no changes in donor lifetime. To tease apart the molecular mechanisms of the observed DA effect we examined the effect of DA agonists and antagonists on aSyn conformation. Primary neuronal cultures were immunostained using antibodies against D1 and D2 receptors to verify the presence of these receptors in our cultured neurons. We then screened a panel of well- characterized DA agonists and antagonists to determine if the previously observed conformational changes in aSyn could be mimicked or impeded. Surprisingly, we were unable to detect an effect on aSyn conformation with any of the DA agonists or antagonists. To further determine if dopamine induced a general effect on the neurons or whether it was required to enter the cells in order to induce the conformational change in aSyn, we used nomifensine (100 µM) to block the activity of the dopamine transporter. Interestingly, we detected a reduction in the dopamine- induced conformational change in aSyn, suggesting that dopamine must gain entry into the cells in order to exert its activity on aSyn. To further investigate how the conformational changes in aSyn affect its biochemical properties, we sought to recapitulate the modulation of conformation in immortalized cell lines, which would enable us to achieve higher transfection efficiencies required for biochemical studies. First, we transfected three cell lines of different origins (H4, MES23.5, and HEK) with the Myc-aSyn-V5 construct. These cells were then treated with dopamine and processed for FLIM. Interestingly, we found that the in the cell lines of neuronal origin (H4 and MES23.5) dopamine induced a conformational change similar to that observed in primary neuronal cultures. In limited experiments using HEK cells, we did not observe such a conformational change. We previously identified, in brain tissue derived from patients with dementia with Lewy bodies (DLB), in aSyn transgenic mice, and in aSyn H4 expressing cells, oligomeric aSyn species which are detergent-insoluble. Here, we hypothesized that the DA-induced conformational changes in aSyn might affect its detergent solubility. To investigate whether DA affected aSyn oligomerization we used native polyacrylamide gel electrophoresis (PAGE). In cells treated with DA, we observed a ∼50% decrease in a ∼250 KDa band and a ∼25% increase in two ∼75 KDa bands, demonstrating that DA induces slight but detectable changes in aSyn oligomerization (arrows). To further assess the effect of DA on aSyn solubility we performed a triton X-100 detergent fractionation of H4 cell lysates expressing Myc-aSyn-V5 treated or untreated with DA. We found no significant differences in the solubility of aSyn in either groups of cells. In order to determine whether DA induces changes at the level of the secondary structure of aSyn we used circular dichroism (CD). Spectra were taken at 37°C in the absence or in the presence of different DA concentrations (10, 100 and 1000 µM of DA, corresponding to ratios of DA:aSyn of 0.14; 1.4 and 14.0) for a fixed concentration of aSyn of 70 µM. The insets show small, but significant changes (values represent the average of 5 different experiments +/− SD) at the level of the secondary structure of aSyn, in agreement the conformational changes observed in neuronal cells, where the distance between the N- and C-termini of aSyn is modified by DA treatment. # Discussion In PD, cell death affects primarily the dopaminergic neurons of the substantia nigra, but the nature of this selective vulnerability is still unclear. A common pathway, involving DA-dependent oxidative stress, has been put forward to explain the death of dopamine neurons. Defects in the sequestration of dopamine into synaptic vesicles in dopaminergic neurons from the substantia nigra, enabling undesired DA-aSyn interactions, may explain their increased vulnerability. It has been reported that DA can undergo auto-oxidation and form DA-quinone adducts with aSyn which prevent aSyn fibrillization and lead to the accumulation of toxic intermediates, but the relevance of these findings in the context of living cells has been difficult to determine. In the current study we sought to investigate whether DA influences the conformation of aSyn in primary neurons in culture. We used FLIM to study alterations in the conformation of aSyn by monitoring the interactions between the N- and C-termini of the protein, as we had previously reported for studies in mammalian cell lines. Here we demonstrate that aSyn adopts different conformations throughout the axon and dendrites. In vitro, purified aSyn does not display any secondary structure, and is considered a natively unfolded protein, but it is highly likely that it adopts specific conformations inside neurons in order to perform its normal function(s). Our data show that different subcellular microenvironments, with potentially different redox conditions, lipid compositions, or other conditions known to influence the behavior of aSyn in vitro, afford aSyn the possibility of adopting distinct conformations inside living cells. For example, lipid rafts mediate the synaptic localization of aSyn in neurons, which may also explain the selective distribution of aSyn. Interestingly, we were not able to identify a specific association of a particular aSyn conformation with any subcellular organelle, suggesting local microenvironments may be more important in determining the structure/function of the protein. The interaction of aSyn with synaptic vesicles is highly dynamic, which may also explain the variety of aSyn conformations detected throughout the axons and dendrites. Our data also demonstrate that aSyn changes its structure in response to DA, or possibly dopamine oxidation by-products, adopting a conformation where its N- and C-termini become closer together. DA or DA by-products inhibit aSyn fibril formation, which may, in turn, lead to the accumulation of aSyn oligomeric species via an alternative folding pathway. Although our results do not show whether the DA-induced change in conformation of aSyn is the precursor for the formation of toxic oligomeric species, our data support the model that DA- induced conformational changes in aSyn, either through a direct covalent interaction or indirectly, may favor changes in the oligomerization state of the protein which may explain the increased vulnerability of dopaminergic neurons in comparison to others. These conformational changes may underlie the recently reported effect of dopamine-modified aSyn on autophagy mediated degradation of the protein, and its subsequent impact on misfolded protein degradation in cells. Defining the role of the identified aSyn conformations will shed light into the pathogenic mechanisms involved in PD, and may pave the way for the identification of novel targets for therapeutic intervention in different synucleinopathies. [^1]: Conceived and designed the experiments: TFO AQ PJM BTH. Performed the experiments: TFO JK KB JT PP LMAO. Analyzed the data: TFO. Wrote the paper: TFO PJM BTH. [^2]: The authors have declared that no competing interests exist.

# Introduction Preterm birth is defined as delivery before the completion of the 37th week of gestation and affects 13 percent (542 893 births in 2006) of live births in the United States (<http://www.cdc.gov/nchs/fastats/birthwt.htm>). Approximately 40 percent of preterm births occur after the spontaneous onset of preterm labor. Long-term neonatal sequelae of prematurity such as bronchopulmonary dysplasia, grade III/IV intraventricular hemorrhage and retinopathy of prematurity determine the overall quality of life for the child and the family. Terbutaline sulfate has been used off-label in selected patients as a maintenance therapy to inhibit uterine contractions for extended periods of time following primary tocolysis with first-line agents. Terbutaline, a β-sympathomimetic drug, acts to relax smooth muscle in the bronchial tree, blood vessels and myometrium. Maternal side effects are common, and can include serious adverse reactions such as pulmonary edema, myocardial ischemia, cardiac arrhythmias, hypotension, and metabolic alterations. Despite previous reviews which questioned the effectiveness and safety of subcutaneous terbutaline infusion, the use of such therapy is not uncommon in the United States. The exact frequency of use of subcutaneous terbutaline infusion for the prevention of preterm birth is not known. This review, commissioned by the Agency for Healthcare Research and Quality through its established stakeholder topic nomination process, aims to systematically review and meta-analyze the evidence examining the efficacy, effectiveness, and harms of SQ terbutaline pump for preventing preterm labor, compared with placebo, conservative treatment, or any other active intervention. We investigated the clinical effectiveness and harms of pump therapy by systematically retrieving, appraising and synthesizing evidence on neonatal health outcomes and outcomes of maternal and neonatal harm. Surrogate outcomes, such as birthweight and prolongation of pregnancy were also examined. The potential confounding effects of maternal activity and maternal care on the above endpoints were explored, as was the incidence of pump-related outcomes. Particularly in light of a recent “black box” warning issued by the FDA, this review provides a contemporary and definitive summary of the available literature on the benefits and harms of terbutaline maintenance tocolysis. # Methods We followed a pre-specified and peer-reviewed study protocol. The full evidence report, including search strategies and a detailed list of *a priori* outcomes, risk of bias assessment and detailed evidence tables are available at [www.effec tivehealthcare.ahrq.gov/reports/final.cfm](http://www.effectivehealthcare.ahrq.g ov/reports/final.cfm). ## Ethics Ethics approval was not required for this review, as there is no potential for individual patient identification. ## Searching We searched Ovid MEDLINE® In-Process & Other Non-Indexed Citations and Ovid MEDLINE® (1950 to April 1 2011); OVID EMBASE (1980 to April 1 2011); CINAHL via EBSCOhost (1985 to December 7, 2009), the Cochrane Library via the Wiley interface (April 1, 2011) (including CENTRAL, Cochrane Database of Systematic Reviews, DARE, HTA, and NHS EED), and the Centre for Reviews and Dissemination (CRD) databases (January 2, 2010). We hand-searched the bibliographies and text of review articles, letters to editors, and commentaries and the reference lists of included studies for additional references. We also reviewed grey literature sources and information received from pharmaceutical companies. Finally, we reviewed the FDA summary of post-marketing data to assess risk of maternal harm. ## Selection Two reviewers screened abstracts and full-text reports with conflicts resolved by consensus or third party adjudication. Studies were included if they met the following criteria: evaluated pregnant women between 24–36 weeks gestation having had acute preterm labor arrested with primary tocolytic therapy; included at least one group that was administered SQ terbutaline pump; and assessed one of the specified outcomes (primary neonatal outcomes, surrogate outcomes, maternal harms, neonatal harms, pump-related outcomes and long-term childhood outcomes). ## Validity Assessment For each study outcome, we assessed confounding and risk of selection, performance, detection and attrition bias. Selected items from the McMaster Quality Assessment Scale of Harms (<http://hiru.mcmaster.ca/epc/mcharm.pdf>) were also included. The overall risk of bias ratings were designated as high, medium, or low. Outcomes were rated as high risk of bias if there was a major flaw in the study. Separately, we evaluated the potential for financial conflict of interest. Following published guidance for the Effective Health Care Program two reviewers graded the strength of evidence for incidence of delivery at various gestational ages, mean prolongation of pregnancy, bronchopulmonary dysplasia, significant intraventricular hemorrhage (grade III/IV), neonatal death, death within initial hospitalization, and maternal withdrawal due to adverse effects. The guidance stipulates that the strength of evidence be rated as insufficient, low, moderate or high to reflect our confidence on the validity and reproducibility of evidence synthesis. According to the guidance, evidence was to be considered insufficient (i.e. inability to conclude) when there were no studies, studies showing opposite direction of effect, or when confidence intervals were wide enough to incorporate the possibilities of benefit, no difference or harms. Also, the strength of evidence based on observational studies conventionally starts off with a grade of low which is upgraded only when there is demonstration of a dose-response relationship, large effect size, or an effect despite confounding towards null. Generalisability or applicability of evidence was also rated as per previous guidance. ## Data Abstraction One reviewer extracted data into a standardized electronic form and assessed study risk of bias and applicability. Extraction items included general study characteristics (e.g. year of publication, study design), population characteristics (e.g. inclusion/exclusion criteria, age, race, level of activity), intervention characteristics (e.g. dose, duration, details about comparators, level of care), and outcomes with their estimates. A second reviewer verified outcomes data and study risk of bias assessments. Ratings for level of care, level of activity, and assessments of applicability were verified by a clinical expert. ## Study Characteristics Aside from case reports, all types of study designs were considered because evidence from experimental studies is often limited for reviews of comparative effectiveness. Non-comparative studies (i.e. case series) were assessed only for pump-related harms, such as incidence of pump failure, missed doses, or overdose and maternal harms. Non-English language records without an English abstract were excluded. We also excluded case reports, but in a *post hoc* decision sought FDA summaries of post-marketing data highlighting serious harms. Pump efficacy was examined for pre-specified subpopulations of women but harms were investigated across subgroups. The subgroups included women delivering extremely preterm (\<28+0 weeks), very preterm (28+0 to 31+6 weeks), preterm (32+0 to 33+6 weeks), and later preterm (34+0 and 36+6 weeks); with multiple gestation; of different racial subgroups; with previous preterm birth; with history of preeclampsia and; with and without recurrent preterm labor. ## Quantitative Data Synthesis We performed a meta-analysis of the RCTs with a random effects model when they were clinically and methodologically similar. To assess statistical heterogeneity and its magnitude, we used Cochran's Q (α = 0.10) and the I² statistic respectively. Odds ratios (ORs) were calculated for dichotomous outcomes and mean differences for continuous outcomes. Analyses were performed using Comprehensive Meta Analysis version 2.2.046 or version 2.2.055 (NJ, USA). We did not perform meta-analysis of the observational studies because of potential differences in confounders, nor did we combine studies of singleton and multiple pregnancies. Small number of included studies precluded meta- regression and exploration of heterogeneity in effect estimates. # Results ## Flow of Included Studies The flow of retrieved records through the phases of literature screening is detailed in. Fourteen independent studies and 1 companion article were included in the review. ## Study Characteristics presents general summary characteristics of the included studies. Most studies were observational, and included cohorts and case series. Two studies were RCTs and one was a nonrandomized trial. Sample sizes ranged from 9 to 1 366, but over 70 percent of studies included at least 200 subjects (average 291±395). Despite meta-analysis, evidence from the two small RCTs remained underpowered for differences in the outcomes of benefit and harms (total N = 94). All studies were from the United States and patients were recruited either from single center study sites or from a national proprietary database run by Matria Healthcare. This database provides an outpatient perinatal program consisting of 24-hour nursing and pharmacy support, home uterine activity monitoring, individualized education, and provision of tocolytic therapy to women with preterm labor. Because five studies originated in the Matria database, and not all reported geographic region and/or years over which participants were recruited, the question of overlap in patients across these studies was an important concern. Comparator groups included placebo, no treatment, oral terbutaline, oral nifedipine, and mixed oral tocolytics. The definition of labor was unclear in 36 percent of the included studies. The remaining studies included women with persistent contractions and cervical change. Parenteral magnesium sulfate was often the primary tocolytic to arrest acute preterm labor. In several studies, only women with two or more episodes of preterm labor (i.e. recurrent preterm labor) were eligible for inclusion,. Some studies were conducted exclusively in women with singleton gestation, while a few studies evaluated women with twins only. Some studies may have included women less than 24 weeks gestational age, but data for such participants could not be separated,. Maternal characteristics of participants in the included studies are summarized in. No studies presented data on concomitant medications, body mass index (BMI), history of preeclampsia, cervical position, cervical consistency, cervical station, Bishop's Score, or fetal fibronectin. ## Risk of Bias Assessment Studies with important group imbalances in baseline characteristics or prognostic factors were rated as high risk of bias. Those with no identifiable flaws but with incomplete reporting of information for risk assessment were judged to be of medium risk of bias. Although the randomization procedures in the two RCTs were appropriate, we rated one RCT as high risk of bias because more than 90 percent of eligible participants declined to participate, the study was underpowered, and blinding was ineffective. The two case series were judged to be of medium risk of bias because neither study provided clear definitions for the pump-related harm outcomes and several criteria such as compliance, adequacy of sample size, and selective outcome reporting, were unclear. ## Quantitative Data Synthesis ### Neonatal Health Outcomes One retrospective cohort with medium risk of bias, presented evidence of low strength suggesting SQ terbutaline pump may decrease the risk of neonatal death compared with oral tocolytics in women with recurrent preterm labor and twin gestation (OR = 0.09, 95 percent CI: 0.01, 0.70). Three retrospective cohort studies reported non-significant differences in rate of stillbirth in women with recurrent preterm labor and single or twin gestation. However, these studies were likely underpowered, given the small number of events (\<1%). Sparse evidence from underpowered studies addressed clinically important neonatal outcomes including necrotizing enterocolitis, retinopathy of prematurity, and sepsis. Results were, therefore, inconclusive. No data were available for bronchopulmonary dysplasia, death within initial hospitalization, periventricular leukomalacia, and seizures. ### Mean Gestational Age at Delivery Larger cohort studies in women with recurrent preterm labor and single or twin gestation demonstrated consistent benefit of SQ terbutaline pump compared with oral tocolytics or no treatment (although there is a high risk of bias in the available data). For women with recurrent preterm labor and singleton pregnancies, the difference in gestational age ranged from 0.70 to 3.40 weeks (95% CI, lower bound range 0.28 to 1.80; upper bound range 0.98 to 5.00) favoring SQ terbutaline pump. In pregnancies complicated by recurrent preterm labor in twins, the difference in gestational age from two cohorts was 0.70 weeks (95 percent CI, lower bound range 0.43 to 0.48; upper bound range 0.92 to 0.97),. ### Incidence of Delivery at Various Gestational Ages As with other outcomes, the strength of evidence surrounding gestational age at delivery was low. The SQ terbutaline pump consistently appeared to reduce the odds of delivering \<32 weeks in women with recurrent preterm labor with twin pregnancies in the six Matria-based cohort studies. The odds ratios ranged from 0.04 to 0.52 (95% CI, lower bound range 0.00 to 0.35; upper bound range 0.50 to 0.76),. The risk of any preterm delivery (\<37 weeks) was also assessed. Low strength of evidence favored SQ terbutaline pump compared with oral tocolytics or no treatment in women with recurrent preterm labor. Four of five cohort studies with important risk of bias reported statistically significant reduction in the odds of delivery \<37 weeks (OR 0.04 to 0.75; 95 percent CI, lower bound range 0.01 to 0.58; upper bound range 0.23 to 1.20). ### Prolongation of Pregnancy As with other outcomes, the strength of evidence for prolongation of pregnancy was insufficient or low. Evidence favored SQ terbutaline pump in women with recurrent preterm labor or twin gestation in cohort studies. The range for mean number of days of pregnancy prolongation was 5.50 to 25.30 (95 percent CI, lower bound range 0.79 to 16.77; upper bound range 8.72 to 33.83),. This evidence came from five cohort studies of medium to high risk of bias. In one Matria-based cohort study, more women in the SQ terbutaline pump group had pregnancy prolonged \>7 days compared with women who received oral nifedipine (OR = 7.84, 95 percent CI, 3.59, 17.12). Other Matria-based studies reported statistically significant benefits in favor of the pump compared with oral tocolytics for prolongation \>14 days (OR range = 1.93 to 3.47, 95 percent CI, lower bound range 0.87 to 2.34; upper bound range 2.65 to 5.15). ### Birthweight Cohort studies of women with recurrent preterm labor and single or twin gestation demonstrated statistically significant differences in mean birthweight in favor of SQ terbutaline pump compared with oral tocolytics or no treatment (range of mean difference in grams = 136 to 721, 95 percent CI, lower bound range 83 to 355; upper bound range 189 to 1087),. Studies reporting the incidence of low birthweight (\<2500 g) found statistically significant differences in favor of SQ terbutaline pump compared with no treatment or oral tocolytics (OR range = 0.24 to 0.64, 95 percent CI, lower bound range 0.06 to 0.51; upper bound range 0.62 to 0.96),. Most of the studies reporting the incidence of very low birthweight (\<1500 g) also found statistically significant differences in favor of the pump (OR range = 0.22 to 0.46, 95 percent CI, lower bound range 0.07 to 0.29; upper bound range 0.60 to 1.06). The studies that reported birthweight were mostly of medium or high risk of bias. ### NICU Admission For incidence of NICU admission, statistically significant differences favoring SQ terbutaline pump were reported in studies of medium or high risk of bias (OR range 0.28 to 0.72, 95 percent CI, lower bound range 0.08 to 0.58; upper bound range 0.63 to 0.97),. Statistically significant differences in favor of SQ terbutaline pump were also reported for NICU length of stay in these cohort studies (range of mean difference in days: −3.50 to −17.90, 95 percent CI, lower bound range −5.26 to −32.88; upper bound range −1.74 to −3.54). ### Other Surrogate Outcomes One cohort study reported a non-significant difference between SQ terbutaline pump and oral tocolytics in requirement for ventilation among infants with NICU admission. Pregnancy prolongation index was reported in two cohorts studies of medium and high risk of bias. Both found statistically significant differences in favor of SQ terbutaline pump compared with either no treatment or oral terbutaline (mean difference range = 0.14 to 0.41, 95 percent CI, lower bound range 0.26 to 0.56; upper bound range 0.02 to 0.26). ### Maternal Harms There were no reports of maternal death in the included studies. Underpowered studies demonstrated indeterminate results for pulmonary edema, therapy discontinuation and maternal hyperglycemia (i.e., type II error cannot be excluded). One prospective cohort of women with singleton pregnancies and recurrent preterm labor suggested that pump use was associated with increased tachycardia/nervousness (OR = 25.48, 95 percent CI:1.23, 526.6). No data within the study settings were identified for withdrawal due to adverse events, heart failure, hypokalemia, myocardial infarction and refractory hypotension. By 2009, 16 maternal deaths and 12 cases of maternal cardiovascular events (hypertension, myocardial infarction, tachycardia, arrhythmias and pulmonary edema) in patients receiving terbutaline tocolysis had been reported to the FDA. Three of the maternal deaths and three cardiovascular adverse events were reported in patients receiving SQ terbutaline pump therapy (<http://www.fda.gov/Drugs/DrugSafety/ucm243539.htm>). ### Neonatal Harms Data for neonatal harms were very sparse. Only one small RCT comparing SQ terbutaline pump with placebo and oral terbutaline demonstrated non-significant differences for neonatal hypoglycemia. ### Incidence of Pump Failure Two case series and one RCT reported outcomes related to the pump device. In a case series of 51 women, one subject had dislodgment of catheter (2 percent, exact central CI: 0.5 percent, 10 percent) and there was one pump malfunction (2 percent, exact central CI: 0.5 percent, 10 percent). No infusion site infections or mechanical failures were observed in a case series of 9 women. An underpowered RCT demonstrated indeterminate results for the outcomes of local pain and local skin irritation. No data were available for missed doses or overdoses. ### Assessment of Confounding by Level of Activity and Level of Care Only a small number of studies could be rated for level of activity and level of care, precluding an exploration of the effect of these variables on maternal and neonatal outcomes. Qualitative assessments did not yield any further insights. ### Applicability summarizes the applicability of the body of evidence. The majority of evidence pertained to women with recurrent preterm labor and singleton gestation. Very little is known about the study population's demographic and clinical characteristics, placing significant restrictions on the generalisability of results. Furthermore, the possibility that subjects represented a selective group of participants remains a concern. # Discussion Across several outcomes, the evidence favors subcutaneous terbutaline pump as maintenance tocolytic therapy for women with arrested preterm labor. However, our confidence in the validity and reproducibility of this evidence is low. Most of the evidence came from biased observational studies that reported surrogate outcomes only. Furthermore, the safety of the pump therapy remains unclear, largely because studies lacked power to detect differences in outcomes of harm. A total of 14 unique studies comprised the body of evidence investigating efficacy and harms of SQ terbutaline pump therapy as maintenance tocolysis in women with arrested preterm labor. Evidence from the only two included RCTs was underpowered to detect differences in outcomes of efficacy and harms. Most data came from observational studies, several of which recruited subjects from a single Matria database. These studies were at significant risk of bias and exhibited considerable clinical and methodological diversity. While a single study demonstrated improvement in neonatal death in women with recurrent preterm labor and twin gestation receiving SQ terbutaline maintenance tocolysis, several studies presented evidence favoring the pump therapy on surrogate outcomes of preterm birth. However, the evidence for important neonatal health outcomes, neonatal harms, and maternal harms was inconclusive because the studies lacked power to detect differences in clinical events. Furthermore, there was no data on the long-term effects of terbutaline infusion on offspring. Although many decisions regarding SQ terbutaline pump are currently made on the assumption that short-term outcomes will correlate well with improved long-term outcomes, rigorous scientific evaluation is needed to confirm whether such factors lead to better outcomes in this population. These findings are consistent with those of two existing reviews of SQ terbutaline pump. As reported by Nanda et al., we found that the available RCT evidence demonstrated non-significant differences between the pump and placebo or oral terbutaline. In agreement with another review, we found that the RCT and observational evidence is conflicting. We noted that the RCT evidence did not demonstrate any benefit from the pump while cohort studies of limited methodological validity demonstrated statistically significant effects in favor of the pump for several outcomes. Based on post-marketing surveillance data, the FDA has issued a new warning against the use of terbutaline in general, and particularly as an injection, as maintenance tocolysis (i.e. beyond 48–72 hours) in pregnant women. The warning is a response to several cases of poor maternal outcomes in pregnancies treated with subcutaneous terbutaline. These cases raised concern regarding a potential causative relationship. It is important to consider that the outcomes in question occur infrequently in the pregnant population, even in the absence of terbutaline use. Assessment of the magnitude of the association between terbutaline use and harm continues to be challenging, given the rarity of events and the lack of good quality, well powered studies. Until this relationship is further delineated, use of terbutaline for the prevention of preterm birth should be limited to carefully-controlled study settings. ## Limitations The evidence base for this review contained several limitations. Most of the evidence originated from observational study designs with significant risk of bias. Important prognostic factors such as race, socioeconomic status, and fetal fibronectin level were not reported and co-interventions, such as administration of corticosteroids, were rarely described. Moreover, it is uncertain how free the available evidence is from confounding imposed by restriction of maternal activity and level of care. While our review of the literature was comprehensive in capturing comparative evidence estimating the magnitude of benefits and harms of pump therapy, one potential and practical limitation was restriction to evidence reported in the English language. There are several factors of applicability that should be considered by maternity care providers and policymakers when translating the evidence from this review. The majority of available evidence included women with recurrent preterm labor (i.e. those with arrested preterm labor following first-line tocolytic therapy for 48 hours and then presenting with a second episode) and singleton gestation, with some evidence including women with twin gestation and recurrent preterm labor. Several studies included patients from a national proprietary database run by Matria Healthcare, which provides an outpatient perinatal program consisting of 24-hour nursing and pharmacy support, home uterine activity monitoring, individualized education, and provision of tocolytic therapy to women with preterm labor. These women received a high standard of care. It is notoriously difficult to conduct trials to assess the efficacy of tocolytics. Studies regarding tocolytics have been plagued by the elusive diagnosis of preterm labor as up to 40 percent of women diagnosed with preterm labor may not actually be in labor. As such, a significant proportion of women enrolled in clinical trials of tocolytic efficacy may not be destined to deliver preterm. A definitive trial in this domain must include a focus on accurate diagnosis of preterm labor, perhaps, combining stringent clinical criteria with factors such as positive fetal fibronectin and shortened trans-vaginal cervical length. Outcomes to be investigated should go beyond those of prolongation of pregnancy and birthweight to hard clinical endpoints of neonatal morbidity. Furthermore, the trial should include long-term follow-up to assess subsequent childhood outcomes. It is evidence of improvements in clinical effectiveness outcomes that can impact clinical decision making, societal healthcare costs and guideline recommendations. In conclusion, our systematic review calls into question the evidence base supporting the current practice of using terbutaline pump as a maintenance tocolytic agent. We feel strongly that use of terbutaline infusion for maintenance tocolysis should be restricted to well-designed, carefully- controlled study settings until there is clear evidence supporting its use. Further, decision and policy makers should take into consideration the limitations of the available data, both in terms of benefit and harm, when formulating recommendations. # Supporting Information The authors thank Steve Doucette, Raymond Daniel, and Sophia Tsouros, for their assistance with statistical analysis, data extraction, editorial assistance and preparation of the original evidence report and this manuscript. [^1]: Conceived and designed the experiments: LG MTA KS LW. Performed the experiments: LG MTA KS LW BS AT. Analyzed the data: MTA KS. Wrote the paper: MTA KS LG. [^2]: The authors have declared that no competing interests exist.

# Introduction In adolescence, significant increases in alcohol consumption are usually found, with prevalence rates of last month alcohol use rising from 16.1% at age 12 to 84.8% at age 16 in Dutch adolescents. Alcohol drinking in adolescence has been associated with several negative consequences. Of particular concern are the neurotoxic effects of alcohol, since the significant maturation of both brain structure and function are assumed to underlie a specific vulnerability of the developing adolescent brain to the adverse effects of alcohol. However, the findings from empirical studies that tried to assess these effects so far remain inconclusive due to a number of methodological pitfalls. Most studies that compared alcohol abusing adolescents with non-abusing adolescents on a broad range of neurocognitive functions, such as language and general intelligence, attention and intelligence, learning, memory, and visuospatial functioning, are not convincing in their conclusions because of their cross-sectional designs. As a result, the reverse effect of neurocognitive impairments on heavy alcohol use is neglected. Furthermore, studies were conducted among adolescents diagnosed with Alcohol Use Disorder (AUD), which limits the generalizability of the findings because this specific group has behavioural problems associated with controlling their behaviour (according to DSM-IV-criteria) and often psychiatric comorbidity. This amplifies the limitations to assessing causal relations. In contrast, population studies have shown almost no significant differences between excessive drinkers and controls on neurocognitive functioning. However, these population studies are again limited by cross-sectional designs and small sample sizes. Also, definitions of excessive or heavy drinking are not consistent across the studies. To the best of our knowledge, only three small scale studies (*n* = 75 and 40 respectively) have analysed the effects of alcohol use on neurocognitive maturation in adolescence using a longitudinal design with pre- and post- measurements of neurocognitive functioning in a general population. The results of these studies do not support the damaging effects of alcohol use in adolescence either. One study found differences between heavy drinkers (average drinks per month: 9.9 for girls and 6.1 for boys) and controls on only one out of seven neurocognitive tasks, and this difference was significant for girls only. However, there was a relation between hangover symptoms and sustained attention in boys. Two other studies showed increased brain activation with fMRI measurements (but not in all hypothesized brain areas) in adolescents who transitioned to heavy drinking (drinks per drinking day: 4.2 and 6.1 respectively), while no differences between drinkers and non-drinkers were found on task performance. Thus, empirical research on the effects of heavy alcohol use on neurocognitive maturation does not result in undisputed findings and calls for large scale population studies on this subject. The aim of the present study was to investigate whether adolescent alcohol users show a distinctive maturation of basic executive functions compared to non- drinkers in a large population-based sample. We focus on Executive Functioning (EF), since this set of cognitive control functions mediates the ability to guide and direct behaviour and is therefore essential for success in school and at work. Furthermore, executive functioning is hypothesised to develop specifically during adolescence, as it parallels the maturation of the parietal and prefrontal cortices. As a result of this prolonged maturational trajectory, executive functioning is assumed particularly vulnerable to the effects of alcohol. We used four measures of EF, inhibition, working memory, sustained attention, and shift attention, as our main outcomes in this study. We used computerised tasks to assess basics forms of these functions, which allowed us to use the same tasks in both early and late adolescence, and compare maturation of task performance between adolescents with different drinking habits. We conducted a pre-exposure measure of executive functioning at age 11 and follow- up measurement at emerging adulthood (age 19). Furthermore, since girls are supposedly more vulnerable due to differences in neuromaturation, hormonal fluctuations, and alcohol metabolism, gender was considered as a possible moderator. # Methods ## Study design The present study used data from the first, second (for descriptive statistics only), third, and fourth wave of the TRacking Adolescents’ Individual Lives Survey (TRAILS). This is a prospective cohort study of Dutch pre-adolescents at age 11. The target sample involved children living in the North of the Netherlands, covering urban and rural areas Demographic information from five municipalities for all inhabitants born between October 1, 1989, and September 30, 1990, in two of the municipalities, and October 1, 1990, and September 30, 1991, in the other three was obtained. These children attended 135 schools that were approached for participation, of which 13 refused to participate (excluding 338 children). Next, parents of eligible children were informed about the study and then invited to participate. After excluding 210 children who were unable to participate because of serious health or language problems, we invited 2,935 eligible children and their parents to enter the study. Seventy-six percent of eligible adolescents and their parents agreed to participate and were enrolled in the study at baseline (T1) (*n* = 2,230, mean age 11.1 years, *SD* = 0.56, 49.2% male) (for more details on the procedure see:. At baseline, children of lower socioeconomic background, boys, and children with poor school performance were slightly less likely to participate. Participants and nonparticipants did not differ in emotional and behaviour problems (De Winter et al., 2005). At the third (T3) wave (*n* = 1,816, mean age 16.3 years, *SD* = 0.70, 47.7% male), the response rate was 81.4%. At the fourth (T4) wave (*n* = 1,596, mean age 19.2 years, *SD* = 0.57, 46% male), the response rate was 70%. ## Procedure On the first assessment, trained undergraduate psychology students administered neuropsychological tests in adolescents’ schools or designated testing centres (for more information, see. Participants who were unable to attend these assessments were tested at home. On the first to third assessments, adolescents completed self-report questionnaires in groups in school, supervised by an assistant. Their parents also completed a written questionnaire. On the fourth assessment, most adolescents were no longer in secondary education. Therefore, trained professional interviewers conducted the neuropsychological test battery individually at participants’ homes or in a nearby community centre (for more information, see. Parents and their children were asked to fill out a computerised questionnaire (or, per request, a paper-and-pencil questionnaire). The Dutch Central Committee on Research Involving Human Subjects approved the study. Parents and adolescents’ written informed consent was obtained. The confidentiality of the study was emphasised. ## Measures ### Alcohol use *Descriptive statistic*: At T1, adolescents were asked: “How often have you been drinking alcohol (for example, a bottle of beer or a glass of wine)?” up until that time point, on a 5-point scale ranging from 0 to 7 times or more. We dichotomised these answers into ‘never or once’ and ‘twice or more’. At T2-T4, adolescents were asked to report their average drinking habits between the previous data collection wave and the present. They were asked four questions: “On how many week(end) days do you normally drink alcohol” and “On an average week(end) day on which you drink alcohol, how much alcohol (glasses, cans, bottles) do you drink?” By multiplying and adding the answers, average weekly quantity of alcohol use can be computed. A Dutch standard drink contains 10 grams of alcohol. Furthermore, at T2-T4 adolescents were asked how many times they had been drunk in the last 12 months. We dichotomised these answers into never and once or more. These measures of alcohol use were used as descriptive statistics. *Drinking groups*: Since the legal age for buying alcohol in the Netherlands was 16 years at time of the data collection, we used data from T3 and T4 to determine group assignment. As to be expected, prevalence rates of frequent heavy drinking (see below) at T2 were indeed very small (\<3%). For constructing drinking groups we used measures of both quantity and frequency of alcohol use. The average amount of glasses on a weekend day (since previous measurement wave) was used since weekend quantity of alcohol use has been shown to be a useful and specific measure of alcohol use at this age. We furthermore used the question “On how many occasions in the last month have you had an alcoholic beverage to drink?” Group assignment was done in two steps; in a first step, participants were classified into four groups according to their answers to these questions per measurement wave. In the second step, these groups per measurement wave were combined, resulting in six final drinking groups. In step one, per measurement wave, we started with identifying participants that were *not drinking*; those who indicated that they did not consume alcohol on a regular weekend day. Next, for the respondents who consumed alcohol, we set a cut off score at 6 glasses on a weekend day for boys and 5 glasses for girls. Participants who indicated that they drank alcohol on a regular weekend day, but less than the cut off, were qualified as *drinking*, *not heavy drinking*. For participants scoring above the cut off (i.e., drinking 5–6 glasses or more on a regular weekend day), we looked at frequency of alcohol consumption. Those scoring above the cut off but drinking on an irregular bases (last month prevalence \< 4 times heavy drinking) were considered to be *infrequent heavy drinking*. Finally, respondents falling above the cut off and drinking regularly (last month prevalence ≥ 4 times drinking, i.e., weekly drinking) were classified as *heavy drinking*. In step two, these categorizations per measurement wave were combined into longitudinal drinking groups. Participants were divided into six groups: non- drinker, light drinker, infrequent heavy drinker, increasing heavy drinker, decreasing heavy drinker, and chronic heavy drinker. For example, a participant *not drinking* at T3 and *heavy drinking* at T4 was longitudinally classified as <u>increasing heavy drinker</u>. For all possible combinations and descriptive statistics, see. ### Executive functioning Baseline executive functioning was examined at T1, using three computerised reaction time tasks from the Amsterdam Neuropsychological Tasks (ANT). The ANT has proven to be a sensitive and valid tool in non-referred samples, as well as in referred samples of various clinical domains. We assessed four basic executive functions: inhibition, working memory, shift attention, and sustained attention, since they have sufficient longitudinal stability and the functions are theoretically hypothesized to be associated with alcohol deficits. The ANT has shown to be applicable for longitudinally assessing these basic executive functions in an adolescent population. In a description of the tasks used is provided. The use of computerised tasks guarantees standardised assessment while working with reaction times allows detection of subtle improvements in performance. Working memory, inhibition, and shift attention were calculated by subtracting the performance time of a baseline version from a more difficult version of the task. Performance results reflecting more than 50% errors were coded as missing and imputed because they were assumed to reflect either misunderstood instructions or false computer settings, undermining the validity of the testing. In accordance with previous research on neurocognitive maturation, we calculated change scores for each of the four executive functions by computing z-scores for the T1 and T4 measures and subsequently subtracting these scores from each other (T1-T4). ### Measurement of covariates In accordance with previous studies, age at the first assessment, socioeconomic status (SES), paternal and maternal alcohol use, delinquent behaviour at T1, and smoking and cannabis use at age T3 and T4 were included as covariates if they significantly correlated with the outcomes measure (*p* \<.05) (see for bivariate correlations). *SES* was assessed using income level, educational level, and occupational level of both parents (occupational level was based on the *International Standard Classification for Occupations*). These five variables were standardised and combined into one scale with an internal consistency of.84 (for more information, see. *Paternal and maternal alcohol use* was assessed at T1 by asking the parent who filled out the questionnaire how much alcohol (s)he drinks per week on average and how much his or her partner drinks. *Delinquent behaviour* was measured at T1 Youth Self Report (YSR), consisting of 15 items (α =.71). *Smoking* was also measured at both T3 and T4 by asking: *“*How many times have you smoked cigarettes during the last month?*”*. *Cannabis use* was assessed at T3 and T4 by asking: “How many times have you used weed (marihuana) or hash during the last year?”. ## Data Analyses Multiple data sets in SPSS 20 were imputed, using fully conditional specification (MCMC) with Predictive Mean Matching because there were missing values due to attrition on both predictors and outcome measures. This approach generated five datasets and one pooled dataset. The six drinking groups were validated by comparing them on alcohol related behaviour, such as prevalence of last year drunkenness and drinking in early adolescence. For variables measuring prevalence, Pearson χ² with standardized residuals was used to identify observed counts that were significantly different (*p* \<.05) from expected counts. For continuous variables, MANOVA’s with post-hoc tests were used. For these two specific analyses, the five datasets were analysed separately, as pooled analyses are unavailable for these tests. To examine the influence of drinking groups on maturation of executive functioning, we conducted four bivariate and multivariate linear regression analyses. Standardised change scores of executive functioning were entered as the dependent variable. Dummy variables of drinking groups were constructed and all dummy variables were entered in the model simultaneously as predictors (non- drinkers served as the reference group). First, we conducted four separate linear regression analyses without confounders. In the second model, we conducted four multivariate linear regression analyses adjusting for control variables. Main effects of drinking groups and drinking patters\*gender interaction were entered in separate blocks and interpreted accordingly. To reduce Type 1 error, we set α at \<.01. For the regression analyses, data from the pooled dataset were used for interpretation. # Results ## Descriptive statistics Descriptive statistics are depicted in. There were no differences between the groups for baseline executive functioning, SES, and age. Pearson χ² yielded significant differences between groups for prevalence rates of other substance use, drinking in early adolescence, and drunkenness. Non-drinkers and light drinkers scored lower than expected on all variables while the chronic heavy drinkers scored higher on all variables. A MANOVA indicated significant differences between groups on all continuous measures (*p* \<.001). Parental alcohol use and delinquency scores in pre- adolescence were largest in the frequent heavy drinking groups (i.e., increasers, decreasers, chronic heavy drinkers). Also, those who later would be classified as frequent heavy drinkers consumed significantly more alcohol at age 13 than the future non-drinkers, light drinkers, and infrequent heavy drinkers. Chronic drinkers, as to be expected, drank most at ages 16 and 19 with an average weekly consumption finally exceeding 14 glasses. Taken together, the results indicate significant differences between the drinking groups on all measures of alcohol related behaviour, validating our identification of the six drinking patterns. The patterns become more differentiated as the adolescents get older. ## The influence of drinking groups on maturation of executive functioning Raw scores of baseline (T1) and follow-up (T4) executive functioning and standardized change scores are depicted in. The results of the bivariate linear regression analyses are depicted in. Drinking group did not significantly predict maturation of executive functioning from age 11 to age 19 for any of the four measures. The results of the multivariate linear regression analyses are depicted in. In step 1, T1 baseline performance, gender, and other confounders were added. For all outcome measures, baseline performance significantly predicted maturation, with a higher initial score (i.e., a less optimal performance at baseline) predicting more maturation. This is to be expected since a less optimal performance leaves more room for improvement. Gender differences were significant for inhibition and shift attention, with boys showing more improvement on these tasks. In step 2, we added drinking groups as dummy variables, with non-drinkers as reference group. This did not significantly improve model fit for any of the measures, and none of the dummy variables predicted maturation of executive functioning compared to the non-drinkers. In step 3, we added drinking group\*gender interaction variable. Again, this did not improve model fit and thus, gender did not moderate the effects of drinking groups on maturation of executive functioning. ## Additional analyses All analyses were also conducted with observed cases only (see and Tables). This did not yield different results from the analyses with imputed datasets; i.e., the results were not significant. Furthermore, we conducted four additional analyses. First, we were interested if the use of a continuous measure of alcohol consumption resulted in different outcomes. Therefore, we conducted linear regression analyses where the average number of glasses per week at T3, at T4, and the sum of these two measures as predictors. This did not result in significant findings (*p*\>.01). Second, we tested whether the findings changed when drinking groups were formed according to the definition of heavy drinking by the National Institute on Alcohol Abuse and Alcoholism (i.e., \>3/4 glasses daily and \>7/14 glasses weekly for women and men respectively). The analysis did not yield significant results (*p*\>.01). Third, non-drinkers may consist of a non-normative group of adolescents; therefore, we also examined whether the findings changed when we used light drinkers or infrequent heavy drinkers as reference group instead of non-drinkers. The results were not significant (*p*\>.01). Finally, since there are differences in maturation in executive functioning between males and females, we also undertook separate analyses for boys and girls, instead of using gender as a moderator. The results were not significant (*p*\>.01). (Data of additional analyses not presented, information available by first author on request). # Discussion Our longitudinal study did not show any measurable differences between groups of alcohol users with respect to maturation of executive functioning, contrary to the hypothesis that the developing adolescent brain is particularly vulnerable to neurocognitive aversive effects of alcohol (e.g.,. The present study found expected maturation on all the executive functioning tests that we used (inhibition, working memory, and sustained and shift attention),–hereby measuring executive functioning over a time span of eight years, i.e., with a pre-exposure measure and after four years of (heavy) alcohol consumption. However, we did not find any differences in maturation between any of the drinking groups, not even of the heaviest drinkers (i.e., drinking every weekend and drinking an average of 15 glasses of alcohol each week) for a period of at least four years in adolescence. Gender did not moderate the relation between drinking groups and maturation of executive functioning, which is in contrast to the previous findings that girls are assumed more vulnerable to the toxic effects of alcohol than are boys. Our findings are thus in sharp contrast with the common assumption that alcohol leads to measurable changes in adolescents’ executive functioning, although they appear to be in line with the results from the few existing longitudinal studies in adolescents in the general population that did not find neurocognitive deficits at a behavioural level in heavy drinkers for the vast majority of tasks. This implies that statements on how alcohol affects the adolescent brain apparently suffer from an overgeneralisation of research in clinical groups and an over-interpretation of cross-sectional research. On the positive side, the effect of alcohol on the developing brain does not appear to affect the basic executive functioning in an irreversible and devastating way. Thus, despite the obvious acute effects of alcohol as a toxic substance on the adolescent brain (being drunk, being unable to reason while being intoxicated, having hangovers and headaches), it may seem to be flexible enough to cope with these effects of alcohol at least at the level of behavioural development and neuropsychological maturation. That is not to say that the damaging effects of alcohol can be neglected *as such*. Alcohol use might have an effect on brain functions not tested in the current study or on the level of brain activation patterns (e.g.. Furthermore, in a recent review paper it was concluded that although there is limited evidence for the effects of alcohol use in adolescence on cognitive functioning per se, there is evidence for alcohol being related to stronger automatic affective towards alcohol cues. Especially adolescents with weak (premorbid) executive functions might be at risk for these processes. As important, heavy alcohol use in adolescents is associated with a large number of other well-established risks, such as developing an alcohol use disorder, driving under influence, and engaging in risky behaviour, including violence and fighting while being intoxicated. Furthermore, our results do not rule out the possibility of irreversible effects of alcohol in the long run, either after continuation of heavy drinking of a longer period of time, or the possibility that adolescent heavy drinking might set the stage for deficits in neurocognitive functioning that would manifest at some point later in life. ## Strengths and limitations Our approach has several strengths and limitations. Strengths are the large population sample and the longitudinal design. To the best of our knowledge, this study was the first to assess heavy drinking in relation to maturation of executive functioning in a longitudinal population cohort of adolescents, covering about eight years, which is longer than in any of the other previous studies. We controlled for several significant confounders that correlated with our dependent variable. We conducted our study in the Netherlands, where legal drinking age is much younger than for example in the United States and alcohol consumption in adolescence is very common, optimising the chances for finding effects of heavy drinking on maturation of executive functioning. The first limitation concerns the basic tasks used to measure executive functioning. We measured the behavioural consequences of heavy and regular alcohol use with straightforward reaction time tasks, with each task measuring specific sub-components of executive functioning. This does not entail any knowledge on whether the underlying neuro-anatomy is affected and to what extent. For example, equal task performance in heavy drinkers and controls can still be accompanied by differences in neural activation while performing this task. It is unclear what these differences represent, but they suggest that alterations in neural processing do not necessarily appear at a behavioural level. In addition, the possibility exists that more complex neuropsychological tasks might have been more sensitive in picking up such alterations. However, an important advantage of starting with the basics was that it allowed using exactly the same tasks at both age 11 and 19, which is a requirement for finding longitudinal change. More complex and strategy-based tasks usually have more stringent age restrictions, and tasks that are both feasible for early adolescents yet still challenging in late adolescence are difficult to find. Furthermore, using straightforward tasks circumvents the problem of ‘task impurity’. Since more complex tasks are assumed to rely on multiple cognitive processes and their integration, it is difficult to identify processes that are responsible for a suboptimal performance. The reaction time tasks we used were not designed to detect deficits, but are able to detect differences between groups on the level of performance. Our findings indicate no performance differences between drinking adolescents and abstaining peers on basic functions, making deficits in these skills unlikely. The second limitation could be that drinking groups were constructed manually using self-reported measures of alcohol use, although this is common in longitudinal research on adolescent alcohol use. Self-report questionnaires have proved to be reliable for assessing alcohol use in adolescence. In addition, our drinking groups showed good and consistent differentiation on validating measures. Although these concern measures obtained from the respondents themselves and not external or independent validators, these findings are reassuring. Chronic drinkers score highest on all alcohol-related behaviours and reveal a heavy drinking pattern at two consecutive waves, which cover at least four years of regular heavy drinking. We are therefore confident that we have adequately identified the most risky drinkers. The final limitation of longitudinal designs is that attrition may have biased the findings, with most at-risk participants dropping out, resulting in an underestimation of effects and possible loss of power. However, using multiple imputation, missing data on alcohol use and executive functioning were imputed based on a wide variety of associated variables. This technique improves validity of datasets with missing data. Therefore, in our study, attrition bias is unlikely to explain the absence of significant results. ## Implications We did not find the effects of adolescent alcohol use on maturation of executive functioning in adolescence. Four years of self-reported weekly heavy drinking did not result in deviancies in behavioural performance on a variety of straightforward executive functioning tasks. However, these finding should not be seen as reassuring about adolescent alcohol use as there are numerous other risk related to heavy drinking, such as developing an alcohol use disorder, driving under influence, and engaging in risky behaviour that include violence and fighting while being intoxicated. Consideration of these risks calls for continuous prevention efforts targeting heavy alcohol use in adolescents. # Supporting Information This research is part of the TRacking Adolescents' Individual Lives Survey (TRAILS). Participating centres of TRAILS include various departments of the University Medical Center and University of Groningen, the Erasmus University Medical Center Rotterdam, the University of Utrecht, the Radboud Medical Center Nijmegen, and the Parnassia Bavo group, all in the Netherlands. We are grateful to all adolescents, their parents and teachers who participated in this research and to everyone who worked on this project and made it possible. [^1]: The authors have declared that no competing interests exist [^2]: Conceived and designed the experiments: JO, FCV, MZ, WAMV. Performed the experiments: SRB. Analyzed the data: SRB, ZH. Wrote the paper: SRB, ZH, MZ, SAR, FCV, JO, WAMV.

# Introduction Human modification of the environment has greatly accelerated the rate of global biodiversity loss in recent decades. A leading driver of these losses is the introduction of non-native species to new environments. Once established, these invasive alien species (IAS) can alter ecosystem dynamics, especially through predation, competition, and disease transmission, resulting in rapid declines to endemic species. IAS effects on biodiversity are profound; 58% of extinctions listed in the International Union for Conservation of Nature (IUCN) Red List name IAS as main contributors. IAS also persist as economically important pests, costing Canada between \$16.6 and \$34.5 billion dollars annually, with approximately \$20 billion of that in the Forestry sector alone. Likewise, the United States spends \>US\$220 billion annually in prevention efforts, damages, and habitat restoration. These negative effects are amplified within island systems, evidenced by IAS being the leading cause of the majority (86%) of recent island extinctions. Moreover, insular systems support disproportionately high levels of endemism relative to mainland areas, punctuating the need for immediate action towards improving IAS management on islands. Invasive mammals are the main cause of animal extinctions on islands and constitute one of the most significant threats to insular biodiversity. Brown rats (*Rattus norvegicus*), black rats (*R*. *rattus*), and Pacific rats (*R*. *exulans*), in particular, make up a triad of the most invasive mammals on the planet, found on every inhabited continent and \>80% of all oceanic islands. To date, invasive rats have caused \>50 species extinctions and are a leading cause of extinction for insular species. Island seabirds, in particular, are adversely affected by rat invasions, as these species are poorly adapted to terrestrial predators due to their natural absence in many island systems; in fact, approximately 60% of all seabird extinctions are linked to predation by invasive rats, 90% of these occurring on islands. As part of seabird recovery efforts, management agencies often use whole-island eradications to manage invasive rat populations. Yet, rodent eradication attempts on islands throughout the world have resulted in a \~10–38% failure rate. In many cases, failure was due to knowledge gaps of key population parameters, including IAS population size, home range size, and dispersal capacity. Genetic and genomic tools provide promising opportunities to inform vertebrate IAS management on islands. For example, genetic evidence was a large factor in the success of the largest rodent eradication in history on South Georgia Island (\~390,000 ha) in the British Overseas Territory of the south Atlantic, providing critical information on population structure and demographic history that allowed for effective planning of a multi-stage eradication. In other cases, genetic and genomic tools can be used to explicitly evaluate the underlying cause of eradication failure, summarized in two prevailing hypotheses. The survivor hypothesis suggests that the eradication effort failed to remove all individuals, leading to these individuals repopulating the island. On the other hand, the re-invader hypothesis posits that the eradication was successful in the complete removal of rats, but that the island was re-colonized from some connected source population. Genetic evidence was used to test these hypotheses following an unsuccessful rat eradication on Pearl Island in New Zealand; genetic analyses of population structure and individual assignment among pre- and post- eradication samples revealed evidence for re-colonization from a neighboring source population on Stewart Island (*i*.*e*., re-invader hypothesis), suggesting that a larger target area may be required to avoid eradication failure in this system in the future. Although useful for informing IAS management, genetic analyses to date have been largely reactive (*e*.*g*., evaluating causes of eradication failure), time-consuming, and costly. Consequently, there is a need to harness leading-edge genomic technologies for providing genetic tools to aid managers in the decision-making process before, during and after IAS management to help improve efficiency, eradication durability, and cost-effectiveness. Here, we developed a Genotyping-in-Thousands by sequencing (GT-seq) panel for rapid species identification and population assignment of invasive brown and black rats (RapidRat) in Haida Gwaii, an archipelago off the central coast of British Columbia (BC), Canada. GT-seq is a targeted, multiplexed amplicon sequencing approach that can simultaneously genotype sample sizes ranging from a single individual up to thousands of individuals at hundreds of single nucleotide polymorphisms (SNPs). We validated the RapidRat panel by comparing estimated population structure and assignment accuracy with results from a previously published double digest restriction site-associated DNA sequencing (ddRAD) analysis (n = 27,686 SNPs) of the same samples within this system. Lastly, we deployed RapidRat to identify the source of recent, novel brown rat invasions on Agglomerate, Hotspring, and Ramsay Islands. # Methods ## Study system Haida Gwaii (<u>X</u>aayda Gwaay in Haida) is an isolated archipelago consisting of approximately 150 smaller and two larger islands located \~80 km off the central coast of British Columbia (BC), Canada. This system contains more unique endemic sub-species than any other equal sized area in Canada and is host to large numbers of seabirds and globally critical breeding habitat. Both brown and black rats have invaded Haida Gwaii, likely as stowaways on European ships using the archipelago as a fishing and whaling port in the 1700’s to early 1900s. Since their introduction, invasive rats have adversely affected native seabird populations, resulting in the extirpation of several large multi-species colonies and population declines in at least half of the species present. To meet seabird recovery goals and to protect ecological and cultural integrity, Parks Canada and other agencies started managing invasive rats in the 1990s primarily through whole-island eradications. Since 1997, invasive rats have been successfully eradicated from five islands throughout the archipelago; however, brown rats were re-detected on the Bischof Islands following two eradication attempts in 2003 and 2011, and brown rats were detected on Faraday and Murchison Islands four years after a successful black rat eradication in 2013. Brown rats have now also spread to Hotspring, Ramsay, and Agglomerate Islands, all of which have been historically rat-free (C. Bergman, pers. comm.). The origins of these invasions are unknown, though both Agglomerate and Hotspring Islands are situated \~1 km from the nearest brown rat population on Murchison Island, while the newly established population on Hotspring Island could have invaded Ramsay Island (900-1000m away); these spans are within the hypothesized swimming distance for brown rats in temperate waters. Additionally, rats could have potentially dispersed from island to island by floating on debris (*i*.*e*., “rafting”) during high tide, and/or by commensal spread on ships. We previously characterized population connectivity among brown and black rat populations in Haida Gwaii using ddRAD sequencing (n = 27,686 SNPs) across the archipelago. We found discrete genetic units largely corresponding to island populations, and we characterized within-island dynamics on the larger islands of Graham, Lyell, and Kunghit. We also assigned individuals from recent invasions on Bischof, Faraday, and Murchison Islands as invaders from neighbouring Lyell Island, demonstrating the utility of genetic analysis in this system, and providing an excellent source of reference data for the current study. ## Study design To identify candidate loci for our GT-seq panel, we used ddRAD data previously generated from brown rats (*n* = 295) and black rats (*n* = 241) collected throughout the Haida Gwaii archipelago. Of these samples, brown rat (*n* = 59) and black rat (*n* = 37) DNA extractions (used in) were re-sequenced and genotyped using GT-seq to optimize panel loci (see below). In addition to these samples, we sequenced and genotyped brown rats using GT-seq from novel invasions on Agglomerate Island (*n* = 18), Ramsay Island (*n* = 1), and Hotspring Island (*n* = 1). See for detailed sample distribution. All sample collection was performed under Parks Canada Agency Animal Care Committee protocol GHNPR11-5. Whole genomic DNA (gDNA) was extracted from 10–20 mg of dried ear tissue using the Qiagen DNeasy^® Blood and Tissue Kit and treated with RNase A (5PRIME) following the manufacturer’s protocol and stored in distilled water until GT-seq library preparation. ## SNP discovery, quality control, and panel selection Raw sequencing reads were first demultiplexed from the ddRAD samples using *process_radtags* as implemented in <span class="smallcaps">Stacks</span> v2.0b. Reads were trimmed to 100 bp to remove low-quality base calls at the 3’ ends. Individual reads were aligned to the most-current brown rat reference genome (Rnor_6.0, GenBank assembly accession: GCA_000001895.4) using the default settings in Bowtie 2 v2.2.9. Single nucleotide polymorphisms were identified and genotyped using *gstacks* and *populations* in <span class="smallcaps">Stacks</span> v2.0b. To ensure only high-quality SNPs were retained, all loci were filtered such that they were genotyped in \>90% of total individuals, had a minimum minor allele frequency of 5%, and had a maximum observed heterozygosity of 50%. Additionally, only a single SNP per RADtag was retained to reduce the likelihood of linkage disequilibrium among loci. To ensure maximum quality and functionality of the panel, we additionally filtered loci such that they were variable in at least one species, genotyped in \>90% of individuals in each species separately, as well as with a mean depth of coverage of 6x in each species separately using VCFtools v0.1.15. Of these, only SNPs found within 40–60 bp of the start of the RADtag sequence were retained to ensure sufficient flanking sequence remained for primer design. We then estimated population differentiation for each locus both within and between species using Weir and Cockerham’s unbiased estimator *θ*. Estimates were calculated for all pairwise observations as implemented in Genetix v4.05.2. To ensure linkage equilibrium, locus pairs were positioned at least 1 Mb apart within the genome; in the event of a conflict, the locus with the lowest *θ* value was removed. We retained the top 350 loci with the highest discriminatory power (*i*.*e*., highest *θ*) within each species as well as an additional eight loci for differentiating between species for a total putative panel of 708 SNPs. The associated RADtag sequences for these loci were sent to GTseek LLC (<https://gtseek.com/>), where custom PCR1 primers were designed for each locus incorporating Illumina primer binding sites. ## Genotyping-in-thousands by sequencing and panel optimization To test the efficacy of our GT-seq panel, we sequenced and genotyped 92 individuals across both species (brown rats, *n* = 59; black rats, *n* = 37), preferentially selecting individuals with the highest quality ddRAD data and incorporating the entire geographic sampling distribution. Of these, four individuals were replicated to explicitly assess genotyping error rate (see below), resulting in a total 96 samples in our initial library. GT-seq library preparation followed the original protocol with some modifications. DNA extractions for all individuals were normalized to 20 ng/uL to ensure even amplification across samples and diluted the PCR1 products 1:10 before use in PCR2. We quantified PCR2 products using PicoGreen^™ (Molecular Probes, Inc.) and manually normalized these products to a concentration of 10 ng/μL. We pooled 2.5 μL into a final sequencing library and purified this library using the MinElute PCR Purification kit (Qiagen^®), eluting into a final volume of 50 μL of nuclease-free water. The library was sequenced using a single lane of Illumina MiSeq paired-end 100 bp sequencing at the McGill University and Génome Québec Innovation Centre. We genotyped raw sequencing data using the GT-seq pipeline found on GitHub (<https://github.com/GT-seq/GT-seq-Pipeline>). We used the *GT-seq_SeqTest*.*pl* and *GT-seq_Primer-Interaction-Test*.*pl* scripts to identify over-represented primer sequences and PCR artifacts and primer-dimers between loci. We removed loci accounting for \>1% of total forward primer reads, loci with total counts \>1% of PCR artifacts, as well as one locus per hetero-dimer pair. The remaining loci were combined into a new primer pool, and the library was re-sequenced and optimized following the above protocol. Loci that did not genotype in the final library, as well as individuals with \>50% missing data, were removed from downstream analysis. ## Assessment of population structure and panel validation To assess genotyping accuracy of the panel within each species, we compared individual genotypes from GT-seq to ddRAD at species-specific loci and measured the percent discordance across methods. We likewise calculated genotyping error rates within each sequencing method by comparing individual genotypes between replicate samples, also measured as percent discordance. Loci with missing data in either one or both of each pair of samples were not included in error calculations. We examined the ability of our panel to differentiate between species across all panel loci using discriminant analysis of principle components (DAPC) as implemented in the R package *adegenet v*2.1.1. This procedure first applies principle component analysis (PCA) to identify genetic clusters and then uses discriminant analysis to maximize the variation among these clusters while minimizing within-cluster variation. These results were compared to species- assignments from Sjodin *et al*. to assess the accuracy of assignment. To ensure our subset of loci resulted in similar patterns of population structure, we ran projected PCA using the ddRAD samples to infer eigenvectors and compared to Sjodin *et al*. for accuracy using the *smartpca* function in EIGENSOFT v7.2.1. We then projected GT-seq samples onto these eigenvectors to compare clustering across methods. Additionally, we estimated admixture coefficients for ddRAD using sparse non-negative matrix factorization (SNMF) as implemented by the *snmf* function in the R-package *LEA v*2.6.0. We averaged admixture coefficients over 20 iterations at the previously identified optimal *k* = 9 for each species and compared these results to Sjodin *et al*.. We also conducted population assignment following the method of Rannala and Mountain as implemented in Geneclass2.0. Reference populations were defined using only the ddRAD samples, and both ddRAD and GT-seq samples were posteriorly assigned to these populations. ## Assignment of novel invasions We evaluated the utility of the panel for assigning individuals of unknown origin to known source populations. We examined samples collected from novel invasions on Agglomerate Island (*n* = 18), Ramsay Island (*n* = 1), and Hotspring Island (*n* = 1) genotyped using GT-seq. To identify the source of these invasions, we conducted projected PCA using the *smartpca* function from EIGENSOFT v7.2.1 and the ddRAD data to define eigenvectors. As the brown rats in Haida Gwaii are strongly clustered to northern, central, and southern regional populations, we ran the projected PCA using only the regional population to which the unknown samples most closely clustered during initial evaluation of population structure (see above). We also assigned individuals to reference populations using the method outlined in Rannala and Mountain as implemented in Geneclass2.0, including all potential reference populations. For this analysis, the Faraday and Murchison Island populations were grouped into a single reference population based on previous results. # Results ## GT-seq panel quality and species determination From the initial 708 putative loci, we were able to design primers targeting 526 loci. Following optimization, we retained 443 loci across both species, including 3 of 8 SNPs that were diagnostically distinct between brown and black rats. Overall, this panel had high discriminatory power to the species-level, resulting in 100% concordance with ddRAD species assignment results based on 27,686 SNPs. In the optimized GT-seq panel, all 443 loci were significantly differentiated between brown and black rats, with 150 loci exhibiting pairwise *θ* \> 0.90. From this optimized set of 443 loci, 315 and 429 were variable within brown and black rats, respectively, and used as species-specific datasets. Mean GT-seq genotyping rates for brown rats was 88.6%; five individuals were removed due to a genotyping rate of \<50%. Mean GT-seq genotyping rates for black rats was 78.2%; two individuals were removed due to a genotyping rate of \<50%. Mean genotyping discordance across methods (ddRAD v. GT-seq) was 2.8% and 3.5% in brown and black rats, respectively. No genotyping error was observed between GT-seq replicates in brown rats; black rats revealed only a 0.2% genotyping error between GT-seq replicates. All GT-seq individuals accurately assigned to species, and all individuals from the novel invasions on Agglomerate, Hotspring, and Ramsay Islands were confirmed as brown rats. ## Assignment accuracy of brown rats We detected the same three regional clusters in the brown rats as previously identified via PCA. Furthermore, GT-seq samples accurately projected to their sampled ddRAD clusters. Ancestry coefficients estimated at the reduced panel were consistent with previous results, though there was a loss of some finer- scale structure relative to the previous ddRAD analysis, especially among centrally located islands. Population assignment resulted in one individual sampled on the Bischof Islands in the GT-seq dataset assigning to a population other than the one in which it was sampled (Tanu Island), though this individual was previously identified as a first-generation migrant from Richardson Island in the full ddRAD analysis. ## Assignment accuracy of black rats Black rats also clustered similar to previous PCA results based on ddRAD. As with the brown rats, admixture coefficients estimated using the reduced SNP panel were consistent with previous results, with all individuals accurately assigned to their sampled population. However, there was a loss of some finer- scale structure among the Graham Island and Sandspit populations, as well as within the Lyell Island populations relative to the ddRAD dataset. ## Identification of origin for novel invasions All samples from novel invasions clustered with the central brown rat populations when projected onto eigenvectors inferred using all reference populations; as such, we only considered central populations as putative sources. The projected PCA for the Agglomerate, Hotspring, and Ramsay Island samples indicated a Faraday or Murchison Island origin, with these samples closely clustering with Faraday and Murchison Island projected GT-seq samples. Population assignment of the Agglomerate, Hotspring, and Ramsay Island individuals strongly indicated a Faraday/Murchison Island origin. # Discussion Invasive rats are having devastating impacts on biodiversity globally, necessitating the development of innovative tools for informing IAS management. Native species, especially seabirds, are unlikely to adapt to the invasive threat in time to prevent extirpation. Without intervention, these losses could result in large scale changes at the ecosystem level. Given the isolation of the Haida Gwaii archipelago, seabirds play a vital role in nutrient cycling as significant transporters of nutrients from off-island via guano deposition. Disruption of this transport chain could lead to increased depletion of resources on the islands and whole-scale regime shifts, a consequence of rat invasions observed in other island systems around the globe. The effects of removing seabirds from Haida Gwaii are not purely ecological; seabirds, such as the rat-threatened ancient murrelet (*Synthliboramphus antiquus*), also have a significant cultural importance to the Haida Nation. To promote seabird recovery in Haida Gwaii, invasive rats must be eradicated. Recent work has shown the utility of genetic data for informing invasive rat management in Haida Gwaii. By harnessing the power of next-generation sequencing and ddRAD, previous work characterized population structure and connectivity among both brown and black rats throughout the archipelago, laying the foundation for robust inference of population origin for nearly any novel invasion. To demonstrate application, this earlier study revealed that failed eradications on the Bischof Islands were due to re-invasion from nearby Lyell Island and not from incomplete removal \[*i*.*e*., the survivor hypothesis; 25\]. The mechanism of reinvasion was likely via swimming given the close geographic proximity (\<500m) to Lyell Island. “Rafting” during high tide and/or commensal spread via ships remain possibilities. Additionally, the origin of novel invasions on Faraday and Murchison Islands was also identified as neighboring Lyell Island. These results not only highlighted the need for potentially increased biosecurity between islands in Haida Gwaii, but also provided insights into the dispersal capacity of brown rats through temperate waters. This initial population genomic analysis was achieved using ddRAD, a reduced representation genome sequencing approach that allows for the genotyping of dozens or more individuals at a large number of SNPs (\>10,000) in a single lane of Illumina sequencing. Yet, this method requires labour-intensive library preparation, where speed and cost-effectiveness do not scale linearly for smaller sample sizes. These considerations are of the utmost importance, as novel rat invasions often occur with relatively few individuals; in some cases, only a single individual may be captured for analysis when populations are at low densities (for example, on Hotspring and Ramsay Islands). In addition, ddRAD requires substantial bioinformatic processing and filtering of data post- sequencing to ensure genotype calls are accurate and not the result of sequencing error, sample contamination, or other confounding factors. While ddRAD does afford connectivity between datasets, these bioinformatic processing steps must be repeated for inclusion of any new individuals. Taken together, ddRAD and similar approaches are not ideal when applied to a single or few samples, especially when the required information is time-sensitive. Rat invasions, in particular, have been likened to other acute impacts to biodiversity such as wildfires or oil spills that require rapid and decisive responses based on timely information. GT-seq overcomes these limitations, providing a more expedient and cost- effective approach for generating needed genetic information, especially when involving a single or few high priority samples. To start, GT-seq library preparation is highly simplified compared to ddRAD. While ddRAD has many time- consuming purification and size selection steps along with multiple ligations and PCR steps, GT-seq only requires two PCRs and a final library purification. Consequently, GT-seq libraries can be prepared in a matter of hours rather than days or weeks as may be the case with ddRAD. Library preparation for GT-seq uses few reagents and a commercially available PCR kit, whereas ddRAD library preparation requires a fuller suite of more costly reagents including restriction enzymes and ligases. Moreover, bioinformatic processing and analyses of raw data are also much more streamlined in GT-seq, which uses open-source software (<https://github.com/GTseq/GTseq-Pipeline>) to simultaneously identify locus-specific primers and genotype the corresponding SNP using minimal computational resources (*e*.*g*., standard laptop). These genotypes require no additional filtering and are immediately suitable for use in population genetic analyses, further positioning GT-seq as an effective rapid response tool. While GT-seq does offer many advantages over other approaches (*e*.*g*., reduced representation sequencing, including ddRAD), panel development does require existing genomic data for locus identification and primer design. For this study, we had previously genotyped brown and black rats across their distributions in Haida Gwaii, as well as characterized population structure among island populations. This previous work not only provided substantial genomic data for GT-seq panel design, but also allowed for an explicit evaluation of the ability of the panel to identify known population structure based on ddRAD results. While we were fortunate in this regard, many systems are lacking these foundational data. In such systems, prior genetic work will be needed to best inform GT-seq panel development; however, once completed, the full advantages of GT-seq over more costly and labour-intensive methods can be exploited. Moving forward, GT-seq and other targeted amplicon sequencing approaches hold great promise for informing conservation and management of biodiversity. Originally demonstrated in steelhead trout, GT-seq has been subsequently applied to the study of other fish species, including redband trout, brook trout, Pacific lamprey, and walleye. More recently, the efficacy of using GT-seq for genotyping minimally-invasive samples has been demonstrated, and successfully applied to further understanding of the molecular ecology and conservation status of the at-risk Western rattlesnake. To the best of our knowledge, our study is the first to demonstrate use of GT-seq to inform invasive species management to help mitigate biodiversity loss. ## Panel efficacy Across both species-specific panels, we were able to accurately and consistently detect previously identified population structure among invasive rat populations in Haida Gwaii. Principle component analysis resulted in near identical clustering when compared to the previous ddRAD study (Figs). We also identified genetic structure consistent with previous results, though there was a slight loss in resolution of fine-scale structure among highly connected populations (Figs) likely due to the reduced number of SNPs in the GT-seq dataset. Both brown and black rats are highly vagile species; significant connectivity and gene flow can occur among island populations separated by as much as 1 km of ocean, and terrestrial populations can remain connected at distances of \>10 km, resulting in less differentiation and more admixture. Using large SNP datasets (≤ 25,686 loci), Sjodin *et al*. were able to detect this fine-scale structure even among minimally differentiated populations. To detect this same structure using our reduced GT-seq panel, targeted SNPs likely will need to be added to further maximize among-cluster variation for these admixed clusters, especially for brown rats sampled on Lyell, Tanu, and Richardson Islands, and for both species with sampling distributions across larger islands like Graham and Kunghit Islands. Nevertheless, these panels still adequately captured among- population variation across both species capable of correctly assigning individuals to their true populations. Furthermore, as GT-seq is a targeted amplicon sequencing approach, this method is highly replicable and completely connectible, so additional reference populations can always be added to the database to allow for even more robust inferences. The ability to cost- effectively genotype individuals is also conducive to applications such as long- term monitoring to better estimate contemporary gene flow and inform biosecurity measures. These estimates would be particularly useful for monitoring brown rat populations on Lyell and surrounding islands, as these populations appear to be rapidly expanding and are currently the highest biosecurity threat in terms of seabird conservation. ## Novel invasions We found strong support for a Faraday and/or Murchison origin for novel invasions on Agglomerate, Hotspring, and Ramsay Islands. The Ramsay Island invasion is particularly concerning as this island hosts the most significant remaining seabird colony in the entire archipelago. Rat invasions have already led to the complete, or near complete, extirpation of large seabird colonies in Haida Gwaii, such as on Langara, Kunghit and Lyell Islands; establishment of a rat population on Ramsay Island will likely lead to the same devasting consequences for extant seabird populations. Prior to this study, we investigated the origin of the incursion onto Hotspring Island using ddRAD (Sjodin & Russello, unpublished report). These results also indicated a Faraday/Murchison Island origin and provided information to support an eradication on Hotspring Island in November 2018 with the goal of preventing spread to Ramsay Island. But for reasons discussed above, ddRAD, while accurate, is not suitable for a single or few individuals as resources and processing time do not scale down linearly for small sample sizes. For island eradications and control of incursions, rapid response is critical and targeting the response at an appropriate scale is crucial to the durability of the eradication. Eradicating the rats on Hotspring Island kept Ramsay Island rat-free, but only for a year; rats subsequently re-invaded Hotspring Island, which likely acted as a stepping-stone leading to recent detections on Ramsay Island (Wojtaszek, pers. comm.). This outcome indicates that the management response to the initial incursions on Hotspring Island was conducted at an inappropriate geographic scale; the distance from Ramsay Island that needed to remain rat free to provide some eradication durability was greater than just the most adjacent islands (Wojtaszek, pers. comm.). Armed with information from the RapidRat GT-seq panel, future initiatives could determine whether all islands within a certain area should be considered a single eradication unit or if a more localized response is appropriate. These data can also point to mechanisms of invasion, which can guide how management agencies deal with the incursion. For example, if the rat on Hotspring Island had come from another location within the archipelago via a vessel moored on the north side of the island (*e*.*g*., human-mediated), rather than via natural dispersal from adjacent islands, the response would be scaled, and biosecurity measures enhanced. # Conclusions Here, we developed, validated and deployed an effective and efficient genotyping tool (RapidRat) for informing invasive rat management in Haida Gwaii. It is important to highlight, however, that GT-seq panel development was greatly facilitated by the existence of an archipelago-wide SNP dataset previously collected via ddRAD. When no such data are available, an initial population genomic analysis would most likely need to be conducted. Moreover, we initially designed RapidRat to be informative for both invasive brown and black rats, likely limiting the resolution at the single species level. Although, this does not appear to be an issue for informing invasive rat management in Haida Gwaii, independent, species-specific panels could be developed using a similar workflow as described here. The potential multi-species ascertainment bias also likely translates to a finer-level; for example, we do not expect RapidRat to be informative in other systems given that locus selection was solely based on among-population genetic variation in Haida Gwaii. Nevertheless, invasive rats constitute a global threat to biodiversity and innovative tools are required to inform management action and minimize impact on novel ecosystems. RapidRat demonstrates the applicability of GT-seq as one such IAS rapid response tool. Importantly, the framework demonstrated here for panel development and validation can be directly applied for informing invasive rat management and biodiversity conservation in other systems. # Supporting information We thank Parks Canada Agency employees from Gwaii Haanas National Park Reserve, National Marine Conservation Area Reserve and Haida Heritage site for the collection of the field samples. We are particularly grateful to Danielle Schmidt and Nathan Campbell for providing support with GT-seq panel development and optimization. 10.1371/journal.pone.0234694.r001 Decision Letter 0 Silva Daniel de Paiva Academic Editor 2020 Daniel de Paiva Silva This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 1 Apr 2020 PONE-D-20-00793 RapidRat: development, validation and application of a genotyping-by-sequencing panel for rapid biosecurity and invasive species management PLOS ONE Dear Dr. Russello, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. We would appreciate receiving your revised manuscript by June 30th 2020. When you are ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission- guidelines#loc-laboratory-protocols> Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. We look forward to receiving your revised manuscript. Kind regards, Daniel de Paiva Silva, Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 1\. Thank you for including the following ethics statement: 'All sample collection was performed under Parks Canada Agency Animal Care Committee protocol GHNPR11-5. Parks Canada is the regulatory agency that permits field research in the study area and were the sole participants involved in sample collection.' Please specify the ethics statement in the Methods section of your manuscript and ensure that your named ethics committee specifically approved this study. Once you have amended this/these statement(s) in the Methods section of the manuscript, please add the same text to the “Ethics Statement” field of the submission form (via “Edit Submission”). For additional information about PLOS ONE ethical requirements for animal research, please refer to <http://journals.plos.org/plosone/s/submission- guidelines#loc-animal-research> 2\. Thank you for including your competing interests statement;"The authors have declared that no competing interests exist." We note that author Michael Russello is an Academic editor for PLOS ONE Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, by including the following statement: "This does not alter our adherence to  PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors [http://journals.plos.org/plosone/s/competing-interests](about:blank)).  If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared. Please include your updated Competing Interests statement in your cover letter; we will change the online submission form on your behalf. Please know it is PLOS ONE policy for corresponding authors to declare, on behalf of all authors, all potential competing interests for the purposes of transparency. PLOS defines a competing interest as anything that interferes with, or could reasonably be perceived as interfering with, the full and objective presentation, peer review, editorial decision-making, or publication of research or non-research articles submitted to one of the journals. Competing interests can be financial or non-financial, professional, or personal. Competing interests can arise in relationship to an organization or another person. Please follow this link to our website for more details on competing interests: [http://journals.plos.org/plosone/s/competing-interests](about:blank) 3\. We note that you have stated that you will provide repository information for your data at acceptance. Should your manuscript be accepted for publication, we will hold it until you provide the relevant accession numbers or DOIs necessary to access your data. If you wish to make changes to your Data Availability statement, please describe these changes in your cover letter and we will update your Data Availability statement to reflect the information you provide. 4\. Additional Editor Comments (if provided): Dear Mr. Russello, after carefull reviews by two independent researchers, they found that your manuscript is nearly to be accepted. Nonetheless, they suggested that minor reviews are performed before it is formally accepted in the journal, especially reviewer \#2, who provided some suggetions in an attached file. Therefore, considering the whole COVID-19 pandemic, I hope you are able to provide a reviewed version of the manuscript by June 30th 2020. Best regards, Daniel Silva \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The paper presents novelty and describes the way the authors obtained genomic data to evalute invasive species. The authors developed, validated and deployed an effective and efficient genotyping tool (RapidRat) for informing invasive rat management in Haida Gwaii. To improve the paper I recomend: a\) Make the results clear in the abstract section b\) Introduction and material and methods: are Ok. c\) Results: ok d\) Discussion and conclusions: Improve them to clarify the novelty, importance and the limitations of such research Reviewer \#2: The manuscript entitled “RapidRat: development, validation and application of a genotyping-by-sequencing panel for rapid biosecurity and invasive species management” describes the development of a rat specific panel for genomic data to access information about invasive species of rats in an archipelago in British Columbia, Canada. The authors proposed a novel and cheaper way to genotype individuals of brown and black rats using the detection of SNPs polymorphisms to track novel and old rat invasions in the islands. I recommend this manuscript to be accepted for publication in PlosOne after the review of the suggestions pointed out in the attached file. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0234694.r002 Author response to Decision Letter 0 1 May 2020 Please see attached response letter 10.1371/journal.pone.0234694.r003 Decision Letter 1 Silva Daniel de Paiva Academic Editor 2020 Daniel de Paiva Silva This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 2 Jun 2020 RapidRat: development, validation and application of a genotyping-by-sequencing panel for rapid biosecurity and invasive species management PONE-D-20-00793R1 Dear Dr. Russello, We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements. Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication. Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at <https://www.editorialmanager.com/pone/>, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. With kind regards, Daniel de Paiva Silva, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#2: All comments have been addressed Reviewer \#3: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#2: Yes Reviewer \#3: No \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#2: (No Response) Reviewer \#3: This article deals with the development of a panel of SNPs in a genotyping-by-sequencing strategy (GT-seq) and its possible application in the management of invasive species of rats in Canada. A review round has already been carried out. Although the authors have made a substantial improvement in the manuscript, I recommend some minor adjustments before being accepted for publication in PlosOne. Introduction: ok Material and Methods: 1\) Information about the ddRAD-seq sequencing data is missing. Which sequencing platform was used to generate this data? How many RAD-tags were generated? What type of sequencing read was generated? Is this data deposited in a database? What is the mean read length? The RAD-tags represent which portion of the total size of the genome used as a reference? 2\) I believe that "Study design" is not a good topic name for methods (line 131). Something like "origin of SNPs" or "SNP loci information from ddRAD-Seq" is perhaps more appropriate. After all, the other topics of the methodology are also part of the "study design". Results: Ok Discussion: Ok Conclusion: Ok Figures and tables: As mentioned by reviewer \# 2 in the previous review, the figures are still in low resolution. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#2: No Reviewer \#3: Yes: Rhewter Nunes 10.1371/journal.pone.0234694.r004 Acceptance letter Silva Daniel de Paiva Academic Editor 2020 Daniel de Paiva Silva This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 10 Jun 2020 PONE-D-20-00793R1 RapidRat: development, validation and application of a genotyping-by-sequencing panel for rapid biosecurity and invasive species management Dear Dr. Russello: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Daniel de Paiva Silva Academic Editor PLOS ONE [^1]: We note that author Michael Russello is an Academic editor for PLOS ONE; this does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction Corneal epithelial stem cells are located in the basal layer of the limbus, which is the narrow transition zone between the cornea and the conjunctiva. The limbal epithelium is a reservoir for replacing corneal epithelial cells that are normally continuously lost from the corneal surface. Severe corneal diseases, such as Stevens–Johnson syndrome, or chemical burns destroy the limbus and cause limbal stem cell deficiency (LSCD). In these cases, corneal epithelial cell sources are exhausted, the peripheral conjunctival epithelium invades inwardly, and the corneal surface becomes enveloped by vascularized conjunctival scar tissue, which results in corneal opacification that leads to severe visual impairment. In cases of severe LSCD, we and others recently demonstrated the successful application of constructs involving ex vivo expansion of autologous oral mucosal epithelium. This method averts the risks of immune rejection and long-term immunosuppression, and thus offers clinical advantages over conventional allogeneic corneal transplantation. We have performed transplantation of oral mucosal epithelial cell sheets in over 20 patients. For these patients, corneal transparency was restored and postoperative visual acuity remained improved for 2–8 years, whereas abnormal corneas were successfully reconstructed using conventional allogeneic transplantation in only 20–30% of patients for 2–3 years. Cultured oral mucosal epithelial cell sheets contain stem/progenitor cells, as demonstrated by colony-forming assays (CFAs) and immunohistochemistry. Clinically successful long-term reconstruction after cell sheet transplantation suggests that these transplanted stem/progenitor cells are maintained in vivo postoperatively. Although a few studies have investigated the existence of stem/progenitor cells in reconstructed cornea, this has yet to be established. Thus, in this study, we assessed the maintenance and distribution of epithelial stem/progenitor cells after corneal reconstruction using oral mucosal epithelial cell sheets in a rat model. # Materials and Methods ## 2.1. Primary culture of GFP rat oral mucosal epithelium Animals were treated in accordance with the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. Our experimental procedures were approved by the Committee for Animal Research of Osaka University Graduate School of Medicine. We created cultured cell sheets fabricated from GFP rat oral mucosal epithelial cells and transplanted it onto the eyes of a nude rat limbal stem cell deficiency model. Oral mucosal biopsy specimens (2 mm radius) were excised from 4 green fluorescent protein (GFP) rats (“green rat CZ-004,” SD TgN (act-EGFP) OsbCZ-004; Japan SCL, Inc., Shizuoka, Japan). Each rat weighed 200 g. Anesthesia was induced by intraperitoneal administration of ketamine hydrochloride (25 mg/kg) and xylazine hydrochloride (10 mg/kg). Biopsy specimens were incubated at 4°C for 4 h with Dispase II (Roche Diagnostics GmbH, Mannheim, Germany) and treated with trypsin/EDTA solution (Invitrogen, Carlsbad, NM) at room temperature for 20 min. Cell suspensions were cultured on temperature- responsive culture inserts (CellSeed Inc., Tokyo, Japan) at an initial density of 4×10⁵ cells/23-mm insert along with mitomycin C (MMC)-treated NIH/3T3 cells that were separated by these cell culture inserts in the keratinocyte culture medium (KCM) (Dulbecco’s modified Eagle’s medium \[DMEM\]/F12 \[3∶1\] supplemented with 10% fetal bovine serum \[Japan Bio Serum, Hiroshima, Japan\], 0.5% Insulin–Transferrin–Selenium \[ITS; Invitrogen, Carlsbad, CA\], 10 µM isoproterenol \[Kowa, Tokyo, Japan\], 2.0×10⁻⁹ M triiodothyronine \[MP Biomedicals, Aurora, OH\], 0.4 µg/mL hydrocortisone succinate \[Wako, Osaka, Japan\], and 10 ng/mL EGF \[R&D Systems, Minneapolis, MN\]). Five days later, oral epithelial cells achieved confluence. After an additional 5–7 days of culture, the resulting cell sheets were harvested by reducing the culture temperature to 20°C for 30 min. ## 2.2. Transplantation of cultured oral epithelial cell sheets A limbal stem cell deficiency model was generated in one eye each of anesthetized nude rats (F344/NJcl-rnu/rnu) by excising all corneal epithelial and limbal tissues (N = 4). Three weeks after surgery, conjunctival scar tissue with some neovascularization covered the entire corneal surface and invaded into the stroma. Prior to cell sheet transplantation, the conjunctivalized ocular surface was surgically removed to re-expose the corneal stroma. Tissue- engineered culture cell sheets, fabricated ex vivo from GFP rat oral mucosal epithelial cells, were harvested and transplanted over the stromal bed. For healing protection, tarsorrhaphy was performed after transplantation. Antibiotics and steroids were topically applied postoperatively 3 times daily. The eyes were carefully observed using a slit lamp biomicroscope and a fluorescence stereomicroscope. Eight weeks after surgery, the rats were sacrificed with an overdose of anesthetic agent (pentobarbital), and their eyes were enucleated for histology examinations and CFA. ## 2.3. Immunofluorescence and histological examinations Oral mucosal epithelium, tissue-engineered epithelial cell sheets, and reconstructed corneas were assessed using immunofluorescence examinations. Cryosections (thick, 10 µm) were treated with 5% bovine serum albumin (BSA) in 50 mM Tris-buffered saline (TBS; pH 7.2) containing 0.4% Triton X-100 at room temperature for 60 min. The sections were then incubated overnight at 4°C with primary antibodies diluted with 1% BSA in PBS containing 0.4% Triton X-100. Primary antibodies included mouse monoclonal anti-p63 (4A4; Santa Cruz Biotechnology Inc., Santa Cruz, CA), mouse monoclonal anti-K14 (CKB1; Abcam, Tokyo, Japan), rabbit polyclonal anti-CD31 (Abcam), and Alexa Fluor 495 or 555-labeled secondary antibodies (Jackson ImmunoResearch Laboratories Inc., West Grove, PA) were used. All sections were counterstained with Hoechst 33342 and observed under a fluorescence microscope (Axiovert 200; Carl Zeiss, Oberkochen, Germany). Transplanted corneal sections were also conventionally stained with hematoxylin and eosin (HE) and observed under a light microscope (BX50; Olympus, Tokyo, Japan). ## 2.4. CFA We used CFA to assess if there were putative progenitor cell populations in the biopsied, cultured, and transplanted cells. Primary cells isolated from GFP rat oral epithelial tissues were seeded in untreated 6-well culture plates for CFAs. Secondary cells isolated from tissue-engineered cell sheets by trypsin digestion were also used for CFAs. Transplanted epithelial cells from reconstructed corneas were also used for CFAs as follows. Corneoscleral tissues were excised from enucleated eyes 8 weeks after transplantation. As assessed using the fluorescence stereomicroscope, all of the epithelium that covered a cornea expressed GFP. The central areas (diameter, 2 mm) and peripheral areas (width, 1 mm) of the corneas were excised and treated with Dispase II at 37°C for 1 h. Epithelial cells were then separated and treated with 0.25% trypsin/EDTA solution at 37°C for 20 min to create single cell suspensions. Cells at a density of 3×10³ cells/well were used for CFAs for biopsied, cultured, and transplanted epithelial cells along with MMC-treated 3T3 feeder cells in 6-well culture plates. After 12 days in culture, the cells were fixed and stained with rhodamine B. The colony-forming efficiency (CFE) of primary, cultured, and transplanted cells was determined by dividing the number of colonies per well by the total number of seeded cells in each well (N = 4, duplicates used for each sample). ## 2.5. Statistical Analysis Results are presented as mean ± SEs. Data were analyzed using t tests; p\<0.05 was considered statistically significant. All statistical analyses were carried out using JMP version 9.0.3. # Results To monitor the cell fates of transplanted cell sheets, we prepared epithelial cell sheets fabricated from GFP rats. Phase contrast microscopy showed that oral mucosal epithelial cells obtained from the GFP rats proliferated and became stratified after culture for 10 days. These cells showed tight, dense packing on culture inserts as well as a cobble stone-like cell morphology, and fluorescence microscopy showed that all of these cells were GFP positive. Epithelial cell sheets were also evaluated on sections. The epithelial cell sheets were well stratified with 2–3 layers of GFP-positive epithelial cells. Immunohistochemistry results showed that the basal cells of cultured epithelial cell sheets expressed p63, a putative epithelial stem cell marker (Fig, 2D). The mean (±SE) CFE values for cells from the primary oral mucosa and cultured cell sheets were 3.17±0.67% and 2.12±0.68%, respectively, and both primary oral mucosa and cultured cell sheets contained sufficient numbers of progenitor cells. Eight weeks after transplantation, a slit lamp photograph showed that the surface of a cornea was completely covered with epithelium. HE staining of a transplanted corneal section showed that 3–4 cell layers of epithelium were reconstructed. Immunostaining results showed that a transplanted cornea expressed p63 in the basal layers and K14 in all epithelial layers. In the peripheral region obvious neovascularization was observed in slit lamp examination. The presence of neovascularization was confirmed by the immunohistochemistry of CD31. The transplanted cell sheet successfully reconstructed the corneal surface, and the entire cornea was covered by the GFP- positive multilayered epithelium. Cross-section analysis also demonstrated that the GFP-positive cells were on a reconstructed cornea. To compare the distributions of putative epithelial stem/progenitor cells in the reconstructed corneas with those in the normal corneas, we used CFAs for epithelial cells harvested from both peripheral and central areas of the corneas in cell sheet-transplanted eyes. The reconstructed cornea contained colony- forming cells, and the mean (±SE) CFE values for the peripheral and central epithelial cells removed from the transplanted corneas were 4.67±1.53% and 0.41±0.26%, respectively, which results were significantly different (p = 0.036; ). In normal rats, the mean CFE value for the peripheral cornea was significantly higher than that for the central cornea (2.78±0.53% vs. 0.73±0.50%, p = 0.024;). # Discussion The aim of this study was to examine the maintenance and distribution of epithelial stem/progenitor cells after corneal reconstruction using oral mucosal epithelial cell sheets in a rat model. Our findings indicate that cultivated oral mucosal epithelial cell sheet survives and contains putative epithelial stem/progenitor cells after transplantation. In addition, epithelial stem/progenitor cells are maintained abundantly in peripheral cornea, which shows a similar distribution pattern of stem cells to normal eyes. When we performed autologous cell sheet transplantation in LSCD patients, it was difficult to distinguish whether the transplanted cell sheets survived, because these sheets were derived from the patients themselves. To resolve this problem, we used cultured cell sheets fabricated from GFP rats in this study. Using this method, the presence of GFP-positive cells after cell sheet transplantation in reconstructed corneas established that the transplanted cell sheets survived postoperatively. We successfully generated the oral mucosal epithelial cell sheets from GPF rat. The epithelial cell sheets expressed p63 in the all of the basal layers and contained sufficient numbers of progenitor cells. Based on these results, we determined that these rat oral mucosal epithelial cell sheets were of sufficient quality for our subsequent transplant experiments. After cell sheet transplantation, the entire cornea was covered by the GFP- positive cells, and they also expressed p63 in the basal layer and K14 in all epithelial layers. These findings suggested that the transplanted cell sheets survived and retained stem/progenitor cells for at least 8 weeks postoperatively. These results strongly support a hypothesis that stem/progenitors cells can survive after this treatment, which results in the long-term success of transplantation of cultured oral mucosal cell sheets in LSCD patients. The CFE values for the peripheral epithelial cells removed from the reconstructed cornea were significantly higher than that for the central cells. This non-uniform pattern over the cornea indicates that, after transplantation, stem cell precursors are predominantly in the peripheral cornea compared with the central cornea. This suggests that stem cell progenitor maintenance is not cell autonomous, possibly owing to different microenvironments. It is interesting that this is also true in normal corneas. The mechanism underlying the enrichment of progenitors in peripheral regions is not entirely clear. It may be due to the peripheral vascularization. The origin of the cell sheet is oral mucosa, a vascularized tissue, inducing neovascularization. We and others reported cultivated oral mucosal cell sheet contains highly level of angiogenic factors, such as basic Fibroblast growth factor or Thrombospondin 1, compared to cultivated corneal epithelial cell sheet. In fact, peripheral neovascularization occurs frequently in human clinical application of oral mucosal cell sheet transplantation reported by us and others,. These neovascularization might contribute to maintain the stemness. For example, in the bone marrow, several reports have shown that vascular endothelial cells regulate hematopoietic stem/progenitor cells through the production of specific paracrine growth factors. Kobayashi et al. demonstrated that the endothelial cells modulates reconstitution of hematopoietic stem/progenitor cells through the modulation of angiocrine factors with Akt- mTOR-activated endothelial cells supporting the self-renewal and expansion of hematopoietic stem/progenitor cells. Recent studies also demonstrated endothelial cells regulate stem/progenitor cell niche in the central nervous system, and adipose tissues. Regarding the corneal epithelium, there is an indirect evidence of vascular niche that limbal niche cells have the angiogenesis potential and prevent corneal epithelial stem/progenitor cells differentiation. In conclusion, we have shown that oral mucosal epithelial cell sheets can survive and contain putative epithelial stem/progenitor cells after transplantation. Because stem/progenitor cell maintenance in transplanted cell sheets is the key for a good prognosis, the results of our study are a first step toward understanding the behavior of these precursor cells in transplanted tissue. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: TS RH MT KN. Performed the experiments: TS RH HS SK. Analyzed the data: TS RH HS. Contributed reagents/materials/analysis tools: TS RH HS SK YO. Contributed to the writing of the manuscript: TS RH HS MT YO KN.

# Introduction Dysphagia occurs in acute stroke patients at high rates, and many of these patients develop aspiration pneumonia. Pneumonia incurs extended hospitalization and decreases the rate of hospital discharge. Therefore, the early intervention in dysphagia is important to prevent aspiration pneumonia in acute stroke patients. Recently, major two clinical trials were reported to assess the effectiveness of antibiotic prophylaxis for reducing pneumonia in acute stroke patients. Unfortunately, antibiotic prophylaxis cannot effectively prevent of post-stroke pneumonia in both two studies. It is now uncertain what way is the effective to reduce pneumonia in acute stroke patients. Team approaches with the cooperation of various professionals have the power to improving the quality of medical care, utilizing the specialized knowledge and skills of each profession. Thus, we consider that multidisciplinary team approaches have the potential for reducing pneumonia onset in acute stroke patients. However, few studies demonstrate that the multidisciplinary participatory swallowing team approach significantly decreases the rate of pneumonia in acute stroke patients. In our hospital, a multidisciplinary participatory swallowing team was organized in April 2011. The aim of this study was to clarify the influence of the team approach on dysphagia by comparing the rates of pneumonia in acute stroke patients between the prior and post team periods. # Materials and Methods All consecutive acute stroke patients who were admitted to our hospital between April 2009 and March 2014 were registered. A multidisciplinary participatory swallowing team was organized in April 2011. Thus, we defined the period before team organization (April 2009 to March 2011) as the ‘prior period’ and the period after team organization (April 2011 to March 2014) as the ‘post period’. The patients in the prior period were analyzed as a historical control. The multidisciplinary participatory swallowing team in our hospital consists of 9 professionals, including doctors, dentists, nurses, physical therapists, occupational therapists, speech therapists, managerial dieticians, dental hygienists, and pharmacists. Two neurologists reviewed the charts of all consecutive acute stroke patients blind the period of admission. Clinical parameters (age, sex, vascular risk factors \[hypertension, diabetes mellitus, and dyslipidemia\], smoking, previous stroke, stroke severity, and stroke subtype), body temperature, laboratory findings (C-reactive protein \[CRP\] and white blood cell \[WBC\] count), and radiological findings (chest X-ray, computed tomography \[CT\], and magnetic resonance imaging \[MRI\]) were recorded for all patients during hospitalization. Hypertension was defined as the use of anti-hypertensive medicines prior to admission or a confirmed blood pressure ≥140/90 mmHg at rest 2 weeks after stroke onset. Diabetes mellitus was defined as an HbA1c of ≥6.5%, a fasting blood sugar ≥126 mg/dl, or the use of anti-diabetic medicines. Hyperlipidemia was defined as a total cholesterol ≥220 mg/dl, a low-density lipoprotein cholesterol ≥140 mg/dl at admission, or the use of anti- hyperlipidemia medications. Smoking was defined according to the definition of the US Centers for Disease Control and Prevention as follows: (1) never smokers, who had never smoked a cigarette or who smoked fewer than 100 cigarettes in their entire lifetime; (2) former smokers, who had smoked at least 100 cigarettes in their lifetime, but said they currently did not smoke; and (3) current smokers, who have smoked 100 cigarettes in their lifetime and currently smoke cigarettes every day (daily) or some days (nondaily). The diagnosis of clinically defined pneumonia was based on the criteria of the Centers for Disease Control and Prevention as follows: Clinically defined pneumonia criteria require the presence of a new and persistent infiltrate or consolidation on at least 1 chest X-ray or CT with one of the following clinical signs: fever, leukopenia or leukocytosis and altered mental status in more than 70-year-olds in the absence of other causes. These should be added to 2 of the following signs: new-onset purulent sputum or change in the character of the sputum, new-onset or progressive cough, rales, and impaired gas exchange. We evaluated the difference in the rate of pneumonia onset between the prior and post team organization periods as well as the factors that influenced pneumonia onset. All patients underwent head CT or MRI. The stroke subtype was determined based on the CT or MRI findings, electrocardiography, and carotid artery and cardiac ultrasound findings by at least two stroke specialists according to the Trial of Org 10172 in Acute Stroke Treatment (TOAST) classification. The neurological severity was evaluated using the National Institutes of Health Stroke Scale (NIHSS) score at admission. This study was approved by the Institutional Review Board (IRB) of Hiroshima university hospital (E-144). All clinical investigation must have been conducted according to the principles expressed in the Declaration of Helsinki. Because the data were analyzed anonymously, no informed consent was given. Statistical analyses were performed using the JMP 12.0.1 statistical software (SAS Institute, Inc., Cary, North Carolina, USA). The data are presented as the mean ± standard deviation (SD) or the median (minimum–maximum) for continuous variables. Statistical analyses of the comparisons of the two groups were performed using Student’s t-test or the Mann-Whitney U test for continuous variables and the chi-square test or Fisher’s exact test for categorical variables. Two tail p-values \<0.05 were considered statistically significant. Univariate and multivariate analyses were performed using a Cox proportional- hazards model to determine the predictors of pneumonia. The Cox proportional hazard model was used to estimate the relative risk (hazard ratio, HR) and the 95% confidence interval (CI). The cumulative incidences of time to the onset of pneumonia were estimated using the Kaplan-Meier method. The cumulative incidence curves for the two groups (patients in the prior period vs. patients in the post period) were compared using a log-rank test. # Results We recruited 132 acute stroke patients from the prior period (April 2009 to March 2011) and 173 patients from the post period (April 2011 to March 2014). Age, sex, vascular risk factors, NIHSS score on admission, and stroke subtype did not significantly differ between the two groups. No patients received prophylactic antibiotics. The rates of patients having a fever with temperatures greater than 38°C did not significantly differ between the two groups, but the rates of patients with an increasing WBC count and CRP were significantly reduced in the post period compared with the prior period. Pneumonia onset was less frequent in the post period compared with the prior period (6.9% vs. 15.9%, respectively; p = 0.01). Using an univariate analysis with a Cox proportional- hazards model, it was determined that NIHSS score on admission and the application of a swallowing team approach were significantly related to pneumonia onset. Using a multivariate analysis with the factors selected in univariate analyses as statistically significant, it was determined that NIHSS score on admission and the application of a swallowing team approach were independently related to pneumonia onset. To understand the difference of swallowing dependent approaches between the prior and post period, frequencies of professional oral care and swallowing evaluations (videoendoscopic examination of swallowing \[VE\] or videofluoroscopic examination of swallowing \[VF\]) were evaluated. The rates of patients receiving professional oral care and swallowing evaluations were significantly increased in the post period compared with the prior period. # Discussion In this study, we demonstrated that the multidisciplinary participatory swallowing team approach effectively decreases the onset of pneumonia in acute stroke patients. Pneumonia after stroke has been implicated in morbidity, mortality, and increased medical costs after acute stroke. Thus, the prevention of pneumonia after an acute stroke is of great importance. Several previous studies reported that stroke severity on admission, dysphagia, and old age are major independent risk factors for the onset of pneumonia after stroke. In our study, NIHSS score on admission and the application of a swallowing team approach were independently related to the onset of pneumonia. The factor of stroke severity on admission is common to both previous studies and our study. Old age was not related to the pneumonia in our study. No studies demonstrate that the multidisciplinary participatory swallowing team approach significantly decreases the rate of pneumonia in acute stroke patients. Therefore, we believe that our results provide great value in improving acute stroke management. There are several possibilities why the multidisciplinary participatory swallowing team approach significantly decreases the rate of pneumonia. The first reason is that an increasing number of patients received professional oral care by dentists or dental hygienists. In the period before the team organization, nurses performed routine oral care to all patients, but the rate of patients receiving professional oral care was only 12.9%. However, the rate of patients receiving professional oral care was raised up to 51.7% in the period after team organization. Several previous studies demonstrated that professional oral care appears to reduce the incidence of pneumonia in elderly individuals but not in acute stroke patients. The second reason is that an increasing number of patients received a swallowing evaluation (VE or VF). The rate of patients receiving a swallowing evaluation was 12.1% in the period before the team was organized but 26.0% in the period after the team was organized. The third reason is that we were able to provide managerial dieticians who created appropriate dysphagia diets and nutritional supplements based on the general condition and neurological prognosis of the patients. One report suggests that under-nutrition is independently related to post-stroke complications (including pneumonia) and clinical outcomes. Therefore, an improvement in nutritional management may decrease the rate of pneumonia. The fourth reason is that physical therapists and occupational therapists improved patient body positions at mealtimes. Acceptable body positions at mealtimes are important to prevent aspiration. We think that an improvement in body position may also contribute to decreasing of pneumonia. Additionally, by organizing the multidisciplinary participatory swallowing team, each profession could communicate very smoothly and share the detailed information of all stroke patients early in the admission. We consider that this face-to-face communication led the intervention of each profession with proper timing and contributed to decreasing pneumonia onset. Before the team organization, the needs to intervene of each profession were decided by only the attending doctors’ assessments. However, the multidisciplinary participatory swallowing team offered a variety of perspectives on the patients and resulted in the improvement of the quality of acute stroke care. All patients before team organization were received oral care by floor nurses including the certificated dysphagia nurse. The attending doctors provided swallowing evaluation of each patient by modified water swallow test at bedside, and decided the food form. When the attending doctors estimated that the patient was needed to receive professional oral care or swallowing evaluations (videoendoscopic examination of swallowing \[VE\] or videofluoroscopic examination of swallowing \[VF\]), dentists and speech therapists intervened in the patients. All patients were received general acute stroke rehabilitation by physical therapists and occupational therapists. Basically, the acute treatment and other general medical care were determined based on the stroke subtype in accordance with the same established guideline (the Japanese Guidelines for the Management of Stroke 2009) both in historical control and study cohorts. The practices from each profession were not different between the prior and post period, although proportions of patients who got professional oral care or swallowing evaluations were substantially higher in the post period. Their increases may come from face-to-face professional communications. Various previous studies reported that pneumonia during hospitalization in acute stroke patients is related to an increasing length of hospital stay. In our study, the length of hospital stay was significantly longer in patients with pneumonia compared with patients without pneumonia (36.5±22.2 days vs. 20.2±11.6 days, p\<0.0001). However, the length of hospital stay did not significantly differ between patients in the prior period and patients in the post period. We attribute this lack of difference to the fact that the rate of patients with pneumonia was relatively low in both groups. Thus, the influence of pneumonia on the length of hospital stay was limited. Some medications are known to prevent aspiration pneumonia. Angiotensin- converting enzyme inhibitors and cilostazol were reported to reduce aspiration pneumonia in stroke patients. In our study, the rates of patients taking these medications were similar (13.6% in the prior period, 15.0% in the post period). Therefore, we consider that the effectiveness of these medications were limited in our study. There are some limitations to our study. First, this is a single-center study that evaluates the influence of the multidisciplinary participatory swallowing team approach on the onset of pneumonia compared with a historical control. Therefore, we cannot exclude the possibility that a selection bias exists. However, it is difficult to conduct a randomized study of the multidisciplinary participatory swallowing team approach. Second, there are few objective indicators of the effects of the multidisciplinary participatory swallowing team approach in this study. To evaluate the effects of the swallowing team approach more objectively, we need quantitative indicators, such as bacterial counts in the oral cavity that reflect the effect of oral care and serum albumin levels that reflect the effect of nutrition management. # Conclusions Our study demonstrates that the multidisciplinary participatory swallowing team approach effectively decreases the onset of pneumonia in acute stroke patients. Further prospective multicenter studies using more objective indicators are necessary to clarify the effect of the multidisciplinary participatory swallowing team approach. # Supporting Information We would like to show our appreciation to the Hiroshima University Hospital Stroke Swallowing Team. [^1]: NH reports honoraria from Mochida Pharmaceutical Co., LTD. outside the scope of the submitted work. Prof. Matsumoto reports grants from Mochida Pharmaceutical Co., LTD., Otsuka Pharmaceutical, and Daiichi Sankyo Co., LTD. as well as honoraria from Sanofi K.K., Bayer Health Care, Otsuka Pharmaceutical, Daiichi Sankyo Co., LTD., Boehringer Ingelheim, and Sumitomo Dainippon Pharma Co., LTD., that are outside the scope of the submitted work. Other authors have declared that no competing interests exist. This does not alter our adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: SA NH JH HM MM. Performed the experiments: SA NH JH MN MY TN SK Y. Nagano AN NY Y. Nishikawa MT HU KO HY. Analyzed the data: SA NH. Contributed reagents/materials/analysis tools: SA NH. Wrote the paper: SA NH. [^3]: ¶ Membership of the Hiroshima University Hospital Stroke Swallowing Team is provided in.

# Introduction Globally, mental health issues are one of the main causes of ill health, accounting for 13% of disease burden and, by 2030, this figure is predicted to rise to 15%. Worldwide, major depression is considered to be the second leading cause of disability, with depression, anxiety and drug use reported as the primary drivers of disability in those aged between 20–29 years. It is also estimated that a quarter of the population will at some point in their lives suffer from a mental health illness. Military studies have reported a wide range of mental health issues with active serving personnel and veterans alike. Indicative of many roles within the military, is the exposure to combat and trauma. High combat exposure has been associated with a deterioration in mental health and an increased risk of suicide. Figures suggest for military personnel the prevalence of depression is between 23–26%, considerably higher than the general population globally, where the figure is estimated at 13–15%. It is also suggested that up to 60% of military personnel with a mental health issue will not seek treatment. In general, a lack of healthcare management models, resulting from underfunding and austerity measures, has led to care provision for those with a mental health issue predominantly falling to family members. The impact on the quality of life for family caregivers can be wide and far reaching. Studies suggest a correlation between care burden and adverse health effects, such as increased stress, physical exhaustion, anxiety and depression for the caregiver. From a military context, the adverse health effects of the caregiver are further compounded, because even without the existence of a mental health issue, it is recognised that there are significant effects on the military family unit, such as long separations, and 24 hour working patterns. This is especially pertinent during times of deployment with military families experiencing a higher prevalence of psychological disorders. It has also been reported that the presence of a mental health issue is the second most leading cause of divorce within serving military populations. Whilst there is a plethora of literature surrounding families and deployment, a preliminary scope of the literature showed that little is known about the experience of military spouses living alongside serving partners with a mental health issue. This literature review aims to explore the experience of the military spouse during their serving partners mental health issue. # Materials and methods In acknowledgment of the limited evidence around the research topic found from the initial scope of the literature, a systematic review with narrative synthesis was executed to enable the inclusion of a wide range of literature and research designs. Qualitative evidence can answer different but often complementary questions to quantitative evidence. A systematic review assumes a narrative synthesis approach concerned with generating new insights and recommendations textually. Narrative synthesis brings together findings from all the included studies to capture conclusions. Using a deductive approach, these conclusions form thematic groups based on the body of evidence as a whole. The review did, however, follow the steps documented in the Preferred Reporting Items for Systematic review and Meta-Analysis (PRISMA) statement **See** : **PRISMA checklist**. This review specifically focuses on the spouse experience, and only aims to include studies whereby the spouse is identified in the aim or outcome. From the research aim, search terms were developed using the framework PICO and a systematic search strategy was utilised to ensure that the searches are comprehensive and transparent. Suitable databases were identified and used for the searches: CINHAL, ASSIA, Proquest Psychology, Proquest Nursing & Allied Health source, Proquest Dissertations & Theses, ETHOS, PsychArticles, Hospital collection, Medline, Science Direct Freedom Collection. All relevant search terms were utilised, and initial searches yielded limited literature specific to serving military spouses, so the parameters of the search were widened to include veteran spouses. The time parameter was also broadened to include any papers from any publication date; however, consideration was given only to those papers written in the English language. Owing to the cultural complexities, studies conducted with westernised military spouses published in peer reviewed English language journals were used. The searches were completed between July 2021 and March 2022. A total of one hundred and fifty-seven papers were retrieved from the initial searches with ninety-three being deemed of some relevance following a title and abstract sift. Twenty six papers were removed as duplicates and following a full-text search, a further forty eight papers were excluded as: only passing mention of spouse (n = 6), the spouse experience was specific to during the time of deployment (n = 13), the focus of the study was the serving/veteran partner (n = 18), the focus was directed at a treatment or therapy (n = 8) and the remaining were literature reviews (n = 3). Reference and citation searches were executed on all relevant papers, resulting in eight further papers eligible for inclusion bringing the total to twenty-seven. From the twenty-seven papers included in the review, the study aim, sample size, method and tools plus the location of study were extracted. Steps 2–4 of the Economic Social Research Council’s (ESRC) guidance on the conduct of narrative synthesis was employed. This guidance proposes four stages; however, the process is iterative, encouraging the researcher to move freely within each stage and not approach them linearly in a sequential manner. Stage one was excluded since developing a theoretical model of how an intervention works and for whom was not an aim of the review. In stage 2 and 3 the initial synthesis of the findings in the included papers was completed followed by an exploration of the relationship between the findings. Stage 4 required the research team to assess the quality of the synthesis. To reduce the risk of bias, all papers included in the review were quality assessed. A process of critical appraisal was executed to determine if the literature was trustworthy, relevant and appropriate to this study. Identifying the strengths and weaknesses in each study allows the researcher the ability to give more weight to stronger papers. The final selection of twenty-seven papers up for scrutiny comprised of eleven quantitative, fourteen qualitative and two mixed methods studies; however, it was the qualitative element within these studies that was of interest for the review. All the quantitative studies selected including mixed methods studies, employed questionnaires/surveys for data collection. An adapted quality assessment tool for quantitative papers was applied to each study. The studies were assessed against seven sections which were ranked as either strong, moderate or weak. An overall ranking was then applied ( Quantitative studies quality ranking). For the remaining sixteen qualitative/mixed methods papers, criteria developed by Kuper, Lingard and Levinson was used to assess domains such as overall coherence of the study, sampling, data collection, analysis, transferability, and ethical considerations. Studies were ranked from unclear, acceptable, good or very good. Papers were included if they ranked acceptable or above in four of the six domains (: Qualitative studies quality ranking). # Results ## Paper characteristics The initial step in identifying thematic groups is to assess the characteristics of the selected studies. The fundamental characteristics within each the twenty- seven studies are identified in. The focus differed amongst the studies, with nine studies including both veteran and/or serving partner as well as the spouse, denoting the spouse element was part of a much larger study. In these cases, the spouse findings have been used within this review. Eighteen studies had specific focus on the spouse. Six studies had a specific focus on the spouse experience whilst their partner was still serving, the other twenty one studies focused on the spouse experience cohabiting with the veteran population. All but six studies paid specific attention to Post-traumatic Stress Disorder (PTSD) and PTSD symptoms only. Within fourteen studies the serving/veteran partner had a clinical PTSD diagnosis \[,, –\]. Seven studies self-reported PTSD with 4 studies using a clinically recognised symptom assessment and classification tool (DSM-III or DSM-IV) to justify participant selection. Nineteen studies identified specific military conflicts. Nine studies specifically focused on more recent conflicts, Operation Enduring Freedom (OEF) in Afghanistan and Operation Iraqi Freedom (OIF) in Iraq and eight studies related only to the Vietnam or Persian Gulf conflicts. One study included participants from both Vietnam and OEF/OIF whilst one study compared and contrasted the findings of two separate studies, where study one focused on OEF/OIF and study two focused on Vietnam. Eight studies did not make any reference to any specific conflict. In all studies included within this review the gender of the spouse was majority female. Nine studies acknowledged the duration of the relationship with six studies identifying an average of up to eighteen years and two studies having an average of twenty-six years plus. Nineteen of the studies did not specify a duration or an average was not calculated. Across the twenty-seven studies a range of recruitment methods were executed. Ten studies utilised formal avenues; four used couples-based marriage enrich workshops, six used outpatient PTSD clinics, and three via random selection from military records. Six studies used advertising, two studies used a snowballing method and two studies utilised a combination of both. Three used third-party services specific to veterans and families and one study recruited from a church group. Twenty-two studies were carried out in the United States of America \[–, , \]. Two studies were carried out in Australia and three studies were completed in the UK. Two studies utilised a mixed method approach, however it is the qualitative element of each study that is relevant for this review, fourteen utilised a qualitative method and eleven utilised a quantitative method \[, –, \]. A range of data collection methods were utilised. Thirteen studies carried out questionnaires or surveys and two studies offered an opportunity for free text within their questionnaire. Seven studies carried out face-to-face interviews, one study used telephone interviews, and a further study choose to use a combination of face-to-face and telephone interviews. Only one study opted for observation and documentation and one opted for focus groups as the method of choice. Whilst there was commonality in the overarching themes being tested within the quantitative studies, that unity was not evident in the selection of tools, inventories and scales used to collect the data. Three different scales were used in more than one study. The PTSD checklist (PCL) was utilised in three studies with a further two studies using a military PTSD checklist (PCL-M). The Burden Inventory was cited in four studies and a further three studies named the Relationship Assessment Scale. All the qualitative studies with the exclusion of one used in-depth semi-structured interviews as the chosen method of data collection. Analysis of the retrieved papers was undertaken to identify emerging themes. Five themes were identified; three distinct themes featured in over half of the studies and a further two themes emerged from over 25% of the studies. See. The five themes are: - Theme 1: Caregiver burden - Theme 2: Relationships - Theme 3: Psychological/psychosocial effects on the spouse - Theme 4: Mental health service provision - Theme 5: Spouse’s knowledge and management of PTSD symptoms. ## Theme 1: Caregiver burden Caregiver burden is defined as the extent to which caregivers perceive their emotional or physical health, social life, or financial status to be affected by their caring for an impaired relative. The concept of caregiver burden includes both an objective element such as strained relationships, financial constraints, and a subjective element such as the reactions and responses as a result of the demand placed on the carer. Partners of ‘cared for’ individuals are potentially at higher risk of experiencing caregiver burden and poorer mental health as opposed to other family, friend or unrelated carers, due to residing together and increased long-term exposure to each other. The notion of caregiver burden was cited in seventeen studies \[,, –, –\]. Yambo et al. and Sherman et al.’s studies identified two differing types of care burden, first the psychological burden, discussed in theme three and secondly, the burden from the practical and physical actions required from the carer. To provide such support very often required a change in role, which was identified in nine of the studies. Within Brown, Lyon, Mansfield et al., Temple et al. and Yambo et al.’s studies, the spouses’ stated that they felt more like a care provider than a wife; being an advocate for their serving/veteran partners’ care. A quote from Temple et al. states: > “*the relationship feels like I’m a nurse v’s the spouse*” > > (p171). In some cases, this change of role was taken on voluntarily; however, for some spouses, the change in role felt forced upon them as a result of their serving/veteran partner being unable or unwilling to perform a role within the relationship. Apparent in four of the studies’ findings, was the naivety and the disillusionment of spouses regarding their appreciation of longevity of the caregiver role, thinking that the role would be temporary rather than the emerging long-term/permanent reality which spouses voiced. One of the ways the long-term impact was identified was the manifestation of the need to constantly maintain the *‘peace’* in order to minimise stress for their serving/veteran partner. Mansfield et al.’s study likened it to > “*walking on eggshells*” > > (p492). Noted across the findings of six studies, was that the physical and mental demands felt from years of providing care, increased stress levels, caused frustration and ultimately, fatigue. The participants in Brickell et al. study highlighted how time-consuming caregiving is and consequently led to exhaustion and being emotionally drained. The burden on the caregiver that this sense of dependency caused was echoed in Waddell et al.’s findings. In addition, a further attributing factor emerged: there was a significant correlation between caregiver burden and the severity of PTSD symptoms. Similarly, three studies, found that serving/veteran partners PTSD severity was a reliable predictor of caregiver burden. As well as exploring the severity of symptoms and caregiver burden, Calhoun et al. study included the level of veteran interpersonal violence; an area not examined in other studies. Findings showed that symptom severity was not solely attributable to caregiver adjustment/burden and that there was a significant association between interpersonal violence and both caregiver burden and partner psychological adjustment. Seven studies also highlighted a link between the need to be engaged in the serving/veteran partner’s treatment and caregiver burden. Within all seven studies there was a hope and expectation that by being involved with their serving/veteran partner’s treatment plan, their serving/veteran partner’s symptoms would decrease and in turn lessen the overall burden of care they felt. ## Theme 2: Intimate relationships Central to all people’s lives are relationships. Relationships come in all shapes and sizes, from casual acquaintances to family/blood connections and to intimate relations. The motivation to establish intimacy with others, is part of a basic human need to belong. Intimacy is a complex concept that is multifaceted, with a range of components within it. Explanations of intimate relationships are founded upon research findings from the fields of psychology, neuroscience, sociology, and from family and communication studies. ‘Intimate relationship’ was a point of discussion within sixteen of the studies. Thandi et al.’s study recognised ‘intimacy’ as a key theme divided into two subthemes: physical intimacy and emotional intimacy. Three studies portrayed positive relationship views, all of which were noted to be from the participants’ discussion of either their relationship prior to deployment and/or the onset of PTSD symptoms, or when the spouses’ talked about their commitment to the relationship. Throughout all the studies there was some degree of negative connotation concerning the spouses’ relationship with their serving/veteran partner. Allen et al. Campbell and Renshaw, and Renshaw and Caska’s findings suggest that the serving/veteran partner’s recent deployment and subsequent increase in PTSD symptoms was indirectly linked to negative marital functioning but were not statistically significant. Overall, PTSD symptoms and their severity were a specific feature in five of the studies, all of which highlighted a major impact on the marital relationship. One other contributing factor to the impact on the marriage relationship was domestic abuse seen in Jordan et al. and Mansfield et al.’s studies. Mansfield et al. reported that 10.6% of their participants were victims of verbal, emotional or physical abuse. Jordan et al. found the prevalence of abuse by asking for the number of violent acts, including threats of violence over the previous year. For spouses of veterans with PTSD, there was greater incidence of abuse both as victims but also as perpetrators of abuse towards their serving/veteran partner. Thirteen studies noted changes in personality, difficulties in communication and long-term withdrawal of the serving/veteran partner ultimately leading to emotional numbing or an emotional disconnect. Waddell et al.’s study illustrated how intimacy problems surfaced because of participant experiences of emotional alienation from being unable to express or share thoughts and feelings with their serving/veteran partner. Renshaw and Caska suggested the generalised symptoms, such as social withdrawal are easily misinterpreted by spouses as a reflection about them and/or the relationship, whereas physical symptoms were commonly linked to an illness and therefore, posed minimal threats to the relationship. Nevertheless, this distancing and abandonment manifested in most of the studies as frustration with/or sadness, grief and loneliness about the relationship changes. Thandi et al. found some participants discussed a change in character in their serving partner and that they were no longer like the person they married, which led less affection and more arguments. Whereas, in Waddell et al. findings, participants viewed their relationships as different to others, prompting the notion that their relationship was not a ‘normal’ one. In addition, Waddell et al. and Waddell et al. found that the spouses’ felt there was a constant striving to intimately connect with their serving/veteran partner, again in order to break down the barrier of emotional detachment. Verbosky and Ryan and Thandi et al. found that for some participants the lack of intimacy enhanced the spouse’s need to be nurturing and caring in order to reconnect. Sherman et al. and Waddell et al.’s studies reported participants expressed loyalty and commitment to their serving/veteran partner and described the importance of providing emotional and behavioural support. This dedication to the relationship was mirrored in Brown, Iniedu, Lyons, Mansfield et al. and Woods findings, although, all reported an ongoing inner struggle as to whether to stay or leave the relationship for most participants. Factors such as children, domestic abuse were listed as reasons to leave; however, these were often overruled by guilt, love, a sense of obligation, and fear that their serving/veteran partner would worsen if they left. Longevity of the relationship was also a consideration which featured in both Allen et al. and Woods studies. Woods’s study showed that those participants with longer relationships were more likely to remain in the marital relationship believing that the relationship was positive whilst those with younger marital relationships, predominantly viewed their relationship, with more negativity. Whereas Thandi et al. and Martinez identified longevity, not by the length of time in the relationship but that a positive relationship was established over a period of time, post diagnosis. ## Theme 3: Psychological/Psychosocial effects on the spouse Seventeen studies reported on the psychological and psychosocial impact experienced by spouses of either serving personal or veterans with PTSD or other mental health illnesses. Psychological distress was a predominant finding throughout the studies. Manguno-Mire et al., Beckman et al. and Iniedu all indicated that the greater the severity of PTSD symptoms experienced by the serving/veteran partner, a greater intensity of psychological distress, dissatisfaction and anxiety was experienced by the spouse. Manguno-Mire et al. and Brickell et al. studies report high individual measures for anxiety, depressive/somatic symptoms and suicidal ideation with the suggestion that the severity of symptoms might warrant clinical intervention. Iniedu’s study found that all spouses experienced secondary trauma as a result of their serving/veteran partners PTSD symptoms and were in receipt of medication and/or face-to-face therapy. Lyon, Waddell et al., Calhoun et al., Murphy et al. and Johnstone and Cogan reported the negative impact that living with a serving/veteran partner with PTSD had on spouses’ mental health by identifying the stress related symptoms they experienced. Manguno-Mire et al.’s study further identified predictors specific to the levels of psychological distress experienced. Psychological distress was found to decrease when there was greater involvement with the serving/veteran partner’s care and treatment; however, if there had been a recent episode of mental health treatment or an increased perceived threat from the serving/veteran partner’s PTSD symptoms, the psychological distress felt by the spouses was also increased. Whereas Martinez examined attachment style and the level of attachment within the relationship and the subsequent effect on psychological and physical symptoms. He found that caregivers with an anxious attachment style were more likely to experience physical symptoms and higher incidents of physiological stress than those with a non-anxious attachment style. Manguno-Mire et al.’s study identified that 60% of participants reported that their serving/veteran partner posed a physical threat to their wellbeing. The threat and psychological distress were also demonstrated in other studies. Murphy et al. identified the volatile environment where some of the participants likened it to the metaphor: > ‘*walking on eggshells*’ > > (p5); Albeit a different interpretation of the same metaphor highlighted in the earlier theme. The ‘*not knowing’* and loss of predictability invariably leads to hypervigilance and hyper-attentiveness which was documented in nine studies. Remaining hypervigilant and hyperattentive to the actions and moods of their serving/veteran partner has been aligned with the need to find a resolution and an attempt to create peace and healing. Yambo et al., and Johnstone and Cogan findings suggest an opposing view of an emotionally unstable environment, resulting from the increased feelings of stress from continuous exposure of symptoms, unpredictability and hypervigilance. As well as spouses being hypervigilant and hyperattentive to their serving/veteran partner’s needs, seven studies highlight a distinct level of responsibility felt by the spouse. Fear for their serving/veteran partner, guilt linked to the inability of being able to rectify their serving/veteran partner’s difficulties and emotional pain led to feelings of self-hate and blame. Lyon’s study demonstrated the move from the early phases of the relationship, referring to spouses’ feelings being: > ‘*compatible with the honeymoon period*’ > > (p72) towards the mid phases, where comprehension of the severity of their serving partner’s PTSD and subsequent impact on the relationship are realised. The feelings reported were numerous and varied from happiness and laughter to frustration, resentment and/or bitterness, guilt and humiliation; being out of control and trapped; grief and loss to pride for both themselves and their serving/veteran partners. Temple et al. simply described the relationship as a ‘*roller coaster’*. This myriad of feelings continues to be identified throughout six studies. The most, negative connotations are described by Brown, she points out: > “*anger was used 103 times to describe their feelings in comparison to > love at 32 times*” > > (p250). Verbosky and Ryan state that the spouses experienced an overwhelming sense of helplessness and uncertainty as they were unable to formulate plans to effectively deal with the symptoms and situations they faced, finding it difficult to be assertive at the appropriate time. In contrast, Iniedu and Johnstone and Cogan suggest that there was evidence of empowerment brought about by the spouses struggles to cope and hold everything together; indicative of the concept of post-traumatic growth. Amongst the extensive array of feelings identified and the change in behaviours required as a result of the serving/veteran partner’s symptoms, a loss of self was identified in five of the qualitative studies. Brown’s study illustrated that participants had exhausted all their intrinsic resources and faced a lack of normality which in turn meant that many had neglected responsibility for themselves and indeed, lost themselves in a sense of powerlessness. Whilst Brickell et al.’s study acknowledged the loss of self from an emotive perspective, the loss of physical self-care was also recognised, emerging in their analysis as the most frequently endorsed theme The transference of loss of self into home and work life was evident in eight of the studies. In most, this psychosocial element was identified as “*tremendously stressful*”. Iniedu, Temple et al. and Verbosky and Ryan’s studies identified that managing either their serving/veteran partners’ symptoms and/or their own stresses had significant ramifications on daily life and in some cases had taken over completely. Temple et al. and Waddell et al., and Brickell et al.’s studies suggested that the adaptations and modifications required to daily life meant that the spouses had to adjust work hours and, in some cases, reduce hours or quit their job. In addition to the impact on their own lives, there was also the identification of children within such scenarios. Brown, Jordan et al. and Temple et al. found similarities in the concern voiced regarding the impact on children and their subsequent behaviours. It could be suggested that as a result of the negative feelings felt by most spouses within the studies there would be correlation with friendship, socialising and external support (discussed in Theme 4). Nine studies explicitly documented findings highlighting friendship and socialisation; Waddell et al.’s study briefly highlighted the spouses’ social isolation whereas Temple et al.’s study explored this in greater depth, finding that the serving/veteran partners’ struggle to leave the house, had an impact on their ability to socialise which led to difficulties maintaining existing friendships or making new ones. Likewise, Brown, Verbosky and Ryan and Bickell et al.’s studies indicated that participants merely gave up on any recreational or social activities. Similarly, Murphy et al. and Brickell et al.’s participants felt others (family and friends) who were not living the same experience, simply did not understand. In contrast however, Jordan et al.’s quantitative study found no significant difference in the levels of social isolation. ## Theme 4: Mental health service provision Mental health service provision emerged as a theme within ten of the papers albeit very briefly in three papers. The ability to liaise with medical or other trained professionals with experience of dealing with PTSD was reflective of the spouses perceived individual needs in three studies. Greater involvement in the serving/veteran partner’s care and treatment by the spouse was also noted. Mansfield et al. and Temple et al.’s studies also identified mental health services; namely, requests for help or receiving care. The spouses’ main aims were to gain information in order to inform their care, receive constructive feedback on how they were managing, sharing information that may not have been disclosed by their serving/veteran partner or merely sharing their experiences of daily life. It is evident from the spouses’ experiences however, that these requests were not always received positively by the mental health services. Johnstone and Cogan voiced: > ‘*a sense of being invisible, forgotten and overlooked*’ > > *(p45)* when it related to their serving/veteran partners’ treatments. Although Murphy et al.’s findings highlighted the value participants felt by being able to share experiences and gain expert in-depth knowledge from specialist practitioners. Serving/veteran partners had a clinical diagnosis and had in the past or were currently receiving treatment for their illness in sixteen studies. It was also identified in theme three that psychological distress was prevalent throughout seventeen of the studies with six studies recognising that the spouse themselves had to seek help or treatment for stress related symptoms. Mansfield et al.’s, Waddell et al.’s and Waddell et al.’s studies described similar feelings albeit related to the attempt as seeking help for themselves, feelings of isolation and invisibility were recognised in comments such as: > “*in general family members seem to be left out” and “…..but there is > no help for the family*” > > (35. p419). In Buchanan et al., Temple et al., Johnstone and Cogan’s studies, spouses were cautious about reaching out to others since their partners were *serving* military personnel and access to mental health care services differed to that offered to veterans. Buchanan et al.’s study highlighted stigma towards PTSD, which is echoed in the majority of narratives gathered by Temple et al.. In addition to the stigma, the narratives also highlighted the mixed messages received from the military unit. Positive messages surrounding PTSD were promoted through adverts in and around the military base, however, direct actions such as accessing services sent a: > “*negative message that the marine was weak*” > > (40. p172) and reactions received when the serving/veteran partner tried to access services was that: > “*a spouse’s cry for help doesn’t matter*” > > (40. p172). Mirrored in another study, spouses, voiced similar feelings of being: > “*silenced by the institution; by having no voice*” > > (27. p240). A further complication to accessing help and support from mental health services and professionals was the belief that doing so, jeopardised the future career prospects for their serving/veteran partner. One narrative in Temple et al.’s study differed, however; this was from a spouse who was a serving member of the military and whose experience varied as a result of being part of the organisation. For those spouses who did have experience of liaising with services, there were a couple of positive comments raised pertaining to service provision. However, the majority of comments made were critical of the services provided. ## Theme 5: Spouse’s knowledge and management of symptoms Six of the studies highlighted spouse’s knowledge around PTSD and the management of symptoms when they occurred. Buchanan et al.’s study specifically focused on the awareness of PTSD from the spouse perspective. They undertook a critical incident survey which included the question *“How would you know if your spouse/partner needed treatment for PTSD*?*”* The findings suggested that two thirds of spouses had received no formal training on PTSD and most spouses had accessed informal sources to learn about PTSD. Media resources such as movies, news broadcasts or internet were identified as primary sources. Murphy et al.’s study suggested that as a result of a sense of responsibility, practical learning about what to do and say was valued by the participants. Temple et al.’s study presented one spouse who differed from the other spouses; she voiced a clear understanding and underpinning knowledge of PTSD symptomology which she attributed to the in-house training she had received as a serving member herself. Buchanan et al.’s study explored further spouses’ knowledge and understanding about PTSD causes, a fifth of spouses were able to identify the causes relating to their serving/veteran partners. 12% of participants declared they had little knowledge of the presenting symptoms. While Murphy et al.’s study didn’t specifically explore an individual’s knowledge, it highlighted the need to share experiences with peers in similar situations in order to gain reassurance and increase confidence in their understanding. One of the key themes emerging from Sherman et al., Temple et al., Yambo et al. and Thandi et al.’s studies into spouses’ experience of living with serving/veteran partners with PTSD, was being unprepared to handle the condition and/or deal with the complexity of the symptoms. Thandi et al.’s participants. > ‘*described how they felt ill-equipped to perform the role as > caregiver*’ > > (p2). Most participants in Temple et al. and Yambo et al.’s studies stated they had never been provided with any information about PTSD either before or after their serving/veteran partners’ deployment and consequently were unable to identify whether their serving/veteran partners had PTSD. As a result, of the lack of information around PTSD, spouses begun to doubt their relationship and own sanity and believed that they were to blame for their serving/veteran partners’ destructive behaviours; and for some spouses, this belief had exceeded 10 years. # Discussion Following completion of the review, it was apparent that there was a limited range of papers where the primary focus was the experience of the spouses of serving military personnel. As explained earlier the parameters of the search had to be widened to include spouses of veterans and the time scale was broadened to include studies undertaken post the Vietnam conflict. On reviewing the available literature, five predominant themes emerged. Interlinked themes were identified it was sometimes difficult to separate findings into one distinct theme since in most cases, they often interlinked. The notion of caregiving burden was evident in several papers. Within most studies’, caregiver burden was viewed negatively. Evident in the literature was how the spouses’ level of burden increased at times when their serving/veteran partners’ symptoms of PTSD were at their most severe. Likewise, when their serving/veteran partners’ PTSD symptoms were minimal and they were responding well to an aspect of treatment, the level of caregiver burden felt by the spouses lessened. As well as the perceived caregiver burden, the impact on the relationship was also apparent and emerged as another key theme. The majority of spouses were married and had been a part of military life whilst their serving/veteran partners were serving in the case of the veterans. The toll on the relationship was evident, with many spouses stating that they had—at times—felt like leaving the relationship. Many spouses blamed themselves for the problems faced in the relationship. In some of the literature, accounts about the relationship prior to their serving/veteran partners’ PTSD illness were taken from the spouses. These were reflected on with fondness and love akin to the ‘honeymoon period’. Once symptoms such as emotional detachment entered the relationship, the relationship became much harder, and problems began to escalate. Many spouses felt a sense of responsibility to stay and ‘*stand by their man’*, and in all but one of the papers the spouses had stayed. Some of this was out of fear that their serving/veteran partner would hurt themselves or become worse. For some, it was out of loyalty, for some it was guilt about deserting them in their time of need and for some it was love. Very often it was mixture of all these reasons, meaning the relationship was no longer viewed as ‘normal’. The decision to stay had ramifications psychologically and psychosocially on the spouse. Throughout many of the studies, it was evident that they found coping with everything- family, home, work and their serving/veteran partner -stressful and anxiety provoking. This stress led to many spouses seeking treatment for their own mental health needs. Spouses described being peacekeepers to prevent triggering their serving/veteran partners’ symptoms. Spouses became hyper- vigilant and hyper-attentive to their serving/veteran partners’ behaviours and needs which, ultimately, placed greater strain on themselves. Spouses also described how their lives had changed socially; some felt forced to reduce their working hours, withdrawal from maintaining existing friendships and/or making new acquaintances due to caring for their serving/veteran partners’. For most, the spouse experiences held negative connotations with few studies exploring resilience, growth and/or transformation of self or the relationship. A small number of studies explored what knowledge and insight spouses held about PTSD or mental health issues. For the majority, no formal training or guidance had been received and most of the spouses had used media such as films, internet, and campaigns to make the connection between their serving/veteran partners’ symptoms and mental health issues. Mixed messages were also highlighted; however, this was predominantly from those studies where the serving/veteran partner was still serving. For these spouses, there was an element of fear about upsetting the ‘*applecart*’; they were frightened that disclosing their serving/veteran partners’ symptomology or seeking help would affect their serving/veteran partners’ career prospects. Further, that their serving/veteran partners would be stigmatised by a diagnosis despite widespread use of flyers and advertisements stating that it was ‘*ok to talk’*. Spouses felt torn between the need to help their serving/veteran partner verses jeopardising their partners’ career. Many spouses felt invisible and isolated with nowhere to turn for support for either their serving/veteran partners or themselves. Barriers to mental health service provision were also recognised; for some it was the financial burden, for others accessibility and/or time and/or not even knowing where to go in the first instance. For those who had accessed mental health services, the experience was far from ideal for most; staff shortages, lack of funding, long waiting times and poor facilities meant disappointment once access was finally gained. # Limitations to current research and systematic review and narrative synthesis Employing a systematic search strategy ensured that the searches were transparent. Despite adopting the systematic approach, only a limited number of contemporary papers specific to the military spouses were yielded. The lack of peer reviewed studies over recent years internationally, provided the rationale for the inclusion of earlier studies. These were identified by increasing the time parameters and by executing a reference and citation search on the papers found; again, this yielded only a few earlier papers for inclusion. From the twenty-seven studies identified, nineteen of the papers focused primarily on the spouses’ experiences. However, only a few specifically pertained to the spouse experiences of serving personnel; the majority were spouses of veterans. Owing to the cultural complexities across military organisations, studies conducted with westernised military spouses published in peer reviewed English language journals were deemed appropriate to expand this review. A major limitation is the distinct lack of studies carried out outside of the USA; only five studies identified in the UK or Australia. Whilst all the studies used were from westernised cultures, the differences in deployment terms, and healthcare systems are noteworthy. This would make the transferability of some of the findings across countries problematic. A further limitation surrounded the specifics of the mental health issue itself. The emphasis in most of the included papers were specific to either experience of service personnel directly after deployment with PTSD, or veterans who were no longer serving with PTSD. PTSD was the single focus for many of the papers; only six papers referred to other mental health illness as well as PTSD. Many of the papers in the review made specific links to war as a precursor to the PTSD. The papers gathered made links to either post service in the Vietnam conflict or after serving OIF & OEF conflicts. It is noteworthy that the conflicts were fought 25 years apart and also in different countries and terrains. They were fought by very different means, in that Vietnam used predominantly guerrilla warfare tactics with a largely unseen enemy, whereas OIF and OEF were more conventional in the type of warfare deployed; for example, soldiers faced a modern military organisation with greater use of armoured and air support. These differences suggest that the experiences and exposure faced by those serving, could have been markedly different. This review is focused specifically on spouses to military personnel or veterans who have served and therefore is not inclusive of the wider literature exploring those spouses’ experience outside of a military context. This focus was intentional, due to the differences in mental healthcare provision for serving personnel. When considering non- military civilian couples, the majority have, and will access the same healthcare provision/organisation. This has advantages such as information sharing between professionals for the provision of holistic family care. Whereas, with most residing military couples, the serving military member accesses different care provision to their family. Accessing separate care provisions provides a potential barrier to information sharing and access to support. It is widely acknowledged that there are a range of programmes/interventions that aim to offer support to spouses who find themselves experiencing life alongside a serving/veteran partner who has a mental health issue; for example, Spencer-Harper et al.’s study of group psychoeducation support. As a result of such programmes/interventions, it is understood that grey literature exists by wider professional, charitable organisation and government publications. Only peer reviewed research was included in this review which meant that all grey literature was excluded. Two further exclusions were domestic violence and secondary PTSD. This was purposeful, as the aim of this review was to explore the experiences of spouses and not the outcome resulting from the experience. It is widely acknowledged, that the potential outcomes of living with someone with PTSD, are, a higher incidence of intimate partner violence and a higher incidence of secondary PTSD for the spouse. It was felt that the inclusion of such studies would detract from and overshadow the limited peer reviewed literature available. # Conclusion The review has identified that there remains a gap in the literature, specifically, studies focusing on military spouses of serving personnel; most of the studies focused on spouses of veterans, but similarities were noted. The majority of the papers reside in the USA (n = 22), with minimal papers from the UK and Australia (n = 3 and n = 2 respectively). While there was a near equal divide between quantitative or mixed methods and qualitative \[n = 11+n = 2 and n = 14\], only nine studies used interviews as the data collection method. Thus, posing a further limitation as the majority of data collected, lacked the rich, in-depth nature required to explore spouse experience. The findings from the review have some implications for policy, practice and research focusing on the military spouses’ experiences of living alongside their serving/veteran partners during a mental health issue. Care burden from both a psychological and a physical/practical aspect was evident, as was the longevity of their partners’ mental health issues. All led to long-term impact, where for most military spouses felt more like care providers than partners. The impact was also felt in the intimate relationship between military spouse and partner; difficulties in communication and emotional numbing were identified. However, dedication and commitment to the relationship was also noted. For the military spouses’ themselves, there was a sense of ‘*loss of self’* as a direct result of caring for their partner. In addition, there was a felt sense of being invisible and/or overlooked by the mental health services; when all that was required was inclusion to gain information, so that they could better manage their partners’ care. Understanding the experiences, perspectives and difficulties of military spouses whilst living alongside their serving partner/veteran during a mental health issue, will assist in better understanding of how their interactions can support or implicate their partners’ recovery. Inclusion from services needs to be considered as a protective factor for both the military spouse and their serving partner. # Supporting information [^1]: I can also confirm that the authors have declared that no competing interests exist.

# Introduction Between 8% and 10% of the North American population has some speech disorder, including 3 million stutterers, and 7.5 million individuals with dysarthria (caused, e.g, by cerebral palsy, Parkinson’s, or multiple sclerosis) according to the U.S. National Institute of Health. Moreover, since linguistic change is often among the first symptoms of neuro-degenerative cognitive decline, the broader set of speech and language disorders are expected to increase with the rising prevalence of dementia in the aging population. It is therefore imperative to build tools for earlier detection and management of change in language. To the extent to which these tools will be based on machine learning, this will require large datasets; unfortunately, the available data tend to be prohibitively small for rarer diseases, and prohibitively difficult to collect for more at-risk populations. We therefore developed a language assessment tool, called Talk2Me, designed for large-scale self-administered collection of spoken and written language data. This includes new open-source software for feature extraction, a publicly-available data set on which those features were applied, and analysis of relevant linguistic patterns in those features. After describing the dataset and the feature extraction framework, we present examples of how the dataset might be used in practice, including. This includes unsupervised learning to detect clusters of participants in unlabelled data, and analyzing the relationships between different features within a single task, as a means to reduce redundancy in prospective data collection. We also apply normative healthy control data, of the type we obtain, to the classification of Alzheimer’s disease (AD) in a smaller dataset. ## Assessing AD with automated analysis of language Language decline is one of the most salient symptoms of AD. Linguistic and acoustic features, such as information content, noun-to-pronoun ratios, and changes to the pitch contour, are indicative of cognitive decline. Connected speech is often used for assessing AD, as it provides insight into semantic processing and syntactic complexity. In a study with 30 healthy participants and 36 participants with AD, Meilán *et al* extracted temporal and acoustic features from speech of simple sentences spoken by older adults with and without AD, and achieved an accuracy of 84.8% in a binary classification task (Healthy vs. AD) Often, connected speech is elicited through the picture description task. Forbes-McKay and Venneri found that deficiencies in semantic processing were apparent not only in participants with AD, but also in those diagnosed with mild or moderate forms of cognitive decline. Jarrold *et al* collected speech from 48 participants completing a picture description task. They then automatically extracted lexical and acoustic features and identified participants with dementia using different machine learning approaches. Similarly, de Lira *et al* extracted linguistic features from transcripts of 121 older adults completing a picture description task and concluded that participants with AD displayed more word-finding difficulties, revisions, and repetitions than healthy controls. In a study involving 48 participants, Giles *et al* used information content to classify participants into four categories of impairment: none, minimal, mild, and moderate. Interestingly, they found that participants with minimal cognitive decline had significant impairments in the production of information units, indicating that decline in the use of semantic content can be detected at early stages of cognitive impairment. DementiaBank is a popular dataset for studying language in AD in which 167 adults with dementia and 97 adults without, all above the age of 44, completed the ‘Cookie Theft’ picture description task, from the Boston Diagnostic Aphasia Examination, which includes the raw audio, the textual transcripts, and the validated clinical assessments. Participants completed these tasks once a year along with a mini-mental state examination. However, longitudinal data points in DementiaBank, especially for AD participants, are sparse. On DementiaBank, Yancheva and Rudzicz automatically detected information content using word embeddings for binary classification of AD and achieved an F-score of 0.74. On the same dataset, Wankerl *et al* employed an *n*-gram based approach and built two language models, one for AD participants and one for healthy participants and achieved a classification accuracy of 77.1%. Fraser *et al* used over 370 automatically extracted acoustic, lexical, and syntactic features to detect AD based on acoustics and speech transcripts. Other tasks have been used to study the relationship between language and AD. Kirshner *et al* found that people with AD had difficulty on naming tasks even though their language showed no qualitative signs of deterioration, except perhaps that they used more generic terms instead of specific ones. Other work has explored verbal fluency and story recall, for example, as means to assess AD. ## Interfaces for self-administered data collection Clinical cognitive assessments are generally performed in person using physical booklets directly with a clinician. Recently, there has been a push for more remote approaches including telephone-based versions of existing cognitive assessments such as the MMSE and the Montréal Cognitive Assessment (MoCA). Rapcan *et al* administered a battery of language assessments over the telephone and found that speech features (e.g., number of pauses, length of utterances) could be reliably extracted by telephone recordings and did not significantly differ from in-person clinic recordings. Using an interactive-voice-response telephone system, Yu *et al* extracted speech features and achieved an AUC of 0.77 in classifying between healthy and cognitively impaired participants. Van Mierlo *et al* built a web- and telephone-based system for administering cognitive self-tests as a method of automatic screening. In their study, 117 participants used their system and were classified into one of the following categories: subjective cognitive decline, mild cognitive impairment, and dementia. They achieved an AUC of 0.86 with the web-based system, and an AUC of 0.78 on the telephone assessment. The tasks employed in that work, however, did not include free-form speech or language production, which is our focus here. # Materials and methods Talk2Me collects data through tasks similar to those used in standard assessments of cognition (including the Mini-Mental State Examination, the Montréal Cognitive Assessment, and the Western Aphasia Battery). Users register on the website and provide consent, then complete a demographics survey. The survey collects information on their sex, age, ethnicity, language fluency, education level, country of origin, and country of residence. Users are also asked if they have ever been diagnosed with dementia, if they are currently taking dementia medication, and if they have been a regular smoker of tobacco within the last 3 years. After answering the survey questions, they can then complete multiple sessions of data collection, some through typing and others through speaking. In order to be as generic as possible, no restrictions are placed on the environment or channel, except that the browser must support HTML5, which is the case for all major browsers. The source code for this tool is being made available publicly (<https://github.com/SPOClab- ca/talk2me_interface>). All data were recorded given informed consent by the participants, according to Research Ethics Board protocol \#31127 of the University of Toronto, which specifically approved this study. ## Language tasks Website users complete eight different types of language task, each designed to evaluate various aspects of cognition. During each session, users complete one or more instances of each task, with each instance corresponding to a different stimulus (e.g., a different word to define, or different picture to describe), as summarized in. 1. **Image naming** In each image naming session, six pictures are displayed and the participant types the name of each object depicted. Images are taken from the Caltech-256 Object Category dataset. 2. **Picture description** Participants verbally describe images that convey more complex scenes that show interacting objects. There is no time constraint on this task, although participants are encouraged to speak for at least one minute. Images used for this task include the *‘Cookie Theft’* picture from the Boston Diagnostic Aphasia Examination, the *‘Picnic scene’* from the Western Aphasia Battery, twelve re-distributable images from Flickr (<https://www.flickr.com/>), and twenty-nine images from the Webber Photo Cards: Story Starters collection. 3. **Fluency** In each session, participants type as many words as possible that match the category. Categories typically consist of a semantic variant (e.g., types of animal) or a phonemic variant (e.g., words that begin with *F*). Verbal performance on this task can differentiate a variety of conditions, including traumatic brain injury and dementia. 4. **Story recall** A short story is displayed to the participant. The text then disappears, as expected, and participants verbally re-tell the story in their own words. There is no time constraint on either phase, but participants are encouraged to speak for at least a minute. Stories used in this task are the *‘My Grandfather’* short story, the *‘Rainbow’* passage, and the *‘Limpy’* passage (<http://itcdland.csumb.edu/~mimeyer/CST251/readingpassages.html>) which are standardized among speech-language pathologists to assess speaking and memory skills. 5. **Vocabulary** Participants define five words by typing definitions using their own words. Words used in this task are taken from the Brown corpus. Each word is assigned a difficulty based on its age-of-acquisition, derived from the Kuperman norms. Specifically, the set of all words is sorted by increasing age-of-acquisition and subsequently trisected into partitions of equal size, uniformly across scores, representing ‘easy’, ‘moderate’, and ‘difficult’ words. 6. **Winograd schema** The Winograd Schema challenge consists of questions with two possible answers (e.g., one stimuli is ‘*The trophy could not fit into the suitcase because it was too big. What was too big—the trophy or the suitcase?*’). Instances are taken from the publicly available Winograd Schema challenge (<https://www.cs.nyu.edu/davise/papers/WS.html>). Participants simply select an answer from the available pair. 7. **Word-colour Stroop** In the Stroop inference task, the user is presented with the name of a colour, presented in a coloured typeface. The user says the colour of the given font out loud, ignoring the orthography. The Stroop test has a high degree of discriminative power in Alzheimer’s disease, depression, and bipolar disorder, for example. 8. **Self-reported disposition** Participants answer five questions taken from a validated short-form version of the Geriatric Depression Scale (GDS), which is a 30-item self-assessment used to identify depression in the elderly. We collect these responses, since mood can affect a person’s performance in language tasks, and since a focus on dementia is ongoing in a parallel study. From the GDS, we ask yes/no questions on life satisfaction, general happiness, and everyday activities. Participants are also asked to rate their current mood on a scale from 1 (very sad) to 10 (very happy). # Talk2Me database In this section, we describe the demographics of individuals in the database, then we describe how different tasks are scored. Lastly, we describe the extracted features. A task *score* is a quantitative measure of how well that task was performed towards some *goal*. Naturally, this only applies to tasks that have an explicit purpose to be achieved. Unlike scores, *features* don’t directly measure success in performing a task, but rather evaluate intrinsic aspects of how the task was performed. ## Demographics Collection of this database is ongoing, and subsequent releases or “snapshots” that we make publicly available will be versioned and time-stamped. We report results and analysis on 1369 sessions completed by 339 unique users, of whom 206 have completed more than one session. The released dataset includes sessions from all participants who have agreed to the public release of their data, and contains 1033 sessions from 196 users. Participants were recruited on a voluntary basis, self-assessing for an adequate level of proficiency in English. 96% of users report being native or fluent speakers of English and 92% of users report being Canadian residents but the tool is not built for any particular country or accent; 3% are from United States and the rest are from other countries. We do not restrict age, sex, or other demographics. While most participants using Talk2Me are less than 30 years of age, approximately 50 users are older adults and 36% are female. shows the distribution of age and education level over all participants. ## Scoring the tasks We automatically transcribe audio files for the picture description, story recall, and word-colour Stroop tasks with Kaldi, an open-source speech recognition toolkit, using a long short-term memory neural network with i-Vector input and a reverberation model, trained on the Fisher data. An *ad hoc* evaluation of a random 10% of transcripts generated from the story recall task reveals a word-error rate of 28.08%, which is approximately state-of-the-art for large-vocabulary speech recognition. Each transcript is then aligned with its corresponding audio file using the Gentle forced aligner (<https://github.com/lowerquality/gentle>), and then segmented into sentences based on pitch, pause, and parts-of-speech features. The fully segmented transcripts are then scored, as described below. ### Image naming scores Stimuli used in the image naming task are taken from the Caltech-256 Object Category dataset, which are labeled. We measure the similarity (on \[0..1\]) between user input and the set of provided annotations using Wu-Palmer Similarity (WuP) on the ontology provided by WordNet. WordNet is a lexical database that groups English words into synonym sets, and maintains a number of relations among these sets and their members. WuP returns a score denoting how similar two synonym sets (*c*₁ and *c*₂) are, based on the depth of the two senses in the ontological graph, from the root node, and that of their least common subsumer *LCS* (i.e., their most specific ancestor node). Specifically, $$\begin{array}{r} {sim_{wup} = \frac{2 \times depth\left( LCS\left( c_{1},c_{2} \right) \right.}{depth\left( c_{1} \right) + depth\left( c_{2} \right)}.} \\ \end{array}$$ Since words can have multiple senses, we choose the most frequent one. There are 257 stimuli in this task. The average score per stimulus is computed and the distribution of the average score for each stimulus is summarized in. The overall average score for this task is 0.89 with a variance of 0.02 and a skewness of −1.5. ### Picture description scores Picture description is often scored in terms of both syntactic and semantic properties, such as agrammatical deletions and ‘emptiness’, respectively. For the former, we measure language complexity automatically using Lu’s Syntactic Complexity Analyzer (SCA). For the latter, we count the number of information content units (ICUs) in produced transcripts. These ICUs constitute entities, actions, or relations in the scene, and were initially determined through annotation by speech-language pathologists. Since participants may describe an ICU in different terms (e.g., ‘mom’ instead of ‘mother’, or ‘kid’ instead of ‘boy’), we use the Lin Similarity (LS) metric from NLTK to account for lexical variety. LS computes the similarity of two synonym sets (*c*₁ and *c*₂) based on the Information Content (IC) of the the Least Common Subsumer and that of the two input synonym sets. Specifically, $$\begin{array}{r} {sim_{LS} = \frac{2 \times IC\left( LCS\left( c_{1},c_{2} \right) \right)}{IC\left( c_{1} \right) + IC\left( c_{2} \right)}.} \\ \end{array}$$ For the words in the input sentence, all possible senses are considered and we accept an input word as an ICU if the similarity of its closest synonym set is greater than 0.75, determined empirically. For each picture, 10 examples were randomly selected and manually verified against different thresholds. If a word is determined to be synonym of an ICU in the context of the picture but their similarity does not satisfy the above threshold, the word is manually added to the list of ICUs for that picture. A very low value for the threshold results in many words being falsely detected as ICUs. A very large value results in many ICUs not be detected and therefore many synonyms should be added manually to the list of ICUs. The threshold of 0.75 empirically balanced accurately detecting ICUs while minimizing manual annotation. ICUs can also take the form of multi-word phrases (e.g., ‘*hard drive*’). To compare an ICU with *m* words with an input window of *n* words (where *m* ≥ *n* by definition), each word in the ICU must be paired with a word in the input. Note that, as illustrated in, a greedy strategy can result in suboptimal pairings, called maximum weight matching in bipartite graph theory. Therefore, using maximum weight matching, an ICU is detected if the similarities for all words in the candidate are greater than the empirical threshold 0.75. Note that this does not incorporate grammatical dependencies or negations. Pictures can have a relatively arbitrary number of ICUs. Some pictures elicit more or less speech, as shown in Figs and, respectively. ### Fluency scores We extract the same scores as in the Wisconsin Longitudinal Study for the same type of task, including the number of tokens in a category and the number of tokens out-of-category. For the semantic fluency task, we manually construct dictionaries for each stimulus, based on a subset of user responses (e.g., the *‘animal’* dictionary contains the words *‘lion’*, *‘tiger’*, and *‘cat’*. To determine if a word is in- or out-of-category, we first check if it belongs to any of the dictionaries. If the word is not found, we use WordNet to check if the category word is its hypernym. For the letter fluency, we check that each word begins with the given letter, and then verify that the word exists by checking if it can be found in WordNet. Alternative dictionaries may be used, in general. Alzheimer’s disease, for example, has a greater impact on semantic fluency than on other types of fluency. shows the pairwise Kullback—Leibler (KL) divergence between the distributions of the number of tokens in-category, for different instances of the fluency task. For visualization, values are linearly normalized on \[0..1\]. Dark blue elements indicate smaller distances between distributions. Here, *‘drinks’*, *‘occupations’*, *‘cities’*, and *‘fruits’* show similar distributions. As one may expect, *‘letters’* and *‘numbers’* have also similar distributions. The distributions of the number of tokens *out* of category for different instances are examined in. These results suggest that latent subgroups exist in the fluency task, which may be useful in mitigating the practice effect that often occurs in longitudinal analysis. ### Story recall scores We transcribe the audio recordings of story recall, and score the task using the ROUGE score (i.e., ‘recall-oriented understudy for gisting evaluation’). ROUGE is typically used to evaluate automatic summarization software, and compares a candidate summary to a list of reference summaries using the overlap of their *n*-grams. When scoring the story recall task, we use the original text of the short story as the reference and the transcript of the participant’s story retelling as the candidate. We extract ROUGE metrics on unigrams (ROUGE-1) and on bigrams (ROUGE-2). shows the distribution of these scores for different stories, including the ‘Grandfather’ passage, whose lower scores suggest that it is harder to recall. ### Vocabulary scores We use the BLEU measure (i.e., ‘bilingual evaluation understudy’) to score the vocabulary task. BLEU is similar to ROUGE in that it compares oracle-provided reference sentences and candidate sentences, but its focus is precision rather than recall. Specifically, given a brevity penalty: $$BP = \begin{cases} 1 & {\text{if}\mspace{720mu} r < c} \\ e^{1 - r/c} & {\text{if}\mspace{720mu} r \geq c} \\ \end{cases}$$ where *c* is the number of word tokens in the candidate and *r* is the nearest length among references, and $$\begin{array}{r} {p_{n} = C/N} \\ \end{array}$$ is the *n*-gram precision given the number *C* of *n*-grams in the candidate that are in at least one reference and the total number *N* of words in the candidate, then: <img src="info:doi/10.1371/journal.pone.0212342.e008" id="pone.0212342.e008g" /> B L E U = B P ( p 1 p 2 . . . p n ) 1 / n In our case, the user provides the candidate definition, and reference definitions are derived from WordNet, Wiktionary (<http://www.igrec.ca/projects/wiktionary-text-parser/>, and the Merriam-Webster dictionary (<https://www.dictionaryapi.com/>). There are 301 different stimuli. The average BLEU score per stimulus is computed and the distribution of the average score of the stimuli is shown in. We provide examples of vocabulary items with very high and very low scores. The average score of the Vocabulary task is 0.49 with variance of 0.03 and skewness of -0.41. Stimuli around the average may be good candidates for future studies of vocabulary. ### Winograd schema scores The dataset used for the Winograd schema is annotated with correct answers. Participants receive a score of 1 for every correct response, and 0 otherwise. There are 274 Winograd stimuli. The average score per stimulus is computed and the distribution of the average score of the stimuli is shown in. The average score of the Winograd task is 0.75 with a variance of 0.01 and a skewness of −0.46. Similar to the vocabulary task, the stimuli around the average may be a good candidate for future studies. ## Feature extraction We extract both textual features where available (including transcripts from speech recognition), and acoustic features from audio, as described below. From text and transcribed audio of the image naming, fluency, story recall, and vocabulary tasks, we extract lexical, syntactic, semantic, and pragmatic features, as described in. ### Lexical features We automatically extract features related to each <u>word</u> (e.g., the number of syllables per word, and the number of characters per word). We count the number of <u>fillers</u> (e.g., *“uh”, “um”*) and normalize by the total number of word tokens in the sample. To compute <u>vocabulary richness</u>, we calculate the type-token-ratio and the moving-average-type-token-ratio with window sizes of 10, 20, 30, 40, and 50. We also calculate the Brunet index and the Honoré statistic; i.e., $$\begin{array}{r} {BI = N^{(U^{- 0.165})},} \\ \end{array}$$ where *N* is total number of word tokens and *U* is the total number of unique word types, and $$\begin{array}{r} {HS = \frac{100\log N}{1 - \frac{N_{1}}{U}},} \\ \end{array}$$ where *N* is the total number of word tokens, *U* is the total number of unique word types, and *N*₁ is the number of *hapax legomena* (i.e., words used only once). The <u>readability</u> of transcripts is calculated by the Flesch reading score, and the Flesch-Kincaid grade level; i.e., $$\begin{array}{r} {F = 206.835 - 1.015\frac{total\mspace{720mu} words}{total\mspace{720mu} sentences} - 84.6\frac{total\mspace{720mu} syllables}{total\mspace{720mu} words}} \\ \end{array}$$ and $$\begin{array}{r} {FK = 0.39\frac{total\mspace{720mu} words}{total\mspace{720mu} sentences} + 11.8\frac{total\mspace{720mu} syllables}{total\mspace{720mu} words} + 15.59} \\ \end{array}$$ We measure the <u>polarity</u> of transcripts by computing averages and standard deviations of norms derived from the Multi-Perspective Question Answering (MPQA) lexicon and the Stanford Sentiment analyzer. The MPQA lexicon provides values of polarity of words as *“strong negative”, “strong positive”, “weak positive”*, or *“weak negative”*. The Stanford Sentiment analyzer provides values of polarity of words as *“very negative”, “very positive”, “neutral”, “negative”*, or *“positive”*. We extract mean values of frequency, age-of-acquisition, imageability, familiarity, arousal, dominance, and valence based on <u>lexical norms</u>. We compute the mean frequency (with which a word occurs in a corpus) of words based on the SUBTL frequency norms. Age-of-acquisition (i.e., the age at which a person learned a word), imageability (i.e., the ease at which a word can give rise to a mental image), and familiarity (i.e., how often a word is used, seen or heard) are determined from the Bristol and Gilhoolie-Logie ratings. Arousal (i.e., the intensity of emotion), dominance (i.e., the degree of control), and valence (i.e., the pleasantness) of words are derived from the Affective Norms for English Words (ANEW) ratings and the Warriner norms. We also obtain average values for psycholinguistic measures from the Linguistic Inquiry and Word Count (LIWC) corpus and the Receptiviti platform (<https://www.receptiviti.ai/liwc- api-get-started>). ### Syntactic features We count constructs extracted from <u>Lu</u>’s Syntactic Complexity Analyzer (SCA). SCA computes various ratios involving T-units (i.e., main clauses plus their dependent clauses) and complex nominals (i.e., groups of words that describe an entity). We compute the Yngve measure, which is computed from Stanford context-free <u>parse trees</u> and quantifies to what extent a sentence is left-branching rather than right-branching. We extract propositional and content <u>density</u>, respectively: $$\begin{array}{r} {density_{prop} = \frac{verbs + adjectives + adverbs + prepositions + conjunctions}{words},} \\ \end{array}$$ and $$\begin{array}{r} {density_{content} = \frac{nouns + verbs + adjectives + adverbs}{words}.} \\ \end{array}$$ Next, we measure the part-of-speech <u>(POS) counts</u> using the Stanford POS tagger (<https://nlp.stanford.edu/software/tagger.shtml>). These include adjectives, adverbs, coordinate conjunctions, demonstratives, determiners, function words, inflected verbs, light verbs, nouns, prepositions, pronouns, subordinate conjunctions, verbs. We also compute the following <u>POS ratios</u>: $$\begin{array}{r} \begin{aligned} {noun - verb\mspace{720mu} ratio} & {= \frac{\# nouns}{\# verbs}} \\ {noun\mspace{720mu} ratio} & {= \frac{\# nouns}{(\# nouns + \# verbs)}} \\ {pronoun\mspace{720mu} ratio} & {= \frac{\# pronouns}{(\# pronouns + \# nouns)}} \\ {subordinate - coordinate\mspace{720mu} ratio} & {= \frac{\# subordinate\mspace{720mu} conjunctions}{\# coordinate\mspace{720mu} conjunctions}} \\ \end{aligned} \\ \end{array}$$ ### Semantic features We compute semantic similarity using the average and minimum <u>cosine distance</u> between each pair of one-hot embeddings of utterances, and the cosine cutoff (i.e., the number of pairs of utterances whose the cosine distance is below a certain threshold). We compute word specificity and ambiguity based on tree depth and the number of senses in <u>WordNet</u>. We also extract multiple WordNet measures of similarity: Resnik, Jiang-Coranth, Lin, Leacock- Chodorow, and Wu-Palmer. ### Pragmatic features We train a general 100-topic latent Dirichlet allocation (LDA) model on the Wikipedia corpus for generalizability. <u>LDA</u> is a generative statistical model used to determine unlabeled topics in a document. For each transcript, we extract the probabilities of each LDA topic. Next, we extract features related to rhetorical structure theory (<u>RST</u>), which is a classic framework for discourse parsing in which partitions of text are arranged in a tree structure by pragmatic relations such as *Elaboration* or *Contrast*. ### Acoustic features We extract acoustic features from all tasks in which the response is spoken, i.e., the picture description, story recall, and word-colour Stroop tasks. We extract acoustic features with the openSMILE open-source tool, which includes features related to formants, loudness, approximations of pitch, including zero- crossing rate and Mel-frequency cepstral coefficients (MFCCs) among others. Additionally, we extract the following features that are not extracted by openSMILE: 1) total duration, 2) total duration of active speech divided by total duration of the sample, 3) mean length of all pauses (pause \> 150 ms), short pauses (150 ms \< pause \< 400 ms), and long pauses (pause \> 400 ms), and 4) ratio of pauses \> 150 ms to non-silent segments. # Results ## Correlation across different tasks In this section, we evaluate the relations between the performance of subjects on different tasks through correlation analysis. For tasks that are scored with multiple measures, e.g., ROUGE-1 and ROUGE-2 in story recall, we consider all the measures and the results are shown in. We also include age, sex, and the education level in the analysis. Scores within the same task are usually very highly correlated across subjects, as one might expect; therefore, for visualization, we only show correlations between scores *across different* tasks. Additionally, correlation values that are *not significant*, with respect to the *p* = 0.05, are also ignored. We have normalized the scores as follows: Tasks such as picture description have different stimuli, which may affect the scores. To alleviate this effect, scores are equalized according to their cumulative distribution function (CDF). The resulting scores are therefore uniformly distributed between 0 and 1. This technique is also known as ‘histogram equalization’ or ‘dynamic range expansion’. illustrates this process for story recall. The original scores for the ‘grandfather story’ are lower than the ‘rainbow story’, suggesting that it is a harder story to recall (bottom). This has been alleviated in the normalized scores, where all stories have similar distribution (left). For tasks that involve binary questions, such as Winograd, CDF is not helpful because the probability distribution function is a Bernoulli process. However, in those tasks, there are multiple stimuli per session that allows for computing an average over stimuli. Taking into account the fact that some stimuli are harder than others, we adopt a weighted average strategy such that the effect of ‘hard’ questions are reduced. That is, the average score is more degraded if a subject answers an easy question incorrectly. The weight of a question represents its ‘simplicity’ and is defined as the rate of correct responses to that question, computed over all available responses to that question. For subjects with more than one session, normalized scores are averaged over all available sessions. In the Fluency task, the number of tokens ‘in category’ has a moderate correlation with the number of ICUs in the picture description task (*ρ* = 0.44, *p* = 9 × 10⁻⁹) and with performance on the image naming task (*ρ* = 0.43, *p* = 6 × 10⁻⁸). Similarly, the Rouge-2 score in story recall is correlated with number of ICUs in the picture description task (*ρ* = 0.41, *p* = 2 × 10⁻⁷). The Winograd task is correlated with the life satisfaction response in the GDS task (*ρ* = 0.71, *p* = 5 × 10⁻²⁰). ## Principal component analysis of scores within tasks Tasks such as GDS and picture description are scored based on different scoring metrics, which we can combine using principal components analysis. The picture description task includes features of both information content and language complexity. In, the direction and length of the vectors indicate how each scoring metric contributes to the two principal components. For example, from, the Dependent clause ratio (DC/C) and Dependent clauses per T-unit (DC/T), which reflect the amount of subordination, are approximately orthogonal to the Coordinate phrases per clause (CP/C) and Coordinate phrases per T-unit (CP/T), which reflect the amount of coordination. They are also approximately orthogonal to the number of ICUs. This suggests that DC/C and DC/T measure a very different aspect of the task compared to the CP/C and CP/T metrics. Similarly, from regarding GDS, the question about staying at home is approximately orthogonal to the other four questions, which are positively associated with happiness. ## Unsupervised analysis of data In order to further evaluate the generalizability of tasks involving spontaneous speech production, we look for the homogeneity across picture description and story recall tasks. We use t-SNE to visualize *features* across these tasks. This analysis reveals a cluster, indicated with green ellipsoids across. We further investigate characteristics of the cluster with respect to different *scores*. We colour the samples by comparing their score against a threshold, to highlight the homogeneity of the cluster with respect to that score. From, it can be seen that the cluster is associated with high *GDS-Happiness scale* and high *story recall (Rouge-1)* score. ## Data augmentation for the assessment of Alzheimer’s disease In this experiment, we combine our normative data with DementiaBank (DB), described above. We adopt the approach used by Vasquez-Correa *et al* for multi- view representation learning via canonical correlation analysis (CCA) to improve the classification of dementia from healthy controls. CCA computes a projection matrix for every view such that, in the shared space, the correlation between the projected samples from different views are maximized. We construct two views using the Talk2Me database. We consider features from the picture description task as the first view and the concatenation of features from the remaining tasks as the second view. Applying CCA on the Talk2Me database provides two projection matrices corresponding to the two views. We then project the DementiaBank data to the shared space using the projection matrix corresponding to the picture description task. We apply feature selection on the original features of DementiaBank and the selected features are concatenated with the CCA embeddings. Classification is done using an SVM with a radial-basis function kernel. shows that, across five feature selection methods, the normative Talk2me data improves overall accuracy; however, an ANOVA test reveals no significant difference. We would encourage exploring additional tools in domain adaptation to handle domain shift and leverage complementary information could be a direction for future research. Moreover, the Talk2Me dataset may be more effective in application domains where participants have demographics more similar to Talk2Me. In the context of AD, it is important to recruit older adults. We will discuss this in the next section. Following, the number of embeddings is set to 20. We also examined a wide range of values for the number of embeddings (i.e., 10, 20, 30 and 40) and also different combinations of tasks to increase the number of views, but no significant difference was observed. We use 10-fold cross-validation in all cases. In addition, hyper-parameters are selected through an internal 10-fold cross-validation where the RBF kernel width is selected among \[0.01, 1, 10, 100\] and the error penalty parameter is set to 1. The number of selected features is selected among 50 to 300 in steps of 50. These settings are determined empirically. # Limitations We aim to design a platform that can be generalized to various populations, conditions, and tasks; in fact, we have recently applied it to a project involving language delays in elementary school children. However, a limitation of the existing data snapshot is that the age range is skewed towards young adults, and the majority of users have at least an undergraduate degree. For our work in specific demographics, e.g., in the detection of Alzheimer’s disease, it will be important to recruit more data from a wider range of people. Some potential barriers to recruitment include: older adults not knowing about the study, not being able to access the website, not wanting to put their personal information online, or not understanding the interface. These concerns may dissipate over time, as a growing proportion of older adults are using computers and accessing the Internet. For instance, Statistics Canada reports that Internet use among 65- to 74-year-olds rose from 65% to 81% in the period between 2013 and 2016, and from 35% to 50% in those aged 75 and older. We intend to increase recruitment of this population through promoting the study on forums and mailing lists for older adults, and in retirement homes, assisted living facilities, and day programs. Another limitation is the lack of control over recording conditions and environmental noise, which can present a challenge for audio processing. However, this is a consequence of collecting data with a set of microphones and recording conditions representative of the intended use. To be of practical use outside of controlled environments, analyses must be robust against changing channel conditions. In our reported analyses, we have previously attempted to mitigate such factors using spectral noise subtraction, and we have shown that software can reduce the effect of the channel in identifying differences in the voice. Moreover, recent research suggests that “training on different noise environments and different microphones barely affects \[speech recognition\] performance, especially when several environments are present in the training data”. The demographic and personal health information associated with the dataset are self-reported and have not been clinically validated. This can also be a limitation due to the potential for deliberate participant misrepresentation. However, the cost and complexity associated with obtaining individual clinical assessments are not compatible with our goals of large-scale data collection and repeated, on-going participation. # Conclusion and future work We have developed a public portal for ongoing *longitudinal* language data collection from a naturalistic population—there are very few barriers to inclusion. We are releasing the first public ‘snapshot’ of normative data, consisting of 1033 sessions from 196 healthy subjects, including raw data, computed transcripts, features, and scores. We are also releasing a new software package (<https://github.com/SPOClab-ca/COVFEFE>) that extracts a variety of lexical, syntactic, semantic, pragmatic, and acoustic features for generic speech and language analysis. To our knowledge, this is the most comprehensive publicly available software pipeline for extracting linguistic features. The data and tools enable a common dataset to benchmark models, extend existing data sets with more data, including longitudinal data, and more diverse demographics and tasks. To describe these data, we analyze relations between tasks, and provide normative scores. This enables baselines against which smaller clinical data sets can be compared in the future. The Talk2Me dataset may be used to augment smaller datasets, especially those with demographics similar to Talk2Me. Along these lines, we have started to take a multi-view approach based on canonical correlation analysis, trained on Talk2Me, to improve the accuracy of classification. We are currently recruiting more older adults to use the Talk2Me interface through various means, such as in retirement homes, assisted living facilities, and day programs. Also, we are currently extending the Talk2Me data collection tool to include a telephone-based interface. The telephone-based version of Talk2Me relies on interactive-voice-response and uses the same tasks as in the web-based version, except for Stroop. Data collection for both the web- and telephone- based systems is ongoing, and we are focusing our efforts on populations of older adults with and without dementia and cognitive decline. # Supporting information [^1]: The authors of this manuscript have read the journal’s policy and have the following competing interests: KCF, MY, and FR are co-founders of a company, WinterLight Labs Incorporated, that commercializes the automated analysis of spontaneous speech and other language tasks. WinterLight Labs provided no influence or financial support, nor has WinterLight Labs received any benefit from this work beyond what is being made public. MY, who was a student at the inception of the study at the University of Toronto, was an employee of WinterLight Labs during the final editing of the document. There are no patents, products in development or marketed products associated with this research to declare. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

# 1. Introduction Eco-cities, which pour effort into eliminating the overall carbon footprint of the city whilst helping the humans and the nature to coexist aiming towards sustenance and sustainability. Eco-cities have witnessed rapid growth in these years worldwide. As the Eco-cities entering operation stage gradually, more and more researchers have found that users (who are living or working in the Eco- cities) satisfaction is one of the most important factors to determine the success or failure of Eco-cities. Kitakyushu eco-city is recognized the most mature model Eco-city project in Japan. The KPIs (key performance indicators) of this project are considered successful based on the quantified data of residents’ satisfaction degree, which indeed improved conditions for the aging population of this city. Another well-known example of Eco-city is Sino- Singapore Tianjin Eco-city (SSTEC) of China, which generally accepted as the flagship project in China and one of the most successful Eco-cities in the world. The construction and operation efforts of this Eco-city have been mainly put into aspects which residents able to truly experience, such as natural environment, man-made environment and life style. By contrast, one of the key reasons why Masdar city (Abu Dhabi, UAE) and Dongtan city (Shanghai, China) considered as failure projects, is that both of them placed too much emphasis on realization of zero-carbon, and mandated the residents to adjust with lesser ‘comforts’ than they are used to. Therefore, it is very important to investigate the user demands to attract more citizens willing to live or work in the Eco- cities, which will make the development of Eco-cities more sustainable and solid. Caprotti et al. carried out fifteen interviews with the residents in Sino- Singapore Tianjin Eco-city of China in terms of the lived experiences, and found that social sustainability need to be paid more attention. Liu et al. utilized Structural Equation Model (SEM) to investigate the residents’ repurchase intention of Green Residential Building (GRB) in Eco-city of China. Marciniak et al. conducted structural interviews with the visitor perception of informal green spaces in Poland and concluded that informal green spaces are important complementary to formal green spaces in the city. To examine residents’ WTP (willingness to pay) for GRBs (green residential buildings) and its determinants, Liu et al. conducted a survey among 511 current GRB occupants living in Sino-Singapore Tianjin Eco-city in China, and latent class regression was used to analyze the heterogeneity of their preferences. Sun et al. collected the social media data, by applying the artificial intelligence technique to analyze the visitor responses to the green and open spaces in Shenzhen, China. Friederike et al. conducted a quantitative questionnaire survey among people aged 50 years and older throughout the city of Berlin, to explore the older people’s urban green space visitation patterns. Erik et al. used a discrete choice experiment to explore people’s preferences and willingness to pay for green features in an urban Neighborhood Management development zone in Berlin. Monteiro et al. investigated the citizen demand of sustainability for urban forests based on i-Tree Eco surveys. The above recent researches on user demands investigation and analysis in the Eco-cities mainly focused on understanding the user need itself, yet lack of research on the relationship between the user demand and user satisfaction. Because improving the user (citizen who lives or works in the Eco-city) satisfaction will be the ultimate purpose of construction and operation of the Eco-city, yet user demands survey is just the process. Actually, the classification of user demands are various, same efforts to different classification of user demands may lead to different degree of satisfaction. In other words, different classification of user demands in Eco-city may need different strategies to satisfy. Therefore, an analysis method on revealing the relationship between user demands and user satisfaction was initially introduced to research field of Eco-city in this paper. The Kano model is a famous and important theoretical and quantitative model for the research of customer satisfaction towards product quality attribute for different industries. Aliyu et al. utilized integration method of Kano model and quality function deployment (QFD) to investigate the user demands for sport earphone. Wu et al. conducted a study adopted Kano two-dimensional quality model to investigate the users’ needs on twenty service attributes of rehabilitation buses. Juan et al. exploring sustainable planning strategies for public housing in Taiwan, the Kano model was adopted as a theoretical base. Ma et al. used Kano model to differentiate between future vehicle-driving services demands. Xu et al. presented a requirements analysis method based on Fuzzy Kano Model, which to improve the quality of virtual reality interior design software. Yao et al. conducted the Kano model analysis of features for mobile security applications to gain more customer satisfaction. Chen et al. investigated pharmaceutical logistics service quality with refined Kano’s model involving 104 respondents from medical institutions. Zhang et al. studied on enhancing readers’ satisfaction model of electronic service quality in library based on LibQUAL+ and Kano model, and result shows that the model they built is feasible through the empirical analysis. However, for the research field of Eco-city, existing studies do not involve the Kano model to explore user (citizen who lives or works in the Eco- city) demands and the affection for the user satisfaction to date. This paper aims to initially introduce the Kano model analysis method to the research field of user demands and satisfaction in terms of the Eco-city, intends to explore the relationship between the user demands and user satisfaction. Based on the Kano model, the user demands classification and importance for the Eco-city can be defined accordingly. The user demands analysis method of Eco-city proposed in this paper, can be used for other researchers worldwide to investigate and quantitively analyze user demands according to their local development situation and preference of Eco-city. The user demands analysis results obtained through this method, can benefit different stages of Eco-city. For the planning and design stage of Eco-city, user demands analysis can help avoid ‘empty city’ phenomenon after construction, so as to save huge amount of fund and time. Because the construction of Eco- cities is extremely complicated and time-consuming, the developing organization of Eco-city needs to allocate limited resources investing in facilities and services of more critical user demands. For the operation stage of Eco-city, user demands analysis can help conduct POE (Post Occupancy Evaluation) process, in order to reveal aspects that residents dissatisfied in terms of operation and management of Eco-city. The O&M organization of Eco-city will be able to undertake relevant renovations accordingly regarding facilities and services to maximize user satisfaction. # 2. Materials Since the Chinese national standard "Assessment Standard for green eco-district" (GB/T 51255–2017) has been promulgated and implemented, the standard has strong authority, a large number of credits, and rich evaluation content. Therefore, the user demands library had been proposed based on the national standard. Regarding the national standard "Assessment Standard for green eco-district", from the perspective of evaluation dimensions, it mainly includes government management dimension and user dimension. For example, the credit "4.1.1 urban planning should meet the urban and rural planning requirements of the region." It can be classified as the content of government management dimension. The credit "4.2.5 Public service facilities in residential areas has better convenience." can be classified as the content of user dimension, which reflects the actual needs and expectations of the people for an Eco-city. Therefore, this article had started from the user dimension, combed and categorized the content of the national standard "Assessment Standard for green eco-district" (GB/T 51255–2017), and initially summarized the user demands library including six major categories. The categories are land use, ecological environment, green building, energy utilization, green transportation and humanities, including 25 specific user demands, which are shown in. # 3. Methodology China’s new urbanization process and the implementation of green development policies have jointly promoted the development of the Eco-cities. With the construction and operation of the Eco-city, user demands are constantly changing. According to the actual needs of users and feedback of problems encountered in the operation process, the Eco-city shall be continually improved and updated, and the functions and services need to meet user demands accordingly. In order to specifically understand the real demands of users in the operation of the Eco-city, and to improve the service quality, thereby enhancing the happiness index of users. The research process undertaken in this paper was as follows. Firstly, the user demands classification analysis for Eco-city based on Kano model was researched, including the questionnaire method and computation formula. Secondly the user demands importance analysis for Eco-city was studied. AHP method was adopted to obtain the initial weights of user demand items. Then weight adjustment was conducted based on the Kano demand category. Consequently, the final ranking of user demands importance was determined accordingly. At last, two Eco-cities in Tianjin and Chongqing city of China were selected to apply the methodology as case study. See for an illustration of the research process undertaken in this article. ## 3.1. User demands classification analysis of Eco-city based on Kano model ### 3.1.1. Introduction of the Kano model The Kano model was an important theoretical model which was initially proposed by Japanese quality management professor—Noriaki Kano in 1984 aiming to illustrate and identify quality attributes for the research of customer satisfaction. Kano model provides a framework which could enable the elicitation of product requirements. It helps increase customer satisfaction if the elicited requirements are met. Kano model not only could be used for requirement clarification in the early stage, but also for different stages in the service delivery lifecycle. Kano survey helps to add value by focusing efforts in service design, development and verification stages to encompass features on use case level, supported by early prototypes and conducted with real customers. Compared with other models, the Kano Model does not assume the existence of a linear relationship between product/service performance and customer satisfaction. Kano noticed that customers’ requirements are not equivalent and that some requirements, in fact, are capable of generating more satisfaction than others. Moreover, customer satisfaction is not always proportional to the functionality of the good, which implies that higher quality does not necessarily lead to higher satisfaction for all product attributes or services requirements. The customer satisfaction over specific quality of a product or service may vary with customer’s preference over the quality attribute as shown in. The X -axis stands for the level of quality performance (from Insufficient to Sufficient) and the Y -axis represents the customer satisfaction level (from Dissatisfaction to Satisfaction). There are 5 quality attributes in Kano Model as follows: 1. One-dimensional quality (O): Customer is satisfied when this quality attribute is sufficient and vice versa. Customer satisfaction level is in liner relation with quality attribute adequacy. 2. Attractive quality (A): Sufficiency of this quality attribute will lead to more satisfaction of customer, yet not cause dissatisfaction with the absence of this attribute. 3. Must-be quality (M): Customers tend to consider this quality attribute for granted. Which means improving this attribute will not result in more satisfaction, but the customer will become very dissatisfied if this quality attribute not provided. 4. Indifferent quality (I): Whether this quality attribute sufficient or not, will not affect customer satisfaction. The customer is not interested with the product or service quality. 5. Reverse quality (R): Customer does not desire this product attribute and also expects the reverse. ### 3.1.2. The questionnaire of Kano model A questionnaire can be constructed according to Kano model. The questionnaire consists of questions about the functional requirements of a product. Each functional requirement consists of two inverse questions, one functional and one dysfunctional. For example, for the functional question, the customer (user) might be asked “If there are sufficient service facilities around the residential area, which are very convenient and accessible, how do you feel?”. For the dysfunctional question, the customer (user) might be asked “If there are insufficient service facilities around the residential area, which are not convenient and accessible, how do you feel?”. Each functional and dysfunctional question has five possible answers, (1) I like it that way; (2) It must be that way; (3) I am neutral; (4) I can live with it that way; (5) I dislike it that way. ### 3.1.3. User demand category analysis of Kano model The form shown in can be utilized to evaluate the quality attribute category of demand by each customer (user), including One-dimensional (O), Attractive (A), Must-be (M), Indifferent (I), and Reverse (R). “Q” represents unusable response which will be eliminated. Then all the Kano categories for each demand item of all customers (users) are summarized to determine the final Kano category for each demand item, with adopting the principle of relatively majority. ### 3.1.4. Computation of CS and DS of Kano model After collecting all the responses from customers (users) towards each demand item in terms of the Kano category (One-dimensional, Attractive, Must-be, etc.), it is possible to calculate two coefficients, namely customer satisfaction (CS) and customer dissatisfaction (DS). The customer satisfaction coefficient has a value between 0 and 1 (values close to 1 represent great satisfaction while values close to 0 indicate low satisfaction). The customer dissatisfaction coefficient has a value between -1 and 0 (values close to -1 represent great dissatisfaction while values close to 0 indicate low dissatisfaction). The calculated equations are as follows: $$CS = \frac{A_{i} + O_{i}}{A_{i} + O_{i} + M_{i} + I_{i}}$$ $$DS = \frac{M_{i} + O_{i}}{A_{i} + O_{i} + M_{i} + I_{i}}.\left( {- 1} \right)$$ When *A*_*i*, *O*_*i*, *M*_*i*, *I*_*i* stands for the number of Attractive, One- dimensional, Must-be, and Indifferent attribute respectively of demand item i. ## 3.2. User demands importance analysis of Eco-city based on Kano model ### 3.2.1. The questionnaire of user demands importance The user demands survey questionnaire of Kano model can carry out qualitative analysis of classification for each demand. However, to understand the comprehensive importance of each user demand in detail, further user surveys are needed. Therefore, Likert five-point scale method was used for the importance survey questionnaires, setting "very unimportant", "relatively unimportant", "generally important", “relatively important” and "very important" for each user demand item. These five levels correspond to scores of 1, 2, 3, 4, and 5 respectively. The importance questionnaire sample form of user demand in Eco- city indicates as follows in. ### 3.2.2. Determination of initial weight The most mature AHP (Analytic Hierarchy Process) method in the industry was used to determine the initial weights of the importance of 25 user demands in the Eco-city. This method can summarize different factors related to decision- making, and perform qualitative and quantitative evaluation. When using this method for calculation, the judgment matrix needs to be constructed firstly. The way to construct the judgment matrix is: calculate the average score of each analysis item, and then divide the average score for two items successively to form judgment matrix. For the user demands importance analysis of eco-city, 25-order judgment matrix was constructed (because 25 user demand items in the survey). After summarizing the results of importance questionnaire for eco-city, the average score for each demand item was calculated, and the judgment matrix was shown in (as the whole matrix was too large, only 10-order matrix shown for demonstration). The judgment matrix analysis used to get the eigenvector and initial weight of each user demand item. Then a consistency test should be conducted. The maximum eigenvalue according to the vectors of matrix needs to be calculated, and the maximum eigenvalue utilized to get the CI (consistency index = (maximum eigenvalue–n)/(n-1)). RI (Random Consistency Index) can be referred to according to the order of matrix (For example, the RI of 25-order matrix is 1.6556). Finally, a consistency test is needed to calculate the CR (consistency ratio = CI/RI). Under normal circumstances, the smaller the value obtained by CR, the better, because this represents better consistency of the key calculation matrix and the normalization. Generally speaking, the value of the calculated correlation matrix should not be greater than 0.1. ### 3.2.3. Weight adjustment of user demand items based on Kano model Initially, the user’s awareness is the only way to determine the initial importance, and it is difficult for users to notice the impact of their own needs on satisfaction, which leads to the great limitations of this evaluation method. Through the analysis of user feedback after the return visit, it is found that most users will emphasize their essential functions when using them, or hope that the product can increase the actual functions they need, but almost no users express their needs for attractive attributes. Attractive attributes can bring unimaginable parts to users in the subconscious to increase their satisfaction, which is an aspect that users often overlook when using products. Therefore, in the final weight analysis, the initial weight of the user demand item cannot be directly used to analyze the importance of the user demand item. It is also necessary to consider whether the demand item is helpful for improving user satisfaction. This paper adjusted the initial weight of each demand item based on the theory of the Kano model, so as to obtain the final weight of user demand in the Eco-city. The adjustment formula is as follows: $$w_{i}^{\prime} = \frac{w_{i} \times m_{i}}{\sum_{i = 1}^{n}w_{i} \times m_{i}}\left( {i = 1,2,\ldots,n} \right)$$ Where *w*_*i*—Initial weight of user demand item i; $w_{i}^{'}$—Final weight of user demand item i; *m*_*i*—Kano classification adjustment coefficient of user demand item i; When the user demand item classified to Attractive attribute, *m*\>1; When the user demand item classified to One-dimensional attribute, *m* = 1; When the user demand item classified to Must-be attribute, 0\<*m*\<l. ### 3.2.4. Ranking of user demands importance Because the final weight of demand items not only reflects the user’s evaluation of the importance of each demand item, but also reflects the impact of each demand item on improving user satisfaction, so the comprehensive importance of user demand items can be ranked according to the final weight. In the operation and management stage of the Eco-city, the operation and management works are extremely complicated. The Eco-city operation manager can refer to the importance of the user demands, and put more resources and energy on the most important demand items. Which can effectively improve user satisfaction and reputation, and truly achieve the goal of "people-oriented". ## 3.3. Case study of user demands analysis of Eco-city based on Kano model ### 3.3.1 Basic information of eco-cities selected for case study Sino-Singapore Tianjin Eco-city and Chongqing Yuelai Eco-city of China were selected to apply the methodology of user demands analysis as a case study. The specifications of the two Eco-cities are described in. ### 3.3.2 Survey questionnaire preparation After determining the user demands items that need to be investigated in section 2, this paper designed a user survey questionnaire according to the requirements of the Kano model. The survey was mainly divided into three parts. The first part was to conduct a basic background survey of the users in the Eco-city. This part sets a total of eight questions, such as gender, age, education, occupation, location of work and living, and the awareness of the concepts of Eco-city. The second part was to ask functional and dysfunctional questions for the demand items of the Eco-city mentioned in this article, so as to understand the attitudes and expectations of the surveyed users for each user demand item. The third part is the survey of the importance of user demands in the Eco-city. ### 3.3.3 Distribution and collection of survey questionnaires The user demands questionnaire survey was conducted from September 1^st to October 31^th, 2019. Sampling size (n) was computed which should be more than 123 by using the formula below. $$n = \frac{Z^{2}NPQ}{\left( {ND^{2}} \right) + \left( Z^{2}PQ \right)}$$ Where n = Sampling size, N = Population size, Z = confidence coefficient (1.96 for % 95), P = probability of measurement in population (0.8), Q = 1- P (0.2) and D = sampling error (0.1). The online questionnaires via a website (<https://www.wjx.cn>) were distributed to users of the selected Eco-cities for case study (people who work or live in Sino-Singapore Tianjin Eco-city and Chongqing Yuelai Eco-city of China). The survey participants were able to finish the online questionnaires generated by the website either by computer or mobile phone. All the data was collected anonymously. A statement “All data collected will only be used for research of user demands analysis in terms of eco-city, submission of the questionnaire means that you give consent to the data collection. Thank you!” was included in the questionnaire. A total of 156 questionnaires were returned in this survey which could fulfil the sampling size computed (more than 123). # 4. Results ## 4.1. Distribution and collection of survey questionnaires A total of 156 questionnaires were returned for this survey in the selected eco- cities. Among the 156 collected questionnaires, 26 invalid questionnaires were excluded, because they both selected the “Like” or “Dislike” for functional question and dysfunctional question corresponding to “Q” according to in this paper. According to the collected 130 valid questionnaires, the basic information summary of the surveyed users is shown in. Reliability and validity tests of collected valid survey questionnaires were conducted before further analysis of Kano category and importance of user demands. This process requires the use of SPSS software. The final results of the reliability and validity of the Kano questionnaire and importance questionnaire are shown in Tables and. The Cronbach’s α coefficient of the responses to functional questions, dysfunctional questions and importance questions are 0.948, 0.986 and 0.993 respectively. Obviously, these three data are very close to 1, which shows that the data obtained through the Kano questionnaire and importance questionnaire with high quality of reliability. The KMO values of the responses to functional questions, dysfunctional questions and importance question are 0.819, 0.951 and 0.960 respectively which are all more than 0.8, which demonstrates good validity. ## 4.2. Analysis of Kano category results of user demands The answer data of each surveyed user was sorted and analyzed correspond to the Kano evaluation form, to determine the Kano category for each demand item for the user. Then all the Kano categories for each demand item of all users were summarized to determine the final Kano category for each demand item, with adopting the principle of relatively majority. According to the Kano category distribution of each demand item, the CS value (the Customer satisfaction coefficient) and the DS value (the Customer dissatisfaction coefficient) were calculated for each demand item, to obtain the impact of each demand item on the level of user satisfaction. The Kano category distribution of each user demand item in the Eco-city, and the calculated CS value and DS value for each demand item are summarized in. According to the analysis results of the Kano category of each demand item in, the user demands of the Eco-city were mainly including three categories: Attractive demand (A), One-dimensional demand (O), and Must-be demand (M). The Kano categories of user demands are summarized in. There was only one user demand item that belongs to the Kano category of Attractive demand, which was D-14 Green operation, and the demand item belongs to major category “Green building”. It can be seen from that there was a large gap between the absolute value of the CS value (the influence of the demand item on improving user satisfaction) and the absolute value of the DS value (the influence of the demand item on reducing user satisfaction). The absolute value of CS value was above 0.5, but the absolute value of DS was small, which means that when this demand item has a high degree of satisfaction, it can significantly enhance user satisfaction, but with a low degree, it will not bring users strong dissatisfaction. It is widely known that green buildings are the basic units and cells of an Eco-city. Studies have shown that 80% to 90% of a person’s life spent inside buildings. If the buildings in the Eco-city cannot be truly green, then the overall Eco-city also difficult to truly establish a reputation. Therefore, the improvement of Attractive demand (D-14 Green operation) can make the overall service level of the Eco-city more perfect. There were three user demand items belonging to the Kano category of Must-be demand, with the serial number D-1, D-13, and D-22. The major demand categories involved were “Land use”, “Green building”, and “Green transportation”. It can be seen from that the absolute value of DS of the three demand items was relatively high, while the absolute value of CS was relatively low. When such demand items are not available or have a low degree of supply, it will greatly reduce the user satisfaction level. Therefore, the operation and management organization of Eco-city needs to ensure that the Must-be demands are met for users. From the perspective of the major demand categories, the operation and management organization of Eco-city should continuously improve the supporting service facilities around the residential area (such as kindergartens, nurseries, primary and secondary schools, elderly service facilities, health service centers, commercial service facilities, etc.) to ensure that basic service requirements of the residents in the Eco-city are met, and service levels constantly improved. In addition, because the construction of an Eco-city is not achieved overnight, the construction project basically in the state of being constructed and put into use at the same time, and the goal of realization for green construction must be achieved. With the improvement of people’s living standards, motor vehicles have become a necessity for families, and as a harmonious and livable Eco-city, "difficult parking" should not become a problem that plagues residents’ daily life. There are 21 user demand items that belong to the Kano category of One- dimensional demand, and the major demand categories cover “Land use”, “Green building”, “Energy utilization”, “Green transportation” and “Humanities”. The absolute value of CS value and DS value for the 21 demand items were all relatively high above 0.5, and the difference between CS and DS was very small, which has a greater impact on user satisfaction and dissatisfaction, directly proportional to user satisfaction. One-dimensional demand items are the key functional requirements that the Eco-city needs to pay attention to. Therefore, the operation and management organization of Eco-city needs to pay attention to the maintenance and improvement of such demand items, and try their best to meet such needs, so as to maximize the user satisfaction and reduce user dissatisfaction. The operation and management organization of Eco-city needs to pay attention to the expectations and needs of users from multiple perspectives of land use, green buildings, energy utilization, green transportation, and humanities. However, the operation and management of Eco-city are extremely complicated, and the organization needs to allocate limited resources investing in more critical user demands. ## 4.3. Analysis of the importance of user demands ### 4.3.1. Analysis of initial weight results for user demand item In this paper, AHP analysis method was used to calculate the initial weight of the importance of each user demand item of the Eco-city, and the calculation tool was SPSS software. 25-order judgment matrix was constructed, the eigenvector and initial weight value for each user demand item calculated by AHP method are Shown in. And the calculated results have passed the consistency test of AHP method (CI = 0.000, RI = 1.656, CR \< 0.1). ## 4.3.2. Analysis of adjustment weight results for user demand item Weight adjustment analysis were conducted for each user demand item in the Eco- city combining with the classification of the Kano categories, which mainly involving Attractive demands (A), One-dimensional demands (O), and Must-be demands (M) in this paper. When a user demand item belongs to the category of One-dimensional demand, the user’s satisfaction with such a demand item is directly proportional to its degree of availability. Therefore, the value of adjustment coefficient (m) usually equals to 1.0. When the user demand item belongs to the category of the Attractive demand, compared with the One- dimensional demand item, it is an unexpected demand of the user, which is more helpful for improvement of user satisfaction. Hence the value of m is usually greater than 1.0. When the user demand item belongs to the category of the Must- be demand, the user considers that this kind of demand item is the prerequisite demand item. When such demand is not fulfilled, the user will be very dissatisfied. However, even if such demand items have a high degree of availability or exceed user expectations, the degree of user satisfaction will not improve significantly. Such demand items are not very helpful in improving user satisfaction. Therefore, the value of m for Must-be demand items is often selected from 0 to 1.0. For the user demand items of the Attractive demand and Must-be demand, the value of the adjustment coefficient m has not been fixed in the existing research. By referring to the relevant literatures, the value of adjustment coefficient m for the Attractive demand was assigned as 1.1 in this paper, and the value of adjustment coefficient m for the Must-be demand was assigned as 0.9. Then the final weight of each demand item was calculated with initial weight and adjustment coefficient according to the formula in section 3.2.3. The values of the adjustment coefficient m and final weights for the 25 user demand items are shown in. ### 4.3.3. Importance ranking result analysis for user demand item The 25 user demands items were sorted in descending order according to their final weights, the overall importance ranking of user demand items in the Eco- city are shown in. Compared with Kano category classification analysis for the 25 user demand items, it is found that the Attractive demand and the One-dimensional demand ranked higher than the Must-be demand. These two types of demand items will have a relatively large impact on improving user satisfaction, indicating that the demand items that users are more willing to accept, and also the more important demand items. The importance ranking results reflect not only the importance of each demand item, but also the degree of influence on the improvement of user satisfaction. From the perspective of the major demand categories of user demands library in Eco-city mentioned in section 2 of this paper, as for the top 10 demand items, "Green operation", "Walking system" and "Green space opening" belong to the major demand categories of "Green building", "Green transportation" and "land use" respectively. The other demand items are all demand items in the major demand category of "Ecological environment", which means that the users of the Eco-city pay more attention to the ecological environment. Ecological environment is an important field and the carrier of key indicators for the operation monitoring of the Eco-city. The major categories of demand items ranked 11–25 are basically sorted into "Green transportation", "Humanities", and "Energy utilization" according to their importance. It can be seen that with the gradual improvement of people’s living standards, users of Eco-cities have relatively low attention to the effective use of resources. Compared with them, they pay more attention to the convenient transportation and humanized functions of Eco-cities. # 5. Discussion This study mainly focused on the analysis of user demands of Eco-city based on the Kano Model. The methodology can be used for other researchers to investigate and quantitively analyze user demands according to local development situation and preference of Eco-city. However, there are still some limitations of results in this paper. Firstly, due to time constraints, the questionnaires were mainly distributed to the citizens who live or work in Sino-Singapore Tianjin Eco-city and Chongqing Yuelai Eco-city of China. As the development and construction of Eco-cites in China, there were several Chinese Eco-cites entering the operation stage successively. Extensive surveys need to be conducted in the future for different types of Eco-cities in different regions of China, to explore the similarity and difference of user demands in the Eco-cities of China. Secondly, based on the user demands importance ranking results in this study, the green operation of green buildings and ecological environment are the most important demands of the two surveyed Eco-cites. The results in this paper in terms of user demands importance will be delivered to the management organization of Tianjin and Chongqing Eco-cites of China. They will be able to improve the facilities and services according to the results, and user satisfaction survey will be conducted to further validate the analysis results proposed in this paper. # 6. Conclusions This paper initially introduced the Kano model analysis method to the research field of user demands in Eco-city, to explore the relationship between the user demand and user satisfaction. Based on the user demand analysis of two Eco- cities as case study, the following conclusions can be drawn: 1. The user demands analysis method based on Kano model for Eco-city proposed in this paper has been validated to be feasible. Comparing with the existing literatures in terms of user demands research for Eco-city, the user demands analysis method based on Kano model of this paper, is able to reveal the influence degree of user satisfaction towards the facilities and services provided in the Eco-city. Referring to the importance ranking adjusted by Kano model of user demands, the operation and management organization of Eco-city will be able to allocate limited resources investing in facilities and services for more critical user demands to maximize the user satisfactions, which will truly achieve the goal of "people-oriented" for Eco-city. 2. Based on the case study for two Eco-cities in China, it can be found that: 1) Compared with Kano category classification analysis for the 25 user demand items, the Attractive demand and the One-dimensional demand ranked higher than the Must-be demand. These two Kano categories of demand items will have a relatively large impact on improving user satisfaction, indicating that the demand items which users are more willing to accept, and also the more important demand items. 2) From the perspective of the major demand categories of user demands library in Eco-city mentioned in this paper, the top 10 demand items obtained, "Green operation", "Walking system" and "Green space opening" belong to the major demand categories of "Green building", "Green transportation" and "land use" respectively. The other demand items are all demand items in the major demand category of "Ecological environment", which means that the users of the two Eco-cities as case study pay more attention to the ecological environment. Although this study conducted user demands analysis selecting two Eco-cities in China as a case study. The method of analysis of user demands classification and importance ranking for Eco-city can be applied to other areas around the world. Firstly, the researcher should select the most relevant local standards and regulations in terms of Eco-city to establish a user demands library in Eco- city. Secondly, based on the user demands library in Eco-city, the questionnaires of Kano model and importance survey need to be distributed to the local citizens who live or work in the sample Eco-city. Thirdly, the quantitative analysis method of Kano model, the initial weight and weight adjustment analysis method for user demands can be introduced to figure out the final importance ranking for the user demands of local citizens in Eco-city. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Tuberculosis and tobacco smoking are two major public health issues that cause millions of deaths every year. An estimated 830,000 people diagnosed with tuberculosis (TB) were linked to tobacco smoking in 2017 worldwide. A study based on mathematical modeling estimated that smoking could increase the number of TB cases by 18 million between 2010 and 2050. The effects of smoking are limited not only to lung health but also to the general defensive immune system of humans, which could be why smokers are also at increased risk for extrapulmonary tuberculosis. There is consistent evidence that tobacco smoking is associated with poor TB treatment outcomes and delayed sputum and culture conversion, which indicates a longer period of infectiousness among TB patients who smoke and increases the risk of recurrence after successful anti-TB treatment. Generally, TB cases who were LTFU represent approximately 5.3% of the total TB population, while TB patients who smoke had a higher proportion of loss to follow-up from TB treatment at double to triple rates. This situation warrants significant concern, as these patients are at higher risk for reactivation of TB infections as well as developing multiple drug-resistant tuberculosis (MDR-TB). TB LTFU is influenced by a myriad of interrelated factors, such as patient beliefs and personal factors, health system and service factors, economic factors, including poverty and gender discrimination, and the social context in terms of social support and TB-related stigma. Many of these factors require qualitative input and further exploration of losses to follow-up TB cases, which is beyond the scope of this study. In industrialized urban areas, where smoking and TB are prevalent, the determinants for LTFU consist of low socioeconomic status, working-age population, previously treated TB cases, TB-HIV coinfection, and patients with moderate lesions on chest X-rays. Smokers tend to have a lower socioeconomic status, which suggests potential additional financial barriers for them to travel and complete their TB treatment. The current TB control program in Malaysia utilizes a directly observed treatment (DOT) strategy to manage patients receiving TB treatments and to prevent the LTFU issues among TB patients. Newly diagnosed TB patients must come to the nearest chest clinic to take their TB medication through daily dosing anti-tuberculosis regimens during the intensive phase (1^st two months of TB treatment). Continuation of DOT during the maintenance phase depends on the sputum seroconversion status, clinical manifestations, and adherence to TB treatment. Patients with fewer complications during TB treatment and who exhibit good adherence to their medication regimen will be allowed to continue taking their anti-TB drugs at home with supervision by family members, community representatives or NGO volunteers through a modified DOT strategy, while patients who live alone are required to take their medication through self- administered treatments. Then, all patients must attend physical visits for follow-up at 1- to 2-month intervals at TB clinics. These methods have been practiced for more than a decade, yet the LTFU rate in the country remains high. In addition, patients with a previous history of LTFU will tend to redefault, as the underlying factors leading to this outcome were not identified and remain unsolved. The development of a multivariable prognostic model includes identifying the important predictors, assigning relative weights to each of the predictors and estimating the performance of the model through calibration and discrimination, and its potential for optimism using internal validation techniques. Several predictive models have been developed to identify patients with TB who are at risk for unsuccessful TB treatment outcomes. However, to date, there has been little information or synthesis of the prognostic factors for LTFU among TB patients who smoke. Targeted interventions for high-risk subpopulations is a way to improve the successful treatment outcome rate among the TB population. In this study, we aim to develop a simple prognostic scoring checklist to predict LTFU outcomes among TB patients who smoke using readily available information during initial patient follow-ups at TB clinics. The model can stratify the targeted population into different risk groups for intervention and management prioritization. # Materials and methods ## Study design This was a retrospective cohort study that involved the development and internal validation of a prognostic model. We utilized a cohort of confirmed and registered TB patients in Selangor state that was obtained from the National MyTB database version 2.1 from 2013 until 2017. The population of Selangor state was selected because it had recorded the highest number of TB cases in peninsular Malaysia each year and has the highest population percentage in Malaysia (20.1% of the total Malaysian population). Selangor is also one of the urbanized states in peninsular Malaysia, with the highest employment rate and job opportunities, especially in the industrial sector. ## Database The National MyTB database is a national surveillance database that consolidates all data on TB cases in Malaysia and is owned by the Disease Control Division, Ministry of Malaysia. TB is a notifiable disease in Malaysia by law under the Prevention and Control of Infectious Disease Act, 1988 (Act 342). Every medical practitioner who treats or becomes aware of a TB infection at any location shall notify the case to the nearest health office or government health facility within seven days of diagnosis. TB cases are notified via the tuberculosis information system form (TBIS 10A-1) and recorded in the state MyTB database, which is managed by the state TB task force team. Patients’ data in the surveillance database are consolidated at the national level, and TB treatment outcomes were defined and recorded after one year of surveillance. Information available from the database includes patient sociodemographic profiles, past medical and comorbidity profiles, tuberculosis disease profiles, laboratory and radiology information and TB treatment profiles. ## Data processing and eligibility criteria All adult TB patients aged \>18 years who were current smokers at the time of their diagnosis were considered for analysis. For TB cases that were initially registered as TB but were changed to other than TB cases, cases with missing data on treatment outcomes and duplicated cases were removed from the dataset. Cases with multiple drug-resistant TB (MDR-TB) were also excluded from the analysis, as the treatment outcome definition for MDR-TB cases is different from the definition for non-MDR-TB. The outcome being measured in this study is the loss to follow-up (LTFU) outcome. Other TB outcomes such as treatment failure, death and outcome not evaluated, were not included in the analysis. ## Operational definition The classification of TB treatment outcomes are defined according to the Clinical Practice Guideline of Tuberculosis by the Ministry of Health, Malaysia and the definition and reporting framework for TB by the WHO as follows: 1. Loss to follow-up (defaulted in the past): TB patient who did not start treatment or whose treatment was interrupted for two consecutive months or more. 2. Cured: Bacteriologically confirmed TB patient who was subsequently smear-or-culture-negative during the last month of treatment or on at least one previous occasion. 3. Completed treatment: Patient who completed TB treatment without meeting the criteria for cure or treatment failure. 4. Treatment failure: TB patient whose sputum smear or culture was positive at five months or later during TB treatment. 5. Died: TB patient who died for any reason before starting or during TB treatment (all-course mortality) 6. Outcome not evaluated: TB patients with no assigned treatment outcome, including TB cases who were "transferred out" to another country for whom the treatment outcomes were not known. Successful TB treatment outcomes were cases that were cured and had completed TB treatment, while unsuccessful treatment outcomes included all other outcomes (e.g., loss to follow-up, treatment failure, death and outcome not evaluated). The smoking status in this study was obtained by confirming whether the patient was a "current smoker" or not at diagnosis. A current smoker was defined as a patient who currently smoked at least one tobacco product every day (daily smoker) or less than daily (occasional smoker). ## Statistical analysis The dataset was randomly divided into two cohorts with a 1:1 ratio using IBM SPSS version 26.0 software. The development cohort was used to develop the prognostic model, while the other cohort in the dataset was used for internal validation. Sociodemographic profiles, disease profiles and patient comorbidity variables were reported as frequencies and percentages. The differences across both cohorts were compared using the chi-square test to ensure homogeneity between the two cohorts. To determine the potential prognostic factors for loss to follow-up, univariate and multiple logistic regression were applied. Variables with P values \<0.25 in the simple logistic regression (S-LogR) were fitted into multiple logistic regression models (M-LogR). Weighted points were assigned to each of the final predictors using the linear shrinkage factor formula described by Van Houwelingen and Le Cessie \[(model χ2 –df)/model χ2\]. The calculated shrinkage factor using the above formula was multiplied to the beta coefficient value of each factor and were rounded to the nearest integers. The shrunken and rounded method was used to avoid overfitting of the developed model. ## Variable selection for the LTFU prognostic model A total of 2346 patients were included in the univariate analysis. Crude associations of sociodemographic profiles, disease profiles and comorbidities with LTFU were analysed using a simple logistic regression analysis. By using a significance value of p\<0.25, the variables that were included in the multiple logistic regression analysis consisted of age, nationality, locality, ethnicity, educational level, monthly individual income, occupation, TB case category, TB anatomical location, TB detection method, chest X-ray status, BCG status, sputum status, and HIV status. For the monthly income level, the cut point at RM2160 (488.30 USD) used was based on the median personal income of population in this country according to the department of statistics Malaysia in 2018 (DOSM). Sex (p = 0.916) and DM comorbidity (p = 0.315) were not included in the multivariable analysis. A backward selection process was conducted to identify the most relevant predictors for outcomes by retaining variables with p values \<0.1. Among the 14 variables included in the multiple logistic regression analysis, 12 variables were retained in the final model (e.g., age group, ethnicity, locality, nationality, educational level, monthly income level, working status, TB case category, TB detection methods, chest X-ray categories, HIV status and sputum status). The discrimination of the predictive model was evaluated by using the area under the receiver operating characteristic (ROC) curve (AUC). AUC value ranges from 0.5 to 1, where a greater AUC value indicates that the model has better ability to distinguish patients who were LTFU and had successful TB treatment outcomes. Model calibration (predictive accuracy) was determined by the nonsignificant Hosmer‒Lemeshow goodness of fit test and visualized in the calibration plot. The performance of the development model was validated based on internal validation cohort dataset. To facilitate the applicability of the model in clinical settings, a risk-scoring table was constructed. Risk scores were calculated for each patient in the development cohort according to the weighted points for each predictor. The summed score for each patient was categorized into three groups based on percentiles of 33.33%, 66.66% and 100.00%, which were significantly distinct in predicting risks for LTFU. This study was conducted in accordance with the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) checklist as a guide. ## Ethical considerations This was a retrospective study that involved secondary data analysis from the National MyTB database version 2.1 from 2013 to 2017 by the Disease Control Division, Ministry of Health. Thus, no consent from respondents was needed. A formal request for data utilization was provided to the Disease Control Division, Ministry of Health Malaysia prior to study initiation. Registered TB cases were kept anonymous and given a unique identification number. This research involves no more than minimal risk to the subjects; thus, the requirement to obtain informed consent to participate was waived by the Research Ethics Committee (REC/04/2020), Universiti Teknologi MARA (UiTM) and Malaysia Ministry of Health Medical Research Ethics Committee (reference number NMRR-21-592-58245). # Results A total of 22785 TB patients were registered in the Selangor MyTB database from 2013 until 2017. A proportion of adult TB patients who smoked, 27.4% (N = 6242), was extracted from the database, and 14.0% of the patients who met the exclusion criteria were excluded from the dataset. Cases were randomly split into two cohorts with a 1:1 ratio, where 2346 cases were included in the development cohort and 2398 cases were included in the internal validation cohort. The data processing flowchart is shown in. The comparative characteristics of patients in the development and internal validation cohorts are summarized in. Both cohorts were homogeneous with nonsignificant differences in all variables with p values \>0.05 using the chi-square test. ## General characteristics of adult TB patients in the development cohort who smoked The majority of included patients were aged \<50 years old (70.4%), male (95.6%), Malaysian citizens (89.2%), lived in urban areas (83.4%) and were of Malay ethnicity (57.2%). They were mostly employed (63.36%), had monthly incomes of less than RM2160 (87.9%) and attained education to at least the secondary school level (63.8%). In terms of disease profiles, more than 80% of the patients were new cases, had pulmonary TB and were detected through passive case detection. Their chest X-ray status mostly indicated no lesions to minimal lesions (68.5%), and they had positive sputum smears (69.2%). Only 20% of TB patients who smoked had diabetes mellitus (DM), and fewer than 10% tested positive for HIV. Regarding treatment management, 93.2% of TB patients in the development cohort who smoked were on directly observed therapy (DOT) during the intensive phase; however, the percentage dropped to 79.5% during the continuation phase. A descriptive analysis of the treatment outcomes among the overall adult TB patients who smoked (N = 5703) showed that 69.1% had successful treatment outcomes (e.g., cured = 2517 or completed treatment = 1424) and 30.9% had unsuccessful treatment outcomes (e.g., loss to follow-up = 803, died = 580, treatment failure = 19 or outcome not evaluated = 360). ## Performance of the final model A total of 2280 of 2346 cases were analyzed in the logistic regression model with 2.8% (n = 66) of missing information in the development cohort. Out of 16 variables analysed in the univariate analysis, 14 variables with p-value \>0.25 were included in the multiple logistic regression and 12 variables were retained in the final model. The final logistic model had fair discrimination with a c-statistic of 0.681 (95% CI 0.657–0.713). There was a nonsignificant chi-square Hosmer‒Lemeshow goodness of fit test value, χ2 = 4.893, and the accompanying p value was 0.769, which indicated good calibration. These results are comparable to the internal validation model with a c-statistic of 0.668 (95% CI 0.639–0.698) and nonsignificant Hosmer‒Lemeshow goodness of fit test with χ2 = 2.223 and (p = 0.563) The AUCs between the two models (development and the internal validation cohort) showed an overlapping confidence interval indicates that the difference between the curve areas is not statistically significant. A graphical assessment of calibration (calibration curve) of the final model and the internal validation are shown in. ## Prognostic scoring system for LTFU Twelve variables were used to develop the LTFU score among TB patients who smoked (T-BACCO SCORE). Weighted points were assigned to each of the prognostic factors as shown in. The total point scores are normally distributed and range from 4 to 59 (mean = 26.04, SD = 7.10). They were divided into three categories based on percentiles (e.g., 33.33^rd, 66.66^th and 100^th). The low-risk group has scores \<15 points, the medium-risk group has scores from 15 to 25 points and the high-risk group has scores \> 25 points, as summarized in. Most patients who were lost to follow-up in the development cohort were in the high-risk group (74.7%). Compared with low-risk patients, patients in the medium and high-risk groups had greater odds for loss to follow-up during TB treatment with OR 2.533 and OR 6.717, respectively. In the development cohort, each risk category was significantly different in predicting the risk for loss follow-up with χ2(2) = 81.94 and p value of \<0.001. The percentage of missing data for all 16 variables was 2.8% in the development cohort and 2.9% in the internal validation cohort. Variable with incomplete data were the sputum status and chest x-ray status (both are categorical data). The little’s MCAR test showed a non-significant p-value in both development cohort (p = 0.516) and internal validation cohort (p = 0.510) indicating that the missing data is MCAR (missing completely at random) and no patterns exist in the missing data. No data imputation was done for cases with missing data in both the development and validation cohort since the missing data number is insignificant (less than 5%). Model development and internal validation were performed using a complete case analysis. Additional analyses were conducted by including the directly observed therapy (DOT) variable during the intensive phase and DOT during the continuation phase in the model. By including the DOT variable during the intensive phase in the model, the discrimination value improved to 0.763 (95% CI 0.745–0.780) however the calibration value of the model was unacceptable with significant Hosmer‒Lemeshow goodness of fit test. Another analysis trial was conducted by including both variables of DOT during the intensive phase and DOT during the continuation phase in the model, and this model showed an excellent c-statistics value with an AUC value of 0.936 (95%CI 0.925–0.946) and good calibration effect with a nonsignificant Hosmer‒Lemeshow goodness of fit test. A model with p values \>0.05 for calibration (Hosmer‒Lemeshow test) and relatively higher value of the c-statistic (area under the ROC curve) indicated relatively better- performing models. The ROC comparison between these models is shown in. This finding emphasizes the importance of DOT in both the intensive and maintenance phases for all TB patients who are receiving treatment. Looking back at the objective of this study is to develop a simple prognostic scoring tool for TB patients who smoke to predict their risk for loss to follow-up at the initial phase of TB treatment, preferably after being diagnosed with TB. However, the DOT variables can only be determined once patients have completed their treatments in both the intensive and maintenance phases. Therefore, the variable selection was mainly based on information that was readily available during initial patient consultation sessions with their health care practitioner, which justifies the selection of our final model. # Discussion We developed a prognostic model to predict LTFU from treatments among TB patients who smoked in a region where TB infections are endemic. The scores allocated to each predictor allow health care personnel to stratify patient risks for LTFU at the time of diagnosis, as this model utilizes easily accessible information on patient profiles and their disease profiles during initial routine follow-up visits at the TB clinic. The variables included consist of age, ethnicity, locality, nationality, educational level, monthly income level, employment status, TB case category, TB detection methods, chest X-Ray categories and HIV status. The final model provides good discrimination results and exhibits good calibration effect with insignificant variations in the internal validation. The results obtained from the final statistical model suggest that this prognostic model is valid for evaluating the probability of LTFU among TB patients who smoke in the early phase of TB treatment initiation. The factors that contributed to high scores (\>5 points) for the risk of LTFU in this study included no formal education or primary school education levels, patients living in urban areas, patients with low monthly income levels and previously treated cases. This indicates that low socioeconomic status, urban living conditions and past TB infections play important roles in predicting loss to follow-up in current TB treatments among smokers. The independent prognostic factors contained in the final model used in this study were consistent with evidence from a previous study regarding the factors associated with LTFU among TB patients. Several predictive models for TB treatment LTFU are available from other countries; for example, in Spain, those who live alone or in an institution, are immigrants, intravenous drug user (IVDU), patients with poor knowledge of TB and previously treated TB cases were identified as predictors for LTFU. Another study from Brazil that used the Classification and Regression Trees (CART) prediction model ranked several important variables that contribute to LTFU, such as the number of doses taken during treatment, age, total number of people in the household, non-IVDU and HIV-negative patients. In comparison with other prognostic models for loss to follow-up among general TB patients, our model identifies several predictive factors that are significant for TB smokers, such as young adults (age \<50 years old) and HIV-positive individuals. Young adult who are in their working age population are mostly associated with smoking behavior and are at risk for LTFU from TB treatment. While, smoking among TB-HIV patients had a synergistic impact on the burden of both TB and HIV, as smoking leads to worse TB treatment outcomes and inhibits the effectiveness of life-saving ART(anti-retroviral treatment), which worsens the overall health outcomes of patients. At the same time, financial limitations and poor social support aggravated the issues of treatment interruption and LTFU among TB-smoking patients with HIV. It also reflects the performance of our management and control activities at the ground level for TB-HIV patients under the National TB and HIV/AIDS control program. Noncompliance with anti-tuberculosis treatments and LTFU among TB patients are serious issues in our TB control program, yet this is preventable. In Selangor state itself, the rate of LTFU among general TB patients ranged from 5.8% to 13% from 2014 to 2017. This rate is higher than the national target rate for loss to follow-up of \<2% (NSPTB 2015–2020). In addition to directly observed therapy (DOT) or directly observed therapy short course (DOTS) as the standard patient management approaches for TB, several supporting actions have been undertaken to decrease the rate of LTFU among TB patients, including video observed therapy (VOT) and financial incentives. VOT has been implemented in some health care facilities since 2019 in Malaysia not only for TB patients but also for managing other chronic diseases to improve patient adherence to treatment. We have been focusing on methods to ensure treatment completion, but we have not reached people who are vulnerable and patients who are at risk of TB and developing poor outcomes. Risk identification and risk management would be ways to supplement the current TB patient management methods to ensure that all patients complete their treatment successfully. At present, no specific management procedure is described for TB patients who smoke in the Clinical Practice Guideline (CPG). An integrated TB and tobacco control program should be implemented through the health care system to improve overall TB treatment outcomes. Studies have shown that smoking cessation intervention using behavioural therapies (i.e., brief advice from a physician or individual or group counselling sessions) is effective in reducing the loss to follow-up rate, improving recovery, and preventing treatment failure. A local study in Malaysia, as part of the SCIDOT project by Awaisu et al., found that integrated TB tobacco treatment using DOTS plus smoking cessation intervention could also improve overall quality of life outcomes (HRQoL) among TB patients who were smokers during the 6-month treatment duration compared to those who received conventional TB care. It is understood that implementing a new program or additional actions in TB management require additional resources such as workforce levels, money, training, and promotional activities. However, if we are committed to achieving the goal of “zero TB cases by 2035”, we must be willing to invest more. Human resources are a crucial determinant in managing TB, especially in high-burden states such as Selangor. Human resources not only refer to doctors but also include other supporting personnel such as nurses, medical assistants and other positions that help in managing TB cases. All working staff require proper training to provide quality service to all TB patients, which means that larger budget allocations need to be reserved for training and educational purposes. In line with the recommendation from the Global Plan to End TB 2018–2022, we are committed to accelerate the innovation of new tools to be used in care delivery to combat TB. This screening tool helps to stratify and prioritize high-risk TB patients who smoke and require specific intervention. Other than that, the score estimated in our study is calculated based on patient’s characteristics and disease profiles of a large cohort of TB patients under the current TB controlled programme and healthcare system, which may indicate where improvement can be made. At the point when a person is diagnosed with tuberculosis would be a starting point or “golden opportunity” to advise TB patients to quit smoking, and DOT is an effective way to encourage and support the pathway to quitting smoking. Most smokers with TB are receptive to quitting advice; however, they often begin smoking again when they feel better, which justifies the requirement of proper smoking cessation interventions along with their DOT. This is also part of our initiative to reduce the financial catastrophes faced by families and patients that can be avoided by quitting their smoking behaviour. A limitation in our study is that the model was developed using secondary data obtained from a surveillance database. Thus, the variables included in the model were limited to the patient information that was available in this database. Other important predictors, such as social support (e.g., living conditions, numbers of people in households and marital status), knowledge of TB infections and TB treatment details (e.g., drug dosage, duration, and side effects from the drugs), were unavailable. This would create further research opportunities for an updated LTFU prognostic model in the future. However, reflecting on our initial objective in this study, it is unfeasible to obtain all the above information to assess patients on their risk for LTFU at the initial phase of the treatment. For example, the treatment durations and side effects from TB drugs can only be assessed after a certain treatment period, while patient knowledge of TB would require a separate assessment session. Improving treatment compliance and preventing LTFU among TB patients who smoke is a great challenge that should be addressed through strong support and engagement from political will, health care providers, family, and social organizations. Smoking cessation interventions using the quit smoking program available in the country must be integrated into the TB care routine through appropriate channels. Our early screening tool to predict LTFU among TB patients who smoke will guide health care personnel to identify which smoking TB patients require further intervention during the 6 months of TB treatment, and key activities under the National TB control program must act on the identified prognostic factors highlighted from this study. # Conclusions This simple prognostic score (T-BACCO SCORE) facilitates risk stratification for TB patients who smoke for LTFU from TB treatment and allows personalized TB monitoring and management options. Further external validation using the tool across the population shall be conducted prior to its application in clinical settings. The authors would like to thank the Director General of Health, Malaysia for permission to publish this paper. We would also like to express our gratitude to all health care personnel involved in the national TB surveillance system in Malaysia for their dedication and give special thanks to the Big Data Task Force, UiTM Selangor Branch for their kind help with data cleaning processes. 10.1371/journal.pone.0287374.r001 Decision Letter 0 Krishnamoorthy Yuvaraj Academic Editor 2023 Yuvaraj Krishnamoorthy This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 22 Nov 2022 PONE-D-22-25828T-BACCO SCORE: A predictive scoring tool for tuberculosis (TB) loss to follow-up among TB smokersPLOS ONE Dear Dr. Ismail, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jan 06 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Yuvaraj Krishnamoorthy Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well. 3\. We note that you have stated that you will provide repository information for your data at acceptance. Should your manuscript be accepted for publication, we will hold it until you provide the relevant accession numbers or DOIs necessary to access your data. If you wish to make changes to your Data Availability statement, please describe these changes in your cover letter and we will update your Data Availability statement to reflect the information you provide. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The manuscript is well conceived and written and needs some major revisions to be acceptable for publication. Major Concerns: 1\. There is no reporting of how well the model performed in the internal validation cohort. 2\. The HOsmer-Lemeshow is not a reliable test for calibration, authors should present calibration plots as illustrated in the TRIPOD statement for reporting of risk prediction models. 3\. Early in the paper it is not clearly and consistently articulated that a simple binary indicator of 'loss to follow-up (LTFUP)' is the outcome being modeled. It is confusing with all the information presented on different outcome states. is the outcome LTFUP relative to the occurrence of all other outcome types? 4\. cross-sectional: the study uses fixed, baseline values of predictors to calculate the occurrence of LTFUP, which happen at some point in the future, so the repeated use of cross-sectional seems not quite accurate 5\. On line 29 no statement about how much data is missing, whether it can be assumed to be missing at random, and whether any form of imputation was used. If not, how was this justified? 6\. Process of multivariable model selection should be in the stats section rather than in results. 7\. On line 188 test of equality for two AUC values should use DeLong's test rather than chi-square. 8\. The tiny fraction of women in the study suggests these results should be limited to males only. Minor comments: 1\. multivariable instead of multivariate when modeling a single outcome. 2\. AUC provides fair, not good, discrimination. 3\. Line 134 confusing. 4\. lines 187-8, chi-square not appropriate for comparing two AUC. 5\. line 220: need to define RM2160 which presumably is transparent to Malaysians. 6\. Line 244 - state 14 variables retained but only 13 listed. 7\. line 260: no mention of model performance in validation cohort. 8\. lines 282-3: no comprehensive discussion or treatment of missingness 9\. lines 288-9: that chi-square test does suggest anything about missing values being completely at random. 10\. Table 5: okay to present Hosmer-Lemeshow as supplement to calibration plots. 11\. line 351: need to define IVDU \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0287374.r002 Author response to Decision Letter 0 11 Jan 2023 Major Concerns: 1 There is no reporting of how well the model performed in the internal validation cohort. Reports on the performance of the internal validation cohort is available in the results section under the performance of final model. 2 The HOsmer-Lemeshow is not a reliable test for calibration, authors should present calibration plots as illustrated in the TRIPOD statement for reporting of risk prediction models: As per suggestion, calibration plots have been added in addition to the Hosmer- Lemeshow value for the model calibration reporting. The calibration curve for the final model and the internal validation model are available in Fig 3 and Fig 4. 3 Early in the paper it is not clearly and consistently articulated that a simple binary indicator of 'loss to follow-up (LTFUP)' is the outcome being modeled. It is confusing with all the information presented on different outcome states. is the outcome LTFUP relative to the occurrence of all other outcome types? The outcome being measured and modelled in this study is the LTFU. The LTFU outcome is being compared with the successful TB outcome (cured and completed treatment) 4 Cross-sectional: the study uses fixed, baseline values of predictors to calculate the occurrence of LTFUP, which happen at some point in the future, so the repeated use of cross-sectional seems not quite accurate: We have look back at the methods of this study. We agree that the term “cross- sectional study” used in this study is inappropriate. This study involves secondary data analysis of a cohort of TB patients from year 2013-2017 in the national TB database. We looked back at the predictors that lead to the LTFU outcome. Thus, the correct study design for this study would be a retrospective cohort study. 5 On line 29 no statement about how much data is missing, whether it can be assumed to be missing at random, and whether any form of imputation was used. If not, how was this justified? Additional statement on the missing data information has been added in the abstract and result section. We have run the missing value analysis and the little MCAR test to report on the data missingness. Certain variable in the database has been recategorized and subsequent changes on the results and figure has been amended. 6 Process of multivariable model selection should be in the stats section rather than in results: The paragraph on variable selection for the final model has been changed to the methodology section under the statistical analysis part. 7 On line 188 test of equality for two AUC values should use DeLong's test rather than chi-square: As per suggestion we have analyse the performance between the two-model using the DeLong’s test. The outcome of this test is included under the result section in the performance of the final model paragraph. 8 The tiny fraction of women in the study suggests these results should be limited to males only: The prevalence of woman who smokes in our country based on the latest national health morbidity survey (NHMS, 2015) is 1.3% which correspond to our finding in this study. Thus, we choose to remain the female subject in this study. Besides that, gender was not a significant predictor in this study, therefore the small proportion of female gender do not affect the performance of the model. Minor comments: 1 Multivariable instead of multivariate when modeling a single outcome: The term multivariate used in this manuscript has been changed to multivariable throughout the manuscript. 2 AUC provides fair, not good, discrimination: Correction has been made. The term used on the AUC has been changed to fair discrimination. 3 Line 134 confusing: The sentence has been rearranged. The outcome being measured in this study is the LTFU TB outcome. 4 Lines 187-8, chi-square not appropriate for comparing two AUC: As per suggestion we have analyse the performance between the two-model using the DeLong’s test. The outcome is available under the result section. 5 line 220: need to define RM2160 which presumably is transparent to Malaysians: The categorization used for the income level is according to the median personal income for the population of Selangor state based on the department of statistic Malaysia (DOSM, 2018). Details on this variable has been added under the statistical analysis paragraph. 6 Line 244 - state 14 variables retained but only 13 listed: We have corrected the sentence. The correct number of variables included in the multivariable analysis were 14, and 12 were retained in the final model. 7 line 260: no mention of model performance in validation cohort: The performance of the Internal validation model was mentioned under the result section in the performance of the final model paragraph. 8 lines 282-3: no comprehensive discussion or treatment of missingness Management and reports on the missing data has been included in the result section under the prognostic scoring system for LTFU paragraph. 9 lines 288-9: that chi-square test does suggest anything about missing values being completely at random: We have conducted the missing value analysis and the little’s MCAR test to handle the missing data in the database. The report on the chi-square test has been corrected. 10 Table 5: okay to present Hosmer-Lemeshow as supplement to calibration plots: Calibration curve for both final and internal validation model has been plotted under Fig 3A and Fig 3B in the manuscript using GGplot R statistical package. 11 line 351: need to define IVDU IVDU has been defined as intravenous drug user in the sentence. Thank you 10.1371/journal.pone.0287374.r003 Decision Letter 1 Krishnamoorthy Yuvaraj Academic Editor 2023 Yuvaraj Krishnamoorthy This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 8 May 2023 PONE-D-22-25828R1T-BACCO SCORE: A predictive scoring tool for tuberculosis (TB) loss to follow-up among TB smokersPLOS ONE Dear Dr. Ismail, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jun 22 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Yuvaraj Krishnamoorthy Academic Editor PLOS ONE \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors have done a nice job responding to all my previous concerns. I advise acceptance with minor revisions and do not need to see the manuscript again. 1\. On line 51 put a comma in 835,000. 2\. re-vise line 178 as follows: '... with LTFU were analyed using... ' 3\. revise line 183 as ' For the monthly income level, the cutpoint at... (488.30 USD) was based on... ' 4\. Line 227: homogeneous 5\. In Fig 2 lease add the labelled AUC for each curve. 6\. For Fig 3 recommend confidence limits for blue model performance to show overlap with diagonal. Reviewer \#2: Authors intended to develop a prognostic scoring tool in predicting loss to follow-up (LTFU) among TB smoking patients. The constructed scoring was based on T-BACCO SCORE and predicted the risk for LTFU with low-, medium- and high-risk. The results showed a fair discrimination, good calibration. 1\. The outcome is loss to follow-up. What’s the length of follow-up time for each individual? Are they very different or the same across the individuals? If vary, will this confound the results? 2\. Line 179. The significance level was set to be 0.025. Please explain why this threshold is selected? 3\. Line 194. AUC of 0.5 is selected as a discrimination threshold. However, 0.5 is just the value with random classification. Will such a low threshold be a concern? 4\. Line 199. De Long test was used to compare the AUC between models. Please clarify what two models were compared? development and the internal validation cohort? If so, does it really make sense to compare their AUC when they are using different data? 5\. It’s good to see the missingness doesn’t differ between development cohort and internal validation cohort. However, shouldn’t the missing data be taken care before split the sample into two cohorts? \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0287374.r004 Author response to Decision Letter 1 18 May 2023 Response to Reviewers Dear PLOS One Editors, AMENDMENT TO THE MANUSCRIPT ENTITLED “T-BACCO SCORE: A PREDICTIVE SCORING TOOL FOR TUBERCULOSIS(TB) LOSS TO FOLLOW-UP AMONG TB SMOKERS Response to Reviewer 1: 1 On line 51 put a comma in 835,000 Response: Correction has been made accordingly on-line 51 2 Revise line 178 as follows: '... with LTFU were analysed using... ' Response: The sentence has been corrected on line 178 “with LTFU were analysed using a simple logistic regression analysis”. 3 Revise line 183 as ' For the monthly income level, the cut point at... (488.30 USD) was based on... ' Response: The cut of point at RM2160 (488.30 USD) used was based on the median personal income of population in this country according to the department of statistics Malaysia in 2018 (DOSM). 4 Line 227: homogeneous Response: Spelling correction has been made accordingly, from “homogenous” to “homogeneous”. 5 In Fig 2 please add the labelled AUC for each curve. Response: Amendment has been made in Fig 2. AUC value for each curve has been added. 6 For Fig 3 recommend confidence limits for blue model performance to show overlap with diagonal. Response: Calibration curve Fig 3(A) and Fig 3(B) has been amended. The confidence limits each calibration curve was inserted as per-recommendation. Response to Reviewer 2: 1 The outcome is loss to follow-up. What’s the length of follow-up time for each individual? Are they very different or the same across the individuals? If vary, will this confound the results? Response: The length of follow-up time unlikely confound the results. Majority of each individual follow-up duration is limited to the recommended six-month of TB treatment duration. Patients with TB MDR status (who needs longer treatment duration) has been excluded from the databased. 2 Line 179. The significance level was set to be 0.025. Please explain why this threshold is selected? Response: For variable selection into the multiple logistic regression analysis, significance level was set at 0.25 (not 0.025). This threshold was selected to ensure that adequate variable can be considered in the final logistic model. 3 Line 194. AUC of 0.5 is selected as a discrimination threshold. However, 0.5 is just the value with random classification. Will such a low threshold be a concern? Response: We are aware that AUC of 0.5 indicates no discrimination and AUC 0.7 is considered acceptable. However, in clinical setting within the TB domain especially on LTFU issue an AUC more than 0.6 is considered satisfactory as compared to the common acceptable threshold of 0.7. We have revised the sentence in line 194 to “AUC value ranges from 0.5 to 1, where a greater AUC value indicates that the model has better ability to distinguish patients who were LTFU and had successful TB treatment outcome”. 4 Line 199. De Long test was used to compare the AUC between models. Please clarify what two models were compared? development and the internal validation cohort? If so, does it really make sense to compare their AUC when they are using different data? Response: Thank you for your comment. Yes, the two model that were compared are the development and the internal validation cohort. Initially, De Long test was used to inform that the performance between the two model has not much difference. However, after careful reading, we realized that De Long test is used to compare two correlated model. Therefor this analysis will be removed from this study finding. The comparison of the ROC between the two cohort will be assessed based on their overlapping confidence interval: Development cohort: 0.681 (95% CI 0.657-0.713) Internal validation cohort: 0.668 (95% CI 0.639-0.698) 5 It’s good to see the missingness doesn’t differ between development cohort and internal validation cohort. However, shouldn’t the missing data be taken care before split the sample into two cohorts? Response: We agreed that missing data analysis should be managed earlier, however the two cohort (development and the internal validation) are coming from the same source of data based and it was divided randomly, therefor it should not be a concern. The purpose of analysing the missing data separately is to provide a more representative picture on the data missingness for each cohort to the reader. 10.1371/journal.pone.0287374.r005 Decision Letter 2 Krishnamoorthy Yuvaraj Academic Editor 2023 Yuvaraj Krishnamoorthy This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 5 Jun 2023 T-BACCO SCORE: A predictive scoring tool for tuberculosis (TB) loss to follow-up among TB smokers PONE-D-22-25828R2 Dear Dr. Ismail, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Yuvaraj Krishnamoorthy Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: (No Response) Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0287374.r006 Acceptance letter Krishnamoorthy Yuvaraj Academic Editor 2023 Yuvaraj Krishnamoorthy This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 8 Jun 2023 PONE-D-22-25828R2 T-BACCO SCORE: A predictive scoring tool for tuberculosis (TB) loss to follow-up among TB smokers Dear Dr. Ismail: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Yuvaraj Krishnamoorthy Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction The lung continuously encounters oxidants from inhalation and is therefore well equipped with a high concentration of the antioxidant glutathione (GSH). GSH acts as an electron donor and is used by glutathione peroxidase to reduce peroxides, resulting in oxidized glutathione (GSSG). Cigarette smoke is known to acutely deplete GSH, for instance by directly reacting with GSH to form non- reducible glutathione-aldehyde derivatives, thereby decreasing the lungs’ anti- oxidant capacity and making it vulnerable to oxidant-induced injury. On the other hand, as an adaptive response to oxidative stress, such as upon chronic smoking, levels of GSH increase in the epithelial lining fluid due to upregulation of the rate limiting enzyme in GSH synthesis, γ-glutamylcysteine ligase. Besides being oxidized itself, glutathione can in conditions of mild oxidative stress, also bind to cysteine residues in proteins. This posttranslational modification is known as S-glutathionylation and protects proteins from irreversible oxidations. Glutaredoxins (Grx) or thioltransferases are redox enzymes that, under physiological conditions, can reverse S-glutathionylation. S-glutathionylation does not only protect the targeted protein thiol groups from further irreversible oxidations, but also has been shown to modulate protein function when the targeted cysteine residue is critical to its function. Examples include mediators of cell death and inflammation such as procaspase-3, multiple members of the NF-κB pathway (reviewed), and matrix metalloproteases. Inhibition has been shown for caspase 3, as well as NF-κB, whereas MMP9 has been shown to be activated by this redox modification. Therefore glutaredoxins play an important role in redox-modulated protein function by regulating S-glutathionylation. Several mammalian Grxs have been identified. Grx1 localizes primarily to the cytosol and Grx2 is present in the mitochondria and nucleus. In many pulmonary diseases, including COPD, the importance of glutathione homeostasis is described, whereas S-glutathionylation and Grxs have hardly been investigated. Grx1 expression in the lungs has been found to be predominantly localized in macrophages and bronchial epithelium. In a mouse model of allergic airway disease and after acute exposure to LPS Grx1 expression was increased. In patients with COPD on the other hand, Grx1 was decreased and specifically the number of Grx1 positive macrophages was found to be positively correlated with lung function. In line with these clinical findings, we have previously reported that cigarette smoke extract downregulated Grx1 levels, which was associated with increased protein S-glutathionylation in lung epithelial cells. Moreover, primary epithelial cells from Grx1 knock out mice were more prone to smoke- induced cell death and displayed higher levels of protein S-glutathionylation compared to controls. I*n vivo* on the other hand, we found smoke exposure to decrease protein S-glutathionylation, while also decreasing Grx1 levels and total Grx activity. Targeted S-glutathionylation is described to inhibit multiple members of the pro-inflammatory NF-κB pathway, including IKKα, IKKβ and Rel A. We have previously described that LPS exposure in the context of ablation of Grx1 failed to activate NF-κB and decreased inflammatory cytokine levels. In the current study we set out to investigate whether absence of Grx1 similarly represses cigarette smoke-induced inflammation in a subacute exposure model in mice. Rather than focusing on individual NF-κB members, we investigated the differential inflammatory response of mouse lungs as well as primary epithelial cells and macrophages to cigarette smoke. # Materials and Methods ## Mice and Primary Cell Culture Male *Grx1^−/−* mice, a kind gift of Dr. Ho (Wayne State University, Detroit, MI), and WT C57BL/6 controls (n = 10 per group) were exposed to cigarette smoke for four weeks as described previously. Briefly, mice were exposed whole body to the tobacco smoke of 5 Reference Cigarettes 3R4F without filter (University of Kentucky, Lexington, KY) four times a day with 30 min smoke-free intervals, 5 days a week for 4 weeks. During the exposure an optimal smoke to air ratio of 1∶6 was obtained. The control groups were exposed to room air. Additional unexposed mice were used to isolate primary tracheal epithelial cells (MTE) as described previously with minor modifications and pulmonary macrophages by saline lavage. Cells were cultured in full medium lacking phenol red for 24 h prior to stimulation. The local ethics committee for animal experimentation of the faculty of medicine and health sciences of Ghent University, Belgium granted approval for all *in vivo* procedures under ECD07/04. ## Bronchoalveolar Lavage (BAL) 5 days after the last exposure mice were euthanized with an overdose of pentobarbital and a cannula was inserted into the trachea. Three times 300 µl HBSS, free of Ca²⁺ and Mg²⁺ and supplemented with 1% BSA, followed by 3 times 1 ml HBSS supplemented with 0.05 mM EDTA, was instilled through the cannula and recovered by gentle aspiration. All lavage fractions were pooled, centrifuged and the cell pellet washed twice and resuspended in 1 ml HBSS. Total and differential cell counts were performed in a Bürker chamber and cytocentrifuged preparations stained with May-Grünwald-Giemsa respectively. Flow cytometric analysis of BAL cells was performed as described previously to enumerate dendritic cells, macrophages, neutrophils and T-lymphocyte subsets. ## Lung Tissue Processing and Cell Counts After rinsing of the pulmonary and systemic circulation, the left lung was used for histology by intratracheal infusion of 4% PFA and embedding in paraffin. Single cell suspensions were prepared from the right lung by mincing thoroughly, digesting and RBC lysis. Cell counts were performed with a Beckman Coulter counter and flow cytometric analysis was performed as described previously to enumerate dendritic cells, macrophages, neutrophils and T-lymphocyte subsets. The other part of the right lung was snap frozen in liquid nitrogen for biochemical assessments. ## Cigarette Smoke Extract 3R4F Research Cigarettes, from the University of Kentucky (Lexington, KT, USA), were removed from their filters and cigarette smoke extract (CSE) was made fresh before every experiment according to. ## Quantitative Determination of S-Glutathionylated Proteins Using 5,5′-dithio-bis(2-Nitrobenzoic Acid) (dTNB) 200 µl of BAL fluid or 200 µg of lung protein homogenate was acetone precipitated for 20 minutes at −20°C and spun down for 5 minutes at 3000×g. Pellets were next resuspended and sonicated in 200 µl of ice-cold extraction buffer containing 0.2% Triton-X 100 and 0.6% sulfosalicyclic acid in 0.1 M potassium phosphate buffer with 5 mM EDTA disodium salt (KPE), pH 7.5. After 2 freeze-thaw cycles, samples were centrifuged at 3000×g for 4 min at 4°C. To remove glutathione (GSH) from proteins, the pellet was treated with 100 µl of 1% NaBH₄ in water and neutralized with 40 µl of 30% metaphosphoric acid. Samples were centrifuged at 1000×g for 15 min and the supernatant was used to determine the GSH content using the dTNB GSSG reductase recycling method. 20 µl of KPE, GSH standards and samples were pipetted into a 96-well microtiter plate and freshly prepared, equal volumes of dTNB and GSSG reductase were added in the dark. After 30 seconds, β-NADPH was added to start the conversion of dTNB to TNB and the absorbance at 412 nm was read every 30 seconds for 2 minutes. A standard curve was performed using a concentration range of GSH. NaBH₄ was omitted for each sample, as a negative control. Values were corrected for protein content and data are expressed as nmol GSH per milligram of protein. ## Grx1 Catalyzed Cysteine Derivatization for In Situ Detection of S-glutathionylated Proteins Frozen cytospins were thawed and washed twice with PBS before being fixed with 4% paraformaldehyde (PFA) for 10 minutes at RT. After three washes with PBS slides were permeabilized and free thiol groups were blocked using a buffer containing 25 mmol/L 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid, pH 7.4, 0.1 mmol/L EDTA, pH 8.0, 0.01 mmol/L neocuproine, 40 mmol/L *N*-ethylmaleimide (Sigma) and 1% Triton (Sigma) for 30 minutes to an hour. After three washes with PBS, S-glutathionylated cysteine groups were reduced by incubation with 13.5 µg/ml human Grx1 (Lab Frontiers), 35 µg/ml GSSG reductase (Roche), 1 mmol/L GSH (Sigma), 1 mmol/L NADPH (Sigma), 18 µmol EDTA and 137 mmol/L Tris · HCl, pH 8.0, for 20 minutes. As a control GSH was left out of this mix. After three washes with PBS, newly reduced cysteine residues were labelled with 1 mmol/L *N*-(3-maleimidylpropionyl) biocytin (MPB) (Roche) for 1 hour, after which excess MPB was removed by three washes with PBS. Next, cells were incubated with 0.5 µg/ml streptavidin-conjugated Alexa Fluor 568 for 30 minutes. Nuclei were stained using 0.5 µg/ml DAPI Blue. Cells were then mounted, coverslipped and analyzed by fluorescent microscopy using a Nikon Eclipse E800 microscope`. `All conditions were scanned using identical instrument settings that did not result in saturation of pixel intensities. Semi-quanititative assessment of the staining intensity was conducted by dividing mean red fluorescence intensity (PSSG staining) by the mean blue fluorescence intensity (nuclear DAPI staining) using Image J software. Mean relative fluorescence intensity (RFI) values and SEM were thus obtained. ## Grx1 Staining Macrophages were fixed with 4% PFA for 10 min at RT. After permeabilization and blocking non-specific binding sites using 0.1%triton, 1% BSA in PBS, primary antibody against Grx1 (Imco) was incubated for 1 h followed by Alexa fluor 488 labelled secondary anti-goat antibody for 1 h. Nuclei were counterstained with DAPI and cells were coverslipped. Semi-quanititative assessment of the staining intensity was conducted by dividing mean green fluorescence intensity (Grx1 staining) by the mean blue fluorescence intenstiy (nuclear DAPI staining) using Image J software. Mean relative fluoresence intensity (RFI) values and SEM were thus obtained. ## Multiplex for Cytokine Measurement To quantify concentrations of 23 cytokines and chemokines in BALF we used a Bio- Plex mouse cytokine 23-plex Panel (IL-1a, IL-1β, IL-2, IL-3, IL-4, IL-5. IL-6, Kc, IL-9, IL-10, IL-12(p40), IL-12(p70), IL-13, IL-17, Eotaxin, G-CSF, GM-CSF, IFN- γ, MCP-1(MCAF), MIP-1α, MIP-1β, RANTES and TNF-α). Assays were performed as described by the manufacturer’s instructions. The analysis was done with a Luminex 100 IS 2.3 system using the Bio-Plex Manager 4.1.1. software. ## Kc ELISA Kc levels in cell culture medium were measured using a commercially available ELISA kit (R&D systems, Inc. Minneapolis, USA) according to the manufacturer’s instructions. ## QPCR Total RNA was isolated from lungs or cells using the RNeasy Mini kit (QIAGEN, California, USA) and an equal amount was reverse transcribed into cDNA using the Reverse-iT 1st strand Synthesis Kit (Abgene, Epsom, UK). Primers for human HPRT (FW:AGAATGTCTTGATTGTGGAAGA; REV:ACCTTGACCATCTTTGGATTA), Grx1 (FW:TTTACAACAGCTCACCGGAG; REV:TCACTGCATCCGCCTATG) and Kc (Fw: CACTGCACCCAAACCGAAG; REV: TCAGGGTCAAGGCAAGCC) were used. PCR reactions were performed on an *iCycler iQ* Real-Time PCR system (BioRad, Hercules, California, USA) using the SYBRgreen dye (BioRad). Relative mRNA expression of genes was calculated using the standard curve method. ## Statistical Analyses Between-group comparisons were analyzed using the Kruskal-Wallis test, followed by Mann-Whitney *U* test (SPSS 17). Unless indicated otherwise, data are expressed as mean and standard deviation. A *p*-value \<0.05 was considered statistically significant. # Results ## BAL Fluid Cell Counts and Differentials in Wild Type Versus Grx1 KO Mice Exposed to Air and Smoke When analyzing lavaged cells, the total number of BALF cells was found to be significantly elevated in wild type and Grx1 KO mice due to cigarette smoke compared to respective air exposed controls. The level of increase in total cell numbers in the BALF did not differ between the two mouse strains. Macrophage cell counts in BALF also increased with cigarette smoke compared to respective air exposed controls, but significantly more so in Grx1 KO mice than in WT. The smoke-induced increase in the number of neutrophils, dendritic cells, CD8+, CD4+ and CD3+ cells on the other hand was significantly dampened in Grx1 KO compared to WT mice ( C–G). ## Lung Tissue Differential Cell Counts in Wild Type versus Grx1 KO Mice Exposed to Air or Smoke Next, we investigated total and differential cell numbers in lung tissue of the wild type and Grx1 KO mice. shows that there is no significant difference in total cell numbers of the lung tissue between the Grx1 KO and wild type mice, under basal conditions and after exposure to cigarette smoke. Although the numbers of macrophages tended to be higher in lung tissue of Grx1 KO mice compared to wild type controls, this was not statistically significant. In addition, the numbers of lung dendritic cells, lung CD3+ cells, lung CD8+CD69+ and GR1+ cells were not affected by smoke exposure in the WT animals, or in the Grx1 KO mice. The numbers of lung CD4+CD69+ cells on the other hand was significantly increased in wild type mice exposed to cigarette smoke. This increase was however not present in the Grx1 KO mice after cigarette smoke exposure. ## BAL Fluid Cytokines in Wild Type Versus Grx1 KO Mice Exposed to Air and Cigarette Smoke We next assessed a broad panel of inflammatory mediators in BALF in order to determine a cause for the diminished influx of inflammatory cells into the lungs of Grx1 mice after smoke exposure. While analyzing the inflammatory mediators measured in the BAL fluid of wild type versus Grx1 KO mice, the general observation was that these are decreased in Grx1 KO mice after exposure to cigarette smoke compared to wild type controls exposed to cigarette smoke. Specifically, cigarette smoke exposure significantly increased the BALF concentration of IL12(p40), GCSF, MCP-1, KC, RANTES and MIP-1alpha in wild type mice. In the Grx1 KO, the cigarette smoke-induced upregulation of these cytokines was significantly impaired, compared to the wild type controls. Moreover, the baseline levels of IL12(p40), GCSF, RANTES, MIP-1α, TNFα (data not shown) and IFNγ (data not shown) were found to be lower in the BAL fluid of Grx1 KO mice compared to wild type mice. ## S-glutathionylation in Lungs and BAL Fluid Cells of Grx1 KO Compared to Wild Type Mice Exposed to Cigarette Smoke and Air Since the major function of Grx1 is to catalyze deglutathionylation under physiological conditions, we next investigated the levels of total protein S-glutathionylation in lung tissue. As demonstrated in, exposure to cigarette smoke for four weeks lead to a significant decrease in S-glutathionylation of proteins in the lung tissue of wild type mice as reported previously. In the Grx1 KO mice, the smoke-induced decrease in protein S-glutathionylation did not reach statistical significance, but the levels observed after smoke exposure were significantly elevated compared to those in the wild type mice exposed to smoke. The levels of free GSH were decreased after smoke exposure as well, but no differences were observed between Grx1 KO and wild type mice (data not shown). When previously visualizing protein S-glutathionylation in whole lung tissue, we noted that this was not decreased in inflammatory cells after smoke exposure. We therefore analyzed protein S-glutathionylation in cells obtained by BAL using the biotin-switch approach and found that protein S-glutathionylation in these BAL fluid cells of mice exposed to smoke was increased compared to BAL fluid cells from air exposed mice. Moreover, S-glutathionylation in BAL fluid cells of Grx1 KO mice was higher after cigarette smoke exposure than in the wild type controls. As expected, most of the cells in the cytospins had the morphological characteristics of macrophages. In cell free BALF, similar trends were observed with increased protein S-glutathionylation after smoke exposure and heightened levels in Grx1 KO mice compared to wt controls. At baseline, no differences were observed between WT and Grx1 KO mice. ## Grx Expression in Lung Tissue and Macrophages of Mice Exposed to Cigarette Smoke We have previously shown that exposure of pulmonary epithelial cells to cigarette smoke extract (CSE) leads to decreased expression of Grx1 mRNA and protein. In we first confirmed these data in lung tissue of smoke exposed mice. Indeed, Grx1, but not Grx2 (data not shown) mRNA levels are significantly decreased after four weeks of cigarette smoke exposure compared to air exposed controls in wild type mice. Given the differential response of structural and inflammatory cells to smoke with respect to protein S-glutathionylation, we also assessed the effects of CSE on Grx expression in primary macrophages isolated from mice by fluorescent staining for Grx1 (represented) and confirm that smoke exposure also represses Grx1 protein levels in these cells. ## KC in Primary Macrophages Versus Primary Epithelial Cells from Wild Type and Grx1 KO Mice after Exposure to Cigarette Smoke Extract Protein S-glutathionylation appears thus to be differentially regulated in structural and inflammatory cells after smoke exposure, albeit independently from Grx1. We therefore asked whether this would differentially impact inflammatory mediator production in these cell types in response to cigarette smoke, given that we have previously shown that the Grx1-PSSG axis is important in determining the extent of NF-κB activation and the levels of cytokines and chemokines such as KC in response to pro-inflammatory stimuli. When performing an ELISA for KC on culture supernatants of macrophages isolated from wild type mice, we found a significant increase in KC after exposure to CSE. In contrast, the Grx1 KO macrophages displayed decreased KC concentrations in their medium after exposure to CSE. Also in MTECs the production of this chemokine was increased by cigarette smoke from both WT and Grx1 KO mice. Moreover, this response to CSE was significantly decreased in the Grx1 KO MTECs compared to the cells from wild types. # Discussion It is becoming increasingly clear that GSH and its associated enzymes and the redox state of cells in general do not only play a role in protection against oxidative stress and damage, but also determine the outcome of discrete receptor-induced ROS-mediated signaling events involved in immune responses. Specifically shown here is that Grx1 and PSSG are key modulators of the *in vivo* response to cigarette smoke. Although there was no difference in the total amount of BAL fluid and lung cells between wild type and Grx1 KO mice after smoke exposure, the pattern of inflammatory cells found in the BAL fluid was different. The Grx1 KO mice actually accumulate macrophages in the BAL fluid after 4 weeks of cigarette smoke exposure, while in contrast to the wild type mice the increase of neutrophils, dendritic cells and CD3+, CD4+ and CD8+ T-cells was significantly impaired. This altered pattern of inflammatory cells in the BALF was mirrored by significant differences in the levels of chemokines and cytokines. After four weeks of cigarette smoke exposure the BALF of Grx1 KO mice contained significantly less KC, RANTES, MCP-1, IL12, GCSF and MIP1α compared to the wild type controls exposed to the same amount and duration of cigarette smoke. The decreased levels of inflammatory mediators in the BAL fluid of Grx1 KO mice can be caused by the diminished levels of inflammatory cells other than macrophages. On the other hand, it could be attributed to the immaturity of these macrophages as we reported previously that alveolar macrophages isolated from Grx1 KO mice are smaller, express lower levels of the hematopoietic cell specific transcription factor PU.1 and displayed decreased phagocytosis capacity *in vitro* compared to alveolar macrophages from wild type animals. Also, LPS-induced NF-κB activation and inflammatory mediator production were found to be attenuated in the Grx1 KO macrophages as well as primary tracheal epithelial cells. The Grx1 KO mice have been used by others to investigate the role of Grx1 in cigarette smoke induced lung inflammation. This study however employed a three day smoke exposure protocol and showed a much different outcome compared to the results obtained here. After a three day exposure regimen, the number of neutrophils were reported to be elevated in the Grx1 KO mice, in conjunction with increased levels of KC and MCP-1 compared to wild type controls. The main difference between studies is the length of exposure and might indicate that the role of Grx1 is opposite in an acute versus a more chronic exposure to cigarette smoke. This is also supported by the early increase in KC levels observed after LPS exposure in the Grx1 KO mice, compared to the attenuation of other mediators at more protracted time points. Not only the role of Grx1 in smoke-induced inflammation was found to be different, but in our model smoke exposure was found to decrease protein S-glutathionylation compared to the increase observed in the acute three day model of cigarette smoke exposure. We did not observe a difference in basal level of protein S-glutathionylation between WT and Grx1 KO mice, but smoke failed to affect PSSG in Grx1 KO mice. The level of PSSG in the lungs of Grx1 KO mice was significantly higher after smoke compared to the wild type animals. The differential response in S-glutathionylation of proteins in the two models might thus be involved in the discrepancies found in lung inflammation between acute and chronic smoke exposure. It has been reported by our laboratory that upon oxidative stress, targeted S-glutathionylation of IKKβ inhibits its activity and thus inhibits Rel A nuclear translocation. Binding of NF-κB to its consensus sequence has also been shown to be negatively affected by S-glutathionylation. Secondly, we have previously shown that Grx1 KO mice and primary epithelial cells isolated from these mice, show a markedly diminished response to LPS with respect to NF-κB activation and inflammatory mediator production. Since the Grx1 KO mice have more overall S-glutathionylation in the lung and in the BAL cells after smoke exposure, this might, through the inhibition of nuclear Rel A translocation, contribute to the decreased inflammatory cytokine production observed in the Grx1 KO mice after four weeks of cigarette smoke exposure, compared to wild type mice. So it is likely that the differential response in PSSG levels in the acute versus the more chronic cigarette smoke exposure causes the difference in phenotype in the mouse lung. When we assessed PSSG in lavaged cells and fluid from smoke compared to air exposed mice, we found increased S-glutathionylation of proteins in contrast to the observed decrease in whole lung tissue. In lung tissue, S-glutathionylation was decreased despite a decrease in Grx1 mRNA expression. When we isolated primary macrophages from mice and exposed them to cigarette smoke extract, there was a decrease in Grx1 protein levels. These findings in macrophages are in line with our previously published data showing decreased Grx1 levels in various pulmonary epithelial cells after smoke exposure. Together these data indicate that overall protein S-glutathionylation can alter independently of differences in Grx1 and that the patterns of S-glutathionylation depend on cell type as well as stimuli and duration of stimulation. It should be noted that the alteration in the overall protein S-glutathionylation pattern does not mean that all individually targeted proteins would be affected in the same direction. Although we did not investigate individual targets, in of this manuscript we demonstrate that a differential effect on overall S-glutathionylation of proteins in response to smoke observed in different cell types, can still result in the same outcome. In particular, we demonstrate that smoke-induced KC production is decreased in both tracheal epithelial cells and macrophages isolated from Grx1 KO mice, irrespective of the effect on total protein S-glutathionylation. Taken together, we demonstrate in this manuscript that by using a knock out mouse model for glutaredoxin 1, this redox mediating enzyme has an important role in regulating cigarette smoke induced lung inflammation. The authors thank Yvonne Janssen-Heininger and Y-S Ho for providing glutaredoxin 1 k.o. mice, and Esther Theunisz, Gonda Konings, Greet Barbier, Eliane Castrique, Indra De Borle, Philippe De Gryze, Katleen De Saedeleer, Anouck Goethals, Marie-Rose Mouton, Ann Neessen, Christelle Snauwaert and Evelyn Spruyt for their technical assistance. [^1]: Conceived and designed the experiments: IK GGB EFMW NLR. Performed the experiments: IK KRB SWA RK ICA NLR. Analyzed the data: IK NLR. Wrote the paper: IK KRB GGB SCA EFMW NLR. [^2]: The authors have declared that no competing interests exist.

# Introduction CD8+ T cells are a key component of the adaptive immune response to HIV-1, both in acute, and chronic, infection. This response is directed by the presentation of HIV-1 epitopes on the surface of infected cells by host HLA Class I molecules. The HLA-B locus is the strongest genetic determinant of disease outcome, but beneficial effects of certain HLA-A, and HLA-Cw, alleles have also been reported. Although a small number of disease-protective and disease- susceptible alleles have been well characterised, ascertaining the impact of many alleles can be difficult due to factors including low phenotypic frequency, linkage disequilibria between alleles, and small effects on disease outcome. Based on these observations, and the known benefits of HLA Class I heterozygosity in mediating virologic control, we have recently investigated the potential for a co-operative additive effect between HLA alleles in suppressing viraemia, and demonstrated that certain combinations of alleles can work in tandem to mediate HIV-1 disease control. This effect is exemplified by HLA-A\*74 and HLA-B\*57, alleles that occur in linkage disequilibrium in some Southern African populations, making the role of each individual allele on disease control potentially difficult to ascertain. Larger cohorts allow for more refined analysis, enabling us to demonstrate that when each of two alleles independently exert a favourable impact, their co- occurrence may additionally have a combined effect. The test we have used here measures an effect where having two alleles working together additively has more impact on outcome (e.g. viral load or CD4+ T cell count) than having either one of them alone. This contrasts with a standard additive test which tests whether one allele has an additive effect above and beyond that of another. In the case where the first allele has little effect and the second allele a substantial effect, testing the two alleles against the first with a standard additive test would yield a positive result, whereas it would not with our test. We refer to the effect measured by our new test as a ‘co-operative additive effect’. The mechanism behind such effects is not clearly understood, but we have previously hypothesized that the reason for a combined benefit of HLA-A\*74 and HLA-B\*57 is – at least in part - the expanded repertoire of unique and complementary CD8+ T cell epitopes presented by the two alleles in combination. We here built upon our previous methods to further develop an extended systematic approach studying an enlarged Southern African cohort. This aims to identify, first, the contribution of individual alleles to HIV-1 disease control, and second, any potential co-operative additive effects between pairs of HLA Class I alleles. We have generalized our previous method so as to allow identification of these effects irrespective of locus and linkage disequilibrium. We also sought to explore the hypothesis that these co-operative additive effects are accounted for by the enhanced breadth of CD8+ T cell epitopes presented by pairs of co-operative alleles, developing a new ‘sharing score’ to quantify breadth of unique CD8+ T cell responses, and demonstrating a correlation between breadth of responses and viraemic control. # Results ## Univariate analysis confirms individual alleles that predict HIV-1 disease control or progression We first sought to identify single HLA alleles that are predictive of better or worse disease outcome with respect to viral load and CD4+ T cell count in our cohort of 2031 Southern African adult subjects with C-clade HIV-1 infection. Using the more stringent cut-off of q\<0.05 (FDR 5%), we identified nine HLA alleles significantly associated with viraemic suppression (highlighted in, upper panel), and ten alleles associated with preservation of CD4+ T cell count (highlighted in, upper panel), representing a total of 14 different HLA class I alleles that are of benefit in disease control in this cohort (q\<0.05). Five alleles, HLA-A\*74, -B\*57, -B\*58:01, -B\*81 and –Cw\*18 were statistically associated with both lower viral load and higher CD4+ T cell count with q\<0.05 (upper panels of). We also identified a total of nine different HLA alleles associated with a ‘hazardous’ (detrimental) outcome either with either respect to viraemic control (highlighted in, lower panel) and/or CD4+ T cell count (lower panel) (q\<0.05). Seven alleles were predictive of worse outcome with respect to both viral load and CD4+ T cell count: HLA-A\*66, -B\*08, -B\*18, - B\*45, -B\*58:02, -Cw\*06 and -Cw\*16 (with q\<0.05; lower panels of). Using the less stringent criterion of q\<0.2 (FDR 20%), we identified 17 alleles that are associated with favourable viraemic control and 13 alleles associated with poor viraemic control. Likewise, for CD4+ T cell count with q\<0.2, we identified 13 alleles associated with good outcome, and 11 alleles associated with lowered CD4+ T cell counts. Many of these associations between HLA Class I alleles and HIV-1 disease outcome have previously been reported by ourselves and other groups studying C-clade infected cohorts. However, a previous univariate analysis in 1211 South African subjects demonstrated fewer alleles that are significantly associated with viral set-point (in this earlier work, only five such HLA associations remained significant after correction for multiple comparisons). All five of these alleles previously reported to be significantly correlated with steady-state viral load feature again here in the extended list generated from analysis of an enlarged cohort. ## Identification of pairs of HLA alleles that co-operate to influence HIV-1 disease outcomes We identified six pairs of protective alleles that have a co-operative additive effect in mediating disease control (with q\<0.05); these pairs are highlighted in. Three of these six HLA pairs that are associated with disease control are in significant linkage disequilibrium (that is, the two alleles in the pair are in linkage disequilibrium). Based on the computational method used, which accounts for linkage disequilibrium, linkage between alleles does not drive these results. That is, if two alleles are either in complete linkage, or are never observed together, then our test yields no statistical power to detect a co- operative additive effect, because the test needs enough examples of alleles to observe together, and apart, in order to assess the impact of having both as compared to just one. That two alleles arise together more frequently together than expected by chance cannot alone drive the test statistic. The enhanced size of this current cohort allowed us to identify many more associations than were previously described ; in fact, of all the pairs of alleles we here identified to have a beneficial effect on disease outcome, only the HLA-A\*74/HLA-B\*5703 combination was previously noted to impact favourably on viraemic control. Another two allele pairs that were earlier reported to mediate a co-operative effect on disease outcome, HLA-B\*81/HLA-Cw\*04, and HLA-B\*39/HLA-Cw\*12, did not reach statistical significance in this current analysis (i.e q\>0.2 in each case). Using the same approach, we also detected four ‘hazardous’ pairs of alleles with q\<0.05 for which the expression of both alleles predicts worse disease outcome than expression of either one alone (these pairs are highlighted). All pairs of alleles mediating a significant co-operative additive effect (with the less stringent FDR of q\<0.2) are shown in (beneficial pairs) and (hazardous pairs). ## Alleles that mediate a co-operative additive effect to control disease target a greater breadth of the HIV proteome We hypothesized that co-operative additive effects in disease control might hinge on the presentation of combinations of alleles that present distinct epitopes from each other, as previously suggested for HLA-A\*74 and HLA-B\*57. Using the approach of calculating a ‘sharing score’ to quantify breadth of epitope coverage, as described in, we demonstrated a significant correlation between the sharing score and the p-value of an additive effect for VL (R = −0.08, p = 0.03; data not shown). The direction of this correlation is in the expected direction, i.e. the negative R-value demonstrates that a larger sharing score (reflecting a greater breadth of epitope coverage) correlates with a smaller p-value (indicative of a stronger co-operative additive effect between alleles); thus, the greater the epitope coverage, the more the co-operative effect. This suggests that some of the co-operative additive effect mediated by a pair of alleles can be accounted for by an increased breadth of CD8+ T cell targeting as compared to either allele alone. The effect was in the same direction (that is, a direction in which less sharing is correlated with being more co- operative, as expected), but not statistically significant, for CD4+ T cell count (R = −0.05, NS; data not shown). # Discussion These studies provide a useful resource in identifying HLA Class I alleles that mediate a co-operative additive effect in control of HIV-1 in C-clade infected African cohorts. The extended size of this cohort (\>2000 individuals) and adaptation of methodology to identify co-operative additive effects has allowed us to build on previous analyses, and to identify the impact of individual or paired HLA alleles with greater sensitivity. Importantly, however, in spite of this large cohort size, the analysis remains underpowered given the large number of HLA-pairs and the necessity of a multiple testing correction. These results are therefore likely an underestimate of the true extent of HLA co-operativity, and future studies employing more individuals, or a more restricted set of tests, are likely to reveal further instances of HLA co-operativity. Furthermore, our approach of using most HLA data at two-digit resolution was aimed to maximize statistical power to detect Class I influences on disease control. However, a caveat of this approach is that it limits the detection of possible differences occurring at high-resolution (often a micropolymorphism) level ; this could be addressed in future by use of larger cohorts. Effects on disease control were not always seen for both CD4 count and VL. Reasons for this likely include imperfect correlation between CD4 count and VL (r² = 0.22, p\<0.0001 by linear regression; data not shown), and that the linear models are only idealizations. Our analysis supports previous evidence that even highly beneficial responses, such as that restricted by HLA-B\*57, can be improved upon by addition of other T cell responses. The mechanism of this phenomenon has not previously been clearly characterised, but we have here demonstrated that – at least in part - the effect may be explained by the targeting of non-overlapping CD8+ T cell epitopes across the HIV proteome. The correlation between our ‘sharing score’ (reflecting breadth of epitopes targeted by a pair of alleles) and the probability of a co-operative additive effect mediated by these alleles was only weak (R = −0.08). Any computational method to assess breadth of epitope targeting is a challenge, especially given the density of overlapping CD8+ T cell epitopes in certain regions of the HIV proteome, the bias towards restricting highly targeted epitopes restricted by prevalent Class I alleles, and the complexity of immunodominance patterns. In addition, any single pair of alleles will also be impacted by the other four HLA Class I molecules expressed by a given individual, and the overall disease outcome will be influenced by many factors in addition to HLA genotype. Furthermore, there is no obvious effect size obtainable for the co-operative additive test, and even if there were it would be possible to have large effects for pairs which were not statistically significant. For these two reasons, we chose to measure correlation with the p-value from our test. These difficulties notwithstanding, these data nevertheless do highlight that two alleles which present different epitopes can each confer a separate benefit (or hazard) to the individual; thus having both of them is better (or worse) than having just one of them and a co-operative additive effect is at play. However, if two alleles present many of the same epitopes (as exemplified by HLA-B\*57 and -B\*58:01, or HLA-B\*42 and -B\*81), they are less likely to act together co-operatively – having one of them may be little different from having both. This effect is also underscored by the phenomenon of heterozygote advantage, which may be mediated by increased breadth of epitopes presented by HLA class I heterozygotes compared to homozygotes. As HLA-peptide complexes are ligands not only for T-cell receptors on CD8+ T cells, but also for KIR receptors on NK cells, another potential reason for the favourable (or hazardous) interaction of some pairs of HLA alleles is the combined effect of a CD8+ T cell response and an NK-cell response. Homozygosity for KIR ligands may also explain poor disease outcomes in subjects with certain HLA Class I combinations, although many of our pairs involved at least one allele that is not a known KIR ligand. Characterising interplay between HLA alleles is made difficult by the presence of linkage disequilibrium between alleles. However, our test statistic will not be significant for two alleles simply because they are in linkage disequilibrium, but rather the test can find two alleles to have a co-operative additive effect *despite* their being in (incomplete) linkage disequilibrium, albeit with reduced power owing to fewer observations of the alleles acting one without the other. That is, if one observes each allele only in the context of the other, or never together, it is impossible to determine whether nor not they have a co-operative additive effect (hence these pairs removed from analysis; see section). However, because one needs to observe enough co-occurrences of the alleles, having alleles in incomplete linkage disequilibrium increases the power to detect co-operative additive effects. In summary, these data highlight the potentially potent interactions between HLA class I alleles to mediate HIV-1 disease control. Even CD8+ T cell responses which are independently associated with strong viraemic suppression and sustained immunological control can be improved upon by the co-expression of certain other favourable HLA class I molecules. This finding underscores the potential benefit of harnessing co-operative effects of multiple CD8+ T cell responses in the development of CD8+ T cell vaccines. # Materials and Methods ## Ethics statement Ethics approval was given by University of KwaZulu-Natal Review Board and the Massachusetts General Hospital Review Board (Durban cohort), the Office of Human Research Administration, Harvard School of Public Health and the Health Research Development Committee, Botswana Ministry of Health (Gaborone cohort), and the University of the Free State Ethics Committee (Durban and Kimberley cohorts). All subjects provided written informed consent. ## Recruitment and characterization of patients We recruited 2031 HAART-naïve, southern African adult subjects with chronic C-clade HIV-1 infection via four cohorts: (i) Durban, South Africa, ; (ii) the Gaborone region, Botswana ; (iii) Bloemfontein, South Africa ; (iv) Kimberley, South Africa. The exact timing of infection in each individual was not known, but all these subjects were either presenting with clinical features of HIV infection, or diagnosed by routine screening in pregnancy – in both cases, in keeping with chronic infection. Viral loads (VL) were obtained for 1873 subjects using the Roche amplicor 1.5 assay and CD4+ T cell counts were determined for 1871 subjects using flow cytometry. All subjects had either a viral load or a CD4+ count available for analysis; the majority (84%) had both. Although a single measurement of VL and CD4+ T cell count for each individual is a limited ‘snap-shot’ of disease, these parameters are known to correlate well with disease outcome/time to AIDS. HLA typing was performed from genomic DNA by sequence-based typing. We collapsed all HLA data to two-digit HLA-types, with three exceptions in which the four- digit type is most likely to be critical to disease outcome: HLA-A\*68:xx, HLA-B\*15:xx, and HLA-B\*58:xx. An HLA imputation tool was used to infer those alleles not collapsed if they were only typed to two-digit level for any individual. Data were removed for 75 subjects in whom the four-digit type for HLA-A\*68:xx, HLA-B\*15:xx or HLA-B\*58:xx could not be determined. Gag population sequences (p17+p24) were generated from genomic DNA for 1256 individuals, as previously described. ## Univariate analysis of impact of HLA Class I alleles on HIV-1 disease control We undertook a univariate analysis to assess the impact of individual HLA alleles on disease control. Such a scan has not always been applied in previous studies that have examined HLA associations with HIV viral setpoint or absolute CD4+ T cell count. As such, the contributions of HLA-A and HLA-Cw alleles that often have less impact than HLA-B have tended to be obscured. Disease control was defined as previously, using continuous-valued data (absolute CD4+ T cell count and absolute viral load) and discrete targets (‘controller’ defined as CD4+ T cell count \>250 cells/mm³; viral load ≤2000 RNA copies/ml plasma). The univariate analysis was performed using an LRT test with linear or logistic regression for, respectively, continuous-valued and discrete targets (for example, CD4+ T cell count is real-valued, whereas ‘controller’ was the binary, thresholded version of CD4+ T cell count). We evaluated only HLA Class I alleles occurring at a phenotypic frequency of ≥0.5%, and used False Detection Rate (FDR) q\<0.05 (5% false positive) or q\<0.2 (20% false positive). ## Multivariate analysis of impact of pairs of HLA Class I alleles on HIV-1 disease control To identify any two HLA Class I alleles with co-operative additive effects on disease control, we used previously published methodology. As described in the, we used the term ‘co-operative additive’ to describe interplay between two alleles that is more than a simple additive effect. Briefly, each HLA combination was tested to see whether an additive model for two alleles together performed better in predicting disease outcome than a model that did not allow both alleles to interact. In contrast to previous analysis, we here generalized the test by removing the restriction that correlations need to be in the ‘direction of control’, allowing for detection of combinations of ‘beneficial’ and ‘hazardous’ alleles, or two ‘hazardous’ alleles. Note that P-values were computed for our test in a non-parametric way - using permutations. The test statistic for an HLA pair was the difference in log likelihood between these two models after fitting each by maximum likelihood. P-values were obtained by 50 K permutations of one HLA allele in the test. As previously, correcting the analysis for cohort origin using cohort covariates was highly statistically significant, but richer lineage-correction using a linear mixed model (with a phylogenetic tree-based variance component using Gag sequences) provided no further benefit; therefore, cohort covariates alone were added to the analysis. In all paired analyses, we set two criteria for inclusion of a pair of alleles in the analysis (i) alleles must be expressed together in at least five subjects, and (ii) alleles must occur independently of one another in at least five subjects (thereby removing any pairs in near or complete linkage). The value five was chosen based on other similar work (e.g. Microsoft PhyloD, which routinely uses a ‘min count’ filter for the minimum number of times an HLA allele must appear). Because this filtering step does not consider VL or CD4+ T cell count values, it is a statistically valid approach, and is conservative in that it can only cause us to miss real associations, not to detect false associations spuriously (regardless of the actual filtering threshold used). Specifically, this filtering threshold was not manipulated in response to the data (we only ever used this one threshold). Such ‘min count’ thresholds are widely used in similar contexts – e.g. all genome-wide association studies where mean allele frequency and Hardy-Weinberg equilibrium thresholds are employed as a preprocessing step (for example, see). ## Statistical tests Linkage disequilibrium between HLA class I alleles was computed using Fisher's Exact Test using the on-line tool at the Los Alamos HLA molecular immunology database: <http://www.hiv.lanl.gov/content/immunology/hla/hla_linkage.html>. This method reports significant linkage following correction for the number of tests performed (in this case, threshold for significance is p\<1.9E-05). Statistical correction for multiple comparisons was performed using a False Discovery Rate (FDR) with thresholds of q\<0.05 (5% FDR) or q\<0.2 (20% FDR). ## Methods to identify correlation between breadth of epitope targeting and disease control In order to investigate any relationship between HIV-1 disease control (mediated by any pair of HLA alleles) and breadth of CD8+ T cell responses, we used IFN-g ELISpot data for 1010 South African subjects tested against a panel of 410 C-clade overlapping peptides (OLPs) spanning the entire HIV-1 proteome, as previously described. We first assigned likely HLA allele restriction(s) to each OLP using stepwise Fisher's Exact Test (FET) to control for linkage disequilibrium. In each iteration, the most significantly associated HLA allele was determined using FET, then all individuals who expressed that allele were removed and the next most significant allele (with corresponding p-value) was identified. All alleles associated with the OLP at q\<0.2 were considered restricting alleles. For each pair of alleles, we computed a ‘sharing score’ as a means of quantifying the breadth of unique epitopes targeted by any given HLA pair. This sharing score was calculated as the number of shared OLPs divided by the number of unique OLPs targeted by the pair. Thus a higher sharing score indicates less total breadth of epitope coverage across the proteome. For each pair of alleles, this sharing score was correlated with the p-value from the additive pairs analysis using a Pearson correlation. We confirmed the analytically-computed P-values yielded by Pearson by using a permutation test with 1000 permutations, and the P-values from both approaches were in agreement. Note that there is no obvious effect size obtainable for the co-operative additive test, and even if there were it would be possible to have large effects for pairs which were not statistically significant. For these two reasons, we chose to measure correlation with the p-value. Based on this approach, a larger sharing score (indicative of wider OLP targeting) correlating negatively with the p-value for an additive effect (where a smaller p value is indicative of a stronger co-operative additive effect between alleles) points to a relationship between breadth of coverage and two alleles acting co-operatively toward immune control. [^1]: JL, JC, and DH are employees of Microsoft and own stock in the company. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: PM JL JC JF DH PG. Performed the experiments: JL JC DH. Analyzed the data: PM JL JC. Contributed reagents/materials/analysis tools: JL JC DH. Wrote the paper: PM JL JC DH PG. Designed the software used in analysis: JL JC DH. Managed clinical cohorts/contributed patient data for analysis: PM RP KH JF DG DS CV PP PJ AO RS ZM TN BW.

# Introduction Economic globalization refers to flows of trade and capital among and between countries. Within sociology, a common, critical view on economic globalization is that the historical forces influencing how and when a given country becomes ‘integrated’ into global trade (and the world economy as a whole) conditions the potential paths of development open to that country (e.g.). Here, trade and other relations between countries act as structural mechanisms enabling wealthier, more core countries to maintain favorable terms of trade, which in turn negatively impacts less developed, more peripheral ones in a variety of ways. In this paper, our primary interest is in tracing how countries’ level of integration in international trade gives rise to between-country inequalities. By integration, we mean the extent to which countries are embedded in the global network of international trade, and we use network analysis to capture this level of embeddedness. A number of scholars (e.g.) use network analysis for measuring countries’ level of integration as an alternative to measures based on exports and/or imports over GDP, noting that such network measures better incorporate implicit notions of ‘economic integration,’ namely, the number and volume of trading ties, as well as the structure of regional trading, that demonstrate the level of connectivity, and hence ‘integration’ in the world economy. By between-country inequalities, we are interested in emissions, wealth and mortality. For emissions, we make use of multiregional input-output (MRIO) analysis to distinguish between two forms of emissions, these being emissions produced within a country via that country’s manufacturing activities (referred to as *production-based emissions*), and emissions that is triggered by a country’s purchase by accounting for all emissions that is triggered throughout the whole global production chain and then allocated to the final consumer (referred to as *consumption-based emissions*). As means of an example, when a person purchases a toothbrush in the USA, this purchase triggers a supply-chain of production around the entire globe. Some parts hail from Asia, others from Europe, and all get shipped to Northern America for assembly. At each step in this global supply-chain, some form of air polluting emissions occurs, be it through manufacturing, assembly, or transportation processes. In a production based accounting approach, the emissions occurring within a specific country’s border, as it relates to the toothbrush’s production, would be assigned to that country. In contrast, with a consumption-based approach, all emissions along the global commodity chain of the toothbrush would be allocated to the country where the toothbrush was purchased. By accounting for emissions from these two approaches, we are able to distinguish between countries that generate emissions through their production activities, and those that trigger emissions through their consumption. In combination with the use of network analysis, which captures structural features of global trade networks, we see our use of MRIO analysis as providing a more powerful framework that reflects the globalized nature of emissions-intensive commodities. In addition to better understanding the globalized nature of emissions, we are interested in how emissions and wealth work together to affect countries’ mortality rates. Whereas production-based emissions are often experienced as a ‘burden’ that local populations must endure in exchange for participation in global trade, economic growth and wealth are often seen as the main, potential benefits of such participation. These trade-offs come together differently for different countries, such that the potential impacts of emissions experienced by local populations’ may be ‘buffered’ by the benefits associated with more wealth. We consider whether such wealth indeed translates into a buffer against emissions impacts by considering how countries’ share of global production-based emissions stands in relation to their share of global value added. Efficiency is often seen as one sign of a country’s level of development, in that more economically advanced societies have accumulated more wealth to invest in cleaner, more efficient technologies such as better filter technologies, less emissions intensive production structures, and a less polluting fuel mix. Finally, in looking at how emissions impacts countries’ populations, we have chosen to use a regional pollutant, in this case sulfur dioxide (SO2). This toxic gas is emitted via the combustion of fossil fuels in power plants and manufacturing facilities, and evidence suggests that local exposure to SO2 is linked to respiratory illnesses such as bronchitis or terminal lung cancer in both children and adults. For these reasons, we make use of SO2 (as opposed to a more global pollutant such as carbon dioxide) for our study. Taken together, our paper makes a number of contributions to the literature on international trade and emissions allocation. Although we are not the first to untangle production-based from consumption-based emissions a common shortcoming of this work is that it remains largely descriptive, demonstrating the regional and between-country emission disparities without attempting to explain them statistically (although see for an exception to this trend). In addition, we are aware of no other paper that simultaneously attempts to uncover the drivers of these different forms of emissions, as well as trace their effects on countries’ populations. The rest of the paper is structured as follows: we offer a review on economic and sociological literature on global trade, economic globalization and emissions. This is followed by a description of our longitudinal data, which includes country-by-country trade data on the sector level, as well as a number of country-level attributes. We discuss our methods, which include social network analysis (SNA), multi-regional input-output (MRIO) analysis, and panel data regression models and estimation techniques. We conclude with a discussion and reflection of the study, highlighting our methodological and substantive contributions to ongoing discussions pertaining to the WS and the environment. # Global Trade, Emissions, and the Effects on Mortality International trade is often described as a system of increasing interdependent economic relations. These economic relations form patterns, giving rise to structural features that shape characteristics and outcomes for countries. A number of studies exist that adopt a network approach to studying international trade relations. Some of these studies use network measures to describe the entire network structure, and in doing so, attempt to gauge the extent to which the global economy has become integrated overtime. Other studies consider how individual countries are conditioned by their position within this global trade network, and/or their level of centrality. Here, an important distinction is made between countries that are positioned in the network ‘core’ versus its ‘periphery.’ In network terminology, a core-periphery structure refers to a two- class partitioning where the core consists of a set of actors (or nodes) that are densely connected to one another and central to the entire network, i.e. they form a well-integrated block and share a similar set of ties to others in the network. In contrast, the periphery refers to a class of countries that are more or less isolated from one another and linked to the rest of the network mainly via ties to the core. Within this core-periphery structure, countries situated within the core are seen as being more integrated into the overall global trade network, and consequently accruing larger benefits, principally in the form of economic growth and/or development. Here, the well-integrated core is understood to exploit the periphery in an unequal exchange, such that financial investment and/or high-value goods flow from the core to the periphery, in exchange for undervalued goods produced in, or extracted from that region. Such unequal exchanges, moreover, prompt higher levels of emissions and resource exploitation in these less-developed, more peripheral nations. In addition, less-integrated, less-developed countries are looking for opportunities for economic growth, which often includes attracting foreign investment and/or the relocation of certain manufacturing activities from core-based transnational companies. As such, these countries’ regimes take a number of measures to attract this economic activity, such as relaxing labor laws and/or environmental regulations, with the consequential result that more environmental degradation is often experienced in these regions. Another body of research suggests, however, that populations of well-integrated, core countries tend to emit high amounts of emissions. Here, emissions are seen as rising because of the high presence of energy/pollution-intensive activities found within wealthier, more integrated countries. Although agents within these countries may indeed invest and develop more efficient, less polluting technologies, the gains in efficiency are understood to lead to increases in the overall rate of consumption, thus leading to increases in overall resource use and the pollution emissions associated with this. In a similar way, research that adopts a consumption-based approach to pollution suggests that wealthier, more countries are the biggest emitters. Here, wealthier, more integrated countries are seen as consuming a larger global portion of goods and services, and thus, triggering a disproportionately higher amount emissions through their consumption of goods and services. As such, they are seen as being more accountable for a larger share of global emissions. Taken together, the contrasting narratives presented above offer an unclear picture regarding the extent to which being integrated in global trade conditions the distribution of emissions among countries. Are core countries generating most of the emissions? Or have they managed to decrease levels of production-based emissions through externalization, thus prompting the same (or more) amounts of emissions in other countries? Finally, are the use of cleaner technologies truly capable of combating the negative environmental impacts of high-consuming countries? Part of the confusion in understanding just how much emissions are caused by these more integrated countries may lie in the fact that most studies fail to disentangle, within the context of the same study, pollution that is produced within a countries’ own boundaries (i.e. production- based emissions) and pollution emissions that are triggered through countries’ purchases and consumption habits (i.e. consumption-based emissions). In this paper, we contrast both types of pollution emissions and examine the extent to which being integrated, or ‘core’ conditions these two forms of pollution. In doing so, we seek to offer a better view of the intertwined, interdependent nature of pollution burdens and pollution responsibilities. In addition to disentangling which countries are the main producers of SO2, and which ones trigger the most SO2 through consumption, we are equally concerned with identifying who is most affected by this pollution, and how a country’s level of integration might help mitigate the impacts of this pollution. As core countries tend to accumulate more wealth than less-integrated ones, the fungible nature of this wealth should help societies adjust and potentially buffer the harmful effects of emissions. To assess whether such wealth translates into an economic buffer that lowers the negative impacts of emissions, we consider how countries’ share of global production-based emissions (SO2prod) stands in relation to their share of global value added. In doing so, we are essentially looking at countries’ relative efficiency scores, and asking to what extent such efficiency can help mitigate negative impacts of emissions. In assessing the negative impacts of emissions, we consider countries’ mortality rates, in particular, child and infant mortality, as infants and children are generally seen as the most vulnerable segments of a society, and hence, the ones most vulnerable to air pollution occurring within non-core countries. Research has shown that infants and children born in low socioeconomic conditions have lower access to resources and health services, and at the same time higher exposure to pollutants, thus increasing their risk for preterm births and premature death. In addition, research on water pollution and infant mortality and children’s health has shown that the periphery suffers higher mortality rates than the core, and such high mortality rates are generally seen as resulting from the multiple structural disadvantages found within the periphery, namely weaker institutions and less environmental safe-guards. Taken together, we predict that more efficient countries would be ones that would also hold lower mortality rates, regardless of how much that country might actually pollute. Stated differently, we argue that countries with higher shares of wealth (in relation to shares of emissions) have an economic buffer that would help mitigate the harmful impacts of pollution on infant and child mortality. # Material and Methods This section describes our dataset, our variables and measures, the methods for constructing these measures, and panel regression techniques. ## Data and Data Transformations Our trade data were extracted from the Eora database. Eora is a multi-region input-output database that provides a time series of high resolution input- output (IO) tables with matching environmental and social satellite accounts for 186 countries. The MRIO tables from Eora contains trade flows, production, consumption and intermediate use of commodities and services for 26 sectors, both within and between 186 countries (see <http://www.worldmrio.com/> for more details). Our data covers a 20 year time span (1990–2010). The benefit of using the global MRIO data, (combined with the SO2 data described below), is that we are able to calculate how SO2 is triggered by both consumption and production processes along the entire global production chain. As part of our intention here is to distinguish between consumption-based and production-based emissions on the global level, such detailed economic trade data are necessary. Although Eora contains data on 186 countries and regions, we were not able to find corresponding data for our other variables of interest for all countries in this dataset, as some countries in Eora have been aggregated into supranational regions. In spite of this constraint, we managed to gather health data for 172 of the 186 cases (see the full listing in the SI). For calculating value-added (VA) and consumption-based SO2, we made use of the full multi-regional input-output (MRIO) database without any transformations. For calculating countries’ level of integration, however, we transformed the dataset, following guidelines established by previous sociological work on global trade networks: first, we aggregated all 26 sectors to form one country- by-country trade matrix consisting of valued data. A single, valued trade network dataset is often preferred for operationalizing ideas of coreness or integration, as a country’s position is determined not only through the quantity and patterning of trading ties to others, but also considers the trade volume of those ties. In addition, as rows in the trade matrix correspond to exports and columns to imports, we took the trade matrix and its transpose, then summed the two together to arrive at a symmetrized matrix that combines, for each country, information on that country’s exports and imports. By summing the export matrix to its transpose (i.e. the import matrix), we focused attention on the structure of trade, as opposed to the directionality of trade ties (18, 22). Lastly, to smooth out any skewness, we took the square root of each cell in the symmetrized matrix. Our SO2 emission data are at the sector level, and were also collected from the Eora database. One thing to note about these SO2 data is that they refer to the emissions themselves, and not to concentrations of SO2 emissions. Concentrations of SO2 take into account whether or not air filters and other cleaning technologies are used to trap the emissions and lessen the amounts that escape into the atmosphere. Analyses of SO2 and economic or health implications are often limited to examining SO2 emissions rather than their concentrations. Concentrations would be better correlated to health impacts but less so to production or consumption based emissions due to other ambient and atmospheric factors and cross-boundary pollution. As SO2 concentrations data are not available for many countries in our sample, we are left with examining SO2 emissions. ## Measures A number of analysts have used social network analysis (SNA), a methodological approach for analyzing relational data, to measure structural ideas of global trade. Although no single network measure has been widely adopted for assessing countries’ level of integration in global trade (see for review), our use of SNA is in keeping with recent research focusing on global trade networks (e.g.). In particular, we made use of the continuous coreness procedure for measuring countries’ level of integration within the world trade network. This procedure fits a core/periphery model to an observed network to identify the extent to which the observed network approaches an ideal core/periphery structure. In an ideal core/periphery structure, the ‘core’ is a block of actors who are tied to one another, and in addition, have ties with many other actors in the network. To be core, then, is to be highly central in one’s own right, as well as part of a dense block of other highly central actors. In contrast, peripheral actors form a second block in which members are largely isolated from one another, and any ties they do hold are with the core. The coreness procedure introduced by Borgatti and Everett proceeds in determining what type of partitioning of actors (in our case, countries) in the observed network most closely brings that network towards an ideal core/periphery partitioning. The procedure results in a vector of scores assigned to countries, which range between 0 and 1, the higher values indicating a country being more core, and lower values indicating the country being more peripheral. The advantage of this procedure is that it results in a ratio-scaled vector of scores, thus enabling a higher degree of precision in making cross-country comparisons. We refer to this resulting vector of scores as our level of ‘integration’ measure for countries. A full listing of all countries and their integration/coreness scores can be found in. Our production-based SO2 measure consists of the data collected from the Eora database. Our consumption-based SO2 measure was constructed using MRIO Analysis (see for an additional example). At its core, MRIO analysis is an accounting procedure relying on national economic input-output (I-O) tables (the global MRIO dataset consists of 186 such national I-O tables) and international trade matrices, depicting the flows of money to and from the various sectors of the national and international economies, thus revealing each sector’s entire supply chain. This method has been applied to global trade studies on land use, water consumption, CO₂ emissions, materials use, and biodiversity loss. The process begins with the Leontief inverse of the MRIO matrix, which essentially involves multiplying an economy’s input requirements with each other by an infinite number of times, thus representing infinite rounds of production layers triggered by the previous set of inputs to fulfill production for final consumption. As such, computing the matrix inverse accounts for all the direct and indirect inputs triggered by final demand for a consumption item in any given country. Second, we calculated the total input requirements to satisfy final demand by multiplying the inverse matrix by final demand of a particular consumption item in any country. Finally, to account for total consumption-based SO2 emissions, direct pollution coefficients for each sector and each economy, showing the amount of pollution that is created for the production of each unit of economic output, are multiplied with the total inputs that are triggered by final demand. Technical details on consumption-based emissions can be found in Information. For measuring mortality, we used child and infant mortality data (each per 1000 live births) downloaded from UNICEF’s Millennium Development Goals database (please visit <http://mdgs.un.org/unsd/mdg/Data.aspx> for details). We included these measures to capture the potential impacts of SO2 and a country’s level of integration on vulnerable segments of a country’s population, i.e. children under the age of 5 and 1. In our regressions, we used the natural logarithm of these two mortality variables in order to normalize their distributions. To measure countries’ efficiency levels, we divided countries’ share of global SO₂ emissions (%SO₂) by their global share of value-added (%VA), such that (NEM = %SO₂/%VA). As such, this NEM essentially translates as a *normalized* measure of a country’s efficiency, as all countries are being assessed in relation to global totals. This enables us to see how countries compare in terms of *global* shares of emissions and wealth, as opposed to comparisons based on individual country attributes, as is found in more traditional efficiency measures (e.g. SO₂/GDP). Our major control variables include population size (in 1,000,000s) and urbanization (percentage of population estimated to live in urban areas in a country), both taken from the World Bank’s database ([http://www.worldbank.org](http://www.worldbank.org/)). Research has shown that population size is positively linked to forms of environmental degradation, including air pollution. Similarly, research has shown a positive link between a country’s level of urbanization and environmental degradation. Urbanization has also been shown to be positively linked to infant mortality in the periphery, although in the core, urban centers have historically been places where wealth and other key resources necessary for a higher quality of life become concentrated (e.g.), thus potentially mitigating the health threats of air pollution. In addition, we have also included countries’ health expenditure as percentage of GDP and countries’ fertility rates as control variables for regression models predicting infant and child mortality. Data for both these variables was downloaded from the World Bank ([http://www.worldbank.org](http://www.worldbank.org/)). All variables’ means and standard deviations are shown below in. ## Regression Analyses For assessing which countries are causing SO₂ pollution, we regressed countries’ integration scores against our two SO₂ pollution measures, controlling for population size and urbanization. For assessing the link between countries’ efficiency (NEM), and their level of integration, we regressed countries’ integration scores on their NEM, with the same controls. Finally, to gauge the extent to which countries suffer from their pollution costs, we regressed mortality rates on levels of integration, NEM, and production-based SO₂, alongside other controls. Our actual regression technique was a time-series cross-sectional Prais-Winsten estimation technique with panel-corrected standard errors (PCSE) as used by Jorgensen and Clark (2012). This technique allows for disturbances that are heteroskedastic and contemporaneously correlated across panels. The PCSE correction is capable of avoiding extreme overconfidence often associated with the popularly-used feasible generalized least-square estimator in the case of panel data sets in which the total time period *T* is smaller than total sections *N*. We control for common first-order autocorrelations of the disturbance terms and both period-specific and unit-specific disturbances. Because there are missing values in different places across variables, we employed ‘pairwise’ option associated with the xtpcse command in Stata (ver. 13) to include all available observations with non-missing pairs, thus maximizing the number of observations in the unbalanced panels. Prior to running regression models, we plotted our main predictor (countries’ level of integration) against our main outcome variables. These scatterplots are found below in. # Results To begin, we offer a global map showing countries’ level of integration according to year 2010, where darker shades indicate higher levels of integration. The pie charts depict well-integrated countries’ total imports, again for year 2010. Next, we present our regression results, starting with results for production- based SO₂ and consumption-based SO₂, as shown below in. Here, we see that, for both emission outcome measures, ‘integration’ holds a positive, highly significant coefficient, even after the controls are entered into the model. This finding suggests that the level of countries’ SO₂ pollution is positively associated with their level of integration in global trade. In addition, consumption-based SO₂ is more strongly correlated to integration than production-based SO₂. In particular, models 2a and 4a show that the integration coefficient (elasticity) is much stronger (i.e. more than doubled) in relation to consumption-based SO₂ than production-based SO₂, once we control for population size and urbanization. Finally, the explanatory power of the models increases once the control variables are introduced, as reflected in the R² values. As such, we see evidence that core countries are responsible for larger pollution emission levels, relative to less core countries, both through their at-home manufacturing activities, and also, through their consumption habits. Our findings thus suggest support for both our narratives: core, well-integrated countries are both the larger emitters of SO₂ emissions via manufacturing, and they also appear to be the main ‘externalizers’ of emissions, in this case via consumption. To assess whether being more integrated translates into attaining a higher ‘buffer’, we turn to results presented in. Here, we note the significant, positive relationship between levels of integration and NEM scores, both with and without controls. These findings indicate that as countries become more core, they gain larger shares in global SO₂ than shares of wealth (regardless of how much they might actually emit). Thus, there appears to be a tendency for more integration leading to less efficiency, i.e. *not* developing an economic buffer against emissions. To what extent does an increase in efficiency have a pay-off, in terms of mortality rates, for individual countries? Integration does not seem to go hand in hand with a lower NEM score, but would lower NEM scores translate into an economic buffer to help reduce mortality rates for infants and children? below shows regression model results exploring this question. For both sets of models in, we see that countries’ level of integration holds a negative and statistically significant relationship with mortality rates, suggesting that more core countries have lower mortality rates. In addition, countries’ NEM scores have a positive and statistically significant relationship to mortality, suggesting that countries with higher shares of emissions in relation to shares of VA suffer higher rates of mortality. In short, it appears that more core countries with higher economic buffers (in the form of lower NEMs) experience lower mortality rates. Interestingly, levels of production- based SO2 is not in any way linked to mortality, implying that countries might potentially have high levels of SO2 emissions within their own boundaries, but this is insignificant if the same countries are well-integrated, and have a low NEM score. As such, more core countries may have managed to ‘cloak’ the potential negative impacts of SO₂ emissions through making a number of adjustments (a lower NEM score, more urbanization, more spending on health, and so on), and these adjustments act as causal mechanisms to buffer the real impacts of SO₂ on the population. We also note that our controls operate largely as expected: i) the urbanization coefficient is negative and significant, suggesting that countries with higher portions of an urban population experience lower mortality rates; ii) the health expenditure coefficient is also negative and statistically significant, implying that higher spending on health per unit of GDP coincides with fewer deaths, and iii) fertility rates are positively and significantly linked to mortality, implying that countries with more births tend to suffer more deaths. Taken together, when we ask ourselves who is most affected by SO2 emissions, our findings offer a complex picture. More integrated countries experience lower mortality rates, yet they appear to do so via a number of mechanisms. One such mechanism is a country’s NEM score. Low NEM scores (implying greater efficiency) appear to help reduce mortality, and thus, it appears that higher levels of integration help countries acquire a stronger economic buffer to mitigate the negative impacts of SO₂ on mortality. In addition, higher levels of integration coincide with higher levels of urbanization, higher health expenditures and fertility rates. All these mechanisms thus appear to be working together to give more core, integrated countries an advantage over less-core ones with regards to mortality rates. Thus, the simple answer to the question, ‘who is most affected by pollution emissions?’ is that less core countries are the ones most affected. Yet it is via a number of causal mechanisms associated with one’s position in the global trade network (as implied by Tables –) that this disadvantage arises. As robustness tests to these findings, we re-ran the same models shown in Tables – as stepwise regression models (see Tables – below). The stepwise regression model results largely replicate our findings presented in Tables. The main differences between the results/models include the fact that the ‘pairwise’ regression option (Tables –) includes all available observations with non-missing pairs, thus maximizing the number of observations in the unbalanced panels. In contrast, the standard stepwise regression models (Tables –) exclude those observations with missing values. In addition, the t-test in stepwise regression is less efficient, making the results slightly less reliable than those from the pairwise option. As such, although production-based SO2 appears as a significant predictor of mortality outcomes in, we return to our more conservative result(s) found in, which show production-based SO2 being an insignificant predictor of mortality outcomes. Other than these noted differences, the patterns in the data results are largely the same. # Discussion and Conclusion We began our paper with a distinction between two ways of accounting for SO₂ emissions: production-based and consumption-based SO₂. We showed that both ways of accounting for emissions hold a positive, linear relationship with countries’ level of integration. Well integrated countries not only emit larger quantities of SO₂ than less integrated ones, they also are the ones who trigger the most SO₂ throughout global supply chains through their consumption habits. Our second main research aim was assessing, ‘who is most affected by emissions?’ Answering this question involved two steps. First, we looked at the relationship between integration and NEM scores. The literature suggests that being core gives countries a number of ‘spill over’ benefits, and thus, we argued, it is not so much that more integrated countries stop polluting, but rather, that their shares of wealth should exceed their shares of emissions, and such an excess in wealth should be perceived as an economic buffer that outweighs pollution costs. Our results showed that, contrary to what we expected, higher integration levels corresponded with higher NEM scores. Although more integrated countries tended to have higher NEM scores, our next round of analyses indicated to us that more core, integrated countries with stronger economic buffers appeared to be in a better position to mitigate the negative impacts of pollution, the so-called ‘costs’ of SO₂ emissions. In particular, an increased economic buffer (in the form of a low NEM) coincided with more urbanization and higher spending on health, all of which could help mitigate the negative impacts of emissions in the form of lower mortality rates. Thus, our findings suggest that being well-integrated works in conjunction with a number of mechanisms–these being countries’ NEM scores, their level of urbanization and amount of health expenditure–to effectively lower mortality rates. In other words, countries’ position within the larger trade network implies a number of spill-over benefits, as suggested by the literature. As a robustness check to our analyses, we re-ran the same set of regression models for predicting air pollution and mortality outcomes, replacing level of integration with two alternatives. These two alternatives included i) a categorical variable for trade network integration and ii) GDP per capita. The categorical measure was the same used in other trade network studies, and GDP per capita was used to replicate other globalization studies focused on the relationship between emissions and wealth. The details of these variables, their measurements, and the regression results can be found in our SI (see Tables A-D in Information File). In sum, the results for these alternative models largely replicated those presented here in Tables, and the R² values were very similar. The main difference between results presented in Tables – and those based on the two alternative measures is that the coefficients for the alternative measures proved weaker than those using integration. In terms of our contribution to the literature: although we are not the first scholars to disentangle consumption-based from production-based emissions, we are unaware of any research that has used such an approach to make theoretical arguments about emissions inequalities and mortality outcomes resulting from trade network position in the way we have demonstrated here. By doing so, we have shown, quite clearly, evidence for well-integrated, core countries being both major polluters as well as a major externalizers of emissions. Further, our focus on SO2 has enabled us to see how global trade patterns can have real local impacts in the form of mortality rates. As such, we have moved beyond fundamental concerns pertaining to inequalities resulting from globalization (e.g. wealth and emissions) to show how such inequalities translate into real ‘life or death’ issues for given societies. Our study has, through a series of analytical steps, slowly developed a picture of the role core countries have on both emissions and mortality outcomes. As data becomes available, future research will track these processes over a broader range of environmental outcomes, such as land displacement, water usage, and other forms of emissions. When thinking about environmental justice, such comparisons between consumption-based forms of environmental degradation and degradation occurring within a country’s boarders are needed for deepening our understanding of who is responsible for and who suffers from environmental harm within the global trade system. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: CP KF LS. Performed the experiments: CP KF LS. Analyzed the data: CP KF LS. Contributed reagents/materials/analysis tools: CP KF LS TWM. Wrote the paper: CP KF LS.

# Introduction The recovery of athletes following training and during competition are a constant concern for technical staff since inadequate recovery can lead to fatigue, decreased performance, and increased potential for injury. The ideal balance between training, competition, and physiological recovery are considered important factors in order to optimize performance. However, there is little scientific evidence supporting the effective recovery of athletes by means of commonly used methods. Typical Brazilian jiu-jitsu training sessions involve warm up, technical drills and combat simulation. This type/method of training can cause muscle damage and may decrease competitive performance. Additionally, physiological recovery between training sessions is pertinent to those athletes who participate in training cycles with a high number of competitions. Recent studies indicate that the post-training use of cryotherapy results in the reduction of markers of muscle damage, such as creatine kinase (CK) and lactate dehydrogenase (LDH), while allowing for the maintenance of strength and a hypoalgesic effect in Brazilian jiu-jitsu athletes. Similar results were obtained after a simulated Brazilian jiu-jitsu competition, where the findings indicated a significant reduction of CK, LDH and hypoalgesia among the athletes who participated in the study. However, research carried out by Broatch et al. with physically active individuals who were submitted to intensive training, established that the use of thermoneutral water obtained similar results to those found when using cryotherapy. This indicates that the physiological benefits (improved classification for pain threshold and readiness for exercise) demonstrated by cold water immersion may be related to the placebo effect. Within the context of these findings, evaluation of the various recovery modalities currently being utilized is needed. Hyperbaric oxygen therapy is a treatment where 100% oxygen is administered under pressure greater than one absolute atmosphere (ATA). The process is based on pressurizing the blood vessels between 1.5–3.0 ATA for 60–120 min periods, once or twice a day. The primary function of hyperbaric oxygen therapy is to accelerate the recovery of soft tissue by means of reducing local hypoxia, inflammation and edema. Hyperbaric oxygen therapy has become a recommended form of treatment for recovery from injuries among non-athletes. However, there is no scientific evidence that the method is effective in the treatment of elite athletes after training sessions and competition. As such, the present study aims to examine the effect of hyperbaric oxygen therapy after intensive training sessions using markers of muscular damage and perceived recovery/exertion among Brazilian jiu-jitsu athletes. It is hypothesized that treatment with 100% oxygen will favor the recovery of markers of anabolic/catabolic status and muscular damage as well as the perception of recovery among participating athletes. # Materials and Methods ## Subjects The study included 11 experienced male adult Brazilian jiu-jitsu athletes (age: 29.7 ± 6.6 years old, body mass: 84.6 ± 15.8 kg, height: 179 ± 6 cm) with 8 ± 7 years of regular and systematic practice (5 blue-belts, 1 purple-belt, 3 brown- belts and 2 black-belts). The following inclusion criteria were required of participants: a 3-year minimum period of regular training (to ensure an adequate level of experience among athletes); participation in official competitions in the six months prior to the study; and a minimum Brazilian jiu-jitsu training frequency of three times per week (typical duration of 1.5 h per session). Athletes who were injured, in a process of weight loss or used illicit drugs (i.e., anabolic steroids), antibiotics and/or anti-inflammatory drugs were not accepted for the study. Athletes were requested to consume dinner on the evening prior to the training and breakfast one hour before training. The athletes were allowed to freely hydrate with water throughout the training sessions. In addition, the athletes were instructed not to take dietary supplements before or during data collection. The data collection was carried out on Mondays and blood samples were collected before training by a team of nurses. Training started at 11 am in both conditions. The athletes were instructed not to perform any type of vigorous physical exertion over the weekend, thus allowing for a 48-hour period of rest. Sample-size calculations indicated that 10 athletes would be necessary to detect a statistical difference based an effect size of 0.5, a 1-β error probability of 0.8 and a P \< 0.05 for the main dependent variables. Thus, 12 athletes they were recruited according to the previous study by Shimoda et al.. However, one athlete was excluded from the study due to injury during testing. This research was conducted in accordance with International Ethical Guidelines and Declaration of Helsinki, and was approved by Hospital Celso Ramos and State University of Santa Catarina, Florianopolis, Santa Catarina, Brazil approval number 4530021.1.00005360. The participants provided written informed consent to participate. ## Experimental design This study utilized a randomized cross-over design where Brazilian jiu-jitsu athletes were subjected to two experimental conditions following intensive training sessions. The experimental conditions included either passive recovery for 2 hours or hyperbaric oxygen therapy for a similar period of time. The actual duration of treatment in hyperbaric oxygen was 1 hour and 40 minutes, but approximately 20 minutes were allotted for the athletes to arrive at and enter the hyperbaric chamber. Prior to and throughout the recovery process, blood markers of anabolic/catabolic status and muscle damage were measured, while the perceived recovery of the athletes were recorded. A wash-out period was utilized in order to avoid interference between the first and the second (carry-over) interventions. Subsequently, the alternate condition was undertaken for each participant. Accordingly, a 7-day period was adopted between the two time-points with the intent of avoiding oscillations in the examined variables among participating athletes. ## Training sessions The athletes carried out two typical Brazilian jiu-jitsu training sessions. Each session lasted for one hour and thirty minutes, comprised of a ten minute general and specific warm-up, technical training for twenty minutes (immobilizing, arm joint-locks and strangling) and six combat simulations of six minutes each, interspersed by four-minute periods of passive recovery. The combat simulations portion of the session was carried out with the highest possible intensity and the athletes received verbal encouragement from the coaching staff. In order to assess the intensity and assure that the sessions were similar, blood samples were collected from the ear lobe for blood lactate analysis at rest and after each combat simulation. Similarly, the rating of perceived exertion (RPE) scale was applied at the same time-point as the blood lactate measurements (see below for further explanation). ## Hyperbaric oxygen therapy The hyperbaric oxygen therapy was carried out by means of a multiplace Fogliene^® (model FH 2205) chamber with capacity for 14 people. The athletes were directed to the chamber, where they sat in individual chairs and oxygen was released through individual masks. The session started with a descent rate of approximately 7 meters per minute. The descent was interrupted at a depth of two meters (1.20 ATA) to allow for participants to put on their masks. From this moment, 100% oxygen concentration was released. At depths of five meters (1.50 ATA) and nine meters (1.90 ATA), the descent was interrupted in order to equalize internal ear pressure. Compression proceeded to a depth 13.9 meters (2.39 ATA), where it remained for 89 minutes. Thereafter, the depressurization process of the chamber was initiated at a rate of 4 meters per minute until a state of complete depressurization was reached. When the cycle ended, there was a treatment/recovery period (oxygen release) of one hour and forty minutes. Pre-treatment glycemia (blood glucose: 112.7 ± 19.3 mg/dL) and blood pressure (systolic: 129 ± 8 mmHg and diastolic: 87 ± 9 mmHg) were evaluated for all athletes and values were within normal ranges after the training session (i.e., before going in to the hyperbaric chamber). A medical team was available throughout the whole treatment/recovery process to attend to participants, but no complications were diagnosed and no medical assistance was required. ## Blood samples and Biochemical analysis Blood samples (15 ml) were collected from the antecubital veins at predetermined time points. The collections were performed through the use of Injex^® 20 ml syringes (Ourinhos, Brazil) and BD Vacutainer^® (Franklin Lates, United States of America) type Scalp 21 G disposable needles. After collection, 4 ml of blood were immediately dispensed into Petrodis^® (Sao Paulo, Brazil) 4 ml tubes with ethylenediamine tetra acetic acid (EDTA) to obtain the plasma. The remaining contents were dispensed in glass tubes without anticoagulant for serum extraction. Both tubes, with and without anticoagulant, were centrifuged at 3000 rpm for 15 minutes at 4°C, for separation into serum and plasma. Furthermore, intraclass correlation coefficient (ICC) values for the hormonal concentrations and cellular damage markers were calculated from the pre-training values using both experimental conditions. To measure hormonal action in the plasma, an enzyme linked immunosorbent assay (ELISA) was used to determine cortisol and total testosterone concentrations with the use of Arbor^® (Michigan, United States of America) commercial kits. The ICC for cortisol was 0.71 and 0.91 for total testosterone. For markers of cell damage, lactate dehydrogenase (LDH) was measured in serum by the kinetic method, and creatinine levels by the colorimetric method. Within the plasma, creatine kinase (CK) was measured by the kinetic method and aspartate aminotransferase (AST) and alanine aminotransferase (ALT) levels were measured by the colorimetric method. Analyses were performed with the use of Gold Analisa^® (Belo Horizonte, Brazil) commercial kits and a Bioplus^® UV-2000 (Sao Paulo, Brazil) spectrophotometer. The ICC for CK was 0.90, 0.87 for LDH, 0.79 for AST and 0.57 for ALT. To measure blood lactate concentration, blood samples (25 μL) were collected at the earlobe with heparinized capillaries and were immediately stored in microtubes containing 50 μL of sodium fluoride at 1% for analysis using a Yellow Springs YSI 1500^® lactimeter (Ohio, United States of America). Moreover, two solutions were prepared containing 50 μL sodium fluoride at 1% and 25 μL of standard solution of 5 mmol/L as a standardization method for homogenizing the blood samples according to the environment in which data collection was performed. Furthermore, ICCs for blood lactate were calculated throughout the training sessions, with the following results: pre-training (0.89) and after each-fight (1st: 0.74; 2nd: 0.86; 3rd: 0.78; 4th: 0.99; 5th: 0.67 and 6th: 0.97). ## Rating of Perceived Exertion and Perceived Recovery Status After the training session, the athletes were questioned about their RPE based on the 6–20 scale (10). The ICCs for RPE throughout the training sessions were calculated, with the following results: pre-training (0.81) and after each-fight (1st: 0.94; 2nd: 0.91; 3rd: 0.90; 4th: 0.86; 5th: 0.90 and 6th: 0.94). In order to verify the rating of perceived recovery (RPR) of athletes, the scale proposed by Laurent et al. was used, with scores ranging from 0 (very poorly recovered/extremely tired) to 10 (recovered well/highly rested), before, two hours after and 24 hours after training in both conditions (control and experimental). Moreover, the pre-training ICC for RPR was 0.81. ## Statistical Analysis The parametric data are presented as mean and standard deviation. The Kolmogorov-Smirnov test was conducted to determine normality and the Levene test was utilized to determine homogeneity of the data. A comparison across the different time points was performed using two-way (treatment and time) analysis of variance (ANOVA) with repeated measures, followed by a Bonferroni post hoc test. A Mauchly’s test of sphericity was used to test this assumption and a Greenhouse-Geisser correction was applied when necessary. All interactions (condition×time) are described in the tables. In the text, interactions are reported only for the same time-point in different conditions. Additionally, to evaluate the magnitude of the observed differences, the effect size was calculated (*eta squared*, η²) and interpreted as follows: \< 0.2 (small), \> 0.2 and \< 0.8 (moderate) and \> 0.8 (large). For the ordinal data, RPE and RPR, non-parametric statistics were used in accordance with Bishop and Herron. In order to compare conditions, the Wilcoxon test was used and data was represented by the median and percentiles (25% and 75%). Furthermore, the effect size for non-parametric data was calculated through the equation (*r* = *Z*/√*N*) wherein the r indicated the effect size and N is the number of observations. This values was interpreted in accordance with Field: \< 0.30 (small), \> 0.30 and \< 0.50 (medium) and \> 0.50 (large), respectively. The significance level was set at P \< 0.05. The data were analyzed using the Statistica software version 12.0 (Stasoft, United States of America) and the SPSS software version 20.0 (IBM, United States of America). # Results presents the lactate concentration and RPE among Brazilian jiu-jitsu athletes during the control and hyperbaric oxygen therapy conditions. For blood lactate, there was a time effect (F_6,120 = 225.04; P \< 0.001, η² = 0.918, large), with resting values being lower than any of the post-fight values (P \< 0.01). There was also an interaction effect (F_6,120 = 3.89; P = 0.001, η² = 0.163, small) where resting blood lactate concentrations were different from all post-fight values for both groups (P \< 0.001), but not between conditions for any of the time- points (P \> 0.05). For RPE, there was no difference between the two experimental conditions at the at any of the selected time-points (P \> 0.05). presents the RPR in Brazilian jiu-jitsu athletes during the control and hyperbaric oxygen therapy conditions. For RPR, there was a between-condition difference at post 2 h (Z = 2.52, P = 0.012; r = 0.76, large) and post 24 h (Z = 2.37, P = 0.018; r = 0.71, large), with higher values from hyperbaric oxygen therapy when compared to the control condition. presents the hormonal and cellular damage responses in Brazilian jiu-jitsu athletes during the control and hyperbaric oxygen therapy conditions. There were no condition or interaction effects for any of the hormonal concentrations or cellular damage markers (P \> 0.05). For cortisol, there was a time effect (F_3,60 = 52.6, P \< 0.001, η² = 0.72, moderate), with higher values post-training and post 24 h, and lower values post 2 h when compared to pre-training values (P \< 0.01). Furthermore, the values post-training were higher when compared to 2 h and 24 h post-training (P \< 0.05) and post 2 h values were lower when compared to post 24 h values (P \< 0.001).For total testosterone, there was a time effect (F_3,60 = 15.4, P \< 0.001, η² = 0.43, moderate), with higher values in pre, post-training and post 24 h when compared to measurements performed post 2 h (P \< 0.001). For CK, there was a time effect (F = 10.5, P \< 0.001, η² = 0.35, moderate), with higher values in all moments when compared to pre-training values (P \< 0.001). For AST, there was a time effect (F_3,60 = 22.4, P \< 0.001, η² = 0.53, moderate), with higher values in all conditions when compared with pre-training (P \< 0.001), and high value when compared to post- training values (P = 0.03). For ALT, there was a time effect (F_3,60 = 8.8, P \< 0.001, η² = 0.31, moderate), with higher values in all moments when compared with pre-training (P \< 0.05). For LDH, there was a time effect (F_3,60 = 33.6, P \< 0.001, η² = 0.63, moderate), with higher values post-training and post 2 h (P \< 0.001) when compared to pre-training values and higher values post- training and post 2 h when compared to post 24 h (P \< 0.001). # Discussion The main goals of this study were to investigate the effect of post-training hyperbaric oxygen therapy on physiological, perceptive and hormone responses in experienced Brazilian jiu-jitsu athletes. The blood lactate and RPE values were similar in the two conditions (experimental and control conditions) across time- points, indicating similar efforts and intensities during the two training sessions. No significant differences in physiological or hormonal responses were found between the control and hyperbaric oxygen therapy conditions. However, post-training RPR values (at 2 h and 24 h) were greater following acute hyperbaric oxygen therapy when compared to the control condition. The currently reported improvement in RPR following hyperbaric oxygen therapy appears to be in agreement with the results of Kim et al. in physically active men showing decreased fatigue perception after 40 min exposure to 1.3 ATA. However, with the altered RPR being the only significant difference between conditions, the potential impact of the placebo effect, as demonstrated by Broatch et al. in response to cold water immersion following repeated sprints, must be considered. Despite no concomitant changes in hormone concentrations and cellular damage markers, the current findings with regard to RPR must considered within the context of competitive athletes. Accordingly, the classification of RPR, as proposed by Laurent et al., may be an important indicator of readiness and performance throughout a training cycle. While the hyperbaric oxygen therapy intervention did not modify the cell damage markers or the hormone concentrations (cortisol and total testosterone), the results of this study indicate that the typical Brazilian jiu-jitsu training sessions increased CK, AST, and ALT concentrations immediately post-training, post 2 h and 24 h for both conditions. Likewise, the two training sessions increased LDH concentrations immediately post-training and post 2 h, but returned to resting levels after 24 h. Hence, the adoption of strategies for physiological recovery may be relevant to maintain performance in Brazilian jiu- jitsu athletes, especially during competitive periods. The treatment of injuries arising from athletic training and competition by means of hyperbaric oxygen therapy has increased exponentially over the past three decades. Notwithstanding, the benefits of hyperbaric oxygen therapy for treatment of sports injuries may be limited, because the location of the injury seems to influence the effectiveness of the treatment. Previous reports have noted that muscle injuries appear less likely to benefit from hyperbaric oxygen therapy for the purpose of recovery, than areas of reduced perfusion, such as tendons and ligaments. Nonetheless, there are no studies in the literature that have assessed the physiological recovery from athletic training, or combat sports, by means of hyperbaric oxygen therapy. Furthermore, Babul et al. did not detect a hyperbaric oxygen therapy-related reduction in muscle damage markers in sedentary college-aged women following execution of 300 eccentric maximal contractions (30 sets of 10 repetitions per minute) using an isokinetic dynamometer. In contrast, a recent study reported that acute treatment by means of hyperbaric oxygen therapy was able to reduce muscle fatigue during the execution of plantar flexion exercises in physically active subjects. Notably, the aforementioned study utilized a monoarticular exercise with fixed resistance. This does not reflect the physiological and metabolic demands of Brazilian jiu-jitsu training and competition, which are characterized by multi-joint movements performed at high-intensities with opponents of varying technical and physical fitness profiles. In relation to hormone responses, elevated cortisol levels immediately post- training and post 24 h were observed when compared to resting and post 2 h values for the two experimental conditions. Concerning total testosterone, lower values post 2 h were observed when compared to all other time points (pre, post and post 24 h). In turn, no data were found in the extant literature regarding the effect of hyperbaric oxygen therapy on the hormonal response to combat sports, or sporting activities in general. However, Passavanti et al. showed elevated testosterone concentrations after several hyperbaric oxygen therapy sessions (ranging from 4 to 23 sessions) in a patients with pathologic conditions and healthy controls. The above-mentioned study did not require physical effort and patients were subjected to a chronic protocol rather than a single treatment session as examined in the current investigation. Nonetheless, increased serum testosterone, a potential marker of anabolism and an indicator of recovery status, may occur following multiple hyperbaric oxygen therapy sessions which could be beneficial for Brazilian jiu-jitsu athletes. Conversely, cortisol has been utilized as a marker of catabolism. A previous study showed that 60 min exposure at 2.5 ATA with 100% oxygen was able to reduce cortisol concentrations after intervention. However, this study evaluated divers, who had distinct characteristics and training adaptations as compared to Brazilian jiu-jitsu athletes. Brazilian jiu-jitsu training requires great neuromuscular demand (aerobic power, grip endurance and muscle power to carry out the guard passing techniques, sweeps, takedowns, back control and submissions) as well as moderate activation of the glycolytic pathway within intermittent activities (high and low intensities). Cortisol plays an important role during physical effort, acting as a counter-regulator to maintain the glycemic levels of athletes. The physical effort demanded in the present study was likely higher than in the activities carried out by divers, who are familiarized with the pressurization process, both of which are factors that could have contributed to differences in cortisol response. Cortisol secretion is also influenced by circadian rhythm and may further explain discrepancies following hyperbaric oxygen therapy. Data collection in the current study started at 11:00 am, whereas Lund et al. conducted their diving investigation from 8:00 am. Moreover, the current results appear to be in conflict with those reported in an animal model, which showed an increase of circulating cortisol concentrations after exposure by means of hyperbaric oxygen therapy. Due to the lack of research, studies seeking to investigate the mechanisms triggered by hyperbaric oxygen therapy must be conducted to improve our understanding of this type of intervention. We can regard the absence of post-training (2 and 24 h) performance tests in both experimental conditions as one of the limitations of the present study. Furthermore, the inclusion of a third condition in the study, in which the athletes were situated in the hyperbaric chamber inhaling ambient oxygen through the individual masks would have allowed for a more thorough examination of the placebo effect. This condition was not feasible due to technical limitations of the hyperbaric chamber to provide ambient oxygen during pressuration which would have minimized the effectiveness of a placebo condition. Finally, the combat simulation protocol presented in this investigation yielded highly reproducible data as evident by high ICC values across training conditions for almost all of the analyzed variables. # Conclusion In summary, the currently investigated method of hyperbaric oxygen therapy, utilized during physiological recovery in Brazilian jiu-jitsu athletes, was not effective. As a consequence of these findings, hyperbaric oxygen therapy using altered protocols, such as increasing the session time and exposure, conducting two sessions in the same day, or chronic post-training exposure over the course of a training cycle, should be examined. In addition, future studies should evaluate the effects of this recovery modality on performance and consider including a placebo condition. We would like to thank the staff of the Hyperbaric Chamber of Maringa for conducting the experimental session and the Diagnostic Center of Maringa for carrying out the blood analyses. We are grateful to the medical team, led by Otávio Augusto Lorentte and Master student Gustavo Gillio for their assistance in data collection. Finally, we would also like to thank Júlio Albuquerque for his assistance in the translation of this article, the coach Rui Pedro da Cruz Venâncio and all the athletes who participated. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: BHMB LVA EF. Performed the experiments: BHMB DHF LVA JFSS JVDCE EF. Analyzed the data: BHMB DHF LVA JFSS JVDCE EF. Contributed reagents/materials/analysis tools: BHMB DHF LVA JFSS JVDCE EF. Wrote the paper: BHMB DHF LVA JFSS JVDCE EF.

# Introduction The human gastrointestinal (GI) tract is home to an array of bacteria, some commensals that are vital to human digestion and others that can cause acute or chronic infections. GI pathogens have been the subject of extensive studies, and many host-pathogen interactions in this tissue have been fully characterized. Thus, it is important to address the environmental setting in which these interactions occur and the factors that are involved. The GI tract represents its own microenvironment within the body: a vascularized, oxygenated, subepithelial mucosa bordered by the severely anoxic luminal region. The intestinal epithelial layer has been shown to be in a physiological state of oxygen deprivation, also known as hypoxia, characterized by daily fluctuations in oxygen tensions with oxygen levels ranging from 1 to 7%. This environment can be challenged even more upon onset of acute infections or chronic inflammation. In fact, infection sites often result in severe hypoxia, with oxygen levels dropping below 1% because of decreased oxygen permeation, increased consumption by invading pathogens and infiltration of recruited immune cells. Hypoxia has been shown to lead to numerous changes within host cells, including cytoskeletal rearrangements and alteration of membrane composition. However, it is still not entirely clear whether a hypoxic environment affects internalization of invasive bacteria such as *Yersinia enterocolitica* into epithelial cells. *Y*. *enterocolitica* is a gram-negative, facultative intracellular zoonotic pathogen that infects the gastrointestinal tract, causing a variety of diseases like gastroenteritis, acute enteritis and enterocolitis especially in children. The most common source of human infections with *Y*. *enterocolitica* is ingestion of contaminated food. After ingestion, *Y*. *enterocolitica* transverses the intestinal lumen and overlying mucosal layer, across the intestinal epithelial barrier and colonizes the underlying lymphoid tissues. The preferential entry of *Y*. *enterocolitica* into ileal Peyer’s patches seems to be facilitated by attachment to and penetration of epithelial microfold (M) cells. The uptake by epithelial cells is predominantly mediated by invasin of *Y*. *pseudotuberculosis* and *Y*. *enterocolitica*, but other adhesins like Ail and YadA can contribute to this process. Invasin-promoted internalization is characterized by a “zipper” mechanism. Invasin interacts with high affinity with several members of the β₁ integrin family through its extracellular C-terminal region. Interaction of invasin of *Y*. *pseudotuberculosis* was shown to bind with a 100 fold higher affinity than the integrin’s natural ligand, fibronectin. Integrins are a family of large transmembrane glycoproteins that function as receptors on the surface of cells, existing as heterodimers of one α and one β subunit, which are non-covalently linked. Among the 18 α and 8 β subunits, β1 integrins are the most widespread. They can be activated by internal as well as external cues, and thus are able to promote inside-out and outside-in signal transduction cascades. Several β1 chain integrins, mainly α5β1 along with α3β1, α4β1, α6β1 and αvβ1, were shown to be receptors for invasin. Invasin binding to integrins triggers receptor clustering, a step that is required for *Y*. *pseudotuberculosis* uptake into host cells. Consequently, a series of signaling cues is initiated, promoting the recruitment of tyrosine kinases like the focal adhesion kinase (FAK) and the involvement of the GTPase Rac1 that induces bacterial entry into non-phagocytic cells. The goal of this study is to investigate the effect of hypoxia on the β1 integrin-mediated internalization of *Y*. *enterocolitica* using Caco-2 cells as a polarized intestinal epithelial cell model. We suggest that cellular changes induced by hypoxia lead to a reduction in cell surface localization of host β₁ integrins thus decreasing invasin-integrin-mediated internalization of *Y*. *enterocolitica* into intestinal epithelial cells. # Materials and Methods ## Cell culture, bacterial strains and growth conditions Ethics approval was not required since a commercially available human epithelial colorectal adenocarcinoma, Caco-2, cell line (ATCC® HTB-37™) was used in the project. Cells were maintained in high glucose (4.5 g/L) Dulbecco’s modified Eagle medium (DMEM, Sigma), supplemented with 10% heat-inactivated fetal calf serum (FCS, Gibco BRL), and 50 U/ml Penicillin and 50 μg/ml Streptomycin (Sigma, Germany). Caco-2 cells were grown on polystyrene 24 well plates (Sardstedt, Nümbrecht, Germany) for 6 days post confluency. The bacterial strains used in this study are *Y*. *enterocolitica* 8081v bioserotype 1A/O:8, patient isolate, wild-type Y1/07 bioserotype 4/O:3, patient isolate, wild-type and YE21 (Y1, **Δ***invA*, kanamycin resistant Kn^R). Overnight cultures of *Y*. *enterocolitica* 8081v were grown at 27°C, and *Y*. *enterocolitica* Y1/07 and YE21 were grown at 37°C in Luria- Bertani (LB) broth. The antibiotics used for YE21 selection were carbenicillin 100 mg/ml and kanamycin 50 mg/ml. Normoxic incubations were performed in a tissue culture incubator at 37°C, 5% CO₂ in water saturated air, while hypoxic incubations were performed in an oxygen control hypoxia glove box (Coy Laboratory Products, Grass Lake MI, USA) at 37°C, 1% O₂ and 5% CO₂ in a humidified (100%) incubation chamber within the glove box. Alternatively, the prolyl-4-hydroxylase inhibitor dimethyloxalylglycine (DMOG; Sigma, Germany) was used to chemically stabilize the HIF-1α subunit under normoxia. DMOG, dissolved in water, was added to the media at 450 μM for 7 hrs. ## Oxygen measurements Immobilized PSt3 oxygen sensor spots (PreSens, Regensburg, Germany) were attached to the inside of 24 well plates and a polymer optical fiber (POF) was connected to a fiber optic oxygen transmitter that relayed the emitted light to a Fibox4 microprocessor (PreSens, Regensburg, Germany). In this manner, oxygen was measured non-invasively and was not consumed during the process of measurement. Caco-2 cells were grown for 6 days under normoxia and then either moved to 1% O₂ for 24 hr or kept at normoxia. On day 7 post confluency the cells were infected with *Y*. *enterocolitica* O:8 8081v (see below) and dissolved oxygen was measured at time of infection (24 hr), 1.5, 2.5, 4 and 6 hrs post infection. ## Infection and internalization Caco-2 cells were seeded at 0.82 x 10⁴ cells/cm² in a 24-well plate with growth area of 1.82 cm² and grown for 6 days post confluence in a normal tissue culture incubator. On day 6, media was changed and two plates were placed at 1% O₂ for 24 hr while one plate was left under normoxia for 24 hr. One well of Caco-2 cells was counted and cells were used at 1.65–2.75 x 10⁶ cells/cm². Cells were washed three times with PBS and incubated in DMEM without FCS or antibiotics. *Y*. *enterocolitica* O:8 8081v or O:3 Y1/07 strains were grown till OD₆₀₀ = 0.5 and used to infect Caco-2 cells at MOI 10. Plates were centrifuged at 142 *g* for 5 minutes (min) at 20°C and then incubated for 90 min at 37°C at normoxia or hypoxia, accordingly. After 90 min, media was removed and cells were washed with PBS to remove non-associated bacteria. FCS and antibiotic-free media with gentamicin 100 ***μ***g/ml (Sigma, Germany) was added to half of the wells for 60 min and the other half was kept with media only. The supernatant of cells incubated with bacteria and gentamicin was plated to ensure bacterial killing, and also taken for cytotoxicity assays. Cells were washed with PBS to remove antibiotics, trypsinized for 2 min with Trypsin-EDTA (Sigma, Germany), and then lysed with 0.1% Triton X-100 in media. Cell lysates were serially diluted and plated on LB agar. The total number of associated bacteria was determined by counting the colony-forming units (CFU) from wells without gentamicin and the number of internalized bacteria was determined by counting the CFU from wells with gentamicin. Internalization was calculated as percentage of gentamicin surviving bacteria relative to the total number of associated bacteria. ## Blocking of β1 integrin function Caco-2 cells were grown for 6 days post confluence in a normal tissue culture incubator. On day 6, media was changed and plates were placed at 1% O₂ for 24 hr while one plate was left under normoxia for 24 hr. One hr before infection, media was removed and replaced with media containing 45 ***μ***g/ml of either 6S6 anti-β1 integrin (Merck-Millipore, Germany) or the mouse IgG1 (Merck-Millipore, Germany). Cells were incubated for 1 hr at 37°C at normoxia or hypoxia. Cells were then washed and infection proceeded as described above using *Y*. *enterocolitica* O8 8081v at MOI 10. ## Infected cell lysis and FAK Western blots Cells were infected as mentioned in the previous section (infection and internalization) and after 1.5 hrs of infection the media was removed and each well was washed twice with phosphate-buffered saline. Cells were lysed with 1% Triton X-100, 2 mM sodium fluoride, 1 mM ethylenediaminetetraacetic acid in phosphate-buffered saline with protease inhibitor mix (Antipain dihydrochloride 1.48 μM, Pepstatin A 1.46 μM, Leupeptin 10.51 μM, Aprotinin 0.768 μM, Trypsin inhibitors 50 μg/ml and phenylmethanesulfonyl fluoride (PMSF) 1 mM; Sigma, Germany). Subsequently, cells were centrifuged at 17, 000 g at 4°C for 10 min and supernatants including cellular proteins were collected and frozen at -20°C until usage. Equal protein amounts (50 μg) of total cell lysates from each sample were denatured in boiling Laemmli buffer plus 50 mM dithiothreitol for 5 min. Samples were then subjected to 8% sodiumdodecyl sulfate polyacrylamide gel electrophoresis and transferred onto a PVDF membrane (Roth, Germany). Total amount of FAK was detected using FAK (D2R2E) rabbit monoclonal antibody and phosphorylated FAK at Tyr397 was detected using P-FAK Y397 (D20B1) rabbit monoclonal antibody (Cell Signaling Technology, Danvers, MA, USA). β-Actin (Santa Cruz Biotechnology, CA, USA) served as loading control. Quantification of band intensities was performed using Image J 1.48v (National Institutes of Health, USA). ## Whole cell lysis and HIF-1α Western blots Whole-cell extracts were obtained from Caco-2 cells grown for 6 days post confluency under normoxia. At day 6, they were either left under normoxia or placed under hypoxia for 24 hr after which they were lysed. Supernatants were removed and cells were washed in cold PBS over ice and scraped into 1 ml of lysis buffer (0.1% Nonidet P40, 300 mM NaCl, 10 mM Tris pH 7.9, 1 mM ethylenediaminetetraacetic acid in phosphate-buffered saline), with protease inhibitor mix (Antipain dihydrochloride 1.48 μM, Pepstatin A 1.46 μM, Leupeptin 10.51 μM, Aprotinin 0.768 μM, Trypsin inhibitors 50 μg/ml and phenylmethanesulfonyl fluoride (PMSF) 1 mM; Sigma, Germany). Subsequently, cells were centrifuged at 17, 000 *g* at 4°C for 10 min and supernatants including cellular proteins were collected and frozen at -20°C until use. Equal protein amounts (50 μg) of total cell lysates from each sample were denatured in boiling Laemmli buffer plus 50 mM dithiothreitol for 5 min. Samples were then subjected to 8% sodiumdodecyl sulfate polyacrylamide gel electrophoresis and transferred onto a PVDF membrane (Roth, Germany). β1 integrin was detected with a purified mouse anti-Integrin β1 antibody (BD Transduction Laboratories, USA). HIF-1 α was detected with a purified rabbit anti-human HIF-1α antibody (Merck-Millipore, Temecula, CA, USA). β-Actin (Santa Cruz Biotechnology, CA, USA) served as loading control. Quantification of band intensities was performed using Image J 1.48v (National Institutes of Health, USA). ## Brush border membrane isolation and sucrase activity Caco-2 cells grown for 6 days post-confluency under normoxia. At day 6, they were either left under normoxia or placed under hypoxia for 24 hr. Brush border membranes of Caco-2 cells were isolated by the divalent cation precipitation method. Cells were homogenized using a Potter–Elvehjem homogenizer in the hypertonic homogenization buffer (300 mM Mannitol, 12 mM Tris-HCl pH 7.1) supplemented with protease inhibitor mix (Antipain dihydrochloride 1.48 μM, Pepstatin A 1.46 μM, Leupeptin 10.51 μM, Aprotinin 0.768 μM, Trypsin inhibitors 50 μg/ml and phenylmethanesulfonyl fluoride (PMSF) 1 mM; Sigma, Germany). The homogenates were passed through a Luer-21 Gage needle and CaCl₂ was added to a final concentration of 10 mM and then centrifuged at 5,000 x *g* for 15 min to obtain the homogenate fraction (H). Homogenates were then incubated at 4°C for 30 min with gentle agitation and centrifuged again at 5,000 x *g* for 15 min. The pellet was then resuspended in 10 mM Tris-HCl + 150 mM NaCl pH 7.4 to obtain the basolateral and microsomal membrane vesicle fraction (P1). The supernatant was centrifuged at 25,000 x *g* for 30 min and the pellet was resuspended in 10 mM Tris-HCl + 150 mM NaCl pH 7.4 to yield the apical membrane/brush border membrane fraction (P2) while the supernatant contained all other soluble and small vesicular membrane-bound fraction (S). Subsequently, 50 μg of total cell lysates from each sample were denatured in boiling Laemmli buffer plus 50 mM dithiothreitol for 5 min. Samples were then subjected to 8% sodiumdodecyl sulfate polyacrylamide gel electrophoresis and transferred onto a PVDF membrane (Roth, Germany). β1 integrins were detected with a purified mouse anti-Integrin β1 antibody (BD Transduction Laboratories, USA) and sucrase isomaltase was detected using mAb anti-SI antibody HBB 3/705 obtained from Drs. Hans-Peter Hauri and Erwin Sterchi (University of Basel and University of Bern, Switzerland). Sucrase activity in the homogenates, basolateral membranes (P1 fraction), supernatant (S fraction) and brush border membranes (P2 fraction) was measured using 150 mM sucrose added to 25 μl of sample and end glucose was detected using GOD PAP fluid (Axiom Diagnostics, Worms, Germany) at 492 nm. Sucrase specific activity was calculated as μM.hour^-1.mg^-1 of protein. ## Immunofluorescence Caco-2 cells were grown on glass on cover slips in a 24-well plate for 6 days post confluence in a normal tissue culture incubator. On day 6, media was changed and plates were placed at 1% O₂ or left under normoxia for 24 hr. Cells were fixed with ice-cold methanol for 15 min and washed with Tris buffered saline with 0.01% Tween 20 (TBS-T). Coverslips were then incubated in blocking solution of 3% BSA with 0.01% TBS-T for 30 min at room temperature followed by permeabilization using 0.3% Triton X-100 for 15 min at room temperature. After washing, coverslips were incubated with 0.01 mg/ml mouse anti-β1 integrin (Merck-Millipore, Germany) or the mouse IgG1 isotype control (Merck-Millipore, Germany) diluted in 3% BSA with 0.01% TBS-T at room temperature for 2 hr. Coverslips were washed with 0.01% TBS-T and incubated with secondary goat anti-mouse Alexa Fluor® 488-labeled antibody (Invitrogen, Germany) for 45 min at room temperature, protected from light. After washing, coverslips were embedded in ProlongGold + DAPI™ (Invitrogen, Germany). Microscopy was performed using a Leica TCS SP5 confocal fluorescence microscope with a HCX PL APO 40X 0.75–1.25 oil immersion objective. Gain settings were kept the same when acquiring images of cells grown under the two conditions. ## Statistical analysis All experiments were performed in duplicate three independent times. Data were analyzed using Excel 2010 (Microsoft) and GraphPad Prism 6.0 (GraphPad Software). Differences between two or more groups were analyzed by using a One- way ANOVA with Tukey's multiple comparisons test. For Western blots and DMOG internalization statistics, unpaired, two-tailed Student’s *t*-tests were performed. The significance is indicated as follows: ns = non-significant, \* p ≤ 0.05; \*\* p ≤ 0.01, \*\*\* p ≤ 0.001 and \*\*\*\* p \< 0.0001. # Results ## Characterization of oxygen conditions during *Y*. *enterocolitica* invasion into Caco-2 cells In order to study the host pathogen interactions under hypoxia, the experimental settings of the culture conditions needed to be established. For our purposes, we used Caco-2 cells. This human cell line was grown to a monolayer with differentiated polarized intestinal epithelial cells. Differentiated Caco-2 cells develop brush-border microvilli typical of intestinal enterocytes and express a multitude of intestinal enzymes like sucrase-isomaltase. Interestingly, it has been recently shown that in Caco-2 polarized epithelial cell lines, β₁ integrins can be found apically at the tight junctions, colocalizing with the zonula occludens proteins. Furthermore, dissolved oxygen levels in the cell culture media were measured using optical sensors, based on the oxygen-dependent quenching of phosphorescent probes that is proportional to the oxygen level in the immediate surroundings. Infection incubations were performed under normoxia or hypoxia, thus resulting in three distinct conditions: normoxic pre-incubation / normoxic infection, hypoxic pre- incubation / normoxic infection and hypoxic pre-incubation / hypoxic infection. Oxygen measurements were performed over the course of 6 hours (hr) before infection and 6 hr following infection with *Y*. *enterocolitica* 8081v with an MOI of 10 (see experimental procedures for details). Normoxic pre-incubation of uninfected cells resulted in oxygen levels lower than 4% after 6 hr (left panel). After normoxic infection at time point 24 hr, cells show oxygen levels that decreased much faster than uninfected cells before similar levels (5% O₂) are reached after 6 hr (post infection) (right panel). Hypoxic pre-incubated cells reach levels of approximately 0.04% O₂ after 6 hr (left panels). Hypoxic pre-incubated cells that were infected under normoxia show a faster decrease in oxygen levels as compared to uninfected cells and finally reach 7% O₂ after 6 hr post infection (right panel). Hypoxic pre-incubated cells that were infected under hypoxia also show a slight yet significant difference in oxygen levels as compared to uninfected cells and finally reach 0.2% O₂ after 6 hr of infection (right panel). It is important to note that after 1.5 hrs of infection in all culture conditions, fresh media with or without gentamicin was added to the cells and corresponds to the peak in oxygen levels that immediately follow. ## Hypoxic pre-incubation reduces *Y*. *enterocolitica* internalization Caco-2 cells were grown for 6 days under normoxia and then either moved to 1% O₂ for 24 hr or kept at normoxia. After addition of *Y*. *enterocolitica* O:8 8081v at a multiplicity of infection (MOI) 10, plates were centrifuged in order to obtain uniform bacterial attachment to host cells and numbers of intracellular bacteria were identified by gentamicin survival assay. shows that cells pre-incubated under hypoxia had a significantly decreased number of internalized bacteria, after normoxic and hypoxic infection, compared to the normoxic control. Normoxic Caco-2 showed 12% internalized bacteria while hypoxic pre-incubated cells showed 2.4 and 1% internalized bacteria during normoxic and hypoxic infections respectively. shows that there was no significant difference in either the number of associated bacteria (2 B) or in the total bacterial number (2 C) respectively, in the different oxygen incubations. Finally, a lactate dehydrogenase assay (LDH) confirmed no significant cytotoxic effect of hypoxic incubation of Caco-2 cells. ## Beta one (β₁) integrin-mediated internalization In order to confirm the role of host β₁ integrins in *Yersinia enterocolitica* entry into intestinal epithelial cells, β₁ integrins were functionally blocked by using a 6S6 anti-β1 integrin antibody that binds to the extracellular fragment of the receptor. The normoxic or hypoxic incubated Caco-2 cells were treated for 1 hour and were then infected with *Y*. *enterocolitica* O:8 8081v at MOI 10. The results in show a significant decrease in bacterial internalization in β₁-integrin-blocked cells as compared to the controls under normoxia. Percent internalization was 6.8% for blocked as compared to 18.7% in untreated cells and 16% in IgG1 isotype-treated cells, in line with previous blocking studies. Blocking of β₁ integrins under hypoxia resulted in a slight but not significant decrease, 0.3% for blocked compared to 1.5 and 1.4% in untreated and isotype control cells, respectively. In summary, blocking under hypoxia revealed a strong decrease in internalization when compared to blocking under normoxia, whereas the number of associated bacteria was comparable between the blocked and untreated controls under either oxygen condition. ## Invasin-mediated internalization In order to investigate the invasin-β₁ integrin mediated internalization, two strains of *Y*. *enterocolitica* serotype O:3 were used to infect normoxic and hypoxic incubated Caco-2 cells. The *Y*. *enterocolitica* strain Y1/07 is a wild type invasin-expressing strain while YE21 is the respective invasin-deficient mutant strain (Δ*invA)*. Similarly to the infection with 8081v, 6-day post confluent Caco-2 cells were pre-incubated either at normoxia or hypoxia for 24 hr. Infection incubations were also performed under normoxia or hypoxia. The results in show that the wild type strain was internalized significantly less in cells pre-incubated under hypoxia, similar to the O:8 serotype. Infection with the invasin mutant showed a highly significant decrease in internalization as compared to the infection with the wild type strain. Similar to the *Y*. *enterocolitica* O:8 8081v wildtype strain, the number of associated bacteria showed no significant difference between the different oxygen conditions for either bacterial strain. ## Influence of oxygen levels on FAK activation The efficient uptake of *Yersinia* spp. into non-phagocytic cells via the invasin-integrin pathway requires the FAK that plays a central role in downstream signaling events. Binding of integrin leads to an increase in tyrosine phosphorylation levels in the cell, specifically at tyrosine 397, identified as the major site of autophosphorylation in cell adhesion. Therefore, in order to determine whether the hypoxia-induced decrease in internalization is also correlating with altered levels of phosphorylated FAK (p-FAK), total and p-FAK level were analyzed in cells that were pre-incubated under normoxia or hypoxia and then infected with *Y*. *enterocolitica*. Western blots show an increase in p-FAK at the site Y397 after infection as compared to uninfected cells under normoxia. Under hypoxia, the levels of p-FAK are low in uninfected cells, and were further reduced upon infection. Quantification of the Western blots shows that when compared to uninfected cells, infected cells show a distinct but not significant increase in p-FAK under normoxia and a significant decrease in p-FAK under hypoxia. Total FAK levels showed no overall difference between infected and uninfected cells under either oxygen condition. ## Beta one (β₁) integrin and HIF-1 alpha (α) protein levels The uptake of *Y*. *enterocolitica* into Caco-2 cells requires binding to host β₁ integrins, and since decreased bacterial entry was seen under hypoxia, it was important to investigate whether reduced oxygen conditions induce changes in β₁ integrin protein levels. Thus, Western blots were performed on whole cell lysates from 7-day post confluent Caco-2 cells incubated under normoxia or hypoxia for 24 hr. Interestingly, β₁ integrin protein levels were significantly decreased (0.5-fold) under hypoxia. Lower β₁ integrin protein levels may explain the hypoxia-mediated decrease of the *Y*. *enterocolitica* internalization rate. At the same time, protein level of the transcription factor hypoxia inducible factor HIF-1α, a global regulator of cellular response to hypoxia was significantly increased (4-fold) in hypoxic incubated cells. In order to confirm the decrease in β₁ integrin protein levels seen in the Western blots, immunofluorescent visualization of β₁ integrins in 7-day post confluent Caco-2 cells incubated under normoxia or hypoxia for 24 hr was performed. Representative images shown in confirm a distinct decrease in β₁ integrin intensity, visualized in green, and distribution on the cells in hypoxic samples compared to the normoxic controls. The isotype controls for normoxic and hypoxic staining are shown in respectively. ## Apical localization of β₁ integrin Differentiated Caco-2 cells display well-developed, brush-border membranes and express the active, apically enriched, intestinal disaccharidase sucrase- isomaltase (SI). To monitor any changes in cell polarization as a consequence of hypoxic exposure, the enrichment levels and activity of brush border SI was assessed in hypoxia pre-incubated cells. Furthermore, apical cell surface enrichment of β1 integrin was determined in hypoxic pre-incubated cells and compared to normoxic pre-incubated cells. For this, brush border membranes were separated from intracellular and basolateral membranes by the use of divalent ions (Ca²⁺) and subsequent separation by centrifugation (see section). First, sucrase activity was assessed in the brush border (P2) fraction versus the total cellular homogenates (H). As expected for well-differentiated cells, under normoxia, the activity of sucrase in P2 was approximately 3-fold higher than in the homogenate. Interestingly, sucrase activity was significantly lower under hypoxia, with activity of sucrase in P2 less than 2-fold higher than in the homogenate fraction. Next, the patterns of SI protein enrichment in the different membrane fractions was analyzed by Western blots. As expected, SI protein bands were mostly enriched in the P2 fraction under normoxia, in good correlation with the specific activity. Under hypoxia, SI bands were significantly increased in the soluble and homogenate fractions with a slight decrease in P1. Localization of β1 integrin under normoxia was highly enriched in the P2 fraction under normoxia, confirming its apical cell surface localization. Under hypoxia, β1 integrin levels were reduced in the H, P1 and P2 fractions accompanied by a significant increase in soluble vesicular membrane localization compared to normoxia. In summary, these results indicate that under normoxia, in 7-day postconfluent Caco-2 cells, β1 integrin receptors are found on the apical surface. In contrast, after 24 hrs of incubation under hypoxia, apical cell surface localization of β1 integrins was significantly decreased, in agreement with the observed decreased bacterial internalization shown above. ## Treatment of Caco-2 cells with DMOG reduces *Y*. *enterocolitica* internalization In order to determine whether HIF-1α plays a role in *Y*. *enterocolitica* uptake, a pharmacological agent was used to stabilize HIF-1α under normoxia. Dimethyloxalylglycine (DMOG) is a competitive pan inhibitor of prolyl-4-hydroxylases that degrade HIF-1α and it has been effectively used to stabilize HIF-1α in cells under normoxia. Therefore, 7 day post confluent Caco-2 cells were treated with DMOG or with media alone under normoxic conditions. The cells were then infected with *Y*. *enterocolitica* O:8 8081v with a MOI 10 (under normoxia), DMOG was kept in the media of treated cells throughout the infection process. shows a significant decrease in bacterial internalization in cells treated with DMOG (6%) as compared to the untreated control (17.3%). Neither the number of associated bacteria or bacterial growth control showed a significant difference between DMOG treated cells and untreated controls (respectively). Cytotoxicity of DMOG on Caco-2 cells was determined by performing an LDH assay, and no significant cytotoxic effect of DMOG treatment was found. Furthermore, we found that DMOG treatment resulted in a slight, but significant decrease (0.8-fold) in β₁ integrin and a significant increase (1.6-fold) in HIF-1α protein levels as compared to the untreated control. These data imply a possible involvement of HIF-1α in the decreased β₁ integrin levels that in turn lead to a reduced internalization of *Y*. *enterocolitica*. # Discussion The zoonotic bacterium *Yersinia enterocolitica* colonizes the human intestinal epithelium and its uptake is mediated by bacterial invasins that bind to host cell surface β₁ integrins. In human intestinal enterocytes, β₁ integrins are mostly localized on the basal and basolateral surfaces, however, on M cells, they are found mostly apically. Studies have shown that M cells are the primary site of *in vivo* intestinal epithelial invasion by *Yersinia* species. However, since oral *in vivo* infection with *Y*. *enterocolitica* was found to be lethal to mice, a direct correlation to human infections cannot be made. Furthermore, since M cells represent less than 1% of the total human intestinal surface, it is much more relevant and efficient to study bacterial invasion in polarized epithelial cells models such as CHO, HEp-2, MDCK and Caco-2. Caco-2 cells are a well-established model for the intestinal epithelium, due to their ability to polarize and differentiate into intestinal epithelial cells. It has been described that Caco-2 cells exist in three different states in culture: homogeneously undifferentiated at subconfluence, heterogeneously polarized and differentiated between 0 and 20 days after confluence, and homogeneously polarized and differentiated after 30 days. Furthermore, in the intermediate state, all cells exhibit apical cell-cell junctions and polarization but they display a high degree of heterogeneity in the organization of the apical surface and the development of the brush border membrane. In parallel to this differentiation process, a differential pattern of expression of extra cellular matrix (ECM) proteins and integrins along the crypt-to-villus is seen in epithelial cells. While undifferentiated cells express integrins along the entire cell surface, as cells differentiate, the integrins begin to exhibit a shift from a lateral to a more basal distribution. Interestingly, a comparative study of monocultures of Caco-2 on plastic plates and co-cultures with human intestinal mesenchymal (HIM) cells revealed a better basal distribution of β1 integrins in the co-culture system. Conversely, β1 integrins in polarized MDCK and Caco-2 cells displayed an apicolateral distribution, where they colocalized with tight junction proteins and enabled *Y*. *pseudotuberculosis* internalization. This disparity may be due to the varying differentiation stages at which the cells were used. Indeed, studies have shown that at advanced stages of cellular differentiation, bacterial internalization frequencies are substantially reduced. We have shown that in our Caco-2 culture system, 7-day postconfluent cells display an apical cell surface localization of β1 integrins. This may explain that *Y*. *enterocolitica* uptake can still be detected in differentiated Caco-2 cells in a cell culture model. In this study we show that hypoxic pre-incubated cells show less internalization of *Yersinia enterocolitica* compared to cells kept under normoxia. This phenomenon was in line with decreased protein levels of host β₁ integrin in hypoxic cells. The results of hypoxic pre-incubation of Caco-2 cells infected with *Y*. *enterocolitica* under normoxia largely excludes any effects of hypoxia on the bacterial expression of invasin, however, we cannot discount the possibility that such an effect may exist and contribute to the decrease in internalization. Furthermore, two other loci have been identified in *Y*. *enterocolitica*, Ail and YadA that contribute to cell surface attachment to host cells. *Y*. *pseudotuberculosis* strains lacking invasin were still able to associate with Caco-2 cells although internalization was abolished. Our results show that bacterial internalization is decreased under hypoxia and abolished in the absence of bacterial invasin or active β₁ integrins. Furthermore, reduced levels of phosphorylated FAK (p-FAK) were confirmed in hypoxia pre-incubated cells infected with *Y*. *enterocolitica*. Since invasin- mediated binding of β₁ integrins promotes recruitment of tyrosine kinases like FAK, these data support our hypothesis that the invasin-integrin- mediated internalization of *Y*. *enterocolitica* is altered under hypoxia. The GI tract has been described to be in a state of constant, low grade inflammation associated with hypoxia, with intestinal epithelial cells playing a pivotal role in mucosal immunity and response to this inflammation. Furthermore, chronic inflammation can be found in cases of inflammatory bowel disease (IBD), which has also been shown to result in hypoxic conditions. In fact, intestinal epithelial cells have revealed a strong resilience to low oxygen conditions and have efficiently adapted to this physiological state. Among these adaptation mechanisms is the accumulation of HIF-1, a transcription factor consisting of two subunits: the oxygen regulated α- and a constitutively expressed β-subunit. HIF-1α has now been shown to be present ubiquitously in human tissues and plays an important role in the cellular adaptation to hypoxia. Under normoxic conditions, the HIF-1α subunit is rapidly degraded by ubiquitination and subsequent proteosomal degradation mediated by oxygen- and iron- dependent prolyl hydroxylases (PHDs). The HIF-prolyl hydroxylases are dioxygenase enzymes that require oxygen and 2-oxoglutarate, rendering them key oxygen sensors. Hypoxic conditions allow for HIF-1α accumulation due to the interruption of its degradation pathway. Following HIF-1α stabilization, it dimerizes with the HIF-1β subunit and subsequently binds to specific hypoxia response elements (HREs) on target genes. HIF-1α binding regulates the transcription of several target genes that encode, among others, angiogenic factors, proliferation and survival factors, glucose transporters, glycolytic enzymes, and antimicrobial factors. Here, we have shown that the decreased internalization of *Y*. *enterocolitica* in hypoxic-treated Caco-2 cells goes along with increased protein levels of HIF-1α. Furthermore, the pan-hydroxylase inhibitor DMOG that is commonly used to stabilize HIF-1α shows a similar phenotype. These data lead to the hypothesis that HIF-1α may contribute to this process. However, an isoform of HIF-1α, HIF-2α, may also be involved in this mechanism since it shares several regulatory functions with HIF-1α and is subject to a similar oxygen-dependent degradation by prolyl hydroxylases. Therefore, studies with genetically modified cells are needed to verify this hypothesis. Interestingly, several studies described an effect of hypoxia on integrin expression, and in fact, a binding site for the transcription factor HIF-1α has been found on the β₁ integrin (ITGB1) gene promoter in colonic fibroblasts that results in a significant increase in ITGB1 induction under hypoxia. Besides transcriptional regulation of ITGB1, however, many hypoxia-induced post-translational modifications have also been reported. In renal epithelial cells, hypoxia results in over-activation of the calcium-dependent cysteine protease, calpain, which then leads to unrestrained cleavage of integrins. In several cell types, hypoxia increased β₁ integrin mRNA levels but hindered maturation and localization, thus resulting in decreased protein levels or activation. Additionally, oxygen-dependent modifications of the host cytoskeleton significantly affect the paracellular permeability, intracellular transport and the general endocytic uptake of particles. Finally, an exposure to hypoxia induces significant remodeling of the host cell membrane microdomains (lipid rafts) in alveolar epithelial cells. Lipid rafts have been shown to function in cell signaling, intracellular membrane transport, cell adhesion and host- pathogen interactions. Whether the hypoxia-induced changes in the host cytoskeleton and membrane microdomains play a role in decreased bacterial entry into epithelial cells remains unclear. The intricate relationship between hypoxia, infection and inflammation has been thoroughly investigated and besides HIF-1α various transcription factors are involved in this cellular stress response pathways, including Nuclear Factor Kappa Beta (NFκB) and cAMP Responsive Element Binding protein (CREB), among many others. Furthermore, many of these investigations have been performed *in vivo*, where all these factors contribute to the immune response during hypoxia and infection. Future experiments with genetically modified cells will be directed to characterize the cellular pathways involved in the hypoxia-modulated internalization of *Y*. *enterocolitica* into Caco-2 cells. Interestingly, a decrease of bacterial internalization under hypoxia was shown for *Shigella flexneri* into host epithelial cells in a predicted HIF-1α -dependent manner. Moreover, *Pseudomonas aeruginosa* entry into alveolar cells was decreased under hypoxia and after DMOG treatment, confirming the role of HIF-1α in an *in vivo* pneumonia model. However, there has been no mention of the involvement of the β₁ integrins in the hypoxia-induced decrease in internalization. Furthermore, evaluation of HIF-1α stabilization by pharmacological inhibition of prolyl hydroxylases, namely the HIF-1-specific PHD inhibitor AKB-4924, have revealed an important role for HIF-1α in boosting the innate immune response of keratinocytes against skin infections and of the intestinal epithelium in murine colitis. In light of these results, understanding the effects of hypoxia on epithelial cells during infections offers a new potential for pharmacological interference and the use of HIF-1α as a therapeutic target. This work was partially supported by DFG grant KO 3552/4-1 (MvK-B); N.Z. was funded by the German Academic Exchange Service (DAAD). [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MvKB NEZ HYN PD. Performed the experiments: NEZ. Analyzed the data: NEZ MvKB. Contributed reagents/materials/analysis tools: PD MvKB HYN. Wrote the paper: MvKB NEZ HYN PD.

# Introduction In the world, women total approximately 50% of the cases of infection by HIV. In Latin America, they represent 36% of the adults living with HIV/AIDS. In Brazil, women account for 35% of the cases of AIDS registered in the period from 1980 to July 2014, of which 50% were between 25 to 39 years old. The reduction of the risk of the vertical transmission of HIV by the adoption of the ACTG 076 protocol and the treatment of children exposed to HIV, as well as the fundamentals for the control of the epidemic, can provide security and increase the possibilities of the exercise of reproductive rights by women living with HIV. Several studies have investigated the impact of the diagnosis of infection by HIV in the reproductive decisions of women and in the rates of fertility and of sterilization. American women living with HIV/AIDS present lower rates of fertility and higher rates of sterilization than women without this diagnosis. Also, in a Brazilian study, the rates of sterilization were higher among women living with HIV/AIDS compared to the female population in general. As well as the impact of the diagnosis, the prophylaxis of the vertical transmission can also influence the reproductive decisions of women living with HIV/AIDS. Women of reproductive age who discover that they are seropositive for HIV face the difficult decision to have children or not. This decision is even more difficult in societies in which the appreciation of women is linked to maternity. Studies in different contexts demonstrate that the reproductive decisions of women are heavily influenced by the significance that is culturally given to maternity. The sexual and reproductive health of women living with HIV/AIDS comprises both the general aspect of being a woman and the particularities of the experience related to the virus. In fact, decisions related to having or not having children are a right of women living with HIV/AIDS, influenced by social and cultural aspects. This article has the objective of analyzing the factors associated with the occurrence of pregnancies after the diagnosis of infection by HIV in women residing in southern Brazil. # Materials and methods The analyzed data is derived from a cross-sectional study entitled “Sexual and reproductive health of women in the context the epidemic of HIV/AIDS in Porto Alegre”. Data collection took place from January to November 2011. The study population was women of a reproductive age (18 to 49 years old) living with HIV/AIDS and cared for by specialized services in Porto Alegre. The sample size was calculated seeking to estimate the reasons of prevalence of pregnancy after the diagnosis of HIV of 1.6 or greater. This sample was estimated for an expected prevalence of the outcome between 30 and 70% in the group without the interest factor. A power of 80% and a bilateral statistical significance of 5% were considered. The calculation was performed in the PROC POWER procedure of the SAS (Statistical Analysis System version 9.2) program. Applying the correction of the effect of the outlining by complex sampling (deff) equal to 1.6, the final sample size was estimated in 615 women with HIV. All specialized outpatient services in HIV/AIDS in the city of Porto Alegre in 2011 were included in the study. A simple random sampling was carried out at each service for women from 18 to 49 years old, respecting the proportionality of care of the services. The selection was based on the consultation appointment schedules of all health professionals of the service. After reading and signing the informed voluntary consent form, the participants responded to a computer questionnaire prepared by the Sphinks Léxica version 4.0 program, in a private location, filled out by the interviewer on a netbook. The questionnaire investigated the following variables: (a) occurrence of pregnancies after the diagnosis of HIV, which is the outcome of interest for this study; and (b) socio-demographical and behavioral variables, including race-ethnicity, level of education, household income (measured by the minimum wage unit), age at diagnosis of infection by HIV, time of diagnosis of infection by HIV, use of condoms and age of first sexual intercourse, number of sexual partners and stable unions in life, practice of sex for money, use of illegal drugs, occurrence of children before and after the diagnosis of HIV, experience of violence related to the diagnosis of HIV/AIDS (either verbal or physical aggression related to HIV or occurrence of discrimination in health care services). To control the quality of the data, 10% of the questionnaires were randomly selected and repeated by telephone. The study was approved by the following ethics committees: Comitê de Ética em Pesquisa da Universidade do Rio Grande (UFRGS—approval number 2008216), Comitê de Ética em Pesquisa da Secretaria Municipal de Saúde de Porto Alegre, Comitê de Ética do Grupo Hospital Conceição, Comitê de Ética em Pesquisa do Hospital de Clínicas de Porto Alegre and the Comitê de Ética em Pesquisa da Secretaria Estadual de Saúde do Rio Grande do Sul. The management of the database and the statistical analyses were performed using SPSS Software version 18.0. The characteristics of the sample are presented by descriptive statistics. The test of homogeneity of proportions, based on the Pearson chi-squared statistic, was used to compare proportions of exploratory variables between defined groups by the outcome of interest (women with and without pregnancies after the diagnosis of HIV), with a defined level of significance of 5%. Poisson regression models with a strong estimator were used to estimate reasons of prevalence (RP) for each category of the exploratory variables, having as a reference the category of least expected risk. The modelling technique estimated the reasons of raw prevalence for each exploratory variable, using the Wald statistic, with a level of significance of 5%. Subsequently, adjusted reasons of prevalence were calculated in a regression model that simultaneously included all the variables that presented a value p ≤ 0.20. This model was used to examine the independent effect of each exploratory variable regarding the outcome, accordingly identifying the main predictors of the outcome, those whose value p was ≤ 0.05 using the Wald statistic. The time of diagnosis variable was included in the modelling as a possible confounding factor because it could influence both the exposition variables and the outcome of the study. # Results 756 women from 18 to 49 years old living with HIV and cared for in the specialized outpatient services were selected as participants of the study. Of these women, 75 women refused to participate in this study. Therefore, the final sample was composed of 681 women; 90.1% of the 756 women who met the criteria for inclusion. presents the socioeconomic characteristics and experiences of women living with HIV/AIDS, in accordance with the occurrence of pregnancies after the diagnosis of HIV/AIDS. From the total of interviewed women, 35.2% reported pregnancies after the diagnosis of infection by HIV. The total sample consisted predominantly of white women (59.2%), with a level of education below high school (66.5%), a household income of less than two minimum salaries (53.2%) and at least two stable unions in life (58.2%). The use of illegal drugs in their lives was reported by 30% of the sample and 37.5% of them reported experience of violence related to the diagnosis of HIV/AIDS. The diagnosis of HIV/AIDS occurred at the age of up to 25 in 39.5% of the women (the mean age at diagnosis was 28,49 ± 7,57 years). presents the sexual and reproductive behavior of women, demonstrating that 47.1% of the sample had their first sexual initiation at up to 15 years of age (the mean age at the first sexual intercourse was 16,10 ± 2,73 years) and that 74.5% of the interviewees did not use condoms in their first sexual intercourse. The median of the number of sexual partners throughout life was 5, with partners ranging from 1 to 1300, and 33.3% of women had 7 or more partners. The practice of sex in exchange for money was reported by 11.2% of women and 69.7% of the sample already had a child before the diagnosis. The comparison between the groups of women with and without pregnancies after the diagnosis demonstrated that the groups differ in relation to age at diagnosis of HIV/AIDS (p\<0.001), level of education (p = 0.002), household income (p = 0.040), use of illegal drugs in their lives (p = 0.001), experience of violence related to the diagnosis of HIV/AIDS (p\<0.001), age of first sexual intercourse (p = 0.001), use of condoms in the first sexual intercourse (p = 0.025), and if already had previous children (p = 0.036) (Tables). The analysis of the group of women with pregnancies after the diagnosis of HIV/AIDS shows that 60.3% of them had the diagnosis of HIV/AIDS at up to 25 years of age (the mean age at the diagnosis was 24,54 ± 6,14 years), 74.2% had a level of education below high school and 64.6% had two or more stable unions in their lives. In this group, 38.1% had already used illegal drugs and 51.3% reported to have suffered violence related to the diagnosis of HIV/AIDS. With respect to sexuality and reproductive life, 55.6% had their first sexual intercourse at up to 15 years of age (the mean age at first sex was 15,56 ± 2,39 years), 79.7% had not used condoms in this relationship and 64.4% already had a child before the diagnosis of HIV. Raw Poisson regression models demonstrated a higher occurrence of pregnancies after the diagnosis in women with a lower level of education (RP = 1.44; IC95%: 1.13–1.84); lower income (RP = 1.25 IC95%: 1.01–1.54); with the occurrence of the first sexual intercourse before 15 years of age (RP = 1.40; IC95%: 1.15–1.73); who did not use condoms in their first sexual intercourse (RP = 1.35; IC95%: 1.03–1.76); who used illegal drugs in their lives (RP = 1.43; IC95%: 1.17–1.76); with experience of violence related to the diagnosis of HIV (RP = 1.75; IC95%: 1.44–2.15); and women without previous children (RP = 1.27; IC95%: 1.03–1.56). The lower the age at diagnosis, the higher the occurrence of pregnancies (test of significant linear trend, p \<0.001). The adjusted Poisson regression model evidenced that the following variables of exposition maintained an independent effect regarding the occurrence of pregnancies after the diagnosis of HIV: level of education below high school (RP adjusted = 1.31; IC95%: 1.03–1.66), non-use of condoms in the first sexual intercourse (RP = 1.32; IC95%: 1.02–1.70), age at diagnosis of HIV, highlighting women with an age equal to or less than 20 years (RP = 3.48; IC95%: 2.02–6.01); and experience of violence related to the diagnosis of HIV (RP = 1.28; IC95%: 1.06–1.56). In this analysis, the effect of the following variables regarding the outcome was not significant: race/ethnicity, household income, number of stable unions, age in the first sexual intercourse, use of illegal drugs, number of sexual partners in their lives, and the existence of previous children. In the group of women who got pregnant after diagnosis, the first pregnancy was not planned for 65,4%, and 47,2% reported willingness to undergo a tubal ligation surgery (data not presented in any table). # Discussion This study demonstrated that one third of women living with HIV∕AIDS cared for in specialized services in the city of Porto Alegre had at least one pregnancy after the diagnosis. Becoming pregnant after the diagnosis can indicate the exercise of the reproductive right of maternity, especially in a scenario in which the anti-retroviral therapy is universally offered, but can also represent a type of consequence of the situation of vulnerability of women living with HIV/AIDS. The data analyzed here indicates that the pregnancy after diagnosis, for most of the interviewees, indicates the context of vulnerability in which these women find themselves. In this study, post-diagnosis pregnancy is associated with a lower level of education and non-use of condoms in the first sexual intercourse. Furthermore, the lower the age at diagnosis, the more prevalent the outcome. The experience of HIV-related violence was also associated with the occurrence of post-diagnosis pregnancy. Studies demonstrate that the use of condoms in the population in general is higher between young people particularly regarding their first sexual intercourse. Although in the raw analysis the occurrence of pregnancies appears to be associated with the start of the sex life before 15 years old, the adjusted analysis demonstrates that the preponderant factor is the non-use of condoms in the first sexual intercourse. In the studied sample, less than 30% of the women used condoms in their first sexual intercourse, when compared in relation to the occurrence of pregnancies after the diagnosis, the use of condoms fell to 20%. This relates to the gender inequalities which are still significantly present in Latin American countries and which, in addition to the low education and income levels found in our study, suggest that the women find it hard to negotiate the use of condoms. Despite this finding, there are studies that report that the use of condoms tends to increase after the HIV diagnosis, but condoms are still infrequently or irregularly used. The age at diagnosis of infection by HIV appears to influence the occurrence of pregnancies. Our results indicate that the lower the age at diagnosis, the higher the occurrence of pregnancies, as already observed in another study. Studies indicate that age is a factor that interferes in the will to conceive. Despite several studies reporting that many women living with HIV/AIDS wish to become mothers our data suggests that the pregnancy after the diagnosis is a result of factors related to the context of vulnerability in which women with HIV/AIDS live. Although in the multi-varied model the association was not maintained between previous children and the occurrence of pregnancy after the diagnosis of HIV, it is possible that the existence of previous children represents vulnerability in the reproductive trajectory and not the fulfillment of a reproductive right. This hypothesis is sustained by the study of Pilleco which indicates a higher prevalence of induced abortion between women living with HIV/AIDS. Added to the already described situations of vulnerability is the association found between the occurrence of pregnancies after the diagnosis and violence related to the diagnosis. Several studies have demonstrated that women living with HIV report a higher prevalence of violence during their lives. Therefore, the occurrence of pregnancies after the diagnosis of infection by HIV does not indicate, necessarily, the exercise of the reproductive rights by women living with HIV/AIDS, because, as demonstrated, these pregnancies occurred in contexts of vulnerability (low education and income level, non-use of condoms at the first sexual intercourse, significant percentage of women willing to undergo a tubal ligation surgery, and high percentage of unintended pregnancies). The pregnancy can, in this context, express the difficulties of these women in negotiating with partners about the use of contraceptive methods and the prevention of STD/AIDS and the difficulty in the decisions regarding sexual and reproductive health. Pregnancy can also be a strategy of social insertion—since maternity performs an important role in the construction of the feminine identity, particularly in Latin American and African societies, and is a way of dealing with the still significantly disseminated prejudice caused by AIDS. Taking into account the presented scenario, the interventions of prevention in women’s health should occur earlier, before the start of their sex lives, in order to guarantee the appropriate information and inputs for the prevention of sexually transmitted diseases such as AIDS and unplanned pregnancies. It is fundamental that the health services recognize the contexts of vulnerability of women living with HIV/AIDS, as demonstrated in this study, in order to promote educational actions of reproductive planning that favor the autonomy of women regarding their sexual and reproductive decisions. # Supporting information We would like to thank all the women surveyed, the health institutions involved in the research that originated the data for this paper and the students who participated as interviewers. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** LBT FBP AV DRK. **Formal analysis:** LBT FBP MLD JCCL AV DRK. **Funding acquisition:** DRK. **Investigation:** LBT FBP AV DRK. **Methodology:** LBT FBP AV DRK. **Project administration:** LBT. **Resources:** LBT FBP MLD JCCL AV DRK. **Supervision:** DRK. **Validation:** LBT FBP AV DRK. **Visualization:** LBT FBP AV DRK. **Writing – original draft:** LBT FBP MLD JCCL AV DRK. **Writing – review & editing:** LBT FBP MLD JCCL AV DRK. [^3]: Current address: Department of Preventive and Social Medicine, Federal University of Minas Gerais (Universidade Federal de Minas Gerais), Belo Horizonte, Minas Gerais, Brazil

# 1. Introduction The volume and value of transactions in the international art market have been increasing over the years. General interest in artworks has progressively increased in recognition of their financial and cultural value. Simultaneously, art forgery scandals have occurred for monetary gain. Consequently, the reliability of the global art market has diminished, and economic turbulence may occur in global art sales. Counterfeit technologies have grown rapidly, and there are cases in which authentication tests cannot be judged to be reliable, even when using various technologies. To sustain the artistic activities of artists and transactions in the art market, a specialized system needs to be established to distinguish between authentic and forged artworks. Two methods can be used to discriminate counterfeit works: a judging process with expert insight through observations with the naked eye and scientific assessments using analytical techniques. In the first method, the appearance of artworks is visually inspected for pigments, paint aging, canvas aging, and cracking generated on the surface of artworks. In the second method, scientific techniques, such as bright-field microscopy, scanning electron microscopy, transmission electron microscopy, infrared imaging, and ultraviolet imaging, are used to examine the artworks. Recent studies have shown that cracks on painted surfaces contain information on environmental factors, the proportions of pigments, humidity, and the unique characteristics of the artist’s painting technique, making it possible to judge the authenticity of artworks. The physical and visual characteristics of cracks on the painted surface have been investigated to analyze cracks. When studying the physical characteristics of cracks, some pigment layers of the artworks need to be extracted occasionally, and the internal and external structures of paintings cannot be examined physically without irrevocable damage to the original piece. However, important information can be obtained through visual analysis, such as the characteristics of the painting and the types of cracks. The overall shape of the crack in the top view can be detected with the naked eye and can be enlarged significantly using a microscope that provides high- magnification images. Natural cracks were successfully distinguished from forged cracks using an ordinary microscope. The general optical microscope can quickly analyze cracks distributed over the entire painted surface, but only the top surface of the cracks can be observed. However, cracks on painted surface are generated with three-dimensional geometrical characteristics by applying the various factors, such as materials, techniques, and the environmental variables. Combining emerging vertical analysis with conventional horizontal analysis can improve the performance and accuracy of the artwork authentication process. In this study, we focus on the cross-sectional analysis of cracks on a painted surface using an optical coherence tomography (OCT) system for an accurate estimation of the crack dimensions (width, depth, and morphology). OCT laser technology has been utilized to investigate the pigment layers of artworks without inflicting damage to them and has proven to be highly effective for artwork analysis. We investigated differences between natural and artificial cracks forged by counterfeiters to provide a basis for authenticity through OCT analysis. The suggested OCT analysis method can be used as an indirect basis and cannot be an absolute answer for determining the authenticity of artworks because the shape of the crack may change owing to various reasons, such as paint type, characteristics, and surface treatment of the emulsified work. In conclusion, this study is expected to advance understanding characteristics of the crack surface and contribute significantly to the determination of art authenticity. # 2. Materials and methods ## 2.1 Experimental process of OCT analysis A schematic depicting the authenticity assessment process for artworks based on OCT technology is shown in. Three artworks were selected for the OCT analysis according to the time of production, the artist’s expertise, and the characteristics of the cracks. Natural cracks occur owing to various causes such as physical and chemical reactions at the painted surface for each work, and these cracks were analyzed using OCT technology ****. An art restoration expert, who created two of the artworks, produced forgery works for each of his two artworks and the one of other artworks with permission from the original artist ****. Before analyzing the paintings using OCT technology, several cracks that represent the characteristics of the artworks were selected from the cracks in each original work. In the counterfeit works, artificial cracks made at locations similar to those in the original work were selected. OCT scanning was performed by selecting a measurement area of 2 mm × 2 mm × 1.7 mm at the crack position on the artwork surface ****. Subsequently, 3D modeling data containing information on the depth, width, cross section, and shape of the cracks were obtained, and a cross-sectional image of the middle plane of the cracks was extracted as representative for the quantitative analysis of crack morphologies. In the cross-sectional image, various parameters, such as width, depth, and shape, were measured according to the morphological diversity and complexity of the cracks. In particular, the width and depth were subdivided into three categories to analyze the crack characteristics in as much detail as possible ****. Based on the data obtained in this manner, the characteristics of the cracks were classified and analyzed according to the cause of the crack, and the foundation for the authenticity of the artwork was established. The width and depth of crack may vary depending on the cross-sectional image extracted from the three-dimensional image of crack Therefore, the quantified values presented in this study can be used as supplementary data to explain the characteristics of the cracks. ## 2.2 Experimental set-up of the OCT system The OCT system was constructed using a Michelson interferometer with a beam splitter. In this setting, a spectral domain OCT system was selected for high- speed data acquisition **(Fig S1)**. The developed OCT system generates a circularly polarized laser operating at a wavelength of 930 nm. The penetration depth of the laser on the painted surface of the artworks could be reduced compared with the longer wavelength of the laser source; however, the OCT system using a laser with a short wavelength had a higher imaging performance with a better resolution for accurate crack detection. In this study, the experimental OCT system could produce an excellent longitudinal resolution of 8 μm and a lateral resolution of 7 μm in air and a measuring depth range of 1.7 mm from the surface. After laser emissions were transmitted through the beam splitter, they were reflected from the two scanning mirrors and injected onto the surface of the artwork with an objective lens. Interference signals were produced by combining the reflected optical signals from the artwork surface with the reference signals. 3D reconstructed images were generated with accurate dimensional and morphological characteristics of the surface cracks. ## 2.3 Preparation of original and forged artworks The three artworks analyzed in this study were oil paintings by Korean artists who agreed to the contents of this study ****. All the paintings used in this study were stored without any light or dust at a temperature of 25°C and a humidity of 40% to prevent the deterioration of their surfaces. In addition, these artworks were stored in a chemical-free space and were fixed to an easel to avoid contact with the floor and surrounding objects. As cracks are likely to occur in old works, art works that were at least 30 years old were selected. Two of these artworks were created by Eunyeong Ko, who specializes in oil paintings and art restoration ****. The third painting was “*figure*” (45 cm × 38 cm), oil painting created by Joonja Jun in 1964 ****. Joonja Jun is an acclaimed contemporary Korean artist who was selected as a recommended artist for the National Exhibition and won three special awards and 13 awards at the Korean National Art Exhibition held from 1949 to 1981. A major theme of her work is humans and festivals. By making full use of intense gesture drawing and action painting, the artist pursues a fundamental inquiry into painting. With permission from the original artist, a restoration expert produced counterfeit works of the original works using the appropriate oil paint and medium after observing the style, color, and cracks of the original work through the naked eye and a magnifying lens ****. The art restoration expert could accurately understand the color combination of the original work, the thickness of the paint layer, the texture of the surface, and the painting style because these features are essential for restoring damaged areas in artworks. The art restoration specialist used appropriate oil paints and media to replicate the colors and cracks as accurately as possible. The methods for the creation of artificial cracks on the surface of artworks can be broadly classified into two types: ⅰ) physical processes such as cutting with knives and the bending deformation of painted layers, and ⅱ) chemical processes using a crackling medium and crackle paste. As dry cracks tend to be wider than aging cracks, cracks were created using sharp knife tools with the advantage of adjusting the location, network pattern, width, and length of cracks as much as possible to resemble the original cracks. The cracks were artificially created by drawing them on the surface of the artwork with physical force using various sharp knives, and the thickness of the cracks was adjusted by adjusting the force. The main cracks that revealed the characteristics of the paint layer in the original works were similarly produced in counterfeit works. Although the result was not exactly the same as the original crack, the thickness of the painted layers and the composition of the medium were manipulated to produce counterfeit cracks similar to naturally occurring cracks. In this study, 34 works were rented, and only three of them were selected as representative works. All these works were painted in oil paint on canvas frames. Some works were created for more than 10 years ago, but most of them were created within the past five years. Cracks occurred in eight artworks and the representative causes of cracks are drying or aging. The OCT cross-sectional images of cracks in five artworks in which cracks occurred were obtained and the geometrical parameters of the cracks were quantified **(Figs S2-S11)**. These results were referred to analyzing the cracks of three representative works in this main text. # 3. Results ## 3.1 Appearance and location of cracks selected in *Artwork 1* and *Artwork 2* to evaluate cracks induced by drying of the paint layers using the OCT system The target regions for OCT analysis were selected at three locations where cracks were generated in each work. The three selected points represented crack locations that were expected to provide particularly meaningful information for art authentication based on the OCT analysis of cracks. ### Artwork 1 shows the varnish-processed oil painting with light brown and darker brown upper layers on top of a light green undercoat layer applied to a canvas. Cracks occurred in both the light brown and darker brown upper layers, except for the under layer of paint. The cracks occurred because the proportion of oil painting media used in the upper paint layers (light brown/dark brown) was not adequately controlled. Forged cracks were produced in both the light and dark brown paint layers to match the locations at which the original cracks were generated. The cracks were produced by drawing thin lines similar to cracks produced using sharp knives on the paint layers with detailed control so that the color on the bottom side of the paint layer was visible ****. ### Artwork 2 In, the varnish-processed artwork has multiple layers composed of a black underpainting layer and upper layers with a mixture of white, light blue, and green colors. Cracks occurred on the upper layers because the upper painted layers contained a relatively small amount of medium. The under layer of the artwork was painted black, and the upper layer was composed of a mixture of white, light blue, and light green paints. Unlike *Artwork 1*, cracks were generated because there was only a small amount paint medium in both the upper and under layers. Some cracks were relatively thin; therefore, the color of the underlayer was not visible. Although the location of the cracks in the forged work was not similar to that of the original cracks, the forged cracks were generated so that it was difficult to distinguish them from original cracks with the naked eye ****. In addition, the thickness of the forged cracks was controlled to be as similar as possible to that of the original cracks. ## 3.2 Quantitative analysis of the original cracks and forged cracks to explain drying cracks of the paint layers on *Artwork 1* and *Artwork 2* The cracks created in the original and counterfeit artworks were measured using the OCT system ****. The cross-sectional images of the cracks were compared to analyze the differences in the morphologies of the original and forged cracks, such as the depth and shape ****. These images contain adequate information because they can identify the 3D visual features of the crack that cannot be observed by the naked eye or a microscope. The widths (top, center and bottom widths) and depths (left, center, and right depths) of each crack were quantified using cross-sectional image analysis. shows the morphology of the cracks formed in the forged and original artworks. The quantitatively measured values of the scale parameters of each crack were classified into various types. These representative values were selected from the largest values of the crack widths and depths and were used to compare the original and counterfeit artworks ****. ### Artwork 1 In, cracks caused by the combination of media occur for each color, but the shape of the cracks for each color was not significantly different. The most prominent feature in the OCT cross-sectional image of the original cracks in *Artwork 1* is a rectangular shape in which the side interface of the crack is shaped like a cliff and the bottom surface of the cracks is flat. This indicates that, when the media in the upper painted layer dried, several gaps were left in the paint, resulting in a complete loss of internal adhesion. As the upper layer dried, the internal bonding decreased, and hence, the side of the crack was clean. However, the surfaces of the forged cracks tended to be uneven in terms of surface roughness. The top sides appeared similar, but there was an evident difference in the cross-sectional OCT images. The cracks of the original artworks had a rectangular shape; therefore, each parameter of the crack width showed similar values. On the other hand, the cracks in the forged artworks had inverted triangular shapes, the measured values were different for each parameter type, and the bottom parameter values were not measured. The depth of the cracks in the counterfeit and original artworks differed by (Ⅰ) 93.9 μm, (Ⅱ) 97.9 μm, and (Ⅲ) 73.3 μm ****. It appears that the depth of the crack increased as a force was applied to the painting layer with a knife to create a width similar to that of the original cracks. ### Artwork 2 The under layer of paint had a larger amount of medium than the upper layer. As the medium solutions dried, small empty pores were created inside the under painted layer. Consequently, the adhesion inside the under painted layer decreased and this phenomenon appears to have affected the occurrence of cracks in the upper layer of the paint. The main characteristic of the cracks is that the width of the cracks is very small; thus, the color of the under layer is not visible, and the aspect ratio (depth/width) of the cracks is large. The original cracks had thin and deep rectangular shapes, whereas the forged cracks had triangular shapes with rough surfaces. A concentrated force was applied with knives to create forged cracks with a width similar to that of the original cracks. Consequently, the forged cracks appeared to be relatively shallow. A medium combination problem occurs in the under layer of paint. The depth was observed to be approximately 2.5 times larger than the width of the crack ****. In the case of thin and deep cracks, the cross-sectional shape of the crack was either thin rectangular or inverted triangular. The weakened bonding force in the under layer of paint affected the upper layer, and it was judged that the crack at the upper part had small width. ## 3.3 Location and enlarged images for a comparison of original and forged cracks in *Artwork 3* to explain cracks induced by drying, canvas aging, and direct physical impacts of the paint layers We borrowed oil paintings on canvas created by Joonja Jun to analyze the original cracks generated by other factors and the forged cracks. The five cracks were classified into five types according to the characteristics of the original artworks and the cause of crack occurrence. ### Artwork 3 The artist attempted to overlay the surface of the artwork with light purple paint to cover naturally occurring cracks ****. This work was selected for art authentication based on OCT analysis because of the occurrence of various types of cracks in the paint layers with varied thicknesses. As the new canvas has good elasticity and does not crack easily, a canvas that was produced approximately 25 years ago was used as the substrate for the forged artwork. In addition, the thickness of the paint layer on the forged artwork was similarly formed by observing the original product with the naked eye ****. ### Craquelure Ⅰ These cracked areas occurred owing to the aging and shrinkage of the canvas at the paint layers that were thinner than the others ****. The speculative reason is that wrinkles appear in the same direction on the back of the canvas as the cracks advanced on the painted surface. These cracks occurred at locations painted with various pigments, regardless of the paint layer colors. To imitate the visual morphology of the original crack, thin mass tools were used to perform finely tuned slashing to create long paths of dynamic cracks in the painting layer of the forged work ****. Consequently, forged cracks with thicknesses and directions similar to those of the original cracks were created. ### Craquelure Ⅱ Similar to the reason for the occurrence of *Craquelure I*, *Craquelure Ⅱ* also occurred in the aged area of the canvas ****. Cracks in the paint layers commonly occur in the direction of the thick wrinkles on the back of the canvas, but the difference is that the paint layer at this location *(Craquelure Ⅱ)* is relatively thicker. Forged *Craquelure Ⅱ* was produced using the same slashing method used to create forged *Craquelure I* ****. ### Craquelure Ⅲ *Craquelure Ⅲ* may appear on painted surfaces during the drying process owing to chemical environmental changes in the layers of paint ****. Cracks were generated across painted layers with a mixture of three colors that were applied with a thickness greater than that of the surrounding layers of paint. In the middle of the paint layer, the painting mediums were poorly mixed, which led to drying cracks occurring in large areas of the painted surface. Forged cracks were also produced in a pale-yellow color of the undercoated layer through a deliberate process in which the force was appropriately adjusted with a thin knife ****. ### Craquelure Ⅳ *Craquelure Ⅳ* occurred owing to unintentional physical impacts on the completed painting by the artist and appeared to be scratched by an ambiguous tool ****. The ivory color of the under layer of paint was visible in the exposed area. During the drying process, the cracks widened with the formation of several branches. When these cracks were forged, they were created by applying deliberate force to the painted surface using a stainless-steel tool. The width of the forged crack was suitably widened, and the under surface layer was clearly visible, similar to the original cracks ****. ### Craquelure Ⅴ Similar to the reason for the occurrence of *Craquelure Ⅲ*, *Craquelure Ⅴ* occurred in the thickest areas of the artwork ****. Cracks were observed across both the upper layer (white) and the sublayer (blue). To replicate the original cracks, a thicker knife was used to apply a relatively stronger force to the painted surface layer ****. ## 3.4 Quantitative analysis of the original and counterfeit cracks to explain the cracks caused by drying, canvas aging, and direct physical impacts of the paint layers on *Artwork 3* ### Craquelure Ⅰ The cross-section of the original *Craquelure Ⅰ* had an inverted triangular shape in the upper painted layer ****. The shape became a sharp point toward the undercoated layers, and the height of both side layers along the crack remained approximately constant. The upper painted layers were affected by shrinkage that occurred owing to the aging of the canvas, and the cracking process may have advanced in the top layer of the paint. Thin thread-like structures were exposed on the inner sides of the cracks. This may have occurred because the ratio of the medium mixture to oil painting was different for each colored layer, which resulted in incomplete separation inside the painted layers. Similarly, the cross-sectional morphology of the forged cracks was a similar inverted triangle, but these cracks had a larger gap and a slightly shallower depth ****. In the forged cracks, one side of the painted layers had a protruded outer surface because of the physical influence of drawing of the crack lines. The natural cracks were generated by canvas shrinkage and were 70.2 μm in width and 370.2 μm in depth ****. In contrast, the forged cracks were approximately 95.7 μm in width and 142.5 μm in depth. This observation may indicate that forged cracks are wider and shallower than the original cracks. There is a statistically significant difference of approximately 227.7 μm in the crack depth between the original and forged cracks. Although the force was finely adjusted to create a small width, it was almost impossible to control the depth with precision. ### Craquelure Ⅱ The original cracks had an inverted triangle with a slight incline ****. The width of the cracks was narrower than that of the original *Craquelure Ⅰ*. The inner spacing of the cracks had an uneven structure without thread-like fibers. In the cross-sectional image of the forged cracks, one side of the painted layers had a slightly overhung outer painted surface ****. The forged cracks had a relatively shallow depth, and the sidewalls of the cracks were similar to those of the forged *Craquelure Ⅰ*. The forged *Craquelure Ⅱ* were also wider and shallower than the original cracks, as in the case of the forged *Craquelure Ⅰ*. When compared with than *Craquelure Ⅰ*, the crack widths were relatively smaller by 20 μm, but the depth was shallower by 134 μm ****. The original *Craquelure Ⅰ* and *Ⅱ* were created by canvas aging and *Craquelure II* occurred in thicker paint layers than did in *Craquelure Ⅰ*. However, it is difficult to forge cracks based on the tendency of the natural crack shapes produced according to the thickness of the paint layer. Consequently, the shapes and dimensions of the original *Craquelure Ⅰ* and *Ⅱ* were different, but the forged *Craquelure Ⅰ* and *Ⅱ* had similar shapes, widths, and depths. ### Craquelure Ⅲ The cross-section of the crack had an inverted triangle shape with a tip having a relatively blunt shape ****. The central axis of the inverted triangle was slightly inclined. The small width of the forged cracks was produced to reveal the color of the underlying paint layer, similar to the original cracks ****. Although the widths of the two cracks appeared similar, the main difference between these two cracks could be accurately distinguished according to the depth and shape from the OCT cross-sectional images. The aging conditions of the canvas directly contributed to the occurrence of *Craquelure Ⅰ* and *Ⅱ*, but *Craquelure Ⅲ* occurred as a direct result of the drying of the upper paint layers. The width of the crack was approximately 89.4 μm, which is relatively large. Conversely, the depth was relatively shallow at 180.9 μm ****. *Craquelure Ⅰ* and *Ⅱ* were produced by the influence of shrinkage in the canvas below the paint layer and were relatively thinner and deeper. In contrast, *Craquelure Ⅲ* was caused by the drying of the paint layer and was represented a reverse shape with a relatively large width and shallow depth. ### Craquelure Ⅳ Cracks with a large width are visible on the of the outer painted layer, and a sharply thin inverted triangle shape with a sharp tip is evident in the deeper undercoated layer ****. The forging process was conducted under condition in which the coating thickness was unknown. These forged cracks were produced to reveal the ivory-color of the under layer similar to the original cracks, but the difference in depth and shape are evident in the OCT cross-sectional image ****. The original cracks had a larger upper width in the shape of an inverted triangle, and the forged cracks showed a large bottom width. The original crack was approximately 191.5 μm in depth, which is greater than the depth of the forged crack of 68.0 μm ****. This may indicate that the original cracks were initially scratched, and after a considerable amount of time, the cracks deformed in width and depth under paint conditions such as paint drying. In addition, the width/depth ratio of the original and forged cracks were similar. ### Craquelure Ⅴ The sidewall of the cracks appeared to be relatively neat, as the painted layer appeared to have dried the most ****. The original *Craquelure Ⅴ* had a large width, but the color of the under layer was not visible. As the paint layer was thick, the crack was not sufficiently deep to reveal the color of the under layer. The forged *Craquelure Ⅴ* was extended and widened by applying a larger force than that used of producing the other forged cracks ****. The sidewall of the cracks appeared to be relatively neat, as the painted layer appears to have dried the most. Therefore, the upper layers were extruded from crack edges. In addition, the side of the forged crack was neatly formed in a straight line, but the detailed appearance with a zigzag shape from the paint layer of the original crack could not be forged. *Craquelure Ⅴ* was 123.4 μm in width and 219.1 μm in depth ****. Compared with *Craquelure Ⅲ*, which occurred in the thinner paint layer, *Craquelure Ⅴ* was 34.0 μm wider and was 38.2 μm deeper. This was observed in the contrast to the original *Craquelure Ⅰ* and *Ⅱ*, which were affected by the canvas underneath the painted layers. If natural cracks are directly influenced by the condition inside the upper painted layers, thicker painted layers generate cracks with a greater width and depth at the painted surface. In contrast, if cracks occur because of the influence of the canvas layer, the shape of cracks become thinner and shallower. The forged cracks were 68.1 μm in width and 104.3 μm in depth. Various cracks with similar widths can be artificially produced, but the depth is difficult to control. These cracks are not as deep as the original cracks ****. # 4. Discussion Cracks in works of art occur for three reasons: the drying of the paint layer, factors related to canvas aging, and direct physical impacts on the painted layer. First, cracks occurred in all three works owing to the drying of the painted layers. In *Artwork 1*, wide cracks occurred in the upper layers, which contained a larger amount of the medium than the under layer. However, thin cracks indirectly occurred in the upper layers of *Artwork 2* because the under layers had a larger amount of medium than the upper layers. As drying advanced, the medium inside the painted layers evaporated, leaving several empty spaces in the painted layers. The adhesion between the painted layers decreased, resulting in cracking. If this cracking occurs mainly in the upper layer, there is a tendency for the cracks to widen, as in *Artwork 1* ****. The entire thickness of the upper layer was split, and the depth of the crack corresponded to the thickness of the upper layers. Conversely, if cracking occurs mainly in the under layers, the reduced adhesion in the under layers affects the upper layers. Thin and deep rectangular cracks were observed in the upper layers of *Artwork 2* ****. In *Artwork 3*, the paint layer dried over a long period of time in addition to the aging of the canvas. Unlike *Artwork 1* and *2*, cracks occurred owing to factors related to both the upper and under layers. Unlike for the previous cracks, the cross-sectional shape of an inverted triangle was observed. When the paint layer was thicker, cracks were wider and deeper. The second factor was the aging of the canvas, which affected the paint layers and generated cracks. These cracks were observed only in *Artwork 3* and occurred in both the thin and thick areas of the artwork. When the paint layers were thick, the width of the cracks became thinner and deeper. The cause of cracking appeared in the support under the paint layers. A weakened adhesive force is applied in the under layer so that the upper layers have cracks, and the thinner the upper paint layers, the easier it is to produce cracks. Therefore, in this case, larger values were observed for crack width and depth. Third, cracks occurred owing to physical shock and were observed only in *Artwork 3*. A crack had been created in the original work owing to an unintended physical impact, but a long time had passed since then. The cracks in the forgery were also created by physical impact, but the difference from the original cracks was the length time elapsed. The much older crack was observed to be wider and deeper. The analysis of these two cracks reveals how deformation occurs over time in the crack generated by physical impact. Based on the above aspects, it is possible to verify whether a crack has been forged or whether the time elapsed since the occurrence of the crack is appropriate compared with the time of production of the work for authenticity assessment. According to the results of the analysis, the width and depth ratios of cracks in the same artwork differ according to the cause of the crack occurrence; therefore, it is necessary to measure several cracks and compare the width and depth ratios ****. # 5. Conclusion OCT technology has been proven to have the potential to analyze the characteristics of natural cracks and simultaneously identify differences from forged cracks. The differences between natural and forged cracks were analyzed by quantitatively measuring the parameters (width, depth, and cross-sectional shape) of the cracks. In this study, the characteristics of cracks in works that can be measured are summarized. Using artificial and natural methods, it was confirmed that the crack widths were similar. The results inferred that it is not sufficient to determine the authenticity of artwork with only one factor i.e., crack width and that the depth of cracks can provide a useful basis for determining the authenticity of an artwork. Natural cracks can become deep as the thickness of the upper paint layer in rectangle or inverted triangle shape, but artificial cracks are not formed in the same way as these natural cracks. The depth of artificial cracks is not as finely controlled as that of natural cracks. As there is a limit to controlling the depth and width of natural cracks, it is possible to analyze whether the cracks are forged. If the cross- sectional shape is considered in addition to the depth, the difference between natural and artificial cracks is revealed, which can further increase the possibility of the successful detection of authenticity. The cross-section of natural cracks can have various shapes, such as a rectangle, trapezoid, inverted triangle with a clean surface, inverted triangle with an internal microstructure, and inverted triangle with a large width only in the upper layer. However, most artificial cracks have a similar inverted triangle shape and two dimensions; wide/deep or narrow/shallow. The distinct difference between natural and artificial cracks according to the crack parameters makes OCT technology useful for appraising the authenticity of artworks. The OCT results presented in this study are limited in that only artificial cracks made by knives were compared with natural cracks. If cracks made by other factors, such as crackle paste, are evaluated using OCT, the 3D images and geometrical properties of the cracks will be obtained without technical issues; however, sufficient information for the discrimination of cracks in artworks can be unclear to obtain. Therefore, combining the foundation of this study with further studies to analyze other cracks is necessary to establish a more specialized OCT system to understand crack properties better for art authentication. Furthermore, by continuously measuring the surface of an artworks using OCT technology and observing the development of cracks and changes in the surface, various types of information to prevent forgery of the artwork can be collected. This study demonstrated that the difference between natural and artificial cracks in oil paintings can be analyzed using OCT technology, which has shown breakthrough potential for determining the authenticity of artworks. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Over the past decades, Turkey’s integration into the world markets has increased, resulting in the globalization of the world economy. The Turkish markets are easily affected by the policies of developed economies or any global external shocks such as oil prices. In this regard, the main objective of this research is to provide new empirical evidence by testing the impact of the external shocks namely: The U. S interest rate and oil prices on Turkey’s real estate market. Turkey has limited oil reserves, and the oil production in Turkey was reported at 314,000 Barrel/Day in 1980 and reached to 731,000 Barrel/Day in 2014. However, Turkey’s demand for crude oil from global markets has increased constantly in the last 30 years. In 2016, Turkey’s total liquid fuel consumption averaged about 861,000 Barrel/Day and more than 90% of total crude oil came from imports. 70% of Turkey’s imports of crude oil came from Iraq, Iran, and Russia. However, Turkey has faced several challenges in energy security, the first main challenge is the energy supply problem. Turkey’s main energy suppliers are Russia and Iran. Probable any economic or political disagreements with these countries put energy security in Turkey at risk. In this regard, Turkey should try to find new suppliers of oil sources to diversify the suppliers to reduce the dependency on the main suppliers. The second main challenge is the high dependency on imported oil, domestic oil production in Turkey is not enough to meet the country’s energy needs. Despite the limitation of oil production, the oil demand rapidly increasing. The rate of imported oil must be decreased by finding more renewable energy sources for the energy supply formula. At this point, Turkey should evaluate its alternatives to renewable energy sources such as solar, wind, and geothermal. However, these resources can simply be produced and renewed. Also, it diffuses fewer pollutants to nature, and it can never be depleted worldwide. Renewable energy resources in Turkey are hydroelectric, wind solar, geothermal, biomass, and waves. Turkey is the third country in the world with 1.28 million tons of oil equivalent (MTOE) in terms of producing geothermal energy worldwide, especially the Aegean territory has huge geothermal energy potential. Although the advantages of renewable energy resources. Non-renewable consumption of energy in Turkey has consistently elevated with the increase in its population. Nonrenewable energy consumption (kg of oil equivalent) has increased around 220% over the period from 1980 to 2014. Energy consumption was 715,149 kg of oil equivalent in 1980 and reached 1,577,828 kg of oil equivalent in 2014. However, the economy of this country is one of the leading producers in the world in textiles, ships, motor vehicles, consumer electronics, home appliances, construction materials, and other transportation equipment, thus led to increasing energy consumption namely oil consumption. According to the literature, the variations of oil prices have a significant impact on various economic variables. In this regard: have indicated that the effect of the oil prices on economic indexes in developed and developing countries can be different; these different findings can be attributed to various economic factors for instance: (oil-importing countries vs oil-exporting countries). Increases in oil prices can be economically bad for oil-importing countries but it is economically good news for oil-exporting countries. However, Turkey is a country that imports a large part of its oil needs from abroad. Hence an increase in oil prices negatively affects Turkey’s current account balance and economic growth and other economic variables. In this regard, if crude oil prices go up, then inflation in oil-importing countries like Turkey goes up, which leads to an increase in interest rates. Thus, any increase in interest rate can affect the cost of finance, which in turn leads to an increase in the housing prices. In this sense, the main objective of this research is to provide new empirical evidence by testing the impact of the crude oil prices on Turkey’s real estate market through the domestic interest rate channel. Besides, the study aims to test the impact of the U.S interest rate on Turkey’s real estate market through the oil prices and the domestic interest rate. The shocks from developed economies like the U.S. economy can affect Turkey’s economy through various transmission mechanisms, such as interest rate and exchange rate channels. The Turkish markets have the potential for the volatility of financial markets due to any change in the U.S. monetary policy, particularly the interest rates channel. Because any increase in the U.S. interest rate, investors will continue to withdraw their investment of emerging markets like Turkey. Therefore, any change in U.S. interest rates may affect the Turkish markets. This article suggests that the influence of the U.S. interest rates on Turkey’s real estate market depends on how this rate affect capital outflows, international trade, exchange rate, and any excess volatility of exchange rate, financial flows, and international trade can negatively affect the economic and financial stability, particularly on Turkey’s real estate market. However, this article aims to analyze the influence of international spillover effects of the U.S. interest rates and the influence of domestic interest rates on Turkey’s real estate market. In the last decades, Turkey has faced several critical reforms, such as ensuring the operational independence of the central bank to support the banking sector and the financial market and eliminating any restrictions on the capital inflow. As a result of these reforms, international trade as a share of Gross Domestic Product (GDP) has increased from 18% in 1981 to 61% in 2018 and the total exports have risen from 2.9 billion USD in 1981 to 166.5 billion USD in 2018. Furthermore, housing assets in Turkey have experienced spectacular growth over the last decades. The housing market represents the largest asset category for Turkish households, with a 70% housing ratio in 2018. However, real estate market developments stand out as one of the considerable economic concerns in Turkey, and the vitality in the housing market is considered as the most important indicator of macroeconomic performance. Numerically, the Turkish real market economy offers great investment potential with its value of 19.5% of the total GDP. Besides, the real estate and construction sector occupied 4.1 billion USD and 24.8% of the total FDI. According to the Knight Frank Global House Price Index, Turkey was recorded as the 55 in terms of annual price growth index. Also, urban renewal projects have started in different Turkish cities. It seems approximately 6.7 million residential units are expected to be demolished and rebuilt over the next two decades. Therefore, this situation increases the importance of researches on the variables affecting the real estate market. The purpose of this article is to provide empirical evidence of the effects of oil prices on Turkey’s real estate market. Furthermore, this study aims to test the impact of the U.S. interest rates on emerging markets like Turkey, particularly after the 2008–2009 global financial crisis (GFC). Many empirical studies have explored the importance of the real estate market in Turkey. Some of these studies have examined the linkage between the real estate market and macroeconomic activity. This relation has drawn special attention in the literature as housing investment has been considered as a substantial leading indicator of economic activity, especially after the 2008 GFC. The GFC in 2008 has increased concerns about economic stability in Turkey and puts the spotlight on the linkage between financial markets and macroeconomic variables (exchange rate, interest rate) and the spillover influence of the external factors (the U.S. interest rates). After the 2008–2009 crisis, the central bank of Turkey started to monitor financial markets development more closely and has started some macro-prudential measures into monetary policy channels to address and overcome these concerns. Therefore, the credit provided by the banking sectors to the markets as a percentage of GDP increased approximately 70% between 2010 and 2018, and the stock of housing credits as a percentage of GDP ratio increased over the period from 2010 to 2018 and reached above 10% by the end of 2018. The increases in housing credits are expected to continue in the future. In the literature, the impact of changes in interest rate channels on housing prices is defined as the housing price channel of monetary policy. In this regard, the monetary policy can affect the real estate market through various channels (direct and indirect mechanisms). In the direct mechanisms, the monetary policy channels can affect the real estate market through the user cost of housing, housing supply, and expectations of future housing prices movement. However, the user cost of housing effect is considered as the main direct impact of monetary policy on the real estate market. On this basis, when the interest rate increases, the average mortgage rate also increases. Thus, an increase in the average mortgage rate leads to an increase in the user cost of capital and an increase in the user cost will lead to a decrease in demand size for housing, leading to a fall in housing prices in the market. Furthermore, an increase in interest rate may have a significant effect on housing construction costs, causing a decline in the housing output. In indirect mechanisms, the monetary policy channels can affect the real estate market through standard credit-channel. On this basis, when the interest rate is increased, the individual housing wealth value falls as real housing prices decline due to a decrease in demand for houses. A decline in individuals’ wealth may lead to a decline in housing demand, leading to lower housing prices. Contrarily, an increase in interest rates will lead to an increase in mortgage repayments, leading to a decline in credit-constrained households’ cash flow and eventually to a decrease in housing prices. The structure of this research is designed as follows: section (2) of this article shows an overview of the literature review; sections (3 and 4) are data, methodology, and findings of the results; and section (5) is the main conclusion of this the article. ## Contributions to the current literature The research aims to provide new empirical evidence by testing the impact of the external shocks namely: oil prices and the U.S interest rate on Turkey’s real estate market by using three techniques of co-integration tests namely: the newly developed bootstrap autoregressive distributed lag (ARDL) testing approach as proposed by (McNown et al. 2018), the new approach involving the Bayer-Hanck (2013) combined co-integration test, Hatemi-J (2008) co-integration testing approach. The ARDL model is utilized to explore the relationship between the variables. However, our study provides three main contributions to the current literature. First, to the best of our knowledge, no empirical research tested the impact of U.S interest rate on Turkey’s real estate market. Second, the study provides robustness and comprehensive analysis by testing the external shocks namely: oil prices and the U.S interest rate on Turkey’s real estate market by using three techniques of co-integration tests. Third, the study uses a the newly developed bootstrap autoregressive distributed lag (ARDL) testing approach as proposed by (McNown et al. 2018) to test the relationship between the selected variables. ## Literature review In this article, we focus on two sections. First, the article aims to test the impact of oil prices on the real estate market. According to the literature, the variations of oil prices have a significant influence on various economic variables. showed that oil prices have a significant impact on GDP in 12 European countries. tested the impact of oil prices on economic activities in Thailand. The results found that oil prices have a significant effect on macroeconomic variables, such as investment, over the period from 1993 to 2006. examined the impact of oil prices on real income in Turkey. The results showed that any change in oil prices has a powerful impact on real income in Turkey, over the period from 1996–2017. indicated that oil prices have a powerful impact on the exchange rates in Saudi Arabia. Furthermore, several papers have tested the impact of oil prices on the equity market. In this regard, have tested the linkage between oil prices and the equity market in Australia, the study found that there is a significant and positive linkage between oil prices and the equity market in Australia. tested the influence of the oil prices on the equity market of the oil-importing countries. The findings showed that oil price shocks have a powerful impact on the equity market of oil-importing countries. used structural VAR and confirmed a positive linkage between oil prices and s the U.S stock market. While there are limited empirical papers that have tested the effect of oil prices on real estate markets, tested the effect of oil prices on real estate in Saudi Arabia. By employing Markov switching, the results fount that there is a significant influence of crude oil prices on the real estate market in Saudi Arabia during the period 2008 to 2015. has examined the effect of oil prices on the real estate market in Malaysia. Using the Toda-Yamamoto, the results suggested that oil prices are one of the leading factors responsible for the variation of the Malaysian real estate market. tested the effect of oil inflows on the real estate market in Iran. The findings found that there is a statistical and positive relation between oil inflows and the Iranian real estate market. Second, the study aims to test the influence of the U.S interest rates on Turkey’s real estate market. The linkage between the interest rates and the real estate market has attracted extensive attention in recent years. However, some empirical studies support a negative link between interest rates and the real estate market. In this regard, used data from 1965 to 2005 and tested the impact of the U.S monetary policy channels on the real estate market in the USA. Using the VAR model. The empirical findings indicated that the contractionary monetary policy (increasing the interest rate) harms the U.S. real estate markets. Similarly, used monthly data sets from 2000 to 2010 and investigated the relationships between interest rates and housing prices. The findings indicated a negative linkage between the interest rate channel and the housing market in the US. used data from 1974Q2 to 2008Q4 and tested the effect of monetary policy channels on the real estate market of Australia. Using the VAR model, the authors found that a contractionary monetary policy significantly reduced the housing activity. The results suggested that investment in the housing sector can be an alternative to investment in stock and bonds, thus leading to an inverse linkage between interest rates and the housing market. tested the effect of interest rate channels on the real estate market in China from 1998 to 2009. The authors showed that lower interest rates had an accelerative impact on house price growth, and suggested that monetary policy tools are the key driving forces behind the changes in house price growth in China. used a standard multivariate dynamic model and tested the relationship between the interest rate and house price in China from 1998 to 2010, and stated that interest rate negatively affects the real estate market in China. The authors suggested that the monetary policy tools in China are the key drivers behind the real estate market fluctuations in China. used the generalized variance decomposition approach and tested the linkage between interest rates and the real estate market in Turkey from 1961 to 2000. The findings shocks of the interest rates had noticeable effects on the real estate market and suggested that housing investment in Turkey is a leading indicator of economic activity. Contrarily, some empirical studies support a positive link between the interest rate and the housing market. On this basis, used a set data from 1987 to 2007 and investigated the effect of the U.S. monetary policy on housing prices in the USA and observed that monetary policy had a positive effect on house prices, and a strong short-lived effect on risk spreads in money and mortgage markets. tested the causality linkage between interest rate and housing price changes in Malaysia based on the Granger-causality test. The regression findings showed a direction causality relation between the interest rate and house price. has tested the relationship between the interest rate and house prices in Vietnam from 2009 to 2018. Using the ARDL approach to estimate the relation between the interest rate and the housing market, the author showed that the interest rate has a positive influence on the housing market in the short run. Whereas, some empirical studies showed that there is no link between the interest rate policy and the real estate market. For instance, tested the linkage between the interest rate and the real estate market in New Zealand from 1999 to 2009. Based on the two-stage least squares pool regression, the author showed that an increase in the interest rate policy rate may be ineffective in depression the real estate market. According to the literature, the influence of the U.S. interest rate on global markets has attracted extensive attention in last years. tested the spillover impact of the U.S. interest rate on financial markets of 12 countries in the Asia-Pacific. The findings showed a significant negative impact of the U.S. interest rate on the financial markets of 12 countries in the Asia-Pacific. indicated that the U.S. interest rate has a powerful effect on global emerging financial markets. Similar results were found by who indicated that the U.S. interest rate has a powerful influence on many emerging financial economies. Furthermore, this effect has a more powerful impact on markets with economies closely linked to the United States. Similarly, used the VAR model and indicated that there is a significant impact of international spillover influence of the U.S. interest rates on the advanced and emerging economies. tested the impact of U.S. monetary shocks on interest rates and exchange rates in 26 selected countries. Using the VAR model, the results suggested that countries with more stringent controls experienced smaller currency depreciation. tested the effects of the U.S. interest rates on local interest rates and the exchange rate channels in East Asian countries. Using the VAR estimation techniques, the authors found that the local interest rate channel responds robustly to the U.S. interest rate changes. Contrarily, showed that the U.S. interest rates have no impact on India’s financial markets. In testing the influence of the U.S. interest rates on the Turkish markets, utilized the ARDL testing model and suggested that the U.S. interest rates have a significant impact on Turkey’s financial markets from 2002 to 2017. These findings indicated that the U.S. interest rates negatively impact the Turkish stock market through debt and interest rate channels. indicated that Turkey’s financial market is significantly correlated with the U.S. financial markets. suggested that U.S. interest rates have a powerful impact on the Turkish banking sector. However, most empirical studies focused on the effect of the U.S. interest rates on the stock market. To the best of our knowledge, this article is the first to test the impact of spillover effects of U.S. interest rates on Turkey’s real estate market using the ARDL testing approach. shows the summary of literature review. ## Methodology ### Data and model specification A monthly dataset that spans from August 2009 to August 2018 was employed for this article. The data was retrieved from the organization for economic co- operation and development, and the Central Bank of the Republic of Turkey (TCMB). The main assumption of the research was that the oil prices and domestic interest rates, and the U.S. interest rates affect the real estate market in Turkey. Thus, the main equation of this article can be checked as follows: $$lnRE_{t} = \mspace{360mu}\beta_{0} + \mspace{360mu}\beta_{1}\mspace{360mu} lnOP_{t} + \mspace{360mu}\beta_{2}\mspace{360mu} lnDI_{t} + \mspace{360mu}\beta_{3}\mspace{360mu} lnUSI_{t} + \mspace{360mu}\varepsilon_{t}$$ *lnRE*_*t* *and lnDI*_*t* represent the logarithm of the real estate market and short-term interest rates in Turkey, *lnOP*_*t* is the Brent crude oil price, this is generally utilized in Turkey. the *lnUSI*_*t* represents the logarithm of the U.S. interest rates, ε_*t* is the error term. However, short-term interest rates have a strong impact on investment opportunities and capital inflow. ### Unit root and co-integration tests The Dickey and Fuller (1979) unit root test and the Clemente et al. (1998) (CMR) unit root test with (2) structural breaks date (SBD) were utilized to determine the stationary among the examined variables. To examine the effects of oil price, domestic interest rates, and effects of the U.S. interest rates on the Turkey real estate market, the study used autoregressive distributed lag (ARDL). The main advantage of this model is that the ARDL model is more appropriate for small data compared with other cointegration tests. Besides, using the ARDL test, the research aims to demonstrate if the variables are cointegrated in (3) options: at the level I(0), at the first difference I(1) or mixed of I(0), and I(1). The lag length was selected through the Akaike info criterion. In the ARDL model, (F) statistics will be compared to the Pesaran et al. (2001) critical values to capture the cointegration among the examined variables. On this basis, if the value of (F)statistics is higher than the upper bound I(1) the cointegration hypothesis will be accepted. In contrast, the cointegration hypothesis will be rejected if the value of (F)statistics is less than the lower bound I(0). Moreover, if the values of (F)statistics fall between at-level I(0), and the first difference I(1), these values mean that the findings will be indecisive. Recently, this approach upgraded by McNown et al. (2018), the recent version includes additional t-test *t*_*dependent* or F-test *F*_{*independent*} on the coefficients of lagged independent variables. The *H*₀ of *t*_*dependent* test is: σ1 = 0. The *H*₁ of *t*_*dependent* test is: *σ*₁ ≠ 0. While The *H*₀ of *F*_{*independent*} test is:*H*₀: *σ*₂ = *σ*₃ = *σ*₃ = *σ*₄ = *σ*₅ = 0. The *H*₁ of *F*_{*independent*} test is:*H*₁: *σ*₂ ≠ *σ*₃ ≠ *σ*₃ ≠ *σ*₄ ≠ *σ*₅ ≠ 0. The critical values (CV) in the bootstrap ARDL approach, are created based on the specific integration features of each time series data using the procedures of ARDL bootstrap, which in turn lead to eliminating unstable results of the ARDL bounds testing model. However, McNown et al. (2018) upgraded the bootstrap ARDL test by employing a table of CV gained by bootstrap simulation. These steps of the bootstrap test will lead to getting better results than the traditional ARDL bounds test. In particular, the Pesaran et al. (2001) CV allows for (1) variable to be endogenous, while the CV generated with a bootstrap technique allows for the endogeneity of all explanatory examined variables. Also, this approach is more suitable for data includes more than (1) explanatory variable. However, the equation of the ARDL cointegration technique is tested as given below: $$\begin{array}{l} {\Delta lnRE_{t} = \beta_{0}\mspace{360mu} + \mspace{360mu}\sum\limits_{i = 1}^{P}\beta_{1}\Delta lnRE_{t} + \sum\limits_{i = 1}^{q}\beta_{2}\Delta lnOP_{t - i} + \sum\limits_{i = 1}^{q}\beta_{3}\Delta lnDI_{t - i} +} \\ {\sum\limits_{i = 1}^{q}\beta_{4}\Delta lnUSI_{t - i} + \mspace{360mu} y_{1}lnRE_{t - 1} + y_{2}lnOP_{t - i} + y_{3}lnDI_{t - i} + y_{4}lnUSI_{t - i}\mspace{360mu} + e_{t}} \\ \end{array}$$ In, lnRE, lnOP, lnDI, and lnUSI are the natural logarithms of examined variables, p represents of number of lags (RE) variable, *q* represents of number of lags (DI and USI) variables; *e*_*t* is the error term, and Δ means the operator of the first difference level. The error correction features incorporating long and short-run information in the ARDL model is tested as given below: $$\begin{array}{l} {\Delta lnRE_{t} = \beta_{0} + \mspace{360mu}\sum_{i = 1}^{n}\beta_{1}\Delta lnRE_{t - i} + \sum_{i = 1}^{n}\beta_{2}\Delta lnOP_{t - i} + \sum_{i = 1}^{n}\beta_{3}\Delta lnDI_{t - i} +} \\ {\sum_{i = 1}^{n}\beta_{4}\Delta lnUSI_{t - i} + ECT_{t - 1\mspace{360mu}} + e_{t}} \\ \end{array}$$ The ECT is significant with a *negative*⁻ sign. However, ECT aims to determine the speed of adjustment from the short-term to the long-term levels. To enhance the findings of the ARDL testing result, the article applied the H-J (2008) co-integration technique test proposed by Hatemi-J (2008). The H-J (2008) allows two SBD and shows the new critical values tests of the co-integration; namely, ADFt, Z_*a*t, and Z_*t*t, and it is tested as given below: $$y_{t} = \alpha_{0} + \mspace{360mu}\alpha_{1\text{Dv}1} + \alpha_{1\text{Dv}2}\mspace{360mu} + \mspace{360mu}\beta_{0\mspace{360mu} xt} + \beta_{1}D_{1xt} + \beta_{2}D_{1xt} + \mspace{360mu} e_{t}$$ where Dv1 and Dv2 represent the dummy variables. In this test, the hypothesis of absence co- integration will not be accepted if the calculated values of the ADFt, *z*_*at*, *z*_*t**t* tests higher than the H-J (2008) critical values. Furthermore, this article uses the new techniques of combined co-integration tests Bayer and Hanck(2013) to boost the findings of the ARDL test. This test combines four various cointegration technique tests; namely, *EG*1987, *JOH*1988, *BO*1994, and *BA*1998*t* as proposed by respectively. Besides, this test includes the Fishe*r*^*F*statistics (FFS) to provide more conclusive results. The BH (2013) test includes functional estimations through disregarding the feature of multiple testing procedures, and it is tested as given below: $$EG1987t - JOH1988t = - 2\left\lbrack {IN\left( P_{EG1987t} \right) + \left( P_{JOH1988t} \right)\mspace{360mu}} \right\rbrack$$ $$EG1987t - JOH1988t - BO1994t - BA1998t\mspace{360mu} = - 2\left\lbrack {IN\left( P_{EG1987t} \right) + \left( P_{JO1988t} \right) + \mspace{360mu}\left( P_{BO1994t} \right) + \left( P_{BA1998t} \right)} \right\rbrack$$ where p is the values of (*EG*1987*t* − *JOH*1988*t* − *BO*1994*t* − *BA*1998*t*)cointegrations tests. To estimate the long-run cointegration, the FFS will be compared with the BH (2013) critical values. The hypothesis of absence long-run combined co- integration will not be accepted if the FFS values exceed the BH (2013) critical values. Moreover, this article used the followıng diagnostics tests (JB normality, the heteroscedasticity, Breusch–Godfrey serial correlation LM, Ramsey). JB normality test was applied to check the normal distribution of the model. The heteroscedasticity and the Breusch–Godfrey serial correlation LM tests were utilized to check the serial-correlation. Besides, the article used the Ramsey, CUSUM, and the CUSUMSQ tests to check the stability of the model. Furthermore, based on the VECM, the study uses the Granger causality (GC) approach to demonstrate the direction of the causality among *lnRE*, *lnOP*, *lnDI*, *and lnUSI*. The GC approach includes (ECT) to measure the short-run deviations of the time-series data from the long-run equilibrium path. However, the equation of ECM is tested as given below (eqs –): $$\Delta lnRE_{t} = \beta_{0}\mspace{360mu} + \mspace{360mu}\sum\limits_{i. = 1}^{p}\beta_{1}\Delta lnRE_{t - 1} + \sum\limits_{i. = 1}^{q}\beta_{2}\Delta lnOP_{t - 1} + \sum\limits_{i. = 1}^{q}\beta_{3}\Delta lnDI_{t - 1} + \sum\limits_{i. = 1}^{q}\beta_{4}\Delta lnUSI_{t - 1} + \partial_{1}\mspace{360mu} ECT_{t - 1} + e_{t}$$ $$\Delta lnOP_{t} = \beta_{0}\mspace{360mu} + \mspace{360mu}\sum\limits_{i. = 1}^{p}\beta_{1}\Delta lnOP_{t - 1} + \sum\limits_{i. = 1}^{q}\beta_{2}\Delta lnRE_{t - 1} + \sum\limits_{i. = 1}^{q}\beta_{3}\Delta lnDI_{t - 1} + \sum\limits_{i. = 1}^{q}\beta_{4}\Delta lnUSI_{t - 1} + \partial_{1}\mspace{360mu} ECT_{t - 1} + e_{t}$$ $$\Delta lnDI_{t} = \beta_{0}\mspace{360mu} + \mspace{360mu}\sum\limits_{i. = 1}^{p}\beta_{1}\Delta lnDI_{t - 1} + \sum\limits_{i. = 1}^{q}\beta_{2}\Delta lnRE_{t - 1} + \sum\limits_{i. = 1}^{q}\beta_{3}\Delta lnOP_{t - 1} + \sum\limits_{i. = 1}^{q}\beta_{4}\Delta lnUSI_{t - 1} + \partial_{1}\mspace{360mu} ECT_{t - 1} + e_{t}$$ $$\Delta lnUSI_{t} = \beta_{0}\mspace{360mu} + \mspace{360mu}\sum\limits_{i. = 1}^{p}\beta_{1}\Delta lnUSI_{t - 1} + \sum\limits_{i. = 1}^{q}\beta_{2}\Delta lnRE_{t - 1} + \sum\limits_{i. = 1}^{q}\beta_{3}\Delta lnOP_{t - 1} + \sum\limits_{i. = 1}^{q}\beta_{4}\Delta lnDI_{t - 1} + \partial_{1}\mspace{360mu} ECT_{t - 1} + e_{t}$$ To test the causality relation in the short run: the Wald-testing technique (F.statistics) is used to capture the significance of linked estimated coefficient using the Δ stationary variables. To test the causality relation in the long run: the t-test of the lagged ECT is employed. ### Empirical findings Tables and represent the outcomes of the unit root test (ADF and CMR), the outcomes show that the variables of this article (RE, OP, DI, and USI) are integrated at the first level I(1). The findings from show the results of the bootstrap ARDL model, the results show that the hypothesis (no cointegration) is rejected, meaning that the cointegration exists between oil price, domestic interest rate, the U.S. interest rate, and Turkey’s real estate market. The result of the HJ (2008) cointegration test includes two SBD as shown in. The result shows that the estimated statistics exceeds the 5% critical value. Therefore, the results provided evidence to reject the null hypothesis (no cointegration) at a 5% significance level. The findings of the BH (2013) test are presented in. The outcomes indicate that the value of the computed (F)statistics exceeds the calculated (F)statistics in both EG1987T-JO1988T and EG1987T-JO1988T- BO1994T-BA1998T at 5% level of significance. However, the results of BH (2013) and HJ (2008) cointegration tests showed significant evidence to support the ARDL results and confirmed the presence of a long-run association among RE, OP, DI, and USI, and they correspond together in the long-term. The outcomes of the diagnostic tests are presented in. The normality test results showed that P-value exceeds the 5% sig level, and it provides evidence that the model of this article is normally distributed. Furthermore, the results of the LM test indicated that there is no autocorrelation in the tested model and this model is homoscedastic. Besides, the Ramsey-Reset test results suggested that the model is well specified. Furthermore, Figs and show the CUSUM and CUSUM of Squares (CUSUMSQ) charts. The CUSUM chart suggests that the model of this article is not miss specified and CUSUM of Squares shows that there is no structural change in the model over the investigated period. The coefficients of the ARDL test are presented in. The outcomes from short- and long-run estimations indicated a significant and positive effect of the oil prices on Turkey’s real estate market. The results in line with who suggested that there is a significant and positive linkage between oil prices and the real estate market. On the other hand, the results indicated a significant and negative effect of the domestic interest rates on Turkey’s real estate market. The coefficient of interest rates confirms the significant impact of the monetary policy in maintaining economic stability in Turkey using the domestic interest rates channel. This finding is consistent with the literature that monetary policy affects the housing markets through interest rate channels. On this basis, the interest rate affects the housing market through the user cost of housing, housing supply, and expectations of future housing prices movements; thereby, an increase in the user cost leads to a decrease in demand for housing, which in turn leads to a fall in housing market prices. Furthermore, an increase in interest rates may have a significant influence on housing construction costs, causing a decline in housing output. This finding corresponds with the findings of Sari (2014) who used the generalized variance decomposition approach and tested the relation between the interest rate and Turkey’s real estate market over the period from 1961 to 2000, and found findings shocks of the interest rate have noticeable effects on the housing market. Besides, the results showed that the U.S. interest-rate coefficient is negatively and statistically significant in both the short and long-run. Thus, any decline in the U.S. interest rates leads to an increase in Turkey’s real estate market. These results confirm that the financial integration between the U.S. and Turkish markets. These results also correspond with the findings of and that confirmed the U.S interest rates affect the Turkish markets. Also, the results showed that the *ECT* is negative and statistically significant at a 5% level. Thus, this result confirms the long-term association among RE, DI, and USI variables. It also indicated that the fluctuations of series from the short- to long-run are amended back (4.4%) every month. The calculated t-statistics of the lagged value of the ECT indicates that there is a long-run causality from the oil price, domestic interest rates, and the U.S. interest rates to Turkey’s real estate market (OP, DI, USI→RE). The tabulated (F)statistics values indicate that there is a bidirectional causality from Turkey’s real estate market to domestic interest rates (RE→DI) and from domestic interest rates to the real estate market in Turkey (DI→RE), and from oil prices to the real estate market in Turkey (OP→RE). Besides, there is a unidirectional causal relation from the U.S. interest rates to Turkey’s real estate market (US→RE), and there is a unidirectional causal relation from the U.S. interest rates to oil prices and domestic interest rate. Thus, this result confirms that there is a spillover influence of the U.S. interest rate channel on Turkey’s real estate market through oil prices and domestic interest rate factors. Therefore, this article provides evidence that the U.S. interest rates have a powerful effect on Turkey’s real estate market through oil prices, domestic interest rate channels. These empirical findings can be attributed to the presence of economic interdependence between the USA and Turkey, and the majority of the external debts and reserve currency in Turkey are composed in the USD. # Conclusion The research aims to provide fresh empirical evidence by testing the impact of the external shocks namely: oil prices and the U.S interest rate on Turkey’s real estate market by using three techniques of co-integration tests. The article covers the period from August 2009 to August 2018. To achieve the main objective of this research: Firstly, the article used the ADF test and the Clemente, Montanes, and Reyes (CM) test with two (SBD) to determine the order of integration of the tested variables. Secondly, the newly developed bootstrap autoregressive distributed lag (ARDL) testing model as proposed by (McNown et al. 2018), the new approach involving the Bayer-Hanck (2013) combined co- integration tests, Hatemi-J (2008) integration testing approach with (SBD) are used to provide strong evidence that the co-integration exists between the tested variables. Thirdly, the Autoregressive distributed lag testing approach (ARDL) is utilized to explore the coefficients between the variables. Finally, The Granger causality (GC) analysis is used to investigate the direction of causality among the variables. The empirical findings from the ARDL testing model indicated that oil prices have a positive influence on Turkey’s real estate market in the short and long term. Besides, the findings from the GC test demonstrate that is a unidirectional causal relation from oil prices to the domestic interest rate. This result confirms that there is a powerful effect of oil prices on Turkey’s real estate market through the domestic interest rate. However, Turkey heavily depends on imported oil; more than 50% of the energy requirement has been supplied by import. Hence, the oil price fluctuations have severe effects on economic performance in Turkeys, which in turn leads to affect the real estate market. The study suggests that the rate of imported oil in Turkey must be decreased by finding more renewable energy sources for the energy supply formula to avoid any undesirable effects of oil price fluctuations on the real estate market and also to achieve sustainable development. Furthermore, the results of this article from the ARDL model demonstrated that there is a significant spillover influence of the U.S. interest rates on Turkey’s real estate market. Besides, the results from the GC test shows that there is a unidirectional causal relation from the U.S. interest rates to Turkey’s real estate market, and there is a unidirectional causal relation from the U.S. interest rates to oil prices and domestic interest rate. Thus, this result confirms that there is a spillover influence of the U.S. interest rate channel on Turkey’s real estate market through oil prices and domestic interest rate factors. This study suggests the U.S interest rates may affect capital outflows, international trade, oil prices, and economic conditions in emerging economies like Turkey. Besides, the presence of economic interdependence between the USA and Turkey, and the majority of the external debts and the reserve currency in Turkey are composed in the USD, and Turkey’s oil imports hit record high in last years. All these indicators and factors led to an increase in the sensitivity and volatility of Turkey’s real estate market to oil prices and the U.S. interest rate fluctuations. Finally, this article suggests that policymakers in Turkey should pay close attention to the effects of external shocks namely the oil prices and U.S. interest rates on Turkish markets to maintain economic and financial stability. [^1]: The authors have declared that no competing interests exist.

# Introduction Diameter distributions are well known and widely used for describing forest stand diameter structure. Accurate quantification of tree characteristics permits study of the interaction among physical and physiological processes and growth. Quantification of diameter distributions over time allows the manager to relate the parameters of the distribution to stand age or stand density. The stand volume characteristics are calculated using diameter distribution and tree height and volume models. Growth and yield prediction based on the diameter distribution approach has also been widely used. Various probability density functions (PDF) such as normal, log-normal, gamma, beta, Johnson’s S_B, and Weibull have been widely used to describe the diameter frequency distributions of forest stands over the last 30 yr,. Additionally, in studies of cumulative diameter distribution, different theoretical growth equations such as Logistic, Gompertz, Mitscherlich, Bertalanffy, Schumacher, Korf, Weibull and Richards have been utilized to characterize the diameter structure of forest stands. Of these, Richards and Weibull equations showed more flexibility than the others, and were respectively the first and the second most accurate. The methods are used to describe diameter distribution can be classified into parametric and nonparametric methods. The abovementioned functions and equations all belong to parametric methods. Nonparametric methods, like percentile prediction method, and *k*-nearest neighbor estimation method, do not rely on any predefined functional form and adapt to description of multimodal distributions. The disadvantage of nonparametric methods is that their high amount of required reference material is difficult to acquire and time- consuming. Doubtlessly, in parametric methods, the Weibull is the most commonly used probability density function for fitting tree diameter distributions. Since three- parametric Weibull function had been derived by Weibull, due to the relative simplicity of expression formula and its flexibility in fitting a variety of shapes and degrees of skew, this function has proved to be a good distribution model. For the estimation of Weibull parameters, many different methods have been applied, such as moment method, maximum likelihood method, percentiles method and nonlinear regression method. With the presentation of advanced analysis software, nonlinear regression method shows its superiority. The best advantage of this method is that parameters can simultaneously accurately be estimated. Additionally, we should acknowledge the facts that iterated function of the three-parametric Weibull distribution is not easy to converge while using Weibull to fit the diameter distribution data, and the correlativity between the parameters estimates and the whole stand characteristics become weak. Thus we wish to explore a new diameter distribution model that overcomes the disadvantages of Weibull and has the advantages of Weibull such as theoretical meanings of parameters and high simulation precision. The objectives of this study were to explore the application of Richards equation on modelling and prediction of stand diameter distribution, test the theoretical meanings of its parameters, and compare the properties of modelling and prediction for stands diameter distribution between Richards equation and three-parametric Weibull equation using the long-term repeated measurement data sets from *Chinese fir* (*Cunninghamia lanceolata*) plantations in southern China. # Materials and Methods ## Data Source *Chinese fir* (*Cunninghamia lanceolata*) stands located in Fenyi city, Jiangxi Province, China, experience a subtropical climate. The longitude is 114°33′E, latitude 27°34′N. Mean annual temperature, rainfall and evaporation are 16.8°C, 1656 mm, and 1503 mm, respectively. *Chinese fir* stands mentioned as follows in the location all are built and authorized by Research Institute of Forestry of Chinese Academy of Forestry and the data originated from our continuous survey. So no specific permits were required for the described field studies, and the field studies did not involve endangered or protected species. The unthinned stands of *Chinese fir* were established in 1981. Planting density was limited within an optimum range according to managerial purposes. The series of stand planting densities was 1667, 3333, 5000, 6667, 10000 stems ha⁻¹. Every planting density had 3 designed replications. Each plot area was 0.06 ha and two adjacent plots were separated by buffer zone. All trees in each plot were marked for continuous measurement. Stem diameter at breast height (Dbh) was measured after tree height reached 1.3 m. All stands were measured every year before reaching 10 years old, and every two years after reaching 10 years old; all stands were measured 10 times. Self-thinning occurred in all stands during the experimental period. The basic information of 150 unthinned stands is described in. The database obtained from these unthinned stands was used to build models of diameter distribution. Another database comprised of 159 diameter frequency distributions was used to test the models. The data came from a thinning study of *Chinese fir* plantations established in the same environment as the above-mentioned unthinned stands. Among the 159 diameter distributions, 63 distributions came from unthinned stands, the remaining came from thinned stands. Because of thinning only being viewed as a management measure that influences stands diameter distribution as same as density, the thinned stands were included in the test database. All thinning was from below. Each plot area was 0.05 ha. The basic information of used stands data for model evaluation is described in. ## Computation of Observed Cumulative Diameter Distribution The diameter classes applied were 2 cm wide. Diameter class, k, is defined in absolute scale (e.g., 1–2.9 cm for *k* = 2, 3–4.9 for *k* = 4, etc.), namely, diameter class *k* is the midpoint value of the absolute scale. The frequency of stems in diameter class *k* at stand *i* is given by: where is the number of trees of diameter class *k* at stand *i* (*i* = 1, 2, …, 159), and is the total number of trees in stand *i*. The cumulative frequency of stems in diameter class *k* at stand *i* can be obtained by: ## Model Development shows the basic mathematical characteristics of Richards and Weibull equations. In the expression of Weibull equations, *a* is the location parameter, *b* is the scale parameter, and *c* is the shape parameter. What, then, has caused the iterated function to not easily converge while employing SAS’s nonlinear regression method to estimate the parameters of Weibull and, subsequently, for the correlativity between the parameter estimates and the whole stand characteristics to be weak? When the expression formula of the Weibull equation is analyzed, it is found the equation is meaningful only when because there is b\>0, so, this means the value of parameter *a* should be smaller than the midpoint of minimum diameter class. If, Weibull equation is meaningless. Furthermore, it is difficult to give a suitable initial estimation for the location parameter *a*, and the ultimate estimation of the location parameter cannot exceed the two neighboring diameter class of any given initial estimation, which can greatly increase the times of iteration. Obviously, it is the location of shape parameter which leads to the above-mentioned shortcomings of Weibull. Richards equation have been found to be suitable to fitting the sigmoid data sets. In the beginning, the curve is concave up, while in later life it becomes convex. Therefore, on the basis of prior study, further analysis is needed to determine the mathematical characteristics of Richards. While fitting distribution data, the parameter *B* of Richards is \<0. As a result, the following new expression formula of Richards function in can be derived: becomesWhile fitting diameter distributions, the empirical values of parameter *B* in Richards function are\<−3, then \>0, besides, parameter *k* is \>0, it can be concluded that parameter *q* is \>0, and parameter *p* also is \>0. Because parameter *m* is \>1, parameter *r* is \<0. Undoubtedly, Eq. (2) will still have high simulation precision, and it is obvious and important that the parameters of Eq. (2) may have the same theoretical meaning as three- parametric Weibull equation. Namely, parameter *q* may be the location parameter, parameter *p* may be the scale parameter, and parameter *r* may be the shape parameter. Note, whether the parameter *q* of Eq. (2) is\>*x* or not, problems that the equation has no meaning or parameters have difficulty to converge will not appear. We call Eq. (2) “*R* distribution”. The probability density function of *R* distribution for a random variable *x* is Eqs. (2) and (3) have the advantage of a simple expression formula. Like Weibull equation, Eq. (2) has a floating inflection point, the abscissa and ordinate of inflection point of Eq. (2) are respectively given by Accordingly, *R* distribution has a good application prospect in the field of diameter distribution. Our database, including 150 cumulative diameter distributions, was used to fit the *R* distribution and Weibull equation, then the theoretical meaning of parameters of *R* distribution was analyzed by discussing the correlativity between the parameters estimates and the whole stand characteristics. The skewness × kurtosis can describle the shape and modeling properties of distribution function while using for forest stand. The shape of *R* distribution is herein evaluated in terms of its skewness × kurtosis from one side. Skewness and kurtosis values of frequency distributions of original data and *R* distribution are calculated for every 150 stands. The mathematical formulas are: where is the midpoint value of diameter class *k*, and is the average value of DBH of a stand, is the frequency of diameter class *k*, is standard deviation of DBH. ## Model Evaluation Model evaluation was made through three steps: (1) the parameters of *R* distribution and three- parametric Weibull equstion were estimated by using nonlinear regression method (NRM) or maximum likelihood estimates method (MLEM) for each of the 150 unthinned stands, (2) prediction functions between parameters and stand characteristics were built by using parameter prediction method (PPM) and parameter recovery method (PRM) for all 150 unthinned stands, (3) the modelling and prediction properties were tested by adopting the 159 independent stands. For *R* distribution, parameter estimates for *p*, *q*, r were obtained by employing NRM. For Weibull equation, parameter estimates for *a*, *b* and *c* were obtained by employing NRM and MLEM. Because the location parameter *a* of the three-parametric Weibull function was often viewed as the minimum observed diameter or its multiple, parameter *a* was defined as the lower limit of the minimum diameter class when MLEM was adopted. PPM and PRM were used to build stand-level diameter distribution models ,. PPM means direct regression of parameters from stand characteristics, and PRM means to recalculate parameters after estimating percentiles or some key point on equation curves from stand characteristics. Some parameters of *R* distribution and Weibull were regressed against stand characteristics which included stand age, site index, planting density, average height, dominant height and quadratic mean DBH (*D_g*), some parameters were derived from the moments of the diameter distribution, which were themselves estimated from stand characteristics. *D_g* is calculated by, where is the diameter values in a stand, *n* is the total number of trees in the stand. For *R* distribution, when PPM and PRM are simultaneously adopted to predict diameter distributions of stands used for model testing, based on preliminary graphical and correlation analyses, the parameter prediction equations for estimates for, and are given by Eqs. (4), (5) and (6), respectively where are the stand characteristics, and is the estimate of abscissa of inflection point of stand diameter distribution, which can be derived by When only PRM is the adopted method, besides Eq. (6), Eqs. (8) and (9) can be used to recover the parameters where and are the diameter at the percentile 0.333 and 0.9 on the distribution curve. Like, and can be predicted by the same formula as Eq. (7). For the three-parametric Weibull function, the PPM and PRM are simultaneously adopted to predict stand parameters. Parameters and can be estimated as parameters and of *R* distribution. Parameter can be solved from the cumulative distribution function of Weibull by Eq. (10). The reason that the 0.5 percentile was adopted is that this percentile was near the inflection point of the distribution curve. The prediction of can be achieved by using the same method as and of *R* distribution. To further test the modelling and prediction properties of *R* distribution and Weibull, another database comprised of 159 diameter frequency distributions was used to test the models. ## Comparison of the Models The application effect of *R* distribution and three-parametric Weibull function was examined by comparing the residual sum of square (*RSS*) and coefficient of determination (*R²*). The residual sum of square and coefficient of determination were respectively calculated as where *obs_k* and *est_k* are the observed and predicted diameter frequency for diameter class *k*, and *n* is the number of diameter classes in a sample stand. We also used the Kolmogorov-Smirnov test to test the goodness of fit of distribution functions. SAS 9.1 and EXCEL 2003 was adopted for parameter estimation and model evaluation. # Results and Discussion ## Modelling Results of *R* Distribution and Three- Parametric Weibull Function shows the mathematical characteristics of parameters and statistics for *R* distribution and the three-parametric Weibull function derived from the 150 plot measurements. For *R* distribution, the estimates of parameters *p*, *q* and *r* are \>0, \>0 and \<0, respectively, which is identical to the empirical distribution range mentioned). The floating range, mean and standard deviation of parameter *p* or *q* shows that these two parameters are both stable. The total RSS from all stands of *R* distribution (0.1913) was smaller than Weibull distribution (0.2069), and the mean of *R²* (*R²* = 0.9990) was slightly higher than Weibull distribution (*R²* = 0.9988) from every stand while NRM was applied. Although the precision of the two models are both high, *R* distribution presents a more accurate simulation than the three-parametric Weibull function. For the three- parametric Weibull function, the precision of NRM is far higher than MLEM in the view of total RSS and *R²*. The skewness × kurtosis of the original data and *R* distribution is depicted in. Each circle dot represents a stand. The distribution range of skewness is similar between the original data and *R* distribution. Most stands have a negative skewness, which is different from inverted J distribution that often happens for the natural stand with positive skewness. The kurtosis obtained from *R* distribution are a little smaller than the original data for some stands, but the value range of kurtosis is same as the actual values. The results mean the shapes modeled by *R* distribution are undistorted and multiple. shows both the actual data distribution (histogram), the estimated R distribution shapes (bold solid line), the estimated Weibull distribution obtained from NRM (solid line) and the estimated Weibull distribution obtained from MLEM (dashed line) for a selection of stands from different planting densities, stand ages and quadratic mean DBH. In general *R* distribution and Weibull distribution from NRM both have provided a good fit for all the stands analyzed, and Weibull distribution from MLEM provided a relatively bad fit. In comparison, *R* distribution is more stable than Weibull distribution from NRM, which can be seen from the application of the two distributions to stand 2, 4. ## Theoretical Meaning of Parameters of *R* Distribution The modelling accuracy and theoretical meanings of parameters are the two most important indexes used to judge whether a function is suitable to modelling diameter distributions and the two indexes are complementary. It is known that *R* distribution has high modelling precision. However, do the parameters of *R* distribution have good theoretical interpretation? This question can be answered by discussing the relationship between parameters of *R* distribution and the basic stand characteristics such as stand age, stand density, site index and quadratic mean DBH. ## Theoretical Meaning of Parameters *p* of *R* Distribution shows the relationship of parameter *p* of *R* distribution to stand age and quadratic mean DBH. Parameter *p* increased with increasing stand age and quadratic mean DBH. Liu et al. (2004) also found that the relationship between scale parameter *b* of Weibull distribution and stand age was positive. The relationship of parameter *p* and stand age and quadratic mean DBH were well approximated by the brief second polynomial. The resultant parameter prediction equation for predicting *p* is given by Eq. (11). where *t* refers to stand age. The result of analysis of variance showed that Parameter *p* had significant relativity with stand age at the 0.0001 significance level. The coefficients of determination (*R²*) of the second polynomial between parameter *p* and stand age, planting density, site index and quadratic mean DBH were respectively 0.7369, 0.0235, 0.0093 and 0.3185, and the positive or negative relativities of them can reasonably interpret the theoretical meaning of parameter *p* as a scale parameter of diameter distribution. According to the definition of scale parameter of Weibull distribution, parameter *p* can be viewed as the scale parameter of *R* distribution. ## The Theoretical Meaning of Parameters *q* of *R* Distribution The relationship of parameter *q* of *R* distribution to stand age, stand density, site index and quadratic mean DBH is illustrated in. Parameter *q* increased with increasing stand age, site index and quadratic mean DBH, while it decreased with increasing stand density. The coefficients of determination (*R²*) of the second polynomial between parameter *q* and stand age, stand density, site index and quadratic mean DBH were respectively 0.2477, 0.5843, 0.3011 and 0.7886. It is obvious that quadratic mean DBH has the biggest effect on parameter *q* among these stand characteristics. The parameter prediction equation for predicting *q* is given by Eq. (12). The coefficients of Eq. (12) were significant at the 0.0001 significance level. From, it can be known that the positive or negative relativities of parameter *q* and the four stand characteristics can reasonably interpret the theoretical meaning of parameter *q* as a location parameter of diameter distribution. For Weibull distribution, the location parameter *a* also increased with increasing quadratic mean dbh and stand age. However we should know the fact that the Weibull location parameter must be smaller than observed diameter in a stand. Based on our experience with the data set and other data sets, the convergence and precision of Weibull function are sensitive to the location parameter *a*. Some researchers regarded parameter *a* as minmum diameter in a stand or its times \[4,340,41\], such as 1/3, 1/2, and 1. And it is difficult to get a common sense. However, for *R* distribution, there is no strictly restrictive condition about location parameter *q* (eq. 2). According to the form of *R* distributin function, parameter *q* can be regarded as the location parameter of *R* distribution, which may make *R* distribution become a new promising diameter distribution. In previous studies, Duan et al discovered Richards function was suitable for modelling diameter distribution, but did not realize the underlying relationship between parameter *B* and *k* in Richards function. They thought parameter *B* had poor theoretical interpretation. For *R* distribution, parameter *q*, composed of parameters *B* and *k*, obviously has good theoretical meaning, and proved easy to converge. This might lead to the use of *R* distribution as a new diameter distribution. ## Theoretical Meaning of Parameters *r* of *R* Distribution shows the relationship of parameter *r* of *R* distribution to stand age and quadratic mean DBH. Parameter *r* decreased with increasing stand age and quadratic mean DBH. The result of analysis of variance showed that parameter *r* had significant relativity with stand age and quadratic mean DBH at the 0.01 significance level. The coefficients of determination (*R²*) of the second polynomial between parameter *r* and stand age and quadratic mean DBH were 0.0562 and 0.0706, respectively. For a sigmoid curve, the location of inflection point decides its shape at some extent. Parameters *p*, *q* and *r* together decide the abscissa of inflection point of *R* distribution. As shown in, the abscissa of inflection point of *R* distribution increased with increasing stand age, site index and quadratic mean DBH, and decreased with increasing stand density. The coefficient of determination (*R²*) of the second polynomial between abscissa of inflection point and stand age, stand density, site index and quadratic mean DBH were 0.3454, 0.6180, 0.3197 and 0.9649, respectively. The deep relationship between the abscissa of inflection point and quadratic mean DBH can be described by Eq. (13). where *D_g* refers to quadratic mean DBH. The value of, being the ordinate of inflection point of *R* distribution, is decided by parameter *r*. shows the relationship of ordinate of inflection point of *R* distribution to stand age. The ordinate of inflection point decreased with increasing stand age. The coefficient of determination was 0.2402, which means that the ordinate of inflection point of *R* distribution can show the change of distribution shape with age in some extent. Besides, the potential relationship between the ordinate of inflection point of *R* distribution and its skewness is showed in. It is found that the ordinate of inflection point of *R* distribution has significant relativity with its skewness and kurtosis, the linear coefficient of determination are 0.4483 and 0.2824 respectively. This phenomenon shows that the size of inflection point of R distribution can report the shape of stands in some extent. Due to the act of parameter *r* in inflection point of *R* distribution, it can be concluded that parameter *r* has a deep relationship with the shape of *R* distribution. Besides, the significant correlation between abscissa and ordinate of inflection point of *R* distribution and stand characteristics indirectly shows that parameter *r* is flexible and theoretical. Accordingly, parameter *r* may be considered as the shape parameter of *R* distribution. ## Distribution of Inflection Point of *R* Distribution *R* distribution has a flexible inflection point. The variation range of the ordinates of inflection points is 0.3787∼0.6436, and 91.33 percent of them distribute in the range of 0.4∼0.6. Therefore, we can conclude that the main distribution range of inflection points for the cumulative diameter distribution of stands was 0.4∼0.6. ## Functions for the Estimation of Stand Parameters Stepwise regression analysis was applied to build the relationship between the parameters of two distributions and the stand characteristics composed of stand age, planting density, site index and so on at a 0.5 risk level. The regressions are on the 150 estimation stands. When nonlinear regression method were adopted, the resultant parameter prediction equations for predicting, and are given by Eqs. (14)–(22), respectively. When maximum likelihood estimates method adopted, the resultant parameter prediction equations for predicting, and are given by Eqs. (23)–(26), respectively. ^aThe variables are, respectively, dominant height, stand age, site index, planting density, quadratic mean DBH and mean height, and are the diameter estimates at the percentile 0.333 and 0.9 on the *R* distribution curve respectively, is the estimate of abscissa of inflection point of *R* distribution, is diameter estimate at the percentile 0.5 on the Weibull curve. Obviously, because of weak relativity, parameter of *R* distribution was not adaptive to be directly predicted by stand characteristics, which could be indirectly obtained through Eqs. (6) or (13). For parameter, of Weibull distribution, the coefficients of determination of second degree polynomials between the other parameter and above-mentioned stand variable were all smaller than those of Eqs. (19), (20), (21), and (22), respectively. Therefore, the five equations were adopted to predict the unknown distribution parameters. Note that for the three-parametric Weibull function, regardless of whether the nonlinear regression method or maximum likelihood estimates method is used, always has significant relativity with planting density and stand age at the 0.0001 significance level. However, under the consideration of relativity, parameter is obtained through Eqs. (21) and (22) or Eqs. (25), and (26). It can be found that these functions have different stand characteristics and exponentia, which are mainly concluded by the analysis of theoretical meaning of parameters and the needs of brief and high precision of models. In practice, while considering parameter prediction method, the parameter prediction models that have both acceptable high precision and theoretical meaning are firstly selected, which often include single stand characteristics, then for parameters with high relativity with some stand characteristics, stepwise regression analysis can be applied. While some or single parameter in distribution model with weak relativity with stand characteristics, parameter recovery method can be adopted. Of course, parameter recovery method can be applied in other conditions. Based on the above-mentioned parameter prediction equations that have a high coefficient of determination, parameters of distribution models can be evaluated by introducing the related stand characteristics into these equations. ## Goodness-of-fit for Estimating Diameter Distribution Data from 159 evaluation subplots provide an opportunity to analyze and compare the accuracy of the *R* distribution and three-parametric Weibull function. In these calculations, two fitting methods (nonlinear regression method and maximum likelihood estimates method) and two parameter estimation methods (PRM and the combination of PPM and PRM) were used. It was encouraging that *R* distribution was found to have lower *RSS* and higher non-rejection rate than the three-parametric Weibull function (e.g. Table. 6). shows the distribution of the residual sum of square (*RSS*) against quadratic mean DBH, from which we can directly compare the prediction effects of the five methods. Method A has the highest precision among the five methods, which shows *R* distribution can accurately estimate diameter distribution of most future stands using the combination of PPM and PRM under the condition that only two stand characteristics are known (e.g. Table. 6). For method A, the result of the Kolmogorov -Smirnov test showed the null hypothesis that the observed and fitted distributions that are the same cannot be rejected for 128 out of 159 stands (80.50%). However, in view of the practical application of distribution models to stand management, method B, based on *R* distribution, would be a better option because of its dependence on stand characteristics related to stand density and site quality and because its non-rejection rate reached 73.58% for the total test data. Methods B and D used the same parameter recovery method and numbers of stand characteristics and had a nearly equal non- rejection rate, which showed *R* distribution had a distribution function equally as good as three-parametric Weibull function. Furthermore, because the iterated function of *R* distribution was easier to converge than three- parametric Weibull function when using nonlinear regression method to predict parameters, *R* distribution would have a wide application prospect. For *Chinese fir* plantations, the non-rejection rate of unthinned stands and thinned stands did not have obvious differences when methods A, B, C and D were adopted to predict the diameter distributions of future stands (e.g. Table. 6). Additionally, for the three-parametric Weibull function, the nonlinear regression method is a more effective approach than the maximum likelihood estimates method in the estimation system of stand diameter distribution (e.g. Table. 6). # Conclusion Based on analysis of the disadvantages of the three-parametric Weibull function, this study develops a promising distribution function (*R* distribution), which is a new and essential exploration in the study of parametric methods. We conclude that: (1) *R* distribution has a more accurate simulation than the three-parametric Weibull function while modelling diameter distributions of *Chinese fir* plantations; (2) the parameters *p*, *q* and *r* of *R* distribution proved to be its scale, location and shape parameters, and have a deep relationship with stand characteristics; this means the parameters of *R* distribution have good theoretical interpretation; (3) the main distribution range of inflection points for the cumulative diameter distribution of *Chinese fir* plantations was 0.4∼0.6; (4) the goodness-of-fit test showed the diameter distributions of unknown stands can be accurately estimated by applying *R* distribution and with regards to modelling precision and theoretical interpretation, method B may be the most suitable choice due to its good convergence, high precision and including multiple stand characteristics. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: A-GD J-GZ. Performed the experiments: A-GD. Analyzed the data: A-GD. Contributed reagents/materials/analysis tools: A-GD X-QZ C-YH. Wrote the paper: A-GD.

# Introduction A hallmark of many human cancers is genomic instability, and cancer itself can be thought of as the result of an altered ploidy. In order to gain a greater understanding of the causes underlying tumor formation, one must understand the core events and changes of cancerous cells. The genetic identity of each cell determines the fate of the cell and thus the cancer genome is a source of information to be mined in order to identify both the ways cancers arise and how they can be treated. The importance of studying the cancer genome cannot be understated as genomic alterations have been linked to cancer causation both broadly and in specific subgroups of patients. Dysregulation of gene expression is one mechanism by which cells become tumorigenic. It has been shown that alterations on a genomic DNA level are likely to cause associated changes in gene expression. Previous global gene expression profiling studies of breast carcinomas have identified at least five distinct subtypes of breast cancer, with specific patterns of <u>C</u>opy <u>N</u>umber <u>A</u>berrations (CNA) that can also define genetic events associated with different expression subtypes. The interplay between genomic DNA changes and gene expression is something that can yield much information about the underlying processes that contribute to breast cancer formation and development. Continued investigation of copy number abnormalities in breast cancer is likely to yield additional insights into the pathogenesis of the disease. As knowledge about breast and other cancers advances, we are finding that there are numerous complex ways by which the genome can be disrupted. While certainly gross chromosomal aberrations and rearrangements have been seen to initiate disease, more subtle derangements of the genome can also contribute to tumor formation as well. With the continued advancement in technologies to detect DNA copy number changes, previously difficult to detect varieties of genetic abnormalities are continuing to be discovered. Through the use of a high-density array comparative genomic hybridization (HD-aCGH) platform, it is possible to detect both gross and fine-scale aberrations in genes. Small-scale CNA, here termed micro-aberrations, represent a previously under-investigated source of copy number variation that may shed light on breast subtype characteristics and tumorigenesis. These micro-aberrations have not previously been the focus of any dedicated study and thus we sought to design assays to identify and characterize them. Enhanced detection, cataloguing, and validation of these events could be an avenue through which we can gain a greater understanding of breast cancer genomes through the improved ability to detect genetic events affecting gene expression and function. We focused our investigation on 128 genes shown to be of importance in breast and other cancers; we reasoned that because these genes were frequently disrupted in cancer, they might be more likely to harbor detectable micro-aberrations. Furthermore, any micro-aberrations that might be detected within these genes would be more likely to be biologically relevant and functional than ones that might be detected within a randomly selected panel of genes. Thus, in this study, we utilize a fine-resolution platform to identify these small scale events and examine the functional consequences that result when they are present. # Methods ## Ethics Statement All samples used in this study were collected using IRB-approved protocols and all patients signed informed consent forms and the data were analyzed anonymously. ## Breast Cancer Patient Dataset The dataset used here contained both gene expression and high-density array comparative genomic hybridization (HD-aCGH) copy number data from a set of breast tumors from UNC “HD-UNC94” (n = 94). Additionally, the SUM102 and SUM149 breast cancer cell lines were obtained from Dr. Steve Ethier, and assayed using these high-density tiling arrays. Tumors in the dataset were assayed for gene expression patterns using Agilent DNA microarrays as previously described. Log₂ ratio data were taken from the UNC Microarray Database (UMD), filtering for a lowess normalized intensity value of 10 or above for each channel, and 70% present data values, and then used for further analyses. Data is available from Gene Expression Omnibus under GSE36889 Sample information including clinical data, subtype, source, GEO Sample ID and overlap with copy number information, can be found in. ## Classifying Tumors for Gene Expression-based Subtype Classification The Lowess normalized gene expression R/G Log₂ ratio data from the HD-UNC94 data set used different gene expression microarray platforms. The dataset was therefore then limited to the probes/genes shared across both platforms. After column standardization of both platforms (samples at N(0,1)), Distance Weighted Discrimination (DWD) was used to remove platform bias prior to classification for the gene expression arrays. After normalization, the R/G Log₂ ratio data was collapsed (via averaging) from probes to HGNC gene symbols. The PAM50 gene set predictor was used to assign subtypes to the tumors. ## Tiling Array Design The custom HD-aCGH tiling platform was designed using Agilent’s E-array v5.0 online (<https://earray.chem.agilent.com/earray/>) software and built on the Human 244 k Custom Oligo platform (GPL15359 Agilent UNC Perou Lab 1X244 k Custom Tiling CGH Array). 230,606 probes cover a total region of 45 Mb, which includes the full genomic sequence of the 128 genes of interest as well as the region 150 kb upstream and downstream of each of these genes; this design gave an average resolution of 200 bp between contiguous probes. Labeling and hybridization were performed according to manufacturer’s instructions using the Agilent Genomic DNA Labeling Kit PLUS (Catalogue Number 5188–5309). A Human Genomic DNA Pool (Promega, Catalogue Number G3041) was used as reference DNA, which was compared versus every tumor or cell line sample. Microarrays were scanned on an Agilent DNA Microarray scanner (G2565CA) and the data uploaded to the University of North Carolina Microarray Database (UMD, [www.genome.unc.edu](http://www.genome.unc.edu)). ## Identification of CNA and microCNA with SWITCHdna To determine regions of Copy Number Aberration (CNA), we utilized the SWITCHdna algorithm, focusing on individual genes. For the purposes of this study, the analysis window was limited to the genomic region of each gene and its introns, plus the 5 kb upstream of the start codon, and downstream of the end of the 3′UTR. In order to further filter the identified segments, we set the cutoff for the absolute value of the log₂ ratio to be greater than 0.30 in order to reduce false positives. After identifying all genomic segments of alteration using SWITCHdna, we analyzed the distribution of sizes of aberrant segments and established a cutoff of \<64 contiguous SWITCHdna probes, or ∼ ≤15 kb as the definition of a micro-aberration, which equates to the smallest 25% of CNA in this dataset. In order to identify the regions of genes most commonly affected by micro- aberrations, each gene was divided into four quadrants based upon a proportional splitting of each gene into four equal segments: 5′ End (typically being the promoter region, 5′UTR, beginning regions of gene), 5′ Middle (first ½ of gene), 3′ Middle (second ½ of gene), 3′ End (typically being the end regions of gene, 3′ UTR, downstream region). For every micro-aberration instance, the affected quadrants were tallied for each quadrant, and the proportion of affected quadrants out of all possible quadrants was calculated. Similarly, each micro- aberration instance was assessed in terms of whether it encompassed the promoter or 5′ untranslated region (UTR) of each gene, with the promoter region defined as genomic space upstream of the transcription start site. ## mRNA-seq mRNA-seq was performed on total RNA isolated from cell lines and tumors using the Qiagen RNeasy Mini Kit (Cat. No. 74104). Library preparation was performed using the TruSeq RNA Sample Kit from Illumina (Cat. No. RS-930-2001) following the low input protocol detailed in the manufacturer’s guidelines. 1×76 bp nucleotide reads were generated using an Illumina GAII sequencer for the SUM102 cell line. For the two tumor samples (UNC990141B and UNC040182B), we sequenced using a 2×50 bp configuration using an Illumina HiSeq2000. In all cases, the read data were aligned to the human HG19 reference genome from the UCSC genome browser **,** and mapped using MapSplice. The alignments were visualized using the Integrative Genomics Viewer (IGV) for evidence of micro-aberrations. To investigate the molecular mechanisms of these aberrations, we next collected the reads aligning to the target regions and performed *de novo* assembly with the Trinity. Default assembly settings were used, and the *de novo* assembled contigs then compared to the reference using BLAST. ## Survival Analysis The patients in the dataset were rank ordered by total number of SWITCHdna- defined aberrations, and micro-aberrations, and separated in the top 67% and bottom 33%; additional rank order splits were also evaluated, but data not shown. Survival analyses were performed using the Kaplan-Meier test in R. # Results ## Copy Number Micro-aberrations are Present in Breast Tumor Subtypes In order to test the hypothesis that primary breast cancer genomes contain areas of small-scale copy number gains and losses, termed micro-aberrations, we designed a custom, high-resolution, high-density, comparative genomic hybridization tiling array (HD-aCGH) with an average probe spacing of 200 base pairs, and which was focused on 128 selected genes. We assembled a dataset of 94 tumors and 2 cell lines and tested them on this HD-aCGH array. Each tumor was also classified into one of five previously defined gene expression subtypes using the published PAM50 identifier, and genomic DNA copy number aberrations identified using the SWITCHdna algorithm. Using this HD-aCGH array, we were able to identify both previously observed large scale amplifications and deletions, and novel small-scale copy number aberrations, which we have highlighted a few selected examples here; the SUM149 cell line has a previously identified micro- amplification in exon 2 of the PTEN gene, which we also clearly observed using our HD-aCGH platform. A number of other relatively small intra-genic aberrations were also detected, including a focal deletion in PTEN in a basal-like tumor, an intra-genic deletion in RB1 in a different basal-like tumor , and a small RB1 amplification in yet a third basal-like tumor. To objectively define a micro-aberration, we established a size definition of the smallest 25% of SWITCHdna identified aberrations in this dataset, which resulted in the size cut off of approximately 64 contiguous probes (∼15 kb). The previously identified PTEN micro-amplification in the SUM149 cell line is able to be identified using these criteria. The genomic landscape of the basal-like subtype exhibited many of these micro-aberrations, as basal-like tumors showed the highest incidence of these events. The value of the high-density tiling array platform is also shown in another example of a RB1 alteration. Each identified micro- and macro-aberration segment for this gene is plotted with the RB1 exons identified by each gray stripe, along with the location of each RB1 probe on the tiling array (green segments), and also shown are probes from an earlier 109,000 feature single nucleotide polymorphism (SNP) platform, which was used in a previous study examining whole- genome landscapes of breast tumor subtypes (5 probes: red lines). The micro- aberration segments overlap with at most one genome-wide probe from the 109K arrays, and thus would have never been called a loss given how most aCGH programs call “changed segments”, which was the case for SWITCHdna. In addition, several macro-aberrations identified from the tiling array platform have minimal overlap with the 109K genome-wide probes. 49/94 samples assayed here on the HD- aCGH array had previously been assayed on the 109K SNP platform and the results of this overlap set for RB1 were compared directly in terms of CNA assignment agreement by SWITCHdna. Again, focusing on the RB1 gene as our example case, we observed copy number aberrations in six of these 49 samples by the HD-aCGH array for RB1; only two of these six samples’ CNA were detectable by the 109K SNP array and these were the aberrations that spanned the whole gene. The remaining four, all of which were intra-genic events, were missed by the copy number segments generated from the whole-genome array (data not shown). This is illustrated in two example RB1 gene plots directly comparing the probes and segments from the high-density tiling array and the 109 k SNP array. ## Small Scale Copy Number Aberrations can Affect Gene Expression Beyond simply determining the frequency of micro-aberrations present within each gene or tumor, we wanted to assess whether the presence of micro-aberrations would result in functional consequences. We first assigned a copy number status for each gene and each sample (gross copy number gain, gross copy number loss, micro-amplification, or micro-deletion) and then for each gene, determined whether the corresponding gene expression was concordant with the type of genomic aberration observed (i.e. micro-amplifications result in increased expression and/or micro-deletions result in decreased expression of the involved gene). We found similar rates of concordant expression between micro-aberrations and gross aberrations, with 30–40% of the tested genes showing 100% agreement between aberration type and gene expression, meaning that for these genes, every sample that displayed a micro-amplification in the gene also had greater than median expression of the gene and likewise every sample that had a micro- deletion in the gene had less than median expression of the gene. Another 50–75% of the tested genes showed at least 50% concordance between aberration type and gene expression meaning at least half of the samples that displayed a copy number aberration had altered expression of the affected gene in the same direction as the CNA. These findings suggest that the micro-aberrations have functional effects upon gene expression similar to what is seen with larger scale CNA. We also examined the expression status of each micro-aberration by micro-aberration type to determine if there was an association between micro- amplifications and high expression and micro-deletions and low expression across all events instead of within genes. When all micro-aberrations are combined, there is no significant difference in expression level between samples with micro-aberrations versus those without, but significant differences by Fisher’s Exact test were observed when looking within micro-amplifications or micro- deletions. A number of genes containing micro-aberrations also showed differential expression of the involved gene when comparing the aberrant vs. non-aberrant groups by ANOVA. ANOVA box plots are shown for the genes NUF2 and UBE2T, where samples with micro-amplifications had significantly higher expression of the gene than those without micro-amplifications. Also shown is ZNF217 , where samples with micro-deletions had lower expression of the gene than those without micro-deletions, and SLC7A6 where the samples with micro- deletions had higher expression than the samples that do not; we do note that the sample size is small in some cases, but overall these data suggest that micro-aberrations can affect gene expression. ## Genomic Micro-amplification Causes Exon Skipping We performed a closer examination of the micro-amplification of the PTEN gene in the SUM149 cell line, as it is a validated aberration that has now been identified by multiple groups. Using mRNA-seq data, we assessed the expression of the PTEN gene on an exon level to determine the functional consequence of the DNA micro-amplification. For comparison, we also examined the data for the SUM102 cell line, which has no genomic alterations in PTEN (data not shown). The distribution of aligned reads for each exon is shown for each cell line, with the SWITCHdna copy number segments for SUM149 shown in genomic space for reference. In SUM149, there is a lack of PTEN gene expression starting from the middle of exon 2, which coincides with the location of the genomic DNA micro- amplification. In comparison, the SUM102 cell line has aligned reads throughout the entirety of the PTEN gene, thus the micro-amplification in SUM149 causes a loss of expression of all downstream exons. Additionally, we sought to validate other micro-aberrations using mRNA-seq data. We were able to observe instances where the presence of a micro-aberration resulted in production of inter-exon mRNA reads with sequence infidelity. In the 990141B tumor sample, the presence of a micro-amplification in the *EGFR* gene results in many mRNA-seq reads that mapped outside of the exon, and that often contain misalignments. This finding was not observed when examining data from the SUM102 cell line data that does not contain a similar micro-aberration. Similarly, in the UNC040182B tumor sample, a micro-deletion in the *BCL11A* gene results in mRNA-seq reads that align outside of the exon containing multiple sequence errors, a finding likewise not observed in the unaffected SUM102 cell line data. We pursued further analysis by performing *de novo* assembly of mRNA-seq reads that mapped to the region of *EGFR* micro-amplification in the 990141B tumor sample and generated two contigs that aligned to the reference mRNA sequence, aside from 5 bases at the start of the contig and 3 bases at the end of the contig. The 5 base unaligned sequence occurs twice within the aligned reference region and may represent the site joining a duplication of the stretch of genome that results in the micro-amplification. *de novo* assembly of mRNA-seq reads that mapped to the region of *BCL11A* micro-deletion in the UNC040182B tumor sample generates a contig that begins at the end of the of the affected exon and extends into intronic space; this could be attributed to loss of the beginning of the exon, which then causes mis-splicing and the generation of an altered mRNA. Thus in these two cases, the micro-aberrations affected the mRNA structures. ## The 5′and Promoter Regions of Genes are most Commonly Affected by Micro-aberrations In order to assess whether certain regions of genes were more commonly affected by micro-aberrations than others, we portioned each of the 128 genes on the HD- aCGH array into four quadrants: 5′ End, 5′ Middle, 3′ Middle, and 3′ End, based upon a proportional splitting of each gene into 4 equal segments. For every micro-aberration instance (n = 330), we noted the quadrants that it occupied, and then for each quadrant, determined what proportion of the total possible quadrants were affected by a micro-aberration event. We found that the 5′ end of the gene was disproportionately affected by micro-aberrations (88% vs. \<40%). Additional refinement of the affected region was also performed and a large percentage of micro-aberrations also affected the promoter (defined as the area upstream of the coding region) and 5′ UTR regions as well (80.6% and 79.7% respectively). Of the aberrations whose area of effect was limited only to the 5′ End, the promoter region was affected at a higher rate than the 5′ UTR region (92.0% vs. 68.1%). ## Micro-aberration Frequency Associated with Poorer Survival The survival outcomes of patients with varying levels of copy number aberrations were also assessed to determine if an association was present. We identified the number of copy number micro-aberrations per sample using our SWITCHdna criteria and rank-ordered the patients in terms of micro-aberration frequency. Each patient in the HD-UNC 94 dataset was assigned to one of two groups depending on whether they were in the top 67% of microCNA or the bottom 33%. Kaplan-Meier analysis was performed examining overall and relapse-free survival. We saw that patients with the least genomic instability as assessed by SWITCHdna-called micro-aberrations had significantly better outcomes in terms of both overall and relapse-free survival. A caveat to these analyses is that the sample size is small, and additional rank order splits of the data were only trending towards significance, but similar to what has been seen before for large numbers of large scale changes, , tumors with the most numbers of changes tended to show worse survival. Lastly, we examined the frequency of micro-aberrations for each of the 128 genes tested. Here, we list the top 17 most micro-aberrant genes among those that were tested on this tiling array with breakdown by micro-amplification and micro- deletion. We also show the number of micro-aberrations that would be expected by chance for each gene based on the distribution of log-ratio values within our dataset and our cutoffs for micro-aberrations. Genes such as *MYC* and *PIK3CA,* known to be activated in many cancers tend to show more micro-amplifications; others known to be inactivated in cancer such as *RB1* and *PTEN* display comparatively more micro-deletions. Gene Set Enrichment Analysis was performed using DAVID on the genes exhibiting more than one micro-aberrations in our study, where the background for the analysis was limited to the 128 genes present on the tiling array in order to control for our biased initial selection of genes. We observed that micro-aberrant genes were more likely to be involved in interphase of the mitotic cell cycle. # Discussion The previous discovery of micro-aberrations within genes using a high-density aCGH array, and the lack of description of such features in many whole-genome aCGH-based breast cancer studies, suggests that these micro-aberrations may occur regularly in breast cancer genomes and that they simply have not yet been detected in previous studies due to the resolution of typical SNP-based aCGH platforms. To address this hypothesis, we assembled a dataset of 94 breast tumors and two breast cancer cell lines and tested them on a custom-designed aCGH tiling array; this array was targeted to 128 gene panel focused on important cancer relevant genes, and previously identified basal-like cancer specific regions and genes. By utilizing a previously tested segmentation and aberration calling algorithm called SWITCHdna, we analyzed the tiling array data and proceeded to generate a numerical definition of a micro-aberration (\<64 probes, ∼15 kb). Essentially all of these micro-aberrations would be mostly undetectable using lower- resolution genome-wide platforms, as these segments would be covered by at most one probe on such arrays. An analysis of the frequency of micro-aberrations within our dataset samples suggests that the basal-like subtype had the most frequent occurrences of these events, mirroring their high overall genomic instability, ; thus the presence of micro-aberrations did correlate with the presence of large aberrations. We note that the frequency of micro-aberrations observed in our dataset was similar to that for both gross copy number aberrations and single-nucleotide variants, when we examine the same genomic regions targeted in this study for comparable distribution of tumor samples (data not shown). Our data also shows that at least in some cases, these events have functional downstream consequences. One point of discussion is the mechanisms by which these micro-aberrations might lead to altered gene expression. It is intuitive how micro-deletions might result in decreased gene expression of the affected gene. However, it is less clear how micro-amplifications can result in increased expression of the affected gene. One proposed mechanism is that micro- amplifications might preferentially occur in the 5′ promoter site, given the overall predilection for micro-aberrations to occur in that region, and those micro-amplifications that occur in the promoter have higher rates of positive concordance than those that occur elsewhere. This did not appear to be the case within this dataset (data not shown), but is a mechanism worth considering for expanded studies. Another mechanism could be that the affected gene is disrupted in a heterozygous fashion, and upregulation of the remaining copy occurs as a result. The current platform is not designed to distinguish between homozygous and heterozygous change, but this is a mechanism by which altered expression might occur. The finding of exon skipping at the point of the focal amplification in the PTEN gene in the SUM149 cell line is particularly interesting given the otherwise normal copy number for this gene in this cell line. The exact mechanism that induces this exon skipping is yet to be determined, but one can imagine that some aspect of the amplified DNA sequence results in an alteration to the pre- processed transcript that could cause early truncation or some sort of structural interference. Likewise, the presence of the micro-aberrations in the 990141B tumor sample with *EGFR,* and the UNC040182B tumor sample and *BCL11A,* may cause mRNA mis-processing because the micro-aberration alters the DNA sequence in such a way that splice-site junctions are altered and we are able to observe inter-exon mRNA-seq reads. Using targeted *de novo* assembly of 990141B mRNA-seq data, we are able to generate contigs that suggest an in-place tandem duplication of a region of *EGFR* that may be the cause of the micro- amplification detected in the tiling array and the source of expression disruption. When examining data on the UNC040182B tumor sample, we are able to produce a contig that covers only the latter portion of the affected exon, but extends out into intronic space. The micro-deletion located in this region may knock out the initial portion of the exon, resulting in the production of the observed aberrant mRNA. We also found that the 5′ end of genes tended to be the most heavily affected by micro-aberration events, specifically the promoter region of the gene. It is unclear what leads to this predilection, but it does suggest that this specific portion of the gene may be more prone to this type of genomic alteration. The involvement of the promoter region does suggest that the site of active transcriptional processing may lead to a structural genomic weakness that causes a predisposition towards micro-aberrations. This finding may also aid in explaining either the factors involved in the formation of micro-aberrations versus macro-aberrations or the possible downstream consequences of such events. A limitation of our study is that we are currently unable to determine if any of these micro-aberrations are subtype specific. If they are, this would 1) mirror whole genome study findings, and 2) potentially showcase an alternative means of gene disruption that unravels previously unexplained expression data. As one example, RB1 dysfunction has been shown to be associated with the basal-like subtype. We have found in our own studies, that RB1-LOH was highly correlated with gene expression subtype and patient outcomes, while RB1 protein expression on the same samples was not. An intra-genic micro-aberration could potentially explain such cases, as the genetic abnormality may only affect a portion of the gene such that a protein is still produced and the majority of it intact, but it does not function properly. Expanded studies with high-resolution platforms like whole genome sequencing will allow us to answer this question with increased precision. Future studies with whole-genome sequencing technology and data are also an additional avenue by which these micro-aberrations can be validated, detected, and further defined. Varying amounts of sequencing depth would, however, be needed depending on the nature of the micro-aberration. Micro-deletions should be comparatively easier to detect as they are due to the absence of DNA, thus reasonably low coverage should be sufficient to call these events. Micro- amplifications would be more complex, especially if one did not know in advance what the nature of the micro-amplification was. Thus, in order to reliably identify all micro-aberrations, one would need sequencing depth capable of performing a *de novo* whole genome assembly. Targeted assembly is able to yield some insights, but full assembly would undoubtedly generate more answers. From a broader viewpoint, it stands to reason that other tumor types may also exhibit these types of events, but as yet they have not been widely described. However, similar intragenic deletions in *RB1* and *PTEN* were recently described in melanoma cell lines (SKMEL-207, A2058, and SKMEL-178), suggesting that these types of micro-genomic events are present in other cancers. In examining the specific genes on the tiling array that displayed micro- aberrations, we noted that genes that were involved in interphase of the mitotic cell cycle were particularly prone to these small-scale events. Given that our panel of targeted genes was focused on cancer relevant genes, there was some inherent enrichment for genes of this Gene Ontology class, but even within the background of genes on the array itself, there was a statistically significant enrichment for cell cycle genes. Coupled with our finding of micro-aberrations being localized to promoter regions, we speculate that cell cycle genes are more prone to these events because of their consistent and often high level of transcriptional activity. We were also able to make the observation that higher genomic instability in the form of micro-aberrations in our dataset was associated with worse survival outcomes. An overall high level of genomic instability has been found to associate with worse survival, and here we see that finding extended to micro- instability in our dataset. There was however, a high concordance of overlap between patients that had many large scale changes and many micro-aberrations, and overall, patients with higher total numbers of CNA were associated with poorer outcomes (data not shown). Nonetheless, the same rank-ordering split that was performed on the micro-aberrations did not result in identical findings for overall aberrations, suggesting that while there could be confounding of the micro-aberration survival findings by overall genomic instability, there may also be characteristics unique to the micro-aberrations themselves. Furthermore, the concordance between gross aberrations and micro-aberrations suggests that there may be common mechanisms of genomic instability at play, which may yield insights into how micro-aberrations arise. ## Conclusions In addition to exhibiting gross copy number changes, breast tumor genomes contain focal micro-aberrations as well when examined using a high-resolution platform. These micro-aberrations occur within the background of global genomic instability and can have disruptive effects upon gene expression. These micro- events represent a potential means of mutagenesis in genes that have been otherwise determined to be normal in terms of gross copy number or SNV-based somatic mutations. Continued investigation into these events with improved tools will allow their increased detection and likely highlight their importance as an additional means of altering gene function. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: CMP HHC. Performed the experiments: HHC XH. Analyzed the data: HHC WZ. Contributed reagents/materials/analysis tools: JSP WZ. Wrote the paper: HHC CMP.

# Introduction Preterm neonates are vulnerable to lung injuries, especially when they are affected by respiratory distress syndrome (RDS) and mechanically ventilated. Because of rapid changes in lung mechanics after surfactant therapy, lung injury and abnormal or fluctuating carbon dioxide levels may occur if the ventilator setting is not adjusted immediately. Thus, continuous monitoring of the adequacy of breathing and oxygenation is necessary. Although pulse oximetry is widely used as a noninvasive method for continuous monitoring, oxygen saturation may be normal even if there is inadequate ventilation. Previous studies have indicated that both low and high partial pressures of arterial carbon dioxide (PaCO₂₎ are associated with long-term morbidity in preterm and term infants. In addition, fluctuating PaCO₂ may lead to lung and brain damage, and is associated with retinopathy of prematurity. Mainstream measurement of the partial pressure of end-tidal carbon dioxide (PetCO₂) is a continuous and noninvasive method to measure blood carbon dioxide tension using with real-time CO₂ waveforms and numerical values immediately displayed on a monitor. PetCO₂ has several advantages, such as reduced arterial blood sampling frequency. It also provides a means for the continuous assessment of ventilation without accompanying iatrogenic anemia and is cost-effective. There is a gradient between PetCO₂ and PaCO₂ (P(a-et)CO₂), which can be determined based on the relationship between ventilation (V), which is airflow to the alveoli, and perfusion (Q_A), which is blood flow to the pulmonary capillaries. On average, the typical V/Q_A is 0.8 and PetCO₂ is normally 2–5 mmHg lower than PaCO₂, as the mixing volume is diluted in the conducting airways and ends at the alveolar compartment dioxide from the anatomical dead space. V/Q_A mismatch occurs due to heterogeneity in the ratio of ventilation to blood flow in different lung units. Areas of the lung that are perfused but not ventilated are said to possess a shunt. Any physiological perturbation that leads to low blood flow levels relative to ventilation in the alveoli increases physiologic dead space and leads to increased P(a-et)CO₂. P(a-et)CO₂ may be caused by shallow breathing, over-inflation of the lung and other cardiac or respiratory pathologies. However, earlier studies examining the effects of changes in dead space to tidal volume ratios (V_D/V_T) on PetCO₂ and PaCO₂ in newborn infant are scant. The purpose of this study was to evaluate the effects of different V_D/V_T on the correlation between PetCO₂ and PaCO₂ in ventilated preterm infants with RDS before and after surfactant therapy. We hypothesized that the difference between PetCO₂ and PaCO₂ in ventilated preterm infants with RDS after surfactant therapy will decrease due to the decrease in V_D/V_T. # Materials and methods ## Patient population This single-center, prospective, non-randomized, consecutive enrollment study was approved by the Institutional Review Board of Chang Gung Memorial Hospital in Taoyuan, Taiwan. Preterm infants with RDS who were admitted to the neonatal intensive care unit (NICU) at Chang Gung Memorial Hospital and treated with survanta (berectant, bovine-derived natural surfactant, AbbVie) between May 2013 and December 2014 were enrolled. Informed consent was obtained from the parents or legal guardians of the patients. Patients with structural cardiopulmonary malformation, those undergoing high-frequency ventilation, and those requiring extracorporeal membrane oxygenation were excluded from the study. The diagnosis of RDS was made based on the classical radiographic appearance, clinical evidence of respiratory distress, laboratory abnormalities due to impaired gas exchange, and the requirement of respiratory support. Surfactant was administered at a dosage of 100 mg/kg, and was divided into 4 quarters following the manufacturer’s recommendation when patients failed to maintain saturations in the normal range when FiO₂ was \>0.4. A second dose of surfactant may be administered if required at least 6 hours after the preceding dose. The patients were ventilated using pressure-limited, time-cycled ventilators in either assist control mode or synchronous intermittent mandatory ventilation mode. The mechanical ventilators (Babylog 8000 Plus, Dräger Medical) were equipped with basic airway graphic monitors and were calibrated following the manufacturer’s recommendations. The initial settings of the ventilator, which were determined using a standard NICU protocol, included a starting respiratory rate of 20 to 40 breaths per minute (bpm) used to maintain a pH of 7.22 to 7.35 and a PaCO₂ of 40 to 60 mmHg, a peak inspiratory pressure (PIP) of 15 to 25 cmH₂O, a tidal volume of 4 to 6 ml/kg to produce adequate chest-wall movement, a positive end expiratory pressure (PEEP) of 4 to 6 cmH₂O to maintain adequate lung expansion, and FiO₂ adjusted to maintain arterial partial pressure of oxygen (PaO₂) of 60 to 80mmHg. Infants with very low birth weight (VLBW) whose birth weights were less than 1,500 g were intubated with size 2.5 mm or 3.0 mm endotracheal tubes without cuffs. Non-VLBW (NVLBW) infants whose birth weights were between 1,500 and 2,499 g were intubated using size 3.0 mm or 3.5 mm endotracheal tubes without cuffs. ## Blood sampling The sampling of arterial blood gas (ABG) was carried out before and 1 hour after surfactant administration, and at 24 hours of age during routine medical care. ABG was measured mainly at the umbilicus arterial catheter, although it was measured at peripheral arteries if the umbilicus arterial catheter was not available. Blood gas determination was performed using a blood gas analyzer (Siemens Rapidlab 248 Blood Gas Analyzer). ## End-tidal carbon dioxide monitoring PetCO₂ was continuously monitored using mainstream capnography (Philips M2501A Mainstream Capnography). Since the dead space of the sensors and response times may result in false interpretations of PetCO₂ readings, the sensor was designed for infants with \<1 ml of dead space and rise times \<60 ms. The infant-type airway adaptor was placed between the endotracheal tube and the Y connection of the ventilator circuit. The capnography device was calibrated according to the manufacturer’s instructions. The sensor for PetCO₂ was placed prior to blood sampling at each time point. We ensured that the waveform of PetCO₂ was continuous and steady by measuring expired CO₂ throughout the ventilator cycle. This allowed us to obtain simultaneous PetCO₂ and PaCO₂ measurements. P(a-et)CO₂ was recorded along with additional data including the mode of ventilation, tidal volumes, PIP, PEEP, total respiratory rate, mean airway pressure (MAP), oxygenation index (FiO₂ x MAP/PaO₂), PaO₂/FiO₂ ratio, oxygen saturation, blood pressure, and demographic details. ## Dead space to tidal volume ratio (V_D/V_T) V_D/V_T was calculated using the Enghoff modification of the Bohr equation: V_D/V_T = (PaCO2 –PetCO₂) /PaCO₂. ## Statistical analysis Continuous data are expressed as mean ± standard deviation. Statistically significant differences were defined using *P* \< 0.05. P(a-et)CO₂ was assessed using the Bland-Altman technique. The precision of PetCO₂ and the relationship between PetCO₂ and PaCO₂ in various clinical situations was evaluated using Pearson’s correlation coefficients and analyzed using the Statistical Package for the Social Sciences (version 19.0 software). Categorical variables were assessed using chi-square tests. Analyses of variables were performed using independent t tests, while comparisons between the pre-surfactant treatment and post- surfactant treatment groups were carried out using paired t tests. When we compared the parameters according to the treatment stage of surfactant therapy, only the first dose of surfactant was considered. # Results One-hundred and one PetCO₂ and PaCO₂ pairs were analyzed from 34 neonates who required ventilation due to RDS and were treated with surfactant. The ventilator parameters were calculated according to the first admission sample and were as follows: mean total respiratory rate (53.8 ± 10.5 bpm), mean tidal volume (5.9 ± 0.2 ml), mean ventilation volume per minute (0.4 ± 0.2 L/min.), mean PIP (16.8 ± 2.5 cmH₂O), mean MAP (9.3 ± 1.2 cmH₂O), mean PEEP (5.1 ± 0.4 cmH₂O), and mean FiO₂ (40.1 ± 10.5%). Sixteen of the infants were NVLBW (mean gestational age 32.3 ± 1.9 weeks and birth weight 1,967 ± 316.5 g). Eighteen infants were VLBW infants (mean gestational age 28.3 ± 1.8 weeks and birth weight 1,084.6 ± 242.6 g). One- hundred and one paired samples (53 from VLBW infants and 48 from NVLBW infants) were used for analysis. The descriptive characteristics of the enrolled patients are depicted in. There was a significant difference in antenatal corticosteroid use (72.2% vs. 25%, *P* \< 0.001) between the VLBW and NVLBW groups. The incidence of bronchopulmonary dysplasia (44.4% vs.25%, *P* = 0.253) and that of patent ductus arteriosus (50% vs.25%, *P* = 0.134) were not different between VLBW and NVLBW groups, as shown in. We analyzed difference between patients receiving surfactant before vs. after therapy according to the first dose of surfactant. There was a significant change in V_D/V_T, in the post-surfactant treatment group when compared to the pre-surfactant treatment group (*P* = 0.003). The correlation was higher in the post-surfactant treatment group (r = 0.786, *P* \< 0.01) than in the pre-surfactant treatment group (r = 0.235). A significant change in PaCO₂ (42.4 ± 8.6 mmHg vs. 37.8 ± 10.3 mmHg, *P* = 0.018) and P(a-et)CO₂ (9.8 ± 9.9 mmHg vs. 4.1 ± 6.5 mmHg, *P* = 0.004) was noted between pre-surfactant and post-surfactant treatment. When considering the overall sample data, we found a moderate correlation (r = 0.603, *P* \< 0.01) between PetCO₂ and PaCO₂. The mean P(a-et)CO₂ was 5.9 ± 7.6 mmHg. Bland- Altman plots of the comparison of the mean versus the difference in values between PaCO₂ and PetCO₂ are shown in. A scattergram plot of the PetCO₂-PaCO₂ relationship is shown in. # Discussion In this study, we performed mainstream capnography in infants with RDS who were treated with surfactant. We found that there was moderate correlation, but poor agreement, between PetCO₂ and PaCO₂. Some researchers argue that PetCO₂ may not accurately predict PaCO₂. Watkins et al. have reported poor correlation between PetCO₂ and PaCO₂ in 19 infants with pulmonary disease. Garcia Canto et al. also reported that PetCO₂ did not have a good correlation with PaCO₂ in 9 ventilated newborns with severe lung illnesses. More recently, Javier et al. reported that there was larger bias and higher precision between PetCO₂ and PaCO₂ than between PaCO₂ and transcutaneous CO₂. This negative result may have been due to the fact that some samples were obtained from babies who were diagnosed with heart failure, and that the response time for the PetCO₂ reading (\<150 ms) was much longer than normal (\<60 ms). In contrast, Wu et al. observed a higher correlation (r = 0.818, *P* \< 0.001) between PetCO₂ and PaCO₂ in 61 infants. In 2012, Daniele et al. reported a positive correlation (r = 0.69, *P* \< 0.0001) between PetCO₂ and PaCO₂ in 45 infants with VLBW. Most previous studies of PetCO₂ measurements have not considered the severity of lung diseases. Recently, Bhat et al. reported the correlation between PetCO₂ and PaCO₂ in a post-surfactant replacement therapy group and concluded that it was more accurate than that in a pre- surfactant replacement therapy group. Similarly, we found a higher correlation between PetCO₂ and PaCO₂ in the post-surfactant replacement therapy group than the pre-surfactant therapy group. Furthermore, our results showed that V_D/V_T was decreased significantly after surfactant therapy and that the correlation between PetCO₂ and PaCO₂ was higher after surfactant therapy. Based on our finding that the correlation between PetCO₂ and PaCO₂ was higher after surfactant therapy, we speculated that our observations may be due to the fact that lung regions with both high and low V_A/Q can occur simultaneously in patients with RDS, while V_D/V_T decreases and the oxygenation index is improved after surfactant therapy. McSwain et al. reported that the correlation between PetCO₂ and PaCO₂ improved significantly in patients admitted to the pediatric intensive care unit with lower V_D/V_T (\<0.4). Bindya et al. also reported sidestream PetCO₂ monitoring provided a more accurate reflection of the PaCO₂ in patients with lower V_D/V_T (\<0.3). Therefore, PetCO₂ may be more accurate in post-surfactant treated infants because of the improvement in V_D/V_T. Whether sidestream or mainstream PetCO₂ monitoring is more accurate and suitable for neonates is still controversial. Instead of sidestream PetCO₂ monitoring, we used mainstream PetCO₂ monitoring in this study and made similar observation in infants with significant improvements in the PetCO₂/PaCO₂ correlation when V_D/V_T was decreased. This study had some limitations. First, the rate of exposure to antenatal corticosteroids was low in the current study. Only 50% of the patients had received antenatal corticosteroids. However, 72.2% of infants with VLBW received antenatal corticosteroids. Second, we did not measure pulmonary mechanical parameters, such as respiratory resistance and dynamic compliance. Evaluation of these parameters may have been helpful in understanding how physiological abnormalities affect the correlation between PaCO₂ and PetCO₂. # Conclusions This study was the first to explore the effects of different V_D/V_T values on the correlation between PetCO₂ and PaCO₂ in ventilated preterm infants with RDS before and after surfactant therapy. Since ABG analysis is not suitable for the collection of continuous data and the observance of trends, more long-term follow-up studies are required to validate the usefulness of PetCO₂ for monitoring and evaluating the response to respiratory therapies. The study was supported partly by Chang Gung Memorial Hospital (CMRPG3D1021). The authors gratefully acknowledge the Center for Big Data Analytics and Statistics at Chang Gung Memorial Hospital for statistical assistance. All of the respiratory therapists and the nursing staff in the neonatal intensive care units do an outstanding job of taking care of the patients and helped to complete this study. [^1]: The authors have declared that no competing interests exist. [^2]: ‡ These authors also contributed equally to this work.

# Introduction In 2012 it was estimated that 35.3 million people were living with HIV and that there were 2.3 million new infections during that year. Discovering ways to prevent the transmission of HIV is of primary concern to health care authorities worldwide. It is well known from a range of observational and epidemiological studies that the risk of acquiring HIV among heterosexual males can be significantly reduced by 60% through safe male circumcision (SMC). Numerous papers on the topic have been published over the past two decades to elevate HIV prevention awareness, especially in sub-Saharan countries. A modeling study published in 2009 estimated that scaling up SMC to reach 80% of adult males in 14 African countries by 2015 could potentially avert more than 4 million adult HIV infections between 2009 and 2025 and yield annual cost savings of US\$1.4–1.8 billion after 2015, with a total net savings of US\$20.2 billion between 2009 and 2025. To date, there are over 38 million adolescent and adult males in Africa who could benefit from SMC for HIV prevention. The challenge Africa faces is how to safely scale up a surgical procedure in resource limited settings. Uganda has a national plan to offer a voluntary SMC program to 4.2 million adult men over 5 years as part of a comprehensive HIV prevention strategy. To achieve this, a minimum of 820,000 procedures need to be performed per year, however over the past 18 months, this target has not been met, falling short by 250,000 procedures. Approaches that involve quicker but equally safe or safer methods are urgently needed to realize scale up and hopefully reach the set targets. In Uganda, one PrePex pilot study has been published and in addition an active surveillance exercise is under way at four sites. In this paper we describe the skills transfer process for a new non- surgical male circumcision device PrePex at a Ugandan SMC site. # Methods ## Ethical consideration This study obtained approval from the Makerere School of Medicine Research and Ethics Committee and the Uganda National Council of Science and Technology. If eligible, and after the subject signed the written informed consent form (one- on-one with the principal investigator or designee), he was enrolled into the study. ## Study methods The training study took place in the context of a prospective study of PrePex safety when used at an urban SMC site, (International Hospital Kampala) conducted from August to October 2012. A total of 625 subjects were eligible for device placement and removal. Those enrolled were males scheduled to undergo voluntary SMC in an effort to prevent the spread of HIV in resource limited high prevalence settings. Duration of the training period was 3 days and the entire period of the study was eight weeks. ## The training model An eligible trainee was a SMC certified health worker with prior surgical male circumcision experience of at least 50 SMC cases and working as the primary ‘surgeon’. Teaching methods included a seminar, ‘dry’ laboratory practice sessions on models and hands-on practice. Materials available for use were: a PrePex video clip. ([www.prepex.com/clinical](http://www.prepex.com/clinical) procedure.aspx), pictorial paper charts, PrePex demonstration kits, standardized adverse events (AEs) definitions and Standard Operating Procedures (SoPs) for AE management. Assessment and feedback was conducted in real time through face-to- face sessions between the Prepex master and the trainee. During the assessment of performance, reference was made to the steps (tasks) as indicated in and. and show some of the tools used. The Training Model contained two parts. ### Part I– Didactic and Simulation sessions These were conducted in one large room; with one side lecture room, with chairs, a flip charts, LCP Projector and white board screen. Three lecturers (2 PrePex supervisors and 1 product specialist) conducted the lectures and demonstrations. The following were covered in these sessions: PrePex device, PrePex procedure, tools and materials, male genital anatomy, PrePex screening procedure, PrePex removal procedure, managing client flow, the post procedure healing course, the possible side effects and adverse events. shows a sample of the training plan used. Space to accommodate 3 work stations was set up in the other half of the room. One station had a wooden mannequin for the male genitalia, the second station had PrePex device for placement, and the third had PrePex materials for removal. An MCQ examination was conducted at the end of the session. Feedback was given to each participant ### Part II- Clinical This was conducted over two days, included screening placement and removal. The time between placement and removal was 6 days (i.e. removal was done on day 7). Ten providers (3 physicians, 2 clinical officers and 5 nurses) were assigned to five training teams. Three PrePex masters trained the teams. A Clinical Officer cadre is equivalent to a Physician Assistant Cadre. Of the 625 eligible men for device placement and removal, 40 were assigned to each team. ## Assessment of Competence The first 20 procedures for each team were closely monitored and tutored by a PrePex master from Rwanda experienced in performing PrePex procedures. Each trainee was assessed for PrePex knowledge and skills using a predetermined criterion. Testing of clinical competence which allowed decisions to be made about fitness to perform the procedure (practice) by the trainee, was based on directly observing the individual steps performed correctly. Judgment of whether the steps were correctly performed was a global rating (a point-scale of either needs to improve or competently performed procedure in proper sequence and progressed from the step to step efficiently). Feedback was given instant for either correctly done or not correctly done and requiring a repeat (formative assessment). At the end of the training a certificate was issued to each individual to certify competence. She/he began performing placement and removal of the devices independently. The procedure steps for placement and removal are outlined in and, respectively and in the supplementary files. ## Data Collection Rate of adverse events, time taken to perform procedures and number of device placements and removals were collected for each trainee during the eight weeks when devices were placed and removed. A questionnaire was used to collect training outcome data, and these data were analysed for frequencies and trends. # Results In total, 625 placements and removals were performed by trainees and trainers. Those trained included five nurses, two clinical officers, three physicians as shown in. All the trainees passed. After the 3 days of training, all trained workers performed procedures during the rest of the study/project period (561 placements and 529 device removals). The majority of procedures were performed by nurses and clinical officers. There was one AE for every 38 placements by a nurse and 1 AE for every 72 placements by a clinical officer and none for surgeons and the trainers. There was one painful (defined as pain score ≥8) removal for every 4 removals by a nurse, 1 in 11 removals by a clinical officer, 1 in 5 by a specialist surgeon and 1 in 2 by the trainers. The 10 trainees were enrolled trained and performed procedures over the eight- week enrollment period of the study. There were some differences in AEs among physicians compared with non-physicians in using PrePex during the 8 weeks. All the AEs resolved without sequel. As shown in, some ‘difficult’ situations were encountered while performing placements; a borderline or narrow prepuce leads to inner ring insertion difficulties. The elastic ring would repeatedly disengage from the inner ring groove and in a separate incident a client had repeated frequent erections, which made device placement impossible. # Discussion This paper describes a rapid training model for the safe use a of non-surgical circumcision device (PrePex). All 10 trainees were competent in carrying out surgical safe male circumcision. The PrePex screening, placement, removal and counseling skills were mastered with relative ease. AE management was within the capability (competence) of the trainees as they had prior surgical SMC experience. AE rates occurring when nurses performed the procedures were twice as high as when clinical officers performed the procedures. The physicians posted no AEs, perhaps because the numbers of procedures they performed were less than for the nurses. The moderate AE rates of 1 in every 39 clients (2.6%) is close to or within the generally acceptable AE rate for SMC of 2–5%, suggesting that nurses were safe operators for this device. The p- value was \>0.25 for differences in moderate AEs between nurses and clinical officers. The occurrence of pain during removal on day 7 among some of the participants is undesirable even if it is short lived pain (less than a few seconds). In this study, one in six experienced short lived pain ≥8 (on the VAS) on removal. The trainers had a much higher rate (1 in 2) compared to the rest of the operators; possibly because they performed far fewer removals, they removed only six. So this could be a chance occurrence. Although the reduction or elimination of pain occurrence during device removal may be achieved through practice (experience), there is room to explore other factors that might contribute such as removal, prior counseling, analgesia, etc. The desirability for short rapid training fits in with a move towards shorter training periods for surgical or procedural skills transfer and emphasis on operating room efficiency. In these trainings, sheer volume of work/procedure exposure rather than specifically designed curricula, is the hallmark of surgical training and skills transfer. The process of new skills training techniques is based on established theories of the ways in which motor skills are acquired and expertise is developed. Fitts and Posner's three-stage theory of motor skill acquisition is widely accepted in the motor skills and surgical literature. In the cognitive stage (the first stage), the learner intellectualizes the task; at this stage performance is erratic, and the procedure is carried out in distinct steps. With practice and feedback, the learner reaches the integrative stage (the second stage) in which knowledge is translated into appropriate motor behaviour. The learner is still thinking about how to move the hands and execute the task with fewer interruptions. In the autonomous stage, (the third stage), practice gradually results in smooth performance. The learner no longer needs to think about how to execute this particular task and can concentrate on other aspects of the procedure. These processes and stages were observed in the 3 days of training and the 7 weeks of observation in this study. We contend therefore that AEs are likely to reduce the more procedures one does. The duration of time in a profession does not necessarily lead to the development of expertise. It is well reported that development of expertise is dependent on deliberate efforts to change important aspects of performance rather than repetitive execution of routine work. In order to acquire expertise, practice should be challenging in relation to its level of difficulty, informative due to the availability of feedback and repetitive with an opportunity to detect and correct errors. The tasks encountered by our trainees were challenging and of interest in the sense that the device was a new concept, the trainees had no prior experience of the device though the technical concepts underpinning device circumcision are the same as for surgical SMC. The difficulties pointed out in highlight some of the challenges encountered. In the context of limited resources, methods for expediting the pathway to expert performance are essential. Anecdotal evidence traditionally attributes development of expertise to experience accumulated. Mere accumulation may not be sufficient for one to become an expert. For some, performance may decrease after training and there are numerous instances where experience and the amount of knowledge and performance are incongruent. Apart from sheer experience and knowledge, other attributes such as behaviour traits, learning styles and environments conducive for the development of expertise should not be lost or left out. Behaviour traits and learning styles were not individually considered in this study though the environment was conducive for learning, was spacious, well-lit, with comfortable ambient temperatures, free of noise, private and non- threatening. Deliberate practice relates to activities in which learners engage with the specific aim of improving some aspect of performance. In this training model, trainers facilitated trainee progression by creating an environment where the trainees could perform repetitive behaviors and receive feedback and instruction to ensure development of key skills. The steps for placement and removal were clearly stated and it was possible to be instructed upon accordingly. The steps were not complex and repetition was possible. Repeated practice is believed to aid mastery. Assessment was objective, and used a checklist similar to the one used for OSCE (Objective Structured Clinical Examination). Incorrect placement or removal was apparent before the client would leave the room and therefore there were immediate feedback opportunities. In the context of SMC, PrePex skills transfer training is feasible after a short period of didactic lessons and with on job training for all eligible health workers or practitioners. WHO recently issued guidelines on the use of male circumcision devices. These guidelines derive from studies conducted at several sites in different countries. # Study Limitations This study only included those operators that had significant prior surgical SMC experience; extrapolation to those without surgical SMC experience may be done with caution. What this study does not explore is when retraining may be required. In the current context there may be time lapses between training in a research setting to routine PrePex practice. # Conclusions PrePex device skills are relatively easy to transfer to non-physician and physician cadres in a short duration in resource limited settings, but among those with past SMC experience. # Supporting Information International Medical Group management, all clients, and SMC staff. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MG. Performed the experiments: MG JPB. Analyzed the data: MG KD. Contributed reagents/materials/analysis tools: JPB KD MG. Wrote the paper: MG KD JPB NW.

# Introduction The *Gadd45* genes are a family of stress response genes, which are involved in diverse processes, including cell growth, DNA repair, and apoptosis, and function as tumor- and autoimmune suppressors. Expression of these genes is induced by DNA-damage and genotoxic stress, including hyperosmotic stress and UV irradiation. The three *Gadd45* genes encode multifunctional, 18 kDa acidic proteins, which can homo- and heterodimerize and which are predominantly localized in the nucleus. Gadd45 proteins interact with many effectors, including Cdc2/CyclinB1 , PCNA, p21, nuclear hormone receptors, histones and MEKK4, to mediate cell cycle arrest, differentiation or apoptosis. More recently Gadd45 proteins have been implicated in epigenetic gene regulation, promoting active DNA demethylation via a DNA repair mechanism. Gadd45a binds to the repair endonuclease XPG and initiates excision repair at methylated CpG motifs in *Xenopus*, *Zebrafish*, and mammalian cells. Gadd45 proteins exhibit sequence homology to the L7Ae/L30e/S12e superfamily. Members of this family are diverse proteins from archea, eubacteria and eucaryota, including ribosomal proteins (S12, L30e), proteins that bind guiding RNA (L7Ae, 15.5 kD, fibrillarin), as well as components of ribonuclease P. Many of these proteins bind functionally diverse RNAs, including ribosomal RNA, snoRNA, snRNA and mRNA. Rather than binding to a specific consensus sequence, these proteins recognize a common structural motif – the kink turn, formed by both canonical Watson-Crick base pairing as well as and non-canonical interactions. The fact that Gadd45 proteins belong to the L7Ae/L30e/S12e superfamily raises the question whether they may also bind RNA. Importantly, RNAs have been repeatedly implicated in active DNA demethylation although their history in this process is confusing. Most recently ROS3 has been described as an essential mediator of DNA demethylation in *Arabidopsis*. ROS3 resides in nuclear speckle- like structures and binds small RNAs. It was suggested that these RNAs may guide the DNA demethylase towards their substrate. Gadd45a has been shown to associate with chromatin, however, it is unknown whether it directly interacts with nucleic acids. Here we provide evidence that Gadd45a has RNA binding properties and possesses characteristics of a ribonucleoprotein particle (RNP). # Methods ## Expression constructs and antibodies For *Xenopus tropicalis* (xt) Gadd45a overexpression in human cells and *E.coli* we used constructs containing xtGadd45a ORF in vectors pRKW2 and pET28a as well as N-EGFP tagged xtGadd45a in pCS2. Point mutants of xtGadd45a, were obtained by circular PCR. The following antibodies were used: anti-hGadd45a (H165), anti-p68 (H144), anti-Brg1 (N-15) (Santa Cruz), anti-hnRNP A1 and anti-histone H3 (Abcam), anti-GFP (Dianova), anti-SC35 (Novus Biologicals). ## Cell culture and transfections HEK293T cells (ATCC CRL 11268) and RKO cells (ATCC CRL 2577) were grown at 37°C in 10% CO2 for 293T cells and 5% for RKO cells in Dulbecco's Modified Eagle's Medium (DMEM), 10% fetal calf serum, 2 mM L-Glutamine, 100 U/ml penicillin and 100 µg/ml streptomycin. Transient DNA transfections were carried out using FuGENE6 (Roche), TurboFect™ (Fermentas) in case of HEK293T, and for RKO cells a combination of Lipofectamine and Plus reagents (Invitrogen) was used following the manufacturer's instructions. ## Immunofluorescence microscopy Cell detergent extraction and RNase treatment was essentially as described. For detection of overexpressed Gadd45a (wild type and G39A mutant) RKO cells were transfected with pCS2-EGFP-Gadd45a. 24 h after transfection cells were extracted with 0.05% Triton X-100, washed with Hank's BSS 1x buffer (PAA laboratories GmbH); with or without subsequent treatement with RNase A for 7 min with (Roche Applied Science, 1 mg/ml). Cells were fixed with Dithiobis (succinimidyl propionate) (DSP) (Thermo Scientific) according to the manufacturer. Briefly, immediately before fixation DSP was added to a final concentration of 0.5 mM in 100 mM Hepes (pH 7.4) in Hank's buffer. Cells were fixed in the freshly prepared DSP solution for 90 minutes and then incubated for 30 minutes in quenching solution (50 mM monoethanolamin, 0.1% Triton X100 in Hank's buffer). Immunostaining was performed with anti-SC35 (Novus Biologicals) and anti-p68 (Santa Cruz Biotechnology) antibodies; immunofluorescent images were recorded on a Nikon confocal microscope. For statistical analysis, the nuclear pattern of EGFP-Gadd45a was assessed manually (n = 35–50). For detection of endogenous Gadd45a HEK293T cells grown on coverslips and subjected to UV irradiation (40 mJ/cm2) were permeabilized with 0.2% Triton X-100 in 20 mM Tris-HCl (pH 7.4), 5 mM MgCl₂, 0.5 mM EDTA, and 25% glycerol; with or without subsequent RNase treatment as described above. Cells were fixed with 2% formaldehyde and antigen retrieval was performed as described, except that microwave treatment was done at 450W. Immunofluorescence was performed using anti-Gadd45a H165 antibody (Santa Cruz). Statistical analysis of the endogenous Gadd45 nuclear patterns was performed manually (n = 50). ## Nuclear extract preparation HEK293T or RKO cells were harvested, washed twice with DPBS buffer, and homogenized in buffer A (0.3 M sucrose, 10 mM Tris-HCl pH 8.0, 3 mM CaCl₂, 2.5 mM Magnesium acetate, 0.25% Triton X-100, 0.1 mM EDTA, 2 mM DTT, Complete™ proteinase inhibitors (Roche)) in a douncer homogeniser. Homogenates were mixed 1∶1 with buffer B (1.8 M sucrose, 10 mM Tris-HCl pH 8.0, 3 mM CaCl₂, 2.5 mM magnesium acetate, 0.1% Triton X-100, 0.3 mM EDTA, 2 mM DTT, CompleteTM proteinase inhibitors (Roche)), and centrifuged through a buffer B cushion for 20 min at 13,000 g in a Beckman SW41 Ti rotor. The pellet was resuspended in nuclear extraction buffer (50 mM Hepes-NaOH (pH 7.8), 140 mM NaCl, 1 mM MgCl₂, 1 mM DTT, 0.1 mM Vanadyl ribonucleosides (Sigma), proteinase inhibitors (Roche) and sonicated. Nuclear extract was centrifuged at 15000 g for 30 min, and the supernatant was used for further procedures. ## Sucrose gradient sedimentation analysis Sucrose gradient sedimentation analysis was performed using nuclear extract from 2×10⁷ RKO cells. Samples were untreated or treated with 100 µg/ml of ribonuclease A (Roche) or 40 U/ml of DNase I (MBI) for 30 min at room temperature. Soluble nuclear proteins were applied to the top of a 8–40% sucrose gradient and centrifuged for 26 h at 50000 g at 4°C. Samples containing sedimentation markers thyroglobulin (19S), β-galactosidase (16.4 S), catalase (11S) or a cytoplasmic fraction containing ribosomal subunits were run separately. Proteins from gradient fractions were precipitated and analyzed on immunoblots. ## Filter binding assay Filter binding assays were performed essentially as described. Recombinant proteins used were bovine serum albumin (Fraction V, Sigma), His-Gadd45a and M-MLV-reverse transcriptase (Invitrogen). Binding reactions were performed in RNA binding buffer (10 mM Tris-HCl (pH 7.4), 10 mM KCl, 1 mM MgCl₂). In reactions without competitor, 2.5 or 1.3 µM recombinant proteins were preincubated for 20 min with 10,000 cpm (approximately 2 ng) of ³²P-labelled multiple cloning site (MCS) RNA of pCS2 and pXT1 plasmids. In competition assays, unlabeled competitor nucleic acids were preincubated for 10 min with recombinant proteins before addition of labeled RNA. Reactions were applied to nitrocellulose filters that were pre-blocked with 50 µg/ml BSA in RNA binding buffer, washed with RNA binding buffer and quantified by scintillation counting. ## Structural data files Apart from the crystal structures used as templates for xtGadd45a homology modelling, the crystal structures of human spliceosomal p15.5 kDa protein bound to a U4 snRNA fragment (PDB ID 1e7k), yeast L30e-mRNA complex (PDB ID 1t0k), *Haloarcula marismortui* ribosomal protein L7Ae-rRNA complex (PDB ID 1s72), yeast spliceosomal protein Snu13p dimer complex (PDB ID 1zwz) were used. For homology modeling the sequences of xtGadd45a (NCBI AN CAJ82672) and human SPB2 (NCBI AN NP_076982) were used. ## Homology modeling and structure analysis Two close homologues of xtGadd45a were used as templates for homology modeling: human (PDB ID 2wal, citation pending), and mouse (PDB ID 3cg6,) Gadd45g. Initial sequence alignments were generated in ClustalX2 and manually refined from 3D alignments of available crystal structures for hsp15.5, scL30e, hmL7Ae, scSnu13p, hsGadd45g and generated models of hsSBP2_RBD and xtGadd45a. Modeling of xtGadd45a was carried out as described, using the MODELLER package. In the same way a model of the xtGadd45a mutant G39A was generated. For modeling of the human SPB2 RNA-binding domain the crystal structure of yeast spliceosomal protein Snu13p (PDB ID 1zwz;) was used. Initial models were scored for energy content and sterical correctness and the best model further optimized using GROMACS molecular dynamics simulations was used. All models were scored for energy and sterical correctness using the ANOLEA, VERIFY_3D and ERRAT (<http://nihserver.mbi.ucla.edu>) online servers. Structure analysis was carried out using SwissPBD Viewer and PyMol ([www.pymol.org](http://www.pymol.org)). PDB2PQR, PropKa and APBS packages were used for charge surface calculations and the HotPatch web server for hydrophobicity calculations. For protein-protein dockings the GRAMM package was used in hydrophobic mode. ## Purification of recombinant his-xtGadd45a pET28a vectors containing ORF of xtGadd45a and mutants were transformed into BL21DE3 *E.coli*. Bacteria were grown in 100 ml medium and induction was performed by IPTG for 4 h. Disruption was performed by French Press, and the lysate was cleared by centrifugation at 110.000 g for 30 min. Purification was performed by metal affinity chtomatography under native conditions on a Ni-NTA column (Qiagen) according to the manufacturer's instruction except that lysis buffer contained 1 mM of MgCl₂ and 0.04% of NP40. Step elution with 60–100 mM imidazole was performed. Fractions were analyzed by SDS-PAGE with Coomassie staining (Thermo Scientific), and fractions eluting at 100 mM imidazole were dialyzed against lysis buffer and used for further experiments. ## Luciferase reporter assay Dual-Luciferase reporter assays (Promega) were performed as described in. ## Southern blot methylation analysis HEK293T cells were transiently transfected in 10 cm dishes with 5 µg HpaII *in vitro* methylated pOctTK-EGFP and 1.2 µg pBl-KS or xtGadd45a. Transfected plasmid DNA was recovered 72 h after transfection, digested with NotI and either HpaII or MspI and analyzed by Southern blot using a *GFP* probe. The expression of EGFP was additionally analyzed by SDS-PAGE and Western blot using anti-GFP antibody. # Results ## Gadd45a binds RNA *in vitro* and *in vivo* To test whether Gadd45a binds RNA, we carried out filter binding assays using recombinant Gadd45a and radiolabeled synthetic vector RNA, which indicated significant RNA binding compared with M-MLV reverse transcriptase. To further characterize nucleic acid binding, filter binding assays were performed by preloading Gadd45a with unlabeled nucleic acids followed by competition with labeled RNA from a plasmid multiple cloning site. Since Gadd45a is implicated in DNA demethylation, we tested methylated as well as unmethylated DNAs. Neither unmethylated, nor methylated single- nor double stranded DNA efficiently competed for RNA binding. Poly-uridine was the best competitor among RNA homopolymers. Other complex RNAs, including tRNA, vector derived RNA, and notably total cellular RNA, were also effective. To examine if Gadd45a is bound to RNA *in vivo*, we analyzed its sedimentation behaviour in sucrose density gradients. As a source we used RKO cells, which express Gadd45a endogenously. Interestingly, the majority of Gadd45a sedimented in the ribosome-sized fractions, with S-values between 40 and 60. Significantly, RNase treatment shifted Gadd45a to lighter fractions, suggesting that Gadd45a may be present in an RNP-like particle. Two RNA binding proteins, ribonucleoprotein hnRNP A1, a component of ribonucleoprotein (RNP) particles and RNA helicase p68, were analyzed as positive controls. The ATPase Brg1 served as negative control protein. It is part of a nucleosome remodelling complex and not thought to bind RNA. The sedimentation profile of hnRNP A1 was broad, with a peak in the ribosomal fractions, like for Gadd45a and consistent with it being part of RNP particles. In contrast, p68 showed a bimodal distribution, fractionating as a very heavy and a light form, as reported previously. Brg1 was recovered only in heavy fractions. DNase pretreatment showed minor alterations in the sedimentation behavior of the proteins but these were within the margins of sample variability. In contrast, RNAse pretreatment led to a reproducible shift in sedimentation of both RNA binding proteins hnRNPA1 and p68 to the light fraction, while it did not affect Brg1, indicating that the effect was indeed due to RNase and not contaminating protease. The RNase sensitive sedimentaion profile of Gadd45a supports it being associated with RNA endogenously. ## Gadd45a localizes in nuclear speckles in an RNase sensitive manner To further examine whether Gadd45a is a RNA binding protein we analyzed the RNase sensitivity of its localization. Cells were transfected with Gadd45a and soluble proteins were detergent extracted with or without RNase treatment and analyzed by Western blot or immunofluorescence (IF) microscopy. In Western blot analysis Gadd45a is removed from the detergent-resistant fraction upon RNase treatment. IF analysis of detergent-extracted RKO cells showed that EGFP-Gadd45a is localized in nuclear speckles, as described previously. There it was colocalized with the nuclear speckle markers SC35 and p68. Nuclear speckles are the main repository for factors involved in transcription elongation, mRNA processing and export. In some cells overexpressed Gadd45a strongly localized to the nuclear periphery. We tested if this localization is RNase sensitive. Indeed, RNase treatment reduced the number of cells where Gadd45a localized in nuclear speckles from 72% to 20%. In contrast, RNase treatment did not affect localization of Gadd45a in the nuclear periphery (not shown) as well as SC35 staining. In HEK293T cells endogenous Gadd45a showed only a weak, homogeneous nuclear signal. However, following UV-irradiation, which induces Gadd45a expression, the protein was colocalized with SC35, but was also found in SC35 negative punctae and in the nuclear periphery. Once again, RNaseA treatment removed Gadd45a from nuclear speckles. The nuclear speckle co-localization with RNP proteins SC35 and p68 and its RNase sensitivity support that Gadd45a is an RNA binding protein and may be part of an RNP. ## Modeling of Gadd45a-RNA binding To gain insight into the structural basis of Gadd45a-RNA interactions we inspected the three dimensional structures of Gadd45 as well as of structures of other L7Ae members, which were solved in complex with RNA. First, we built a homology model for *Xenopus tropicalis* Gadd45a by employing the available crystal structures of Gadd45g. Next we compared the sequences and structures of the three L7Ae protein family members: human spliceosomal p15.5 kDa protein bound to a U4 snRNA fragment, yeast L30e-mRNA complex, and *Haloarcula marismortui* ribosomal protein L7Ae-rRNA complex. All three L7Ae family proteins are bound to the so-called kink-turn RNA motif. Analyzing general rules of recognition of this type of RNA, we identified two main patches on the RNA-binding surface of these proteins. In patch 1 (shades of blue) RNA- protein contacts are formed by positively charged amino acids from β-strand β1, helix α2 and several highly conserved amino acids. Patch 2 (green) represents a hydrophobic pocket able to accommodate a purine or pyrimidine base flipped from the kink-turn RNA. This pocket is formed by amino acids from different parts of the protein. Both of these patches are present in all L7Ae family member-RNA complexes. Patch 1 is relatively conserved in sequence and structure. Patch 2 varies in size and amino acid composition but is generally composed of hydrophobic core residues surrounded by polar amino acids. We propose that patch 1 plays a role in general protein-RNA interaction and patch 2 may be responsible for sensing the kink-turn RNA motif. Interestingly, the same conserved RNA- binding like patches are present on the surface of *Xenopus* Gadd45a. Besides the two patches, the glycine residue homologous to G39 in Gadd45a (red) is highly conserved in all L7Ae family members and constitutes a third important structural determinant for kink-turn RNA binding. This residue is located at the beginning of helix α2, further referred to as guanine (G)-binding region, since in all L7Ae-RNA structures a guanine base is tightly bound in this region through an extensive hydrogen-bonding network (Figures S2 and S3). A glycine to alanine or lysine mutation of this residue completely abrogates RNA binding and protein function in human SBP2 and p15.5 kDa protein, respectively. Indeed, modeling the G38A mutant of hsp15.5, we discovered that a G38A mutation should result in sterical clashes with bound RNA. Taken together, our modeling and structural analysis suggests a rationale for the ability of Gadd45a to bind RNA despite its acidic pKa and the absence of a distinctly positively charged region. ## Mutations affecting Gadd45a RNA binding and demethylation To test the role of amino acid residues in patch 1 and of G39 in RNA binding and DNA demethylation, we generated four point mutations in Gadd45a, K45A, R34G, V49R and G39A. RNA binding activity of purified recombinant proteins was tested by filter binding assay with radioactively labeled vector derived RNA. Nonspecific RNA binding ability was comparable for the three mutants K45A, R34G and G39A and wild type Gadd45a. The V49R substitution showed three-fold higher RNA binding ability. This may reflect increased ionic interaction upon addition of an extra positive charge in this position. We next tested specific RNA binding by the RNA competition assay described in. For each mutant we compared the binding competition of labeld synthetic (vector) RNA with unlabeled cellular RNA. Since total RNA was a good competitor in the *in vitro* binding assays, we reasoned that it contains relevant- but unknown RNAs, which physiologically bind to Gadd45a with high affinity. Wild type Gadd45a showed a three-fold difference in competition assays between vector- and total RNA. This was similar for the R34G and V49R mutants, which harbor patch 1 substitutions naturally occurring in other L7Ae members. In contrast, substitution of either of the two ultra-conserved amino acids – K45A and G39A - caused a loss of discrimination between vector- and total RNA binding. To test the activity of the mutants in DNA demethylation, we monitored Gadd45a-mediated re-activation of an *in vitro* methylated – and hence silenced - luciferase reporter plasmid. Gadd45a can demethylate and thus transcriptionally activate such reporters. Upon transfection in HEK293T cells wild type Gadd45a as well as R34G and V49R mutants equally activated the methylated reporter. Notably the K45A mutant, which failed to discriminate between specific and non-specific RNA binding, was fully active in the demethylation assay, indicating that the two properties can be uncoupled. This already suggests that specific RNA binding is not absolutely essential for DNA demethylation, at least under these experimental conditions. In contrast, Gadd45a G39A was the only mutant inactive in the reporter assay as well as in specific RNA binding. To test for gene specific demethylation, we transfected a methylated EGFP expression plasmid and monitored its methylation status by digestion with the methylation sensitive endonuclease HpaII. In parallel we measured its expression by detecting EGFP protein in the cell lysates. The analysis showed that cotransfection of wild type, K45A, R34G and V49R led to expression of EGFP protein and to the appearance of a HpaII cleavage product, indicative of demethylation. The G39A mutant, which failed to discriminate between specific and non-specific RNA binding, was also inactive in DNA demethylation as well as activation of EGFP expression. Finally, we tested, if the G39A mutant protein still localizes to nuclear speckles. Nuclear localization of the G39A mutant was much reduced in general and less than 20% showed nuclear speckles. Similarly, Western blot analysis of detergent treated cells showed that G39A was more sensitive to extraction than wild type Gadd45a. From the mutant analysis we conclude that a) specific RNA binding is not absolutely essential for DNA demethylation and b) that G39 is a critical amino acid for the function and localization of Gadd45a. # Discussion The main findings of this study are that Gadd45a is an RNA binding protein and that it appears to be part of an RNP particle. This is in line with the function of other members of the L7Ae/L30e/S12e superfamily, which are either ribosomal components or associated with RNP particles. Gel filtration and cross linking analysis of recombinant Gadd45b,g indicates that the protein forms a dimer of 35 kDa, while Gadd45a can oligomerize, suggesting that the cellular high molecular weight form of Gadd45a may contain multimers. The conclusion that Gadd45a is in an RNP complex is supported by sucrose density gradient centrifugation and its localization in nuclear speckles. It is interesting that nuclear speckles are a site of active transcription, RNA splicing and processing. This raises the possibility that Gadd45a RNPs are associated with genes undergoing active DNA demethylation and transcriptional activation. Of note, p68/Ddx5, which colocalizes with Gadd45a in nuclear speckles, was previously described as a component of a DNA demethylase complex. RNP complexes play prominent roles in RNA processing, RNA transport and RNA translation (for review, see –). In light of our results it is interesting that overexpression of Gadd45 leads to a similar phenotype in *Drosophila* as mutation of *squid*, which encodes an hnRNP. In both cases the chorion of fly eggs is dorsalized due to defects in *grk* mRNA localization and translation, supporting the idea that a Gadd45 RNP function is evolutionary conserved. Our *in silico* modeling suggests a structural basis for the RNA binding of Gadd45a. Like other RNA binding proteins of the L7Ae family, Gadd45a contains two patches, which appear to be involved in RNA binding. The *in vitro* RNA binding assays and point mutagenesis data suggest that Gadd45a has moderate affinity for nonspecific RNAs and high affinity for specific RNAs. The V49R substitution increased non-specific RNA binding. Since arginine instead of valine 49 is a naturally occurring variant in some L7Ae superfamily proteins, we propose that such proteins have a higher general RNA binding propensity. The specific RNA binding of Gadd45a is clearly not absolutely essential for its demethylating activity, as shown by the K45A mutant. However, our DNA demethylation assay chosen for convenience is rather artificial; it involves an abundant *in vitro* methylated reporter plasmid, which is demethylated by overexpressed Gadd45a. In contrast, locus-specific demethylation under physiological conditions may very well require its ability to bind specific RNAs as discussed below. The G39A substitution in the G-binding region abolished both DNA demethylation as well as specific RNA binding ability. Since the K45A mutant still demethylates despite inactivated specific RNA binding, the RNA binding defect of G39A may not be the cause for the demethylation defect. This raises the possibility that G39A interferes with some other important property of Gadd45a. However, at least dimerisation of Gadd45a does not seem to be affected by G39A since by *in silico* docking of Gadd45a dimers, G39 was not found in the vicinity of the putative Gadd45a dimer interface. Our study raises new questions concerning the biology and biochemistry of Gadd45 proteins. Which RNAs are physiologically bound to Gadd45? What other proteins are parts of the Gadd45 RNP particle? Is the role of Gadd45 bound RNAs purely structural or is RNA involved in e.g. specific targeting to demethylated DNA regions? # Supporting Information We thank Gabi Döderlein for help with experiments and Ingrid Grummt for critical reading. [^1]: Conceived and designed the experiments: YAS CN. Performed the experiments: YAS AK AS. Analyzed the data: YAS AK AS AW CN. Contributed reagents/materials/analysis tools: AK CN. Wrote the paper: YAS AW CN. [^2]: The authors have declared that no competing interests exist.

# Introduction According to the reports of world health organization (WHO) cardiovascular diseases (CVDs) are the number one cause of mortality worldwide and an estimated 17.7 million people died from CVDs in 2015, representing 31% of all global deaths. Of these deaths, an estimated 7.4 million were due to coronary heart disease and 6.7 million were due to stroke. The treatment costs of CVD imposed on individual and society are substantial; since it involves majority of young working and productive population especially in Iran and low income countries, its heavy financial burden in these countries is alarming. In Iran, coronary artery disease (CAD) is a major cause of mortality and morbidity and accounts for nearly 50% of all deaths and 79% of deaths due to chronic disease per year. As estimated, the burden of cardiovascular disease in Iran will increase steeply over 2005 to 2025, mainly because of the major epidemiologic and demographic transitions and increase in aging population. These alarming points highlight the need for more attention to deal with the impact of CVD in the following decades in Iran. Primary prevention of cardiovascular disease is an important priority for developers of health policy against cardiovascular disease; healthy dietary habits, smoking cessation, weight management, regular physical activity and stress management all are most important strategies to reduce the risk of CVD later in life. Diet, undoubtedly, is one of the most important factors affecting cardiovascular health and numerous healthy dietary guidelines have been developed like dietary approach to stop hypertension (DASH) recommending diets high in fruits, vegetables, polyunsaturated fatty acids and low in fat and sugar, European society of cardiology (ESC) and National Institute for Health and Clinical Excellence (NICE) recommendations about reducing saturated fat and alcohol and increase in fiber and fish intake as a hallmark of Mediterranean dietary pattern. However, it is important to study the effect of whole diet and food groups rather than single nutrients or foods to achieve a reliable perspective of the diet-disease relationships. Recently, dietary indices and patterns have been developed and attracted much attention because of capturing multiple dietary factors and providing a comprehensive assessment of diet, accounting for the complex interactions between nutrients and foods. Because of the pro-inflammatory and anti-inflammatory roles of nutrients, the inflammatory potential of diet could be associated with numerous inflammation-associated diseases including CVD, diabetes, obesity and so on. Therefore, previous literature review based-cross sectional studies have developed dietary inflammatory indices to reveal the inflammatory potential of diet. Numerous reports are available about the association between dietary inflammatory index and several chronic diseases. However, several inconsistencies about the association between DII and chronic disease are also present. Recently, empirically developed dietary inflammatory potential (EDIP) has been developed in a US-based prospective cohort according to the inflammatory potential of food groups and its validity has been evaluated. Recent studies have reported that EDIP has a higher ability to predict the plasma concentrations of circulating inflammatory parameters compared with DII. Because of the potential role of inflammation in the pathogenesis of cardiovascular disease, and the clear association between dietary parameters with the risk of CVD, in the current study, we aimed to: (a) evaluate the role of EDIP in the prediction of cardiovascular metabolic risk factors and (b) to evaluate the association of EDIP with several other dietary indices including dietary antioxidant quality score (DAQs), dietary phytochemical index (DPI) and Mediterranean dietary quality index (MEDQI) in patients candidate for CABG. # Materials and methods ## Subjects The reports of the current study are a part of the Tehran- Heart Center-Coronary Outcome Measurement (THC-COM) study. Four hundred fifty four patients candidate for the CABG aged between 35 to 84 years old admitted to the cardiothoracic ward for CABG surgery at a large Heart Center in Tehran had been participated in the current study. Demographic, anthropometric and biochemical assessments were performed in participants after obtaining written informed consent. The study protocol was also approved by the ethics committee of Tehran University of Medical Sciences and Ethics committee of Tabriz University of Medical Sciences. Anthropometric assessments included weight and height which was measured by standard methods and body mass index (BMI) was also calculated. Biochemical assessments were also previously described unless about serum telomerase and vitamin D assessments which were measured by ELISA method (MyBiosource, USA). ## Measurement of empirically developed dietary inflammatory potential (EDIP), dietary antioxidant quality score (DAQ), dietary phytochemical index (DPI) and Mediterranean dietary quality index (Med-DQI) Dietary intake was assessed using 138-item semi-quantitative food frequency questionnaire (FFQ) consisting of a list of foods with standard serving sizes commonly consumed by Iranians and was validated and adopted for use in Iran. For measurement of the inflammatory potential of diet, EDIP was developed based on food group intakes. Dietary intakes were assigned into the fifteen food groups, including coffee, tea, dark yellow and leafy green vegetables, tomatoes, other vegetables, processed meat, red meat, organ meat, other fish, snacks, fruit juice, pizza, refined grains and high-energy beverages. Wine, beer and low- energy beverages were not used to construct EDIP score because their consumption is not usual in our population. Mean daily intake of each food group was determined by defined serving sizes and then weighted by the proposed regression coefficients. The weighted intake of food groups was summed to construct EDIP and then rescaled by dividing by 1000 to decrease the magnitude of the score and simplify the interpretation. Total dietary antioxidant quality score (DAQ) was calculated according to a method first described by Rivas et al. For the calculating the score, the intake of several certain antioxidants including zinc, selenium, vitamin A, vitamin E and vitamin C were assessed separately by assigning a score of 0 or 1. This scoring was based on the comparison of nutrient intake with the dairy recommended intake of nutrients (RDA). When the intake was below 2/3 of the RDA, 0 score was signed. While in the dietary intake of nutrient higher than 2/3 RDA, the assigned score was 1. Therefore, the total dietary antioxidant intake (TAC) was ranged between 0 (very poor) to 5 (high quality). Mediterranean dietary quality index (Med-DQI) and dietary phytochemical index were measured as described in our previous report. Briefly, for calculating the Med-DQI daily intake of each of the seven food components including saturated fatty acids, cholesterol, olive oil, meats, fishes, cereals, vegetables and fruits assigned a score of 0, 1 or 2 and then final score was reported as a summation of all nutrient scores ranged between 0 and 14. The lower score of MED-DQI denoted a better nutrition quality and higher adherence to Mediterranean dietary pattern. For calculating the DPI, calories derived from high-phytochemical content foods including fruits, vegetables (except for potatoes), legumes, whole grains, nuts, seeds, fruit/vegetable juices, soy products are enumerated in this index. The higher score denoted the higher phytochemical content of diet. ## Statistics SPSS software (statistical package for social analysis, version 18, SPSS Inc., Chicago, IL, USA) was used for analysis of data. The normality of data was tested by Kolmogorov-Smirnov test. The comparison of discrete and continuous variables between different quintiles of EDIP was performed by Chi- square test and analysis of variance (ANOVA) respectively. Odds ratios and 95% confidence intervals for the association between different quintiles of EDIP and biochemical parameters were estimated using logistic regression models, adjusting for confounders including age, BMI and presence of diabetes and myocardial infarction. All data are expressed as means ± SD or number and percent. *P* values less than 0.05 were considered as statistically significant. # Results Patients in higher quintiles of EDIP were younger and had higher BMI and prevalence of hyperlipidemia (P \< 0.05). The comparison of dietary indices among different quintiles of EDIP among patients showed that individuals in lower quintiles of EDIP had higher scores of vitamin E and total dietary antioxidant scores, higher dietary phytochemical score and lower dietary Mediterranean quality scores. In comparison of OR and confidence interval (CI) for the association between EDIP and biochemical variables, in male subjects, highest scores of EDIP was associated with higher serum cholesterol, creatinine, BUN and Lp (a) concentrations; moreover, being in third quintile of EDIP was associated with lower albumin concentrations compared with first quintile. In female subjects, lower HCT, higher creatinine and higher CRP concentrations were associated with higher EDIP scores. # Discussion In the current work, high empirically developed inflammatory potential of diet was associated with higher body mass index, higher prevalence of hyperlipidemia, lower score of vitamin E and antioxidant score of diet and lower DPI and Med-DQI in patients candidate for CABG. Moreover, among biochemical variables, highest scores of EDIP was associated with higher serum cholesterol, creatinine, BUN and Lp (a) and lower albumin concentrations in men and lower HCT, higher creatinine and higher CRP concentrations in women. This is the first study reveals the association between EDIP of diet and numerous metabolic risk factors of CVD and also dietary indices among patients candidate for CABG. EDIP of diet, give a perspective of the potential of diet in inducing inflammation mostly based on increased interleukin-6 (IL-6), C-reactive protein (CRP), and tumor necrosis factor a receptor 2 (TNFaR2) concentrations. Obesity is a well- known low-grade chronic inflammatory condition mostly because of adipose tissue derived cytokines development and accumulation in the body. The association between pro-inflammatory potential of diet and incidence of overweight and obesity has been studied and revealed before; in the study by Ramallal R among adults of SUN cohort, being in highest pro-inflammatory diet index make people 1.32 times more likely to being obese annually compared with lowest quartile (anti-inflammatory diet). Another population based report among 7236 participants, dietary inflammatory index was directly associated with the incidence of general and central obesity even after adjustment for the known risk factors. The association between obesity and inflammation can be bidirectional meaning that adiposity can induce inflammation, while a pro- inflammatory diet can induce adiposity and therefore this will make a vicious cycle between obesity and inflammation; accordingly, in our previous work, we reported the higher DII in association of serum lipids and CRP concentrations among candidates of CABG. Inflammation leads to HDL reduction and LDL and VLDL production in the cells via impairment in the reverse cholesterol transport system; another possible mechanism is triggering receptor expressed on myeloid cells-1 (TREM-1) induced dyslipidemia and consequently, fat deposition and pro- inflammatory cytokines production. In the current work, EDIP of diet was negatively associated with vitamin E and total dietary antioxidant scores and also was in negative association with DPI and Med-DQI in patients. These findings showed that EDIP is a precise measure of other dietary indices as expected; same as our results, Hodje et al reported an inverse association between Mediterranean dietary score and DII while higher Mediterranean dietary score denoted more healthy diet. In another work by Bawaked et al, higher DII of diet was associated with lower adherence to Mediterranean diet, lower antioxidant capacity of diet and high energy density in youth. These findings confirm that EDIP of diet has a precise association with other dietary indices and is a reliable indicator of unhealthy diet. Lower serum albumin in third quintile of EDIP compared with first quintile in women is another finding indicting the association between diet and inflammation. Albumin is a negative acute phase proteins and inflammation is a suppressor of its synthesis. It has been shown that anti-inflammatory and antioxidant nutritional interventions are able to increase serum albumin concentrations and to reduced hypo-albuminemia. Higher serum BUN and creatinine were associated with higher inflammatory potential of diet in patients candidate for CABG. We also reported similar results in our previous work indicating positive association between DII and serum creatinine. Several other studies have also reported consistent results about the potential of inflammatory diet in inducing chronic kidney disease and reduced kidney function by increasing CRP concentrations, reducing glomerular filtration rate (GFR) and increase in BUN and creatinine. We should also address several limitations of the current study; first of all, the self-reported dietary information obtained by food frequency questionnaire could have a potential recall bias. However, the questionnaire has been validated previously. Secondly, we did not measure WHR or WC as indicators of central adiposity. Taking together, the study’s relatively large sample size and inclusion of multiple confounders in the statistical model are potent strengths of the current study. Moreover, this is the first study evaluated the association between EDIP and CVD risk factors in patients candidate for CABG. # Conclusion We reported that EDIP is a potent predictor of obesity, dyslipidemia and cardio- metabolic risk factors among patients candidate for CABG. The summarized conceptual findings of the work have been presented in. Additionally, EDIP of diet was in inverse association with two dietary indices including dietary anti- oxidant quality score and dietary phytochemical index. Taking together, EDIP could be assumed as a precise nutritional tool for estimating the CVD related risk factors among patients candidate for CABG. ## Ethical approval and consent to participate All participants signed a written informed consent approved by the Institutional Review Board of Tehran University of Medical Sciences. The study design and protocol was approved by the ethical committee of Tehran and Tabriz University of Medical Sciences. # Supporting information The authors appreciate the cooperation of the patients in the current work. ANOVA analysis of variance CAD coronary artery disease CVD cardiovascular disease DAQs dietary antioxidant quality score DPI dietary phytochemical index EDIP empirically developed dietary inflammatory potential HDL high density lipoprotein cholesterol hs-CRP high sensitive C reactive protein LDL low density lipoprotein cholesterol Lp lipoprotein Med-DQI Mediterranean dietary quality index TC total cholesterol [^1]: The authors have declared that no competing interests exist.

# Introduction Humans' facility for dispersal has played a large role in our evolutionary history, yet our understanding of how and why humans have moved throughout history is unclear. Most data on human movement come from ethnographic and archaeological studies, comparisons of birthplaces from birth certificates, and census data. While ethnographic studies offer insight into social and environmental factors that influence human movement, they generally involve seasonal or temporary movements, as in the case of migrant workers or hunter- gatherers. In order to understand how migration has influenced our evolutionary history, it is necessary to address migration as the movement to a new location for permanent settlement. Although archaeological studies can provide information about movement over longer periods of time, they are often limited by the availability of data and restricted to specific regions and time periods. Birth certificate and census data allow us to trace movement across longer periods of time as well, but studies using these data generally focus either on the proportion of migrants or the distance moved, do not usually use multi- generational families, and can typically only be studied in developed countries. A deeper understanding of migration over multiple generations in a developing country offers the possibility of describing more general patterns of human migration and of identifying factors that may have influenced migration throughout human evolution. Since human migration has had the largest effect on genetic variation over human evolution, a better understanding of human migration patterns would allow more accurate reconstructions of demographic processes. Comparisons of empirical genetic data to simulated genetic variation generated from models that realistically represent the demographic process under study offer the possibility of reconstructing prehistoric demographic processes. Values for migration parameters estimated from human migration patterns, such as the proportion of the population that is moving, could define some model parameters in order to generate more realistic demographic scenarios. The ability to include empirically-informed values to fix or set ranges on migration parameters increases the probability of identifying the best model to explain the data. Yemen is a developing country that has a heterogeneous landscape with coastal plains on the west and south, mountain ranges in the west and desert in the north, thus providing a fertile setting in which to investigate environmental factors that may have influenced prehistoric population movements. Yemen has a patrilocal and patrilineal society with a primarily shared language and religion, which are social factors that could play a role in migration, as well. Migration within a population of mostly agriculturalists and pastoralists should provide more realistic values of distance and proportion of migration for prehistoric movements since the advent of agriculture. The values should also provide informative lower limits for describing the migration of prehistoric hunter-gatherers, who typically exhibit more movement than agriculturalists. In this study, we use GPS coordinates from birthplaces and places of residence across four generations in Yemen to calculate the proportion, the distance and the direction of migration between each generation. We test for differences in these values between the generations, we identify factors that influence migration patterns, and we discuss possible effects of the migration patterns on genetic variation. Based on our results, we provide estimates for the proportion and distance of migration in a developing country, which can define parameter values for evolutionary models used to reconstruct prehistoric demographic processes. Our use of empirical data on population movements over four generations in Yemen provides knowledge that will allow for more accurate reconstruction of prehistoric processes of migration. # Methods ## Ethics Statement This study has been approved by the Western Institutional Review Board Olympia, Washington (WIRB project \#20070219). Samples were collected with verbal informed consent approved by WIRB. This modified inform consent was used because a majority of the population is illiterate. Only individuals who gave consent provided both saliva samples and information for the sample collection sheet, and were entered into the database of study participants. ## Samples and Data In 2007, saliva samples were collected throughout mainland Yemen for genetic analysis. Data were also collected from each study participant on current place of residence, place of birth, parents' place of birth and grandparents' place of birth. Since all sampled individuals were adults, their current residence was used as a proxy for the location of the next generation, i.e. their offspring, therefore providing data on residence patterns for four generations in the study. For the purposes of this study, the individuals in each generation were considered independent samples. Location names for all birthplaces (and place of residence) were translated from Arabic and GPS coordinates were obtained using Geonames.org. In instances where a town name was not identifiable in the Geonames database, but the larger district could be identified, a GPS coordinate was obtained for the centroid of the district. Samples for which town or district locations could not be determined were removed. Ultimately, the resulting dataset contained GPS coordinates for the sampled individual's place of residence and place of birth, mother's and father's places of birth, and maternal-grandmother's, maternal-grandfather's, paternal-grandmother's, and paternal-grandfather's places of birth for 351 sampled individuals (2,457 total sample locations). ## Estimation and Analysis of Migration The occurrence of migration was determined by the difference in birthplace or residence location between generations. The current place of residence, considered a proxy for the “offspring” generation (G0) of the sampled individuals, was used to identify migration in the sampled individual's generation (G1). Thus, a migration event occurred in the sampled individual's generation (G1) if the place of residence was different from the birthplace. A migration in the parental generation (G2) occurred if the parent's offspring was born in a different location than the parent's birthplace (i.e. if the sampled individual's birthplace was different from their mother's or father's birthplace). Similarly, a migration event in the grandparental generation (G3) occurred if the parent's birthplace was different from the grandparent's birthplace. Migration events were determined for eight different groups: female sampled individuals (G1_fem), male sampled individuals (G1_male), mothers (G2_fem), fathers (G2_male), maternal-grandmothers (G3_mfem), maternal-grandfathers (G3_mmale), paternal-grandmothers (G3_pfem), paternal- grandfathers (G3_pmale). The frequency of migration events was calculated for each of the eight groups (sample sizes were 70 in G1_fem, 281 in G1_male, and 351 in each group in G2 and in G3. The observed frequencies were compared through goodness-of-fit tests. The age of the sampled individuals ranged from 18 to 69, which meant that each generation group (G1, G2, G3) essentially included two generation time periods. To account for the possibility of migration events occurring over different generation time periods within each generation group, the eight groups were further divided into two age groups with a 25 year generation time between them, based on the ages of the sampled individuals (under and over 40 years old). Only 10% of the samples in any generation were in the over 40 years old sub-group, suggesting that any difference in migration event frequencies could be due instead, to the unbalanced sample size; thus no further analyses were performed with the groups partitioned by age over and under 40 years. Migration distance was calculated from the geographic distance between birthplaces/residences in two different generations using the GPS coordinates. G1 migration distances were calculated as the geographic distance between the sampled individual's birthplace and place of residence. G2 migration distances were calculated from the parent's birthplace and the sampled individual's birthplace. Migration distances were calculated for G3 from the difference in grandparent's birthplace and parent's birthplace. The migration distances were compared between sex in each generation and between generations using Wilcoxon Rank tests and Kruskal-Wallis analyses of variance. Different models including generation group, sex, birthplace location (latitude and longitude), and residence location (latitude and longitude) were tested in logistic regressions to see which model (and parameters) best explained migration. AIC (Akaike information criterion) were used to select the best model. Additionally, the migration events were plotted geographically and the mean direction of the migrations was calculated for each collection site (to account for sampling) using ESRI ArcMap10. # Results The proportion of migrants was calculated from the frequency of migration events for females and males in three generations (G1_fem = 0.314, G1_male = 0.267, G2_fem = 0.376, G2_male = 0.311, G3_mfem = 0.120, G3_mmale = 0.111, G3_pfem = 0.097, G3_pmale = 0.080). Within each generation, the proportion of migrants between male and female groups was not significantly different. However, more recent generations G1 and G2 had a significantly larger proportion of migrants than G3 (p = 0.0005). The proportion of migrants for each generation (males and females combined) was G1 = 0.276, G2 = 0.343, G3 = 0.102. We also calculated a multi-generation proportion of migrants for G3 to correct for back migration events by determining the number of migration events in which the grandparents' birthplace was different than the residence location. This produced a multi-generation proportion of migrants for G3 of 0.086. The distance of migration was also calculated for each of the eight groups. G1 and G2 migration distances were significantly larger than G3 (p\<2.2×10⁻¹⁶). Density plots combining the migration distance (including non-migrants) and the frequency of these distances revealed that G1_fem not only had the largest migration distance, but had more migrations at longer distances (\>250 km), than the other groups. However, when compared by sex within generations, female distances were not significantly different from male distances. Summary statistics on migration distances were calculated on all individuals and on only migrating individuals. Correlation analyses were performed on marital pairs in G2 and G3 to determine whether marital pairs were moving together. A low correlation coefficient (\<0.1) would suggest the marital pair migrations were completely independent from each other and a high correlation coefficient (\>0.9) would suggest that the marital pairs were moving together and should be treated as one group (instead of female and male groups). G2 had a significant (p = 2.2×10⁻¹⁶) Spearman's rho correlation coefficient of 0.589. Maternal grandparents (G3_M) had a rho coefficient of 0.782 (p = 2.2×10⁻¹⁶) and paternal grandparents (G3_P) had a rho coefficient of 0.623 (p = 2.2×10⁻¹⁶). These results showed there was a moderate and significant correlation between all marital pairs. These coefficients suggest that a portion of the marital pairs are moving together, but the correlations are not high enough (\>0.9) to consider the marital pairs as a single group. Female and male marital pair distances were plotted and showed that correlated migrations were of the same distance, which is consistent with marital pairs moving to the same place. Out of the 121 migration events in G3, 56% were of marital pairs moving together. These results suggest that many of the individuals may be moving due to post-marital residence dynamics (i.e. husbands and wives moving together). Logistic regression models, including different combinations of generation, sex, birthplace coordinates and residence location coordinates, were performed to explain presence or absence of migration. The model with the lowest AIC included generation, sex, birth latitude and longitude and residence latitude. This best model demonstrated, that relative to G1_fem (as the baseline group), the probability of migration decreased in G3, decreased in males (consistent with females moving with their husbands' families) and decreased with a more easterly birthplace. In contrast, the probability of migration increased in G2 and increased with more northern birthplaces and places of residence. However, of these factors, only G3 had a coefficient above one, suggesting that G3 contributes the most to the probability of migration, and specifically, belonging to the G3 generation decreases the probability of migration. Although birthplace latitude, birth place longitude and residence location latitude had small coefficients, their statistically significant contribution to the migration probability suggests that there could be factors “pushing” individuals away from a place (leave one's birthplace) or “pulling” individuals to a place (move to a new place). The birthplace and residence coordinates were used to plot the directionality of migration to assess whether or not there was a pattern in directionality that could explain the “pushing” and “pulling” effects. The mean migration direction was calculated from these migration vectors for each sample collection site (to account for the effect of sampling). While the mean migration directions seem to have a southbound tendency, the circular variance (which describes the variation associated with the directional mean, where values close to 0 represent a similar direction for all migration vectors and values close to 1 correspond to vectors in all compass directions) was moderate to high for all collection sites, ranging from 0.675 to 0.867, suggesting movement in all directions. The mean migration directions were further calculated by collection site for each generation group. Within generation groups G2 and G3, female and male migration directions were similar in many collection sites, supporting the idea that marital pairs moved together. The mean migration lengths were generally larger for G1 and G2 than for G3, reflecting the decreased migration distance in G3. For each collection site, the mean migration directions varied greatly between generation groups, suggesting a level of stochasticity to the migration directions. When the mean migration directions were spatially compared to geographic features (i.e. elevation, land use/land cover, and watershed), no pattern arose (data not shown), further supporting stochasticity in the directionality of migrations. # Discussion Our study helps elucidate human migration patterns using empirical population movement data across multiple generations in Yemen. Our results show that the proportion and distance of migration increased in recent generations. While movement in the recent generations may reflect social and political changes that have occurred in the last 50 years, the reduced movement in the oldest generation most likely reflects a lack of technology and associated mobility, suggesting that this generation may be most representative of prehistoric movements. The correlated distance and directionality of migrations within marital pairs illustrate the prevalence of post-marital residence dynamics. The significance of birthplace and residence locations in the probability of migration, but lack of pattern in the direction of migration, suggest a degree of stochasticity in terms of human movements. These cultural factors affecting modern movement have most likely played important roles in prehistoric migrations as well, suggesting that the migration patterns and estimates described in our results provide information to make more accurate prehistoric inferences. ## Patrilocality and Genetic Signals Moderate correlation coefficients for G2 and G3 marital pairs and the plot of migration distances in marital pairs suggests that pairs are moving together and the correlation seems to strengthen with increasing distance. Our best fit model, which shows that females are more likely to move than males when accounting for other contributing variables, suggests that patrilocality (females moving to their husbands' family) may be driving the movement. This is supported by ethnographic accounts that ∼90% of the Yemeni population is patrilocal. However, the coefficient of the effect that being male has on the probability of migration is low (−0.240) and within each generation the migration distance is not significantly different between females and males. This suggests that males are only slightly less likely to migrate than females and that males are travelling similar distances compared to females. In a perfect patrilocal post-marital residence dynamic, males move short distances and stay close to their family, while females move longer distances to be near their husbands' family. The similar migration distances between females and males suggest there is not strict patrilocality in Yemen and that other factors are influencing male movement. This interpretation is supported by ethnographic data showing that males may occasionally migrate large distances from their birthplace for socioeconomic or political reasons. Our data show that male migration has occurred more often in the last 50 years (as shown by the increase in dispersal in G1 and G2 relative to G3). The similar migration distances between females and males, and consequent imperfect patrilocality may be the principal contributor to the lack of association observed between geographic and genetic distance in male lineages (i.e. Y chromosome) in Yemen. Females moving with their husbands may also explain why shared mitochondrial DNA (mtDNA) haplotypes have been found between east and west Yemen, over 750 km apart. ## Patterns of Migration Logistic regressions were used to test the effect of birthplace and residence locations on the probability of migration in order to assess whether there were factors “pushing” or “pulling”, respectively, individuals to a new location. Birthplace latitude and longitude and residence latitude were significant parameters in explaining the probability of migration. Given this result, birthplace and residence coordinates were used to plot migration directions and determine whether a pattern could be observed that could account for the effects of birthplace and residence locations. Mean migration directions were calculated by collection site (to account for sampling bias) to summarize the overall migration direction patterns. Although the mean migration directions had a southbound trend, the circular variances were large, suggesting overall dispersal in multiple directions. Additionally, mean migration directions calculated by collection site for each generation showed that the collection sites had different mean directions between generations, further supporting migration in multiple directions. We also spatially compared the migration directions with different geographic features (i.e. elevation, land use/land cover, and watershed) to identify environmental factors that may influence migration direction. We found no pattern associated with the migration directions and the geographic features (data not shown). These results suggest that while there may be factors “pushing” and “pulling” individuals to move, the overall direction of migration has little or no pattern. These results contrast with island migration patterns (e.g., Polynesia) where migration direction has a pattern from larger islands to smaller islands. Given that continental migrations are less limited by the carrying capacity of new colonization sites than islands, our results are not surprising. While island migrations have been well described by ethnographic and archaeological data, continental migration patterns have been primarily addressed through genetic data. Genetic evidence has suggested that overall continental migrations have a linear pattern, such that increasing distance from Africa is correlated with decreasing genetic diversity. Our data suggest that the smaller scale migrations that led to this continental pattern may have been less directed. Our results are consistent with the idea that smaller migrations, which consider the movement of individuals, tend to be more random, while larger scale movements focused on populations have more directionality associated with them. ## Empirical Estimates of Migration Comparisons of proportion of migrants and migration distances across four generations showed that migration was significantly lower around fifty years ago (G3). Furthermore, the best fit model to explain the probability of migration shows that G3 has not only the biggest effect, but a negative effect on the probability of migration (i.e. belonging to G3 decreases the probability of a migration event). Spatial patterns of migration in G3 show, that although there are some long migration distances, on average, the distances are short. Yemen's less-developed state and poor transportation infrastructure combined with the significantly reduced migration in G3, suggests that our data from the G3 generation can provide empirically-based estimates of migration frequency and distance that are reflective of prehistoric movements. We calculated the mean and median migration distances for G3. The mean migration distance for all individuals (i.e. including both individuals who migrated and those who did not) was 10 km. The mean and median distances for migrating individuals only were 96 km and 26 km, respectively. The shorter migration distance values (10 km and 26 km) are within the range of previously reported average migration distances –. These shorter migration distances potentially demarcate the distances within which post-marital residence patterns (patrilocality in the case of Yemen) have a distinguishable effect on genetic structure. In contrast, at distances beyond these values, isolation by distance is probably more predominant, and sex-biased migration is less detectable. Since most populations before the advent of agriculture (∼10 kya) were hunter- gatherers, we wished to identify whether our results provided estimates that may be informative in reconstructing prehistoric processes throughout these different periods in human history. Our shorter migration distance values (10 km and 26 km) are within the range of 10–30 km that Ammerman and Cavalli-Sforza believe is plausible for migration distance in agriculturalist societies. Furthermore, dividing 26 km by a generation time of 25 years results in a migration speed of 1.04 km/year. This value is comparable to the 1 km/year migration speed for the Neolithic transition estimated from archeological data. These similarities suggest that the shorter distance values, particularly the median distance, are representative of migration distances of agriculturalist groups. Hunter-gatherers generally migrate more and longer distances than agriculturalists. Therefore, our mean migration distance estimated using only migrating individuals offers a potentially informative migration value for the more mobile hunter-gatherer populations. Specifically, a migration speed (3.84 km/year) calculated from the mean value for only migrating individuals (96 km) falls within the broad range of hunter-gatherer migration speeds based on archeological evidence. Fort et al estimated the speed of the hunter-gatherers' recolonization of northern Europe after the last glacial maxima between 0.7 and 1.4 km/year. Hamilton and Buchanan estimated a speed of 5–8 km/year for the colonization of North America, while Hazelwood and Steele obtained estimates of 6–10 km/year. Because our value is intermediate to the values of these region- specific studies, it provides a distance that may be more generally applicable to migration processes, particularly *de novo* colonization migration distances by hunter-gatherers. This can be seen when we compare our migration speed estimate with Macaulay et al's inferred migration speed for the colonization of Southeast Asia. Based on founder time estimates from Eurasian and Australasian mtDNAs and the distance between India and Australasia, Macaulay et al infer a migration speed of 4 km/year. Our empirical estimate of 3.84 km/year suggests that their proposed migration process is in fact plausible. While migration distance has been estimated through different approaches, few studies have estimated the proportion of migrants,. We calculated the proportion of migrants for G3 to be 0.102 (or 0.086 when adjusting for back migration in the four generations). These values are smaller than the 0.4 proportion of migrants that can be calculated from Wood et al's dataset on migration between parishes in Papua New Guinea or the 0.366 estimate obtained from the calculation of individuals that were not born in the same parishes as their parents in La Cabrera, Spain. These differences from our estimates seem reasonable as Wood et al's estimates are from a more recent population (and are closer to our G1 and G2 estimates) and Boatinni et al's estimates are from a more developed country. Our estimates are somewhat larger than the 0.032 proportion of migrants into the island of Pingelap in Micronesia presented by Morton et al. However, our adjusted proportion of migrants (0.087) is closer to Morton et al's value. We also calculated the maximum and average number of individuals moving between a pair of locations, for a proportion of migrants of 0.0036 and 0.0011, respectively. These lower values are consistent with findings by Deshpande et al, where the genetic estimates of proportion of migrants (i.e. migration rates) for a world-wide colonization model are less than 0.01. Our values are similar to findings by Miró-Herrans and Mulligan, where the most probable proportion of migrants exchanged between African and non-Africans populations was 0.001 and are similar to the migration rate for non-African populations (1.5×10⁻³) obtained by Cox et al. The similarity of our estimates with those from other migration studies suggests that our values can be used in different scenarios to generate testable models for prehistoric reconstruction. ## Application of Migration Estimates in Prehistoric Demographic Modeling Model-based approaches for inferring prehistoric processes from genetic variation are becoming increasingly popular. These approaches, such as approximate Bayesian computation, require the generation of explicit demographic models to compare to empirical data. Including specific values for known parameters and informative ranges of values for unknown parameters increases the probability of identifying the best model to explain the data. The results from our study provide estimates that can be used to fix or set ranges on parameters related to migration, such as gene flow or founding population size, so that other parameters of interest can be addressed in greater depth, e.g., time of a demographic event. For example, the maximum and average proportion of individuals moving between a pair of locations (0.0036 and 0.0011) can be used to define gene flow (or migration rates) between populations stretching from southern Asia to northern Africa to create simulated DNA for models that address the back-migration into Africa. The larger migration values (0.102 or 0.086) can be used to define the founding population sizes for each new population out-of- Africa and back-to-Africa. Defining these parameters would allow for an in-depth exploration of the timing of the back-migration. Additionally, our results provide estimates to generate more geographically explicit models. Our mean and median migration distances (96 km and 26 km) provide estimates for the distance between populations, particularly for large scale movements, such as the back-migration from southern Asia. The migration distance between each population would define the number of populations to be simulated for the region under study. For example, a distance of 100 km between each population would require ∼70 populations between southern Asia and northern Africa (approx. 7,000 km). Understanding the possible distances involved in large scale movements also helps us determine how rapidly a migration could have occurred and how levels of gene flow may have been affected between the populations. The lack of migration directionality in our results suggests that explicitly including stochasticity or multidirectionality when describing the movement between populations might more accurately reflect the large-scale migration process. For example, the back-migration to Africa probably included movement through established populations, where the migrants settled in some of the established populations, but not in others. Therefore, a lattice stepping-stone migration model, that includes some randomness in terms of when a migration occurs and between which populations, might better reflect this migration process. Our results show there is over a 58% correlation between female and male movement in marital pairs, in which more pairs move together with increasing distance. Additionally, we show that 56% of migration events in G3 were by marital pairs. This means that at least 50% of the migrants have a 1∶1 female to male ratio. Even if the remaining 50% of migrants are only female or male, the ratio is at most 3∶1. These results argue for, at most, a 3∶1 ratio (for either sex) of sex-biased migration for migrations at short distances, where post- marital residence has a larger effect on population structuring. Alternatively, for longer migrations, such as the migration from southern Asia to northern Africa, our results suggest that a female to male ratio closer to 1∶1 more accurately models demographically balanced populations that would have been reproductively self-sustaining. # Conclusions In this study, we analyzed empirical data on migration patterns over four generations of human populations in Yemen in order to gain insight into the factors that influence migration, and specifically may have affected prehistoric movements throughout human evolution. Our approach to trace migration over generations has enabled the study of migration patterns throughout a developing country that would otherwise have been unfeasible. We provide empirical estimates for migration-related parameters that can be used to generate demographic models in model-based methods of prehistoric reconstruction. Our empirical estimates of generation G3 provide values for proportion of migrants, with values ranging from 0.102 or 0.086 proportion of overall migration, to 0.0036 or 0.0011 proportion of migrants between two specific populations. We also provide migration distances (96 km and 26 km, mean and median, respectively) that can be used to define the distance between populations and therefore the number of populations for the area under study. Using our approach, populations employing other modes of subsistence, such as hunter- gatherers, may be studied to further improve our knowledge on human migration. The findings from this study shed light on human migration patterns and enable more accurate reconstruction of the demographic processes that characterized human evolution. Improved models of human demographic changes and the associated genetic variation can provide a powerful tool to test for selective pressures, as well as to model the evolutionary history of co-evolving organisms. In this way, reconstruction of human demography and evolution may further provide insight into the movement and evolution of human pathogens and other co-evolving organisms. # Supporting Information We thank David Reed, Michael Miyamoto, and Steven Brandt for insightful discussion and comments on this manuscript. We thank Tania Saade for her contribution to the translation of the names of geographic locations used in this study. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: ATMH CJM. Performed the experiments: ATMH CJM AA. Analyzed the data: ATMH. Contributed reagents/materials/analysis tools: ATMH CJM AA. Wrote the paper: ATMH CJM AA. [^3]: Current address: Department of Anthropology, University of Texas at Austin, Austin, Texas, United States of America

# Introduction The scientific literature has repeatedly demonstrated that socioeconomic factors are powerful determinants of health-related outcomes, and socioeconomic status (SES) may be considered as one of the main causes of health disparities between different population groups. However, SES does not directly affect health status, rather it represents a proxy of other proximal and intermediate causal factors. Several competing factors and mechanisms have been proposed to explain the chain of events linking SES to health outcomes. However, the entire pathway by which SES exerts its effect on health has not yet been completely elucidated. Recently, health literacy (HL) has been proposed as one of the potential links between SES and health. According to Sørensen and colleagues, HL is linked to literacy and can be defined as the “people’s knowledge, motivation and competences to access, understand, appraise and apply health information in order to make judgments and take decisions in everyday life concerning healthcare, disease prevention and health promotion to maintain or improve quality of life during the life course”. Therefore, HL appears to be in close correlation with both socio-economic determinants and health outcomes. In a recent study that assessed both antecedents and consequences of HL in the same sample, Bonaccorsi et al found that less educated and poorer population groups have more often limited or inadequate HL and that low HL skills are associated in turn with worse self-reported health status. Indeed, on the one hand, disadvantaged socio-economic groups have been reported to have a greater risk of limited HL compared with more advantaged socio-economic groups. On the other hand, a limited HL has been shown to be associated with unhealthy lifestyle behaviors and several negative health-related outcomes, such as higher mortality and hospitalization rates, greater use of emergency care, lower receipt of mammography screening and influenza vaccine and poorer management of chronic diseases. However, most studies published to date have separately analyzed predictors and outcomes of HL, while only a few have evaluated the possible contribution of HL in the pathway through which socio-economic determinants affect health. Thus far, only one systematic review on the role of HL in the relationship between SES and health has been published. Results of this review suggest that HL partially mediates the effect that SES exerts on health status and on several health related-outcomes. However, the review pointed out that the evidence on the mediating role of HL is scarce, and that data from population-based sample studies are lacking. Yet, given the difficulties to act directly on socio- economic determinants of health, the confirmation of the mediating role of HL may help design and implement health policies and interventions aimed at reducing health inequalities. While assessing the mediation effect, it is also important to determine if results are equal across population subgroups. Indeed, the effect of SES determinants on health via HL might vary depending on the values taken by one or more third variables. Specifically, these third variables–defined as effect moderators—may affect (either enhance or reduce) the relationship between SES and HL and/or the relationship between HL and health. Therefore, the mediating role of HL in the pathway through which SES determinants affect health may be strengthened or weakened depending on the values taken by the moderator. From a public health perspective, the identification of the moderators may help target the population subgroups who would benefit most from HL interventions. However, to the best of our knowledge, no studies have analyzed the moderators of the mediating role of HL in the relationship between socio-economic factors and health. The aim of the present study is to evaluate whether functional health literacy mediates the association between socio-economic factors and health status in a population based-sample. # Materials and methods The study was approved by the Ethics Committee of the “*Area Vasta Centro*” (Local Health Unit of Central Tuscany, Careggi University hospital and Meyer University Children’s Hospital; Ref. CEAVC:10113, 01 December 2016) and was conducted according to the principles described in the Declaration of Helsinki. This study is part of a research project conducted to assess the level of HL in a population-based sample in Florence, and to validate different HL measures in Italian language. For more detailed information regarding methods and projects outputs, the reader is referred to the study protocol and articles published elsewhere. ## Study population and sampling criteria This is a cross-sectional study carried out in a not-representative population- based sample. Participants were randomly selected from a list of residents available from the registers of eleven general practitioners (GPs) working in primary healthcare centers of the municipality of Florence, Italy. According to the regulations of the National Healthcare System and the Constitution of the Italian Republic, every Italian and foreign resident aged ≥18 years has to be registered in a general practice, and people are enrolled in the general practices according to their place of residence (percentage of resident population registered 98.8%). This sampling method was chosen with the aim of increasing the participation rate as the invitation letter was jointly signed by the general practitioners and the researcher in charge of the study. The GPs were recruited using convenience criteria. All the GPs of the municipality of Florence were invited to join the study by both the Provincial Medical Council and the University Hospital of Florence. A total of 11 GPs based in different districts of Florence were recruited on a first-come basis. The number of GPs recruited in the study was increased from what had been originally proposed in the study protocol (i.e. n = 8) in order to extend the geographical coverage of the study. The recruited general practitioners were based in the city center and in the inner and outer suburban areas of Florence. Each GP selected 80 people from his/her register through a random number generator. Inclusion criteria were the following: 18–69 years of age (this criterion was adopted in order to be coherent with the Italian lifestyle surveillance system) and Italian speaking (since the survey was conducted in Italian only). Exclusion criteria included severe cognitive impairment and severe psychiatric diseases as these conditions may significantly impair the ability to understand the questions and provide reliable answers. Furthermore, in line with the principles described in the Declaration of Helsinki, people with end-stage diseases were excluded as all the potential insights derived from the research would not be beneficial to this population group. Inclusion and exclusion criteria were applied by each GP independently. ## Data collection, main variables and HL measure Data were collected between February and December 2017. Each selected person was contacted via postal mail. The selected people received a mail that included a short description of the study, an invitation to participate and a consent form. Participants were asked to sign the consent form and return it via mail to the researchers in charge. The mail also contained the nutritional label of the Newest Vital Sign-Italian (NVS-IT) designed to be easily readable (i.e. large font size and line-spacing). After receipt of the signed consent form, each participant was contacted by phone for the computer-assisted interview. If the consent form was not received within two weeks, a follow-up phone call was made by the research group. The phone call served to clarify any questions and to identify and assist people with difficulties in completing the consent form (e.g. reading difficulties). Nine interviewers conducted the phone interviews. A shared written protocol on how to conduct the interview was followed in order to standardize the interviews and to limit interviewer bias. Each participant was randomly assigned to one of the nine interviewers and contacted up to six times before being considered unreachable. The whole interview took about 20–25 minutes. Several demographic, socioeconomic, and health outcome variables were collected. For the specific purpose of the present study the following variables were utilized: sex, age, presence of chronic diseases, education level, financial status, and SRH. In particular, age in years was calculated from birth year; the presence of chronic diseases—defined as any disease that has lasted or is expected to last for at least 6 months—were coded in five categories: yes, more than one; yes, one; no; do not know; refuse to answer. Each participant’s education level was defined as the highest level of education attainment and was categorized as follows: Bachelor’s degree or higher; high school diploma; lower secondary school diploma or lower. As for the financial status, it was assessed by the item “is your income adequate to meet monthly living expenses?”, response options being: more than adequate; adequate; barely adequate; inadequate; refusal. Finally, SRH was chosen as it represents a valid and valuable measure of the overall health status that is widely employed in general population health surveys. SRH was assessed by a single item: “In general, would you say that your health is excellent, very good, good, fair, or poor?”. Functional health literacy was assessed using the NVS-IT tool. The NVS-IT consists of an ice cream nutrition label, with seven associated questions that measure literacy and numeracy. It produces a final score ranging from 0 to 6, allowing participants to be classified into three levels of functional HL: high likelihood of limited HL (score: 0–1), possibility of limited HL (score: 2–3), and adequate HL (score: 4–6). ## Statistical analysis Education level, financial status and SRH were analyzed as dichotomous variables. Specifically, education level was re-classified into the following categories: *i*. Bachelor degree or higher and *ii*. high school diploma or lower. The following two categories for financial status were defined: *i* adequate or more than adequate income and *ii*. barely adequate / inadequate income to meet monthly living expenses. SRH was classified into the following two categories: *i*. good SRH (*i*.*e*. excellent, very good, and good SRH) and *ii*. fair or poor SRH. Lastly, for all the analyses, subjects with high likelihood of limited HL and those with possibility of limited HL were grouped together in a single group, referred to “inadequate and at-risk HL” and compared with those with adequate HL. In the mediation analysis, a pathway is specified a priori in which an independent variable of interest influences an outcome through an intermediate variable, which is referred to as a mediator. Specifically, to test whether HL mediates the association between socio-economic factors (*i*.*e*. education level and financial status) and health status, a model-based causal moderated mediation analysis was performed separately for each of the two socioeconomic factors considered. The analysis proceeded in two steps. First, two logistic regression models were specified, *i*. *the mediator model* for the conditional distribution of the mediator (health literacy levels) given the independent variable (socio-economic factor), and *ii*. *the outcome model* for the conditional distribution of the outcome (self-reported health status) given the independent variable and the mediator. These models were fitted separately and controlled for age and sex as covariates; in addition, the outcome model also contained an interaction term for the independent variable x the mediator. Subsequently, the outputs of the mediator and outcome regression models served as the main inputs to the “mediate function” (mediation package in R software version 4.4.6), which computes the total effect of the independent variable on the outcome and decomposes this effect into the indirect (*average causal mediation effects*—*ACME*) and direct (*average direct effects*- *ADE*) effects. Specifically, the indirect effect reflects one possible explanation of the observed effect (*i*.*e*. that transmitted through the mediator), whereas the direct effect represents all other possible causal chains. Furthermore, the mediate function estimates the proportion of the effect mediated, which describes the average magnitude of indirect association between the independent variable and the outcome that is due to changes in the mediator variable relative to the average total association. Lastly, “moderated mediation” models were fitted in order to identify whether the magnitude of the ACME of HL varies depending on (*i*.*e*. “is moderated by”) the values taken by other covariates (defined as “moderators”). The following variables were tested as potential moderators: age, sex, and the presence of one or more chronic diseases. In the analyses, age was used as a continuous variable, and multiple comparisons between age classes were made. As for chronic diseases, the following comparisons were tested: no *vs*. one chronic disease; no *vs*. two or more chronic diseases; and one *vs*. two or more chronic diseases. As the missing values were lower than 1% for all the variables considered in the moderated mediation models, a complete data analysis with no attempt to input missing values was performed. For each analysis, an α level below 0.05 was considered as significant. # Results ## Descriptive statistics of the sample A total of 984 individuals were invited to participate in the study, of which 493 agreed to be interviewed (50.1%) and 454 (46.1%) were effectively interviewed. As far as non-participation reasons were concerned, 340 (34.5% of the total sample) people did not respond to any contact attempts, 151 (15.3% of the total sample) people refused to participate, and a further 39 (4% of the total sample) people initially agreed to be interviewed, but subsequently it was not possible to arrange an interview. Non-participants were on average two year younger than participants (51.2 ± 11.8 years and 53.3 ± 11.7 years, respectively). No significant sex differences emerged between participants and non-participants. Two interviewees were excluded from the study because of missing data on several variables; therefore, a total sample of 452 participants were finally included in the analyses. shows selected participants’ characteristics. Females represented 58.8% of the sample and participants with bachelor’s degree or higher represented the 41.1% of the sample; 69.9% of the participants had an income adequate or more than adequate to meet monthly living expenses. As for health status, 31% of the participants referred a very good or excellent health status. In terms of HL, high likelihood of limited HL, possibility of limited HL and adequate HL were found in 11.5%, 24.6% and 63.9% of the sample, respectively. ## Association between health literacy and independent variables (mediator models) The results of the mediator models for education and financial status are shown in Tables and, respectively. Both education level and financial status were positively associated with HL. In particular, participants with a high school diploma or lower education level had an increased odds ratio (OR) of having inadequate or at-risk HL compared with those with bachelor’s degree or higher education level (OR 2.59, 95%C.I. 1.66–4.02). As for the relationship between financial status and HL, participants with inadequate or barely adequate income had an increased odds ratio of having inadequate or at-risk HL compared with those with adequate or more than adequate income to meet the monthly living expenses (OR 2.03, 95%C.I. 1.28–3.21). ## Association between health status, health literacy and independent variables (outcome models) The results of outcome models for education and financial status are shown in Tables and, respectively. As for the former, lower HL and educational level were both independently associated with an increased odds of reporting a worse health status (OR 1.90, 95%C.I. 1.17–3.10; and OR 1.76, 95%C.I. 1.06–2.91, respectively). As for outcome model for financial status, participants with inadequate or barely adequate income and participants with inadequate or at-risk HL levels were found to have an increased odds ratio of worse health status (OR 4.08, 95%C.I. 2–8.34; and OR 3.75, 95%C.I. 1.68–8.37, respectively). No significant interaction between HL and education level was seen in the former model, and the interaction term was therefore not retained in the model whose output was entered in the moderated mediation analysis. Instead, there was a negative interaction between HL and financial status in the latter case (OR for the interaction term 0.032, 95%CI 0.12–0.86). ## Mediation effect of health literacy and analyses of the moderators In is presented the mediation effect of health literacy in the association between education level and SRH. On average, HL was found to significantly mediate 18.5% of the association between education and SRH; in particular, HL significantly mediated 21.1% of the association in participants with high school diploma or lower education levels and 15.8% of the association in participants with bachelor’s degree or higher education level. As far as the relationship between financial status and SRH is concerned, HL was found to significantly mediate 12.9% of the association on average and 24% of the association in participants with inadequate or barely adequate income; the mediating role of HL was not significant in participants with adequate or more than adequate income (*p* = 0.54). As for the analyses of moderators, the mediating role of HL in the association between education and SRH was not significantly moderated by age, sex and chronic diseases (p\> 0.05, results not reported). Similar results were observed in the analysis of moderators of the association between financial status and SRH: the mediation effect of HL was not significantly moderated by age, gender and chronic diseases (p\> 0.05, results not reported). # Discussion The aim of the study was to evaluate whether functional health literacy constitutes a pathway by which socio-economic determinants affect health. This hypothesis was tested in a population-based sample using education level and financial status as socio-economic determinants, self-reported health as outcome measure, health literacy as mediator, and age, sex or presence of chronic diseases as potential effect moderators. Results of logistic regression models showed the existence of positive associations among socio-economic determinants, functional HL and SRH. In particular, lower education and worse financial status were independently associated with a worse SRH and with lower functional HL levels, and inadequate or at-risk HL emerged as a predictor of poor SRH. The mediation analysis models showed that functional HL partly mediates the effect of the socio-economic determinants on health, and that the proportion of the overall effect of socio-economic determinants on SRH mediated by functional HL is higher among study participants belonging to lower socio-economic classes. Lastly, no moderation effects of age, sex or presence of chronic diseases on the proportion mediated by functional HL were found, suggesting that the observed pattern is robust across demographic characteristics and “objective” health status. As regards the interrelationships among socio-economic determinants, HL and health, results of our study are in line with the scientific evidences available to date. Indeed, as for the relationship between SES and health, our findings confirm that SES disparities predict health disparities; also in accordance with the literature, our findings confirm that SES and HL are positively linked and that HL is a predictor of health status. These findings confirm the need of exploring the potential mediating role of HL in the relationship between socio- economic factors and health investigated by the present study and by few other studies. Regarding the mediating role of HL in the relationship between socioeconomic factors and health status, our findings showed that functional HL may constitute one of the possible pathway by which socio-economic factors influence health status. This result is in line with the literature, although it should be pointed out that only very limited studies have investigated the mediating role of HL to date. Should this result be confirmed in futures studies, HL would be entitled to be listed together with the other established mediators that link SES determinants to health status such as behaviors and lifestyles, social and environmental exposures and access to, use of, and quality of health care. Besides deepening the research concerning the underlying mechanisms linking SES to health status, the mediating role of HL may have important implications for the interventions aimed at reducing health disparities as HL can be modified more easily than the SES determinants. Policies and interventions aimed at increasing the level of HL in the population or that take people’s low HL into account might effectively contribute to reduce health inequalities. Furthermore, the results of the moderated mediation analysis highlighted that functional HL constitutes a more important pathway among those with a lower socio-economic status than among those with higher socio-economic status, in which functional HL resulted to play a limited (and possibly, not significant) role. A similar finding was reported by the study of Van der Heide and collaborators in which the mediation effect of health literacy resulted to play a larger role in lower education levels; however, due to the paucity of data to date, further research is needed to confirm the presence of a socio-economic gradient in the mediating role of HL. The analysis of the relative weight of the mediating role of HL among socio-economic classes may help identify population groups that may benefit most from intervention aimed at improving HL. As far as the analysis of moderators is concerned, the proportion of mediated effect by functional HL was not found to vary according to the participants’ age, sex, or presence of chronic diseases, suggesting that the role of functional HL as a mediator of the effect of socio-economic determinants on health status is relatively constant across population subgroups defined in terms of demographics and objectively measured health status. However, it should be underlined that only few effect moderators have been explored in this study and that this study is the first to examine effect moderators of the mediating role of HL in the literature, to the best of our knowledge. Several other potential moderators may affect the mediating role of HL in the pathway through which SES determinants affect health as several third variables—such as ethnicity, occupation, perceived social status or specific chronic diseases—may potentially influence the relationship between SES and HL and/or the relationship between HL and health. As the identification of moderators of the mediating role of HL may help to identify population subgroups who will benefit most from targeted HL interventions, further and more in-depth studies on effect moderators are needed to confirm our results and to explore the presence of other effect moderators. Our study has some limitations that should be acknowledged. First, data cannot be considered representative of the overall Italian or Florentine adult population as the population-based sample was obtained through a combination of convenience and probability sampling procedures; indeed, while participants were randomly selected from a list of residents available from the registers of the GPs, the GPs were selected on a convenience basis. This may represent a major limitation for external comparison of the study results, since the convenience sampling of GPs may have introduced a selection bias. However, since the included GPs were based in different districts of the city, the geographical coverage of the sample included residents of different areas of the city, thus partially counterbalancing the selection bias detailed above. Secondly, as the variables have been self-reported by the participants, a social desirability bias may have affected the accuracy or completeness of the information retrieved, especially for the reported financial status. However, the telephone interview may have limited this potential bias. Thirdly, with regard to self- reported financial status, it should be underlined that the perceived financial status may be influenced by a broad set of factors; in particular, age and economic conditions experienced at the time of labor market entry by each birth cohort may shape the perception of the financial status. However, it should be pointed out that the subjective perception of financial status is not only consistently and robustly related to objective financial capacity but also predicts health-related outcomes and SRH even after controlling for objective economic factors. Lastly, the study had a high non-participation rate that may be attributable to the participants’ recruitment method—*i*.*e*. postal mail and phone call invitations—as most of the selected people resulted unreachable (no answer). This issue may have introduced a non-participation bias; however, it should be underlined that participants and non-participants differed only moderately in terms of age distribution. Concerning potential areas for future research, it should be pointed out that only a uni-dimensional aspect of HL (*i*.*e*. functional HL) was assessed in the present study and that the mediating role of HL may be different if assessed with the use of multidimensional HL measures. Indeed, different HL dimensions may be differently linked to both socio-economic determinants and health status. The use of multidimensional HL measures would allow a better definition of the role of HL—and its sub-dimensions—as a potential mediator of the effect of socio-economic determinants on health. Another potential area of interest for future research is the evaluation of specific and objective health outcome measures in order to better define the mediating role of HL in the different health domains. Lastly, our study could be further improved by simultaneously taking into account the other potential mediators of the relationship by which socioeconomic factors influence health status in order to evaluate the role and positioning of HL as a mediator in a broader framework. # Conclusion Our findings suggest that functional health literacy may serve as a pathway by which socioeconomic status affects health status. If confirmed, these findings would not only provide insight into the underlying mechanisms by which socio- economic disparities contribute to health differences, but would also suggest that functional HL may be a valuable target to address health inequalities. Indeed, HL may be more easily modified compared to the main established socio- economic determinants of health inequalities, on which it is often difficult to act. Policies and interventions aimed at increasing the level of HL in the population (or that otherwise take into account people’s HL) may be an effective means to reduce health inequalities. Therefore, a thorough characterization of the attributes of HL as a mediator of the relationship between socioeconomic status and health is critical in order to orient, tailor and maximize any public health effort aimed at reducing health inequalities. However, there is very limited research on the topic and further studies that take into account possible effect moderators, multidimensional HL measures and specific health outcome measures are needed to better define the role of HL in mediating the relationship between socio-economic determinants and health status. # Supporting information The authors gratefully acknowledge Drs. Flavio Godi, Mauro Grazini and Poste Italiane for their assistance in the study implementation and conduct. The authors also thank the subjects whose participation made this study possible The members of the Florence Health Literacy Research Group are as follows: Elisabetta Alti³, Sergio Baglioni³, Angela Bechini¹, Leonardo Bellino³, Niccolò Berzi³, Jacopo Bianchi², Sara Boccalini¹, Guglielmo Bonaccorsi¹, Giuseppe Burgio³, Alessandro Bussotti⁴, Marco Del Riccio², Martina Donzellini², Angela Galdiero⁵, Alessandro Grassi³, Tommaso Grassi², Vieri Lastrucci¹, Arrigo Lombardi³, Chiara Lorini¹, Sarah Mantwill⁶, Federico Manzi², Alessandro Mereu³, Donatella Messina³, Chiara Milani², Diana Paolini², Marco Targonato³, Marco Toccafondi³, Gino Sartor², Virginia Vettori¹. Department of Health Science, University of Florence, Florence, Italy School of Specialization in Hygiene and Preventive Medicine, University of Florence, Florence, Italy General Practitioner, Local Health Unit of Central Tuscany, Florence, Italy Careggi University Hospital, Florence, Italy Local Health Unit of Central Tuscany, Florence, Italy Department of Health Sciences & Health Policy, University of Lucerne, Lucerne, Switzerland [^1]: The authors have declared that no competing interests exist. [^2]: ¶ The complete membership of the author group can be found in the Acknowledgments section

# Introduction Approximately 14–25% parents with dependent children are diagnosed with cancer which can have a major impact on the entire family. Cancer patients parenting minor children experience increased levels of stress and anxiety compared to patients without minor children. Additional to the burden of the life-limiting disease and its treatment, parents with cancer worry about how to maintain family life and their role as a “good” parent and supporter. Parents often feel insecure if, when and how to communicate with their children about cancer and how to adequately address their children’s needs. Children of parents having cancer experience major challenges in their family routine and increased psychosocial stress. Even without knowing, they feel that something serious is going on. Providing age-appropriate information and timely communication about parental cancer can decrease the risk of developing negative psychological and physical consequences in affected children. Healthcare professionals (HCPs) have a significant role in identifying patients parenting minor children, their specific needs and—if necessary—initiating supportive, psychosocial care. In order to provide high-quality, patient-centred cancer care, involvement of family and their specific needs is essential. Family members are often the primary support for cancer patients and act as caregiver and thus are impacted by cancer as well. As family communication is associated with relationship functioning and adjustment to the cancer diagnosis, it is essential for HCPs to provide support to cancer patients and their families on family communication issues, e.g., open communication. In order to identify potential cancer patients parenting minor children, it is key to know about the patient’s family status and if applicable to proactively address child- and family specific themes within cancer care. Previous studies show that parents with cancer wish for support and guidance from their HCPs about child- and family-specific aspects, especially on communication with their children. However, current results show that less than 50% of HCPs routinely communicate about child- and family-specific themes with their patients. Barriers of HCPs to include child- and family-specific aspects routinely in cancer care are e.g., lack of specific competencies and knowledge as well as time pressure or structural barriers. Additionally, other studies report that HCPs feel insufficiently trained in providing basic adequate psychosocial support to cancer patients parenting minor children. In order to address these major barriers in HCP’s communication about child- and family-specific aspects, adequate trainings are needed to improve communication skills and competencies for HCPs in oncology. Over the last decade, various communication skills trainings (CSTs) have been developed and implemented to improve communication skills in oncology. Findings indicate improvements in HCP’s communication skills, namely increasing empathy, knowledge and self-efficacy or in certain patient-reported outcomes, e.g., patient satisfaction. Considering the described relevance and specific burden of affected parents, CSTs should also address these aspects. However, despite many CSTs being developed for HCPs in oncology in recent years, it remains unclear whether and to what extent child- and family-related aspects are addressed in these CSTs and previous reviews on CSTs have not included this topic. To close this gap, this systematic review aims to a.) provide an overview of existing CSTs for HCPs working in oncology addressing child- and parent-specific aspects in cancer care, b.) explore reported outcome measures associated with the CSTs and c.) gather existing evidence of effectiveness of these trainings. # Materials and methods The systematic review was registered in the International Prospective Register of Systematic Reviews (PROSPERO, registration code: CRD42020139783) and follows the updated guideline for reporting systematic reviews (PRISMA 2020 statement). ## Data sources and search strategy An electronic literature search was performed in the databases of PubMed, Cinahl, PsycInfo and Web of Science with no limitation regarding the publication year. The search was conducted on December 9^th, 2020, was developed in PubMed and adapted to the other databases. A search update was conducted on December 3^rd, 2021 and on August 12^th 2022. A librarian of the Central Medical Library Hamburg was consulted to review the final search strategy. The systematic search strategy consisted of a combination of different terms and keywords from the following four domains: (i) communication skills training, (ii) healthcare professional, (iii) oncology, and (iv) parent/family. Articles on pediatric oncology as well as qualitative studies were excluded. Our primary electronic search strategy was complemented by a hand search, consisting of citation tracking of included articles. ## Eligibility criteria and study selection Due to language restriction of the authors, peer-reviewed publications in English or German were retrieved. We included studies reporting any type of CST with a pre-post design (e.g., single arm intervention studies or studies including a control group) regarding outcomes assessing change of communication competencies, comprising at least one module on child- or parent-specific aspects in cancer care for HCPs caring for adult cancer patients. The applied in- and exclusion criteria are displayed in. However, despite our extensive search strategy only two studies were identified during the study selection process to focus on child- and parent-specific aspects within their CSTs. Therefore, we decided to broaden the focus of this systematic review and to include studies, which entail a child- and family-specific module within their CST. To manage and facilitate the selection process, search results were imported into the reference management software EndNote (Version EndNote X9.3.2) and duplicates were removed. One author (WF) conducted the title and abstract screening. All potentially relevant articles according to the defined inclusion and exclusion criteria were included for full text screening. Full texts were independently assessed for eligibility by two reviewers (WF, WG). Disagreement between reviewers was resolved by discussion; where necessary, a third reviewer (LI) was consulted. ## Data extraction and synthesis As we were expecting a large heterogeneity of studies including a large variation in participants, outcome measures or type of CST being used, we synthesized findings of the included studies in the form of a narrative review. A data extraction form was developed including the following information: aims/background, study design and methods; details of CST (e.g., development, setting, duration, content, teaching strategies); details on child- and family- specific module; characteristics of participants; CST outcome measures and results. The form was independently pilot tested by two reviewers (WF, WG) with one randomly selected study included in this review. Data extraction of included studies was systematically performed by one reviewer (WF), final results were discussed with two other reviewers (WG, LI). Intervention outcomes and findings were categorized based on Kirkpatrick’s framework for training evaluation based on the following levels: 1. Reaction–Participant’s satisfaction with the training; 2. Learning–Participant’s change of attitudes, increase in knowledge and skills; 3. Behavior–Participant’s change in behavior; 4. Results–other improvements in patient-oriented healthcare (e.g., participants well-being). ## Quality assessment Methodological quality of included studies was independently assessed by two reviewers (WF, WG) using a slightly modified version of the National Institutes of Health (NIH) quality assessment tool for pre-post studies without control group. This tool was selected as all included studies were quasi-experimental studies with a pre-post design. none including a control group. Study quality could be rated as good, fair or poor. Any disagreement between reviewers was resolved by discussion and, where necessary, a third reviewer (LI) was consulted. # Results The main literature search identified two studies specifically addressed the subject of cancer patients parenting minor children within their CST and five studies incorporated a brief family module within their CST. The first update added another two studies evaluating a CST for HCPs in oncology, including a brief module on family-specific aspects in cancer care. In total, nine studies were included in this review. ## Description of included studies gives an overview of included studies. All included studies were published between 2008 and 2021. Five studies were conducted in the North America, two in Australia, one in Africa and one in Europe. Included studies used a quasi- experimental design with pre-post measurement only and no studies were identified including a control-group. ## Participants Most studies included qualified HCPs, two studies included nursing students only. In total, 1578 HCPs participated in the included studies. Six studies including nursing professionals only, three studies including HCPs of various disciplines (e.g., nurses, doctors, social workers), in which there was a high proportion of nurses. In two studies only female HCPs participated, four other studies included mainly female participants (range 75–97%). In studies reporting on mean age of participants, mean age ranged from 24 to 47 years. Professional experience varied from overall working experience to working in oncological setting or in palliative care. Two studies assessed previous communication skills training participation. One study assessed if participants had currently a serious illness in family member (34%) and previous history of bereavement of a first degree relative (51%). ## CST characteristics Of the nine included studies, one training was an e-learning training and one a webinar series. The remaining CSTs were face-to-face trainings. The duration of the programs varied substantially in length, ranging from 30–40 minutes to a 2-day program. Group size of trainings varied between studies, with small groups of n = 3–8 participants each and large groups of e.g., up to n = 158 participants per training. The content of the CSTs was either developed based on a literature review, a needs assessment (e.g., focus group or survey) or input through a workshop with experts. One study reported on pilot-testing their intervention. Detailed description of the CSTs, outcome measurements and results of the included studies are presented in. ## Content and development of the CSTs including a child- or family-specific module Of the nine studies included, only two included a CST for HCPs specifically addressing the subject of cancer patients parenting minor children. The remaining seven studies incorporated a brief family module within their CST. This brief family module often entailed themes e.g., how to communicate with families of cancer patients or how to involve the family in cancer care. Detailed information if and in which way the family modules refer to children as relatives in particular or if parental issues were covered was not reported within the studies. Two studies applied the COMSKIL training program and four the original or an adapted version of COMFORT ^{TM SM} Communication Curriculum, a CST specifically designed for nurses. Three studies developed their own CST program. The description of the family module differed slightly between studies using the original COMFORT curriculum without further information or explanation for possible variations in content of their CST within their reports. Fuoto et al. described their module as “Family module: support caregiver involvement and understanding” and Wittenberg et al. (2021) “F-Family caregivers”. ## Didactic techniques/materials All included studies combined various didactic techniques and materials within their training program. Role-play exercises with regular feedback were part of five studies, role-play exercises with simulated patients were incorporated in three studies. All studies but the study using the e-learning gave some kind of presentation (e.g., power-point introduction on training or overview of communication skills). Discussion rounds were part of the training in four studies and videos e.g., to illustrate key skills or family needs included four studies. Moreover, various studies used written material in form of manuals, booklets or pocket-cards. Five studies reported on professional background of CST facilitators, which varied greatly between studies (for details see). ## Outcome measurement Included studies varied considerably in defined outcomes and applied instruments (e.g., number of items, scales, description of adapted instruments). Most instruments have been self-developed without validation (see for details). Two studies applied Kirkpatrick’s framework for training evaluation, focusing on the first two levels: participant’s reaction and learning. ***Participants’ satisfaction*** with the CST was assessed post-training participation. Five studies evaluated satisfaction using quantitative evaluation surveys and two qualitative methods (e.g., open-ended questions; focus groups). Some studies assessed overall satisfaction with CST, others assessed satisfaction with individual sessions/modules. Overall, assessment of participants’ satisfaction varied considerably. Majority of studies (n = 8) included a pre-post participation assessment of ***self-efficacy and/or perceived confidence in communication competencies***. Three studies analyzed change of HCPs’ ***attitudes*** (e.g., towards the importance of learned skills), three studies analyzed change of HCPs’ self- perceived ***communication behavior*** in daily practice and three ***observed communication skills*** assessed through simulated patient assessments (SPAs) pre-post training participation. Two studies measured change in ***perceived importance*** of communication. Four studies assessed change of ***knowledge*** how to support parents and families, e.g., knowledge on palliative care or retrospectively perceived increase of knowledge on supportive needs of parents and families. HCPs’ ***general health*** (burnout and perceived stress as secondary outcomes) and ***patient-reported outcomes*** (adapted version of the patient-family satisfaction with End-of-Live care survey (FEPC), a post-death survey for relatives originally developed by the Natioanl Hospice and Palliatve care Organization in Virginia, USA, however not available) were each reported in one study. ## Evaluation of CST following Kirkpatrick’s framework for training evaluation *Reaction–Participant’s satisfaction with the training*. Overall, participants’ reaction to CST was predominantly positive. Participants rated the trainings beneficial for applying it to their daily practice, to increase their confidence and would recommend it to their colleagues. Reported suggestions were e.g., increasing group sharing exercises or discussion/exchange rounds to share their experiences with affected families. *Learning–Effects on participant’s communication confidence*, *attitudes and knowledge*. Statistically significant improvement on participants’ self-reported ***self-efficacy in communication competencies*** were found in seven studies with considerable variation in defined outcomes and applied instruments (see for details). One study did not report detailed statistic parameters. Two of the three studies assessing participants’ ***attitudes*** reported significant improvements over time. Only one of the two studies assessing ***perceived importance*** of communication found significant improvements over time. Regarding ***knowledge***, three studies reported significant improvements over time, with one study missing clear and detailed statistic parameters. *Behavior–Participant’s change in behavior*. Of the three studies assessing daily ***communication behavior***, only one study reported on significant changes, but did not provide statistic parameters. Semple et al. assessed change of communication behavior only at post-participation without comparison over time and Fuoto et al. with an open-answer format only. Significant changes in ***observed communication skills*** were found in three studies. Banerjee et al. and Cannity et al. reported significant improvements for overall skills using both the same Comskil coding manual, Turner et al. for five of their six categories on measuring General Interaction skills and responses to Scripted Cues. *Results–other improvements in patient-oriented healthcare*. One study assessed participants’ ***general health*** using the General Health Questionnaire 28 (GHQ), the level of perceived stress (self-administered) and burnout with the Maslach Burnout Inventory (MBI). There were no significant changes in stress and burnout or level of perceived stress. Significant decrease in the somatic subscale of the GHQ was reported. Regarding the patient-reported outcomes measuring patient-family satisfaction with care no significant differences between pre- and post-training scores were found. Both studies specifically focusing their CST to provide support for cancer patients parenting minor children found significant changes within the pre-and post training assessment for multiple outcomes. ## Methodological quality assessment The methodological quality of included studies was rated as “fair” in six, “poor” in two and “good” in only one included study. None of the included studies reported on a sample size calculation, the statistical methods of two studies were of poor reporting quality, the eligibility criteria for participants were only partly or not described in eight studies, outcome measures were not or only partly reported in all studies, and only two studies reported on consistent delivery of intervention. # Discussion This review aimed to provide an overview of existing CST interventions for HCPs in oncology explicitly addressing child- and parent-specific aspects in adult cancer care. Second, the review aimed to assess reported outcome measures associated with the CST’s evaluation. The third aim was to report on CST effectiveness. Since only two studies were identified explicitly reporting on a CST solely focusing on parental cancer, we broadened our focus during the screening process to also include studies reporting on a family-specific module within their CST. Thus, in total, we included nine studies with at least one module on child- or family-specific aspects in communication in cancer care. The seven included studies including a family-specific module did not provide details what is included (e.g., parental-specific aspects during cancer care). Hence, it remains unclear if and to which extend children as relatives of cancer patients are explicitly addressed. Findings of the present work are consistent with previous research identifying a lack of communication skills trainings in oncological care especially for HCPs caring for patients experiencing additional burden and needs. In our included studies, nurses represented a large proportion of participants with six studies including nurses only and two studies mainly including nurses. This is not surprising as one frequently evaluated CST is the COMFORT curriculum explicitly developed for nurses. As nurses spend a considerable amount of their working time caring for patients, developing a close relationship with their patients and relatives, they are often confronted with patient’s specific needs and provide emotional support. Additionally, shortage of nursing staff globally and a continuous physically and emotionally draining job increase the need to enhance effective communication with patients and their families to reduce stress experience and emotional exhaustion in nursing profession. Physicians usually are the key contact and person of trust for patients during cancer care. Therefore, they can act as gatekeepers for additional support according to child- and family-related needs. However, in the included studies only few physicians participated. Studies on child- and parental-related issues report lack of knowledge and specific communication skills as well as perceived limited competence on parental issues in clinicians in cancer care. This strongly indicates a need for 1) specifically developed training programs for physicians and oncologists incorporating child- and parent-specific aspects or 2) optimization of access to existing interventions to improve participation of physicians, e.g., by including incentives or adapting trainings to their specific needs and working schedule. Six of the included studies found significant improvements in either self- efficacy and/or confidence, behavior and knowledge for general communication skills, two additional studies for specific communication aspects in parental cancer. This implies that CSTs are a promising approach to improve HCPs communication skills including specific skills on parental cancer and support building a bridge to communicate effectively with affected parents and their families. This implication is supported by previous research, indicating increased self-efficacy, knowledge and skills will in turn improve (a) HCP’s communication behavior, (b) HCP’s satisfaction with communication and their mental well-being health (e.g., reduced emotional burn-out), and (c) outcomes for patients and their families (e.g., reduced stress and feelings of anxiety, improved satisfaction with care). However, findings are not generalizable due to small sample sizes in most studies included in this review and only two included studies applying a specific CST on parental cancer. The overall methodological quality of included studies was fair to poor. Applied outcome measures varied considerably and psychometric properties of measures were insufficient. However, validated and reliable tools assessing specific communication skills and behavior in child- and family-specific aspects in cancer care are rare. Hence, there is a need for rigorously developed and psychometrically sound instruments. Moreover, objective simulated patient assessments (SPAs) should be included in future studies as they are the gold standard for evaluation of CSTs. Clinical case vignettes, as used in one included study, have been found to be comparable to SPAs. However, development of vignettes should be standardized and follow current recommendations. ## Study limitations This study has several limitations. First, this systematic review focused on CSTs with a specific module on child- or family specific aspects in cancer care. Though our search strategy was extensive, the articles reviewed may not represent all CSTs with such specific modules in cancer care given the restrictions of search terms used, databases searched and requirements for English- or German-language due to language restrictions of the authors. However, by including a thorough secondary literature search, additional relevant CSTs were included. Second, as included studies varied considerably in e.g., CST content and outcome assessment and tools used, comparison of CSTs and their quality of evidence is difficult and generalizability is impeded. Additionally, based on our quality assessment, only one study with good methodology design was included. ## Clinical implications Overall, implication for future research is to develop a structured and theory- based communication skills intervention for HCPs in oncology to improve family- centered cancer care, specifically when a parent has cancer. Future studies should develop specific trainings to enhance HCPs communication skills, knowledge and self-efficacy to address child- and family-specific aspects when a parent has cancer. Also, these studies should provide an evaluation using state of the art methodology (e.g., including a control group thorough outcome assessment with validated, and pilot-tested outcome measurements based on e.g., Kirkpatrick’s model of evaluation). Additionally, newly developed interventions should specifically address physicians and oncologists and if possible be adapted to their needs to increase participation of this specific HCP group. Existing studies including a family-specific module should provide further detail on the topic of “family communication”, e.g., if minor children are included as family members. # Conclusion This systematic review gives an overview of existing CSTs for HCPs on parenthood and cancer. Despite a high need for a specific CST to improve HCP’s communication skills regarding parental cancer, only two CSTs focusing on parental cancer were identified, the remaining seven studies only included a brief module on family communication. The quality of evidence for included studies remains insufficient. Due to the lack of specific CSTs and poor or only fair quality of the included studies, further CSTs on aspects of parental cancer should be developed and evaluated rigorously. # Supporting information COMFORT COMFORT TM SM Communication Curriculum COMSKIL COMSKIL training program CST Communication skills training GHQ General Health Questionnaire 28 HCPs Healthcare professionals MBI Maslach Burnout Inventory NIH National Institutes of Health 10.1371/journal.pone.0277225.r001 Decision Letter 0 Silva Junior Manoelito Ferreira Academic Editor 2022 Manoelito Ferreira Silva Junior This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 2 Aug 2022 PONE-D-22-11030Child- and family-specific communication skills trainings for healthcare professionals caring for families with parental cancer: a systematic reviewPLOS ONE Dear Dr. Frerichs Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ============================== ACADEMIC EDITOR: Thank you for choosing Plos One to send the manuscript. According to the reviewers' evaluation, the manuscript is adequate, and only needs a few minor corrections for its acceptance. Therefore, I encourage authors to carefully review the points raised by the reviewers. ============================== Please submit your revised manuscript by  Aug 15, 2022. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Manoelito Ferreira Silva Junior, Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <h ttps://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_b ody.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. Additional Editor Comments (if provided): Thank you for choosing Plos One to send the manuscript. According to the reviewers' evaluation, the manuscript is adequate, and only needs a few minor corrections for its acceptance. Therefore, I encourage authors to carefully review the points raised by the reviewers. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The study presents an adequate methodological structure. Some recommendations will be made to better adapt the proposal and understand the development of the article. 1\. It is suggested to update the databases. The last update is more than 6 months old. 2\. Justify the language restriction. It was important to consider it as a limitation of the study. 3\. Justify the exclusion of other types of quantitative studies. Including justifying the absence of a control group, which allows better comparability of training results. In the results, 2 studies include students in the sample. It is important to point out if the results are treated differently in the studies, in relation to the professionals. In table 4, some numerical data can be inserted into the results for better visualization of the data presented. The review brings only 2 studies that address the relationship between parents and children. Most other articles focus on general family communication guidelines. Thus, I understand that the analyzed articles extrapolate the objective "a" proposed in the review. Can be reviewed. In addition, the Introduction directs toward understanding the relationship. Depending on the targeting of the objective and the results, the text needs to be reorganized, expanding the justification for family communication. The analysis of studies that consider the relationship between parents and children may be highlighted in a sub-analysis. The clinical implications should consider the importance of the study for training that considers the family relationships of parents and children in the oncological context, which is the central objective of the presented study. Reviewer \#2: The topic researched by this article (Communication skills trainings (CST) in oncology) is extremely relevant for the scientific community and for healthcare professionals. The authors meet the PLOS criteria for publication and responded to the objectives proposed in the research. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0277225.r002 Author response to Decision Letter 0 2 Sep 2022 Editor comments: 1\. Thank you for choosing Plos One to send the manuscript. According to the reviewers' evaluation, the manuscript is adequate, and only needs a few minor corrections for its acceptance. Therefore, I encourage authors to carefully review the points raised by the reviewers. � Thank you for the opportunity to review and resubmit our manuscript. We carefully revised the manuscript according to the reviewer’s comments and also revised the Abstract, as word count was exceeded. We hope that it is now considered appropriate for publication in PLOS ONE. Response to reviewer comments Reviewer \#1: The study presents an adequate methodological structure. Some recommendations will be made to better adapt the proposal and understand the development of the article. � Thank you for your valuable peer review. 1\. It is suggested to update the databases. The last update is more than 6 months old. � Thank you for pointing this out. As suggested, we did an update of the search within the databases on August 12th and updated “Fig 1. PRISMA flow diagram for systematic reviews” as well as the Methods section accordingly. In summary, a total of n=1455 additional records were identified, of which n=713 screened for title abstract screening, and of which n=2 full-text articles were assessed for eligibility. N=0 articles met inclusion criteria and therefore no additional articles were included in the qualitative synthesis. 2\. Justify the language restriction. It was important to consider it as a limitation of the study. � Thank you for pointing this out. We have revised the manuscript and included the language restriction of the authors within the Methods - Eligibility section (“Due to language restriction of the atuhors, peer reviewed …”) and further pointed it out as a study limitation within the discussion section (“... requirements for English- or German-language due to language restrictions of the authors.”) 3\. Justify the exclusion of other types of quantitative studies. Including justifying the absence of a control group, which allows better comparability of training results. � Thank you for raising this concern. We did not exclude other types of quantitative studies and would have included studies incorporating a control group, if there were any identified within our search. We agree, that our description can benefit from more clarity and therefore have added a sentence within the Methods – Eligibility criteria and study selection section “… reporting any type of CST with a pre-post design (e.g., single arm intervention studies or studies including a control group) regarding outcomes … “. Further, we clarified the section Description of included studies by adding following information “Included studies used a quasi-experimental design with pre-post measurement only and no studies were identified including a control-group.” 4\. In the results, 2 studies include students in the sample. It is important to point out if the results are treated differently in the studies, in relation to the professionals. � Thank you for raising this concern. As in these two studies the sample consisted of students only, the results could not be treated differently. However, we realized that our description of the two studies entailing students only was limited and therefore revised the manuscript to make it more clear: “Most studies included qualified HCPs (23, 31, 33-36, 38), two studies included nursing students only (32, 35) (28, 31) (see Table 3)”. 5\. In table 4, some numerical data can be inserted into the results for better visualization of the data presented. � Thank you for this suggestion. We agree that the visualization of the data presented so far was not optimal. Therefore, we added significant numerical data to table 4 to aid understanding of presented data of the following authors: o Banerjee et al 2017 regarding outcomes on participants learning; o Cannity et al 2021 regarding outcomes on participant learning; o Fuoto & Turner 2019 regarding outcomes on communication confidence, communication satisfaction; o Quinn et al 2008 regarding outcomes on Generic Palliative Care questionnaire; o Semple et al 2017 regarding outcomes on perceived confidence and competence to communication; o Turner et al 2009 regarding outcomes on General Health Questionnaire, perceived stress, attitudes and confidence, clinical vignettes, simulated patient interviews and subgroup analyses; o Wittenberg et al 2020 regarding outcome one numerical data on pre- post-test on attitudes, knowledge and behavior; however, it is not clear how this was analyzed and no overview (e.g., table) of data is provided; 6\. The review brings only 2 studies that address the relationship between parents and children. Most other articles focus on general family communication guidelines. Thus, I understand that the analyzed articles extrapolate the objective "a" proposed in the review. Can be reviewed. � Thank you very much for this important point. During the search process only two studies focused on child-and parent-specific aspects only. As the remaining seven studies had at least a family-specific module and were identified through our search strategy (which entailed a search term “famil”), but did not further specify the content of the family-module, e.g., if communication with minor children of cancer patients were included within the module, we decided to broaden the scope to child-and family-specific communication skills trainings. But, we agree, that the objective is misleading and have therefore revised the manuscript as described below. 1\. Revised the objective “a” accordingly in the INTRODUCTION section: “a.) provide an overview of existing CSTs for HCPs working in oncology addressing child- and parent-specific aspects in cancer care,” 2\. Have revised the METHOD section why we also include family-specific communication trainings. \- “However, despite our extensive search strategy only two studies were identified during the study selection process to focus on child- and parent- specific aspects within their CSTs. Therefore, we decided to broaden the focus of this systematic review and to include studies, which entail a child- and family-specific module within their CST.” \- And included within the Table 2. Inclusion and exclusion criteria the criteria: Studies evaluating a communication training or educational program including at least a module on child-, parent- or family-specific themes; 3\. Revised the RESULT section accordingly to highlight the results of the content of the training \- by highlighting the two specific trainings on parents with cancer at the beginning of the RESULT section: “The main literature search identified two studies specifically addressed the subject of cancer patients parenting minor children within their CST and five studies incorporated a brief family module within their CST. The first update added another two studies evaluating a CST for HCPs in oncology, including a brief module on family-specific aspects in cancer care. In total, nine studies were included in this review (Fig 1) \- By addressing the significant results of the two specific trainings on parental cancer: “Both studies specifically focusing their CST to provide support for cancer patients parenting minor children found significant changes within the pre-and post training assessment for multiple outcomes (23, 38) (see Table 4).” 4\. Revised the Discussion section accordingly to make the objective and results of the review clear: “This review aimed to provide an overview of existing CST interventions for HCPs in oncology explicitly addressing child- and parent-specific aspects in adult cancer care. Second, the review aimed to assess reported outcome measures associated with the CST’s evaluation. The third aim was to report on CST effectiveness. Since only two studies were identified explicitly reporting on a CST solely focusing on parental cancer, we broadened our focus during the screening process to also include studies reporting on a family-specific module within their CST. Thus, in total, we included nine studies with at least one module on child- or family-specific aspects in communication in cancer care. The seven included studies including a family-specific module did not provide details what is included (e.g., parental-specific aspects during cancer care).” 7\. In addition, the Introduction directs toward understanding the relationship. Depending on the targeting of the objective and the results, the text needs to be reorganized, expanding the justification for family communication. The analysis of studies that consider the relationship between parents and children may be highlighted in a sub-analysis. � Thank you for this valuable comment. We agree and have adjusted the introduction by expanding to family communication within cancer care including applicable references. We hope in combination with adjusting comment No 6, these changes lead to a clearer understanding of the importance of family communication within cancer care, specifically when cancer patients parent minor children. Adjustments to the INTRODUCTION: “In order to provide high-quality, patient-centred cancer care, involvement of family and their specific needs is essential (15, 16). Family members are often the primary support for cancer patients (17) and act as caregiver and thus are impacted by cancer as well (18). As family communication is associated with relationship functioning and adjustment to the cancer diagnosis (18), it is essential for HCPs to provide support to cancer patients and their families on family communication issues, e.g., open communication. In order to identify potential cancer patients parenting minor children, it is key to know about the patient’s family status and if applicable to proactively address child- and family specific themes within cancer care.” 8\. The clinical implications should consider the importance of the study for training that considers the family relationships of parents and children in the oncological context, which is the central objective of the presented study. � Thank you for this important feedback. We have adjusted the CLINICAL IMPLICATION section accordingly by including a specific focus on family communication when cancer patient parent minor children and regrouped the argumentation. “Overall, implication for future research is to develop a structured and theory- based communication skills intervention for HCPs in oncology to improve family- centered cancer care, specifically when a parent has cancer (38, 43). Future studies should develop specific trainings to enhance HCPs communication skills, knowledge and self-efficacy to address child- and family-specific aspects when a parent has cancer. Also, these studies should provide an evaluation using state of the art methodology (e.g., including a control group thorough outcome assessment with validated, and pilot-tested outcome measurements based on e.g., Kirkpatrick’s model of evaluation) (29, 50). Additionally, newly developed interventions should specifically address physicians and oncologists and if possible be adapted to their needs to increase participation of this specific HCP group. Existing studies including a family-specific module should provide further detail on the topic of “family communication”, e.g., if minor children are included as family members. (29, 50).” Reviewer \#2: The topic researched by this article (Communication skills trainings (CST) in oncology) is extremely relevant for the scientific community and for healthcare professionals. The authors meet the PLOS criteria for publication and responded to the objectives proposed in the research. → Thank you for your valuable peer review. 10.1371/journal.pone.0277225.r003 Decision Letter 1 Silva Junior Manoelito Ferreira Academic Editor 2022 Manoelito Ferreira Silva Junior This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 24 Oct 2022 Child- and family-specific communication skills trainings for healthcare professionals caring for families with parental cancer: a systematic review PONE-D-22-11030R1 Dear Dr. Wiebke Frerichs We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Manoelito Ferreira Silva Junior, Ph.D. Academic Editor PLOS ONE Additional Editor Comments: I thank the authors for the effort to answer each of the points raised by the evaluators. Therefore, the items included in the final version of the file, and the answers included in the authors' letter, I inform you that the decision is to approve in the current format. Best Regards. [^1]: The authors have declared that no competing interests exist.

# Introduction Asthma is one of the most common chronic diseases in children, and it is characterised by bronchial hyperresponsiveness (BHR) and reversible airway obstruction. Both genetic and environmental factors play important roles in the development of asthma. Although more than 100 genes have been associated with asthma, most of these associations have proven to be non-replicable in multiple populations, which indicates a complex genetic susceptibility pattern. Integrin β3 (ITGB3) is a serotonin-related gene on chromosome 17 that encodes a beta chain integrin subunit. Integrins are known to participate in cell adhesion and cell surface-mediated signalling. Recent investigations have suggested that ITGB3 is involved in the pathogenesis of asthma, particularly in early childhood. Five SNPs in the ITGB3 gene have been linked with asthma in a Hutterite population, but these results were not observed in three other unrelated populations, most likely due to differences in environmental exposures in childhood. Rogers *et al*. demonstrated that few identified SNPs could be replicated in different populations. Factors such as the criteria used to diagnose asthma, environmental exposures, numbers of subjects, different patterns of linkage disequilibrium, and population stratification could be potential causes of this non-repeatability. Multiple studies in diverse populations would be helpful to identify true candidate genes. To date, few studies have been performed to investigate associations between the ITGB3 gene and asthma in Chinese Han children. Thus, it is necessary to identify single nucleotide polymorphisms (SNPs) in the ITGB3 gene associated with asthma in a Chinese population. In the present study, we investigated the association of SNPs in ITGB3 with asthma in Chinese Han children using HRM analysis for SNP genotyping. Our study revealed significant associations between polymorphisms in the ITGB3 gene and asthma risk. # Materials and Methods ## Ethical Statement This study was approved by the Medical Ethics Committee of Shandong University, and written informed consent was obtained from the parents of every participant. ## Study Population During the 6-year period from 2006 to 2012, 321 unrelated Chinese children with asthma were recruited as case subjects from the QiLu Children’s Hospital of Shandong University, with their parents’ consent; 315 unrelated healthy children were recruited randomly from Shandong province and the surrounding area as control subjects. The diagnosis of asthma in all subjects was performed by asthma specialists according to the modified criteria : (1) recurrent wheezing, coughing, shortness of breath, and chest tightness, closely related to multiple factors, such as inhaled allergens, changes in weather, physical or chemical irritants, or respiratory tract infections, that frequently occur or worsen at night and/or in the early morning; (2) scattered lung wheezing sounds upon exhalation; (3) spontaneous remission of the above symptoms with anti-asthma treatment; and (4) the absence of other diseases that could cause wheezing, coughing, shortness of breath, and chest tightness. ## DNA Extraction Each child contributed a 1 ml whole-blood sample to this study. Genomic DNA was extracted from the blood samples using a QIAamp DNA Mini Kit (QIAGEN). All DNA samples were quantified using a NANODROP 2000 spectrophotometer (Thermo) and stored at −20°C until use. ## SNP Selection and Primer Design SNP genotype data for CHB were obtained from the HapMap database and analysed using Haploview software. SNPs were selected according to the following criteria: (1) r² threshold of 0.8, as analysed by a pairwise tagging algorithm; (2) no C/G alleles; and (3) MAF\>0.05, except for the functional SNP rs5918. Of the SNPs that met the above three criteria, a total of 6 SNPs were selected for genotyping: rs2015729 (Intron 2), rs2317676 (3′UTR), rs5918 (Exon 3), rs5919 (Exon 6), rs3809865 (3′UTR), and rs10514919 (Intron 1). The primers used to amplify the selected SNPs were designed using Primer5 according to their flanking sequences, based on the following criteria: (1) melting temperature (Tm) between 60 and 65°C; (2) absence of dimerisation and mispriming capabilities; and (3) an amplicon size smaller than 100 bp, to ensure high sensibility. Prior to HRM analysis, PCR conditions were confirmed to yield a single band to avoid interference from primer dimerisation and mispriming products. ## Genotyping The PCR was performed in a final volume of 10 µL containing 25 ng DNA, 1.0 U Hot Start Taq-DNA polymerase (Takara), 1.25 µM of each primer, 0.8 µL dNTP MIX (Takara), 1.0 µL 10× PCR buffer (Takara) and 1.0 µL LC green. The amplification was carried out according to the following protocol: initial denaturation at 95°C for 5 min, followed by 30 cycles of 95°C for 30 s, annealing at 60°C for 30 s and elongation at 68°C for 30 s. All PCR amplifications were carried out in 96-well plates on a LightCycler 480 instrument (Roche) according to the manufacturer’s instructions. HRM curves were generated by monitoring the fluorescence of the sample during a temperature ramp from 65 to 95°C at 0.02°C/s. Normalised HRM curves were generated with the following normalisation regions: rs2015729, 82.22–83.33 and 89.85–90.74; rs2317676, 83.78–84.86 and 87.76–88.95; rs3809865, 82.01–83.57 and 87.6–88.99; rs5919, 81.6–83.2 and 86.42–88.47; rs5918, 82.15–84.22 and 89.79–91.44; and rs10514919, 83.04–84.48 and 88.41–90.23. HRM curves were classified into two or three distinct groups. Thirty random samples from each group were sequenced with an ABI 3100 Genetic Analyzer to confirm the accuracy of the genotyping. Samples with known genotypes were used as internal references to generate standard curves for the classification of the unknown samples. ## Luciferase Reporter Assay A 227 bp region of ITGB3 3′UTR containing the putative recognition site for rs3809865 was amplified and ligated into pHSA-MIR-REPORT (Ambion). CRL1730 (human umbilical vein endothelial cells), 293T (human embryonic kidney cells), and A549 (human lung adenocarcinoma cells) were co-transfected with 400 ng of the 3′UTR-luciferase reporter vector and 20 nM mature hsa-mir-124 (final concentration) using Lipofectamine 2000 (Invitrogen). Control cells were transfected with 100 ng pRL-SV40 plasmid (Promega) for normalisation. After a 24 h incubation, luciferase and Renilla activities were measured using a Dual Luciferase Assay Kit (Promega) according to the manufacturer’s instructions. ## The Regulation of ITGB3 by hsa-mir-124 To further confirm the regulation of ITGB3 by hsa-mir-124, A549 was transfected with an hsa-mir-124 mimic (124-M, 5′U<u>AAGGCAC</u>GCGGUGAAUGCCAAG3′) or a control miRNA (CON-M, 5′U<u>AAccCAC</u>GCGGUGAAUGCCAAG3′). After a 40 h incubation, cells were collected, washed with ice-cold PBS and then lysed with RIPA lysis buffer (Cell Signaling Technology) in an ice bath for 5 min, followed by sonication for 8 s. The protein concentration of each sample was determined using a BCA protein assay kit (Pierce) according to the manufacturer’s protocol. Samples with equal amounts (35 µg) of protein were separated via 12% sodium dodecyl sulphate polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to polyvinylidene fluoride membranes. The membranes were blocked for 1 h with 5% fat-free milk in Tris-buffered saline with 0.05% Tween-20 (TBST) and then incubated with the indicated primary antibodies overnight at 4°C. After washing three times with TBST, the membranes were incubated with horseradish peroxidase- conjugated secondary antibodies for 1 h at room temperature and washed three times with TBST. The labelled proteins were detected using the enhanced chemiluminescence method and quantified using Alpha Imager 2200. Cells were fixed in 4% paraformaldehyde in PBS for 20 min at room temperature, rinsed in PBS and permeabilised with 0.5% Triton X-100 for 10 min. Subsequently, cells were blocked with 10% goat serum in PBST (PBS containing 0.1% Triton X-100). After three washes with PBST, the cells were incubated with rabbit polyclonal anti-ITGB3 primary antibody (1∶50, Abcam) overnight at 4°C, washed ten times with PBST and incubated with secondary antibody (1∶200, Anbo) for 1 h. The cells were then washed ten times with PBST and imaged under a fluorescence microscope (Leica). ## Statistical Analysis Hardy-Weinberg equilibrium was evaluated using the Fisher’s exact test. The genotypic frequencies of all SNPs in cases and controls were compared by Chi- square (χ²) tests. The allele frequencies were compared by allelic associations test. Odds ratios with 95% confidence intervals (95% CI) and adjusted p values were calculated using allelic associations tests. The associations between genotypes and asthma were assessed using χ² tests. Statistical analysis was performed in PLINK v1.07. P\<0.05 was considered to be statistically significant. # Results In the present study, 321 paediatric asthma patients (female: 143, male: 178, median age: 6.6) and 315 controls (female: 131, male: 184, median age: 6.7) were recruited. Chi-square tests were used to assess whether the cases and controls were similar. As shown in, the case and control populations were well matched in terms of sex and age distributions. ## Detection of Amplicons The primers used to amplify the selected SNPs are listed in. PCR was performed under the aforementioned conditions, and the amplicons were analysed by 2% agarose electrophoresis. As shown in, each amplicon was unique and appeared at the expected size, indicating that the primers and PCR conditions were adequate for HRM analysis. ## Genotype Assignment The genotype assignments of the selected SNPs were determined by HRM curves, using the sequenced samples as control genotypes. The studied SNPs were successfully genotyped by HRM analysis, as shown in. All of the tested SNPs were in Hardy-Weinberg equilibrium (HWE) in both the case and control groups (P\>0.05). We first analysed the association between SNP genotypes and the risk of asthma. The frequency distributions of the genotypes are shown in. No differences in the genotype frequencies of the six SNPs were observed between the controls and cases (p\>0.05 for all). We further compared the frequency distributions of the minor alleles between the controls and cases using χ² tests. The minor allele (C) of rs5918 was present at a slightly higher frequency in the case group (MAF = 0.024) than in the control group (MAF = 0.016), with an odds ratio of 1.46 (0.65–3.28), but it showed no association with asthma (p\>0.05). The other four SNPs, including rs2015279, rs2317676, rs5919, and rs10514919, were also well genotyped but showed no significant difference between controls and cases (p\>0.05 for each). The SNP within the 3′UTR of ITGB3, rs3809865, showed a significantly different distribution between cases and controls (p = 0.0007). The frequency of the minor allele (T) was obviously higher in the control group than in the case group (MAF = 0.2 and 0.13, respectively). The odds ratio (95% CI) was 0.59 (0.43–0.8) using the major allele (A) as a reference, which suggested that this allele provided some protection again asthma. After Bonferroni correction, rs3809865 still showed a significant association with asthma (p = 0.004). ## Targeting of ITGB3 by hsa-mir-124 We used TargetScan (<http://www.targetscan.org/>) and PITA (<http://genie.weizmann.ac.il/pubs/mir07/mir07_data.html>) to identify miRNAs that were predicted to bind to the 3′UTR of ITGB3. Two miRNAs (hsa-mir-124 and hsa-mir-506) were found to have seed regions that bound to the sequence containing rs3809865. These results suggested that these two miRNAs should bind more stably to the T allele of rs3809865 than to the A allele. Thus, it was deduced that the allele of rs3809865 might affect the miRNA binding to ITGB3 and thus regulate the expression of ITGB3. Luciferase assays were performed to validate this prediction. As shown in, hsa- mir-124 significantly inhibited the expression of the luciferase gene from the vector carrying the rs3809865 T allele compared with the vector carrying the rs3809865 A allele and the control vector in all cell lines tested (CRL1730∶37.6% expression for the T allele versus 61% for the A allele; 293T: 41.1% expression for the T allele versus 76.3% for the A allele; A549∶46.9% expression for the T allele versus 84.4% for the A allele). Having established that hsa-mir-124 could interact with this putative binding site, we further investigated the regulation of ITGB3 by hsa-mir-124. The results showed that the transfection of A549 with hsa-mir-124 resulted in an obvious decrease in expression of ITGB3, as analysed by immunofluorescence and western blotting. # Discussion In the present study, HRM technology was applied to genotype the selected SNPs in the ITGB3 gene. Six SNPs were accurately genotyped using HRM. The genotyping was especially successful for rs3809865 with the similar GC content, which suggested that small amplicons were good substrates for SNP genotyping by the HRM method. Based on the HRM genotyping results, the associations between the six SNPs in ITGB3 gene and asthma were investigated in Chinese Han children. Previous studies have demonstrated that the ITGB3 gene plays an important role in asthma pathogenesis. Among the tested SNPs in ITGB3, the minor allele of rs2015729 exhibited a strong association with asthma in a Hutterite population (0.001\<p\<0.01), and the major allele showed modest association in a population from Madison (0.01\<P\<0.05), but this association was not replicated in populations from Chicago. Our results suggested that this SNP was not associated with asthma in Chinese Han children. A modest association between rs2317676 and asthma has been identified in a Caucasian population from Chicago (0.01\<P\<0.05). Furthermore, rs2317676 was correlated with IgE level in a group of children from Madison and was also associated with asthma in their parents. In the present investigation, however, rs2317676 showed no association with asthma. The coding variant rs5918 has shown a weak association with wheezing and asthma in the Madison population, but no relationship was observed in this investigation. The rs3809865 allele was strongly associated with asthma in Chinese Han children. This SNP tags an asthma-associated SNP (rs3760372, 0.001\<P\<0.01) with an r² value of 1.0, which seems likely to account for this result. The SNPs located in the 3′UTRs of genes have attracted attention due to their roles in regulating gene expression. Our TargetScan and PITA analyses indicated that rs3809865 might influence miRNA binding to ITGB3. The sequence containing the T allele of rs3809865 bound more stably to hsa-mir-124 and hsa- mir-506 than the sequence containing the A allele, which would likely result in the down-regulation of integrin β3. It has been reported that decreases in integrin β1 and β4 levels could prevent the development of asthma, but it is unclear whether the down-regulation of integrin β3 could have a similar effect. Therefore, we further investigated the regulation of integrin β3 by hsa-mir-124. Luciferase reporter assays suggested that the allele of rs3809865 could affect the targeting of hsa-mir-124. Subsequently, it was confirmed that the expression of integrin β3 in A549 cells was suppressed by transfection with hsa-mir-124, which indicated that hsa-mir-124 could regulate the expression of integrin β3. These results clarify the mechanism underlying the association between rs3809865 and the decreased risk of asthma. In this study, we evaluated three SNPs that have been associated with asthma in Western populations but failed to replicate these associations in Chinese Han children. The different findings in distinct populations might be due to differences in environmental conditions. Environmental exposures have been shown to have important roles in triggering asthma. Various environmental factors, such as air pollution, microbial exposure, diet and pet ownership, could affect asthma development. Our subjects lived mainly in Jinan city and the surrounding area and thus were subjected to different environmental risk factors than their Western counterparts, likely explaining the difference in results. Other differences between studies, such as the criteria used to define asthma, the age and race of the subjects and differences in the patterns of linkage disequilibrium are also factors that might explain non-replication in genetic association studies. Overall, the findings presented in this study provide new evidence for the association between SNPs in ITGB3 and asthma. Our results, combined with those of previous reports, suggested that ITGB3 should be considered a true asthma- related gene. # Supporting Information We thank all the subjects for their participation. We also thank the clinicians and other staff from the QiLu Children’s Hospital for their assistance in blood sample and data collection. We are grateful to Dr Lin (Provincial Hospital Affiliated to Shandong University) for performing the HRM analyses. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YZ YH MD. Performed the experiments: YZ YH. Analyzed the data: YZ YH LD HY. Contributed reagents/materials/analysis tools: HY LC XZ. Wrote the paper: YZ YH MD.

# 1. Introduction Kaposi’s sarcoma (KS) is a vascular malignancy highly prevalent in sub-Saharan Africa (SSA). Human herpesvirus type 8 (HHV-8), which is also known as Kaposi’s sarcoma-associated herpesvirus (KSHV), has been implicated as the etiological agent of all four types of KS. These are: i) Classic KS—occurs in elderly men of Mediterranean origin; ii) Iatrogenic KS—a result of immunosuppressive therapy; iii) Endemic KS—seen among the HIV-negative population in SSA; and iv) Epidemic or AIDS-associated KS (EpKS)–associated with HIV-induced immunosuppression. EpKS is the most common type of KS in SSA countries, and is an HIV stage 4 disease according to the World Health Organization. Despite the introduction of ART, the incidence, prevalence, and mortality attributed to EpKS remain high in SSA despite more substantial reductions in higher income settings like North America and Europe. It is clear from the epidemiological association of KS with HIV-induced immunosuppression, immunosuppressive therapy, and aging that immune dysregulation plays a key role in the pathogenesis of KS. However, the fundamental mechanisms underlying KS development are largely unknown. Prior to the introduction of ART, EpKS was considered an AIDS-defining malignancy because it commonly developed when HIV viral loads were high and CD4 counts were very low. However, EpKS incidence levels have remained steady despite effective roll- out and uptake of ART throughout SSA. The majority of EpKS cases are now presenting after HIV viral suppression and at least partial immune reconstitution. In addition, the recurrence rates after KS treatment are high. Management of EpKS can be challenging due to difficulties in disease staging, adverse effects of available treatments, and poor outcomes. Early-diagnosed skin-limited EpKS is usually treated with ART alone, whereas advanced EpKS requires ART plus cytotoxic chemotherapy. Treatment of EpKS with ART results in variable outcomes including remission, stable disease, or progression requiring addition of cytotoxic cancer chemotherapy. Furthermore, EpKS patients initiating ART have increased risk of mortality when compared with non-KS HIV-infected individuals initiating ART. Cytotoxic cancer chemotherapeutics can lead to bone marrow suppression resulting in anemia and leukocytopenia, and may also cause other adverse effects such as peripheral neuropathy, lung fibrosis, and cardiotoxicity in individuals who are already immunocompromised. Factors associated with disease progression or remission as a result of ART are unknown. It is important to identify these factors so that chemotherapy can be avoided when possible, and individuals with early EpKS who will require both ART and chemotherapy can be identified early and treated appropriately. Cytokines have been observed to be dysregulated in KS patients. Anti-KSHV neutralizing antibodies (nAb) are known to be more readily detected in KS patients than in asymptomatic KSHV-infected controls, but it is not known how or whether nAb levels change in association with ART or any other KS treatment. Here, we quantified plasma cytokines and nAb in chemotherapy-naïve, early- diagnosed ART-treated KS patients to investigate their potential as prognostic biomarkers of disease progression or remission. # 2. Materials and methods ## 2.1 Patients and samples This study was conducted on adult KS patients presenting at the University Teaching Hospital (UTH) in Lusaka, Zambia. We recruited histologically- confirmed, early-stage (AIDS Clinical Trials Group Stage T0/S0) HIV-associated KS patients who were antiretroviral therapy (ART)-naïve or on ART for less than 2 weeks. Recruitment was done after obtaining informed consent. All recruited patients were 18 years and older, had a limited number of early KS lesions (patches, plaques, or nodules), lacked lymphedema, and had no evident visceral involvement. At baseline, sociodemographic information was collected, a physical assessment was conducted, and blood samples were collected. ART was initiated in those who were ART-naïve. All recruited patients were followed up monthly for at least 6 months for either remission (responders) or progression of KS (poor responders). Responders all had a follow up period of at least 6 months while poor responders had variable follow up ranging from 1 to 6 months depending on when they experienced disease progression. None of our study participants had co-infections such as malaria or hepatitis, or HIV-associated comorbidities such as tuberculosis or cryptococcal meningitis at baseline or during follow up. Venous whole blood was collected at baseline and at the time of progression or after 6 months for those who responded to ART. All blood samples were collected in EDTA vacutainers. Plasma was separated and then stored at -80°C until analysis. The study was approved by the University of Zambia Biomedical Research Ethics Committee, the Zambian National Health Research Authority, and the Institutional Review Board at the University of Nebraska-Lincoln. ## 2.2 HIV testing and viral load HIV-1 status for all patients was determined as recommended by the Zambian Ministry of Health. The Alere Determine HIV-1/2 kit (Alere Medical Co. Ltd) was used for screening, while the SD Bioline HIV-1/2 kit (Standard Diagnostics Inc) was used for confirmation. HIV-1 plasma viral load was measured on the Hologic Panther (Hologic) using the Aptima HIV-1 Quant Dx Assay kit (Hologic) according to the manufacturer’s protocol. ## 2.3 CD4 T cell quantification CD4 counts were determined with BD FACSCalibur (BD Biosciences) using the BD TriTest kit (BD Biosciences), according to the manufacturer’s protocol. ## 2.4 Measurement of plasma cytokines and chemokines Plasma cytokines and chemokines were quantified using the Beckton-Dickinson Cytometric Bead Array (CBA) Flex Set kits according to the manufacturer’s protocol. The following cytokines and chemokines were quantified: Transforming growth factor-β (TGF-β), Interleukin 5 (IL-5), Interferon-inducible protein 10 (IP-10), Vascular endothelial growth factor (VEGF), Interleukin 6 (IL-6), Tumor Necrosis Factor (TNF), and Interleukin 10 (IL-10). Data was quantified on a BD Accuri C6 Plus cytometer at 448nM and 640nM excitation wave lengths (BD Biosciences, San Jose, CA) and analyzed with FlowJo version 10 software (TreeStar, Ashland, OR). ## 2.5 KSHV neutralizing antibodies The challenge virus, KSHV (rKSHV.219) encoding the green fluorescent protein (GFP) gene under control of the cellular EF1α promoter, was generated by stimulating latently infected Vero.219 cells, as previously described. Flow cytometry was used to titrate the rKSHV.219 stock to determine the amount required to achieve 50% infectivity on the human embryonic kidney cell line (293T cells). A 1:50 dilution of heat-inactivated plasma was incubated with rKSHV.219 virus, then the virus-plasma mixture was used to infect 293T cells. All assays were carried out in triplicate and infection was quantified by flow cytometry for the number of cells expressing GFP. A sample was considered to be neutralizing antibody positive if it inhibited at least 50% of the infectivity. ## 2.6 Statistical analysis Baseline characteristics were analyzed using descriptive statistics. The Kruskal-Wallis test was used to compare cytokine/chemokine differences between groups. Comparison of paired data within groups was done using the Wilcoxon matched-pairs signed-rank test. Testing for correlation of continuous variables was done using the Spearman rank correlation. All statistical tests were two- sided, p values \<0.05 were considered significant. Stata version 15 (StataCorp LLC, USA) was used for all statistical analyses. # 3. Results ## 3.1 Baseline characteristics of study participants We successfully recruited 27 eligible study participants. 5 of the recruited participants were lost to follow up after the baseline assessment, 1 participant was commenced on cytotoxic chemotherapy at another health facility and hence was no longer eligible for follow up, 13 participants had worsening KS within 6 months of ART initiation, and 4 participants had stable disease or regression of KS lesions after at least 6 months of follow up. Among the participants that worsened on ART, only those that were on ART for at least 2 months before worsening of KS were analyzed. We therefore analyzed 11 EpKS patients who had early-stage KS disease and recently initiated ART, 7 of these had EpKS disease progression within 6 months of initiation of ART and follow up while 4 had stable disease or KS tumor regression for at least 6 months. highlights the baseline characteristics of the analyzed study participants by treatment outcome. ## 3.2 HIV disease parameters Among the poor responders, the median HIV plasma viral load at baseline was 45,610 copies/ml \[IQR = 6,477–691,296\] while at determination of ART response (DAR) the median HIV viral load had dropped to 82 copies/ml \[IQR = 0–2664\]. Responders had a baseline median HIV viral load of 577,016 copies/ml \[IQR = 401,840–1,092,995\] which dropped to 353 copies/ml \[IQR = 15–1679\] at DAR. There was no statistically significant difference in HIV viral load at baseline or at DAR between the responders and poor responders. There was a statistically significant decrease in HIV viral load from baseline levels at DAR among the poor responders (p = 0.018), whereas the drop in HIV viral load among the responders was not statistically significant (p = 0.11). The median CD4 count for the poor responders was 166 cells/μl \[IQR = 127–295\] at baseline, and increased to 195 cells/μl \[IQR = 131–339\] at DAR. The median CD4 count for responders was 135 cells/μl \[IQR = 68–211\] at baseline and was 135 cells/μl \[IQR = 101–206\] at DAR. There was no statistically significant difference in CD4 counts at baseline and at DAR between responders and poor responders. In addition, CD4 count comparisons between baseline and DAR in both responders and poor responders were not statistically significant. ## 3.3 Comparison of plasma cytokines between responders and poor responders We quantified the plasma levels of TGF-β, IL-5, IP-10, VEGF, IL-6, TNF, and IL-10 in responders and poor-responders at baseline and at DAR. These cytokines and chemokines have previously been associated with regulation of cell growth and survival, humoral immune response, chemotaxis, tumor proliferation, and enhancing and/or suppressing cancer-associated inflammation. There was no statistically significant difference between the two groups at baseline or at DAR in the plasma levels of VEGF, TGF-β, TNF, and IL-10. However, median IL-5 levels were significantly higher in responders than poor responders at baseline (0.76pg/ml vs. 0.37pg/ml; p\<0.01), and remained higher at DAR (0.56pg/ml vs 0.37pg/ml; p\<0.01). There was no significant difference in median IL-6 levels between responders and poor responders at baseline (3688fg/ml vs. 2390fg/ml; p = 0.57); however, at the time of DAR, median IL-6 levels were 7-fold lower in the responders than in the poor responders (600fg/ml vs. 4272fg/ml; p\<0.05). The median plasma level of the chemokine CXCL10 (IP-10) was not significantly differential at baseline between responders and poor responders (724pg/ml vs. 726pg/ml; p = 0.85). However, at DAR, the level of IP-10 was significantly lower in responders than poor responders (187pg/ml vs. 528pg/ml; p\<0.01). ## 3.4 Comparison of KSHV neutralizing antibodies between responders and poor responders KS patients have much higher neutralizing antibody (nAb) responses than KSHV- infected, but asymptomatic individuals with HIV-1 co-infection. However, whether such responses increase, decrease, or remain unchanged, over the course of ART treatment of early KS is unknown. Moreover, whether the magnitude of the baseline nAb response correlates with eventual ART treatment outcome has not been explored. At baseline, both responders and poor responders were able to neutralize \>50% of the challenge virus. Neutralization trended higher in responders than in poor responders; however, this difference was not statistically significant (85% vs. 67%; p = 0.20). At DAR, the magnitude of the KSHV nAb response in responders was also indistinguishable from that in poor responders (66% vs. 70%; p = 1.0). There was a non-significant decline in KSHV nAb responses from baseline to DAR among responders, whereas the nAb responses trended upward over treatment among the poor responders. Neither differential was statistically significant (p = 0.29 and p = 0.75 respectively). No correlation was detected between the levels of antibody-associated cytokine IL-5, and nAb levels at baseline in responders (⍴ = -0.5; p = 0.67) or poor responders (⍴ = 0.46; p = 0.35). At DAR, there was no correlation between IL-5 levels and nAb levels among the poor responders (⍴ = -0.54; p = 0.27), whereas the correlation could not be determined among the responders. highlights a compilation of HIV viral loads, CD4 counts, cytokines/chemokines, and KSHV nAb results. # 4. Discussion HIV-induced immune suppression and dysregulation are major predisposing factors for the development of EpKS. However, it is not entirely clear what HIV does and to what extent. Therefore, treatment with ART is essential in the management and long-term control of KS. ART-induced immune reconstitution reverses KS progression. However, in a significant proportion of individuals, ART seems to exacerbate the disease. Changes in immunological and/or HIV and KSHV virological factors likely increase or decrease the likelihood of ART-treated KS regression or progression. In this study, we quantified several immunological and virological factors and their potential association with response to ART treatment of early-diagnosed EpKS. We observed a decrease in HIV viral loads in both responders and poor responders to viral suppression levels of less than 1000copies/ml. Previous studies have reported that ART has no direct anti-KSHV or anti-KS activity. In addition, there is abundant evidence of KS development in HIV patients on ART who have high CD4 counts and low viral loads. Consistent with the concept, in our study, HIV viral load was not differential between the responders and poor responders at baseline or follow up. In addition, CD4 counts were also not significantly different at baseline and follow up between responders and poor-responders. This suggests that ART reduction of HIV replication is decoupled from KS control even though uncontrolled HIV-1 replication clearly leads to enhanced risk for KS development. Consistent with previous reports of KSHV nAb primarily in KS symptomatic subjects, both responders and poor responders demonstrated high KSHV nAb. However, due to the small sample size, the observation that nAb was higher among responders than poor responders was not statistically significant. In concert with previous reports, the lack of correlation of baseline nAb level, or changes in nAb, with treatment, suggests that KSHV nAb is not a correlate of protection. The data further suggest that nAb is unlikely to be a marker for KS disease progression or control in the context of ART. We did, however, observe differences in IL-5, IL-6 and IP-10 between responders and poor responders. IL-5 is a Th2 cytokine that induces eosinophilia and also promotes immunoglobulin secretion by B cells. IL-5 levels have been reported to increase following KSHV infection and in KSHV-associated conditions. We observed higher IL-5 levels at baseline and follow up among the responders compared to the poor responders. It was therefore anticipated that anti-KSHV nAb would be significantly higher among responders than poor responders. Yet, IL-5 levels did not correlate with levels of anti-KSHV nAb, suggesting that IL-5 expression may not be a good marker of humoral control of KSHV infection. The high IL-5 levels, may reflect the previously described effect of parasitic infestations on initiation and progression of KSHV pathogenesis. Nevertheless, our study findings suggest that high IL-5 levels at baseline may be a good prognostic marker for ART-treated EpKS perhaps indicative of a more favorable outcome of Th2 versus Th1 skewing of responses. We also observed higher plasma IL-6 levels in individuals who were undergoing KS disease progression compared to those who underwent response. IL-6 is a pleiotropic cytokine known to stimulate the proliferation of KSHV-infected cells, and also promotes humoral responses by driving B cell maturation. Interestingly, KSHV also produces a viral homolog of IL-6 (vIL-6) which upregulates expression of carcinoembryonic antigen-related cell adhesion molecule 1 (CEACAM1), a protein that has been implicated in angiogenesis, endothelial cell migration, and vascular remodeling. Both IL-6 and vIL-6 have been associated with the proliferation of KSHV-infected tumors. Furthermore, detectable plasma KSHV viral load has been found to be associated with elevated plasma IL-6 levels in KS patients. Our findings are consistent with previous reports on the role of IL-6 in promoting the proliferation of KSHV tumors. In turn, it is possible that elevated IL-6 also contributes to the detection of nAb in KS disease, even though such Ab responses are non-protective. IP-10 is a chemoattractant for recruiting leukocytes to involved tissues, and thereby intensifies inflammation that results in tissue damage. It is also closely associated with HIV infection, systemic inflammation, and cellular activation. In cancer, IP-10 has been shown to both inhibit and promote tumor formation and/or metastasis. The effect of IP-10 on a tumor largely depends on the type of CXCR3 receptor expressed by that tumor. IP-10 has been found to inhibit CNS tumors and melanoma, whereas it has been observed to promote breast cancer, some lymphomas, colon cancer, and basal cell carcinoma. The effect of IP-10 on KS tumors is largely unknown. We observed a significant decrease in plasma IP-10 levels among the responders whereas the poor responders experienced no change from baseline, suggesting a potential tumor promoting effect of IP-10 in KS. This is consistent with previous reports on upregulation of IP-10 expression in KS tumors. # 5. Study limitations The major limitation of this study was the low number of complete responders. A larger sample size with increased numbers of both responders and poor responders, and a longer follow up period would be required in future. # 6. Conclusion High plasma IL-5 is a potential marker of good prognosis in ART-treated EpKS, whereas high plasma IL-6 and IP-10 levels are prognostic markers for potentially poor ART-treatment outcomes. [^1]: The authors have declared that no competing interests exist.

# Introduction Urinary tract infections (UTI) are one of the most common bacterial infections in primary care and classified as either complicated or uncomplicated. Uncomplicated UTI occur in healthy individuals with either no structural or functional abnormalities of the urinary tract. Bacterial pathogens are thought to be the cause of the presenting symptoms, and are often treated with antibiotics, which accounts for up to 95% of the antibiotic prescription for UTI in primary care settings. *Escherichia coli (E*. *coli)* is the most common pathogen isolated in around 75% of uncomplicated UTI. However, cases may be complicated by the *E*. *coli* producing extended-spectrum β-lactamase (ESBL), resulting in resistance to multiple β-lactam antibiotics. Three international surveillance studies estimated that the prevalence of *E*. *coli* resistant to several types of antibiotics were around 8% (nitrofurantoin) to 48% (ampicillin) in North American, South American and European populations. Meanwhile, primary care physicians are called to be antimicrobial stewards. Antibiotic resistance has been shown to be associated with patients’ age and sex, previous antimicrobial therapy and can be dynamic, responding to patterns of antibiotic treatment. It is important to understand antibiotic prescribing behaviour of primary care physicians and antibiotic resistance patterns in primary care. Many studies has been conducted in the UK, however, there has been few studies to explore physician prescribing behaviour and antibiotic resistance in a Chinese primary care population. Antibiotic resistance of uropathogens and the prevalence of ESBL production of community-acquired UTI in Hong Kong is high (6.6% and 10% in 2004 and 2005 respectively). This study aimed to explore (1) the antibiotic resistance of urinary isolates of symptomatic patients presenting to primary care, (2) empirical antibiotic prescribing behaviours of primary care physicians and (3) factors associated with antibiotic resistance and physician antibiotic prescription. # Materials and methods ## Design and setting This was a prospective cohort study of public and private primary care clinics in Hong Kong. Public primary care clinics (government funded group primary care clinics) and private clinics provide more than 80% of primary care services for general population in Hong Kong. Invitations were sent to all public and private group teaching practices in the New Territories and Kowloon region from January 2012 to December 2013. Group practices were chosen to increase yield of urinary samples. New Territories and Kowloon regions were selected to facilitate transportation of specimens. ## Characteristics of patients Eligible patients were women above the age of 16 presenting to primary care clinics with symptoms suggestive of UTI. The attending physician assessed eligibility criteria and invited patients with a provisional diagnosis of uncomplicated UTI based on UTI symptoms to participate in the study. Patients who were already on antibiotics or had antibiotics in the preceding 21 days, pregnant, having functional or structural abnormalities of the urinary tract, immunocompromised illness (e.g. AIDS, leukaemia, active viral hepatitis, multiple myeloma) or using immunosuppressants (e.g. oral corticosteroids, medication for modification of autoimmune conditions or transplanted organ and chemotherapy) were excluded from the study. Physicians were required to fill in a case report which included patient’s demographics, symptoms of UTI, hospitalisation history, antibiotic prescription for the present UTI, records of previous bacterial pathogens in the past one year and antibiotics prescribed in the past one year. Chronic medical conditions (e.g. diabetes mellitus, hypertension, ischaemic heart disease, chronic obstructive pulmonary disease and asthma, thyroid disease) were also recorded. ## Ethical statement This study was approved by the Clinical Research Ethics Committee of the Chinese University of Hong Kong and New Territories East Cluster of the Hospital Authority \[CRE-2011.115\]. Written formal consent was obtained from all participants. ## Laboratory analysis Participants were given verbal instructions on the steps of collecting a clean catch mid-stream urine (MSU) sample. The sample was collected in a sterile container refrigerated and delivered within the same day to the centralized accredited University Pathology laboratory for processing. Only one sample per patient was obtained. In order to identify the major pathogen of UTI, the collected MSU specimens were processed for urine microscopy and cultures. Urine microscopy was performed by standard methods according to laboratory operating procedures (PHLS BSOP41, 1998; Urine for Microscopy, bacterial counts, identification and antibiotic susceptibilities, RCPA QAP Ltd). The urine sample was mixed well and a known quantity was examined under the inverted microscope at 20X power field in a flat-bottomed microtitre tray. The number of red and white cells were enumerated and reported as negative, moderate (10,000–100,000 cells/ml), large number (\>100,000 cells/ml) of cells. Specimens were inoculated by the paper strip method (MAST) onto chromogenic medium (CPSE plate, Biomerieux) and incubated aerobically for 18–24 h at 35°C. The number of colonies were enumerated and reported with 0, \<10³, 10³⁻⁴, 10⁴⁻⁵, and \>10⁵ cfu/ml as no growth, insignificant, scanty, moderate and heavy growth respectively. Only urine cultures with heavy growth of a single bacterial type (\>10⁵ cfu/ml) were included in the analysis. To identify antibiotic resistance patterns in the urinary isolates, the samples were further tested for antibiotic susceptibility to amoxicillin, ampicillin, ciprofloxacin, co-trimoxazole, gentamicin and nitrofurantoin and to detect ESBL positivity. Antibiotic susceptibilities were performed and interpreted as according to Clinical and Laboratory Standards Institute (CLSI) (Clinical and Laboratory Standards Institute. Performance Standards for Antimicrobial Susceptibility Testing. CLSI document. Wayne). Specimens which were not processed immediately were stored at -4°C. ## Data analysis Descriptive statistics (mean±SD, percentages) were used to describe the frequency of bacterial pathogens, antibiotic resistance and prescription pattern. Chi-squared test and univariate analysis were used for statistical analysis. *E*. *coli* antibiotic resistance was stratified with age group and chronic medical conditions and analysed by chi-square test. Odds ratios (OR) with 95% confidence intervals (95% CI) were calculated to identify factors associated with *E*. *coli* isolates resistance to antibiotics and for antibiotic prescription. A two tailed *P*\< 0.05 was considered significant and all statistical analyses were performed by using SPSS v 21.0 Window (IBM Inc.,SPSS Inc, Chicago, IL USA). # Results ## Patients and urine samples characteristics Of the invited group clinics, 5 public and 12 private group clinics participated in the study and 306 female patient presentations were recruited into the study. Of these, eight clinical presentations were from previously recruited patients and only their first presentation were included in the study. Thus a total of 298 patients participated in the study of 145 (48.7%) and 153 (51.3%) were recruited from public and private clinics respectively. The mean age of patients was 53.8±17.1 years old. Of the samples received, two were discarded as there was leakage on transit and could not be processed. A positive urine culture was found in 141 (47.3%) of the collected samples whilst 73 (24.5%) showed no growth, 68 (22.8%) as insignificant growth and 14 (4.7%) showed mixed bacteria and probable contamination. ## Bacterial pathogens among positive urine isolates Among the 141 positive urine isolates, the most common uncomplicated UTI pathogen was *E*. *coli* (75.9%). Other uropathogens included *Klebsiella* (9.2%), *Enterococcus* (3.5%), *Proteus Mirabilis* (3.5%) and *Citrobacter* (2.8%). *E*. *coli* infection occurred mostly in women who aged 51 above (63.5%) and most likely to present with two or more symptoms (79.5%) of which dysuria, frequency and urgency were common symptoms experienced. ## Antibiotic resistance rates of *E*. *coli* isolates Of the 107 *E*. *coli* isolates, the resistance rates were 59.8% for ampicillin, 31.8% for co-trimoxazole, 25.2% for gentamicin and 23.4% for ciprofloxacin. The majority of *E*. *coli* isolates were sensitive to nitrofurantoin (98.1%), amoxicillin (78.5%) and ciprofloxacin (76.6%). *E*. *coli* isolates with resistance to one drug can also be resistance to other drugs. Multidrug resistance (co-resistance to ampicillin, ciprofloxacin, and co- trimoxazole) occurred in 12.1% of *E*. *coli* isolates (n = 13). ## ESBL producers ESBL producers (n = 14) were found in 9.9% of the positive isolates (n = 141) and almost all ESBL producers (n = 13) were from *E*. *coli* isolates. 85.7% (n = 12) of ESBL producing isolates were found amongst older women who were aged 51 years and above. All ESBL producing isolates were resistant to ampicillin, whilst almost half were resistant to ciprofloxacin (n = 8, 53.3%) and co- trimoxazole (n = 8, 53.3%). ## Antibiotic resistance rate of *E*. *coli* isolates and age and chronic medical conditions Antibiotic resistance rates of ampicillin and multidrug resistance were statistically significant with increasing age. The OR of patients who aged \>50 were 2.90 (95% CI: 1.28–6.55, *P* = 0.011) for having resistance with ampicillin. In addition, The OR of patients who aged \>50 were 8.14 (95% CI: 1.02–65.26, *P* = 0.048) for having multidrug resistance. Meanwhile, although antibiotic resistance appear higher in those with more than two chronic medical conditions for most antibiotic types, this was not statistically significant (*P* = 0.134, *P* = 0.150*P* = 0.340 and *P* = 0.253 for ampicillin, ciprofloxacin, co-trimoxazole and multidrug resistance respectively). ## Factors associated with antibiotic resistance of *E*. *coli* isolates Patients who had taken antibiotics in the past year had OR of 2.89 (95% CI: 1.09 to 7.68; *P* = 0.029) for resistance to ciprofloxacin than those without antibiotics treatment. In addition, the OR of those exposed to antibiotics in the past year were 3.29 (95% CI: 0.99 to 10.95; *P* = 0.043) to have multidrug resistance than those without previous antibiotics treatment. The OR for those with previous UTI diagnosis in the past 6 months were 3.04 (95%CI: 1.06 to 8.71, *P* = 0.03) to have ciprofloxacin resistance. ## Antibiotic prescription Of the 298 patients presenting with symptoms, 245 (82.2%) were prescribed antibiotics. Amoxicillin (39.6%) was most commonly prescribed, followed by nitrofurantoin (28.6%) and ciprofloxacin (10.2%). The mean duration of antibiotic prescription was 6.86±0.54 days. ## Antibiotic prescription rates among public and private clinics The antibiotic prescription rates were 50.6% and 49.4% in public and private primary care clinics respectively. There was no significant difference in empirical antibiotic prescription between the two groups (*P* = 0.147). Meanwhile, the OR of public primary care physicians to private primary care physicians in prescribing amoxicillin were 2.84 (95% CI: 1.67 to 4.85) and 2.01 (95% CI: 1.14 to 3.55) for nitrofurantoin. The results were statistically significant (*P*\<0.001 and 0.015). In contrast, private primary care physicians prescribed more cefuroxime (5.3 versus 0.8%, *P* = 0.003) and ciprofloxacin (7.8 versus 2.4%, *P* = 0.005). ## Factors for antibiotic prescription There was no association between the rate of antibiotic prescription with age (*P* = 0.370). However, patients aged 65 years or above received more amoxicillin as empirical antibiotic treatment (*P* = 0.047), whilst patients aged between 17 and 50 years old received more cefuroxime (*P*\<0.001) and ofloxacin (*P* = 0.005). The OR of patients with co-existing medical conditions was 3.08 (95% CI: 1.60 to 5.94) to be prescribed amoxicillin than those without chronic medical conditions and this was significantly significant (*P* = 0.001). The OR of prescribing ciprofloxacin was 2.17 (95% CI: 0.80 to 5.89, *P* = 0.122) and co-trimoxazole was 2.58 (95% CI: 0.48 to 13.84, *P* = 0.254) for patients with previous history of a UTI in the past 6 months than patients without a recent UTI. However the results were not statistically significant. The odds of receiving an antibiotic prescription were greater with symptoms of dysuria or urgency. The OR of dysuria was 4.53 (95% CI: 2.33 to 8.83, *P*\<0.001) and urgency (OR: 2.32, 95% CI: 1.18 to 4.58 *P* = 0.013). Meanwhile, The OR of patients presenting with more than two urinary symptoms were 0.33 (95% CI: 0.16 to 0.69, P = 0.003) time than those with one or two presenting symptoms to have empirical antibiotic treatment. ## Empirical antibiotic prescription and uropathogen susceptibility For the 141 positive isolates, 122 received empirical antibiotic treatment. The appropriate use of antibiotics was evident in public primary care. This was demonstrated by the matching of antibiotic susceptibility of the uropathogen and the prescribed antibiotics were statistically significant for both sensitivity and resistance of *E*. *coli* pathogens and for other uropathogens. The OR in matching antibiotic susceptibility and physician prescribing was 6.72 (95% CI: 2.07 to 21.80, *P* = 0.001) for *E*. *coli* isolates and 6.19 (95% CI: 1.04 to 36.78, *P* = 0.034) for other uropathogens between public primary care and private primary care physicians. # Discussion ## Antibiotic resistance Our result showed that *E*. *coli* was the most common pathogen and present in 75.9% of the positive urine cultures, which was consistent with international studies. Regarding antibiotic sensitivity and resistance rate, our results were consistent with a regional study of urinary isolates of outpatients from private hospital laboratories and community laboratories, the sensitivity rates of ampicillin (38.3% versus 37.4–37.6%), ciprofloxacin (76.6% versus 76.3–77.2%), co-trimoxamole (68.2% versus 64.7–65%) were similar. In addition, a lower antibiotic sensitivity to amoxicillin (Sensitivity rates of 81.3%-83.6% vs 78.5%) was observed, although resistance rates were low and comparable. Reduced susceptibility to amoxicillin in our study may indicate patients presenting to primary care which may have less severe symptoms and likely to present earlier or may reflect changes in antibiotic susceptibilities due to physicians’ prescribing behaviour. The Centre of Health Protection in Hong Kong publishes sentinel surveillance information of outpatient urinary samples every month. Meanwhile the sentinel data analyses around 500 routine samples per month, all of which are urine samples collected in 85 public primary care centres (GOPCs); this includes male and female who are likely to those who have complicated or recurrent urinary illness to warrant further investigation from urinary microscopy and culture. Over comparable periods in 2012–13, bacterial pathogens and antibiotic resistance to most antibiotics were similar to our results. However, the antibiotic resistance rates for co-trimoxazole as well as ESBL producers were lower in our study (31.8 versus 38% and 9.9% versus 18.0% respectively). Meanwhile, the website publishes resistance rates only, whilst sensitive and intermediate responses may also be helpful in primary care physicians choice of antibiotics given other factors e.g. co-morbidities and contraindication. Previous studies have suggested an association between antibiotic resistance and increasing age and whilst this trend was observed, only a significant finding of age and resistance to ampicillin and multiple drugs were evident in our study. Our study showed that public primary care clinics and past history of UTI diagnosis (within 6 months) were statistically significant associated with *E*. *coli* isolates resistance to ciprofloxacin. A study in Europe suggests that the prior use of antibiotics may be a contributory factor for antibiotic resistance; Sotto et al found that a past diagnosis of UTI within the past 1 year was associated with *E*. *coli* antibiotic resistance to ciprofloxacin. ## Antibiotic prescription The results of our study suggest that nitrofurantoin is the most appropriate first line antibiotic of choice which had the lowest rate of resistance and the highest sensitivity. Second line treatment would be amoxicillin. Our result is similar with current guidelines on antibiotic prescription for UTI, such as Infectious Diseases Society of America guidelines and inter-hospital guidelines in Hong Kong. High antibiotic resistance rates to ampicillin, ciprofloxacin and co-trimoxazole (resistance rate \> 20%) have been consistent to a local study and demonstrate that these drugs should not be considered in the management of UTI unless there is proven resistance or allergy to nitrofurantoin or amoxicillin. In our study, the most commonly prescribed antibiotic amongst primary care physicians is amoxicillin followed by nitrofurantoin. Selection of appropriate antibiotics should be based on differing resistance patterns in the locality, as well as patients’ medication, contraindications, allergy profile and cost. All these considerations may affect the antibiotic prescription pattern, which may account for the differences in the antibiotic prescription rate. The use of nitrofurantoin may be limited as it is contraindicated in patients with renal impairment and creatinine clearance \<60mL/min and may require careful consideration in the elderly and this may explain the reluctance in prescribing. The threshold of creatinine clearance \<60mL/min however, has been disputed due to non-existent evidence and suggests a lower threshold of creatinine clearance \<40mL/min could be adapted to improve nitrofurantoin use. Meanwhile, even with cautionary use, it may be likely that nitrofurantoin is underutilised and further exploration in clinician’s reluctance and other factors involved in the antibiotic decision making process. Meanwhile, our result shows that private primary care physicians prescribed more ciprofloxacin (*P* = 0.005), while public primary care physicians prescribed more amoxicillin (*P*\<0.001) and nitrofurantoin (*P* = 0.015). This may be attributed to guidelines, drug formulary and the cost of treatment. The public primary care clinics are governed by the Hospital Authority which has guidance for public primary care physicians on their clinical management system. Private group practices have their own clinical management systems and are not restricted by guidelines and have a wider drug formulary. Meanwhile, private group practices are able to absorb higher drug costs. In Hong Kong, a 3-day course of nitrofurantoin and ciprofloxacin costs 2.38 and 25.8 HKD respectively (1 USD = 7.8 HKD). The predominant use of nitrofurantoin and amoxicillin in public clinics may also be attributed to the use of available inter-hospital guidance and clinical governance of its governing body the Hospital Authority and may reflect the benefits of governance on antibiotic prescribing behaviour. Surveillance of prescribing data of private doctors may enhance appropriate antibiotic prescribing, but this would logistically be difficult due to the heterogeneity of clinical systems and documentation by private practitioners and lack of a prescription tracking system as many private practitioners are dispensing practices. International recommendations for treating uncomplicated UTI in primary care is a 3 day course of antibiotics, however 99.5% of the antibiotic prescriptions in this study were of 5–7 days duration. Our results show that patients who attended private clinics were younger (mean 49.50±17.66 years old) and had less comorbidity when compared to patients who attended public clinics. Private primary care physicians also prescribed cefuroxime and ofloxacin which has been documented for use in young women. Further exploration of physician’s prescribing behaviour and facilitators and barriers to appropriate prescribing is warranted. ## Strength and limitations This is the first study to investigate both antibiotic resistance and prescription rates of patients with uncomplicated UTI in Chinese primary care settings. Previous antibiotic resistance data was conducted in outpatients departments and public primary care clinics. Meanwhile, there have been no studies documenting physician prescribing habits in the region. The strength of this study has been to show the use of both antibiotic susceptibility data and antibiotic prescribing behaviour in identifying highlighting gaps associated with antibiotic appropriateness. This study has several limitations. The participation of group practices in selected districts may limit the representativeness of patients across the Hong Kong S.A.R region. Due to the uncommon and sporadic presentation for UTI across a number of sites, the recruitment period was long and we were unable to document the number of patients who were approached but declined to participate in the study which may attribute to selection bias of the study sample. In addition, primary care physicians may also be more likely to prescribe an appropriate antibiotic as they were aware of being observed. Recall bias may also occur in patients recounting prior UTI episodes and prior antibiotic use. However these limitations are likely to be underestimating our study findings. The study was designed to reflect the pragmatic approach of UTI diagnosis and empirical antibiotic prescription in primary care. The use of urine dipstick in the diagnostic criteria, the follow up of presenting symptoms and analysis of urine culture yields of 10³ and 10⁴ cfu/ml may further enhance the study, particularly in detailing threshold for treatment and symptom resolution. Our study has highlighted a gap between physician current prescribing habits and antibiotic resistance. There is a need for increased efforts for education and dissemination to improve appropriate antibiotic prescription. It may also be advantageous for the surveillance data to show intermediate susceptibility to particular antibiotics as well as resistance data to help guide primary care physicians on antibiotic choice. The prospective cohort and multicentre approach in this study yields a more comprehensive picture of the complexities in the patient presentation and doctors’ prescription of antibiotics not addressed in previous regional studies. Additional information on patient’s clinical response and recovery and physician’s decision on antibiotic prescription and choice can be considered in further studies. # Conclusions This study identified antibiotic resistance amongst uropathogens and physician antibiotic prescription in primary care settings. There is scope to encourage nitrofurantoin as first line treatment in primary care settings particularly in private primary care and to explore its use and barriers for use. Enhancing antibiotic guidance for UTI and continued surveillance of antibiotic resistance and physician antibiotic prescription may improve appropriate antibiotic prescription for UTI in Chinese primary care. # Supporting information We would like to thank the medical staff in all the participating primary care clinics and the laboratory technicians in helping out with the study. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** CKMW KK. **Formal analysis:** AF PAD NL SYSW CKMW. **Funding acquisition:** CKMW KK SYSW. **Investigation:** MI AF. **Writing – original draft:** CKMW. **Writing – review & editing:** PAD MI SYSW CKMW.

# Introduction Globally, an estimated 295,000 women died during and following pregnancy and childbirth in 2017. Sub-Saharan Africa (SSA) alone accounted for two-thirds of maternal deaths, a vast majority of which could have been prevented. The Sustainable Development Goal 3 is set to reduce the global maternal mortality ratio (MMR) from 216 deaths per 100,000 live births in 2015 to less than 70 deaths per 100,000 live births by 2030. Achieving this goal could be difficult if preventable maternal deaths continue to occur. The bulk of avertable pregnancy-related deaths and morbidity arises from non-use of skilled maternal healthcare, therefore maternal mortality alone is not a sufficient indicator of a region’s maternal health. Also relevant are indicators related to access and quality of skilled maternal healthcare, particularly among the poorest of the population. Existing studies have established a relationship between favourable maternal health outcomes and the use of skilled healthcare. In countries where the use of specific services such as antenatal care (ANC), in-facility delivery and skilled birth attendance was 98% and above, the MMR was less than 15 (15 deaths per 100,000 live births). In contrast, countries with an MMR of 500 or higher saw relatively low use of skilled pregnancy care. More precisely, these countries registered less than 50% coverage of adequate ANC, in-facility delivery, and skilled birth attendance. The latter mirrors the maternal health situation in Nigeria where, in 2017, only 43% of births were assisted by skilled health personnel and only 39% occurred in health facilities. Rural women in Nigeria are underserved and tend to use skilled healthcare services much less than their urban counterparts. For instance, in 2018, only 26% of rural births occurred in a health facility, compared to 61% of urban births. Consequently, rural women face a disproportionate risk of poor health outcomes including maternal mortality and morbidity compared to urban women. Evidence on reasons for non-use of skilled care across SSA are varied and often context specific. These variations highlight the importance of context in determining mechanisms to address factors that impede women’s use of skilled maternal care. For instance, employed women in rural Kenya and rural Ethiopia are not using skilled maternal care services but for different reasons. In Ethiopia, employment status among pregnant women was associated with non-use and underutilization of skilled maternal healthcare services due to time constraints imposed by women’s jobs and household chores. In rural Kenya, women with seasonal employment were not using skilled maternal healthcare because they could not afford it. Beyond individual factors, it is also essential to consider how factors interact at the level of women’s community and around the organization of health care. For instance, in Nigeria, women in underserved areas were not using free skilled maternal care services due to the limited availability of maternal care services. In other instances, women were deterred by different dimensions of quality of care including 1) structural factors such as medical supplies, drugs and basic infrastructure at health facilities; 2) process factors such as respectful care, and 3) outcomes such as the addressing and averting pregnancy complications. Therefore, understanding these external factors that influence women’s non-use necessitates the involvement of actors across the different sectors, such as community leaders and policymakers, who have vital roles to play in the organization of maternal healthcare. Across SSA, there is a growing appreciation of community leaders as advocates for maternal health. Community leaders, who are often men, are key decision- makers and custodians of culture who have also been shown to wield significant influence over behaviour change, particularly relating to health. Successful maternal health projects across SSA have noted that engaging traditional leaders in maternal health discussions is paramount to achieving change. Not only are community leaders helpful in identifying emerging issues in their communities, but they also create a conducive environment that enables women to define their own health needs and take the necessary actions. Similarly, policymakers are key stakeholders in maternal healthcare because they establish the policies and contexts within which health care is provided. Studies have reported on the insufficient interest and commitment on the part of policymakers to implement evidence-based policies. This is because when policymakers are not included in the research process, they are constrained in their capacity to access, synthesize and utilize the available evidence to improve policies. Therefore, in order to design policies that deliver maximum health benefits, policymakers must be part of the discussion in understanding maternal health-seeking behaviour and exploring factors that could improve women’s use of skilled maternal healthcare. The recent COVID19 pandemic has tested the resilience of health systems and countries’ emergency preparedness and response. Healthcare for women and children is often negatively impacted in these scenarios, it is therefore hypothesized that the COVID19 pandemic will worsen the already poor situation of maternal healthcare utilization in Nigeria. In an effort to understand the challenges of providing essential maternal healthcare services while battling the COVID19 pandemic in Nigeria, a countrywide study is relying on the knowledge and skills of various groups of stakeholders whose insights will provide new direction and policy toward mitigating the negative impact of COIVD19 on an already fragile health system. Such studies are recognising that efforts to address the non-use of skilled maternal healthcare should not occur in isolation of one group of stakeholders, but rather across stakeholders who differ in power, worldviews, and objectives. Previous research in rural Nigeria has leveraged the support of an important group of stakeholders, community leaders, to prioritize maternal health, but it is also important that perspectives on maternal health are expanded to include the voices of policymakers. Therefore, the current study explores factors contributing to the non-use of maternal healthcare services in rural areas of Edo, Nigeria from the perspectives of both policymakers and community leaders. This study extends existing literature on access to and use of skilled pregnancy care services from the perspectives of primary users, women in rural Edo, led by the same research team. Results from this study will be useful in developing policy and programs that prioritize maternal health and also increase women’s access to skilled maternal healthcare in rural Edo, Nigeria. Our findings were reported based on the Consolidated Criteria for Reporting Qualitative Research (COREQ) (See). # Method ## Study design This study uses qualitative description as a qualitative research approach. Qualitative description allows health researchers to examine a phenomenon from a naturalistic perspective and gives a straight description of a phenomenon. In using this method researchers acknowledge existing knowledge of a phenomenon, are able to be flexible in commitment to a theory during the study design and aim for a low-inference interpretation of study findings with an emphasis on describing participants’ views as close to the data as possible. This study presents information from community conversations and key informant interviews with community elders and policymakers respectively, in rural areas of Edo State, Nigeria. Key informants are individuals possessing particular knowledge, status and skill in and understanding of a subject matter. They provide information in various ways including formal interviews, informal conversations, manuscripts, artifacts, or other forms. For this study, policymakers were considered to be in the best position to offer insights from a community and systems-level on non-use of maternal health services and proffer solutions. Policymakers are crucial to the functioning of the health system, and they make decisions that influence service delivery and uptake of skilled maternal health care. In this study, face-to-face in-depth interviews were conducted with various policymakers. Community conversations are interactive processes whereby members of the community are engaged around a common problem and arrive at a solution. One key principle of community conversations is that they create spaces for interaction, change and transfer. This form of qualitative research employs different tools for data collection including conversational dialogues with individuals or reviewing artifacts or group discussions. The common theme in these procedures is an orientation towards problem-solving and the role of the individuals in arriving at solutions. Mainstream approaches to dealing with health issues in African communities have involved assembling people for sensitization sessions or awareness-raising activities. The trade-off has been limited opportunities for dialogue and reflection with and among community members. Community conversations create opportunities for people to discuss health issues away from mainstream social environments thereby rejecting the status quo and enhancing new ways of thinking and questioning. In this study, community conversations occurred as group discussions guided by trained facilitators to support critical thinking and problem solving around the low use of skilled maternal healthcare in their communities. In using this approach for this study, the authors recognize that elders of the various communities possess knowledge and capability to bring about social change individually and collectively around the issues of non-use of maternal health care services. This qualitative study is a part of a larger project by the Women’s Health Action Research Centre in Nigeria and the University of Ottawa, funded under the Innovating for Maternal and Child Health Africa initiative (a partnership of Global Affairs Canada, Canada’s International Development Research Centre, and Canadian Institutes of Health Research). The larger project is a community- based, multi-site, and multi-disciplinary cluster-randomized trial using a mixed methods approach. Details about the larger project have been reported elsewhere. ## Research setting Nigeria’s current population of 206 million makes it the sixth most populous country in the world. Nigeria’s total fertility rate of 5.11 (live births per woman) is projected to drive a population boom and make Nigeria the second largest country in the world by 2100 with a population of 790 million. Nigeria has 36 states and a Federal Capital Territory within which are a total of 774 local government areas (LGAs). It is grouped into six geopolitical zones namely: North West, North East, North Central, South East, South West and South. About 50% of Nigeria’s population reside in rural areas. This study was conducted in Edo State, one of Nigeria’s 36 States and has a population of approximately 3 million people and a land area of 17, 802 square miles. Specifically, this study was conducted in Esan South East (ESE) and Etsako East (ETE), both of which are local government areas (LGA) of Edo state. They are located in the rural and riverine areas of the State. Each LGA comprises 10 political wards within which are several communities, ESE has 100 communities and a total population of 313,717, and ETE has 42 communities and a total population of 145,996. These study sites were chosen because preliminary baseline assessments revealed high maternal mortality rates and low use of primary healthcare facilities. Primary healthcare centres (PHCs) are government-funded facilities and constitute the main source of skilled maternal healthcare in the two LGAs. There are 25 PHCs in ESE and 28 in ETE. Esan South East has one general hospital in the local government’s headquarters (Ubiaja) and Etsako East has two general hospitals; one in the local government’s headquarters (Agenebode) and another in nearby Fugar City. They are used in addition to existing PHCs for referral for maternal health services. ## Characteristics of participants For the community conversations, a total of 151 men elders and 7 women elders aged 50–101 years of age participated in the community conversations. Most of them attained post-primary education, whereas a few had no education. The majority were farmers and artisans. The majority of the participants were Christians, and a few declared no religious affiliation. For the key informant interviews, a total of 6 participants included: one senior official within the State Ministry of Health, one senior official within the State Primary Healthcare Development Agency (SPHCDA), two senior officials responsible for PHCs, with one from ETE and the other from ESE LGAs, two senior Local Government officials, one from ETE and the other from ESE. ## Participant recruitment ### Community elders Community elders were purposefully recruited using locally accepted methods of establishing contact. The lead investigators for the project (FO, LN) identified indigenous guides in ETE and ESE who then introduced them to the traditional ruler of the communities. The project leaders met with the traditional rulers, explained the purpose of the study, and obtained consent to conduct research with community elders. Community elders within a traditional age-based hierarchy are considered influential and agents of change. Target participants were community elders who were often over 50 years old and who were recognized as influential among their communities. Consultations with the traditional ruler identified 10 communities with traditional age-based hierarchies across the two local governments. Traditional rulers in the communities worked with the research team to schedule meetings and communicated with the elders who gathered for the various group discussions. ### Policymakers The lead investigators (FO, LN) identified key informants among known policymakers who held various policy-related positions at the State and Local government levels. Using a purposeful criterion sampling technique, participants’ eligibility to participate were determined based on the following criteria; 1) participants were in key policy positions 2) participants had experience within the PHC system. These criteria were necessary to enhance opportunities to obtain rich insights into the non-use of maternal healthcare services in rural areas. The lead investigators contacted each participant via phone or email with information about the study. Face-to-face in-depth interviews were conducted with 6 policymakers from different institutions in ETE and ESE LGAs and at the state level. ## Data collection ### Community conversations A total of 10 community conversations were conducted, six were conducted in ESE and four in ETE between July 26 to August 16, 2017. Community conversations were conducted as group discussions, each consisting of 12 to 21 participants. Groups were small enough for open dialogue and freedom of expression yet large enough to maximize discussions from a diverse group of elders. Conversations occurred outdoors and were conducted in Pidgin English and a few in the local languages (Ishan or Etsako). The lead investigators trained local researchers to conduct community conversations. The local researchers were conversant with the traditions of the community and spoke English, Pidgin English, Ishan and Etsako. The lead investigators developed a community conversation topic guide which was piloted in a neighbouring village with 12 elders aged 50 years and older. The guide was designed to engage elders in problem-solving. In keeping with cultural practices and values that emphasize oral modes of agreement, all participants provided verbal consent to participate in the study. They all subsequently provided written informed consent. Community conversations proceeded as follows: first, the researchers raised awareness about maternal mortality and preventative strategies. Second, they discussed challenges to women’s access to skilled pregnancy care and finally, elders proffered solutions and mentioned specific ways they will take action to address issues. Conversations lasted about 90 minutes and ended when no further issues arose. Resolutions generated from the discussions were itemized and read back to the elders at the end of the discussion. The elders provided feedback where necessary. ### Key informant interviews Data collection using a key informant interview guide took place from July 16 to August 30, 2017. The principal investigators provided a three-day training session to research assistants who conducted in-depth interviews with participants. In keeping with a qualitative descriptive study, the interview guide was moderately structured to allow for the free description of opinions and experiences. Trained research assistants ran a pilot test of the guide in a community with similar characteristics as the study location. A total of 6 policymakers participated in the study. Quite frequently and for ambivalent reasons, guidelines adopt a sample size of multiples of 10 for interviews. However, studies recommend that in choosing sample size, researchers focus on what has the best opportunity to reach data saturation as that constitutes the gold standard by which purposeful sample sizes are determined in health research. For in-depth interviews, data saturation can be attained in as little as 6 interviews depending on the diversity of data and the sample population. However, the concept of data saturation is also contested within research designs such as qualitative descriptions that stress the uniqueness of each individual’s experience. The authors acknowledge that information obtained from six participants may never truly reach data saturation, the key, however, was to strive to attain thick and rich data. Based on the diverse policymakers interviewed for this study, the authors believe that the data obtained is detailed, nuanced and intricate. Participants signed a consent form prior to participating in the interviews. The interviews lasted for 45 minutes on average and ended when no further issues arose. See for relevant interview questions. ### Ethical considerations The ethical clearance needed for the larger study was granted by the National Health Research Ethics Committee (NHREC) on April 18, 2017 (reference number NHREC/01/01/2007–18/04/2017). Participants gave their free and informed consent to be enrolled in the study. Participants provided written informed consent prior to participating in this study. They were also informed that once they chose to participate, they could withdraw at any time or chose not to answer any questions, to which there would be no negative consequences. To ensure confidentiality, all personal information were not included in transcripts and quoted texts. ### Data analysis The community conversations and key informant interviews were audio-recorded and transcribed by the paid research assistants who were conversant with the spoken languages. Data were transcribed verbatim if in English or translated if in Pidgin English or the local language. Direct translations (word-by-word) were carried out in order to portray the mentality of the participants and present their message as accurately as possible. In cases where syntactical and grammatical structures precluded literal translations, free translations were used to enhance the readability of a text. The primary author (OU) and corresponding author (SY) analyzed the data independently and checked for consistency during frequent discussions. Data analysis was conducted manually and followed the analytical strategies for qualitative description. After immersing themselves in the data, OU and SY read the data line by line, recorded insights, and proceeded to code the data. Next, coded information was sorted to identify patterns and themes from which similarities and differences were identified and extracted for further consideration and analysis. Similar themes generated sub-categories which gave a more general description of the content. Participants’ views of the non-use of maternal health care services in rural Nigeria are presented in the following overarching themes: quality of care, utilization patterns, affordability, and accessibility. ## Trustworthiness The rigour of a study lies in the degree of confidence in methods used, data obtained and interpretation of data. Trustworthiness of a study is necessary for establishing confidence based on various criteria including credibility, dependability, confirmability and transferability. To enhance the credibility of the study, the authors used investigator triangulation; the coding process involved two coders working independently to code the data and working collaboratively to generate themes. Furthermore, this study used method triangulation by having different methods of data collection namely, in-depth interviews with policymakers and community conversations with influential community elders. After data collection, the lead investigators conducted member checks by feeding back data to the participants from whom data was obtained. To enhance confirmability, the primary author (OU) provided thick descriptions of participants’ responses, alongside relevant quotes to confirm interpretations. Quotes were also chosen to represent a typical response relative to the theme. # Results ## Reasons for non-use of skilled maternal health care ### Quality of care Several policymakers and community elders recognized the critical role of skilled healthcare in ensuring optimal maternal health, however, they acknowledged that this knowledge alone was not sufficient to persuade women to use services. See for a summary representation of participants’ responses. Participants opined that the quality of care, or the lack thereof, was a key reason for non-use of skilled maternal healthcare. Poor quality of care manifested as massive shortages in skilled health professionals, apathy and abusive behaviours from healthcare providers, lack of life-saving equipment and lack of safe skilled pregnancy care. Participants indicated that their rural healthcare facilities were in dire need of qualified staff. Reportedly, there was only one doctor attending to the whole local government area. Policymakers attributed this to a massive brain drain underway in the country and a lack of incentives to attract and retain skilled healthcare staff in rural areas. Shortage of qualified healthcare workers has not only created unhealthy work environments for the healthcare workforce, but has also put women in precarious situations. Participants remarked that in some instances, maternal mortality or morbidity had occurred after unqualified staff attended to obstetric emergencies. In other instances, women did not receive care at all due to the lack of healthcare staff. Examples of poor quality of care in rural healthcare facilities were referred to as follows: “ “*A lot of studies have been done to try and find out what the problem is*, *but I can tell you what we know from our end*. *We lack human resources*, *so you can have primary health care facility that has one nurse and 2 community health workers*, *in fact in some places you have just 2 staffs*, *2 community health workers and you will agree with me that it will not be possible to provide 24 hours service*, *so they will share themselves*, *you work in the morning*, *I will work in the evening then the night hours are not covered and many of these deliveries come in the night hours*. *So*, *they \[pregnant women\] get to the healthcare facility and because there are not enough staff*…*she cannot get service*. *So those things affect why the women prefer not to use PHC*. *(Policymaker*, *State Ministry of health)*. “*We \[women\] don’t use PHC because the people are too harsh and do not treat us well*. *They are not qualified to give us injection*…*I feel that if we have qualified doctors and nurses*, *we \[women\] will use PHC regularly than to take self-treatment*” *(Community elder*, *ETE)*. ” Policymakers and community elders expressed their frustrations over disrespectful care provided by healthcare staff who were often hostile to patients. Participants unequivocally cited it as a major deterrent to women’s use of skilled care in rural communities. As participants explained, women in the community acquire information related to quality from the experiences of friends and family which consequently influences their use of skilled healthcare facilities. Community elders recalled instances of apathy and lack of commitment from health care workers, they deemed it offensive that nurses would not attend to their duties at health facilities and would instead demand that women come to their private houses to receive treatment. Participants explained: “ “*What I want to say my people have said it all*. *I was thinking the health centre was built there for the people alone but when you go there*, *they talk to you anyhow*, *you will not see them on duty rather if there is any treatment*, *they take it home to treat*. *There was a day somebody had an accident*, *and the person was rushed to the health centre*, *the nurse was not there to attend to the person*. *When she came*, *she started talking mannerlessly*, *so that is the more reason people do not patronize them*” *(Community elder*, *ESE)*. “*They \[pregnant women\] say I won’t go for antenatal; I don’t believe in it*. *They believe the hospital is a place where if they go there*, *they \[healthcare staff\] will not have time to attend to them and they are too harsh*” *(Policymaker*, *PHC official*, *ESE)* ” ## Utilization patterns The unique perspectives of community elders highlighted a remarkable utilization pattern of health services among pregnant women; women combined traditional, faith-based, and skilled maternal healthcare services at different points of their pregnancies for different reasons. Women were reportedly skeptical of the benefits of skilled pregnancy care and less likely to use it adequately if they or someone they know had experienced pregnancy complications while using skilled care. One community elder contemplated the efficacy of skilled healthcare and posed the question *“is traditional medicine better than hospital medication*? *I don’t know”*. This accentuated the skepticism and doubt surrounding skilled maternal healthcare within the rural community. Some community elders themselves advocated that women use a combination of different types of health services to avoid complications during pregnancy. The extended narratives of elders bring these issues to relief: “ “*I have to say this*, *sometimes*, *though some women register in the health centre and the traditional care centre*, *they still end up having complications in childbirth… if a woman is pregnant and she decides to register in health centre and she delivers safely*, *she will be encouraged*. *But if she has complications in childbirth though registered in a health centre*, *it will not be a good story*. *So*, *we want more enlightenment on this area because the women we have to discuss this with may ask some questions*.” *(Community elder*, *ESE)* “*For me*, *when I got married and my wife was pregnant*, *I registered her in the general hospital*, *and also in a traditional Centre*. *Because my understanding is that there are medications in the hospital and also another type of medication from the traditional*. *Because when it is time for her to get the traditional medications*, *she will get it*, *and when is time for her to get the ones from the hospital she will get them too*. *My advice will be*, *any one whose wife is pregnant should register on both side (traditional care centre and hospital) so that there won’t be complications*.” *(Community elder*, *ESE)* ” However, policymakers expressed concern over the dangers of seeking a combination of health services. They narrated instances where women would patronise skilled health services for fetal ultrasound scans and then patronise traditional birth attendants (TBAs) to interpret the results. The results, they reasoned, are often erroneously interpreted which has caused delays in detecting and addressing pregnancy related complications. Furthermore, narratives of policymakers and community elders contrasted attitudes of TBAs who were perceived as kind, diligent, responsive, and willing to provide care with those of healthcare staff who were perceived as abusive, nonchalant, and sometimes unqualified. Participants opined that women chose to use TBAs or patronized them alongside skilled pregnancy care services because TBAs have an essential role in rural communities. TBAs are non-health professionals who are often older women experienced in providing native care to women throughout pregnancy, childbirth and postpartum. There seemed to be a sense of comfort associated with patronising TBAs as a policymaker explained: “ “*You know*, *some of our people when they get used to a particular way of doing things*, *even when other modern ways are available*, *they think it is more comfortable*, *more convenient for them to reach these TBAs but we are still doing our enlightenment to ensure that people embrace the health facility*.” *(Policymaker*, *ETE)*. ” ## Affordability Among policymakers, affordability was viewed as a major reason for non-use of maternal healthcare services. Policymakers framed affordability of maternal healthcare to include the economic cost of transportation to health facilities, cost of healthcare services and cost of drugs. Policymakers acknowledged that there were individual variations in the cost of care but generally felt that the cost of maternal health services was too high for the rural population. They attributed the high cost of skilled health services to inadequate funding for the rural healthcare sector. They opined that women often sought cheaper healthcare from TBAs whose services were more affordable. > “*The cost of maternity care is I think between N15*,*000 to N20*,*000 > (\$58-\$68) for the average Nigerian*, *but for the average mother in > rural areas*, *it is a huge amount for our population*” > > *(Policymaker*, *PHC official*, *ETE)* > > “*Very few \[pregnant women\] go to the health facilities*. *And even > the health care facilities are not adequately armed to cater to these > women*. *Most of them will opt to go to traditional birth attendants > because they do not have the financial wherewithal to go to these > adequate facilities*. *It is not fair and many of them do not have > jobs so they would rather go to the TBAs that collect their > substance*. *If they are farmers*, *they give TBAs foodstuff and they > are happy to take the delivery*. *If they are probably shoemakers*, > *whatever trade you do*, *the TBAs will collect those materials > instead of the cash that they do not have*. *So*, *they would rather > patronize TBAs*”. > > *(Policymaker*, *SPHCDA)*. However, not all participants were persuaded about the impact of cost on women’s use of maternal healthcare services. Community elders generally seemed convinced that the quality of care superseded the cost of care for rural households. Reportedly, notions of quality care which emphasized compassionate care, the ability of healthcare providers and facilities to manage maternal health conditions, and accessible health facilities determined women’s use of health care facilities in rural areas, not cost. Community elders acknowledged that services from skilled health facilities are often expensive, but they generally reasserted the importance of skilled healthcare in averting maternal and newborn deaths. They also commented on the state of the facilities which they pointed out were in a state of ruin as a result of neglect. They highlighted the importance of a hygienic and functional facility. Views on this were expressed in community conversations: “ “*The reason is not just because of the charges*. *I have never seen anyone who comes back with good attention and complain that the money is too much and tells other women not to go*. *The reason \[for non- use\] are the nurses are not always on duty for their primary assignment*” *(Community elder*, *ESE)*. “*It is not the money that makes us not go there*, *it is the state of the budling*. *I have not been there in many years*, *I hear that the roof of the building is falling*” *(Community elder*, *ETE)*. ” ## Accessibility Participants indicated that women face significant challenges in reaching and accessing care in rural communities. They acknowledged that in communities without a local healthcare facility, women would need to travel long distances to reach health facilities. Even in communities with a local health facility, distance to facilities was further complicated by poor modes of transportation and poor road infrastructure. As participants indicated, poor accessibility often incurred an added cost in seeking care. These factors combined to hinder women’s use of health facilities. Community elders recalled instances whereby women had gone into labour and given birth to their babies on their way to health facilities. They opined that in communities where health facilities were absent, women would often resort to using TBAs who were physically more accessible. Policymakers acknowledged that there was a need for more health facilities to better serve the needs of rural communities. > “*If we need motorcycle to come out from here*, *it is difficult for > us*. *Sometimes*, *if our wives fall into labour at night*, *before we > can come out from here to access the health centre at Eguare (next PHC > community)*, *it will be very difficult*. *You now see that the actual > time a woman would have delivered will now be prolonged because she > does not arrive early*. *Sometimes*, *women give birth on their way*.” > > *(Community elder*, *ESE)* > > “*It is not as if they don’t want to use the PHCs but what I feel is > making them reluctant to use PHCs*…*those that really live very far > can really find it very difficult to access the PHCs*. *So PHCs really > need to spread out more and maybe on a ratio to population*. *Now we > have so many people attending just one PHC or being serviced by one > PHC*… *so we want many more PHCs*…. *so that many more people will > have where to go*…*I think that is what has reduced their usage*. > > *(Policymaker*, *LGA Official*, *ETE)* While the issue of access and distant facilities was highlighted among community elders and policymakers, some policymakers were not cognisant of accessibility issues. They remarked on the easily available and assessable skilled healthcare facilities in the communities. They opined that health centres were at the “doorsteps of women in the communities”. Furthermore, community elders lamented certain health facility policies that limited pregnant women’s access to skilled care. Some facilities required patients to present a registration card prior to accessing care. This was a concern especially during health emergencies when urgent care was needed. A participant recalled an incident: “ “*I notice that in some of our hospital here*, *the most important thing to them is the card*, *Do you have a card*? *somebody is dying and you are asking the person do you have a card*?” *(Community elder*, *ETE)* ” ## Addressing non-use of skilled maternal health services Participants believed that if healthcare centres were running effectively, more women would patronise them for antenatal, childbirth and postpartum care. Participants proffered solutions to addressing non-use of skilled maternal health services in their communities. Participants agreed that the responsibility ultimately falls to the government to improve the state of primary health centres which is under the purview of the local government. Policymakers acknowledged that the health sector was underfunded and assumed the responsibility for collaborating with the government to address issues related to non-use of skilled maternal care. They suggested that the bottom line to increasing women’s use of skilled care is increased funding. They called for more funding for health education and raising awareness of the benefits of skilled maternal care. They called for increased funding to better equip health facilities with medical supplies and equipment. They called for increased funding to offer better remuneration to attract and retain healthcare workers. Policymakers who cited barriers of cost routinely mentioned community health insurance and cost-sharing schemes as a means to increase women’s use of healthcare services. They intimated on forthcoming policies by the government to enact health insurance schemes and increase funding for the healthcare sector. > “*if you run PHCs effectively*, *the women will not prefer to go to > TBAs*. *They will come to your health facilities and because we have > deliveries by skilled attendants*, *we will also reduce the maternal > mortality and the perinatal mortality and the morbidity*, *that’s the > focus of the government and we are working on getting that seriously > working*, *that is*, *our PHCs working effectively*” > > *(Policymaker*, *State Ministry of health)* > > “*We are also trying to capture those people who are self-employed*, > *because they also will need to benefit from health services and also > be able to benefit from this program \[health insurance scheme\] to > reduce out of pocket expenditure because that has been a challenge*, > *paying for health services*, *it has been a challenge*. *It is part > of the reasons why people don’t really patronize orthodox health care > centres*, *they prefer to go to quacks and self-medicate and cause a > lot of problems*, *so the state health insurance scheme is in the > pipeline*, *the state government is working on it*” > > *(Policymaker*, *Ministry of Health)* > > “*Insurance and co-sharing*. *We like it*. *We have been having > community money*, *we use it to help in urgent needs of pregnant women > and children*.” > > *(Community elder*, *ESE)* Community members are not leaving the responsibility to government alone. They disclosed that they already had co-sharing schemes within the communities. These schemes have been used to offset the cost of skilled maternal healthcare services for several women in their communities. They however welcomed the idea of health insurance schemes enacted by the government. Furthermore, community elders indicated their willingness to collaborate with the government in establishing a health facility in their government. Some communities offered their land upon which to build a healthcare facility, others offered to provide hands-on assistance in building PHCs. # Discussion This study explored perceptions of community elders and policy makers on the non-use of skilled pregnancy care in rural Edo State. While community elders and policymakers shared similar views on the topic, their unique understandings and insights into issues faced by women in their communities differed. Findings from the perspectives of community elders and policymakers indicate that women were not using skilled pregnancy care for a variety of reasons. Community elders contended that the use of skilled pregnancy care was driven primarily by quality considerations, not cost. This finding was corroborated by a study in Tanzania where pregnant women in rural and predominantly poor communities were shown to patronize private healthcare facilities more than government facilities even though maternity services in private facilities were twice as expensive as at government facilities. The women reported deficiencies in the quality of care in government facilities as reasons for non-use. From participants’ responses, notions of quality revolved around the severe shortage of skilled healthcare staff, lack of respectful care, and inadequate medical supplies, equipment and infrastructure. This is consistent with studies conducted in rural Nigeria that confirmed that the different dimensions of quality include: 1) structural factors such as medical supplies, drugs and basic infrastructure at health facilities; 2) process factors such as respectful care, and 3) outcomes such as the addressing and averting pregnancy complications determine women’s use of skilled health facilities. In discussing the quality of care, policymakers emphasized the importance of respectful care in determining women’s use of skilled care. This is consistent with studies in Uganda, where respectful maternal care was shown to be a key determinant in women’s use of maternal care services. Respectful care, conceptualized as communication with healthcare staff was a major concern for mothers who reported that respectful communication with their midwives encouraged their use of skilled pregnancy care, while disrespectful communication such as verbal abuse, discouraged mothers from using skilled care facilities. Furthermore, similar findings across Africa indicate the association between the severity of lack of qualified staff and basic medical supplies and pregnant women’s use of healthcare facilities. Furthermore, while out-of-pocket cost has often been cited as an impediment to women’s use of skilled pregnancy care in rural Nigeria, the current study saw a contradictory and somewhat unexpected finding. Community elders did not deem the cost of skilled pregnancy care catastrophic, although both elders and policymakers acknowledged it as a barrier. The insights of community elders on cost are not surprising because men in this community are often ascribed the role of financial providers. Furthermore, studies across Africa have highlighted the prevalence of men and community engagement in maternal health. It is important to recognize that in most rural communities, skilled modern healthcare coexists with indigenous healthcare. Rural residents rely on indigenous medicines for a large proportion of their healthcare needs. As indicated by community elders and policymakers, women often combined different types of pregnancy care, sometimes to their detriment. Compared to policy makers who trusted the efficacy of skilled pregnancy care, some community elders questioned its efficacy and disclosed that their beliefs influenced women’s use of skilled pregnancy care. Similar findings in Ethiopia and Kenya corroborate the influence of community leaders over health behaviors in their community. Furthermore, policymakers were aware of widespread beliefs in communities that the different types of health providers specialize in different pregnancy related health conditions and opined that this influenced women’s utilization patterns. Their observations are comparable to similar studies in rural Nigeria where women were shown to have utilization patterns that combined different types of maternal care including formal, traditional and religious pregnancy care. Reasons for this were attributed to women’s attitudes and beliefs, as well as geographic and financial limitations. Other studies across Africa have noted that women’s continued use of TBAs even with the availability of skilled pregnancy care is driven by various contextual factors such as notions of provider expertise, health experiences and outcomes of family members and friends. This finding raises policy implications. TBAs are well established in the community, autonomous of the formal health system and are integral parts of the religious and cultural systems of their community. They use practical approaches and experiential knowledge, steeped in religious and cultural norms and practices to provide care to pregnant women. Their influence can be leveraged to work collaboratively with the formal healthcare system to provide support and health promotion activities during pregnancy and childbirth. Views from policymakers and community elders on accessibility issues underscores the gaps in the physical coverage of health facilities in their communities, which consequently leaves some members of the community underserved. As participants noted, geographic distance to facilities impedes women’s use of skilled pregnancy care. This is worsened by the limited availability of infrastructure such as good roads and emergency or ambulatory services. Studies in rural Nigeria and rural Ghana showed that distance was a determinant of women’s use of maternal health services. Participants suggest that building more facilities will ensure wider coverage among the community and essentially increase healthcare use. In addition to building facilities, participants recommended improving infrastructure in facilities, availing communities with emergency transportation services, improving medical supplies and equipment. Prior research has identified similar strategies to enhance use of health facilities in rural communities in Nigeria. This prior research also suggests integration and coordination among health providers to provide holistic care during pregnancy, delivery and postpartum. For instance, the use of health cards with relevant health information would enhance access to different health facilities and prompt management of health conditions. It is widely acknowledged that community participation is key to successfully implementing and maintaining health services in rural areas. Community elders offered practical assistance in the form of their community land and labour to support health facilities in their communities. The community was also mobilizing financial resources to aid community members utilize skilled health facilities. These narratives of community support for skilled healthcare are promising. Evidence has shown that skilled health service delivery and uptake are enhanced with active community participation whereby communities have a stake in ensuring quality services are provided and sustained. Health insurance schemes were proposed by policymakers. Likewise, community elders welcomed the idea of a more structured health insurance scheme. Nigeria continues to face challenges in enrollment and uptake of community health insurance, however, evaluation of the scheme in Rwanda where more than 90% of the population were covered by health insurance showed improved skilled medical care utilization and protection of households from catastrophic cost of healthcare. Furthermore, recent policy initiatives in Nigeria, as discussed by policymakers, could strengthen maternal health service delivery, and improve women’s use of skilled care. Policy makers mentioned their plans to embark on a cost recovery mechanism for rural public health facilities known as the drug revolving fund. Through this system, revenue generated from the sale of drugs to patients is used to purchase new drugs. The system will ensure sustainability and continuity of essential drugs while also making them affordable for patients. Furthermore, ongoing policy initiatives in Nigeria such as the Primary Health Care Under One Roof (PHCUOR), aims to integrate primary health care under one authority and enhance access to funds for PHCs. Funds obtained through this policy can be allocated towards the provision of essential drugs, the maintenance of healthcare facilities, health care transportation, and the development of human resources for rural PHCs. Triangulation of methods and participants elicited rich and diverse views of reasons for non-use of skilled maternal health services in rural Edo State, however, this study is not without limitations. Majority of the community elders in the study sites are men, therefore the perceptions of men were disproportionately represented. The results from this study could have differed if more data was obtained from women elders who are likely to have lived experience of pregnancy and childbirth. The authors reported and interpreted the data with this in mind. Furthermore, this study did not account for socioeconomic differences between participants as this could impact their perspectives on the affordability of skilled healthcare. However, the authors noted general characteristics of the participants including educational level and sources of employment. Furthermore, it is impossible to rule out perspective bias from policymakers who themselves are responsible for health programs and policies. In analyzing the data, free translations were used to enhance readability of the texts in certain instances, the authors acknowledge that in choosing this method, there is the likelihood of misinterpreting participants’ responses. Findings might not be generalizable to all of rural Nigeria as each community’s priorities and experiences with skilled pregnancy care differ. # Conclusion This study explored perceptions of community elders and policy makers on the non-use of skilled maternal healthcare in rural Edo State. Both groups expressed similar knowledge of issues preventing women’s use of skilled care, they also offered unique and sometimes different perspectives on why women do not use skilled maternal healthcare in ETE and ESE. Community elders’ influence on women’s combination of different types of care highlights a need to improve awareness of the importance of skilled pregnancy care at the community level. Findings also highlight the need to strengthen the availability, accessibility, and the affordability of rural healthcare facilities. In addition, the provision of high-quality maternal healthcare should be prioritized. Both policymakers and community leaders deemed disrespectful care a barrier to women’s use of skilled care. It is important that interventions that emphasize patient-centered care be made a priority in order to enhance healthcare staff attitudes. As indicated by the policymakers, substantial funding of the rural healthcare sector is required alongside policy changes that prioritize health systems reforms to address quality issues. It Is also important that pro-poor policies are enacted to reduce the financial burden on the poorest of the population. It is noteworthy that the situation described in this study preceded the COVID19 pandemic which continues to test the resilience of health systems and countries’ emergency preparedness and response especially in underserved communities. # Supporting information We thank Francis Igberaese, and Joab Oghene who coordinated the data collection in Esan South East and Etsako East respectively; and Michael Ekholuenetale, Michael Alli, Mary-Jane Emiowele, Precious Ntulu, Best Ojemhen, Jessy Ezebuihe, Peace Oppogen, Progress Emoitotoga, Abubakar Zuleya Ogechukwu Onwuma who were data collectors in the two LGAs. We are also grateful to Evans Ejedenawe, Raphael Okpaire, Cynthia Okojie, Tayo Ozobo, Akingbe Aminat, Ebunu Fatimetu, who served as community focal persons in both LGAs. They were instrumental in helping the project teams to gain access to the project communities. ANC Antenatal care ESE Esan South East ETE Etsako East KII Key Informant Interview LGA Local Government Area MMR Maternal Mortality Rate PHC Primary Health Care TBA Traditional Birth Attendants [^1]: The authors have declared that no competing interests exist.

# Introduction The sample size selected from each stratum in a stratified random sample survey (SRSS) must be known and chosen to lower the survey cost, or the estimator’s sample variance must be determined to determine the estimate’s precision. When the population mean of the characteristics is essential for the L strata of the stratified sampling, sample allocation $n_{h}^{*}(h = 1,2,\ldots,L)$ helps maintain the variance of the stratified sample. In the *h*^th stratum (*h* = 1,2,…,*L*), *N*_*h* denotes stratum size, ${\overline{Y}}_{h}$ stands for the population mean, $W_{h} = \frac{N_{h}}{N}$ denotes the stratum weight, and $S_{h}^{2}$ indicates the population mean square. The population mean for the *j*^th character ${\overline{Y}}_{j}(j = 1,2,\ldots,p)$ is denoted by ${\overline{y}}_{jst}$. Then, the sampling variance of ${\overline{y}}_{jst}$ is given by: $$V({\overline{y}}_{jst}) = {\sum\limits_{h = 1}^{L}{\frac{W_{h}^{2}S_{hj}^{2}}{n_{i}} -}}{\sum\limits_{h = 1}^{L}\frac{W_{h}^{2}S_{hj}^{2}}{N_{i}}},\mspace{14mu} j = 1,2,\ldots,p$$ where, $S_{hj}^{2} = \frac{1}{N_{h} - 1}{\sum\limits_{i = 1}^{N_{h}}\left( {y_{ihj} - {\overline{Y}}_{hj}} \right)^{2}}$ is the variance for *j*^th character in the *h*^th stratum. In the case of nonresponse, let the *h*^th stratum be split into *N*_*h*1, the size of the respondent’s group and *N*_*h*2 = *N*−*N*_*h*1, the size of the non-respondent’s group. The sample size of the *h*^th stratum (*n*_*h* unit) is split into *n*_*h*1 units for the respondent’s group and those left over for the non-respondent group with *n*_*h*2 = *n*_*h*−*n*_*h*1 units. For the nonresponse problem, subsamples of size were obtained from the non-respondent group for the second attempt as follows $$r_{h} = {n_{h2}/k_{h}};\mspace{9mu}\forall k_{h} \geq 1,\mspace{9mu} h = 1,2,\ldots,L$$ where, among the *h*^*th* stratum of people who did not fill out the survey, the sample proportion is represented by 1/*k*_*h*. The unbiased estimate of the *N*_*h*1 and *N*_*h*2 are the ${\hat{N}}_{h1} = n_{h1}{N_{h}/n_{h}}$ and ${\hat{N}}_{h2} = n_{h2}{N_{h}/n_{h}}$. When utilizing the Hansen–Hurwitz method, the following formula is used to calculate the estimator of the stratum mean ${\overline{Y}}_{jh}$ in the *h*^*th* stratum for the *j*^*th* character: $${\overline{y}}_{jh(w)} = \frac{n_{h1}{\overline{y}}_{jh1} + n_{h2}{\overline{y}}_{jh2(r_{h})}}{n_{h}}$$ for the *h*^*th* stratum, the sample means of *n*_*h*1 units from the respondent group and *r*_*h* units from the subsample from the non-respondent group are denoted by ${\overline{y}}_{jh1}$ and ${\overline{y}}_{jh2(r_{h})};j = 1,2,\ldots,p$. The unbiased estimate of the variance ${\overline{y}}_{jh(w)}$ for the *j*^*th* characteristic is represented by: $$V({\overline{y}}_{jh(w)}) = \left( {\frac{1}{n_{h}} - \frac{1}{N_{h}}} \right)S_{jh}^{2} + \frac{W_{h2}^{2}S_{jh2}^{2}}{r_{h}} - \frac{W_{h2}S_{jh2}^{2}}{n_{h}}j = 1,2,\ldots,p\mspace{9mu}{and}\mspace{9mu} h = 1,2,\ldots,L$$ and $$S_{jh}^{2} = \frac{1}{N_{h} - 1}{\sum\limits_{i = 1}^{N_{h}}\left( {y_{jhi} - {\overline{Y}}_{jh}} \right)^{2}}\mspace{9mu}{and}\mspace{9mu}{\overline{Y}}_{jh} = {\sum_{i = 1}^{N_{h}}{y_{jhi}/N_{h}}}$$ $$S_{jh2}^{2} = \frac{1}{{\hat{N}}_{h2} - 1}{\sum\limits_{i = 1}^{{\hat{N}}_{h2}}\left( {y_{jhi} - {\overline{Y}}_{jh2}} \right)^{2}}\ {and}\ {\overline{Y}}_{jh2} = {\sum_{i = 1}^{{\hat{N}}_{h2}}{y_{jhi}/{\hat{N}}_{h2}}}$$ *W*_*h*2 = *N*_*h*2/*N*_*h* is the weight for the non-respondent group in the *h*^*th* stratum. An unbiased estimate of the population mean ${\overline{Y}}_{j}$ for the *j*^*th* characteristics can be determined by $${\overline{y}}_{j(w)} = {\sum\limits_{h = 1}^{L}{W_{h}{\overline{y}}_{jh(w)}}}\mspace{9mu}{and}$$ $$V\left( {\overline{y}}_{j(w)} \right) = {\sum\limits_{h = 1}^{L}{W_{h}^{2}V\left( {\overline{y}}_{jh(w)} \right)}}$$ after that $$V\left( {\overline{y}}_{j(w)} \right) = {\sum\limits_{h = 1}^{L}\frac{W_{h}^{2}\left( {S_{jh}^{2} - W_{h2}S_{jh2}^{2}} \right)}{n_{h}}} + {\sum\limits_{h = 1}^{L}{\frac{W_{h}^{2}W_{h2}^{2}S_{jh2}^{2}}{r_{h}} = V_{j};}}\mspace{9mu} j = 1,2,\ldots,p.$$ Because more than one characteristic is evaluated in multivariate SRSS, it is possible that the optimal allocation of resources for one characteristic will not be optimal for other characteristics. In compromise allocation, to arrive at a functional allocation that is, in some respects, optimal for each characteristic, there is a condition that must be satisfied, known as a compromise. The estimator of sampling variance for the total population (or mean) with a fixed cost is optimized in the sampling literature by defining the units of sample sizes among strata. The allocation problem is defined as minimizing the overall survey cost given a fixed estimator precision. The sampling process entailed assessing the properties of each unit randomly selected from the sample. This is known as "multivariate or multiple response sampling." The resulting allocation is well known as the optimal allocation. Because the cost measurement is set and does not fluctuate from one stratum to the next, one way to decrease the total expense of the survey is to decrease the number of respondents included in the sample. Khan et al. discussed the nonresponse case of multivariate SRSS in the presence of nonlinearity to obtain the optimal allocation and size of the subsample using the Lagrange Multiplier (LM) method. Díaz-García and Garay-Tápia considered the allocation problem of SRSS of stochastic nonlinear programming (NLP) and solved it using different techniques such as the modified ∈-model, LMs, V-model, chance constraints and E-model. Varshney et al. discussed double sampling for multivariate SRSS of the multiobjective NLP problem and solved it with goal programming (GP) to obtain an optimal compromise allocation. Varshney et al. described a solution procedure using GP for multiobjective integer NLP problems for nonresponse cases of SRSS and showed practical utility with a numerical illustration. Ali et al. considered the multivariate SRSS of the nonlinear stochastic programming problem and used the D₁-distance, Chebyshev approximation method, and GP to obtain a compromise allocation. Ali and Haseen provided geometric programming for multivariate SRSS for the case of nonresponse and used the first-phase solution to obtain the optimal allocation of the second phase by applying it as a role model of the primal-dual relationship theorem. Raghav et al. discussed the solution procedure of the GP, value function, distance-based, and ∈-constraint techniques for the multiobjective NLP problem of nonresponse cases with a numerical example. Varshney and Mradula estimated the p-population mean for nonresponse cases and a solution procedure for lexicographic GP with numerical illustrations for practical utility. Haseen et al. formulated a multivariate SRSS for the nonresponse case of the multiobjective stochastic programming problem, used the chance constraint to convert the NLP problem, and modified ∈ – model. GP, D₁-distance and fuzzy programming were used to determine the Pareto optimal allocation of the expressed model. Khowaja et al. described the multiobjective NLP problem in the presence of a quadratic cost function for different circumstances, namely, partial, complete, or null stratified sampling information. Ghufran *et al*. discussed all integer NLP problems and used the measure of "minimizing the sum of the squares of the coefficient of variation for different characteristics." Using the fuzzy technique, the author developed a methodology to determine the optimal allocation for a multivariate SRSS. The NLP problem of the two-stage stratified Warner’s randomized response model was studied by Ghufran et al. to establish the best allocation in the presence of nonresponse. Gupta and Bari dealt with a multivariate SRSS with uncertainties of fuzzy parabolic numbers and a fuzzy multiobjective NLP problem with a developed quadratic cost function. An α-cut was used to defuzzify the fuzzy parabolic number. Khan *et al*. determined the sample sizes of the formulated multistage decision problem, where various strata were present with several characteristics and used dynamic programming approaches to obtain an integer solution. below summarises some of the related works on multivariate stratified sampling topic. Khowaja *et al*. discussed linearizing the nonlinear objective functions of a multiobjective NLP problem and used the GP technique to solve the approximation of an integer linear programming problem. Ghufran *et al*. dealt with a two- stage stratified Warner’s randomized response, and the GP technique was used for the formulated multiobjective integer NLP problem with linear and quadratic cost functions. Khan et al. used a dynamic programming technique to determine the Pareto optimal allocation for the integer NLP problem of multivariate SRSS. Khowaja *et al*. proposed a technique for attaining a compromise Pareto optimal allocation for multivariate SRSS. The value function technique converts the formulated multiobjective integer NLP problem into a single-objective problem. The LMs technique was used to obtain an optimal compromise solution for continuous sample sizes. Varshney *et al*. considered a multivariate SRSS with unknown stratum weight, and an optimal sampling was provided to estimate the unknown population means for nonresponse by utilizing double sampling for stratification. The authors used the GP method in the solution procedure development process. Varshney *et al*. discussed multiobjective NLP problems for multivariate SRSS. The author minimized the individual estimated coefficients of variation using auxiliary information and a nonlinear cost function for various characteristics with a fixed budget to find a compromise solution using the GP technique. Varshney *et al*. described the optimal allocation in multivariate SRSS for nonresponse, and a solution procedure was developed using the GP method for the multiobjective integer NLP problem. Ghufran *et al*. discussed the optimum allocation problem by minimizing the variation coefficient for various characteristic estimators under the travel cost constraint for a multivariate SRSS. To obtain a compromise solution, the multiobjective NLP problem was solved using different techniques, such as distance-based, ∈-constraint, and value function methods. Dáz-Garcá and Cortéz explored the allocation problem in multivariate SRSS and formulated it as a nonlinear problem of matrix optimization of integers with the constraints of a cost function and a predetermined sample size. They found that this problem could be resolved. Khan *et al*. determined the optimal allocation and subsample size for different strata of multivariate SRSS for the nonresponse case and formulated it as an NLP problem. The LMs method was used to obtain the optimal allocation. An explicit formula was obtained for the optimal allocation and subsample sizes. Khan *et al*. discussed optimum allocation using various compromise criteria for a multiple-response SRSS. Goli and Golmohammadi proposed a multiobjective mathematical model to determine locations and perform distributions in a closed-loop supply chain under competitive conditions. Alike, Raghav *et al*. explored a multiobjective mathematical programming problem to maximize the profit and minimize the cost function utilizing various methods for preventative system maintenance. Gupta et al. discussed the stratified random sampling with minimization of variance and solved it with IF programming. Additionally, using Pareto-based algorithms, Tirkolaee *et al*. discussed multiobjective optimization for a sustainable pollution-routing problem using cross-dock selection and found solutions using different optimization techniques. Similarly, Sangaiah *et al*. addressed a robust mixed-integer linear programming model for LNG sales planning over a given time horizon, aiming to minimize the costs of the vendor. Besides, Goli *et al*. developed a product portfolio optimization problem using a robust optimization approach and multiobjective invasive weed optimization algorithm. They used a robust optimization approach to address this issue, considering the profit margin uncertainty in real-world investment decisions. ## Methodology-optimistic and pessimistic method of intuitionistic fuzzy programming The sampling problem is formulated as a univariate nonlinear mathematical programming problem. This is intended to satisfy all non-negativity criteria, and minimize the variance, while the cost is a constraint. Many scholars have conducted similar research to find the optimum or best solution by using the aforementioned or other related techniques. The general single objective problem after the formulation is of the form: $$Maximize\ \left( {or\ Minimize} \right)\ z = f(X)$$ $$Subject\ to\ g(X)( \leq or = or \geq)b$$ $$X \geq 0$$ The multiobjective multivariate stratified sampling (MOMSS) model where the variances form the objectives and cost works as a constraint. The MOMSS model is expressed as follows: $$\begin{array}{l} {{Min}\ \left\{ {V_{j}(n)} \right\}\ (j = 1,2,\ldots,p)} \\ {s.t.\ n \in F} \\ \end{array}$$ *F* is a feasible set and *V*_*j*(*n*) (*j* = 1,2,…,*p*) represents the multiple objectives to be minimized. The intuitionistic fuzzy MOMSS model is expressed as $$\begin{array}{l} {{Min}\ \left\{ {{\widetilde{V}}_{j}(n)} \right\}\ (j = 1,2,\ldots,p)} \\ {s.t.\ n \in F} \\ \end{array}$$ ${\widetilde{V}}_{j}(n)\ \left( {j = 1,2,\ldots,p} \right)$ are various objectives that need to be minimized under an intuitionistic fuzzy (IF) sense. The objective functions under intuitionistic fuzzy sense with the aspiration levels *g*_*j*; the positive admissible acceptances *t*_*j*(*j* = 1,2,…,*p*), optimistic quantities *o*_*j*(*j* = 1,2,…,*p*), and pessimistic quantities *q*_*j*(*j* = 1,2,…,*p*) given by decision-makers (DMs), where *o*_*j*≤*q*_*j*≤*t*_*j*. IFMOMSS is as follows: $${Find}\ n\ s.t.\left\{ \begin{array}{l} {{Min}\ \left\{ {{\widetilde{V}}_{j}(n)} \right\}\widetilde{<}g_{j}\ (j = 1,2,\ldots,p)} \\ {s.t.\ n \in F} \\ \end{array} \right.$$ The membership functions (MFs) of the objective functions under the IF sets are as follows: $$\mu_{j}(V_{j}(n)) = \left\{ \begin{array}{l} {1,\mspace{166mu} V_{j}(n) \leq g_{j}} \\ {1 + \lbrack{{(g_{j} - V_{j}(n))}/{t_{j}\rbrack,\mspace{9mu} g_{j} < V_{j}(n) \leq g_{j} + t_{j}}}} \\ {0,\mspace{166mu} V_{j}(n) > g_{j} + t_{j}} \\ \end{array} \right.$$ We construct the non-MF of *υ*_*j*(*V*_*j*(*n*)), which must satisfy $\mu_{j}(V_{j}(n)) + \upsilon_{j}(V_{j}(n)) \leq 1$. For *V*_*j*(*n*)≤*g*_*j* and $\mu_{j}(V_{j}(n)) = 1$, $\upsilon_{j}(V_{j}(n)) = 0$; while for *V*_*j*(*n*)≥*g*_*j* and 0≤*μ*_*j*(*V*_*j*(*n*))\<1, $0 \leq \upsilon_{j}(V_{j}(n)) < 1$. The non-MF of *V*_*j*(*n*) can be defined as follows: $$\upsilon_{j}(V_{j}(n)) = \left\{ \begin{array}{l} {0,\mspace{198mu} V_{j}(n) \leq g_{j} + (1 - \lambda)(t_{j} - q_{j})} \\ {1 + \frac{V_{j}(n) - g_{j} + (1 - \lambda)(t_{j} - q_{j})}{\lambda(o_{j} + t_{j}) + (1 - \lambda)q_{j}},\mspace{9mu} g_{j} + (1 - \lambda)(t_{j} - q_{j}) < V_{j}(n) \leq g_{j} + (1 - \lambda)(t_{j} - q_{j})} \\ {1,\mspace{198mu} V_{j}(n) > g_{j} + (1 - \lambda)(t_{j} - q_{j})} \\ \end{array} \right.$$ The parameter *λ*∈\[0,1\]; DM’s preference information. The DM prefers positive feelings *λ*∈\[0.5,1\], negative feelings *λ*∈\[0,0.5\]and *λ* = 0.5 demonstrates that DM does not favor either positive or negative feelings. The unique forms of the non-MF were obtained for different parameter values *λ*∈\[0,1\]. For *λ* = 1, it is easy to see that $\mu_{j}(V_{j}(n)) = 0$, $\upsilon_{j}(V_{j}(n)) < 1$ in the interval $\lbrack g_{j} + t_{j},g_{j} + t_{j} + o_{j}\rbrack$, the values of MFs will be zero while the values of non- MFs will be less than one. Yu *et al*. called MF an optimistic method. The non- MF of *V*_*j*(*n*) can then be defined as follows: $$\upsilon_{j}(V_{j}(n)) = \left\{ \begin{array}{l} {0,\mspace{162mu} V_{j}(n) \leq g_{j}} \\ {1 + \frac{V_{j}(n) - (g_{j} + t_{j} + o_{j})}{o_{j} + t_{j}},\mspace{9mu} g_{j} < V_{j}(n) \leq g_{j} + t_{j} + o_{j}} \\ {1,\mspace{162mu} V_{j}(n) > g_{j} + t_{j} + o_{j}} \\ \end{array} \right.$$ When *λ* = 0, for the interval $\lbrack g_{j},g_{j} + t_{j} - q_{j}\rbrack$, we see that $\upsilon_{j}(V_{j}(n)) = 0$ $\mu_{j}(V_{j}(n)) < 1$, the values of MFs will be less than one while non-MFs will be zero. Then, the MF is called a pessimistic approach. Therefore, the non-MFs of *V*_*j*(*n*) can be defined as follows: $$\upsilon_{j}(V_{j}(n)) = \left\{ \begin{array}{l} {0,\mspace{162mu} V_{j}(n) \leq g_{j} + t_{j} - q_{j}} \\ {1 + \frac{V_{j}(n) - (g_{j} + t_{j})}{q_{j}},\qquad g_{j} + t_{j} - q_{j} < V_{j}(n) \leq g_{j} + t_{j}} \\ {1,\mspace{162mu} V_{j}(n) > g_{j} + t_{j}} \\ \end{array} \right.$$ The grouping of the optimistic and the pessimistic methods for 0\<*λ*\<1 is called a mixed method. The IFMONLP problem’s flowchart can be seen in. Using Eqs and, we solved our model as follows: $${Find}\ n\ s.t.\left\{ \begin{array}{l} {{Max}\ (\mu - \upsilon)} \\ {\upsilon_{j}(V_{j}(n)) = 1 + \frac{V_{j}(n) - (g_{j} + t_{j})}{q_{j}}} \\ {\mu_{j}(V_{j}(n)) = 1 + \frac{g_{j} - V_{j}(n)}{t_{j}}} \\ {\mu \leq \mu_{j}(V_{j}(n)),\upsilon \geq \upsilon_{j}(V_{j}(n))} \\ {\mu_{j}(V_{j}(n)) + \upsilon_{j}(V_{j}(n)) \leq 1} \\ {n \in F,\mu_{j}(V_{j}(n)) \in \lbrack 0,1\rbrack,\upsilon_{j}(V_{j}(n)) \in \lbrack 0,1\rbrack} \\ \end{array} \right.$$ # Multivariate SRSS designs This section discusses the multivariate stratified sampling design for optimum allocation problems under complete response and nonresponse. ## Multivariate SRSS- a case of complete response **Defination 1:** Stratified sampling is a type of response case sampling technique in which the population is divided into subgroups (strata) based on one or more characteristics, and a sample is taken from each stratum. The goal of stratified sampling is to ensure that each stratum is represented in the sample in proportion to its representation in the population. Several authors have discussed multiobjective optimization problems in multivariate stratified sampling subject to sampling cost and other feasibility conditions. It is also observed that the sampling costs may not always be linear; hence, the sampling costs are considered nonlinear. For instance, the cost of travel between the chosen strata units is large and cannot be ignored; it is possible that the cost function does not accurately estimate the total cost acquired. Beardwood *et al*. proposed that the visiting cost *c*_*h* selected for units from the *h*^th stratum can be represented as $t_{h}\sqrt{n_{h}},h = 1,2,\ldots,L$ for the cost of travel per unit from the *h*^th stratum. The distance-based estimation between *k* randomly distributed points is related to $\sqrt{k}$. As a result of this circumstance, the sum of the costs associated with travel, measurement, and overhead will constitute the overall cost of an SRSS. The total cost *C* is expressed as follows: $$C = c_{0} + {\sum\limits_{h = 1}^{L}{c_{h}n_{h}}} + {\sum\limits_{h = 1}^{L}{t_{h}\sqrt{n_{h}}}}$$ Based on this discussion, the compromise allocation of a multiobjective NLP problem with a quadratic cost function is stated as follows: $$\left. \begin{array}{l} {MinV_{j}(n) = {\sum\limits_{h = 1}^{L}{\left( {\frac{1}{n_{h}} - \frac{1}{N_{h}}} \right)W_{h}^{2}S_{jh}^{2}}}} \\ {subject\ to} \\ {{\sum\limits_{h = 1}^{L}{c_{h}n_{h}}} + {\sum\limits_{h = 1}^{L}{t_{h}\sqrt{n_{h}}}} \leq C_{0};n_{h} \geq 0,\ h = 1,2,\ldots,L} \\ \end{array} \right\},j = 1,2,\ldots,p$$ The available cost to meet travel and measurement expenses is *C*₀ = *C*−*c*₀; *V*_*j*(*n*) is the sampling variance and $S_{jh}^{2},h = 1,2,\ldots,L$ are the known population variance. **Example 1:** Data with five strata and three characteristics were simulated for a budget of 1,000. Then, the compromise allocation of a multiobjective NLP problem with a quadratic cost function using the data in is as follows: $$\begin{array}{l} {{Minimize}\ V_{1}(n) = \frac{1.299435}{n_{1}} + \frac{4.321858}{n_{2}} + \frac{2.419448}{n_{3}} + \frac{8.603727}{n_{4}} + \frac{3.192979}{n_{5}}} \\ {{Minimize}\ V_{2}(n) = \frac{37.31234}{n_{1}} + \frac{2.70621}{n_{2}} + \frac{24.33143}{n_{3}} + \frac{33.95098}{n_{4}} + \frac{3.543856}{n_{5}}} \\ {{Minimize}\ V_{3}(n) = \frac{0.742534}{n_{1}} + \frac{1.413692}{n_{2}} + \frac{0.867349}{n_{3}} + \frac{1.391779}{n_{4}} + \frac{0.91228}{n_{5}}} \\ {{subject}\ {to}\ 4n_{1} + 4.5n_{2} + 5n_{3} + 5.5n_{4} + 6n_{5} + 7\sqrt{n_{1}} + 9\sqrt{n_{2}} + 10\sqrt{n_{3}} + 12\sqrt{n_{4}} + 13\sqrt{n_{5}} \leq 1000} \\ {\mspace{130mu} 2 \leq r_{h} \leq {\hat{n}}_{h2},2 \leq n_{h} \leq N_{h}\& n_{h},r_{h} \in {integers}\ \forall h = 1,2,3,4,5.} \\ \end{array}$$ The MFs of the IF sets that represent the objective functions are as follows: $$\mu_{1}(V_{1}(n)) = \left\{ \begin{array}{l} {1,\mspace{166mu} V_{j}(n) \leq 0.6323559} \\ {1 + \frac{\lbrack 0.6323559 - V_{1}(n)\rbrack}{0.4}\mspace{27mu} 0.6323559 < V_{1}(n) \leq 1.0323559} \\ {0,\mspace{166mu} V_{1}(n) > 1.0323559} \\ \end{array} \right.$$ $$\mu_{2}(V_{2}(n)) = \left\{ \begin{array}{l} {1,\mspace{162mu} V_{2}(n) \leq 2.752765} \\ {1 + \frac{\lbrack 2.752765 - V_{2}(n)\rbrack}{1.29}\mspace{32mu} 2.752765 < V_{2}(n) \leq 4.042765} \\ {0,\mspace{162mu} V_{2}(n) > 4.042765} \\ \end{array} \right.$$ $$\mu_{3}(V_{3}(n)) = \left\{ \begin{array}{l} {1,\mspace{166mu} V_{3}(n) \leq 0.1802541} \\ {1 + \frac{\lbrack 0.1802541 - V_{3}(n)\rbrack}{0.09}\mspace{27mu} 0.1802541 < V_{3}(n) \leq 0.2702541} \\ {0,\mspace{162mu} V_{3}(n) > 0.2702541} \\ \end{array} \right.$$ **Optimistic approach (*λ* = 1):** The non-MFs of *V*_*j*(*n*)∀*j* = 1,2,3 are as follows: $$\upsilon_{1}(V_{1}(n)) = \left\{ \begin{array}{l} {0,\mspace{166mu} V_{1}(n) \leq 0.6323559} \\ {1 + \frac{\lbrack V_{1}(n) - 1.1323559\rbrack}{0.5},\mspace{27mu} 0.6323559 < V_{1}(n) \leq 1.1323559} \\ {1,\mspace{166mu} V_{1}(n) > 1.1323559} \\ \end{array} \right.$$ $$\upsilon_{2}(V_{2}(n)) = \left\{ \begin{array}{l} {0,\mspace{166mu} V_{2}(n) \leq 2.752765} \\ {1 + \frac{\lbrack V_{2}(n) - 5.012765\rbrack}{2.26},\mspace{27mu} 2.752765 < V_{2}(n) \leq 5.012765} \\ {1,\mspace{166mu} V_{2}(n) > 5.012765} \\ \end{array} \right.$$ $$\upsilon_{3}(V_{3}(n)) = \left\{ \begin{array}{l} {0,\mspace{166mu} V_{3}(n) \leq 0.1802541} \\ {1 + \frac{\lbrack V_{3}(n) - 0.3102541\rbrack}{0.13},\mspace{27mu} 0.1802541 < V_{3}(n) \leq 0.3102541} \\ {1,\mspace{166mu} V_{3}(n) > 0.3102541} \\ \end{array} \right.$$ Then, the optimistic model of the response multivariate stratified sampling is as follows: $$\begin{array}{l} {{Max}\ (\mu - \upsilon)} \\ {{subject}\ {to}} \\ {\mu_{1}(V_{1}(n)) = 1 + \frac{\lbrack 0.6323559 - V_{1}(n)\rbrack}{0.4},\mu_{2}(V_{2}(n)) = 1 + \frac{\lbrack 2.752765 - V_{2}(n)\rbrack}{1.29}} \\ {\upsilon_{1}(V_{1}(n)) = 1 + \frac{\lbrack V_{1}(n) - 1.1323559\rbrack}{0.5},\upsilon_{2}(V_{2}(n)) = 1 + \frac{\lbrack V_{2}(n) - 5.012765\rbrack}{2.26}} \\ {\mu_{3}(V_{3}(n)) = 1 + \frac{\lbrack 0.1802541 - V_{3}(n)\rbrack}{0.09},\upsilon_{3}(V_{3}(n)) = 1 + \frac{\lbrack V_{3}(n) - 0.3102541\rbrack}{0.13}} \\ {V_{1}(n) = \frac{1.299435}{n_{1}} + \frac{4.321858}{n_{2}} + \frac{2.419448}{n_{3}} + \frac{8.603727}{n_{4}} + \frac{3.192979}{n_{5}}} \\ {V_{2}(n) = \frac{37.31234}{n_{1}} + \frac{2.70621}{n_{2}} + \frac{24.33143}{n_{3}} + \frac{33.95098}{n_{4}} + \frac{3.543856}{n_{5}}} \\ {V_{3}(n) = \frac{0.742534}{n_{1}} + \frac{1.413692}{n_{2}} + \frac{0.867349}{n_{3}} + \frac{1.391779}{n_{4}} + \frac{0.91228}{n_{5}}} \\ {4n_{1} + 4.5n_{2} + 5n_{3} + 5.5n_{4} + 6n_{5} + 7\sqrt{n_{1}} + 9\sqrt{n_{2}} + 10\sqrt{n_{3}} + 12\sqrt{n_{4}} + 13\sqrt{n_{5}} \leq 1000} \\ {\mu \leq \mu_{j}(V_{j}(n)),\upsilon \geq \upsilon_{j}(V_{j}(n)),\mu_{j}(V_{j}(n)) + \upsilon_{j}(V_{j}(n)) \leq 1} \\ {\mu_{j}(V_{j}(n)) \in \lbrack 0,1\rbrack,\upsilon_{j}(V_{j}(n)) \in \lbrack 0,1\rbrack,\forall j = 1,2,3;} \\ {2 \leq n_{h} \leq N_{h},2 \leq r_{h} \leq {\hat{n}}_{h2};n_{h},r_{h}\ {are}\ {integers};\ h = 1,2,3,4,5.} \\ \end{array}$$ We can obtain a compromise result for the integer NLP problem by utilizing the Optimistic Approach of Intuitionistic Programming. As $$\begin{array}{l} {n_{1} = 38,n_{2} = 23,n_{3} = 29,n_{4} = 40,n_{5} = 18,\mu = 0.8205657,\upsilon = 0.1313135} \\ {V_{1} = 0.6980127,V_{2} = 2.984235,V_{3} = 0.1963905,\ {Required}\ {Cost} = 999.714,\ {Trace} = 3.878638} \\ \end{array}$$ **Pessimistic approach (*λ* = 0):** The non-MFs of *V*_*j*(*n*)∀*j* = 1,2,3 are as follows: $$\upsilon_{1}(V_{1}(n)) = \left\{ \begin{array}{l} {0,\mspace{166mu} V_{1}(n) \leq 0.7323559} \\ {1 + \frac{\lbrack V_{1}(n) - 1.0323559\rbrack}{0.3},\mspace{22mu} 0.7323559 < V_{1}(n) \leq 1.0323559} \\ {1,\mspace{166mu} V_{1}(n) > 1.0323559} \\ \end{array} \right.$$ $$\upsilon_{2}(V_{2}(n)) = \left\{ \begin{array}{l} {0,\mspace{166mu} V_{2}(n) \leq 2.992765} \\ {1 + \frac{\lbrack V_{2}(n) - 4.042765\rbrack}{1.05},\mspace{27mu} 2.992765 < V_{2}(n) \leq 4.042765} \\ {1,\mspace{166mu} V_{2}(n) > 4.042765} \\ \end{array} \right.$$ $$\upsilon_{3}(V_{3}(n)) = \left\{ \begin{array}{l} {0,\mspace{166mu} V_{3}(n) \leq 0.2002541} \\ {1 + \frac{\lbrack V_{3}(n) - 0.2702541\rbrack}{0.07},\mspace{22mu} 0.2002541 < V_{3}(n) \leq 0.2702541} \\ {1,\mspace{166mu} V_{3}(n) > 0.2702541} \\ \end{array} \right.$$ Then, the pessimistic model of the response multivariate stratified sampling is as follows: $$\begin{array}{l} {{Max}\ (\mu - \upsilon)} \\ {{subject}\ {to}} \\ {\mu_{1}(V_{1}(n)) = 1 + \frac{\lbrack 0.6323559 - V_{1}(n)\rbrack}{0.4},\mu_{2}(V_{2}(n)) = 1 + \frac{\lbrack 2.752765 - V_{2}(n)\rbrack}{1.29}} \\ {\upsilon_{1}(V_{1}(n)) = 1 + \frac{\lbrack V_{1}(n) - 1.0323559\rbrack}{0.3},\upsilon_{2}(V_{2}(n)) = 1 + \frac{\lbrack V_{2}(n) - 4.042765\rbrack}{1.05}} \\ {\mu_{3}(V_{3}(n)) = 1 + \frac{\lbrack 0.1802541 - V_{3}(n)\rbrack}{0.09},\upsilon_{3}(V_{3}(n)) = 1 + \frac{\lbrack V_{3}(n) - 0.2702541\rbrack}{0.07}} \\ {V_{1}(n) = \frac{1.299435}{n_{1}} + \frac{4.321858}{n_{2}} + \frac{2.419448}{n_{3}} + \frac{8.603727}{n_{4}} + \frac{3.192979}{n_{5}}} \\ {V_{2}(n) = \frac{37.31234}{n_{1}} + \frac{2.70621}{n_{2}} + \frac{24.33143}{n_{3}} + \frac{33.95098}{n_{4}} + \frac{3.543856}{n_{5}}} \\ {V_{3}(n) = \frac{0.742534}{n_{1}} + \frac{1.413692}{n_{2}} + \frac{0.867349}{n_{3}} + \frac{1.391779}{n_{4}} + \frac{0.91228}{n_{5}}} \\ {4n_{1} + 4.5n_{2} + 5n_{3} + 5.5n_{4} + 6n_{5} + 7\sqrt{n_{1}} + 9\sqrt{n_{2}} + 10\sqrt{n_{3}} + 12\sqrt{n_{4}} + 13\sqrt{n_{5}} \leq 1000} \\ {\mu \leq \mu_{j}(V_{j}(n)),\upsilon \geq \upsilon_{j}(V_{j}(n)),\mu_{j}(V_{j}(n)) + \upsilon_{j}(V_{j}(n)) \leq 1} \\ {\mu_{j}(V_{j}(n)) \in \lbrack 0,1\rbrack,\upsilon_{j}(V_{j}(n)) \in \lbrack 0,1\rbrack,\forall j = 1,2,3;} \\ {2 \leq r_{h} \leq {\hat{n}}_{h2},2 \leq n_{h} \leq N_{h};n_{h},r_{h}\ {are}\ {integers};\ h = 1,2,3,4,5} \\ \end{array}$$ We can devise a compromise solution for the integer NLP problem by adopting a pessimistic approach to intuitionistic programming. As $$\begin{array}{l} {n_{1} = 37,n_{2} = 23,n_{3} = 32,n_{4} = 38,n_{5} = 18,\mu = 0.8247996,\upsilon = 0} \\ {V_{1} = 0.7024361,V_{2} = 2.976788,V_{3} = 0.1959460,\ {Cost} = 999.9377,\ {Trace}\ {value} = 3.875170} \\ \end{array}$$ ## Multivariate stratified sampling- a case of nonresponse **Defination 2:** Stratified sampling is a type of nonresponse case sampling technique in which the population is divided into subgroups (strata) based on one or more characteristics, and a sample is taken from each stratum. The goal of stratified sampling is to ensure that each stratum is represented in the sample in proportion to its representation in the population, even if some members of the population do not respond to the survey. The sample size $n = {\sum_{h = 1}^{L}n_{h}}$, from the *n*_*h* units; the respondent’s group belongs to *n*_*h*1 units, and the non- respondent group belongs to *n*_*h*2 = *n*_*h*−*n*_*h*1 units. With subsamples of sizes made for nonresponse problems as: $$r_{h} = {n_{h2}/k_{h}}\forall k_{h} \geq 1,h = 1,2,\ldots,L.$$ It was taken from the non-respondent’s collection, the sampling fraction 1/*k*_*h* between the *h*^th stratum of the non-respondents. The unbiased estimates of *N*_*h*1 and *N*_*h*2 are ${\hat{N}}_{h1} = n_{h1}{N_{h}/n_{h}}$ and, ${\hat{N}}_{h2} = n_{h2}{N_{h}/n_{h}}$ respectively. Similarly, the multivariate sampling problem in the case of nonresponse can be formulated as a multiobjective nonlinear multivariate stratified sampling problem, as follows: $$\begin{array}{l} {{Minimize}\mspace{22mu} V_{j}(n) = {\sum\limits_{h = 1}^{L}\frac{W_{h}^{2}\left( {S_{jh}^{2} - W_{h2}S_{jh2}^{2}} \right)}{n_{h}}} + {\sum\limits_{h = 1}^{L}{\frac{W_{h}^{2}W_{h2}^{2}S_{jh2}^{2}}{r_{h}} - {\sum\limits_{h = 1}^{L}\frac{W_{h}^{2}S_{jh}^{2}}{N_{h}^{2}}};}}\forall j = 1,2,\ldots,p} \\ {{subject}\ {to}\mspace{22mu}{\sum\limits_{h = 1}^{L}{\left( {c_{h0} + c_{h1}W_{h1}} \right)n_{h} + {\sum\limits_{h = 1}^{L}{c_{h2}r_{h} + \omega_{1}{\sum\limits_{h = 1}^{L}{\frac{n_{h}}{\lambda_{1}} + \omega_{2}{\sum\limits_{h = 1}^{L}{\frac{r_{h}}{\lambda_{2}} \leq}}}}}}}}B_{0}} \\ {\mspace{104mu} 2 \leq r_{h} \leq {\hat{n}}_{h2},2 \leq n_{h} \leq N_{h}\& n_{h},r_{h} \in {integers};\ h = 1,2,\ldots,L.} \\ \end{array}$$ **Numerical Example 2:** With a total budget of 10,000, five strata and three data characteristics were chosen from Sukhatme et al.. Then, the compromise allocation of a multiobjective NLP problem with a quadratic cost function using the data in is as follows: $$\begin{array}{l} {{Minimize}\mspace{9mu} V_{1}(n) = \frac{0.4388203}{n_{1}} + \frac{2.663113}{n_{2}} + \frac{49.60277}{n_{3}} + \frac{2.66616}{n_{4}} + \frac{9.938173}{n_{5}} + \frac{0.002437891}{r_{1}}} \\ {\mspace{144mu} + \frac{0.08937845}{r_{2}} + \frac{1.206554}{r_{3}} + \frac{0.044436}{r_{4}} + \frac{0.2094497}{r_{5}}} \\ {{Minimize}\mspace{9mu} V_{2}(n) = \frac{2.047828}{n_{1}} + \frac{70.97196}{n_{2}} + \frac{25.1802}{n_{3}} + \frac{11.2677}{n_{4}} + \frac{2.736015}{n_{5}} + \frac{0.03413047}{r_{1}}} \\ {\mspace{144mu} + \frac{2.381936}{r_{2}} + \frac{0.6124914}{r_{3}} + \frac{0.187795}{r_{4}} + \frac{0.05766226}{r_{5}}} \\ {{Minimize}\mspace{9mu} V_{3}(n) = \frac{1.510273}{n_{1}} + \frac{0.7689739}{n_{2}} + \frac{0.4857774}{n_{3}} + \frac{0.36501}{n_{4}} + \frac{1.561138}{n_{5}} + \frac{0.02517122}{r_{1}}} \\ {\mspace{144mu} + \frac{0.02580803}{r_{2}} + \frac{0.01181621}{r_{3}} + \frac{0.0060835}{r_{4}} + \frac{0.03290141}{r_{5}}} \\ {{subject}\ {to}\quad 4n_{1} + 4.9n_{2} + 5.9n_{3} + 7.75n_{4} + 8.92n_{5} + 6r_{1} + 7r_{2} + 9r_{3} + 11r_{4} + 12r_{5}} \\ {\mspace{90mu} + 100{\sum_{h = 1}^{5}\frac{n_{h}}{4}} + 100{\sum_{h = 1}^{5}\frac{r_{h}}{4}} \leq 10000} \\ {\mspace{90mu} 2 \leq r_{h} \leq {\hat{n}}_{h2},2 \leq n_{h} \leq N_{h};n_{h},r_{h}\ {are}\ {integers};\ h = 1,2,3,4,5.} \\ \end{array}$$ The MFs for the objective functions under the IF sets are as follows: $$\mu_{1}(V_{1}(n)) = \left\{ \begin{array}{l} {1,\mspace{166mu} V_{j}(n) \leq 0.8441696} \\ {1 + \frac{\lbrack 0.8441696 - V_{1}(n)\rbrack}{0.8464674}\mspace{27mu} 0.8441696 < V_{1}(n) \leq 1.690637} \\ {0,\mspace{166mu} V_{1}(n) > 1.690637} \\ \end{array} \right.$$ $$\mu_{2}(V_{2}(n)) = \left\{ \begin{array}{l} {1,\mspace{166mu} V_{2}(n) \leq 1.663423} \\ {1 + \frac{\lbrack 1.663423 - V_{2}(n)\rbrack}{1.804234}\mspace{32mu} 1.663423 < V_{2}(n) \leq 3.467657} \\ {0,\mspace{166mu} V_{2}(n) > 3.467657} \\ \end{array} \right.$$ $$\mu_{3}(V_{3}(n)) = \left\{ \begin{array}{l} {1,\mspace{166mu} V_{3}(n) \leq 0.0917367} \\ {1 + \frac{\lbrack 0.0917367 - V_{3}(n)\rbrack}{0.1124883}\mspace{27mu} 0.0917367 < V_{3}(n) \leq 0.2042250} \\ {0,\mspace{166mu} V_{3}(n) > 0.2042250} \\ \end{array} \right.$$ **Optimistic approach (*λ* = 1):** The non-MFs of *V*_*j*(*n*)∀*j* = 1,2,3 are as follows: $$\upsilon_{1}(V_{1}(n)) = \left\{ \begin{array}{l} {0,\mspace{166mu} V_{1}(n) \leq 0.8441696} \\ {1 + \frac{\lbrack V_{1}(n) - 2.28128374\rbrack}{1.4389348},\quad 0.8441696 < V_{1}(n) \leq 2.28128374} \\ {1,\mspace{166mu} V_{1}(n) > 2.28128374} \\ \end{array} \right.$$ $$\upsilon_{2}(V_{2}(n)) = \left\{ \begin{array}{l} {0,\mspace{166mu} V_{2}(n) \leq 1.663423} \\ {1 + \frac{\lbrack V_{2}(n) - 4.647891\rbrack}{2.984468},\mspace{27mu} 1.663423 < V_{2}(n) \leq 4.647891} \\ {1,\mspace{166mu} V_{2}(n) > 4.647891} \\ \end{array} \right.$$ $$\upsilon_{3}(V_{3}(n)) = \left\{ \begin{array}{l} {0,\mspace{166mu} V_{3}(n) \leq 0.0917367} \\ {1 + \frac{\lbrack V_{3}(n) - 0.29577133\rbrack}{0.2039766},\quad 0.0917367 < V_{3}(n) \leq 0.29577133} \\ {1,\mspace{166mu} V_{3}(n) > 0.29577133} \\ \end{array} \right.$$ Then, the optimistic model of the nonresponse multivariate stratified sampling is as follows: $$\begin{array}{l} {{Max}\ (\mu - \upsilon)} \\ {{subject}\ {to}} \\ {\mu_{1}(V_{1}(n)) = 1 + \frac{\lbrack 0.8441696 - V_{1}(n)\rbrack}{0.8464674},\mu_{2}(V_{2}(n)) = 1 + \frac{\lbrack 1.663423 - V_{2}(n)\rbrack}{1.804234}} \\ {\upsilon_{1}(V_{1}(n)) = 1 + \frac{\lbrack V_{1}(n) - 2.28128374\rbrack}{1.4389348},\upsilon_{2}(V_{2}(n)) = 1 + \frac{\lbrack V_{2}(n) - 4.647891\rbrack}{2.984468}} \\ {\mu_{3}(V_{3}(n)) = 1 + \frac{\lbrack 0.0917367 - V_{3}(n)\rbrack}{0.1124883},\upsilon_{3}(V_{3}(n)) = 1 + \frac{\lbrack V_{3}(n) - 0.29577133\rbrack}{0.2039766}} \\ {V_{1}(n) = \frac{0.4388203}{n_{1}} + \frac{2.663113}{n_{2}} + \frac{49.60277}{n_{3}} + \frac{2.66616}{n_{4}} + \frac{9.938173}{n_{5}} + \frac{0.002437891}{r_{1}}} \\ {\mspace{90mu} + \frac{0.08937845}{r_{2}} + \frac{1.206554}{r_{3}} + \frac{0.044436}{r_{4}} + \frac{0.2094497}{r_{5}}} \\ {V_{2}(n) = \frac{2.047828}{n_{1}} + \frac{70.97196}{n_{2}} + \frac{25.1802}{n_{3}} + \frac{11.2677}{n_{4}} + \frac{2.736015}{n_{5}} + \frac{0.03413047}{r_{1}}} \\ {\mspace{90mu} + \frac{2.381936}{r_{2}} + \frac{0.6124914}{r_{3}} + \frac{0.187795}{r_{4}} + \frac{0.05766226}{r_{5}}} \\ {V_{3}(n) = \frac{1.510273}{n_{1}} + \frac{0.7689739}{n_{2}} + \frac{0.4857774}{n_{3}} + \frac{0.36501}{n_{4}} + \frac{1.561138}{n_{5}} + \frac{0.02517122}{r_{1}}} \\ {\mspace{90mu} + \frac{0.02580803}{r_{2}} + \frac{0.01181621}{r_{3}} + \frac{0.0060835}{r_{4}} + \frac{0.03290141}{r_{5}}} \\ {4n_{1} + 4.9n_{2} + 5.9n_{3} + 7.75n_{4} + 8.92n_{5} + 6r_{1} + 7r_{2} + 9r_{3} + 11r_{4} + 12r_{5}} \\ {+ 100{\sum_{h = 1}^{5}\frac{n_{h}}{4}} + 100{\sum_{h = 1}^{5}\frac{r_{h}}{4}} \leq 10000} \\ {\upsilon \geq \upsilon_{j}(V_{j}(n)),\mu \leq \mu_{j}(V_{j}(n)),\mu_{j}(V_{j}(n)) + \upsilon_{j}(V_{j}(n)) \leq 1} \\ {\mu_{j}(V_{j}(n)) \in \lbrack 0,1\rbrack,\upsilon_{j}(V_{j}(n)) \in \lbrack 0,1\rbrack,\forall j = 1,2,3;} \\ {2 \leq r_{h} \leq {\hat{n}}_{h2},2 \leq n_{h} \leq N_{h}n_{h},r_{h}\ {are}\ {integers};\ h = 1,2,3,4,5.} \\ \end{array}$$ We can devise a compromise solution for integer NLP problems using the optimistic approach of intuitionistic programming. <img src="info:doi/10.1371/journal.pone.0284784.e085" id="pone.0284784.e085g" /> n 1 = 35 , n 2 = 69 , n 3 = 88 , n 4 = 33 , n 5 = 49 , r 1 = 5 , r 2 = 13 , r 3 = 14 , r 4 = 4 , r 5 = 7 V 1 = 1.032989 , V 2 = 2.059496 , V 3 = 0.1168203 , Cost = 9990.13 , Trace value = 3.209306 **Pessimistic approach (*λ* = 0):** The non-MFs of *V*_*j*(*n*)∀*j* = 1,2,3 are as follows: $$\upsilon_{1}(V_{1}(n)) = \left\{ \begin{array}{l} {0,\mspace{166mu} V_{1}(n) \leq 1.0142696} \\ {1 + \frac{\lbrack V_{1}(n) - 1.690637\rbrack}{0.6763674},\mspace{27mu} 1.0142696 < V_{1}(n) \leq 1.690637} \\ {1,\mspace{166mu} V_{1}(n) > 1.690637} \\ \end{array} \right.$$ $$\upsilon_{2}(V_{2}(n)) = \left\{ \begin{array}{l} {0,\mspace{166mu} V_{2}(n) \leq 2.054523} \\ {1 + \frac{\lbrack V_{2}(n) - 3.467657\rbrack}{1.413134},\mspace{27mu} 2.054523 < V_{2}(n) \leq 3.467657} \\ {1,\mspace{166mu} V_{2}(n) > 3.467657} \\ \end{array} \right.$$ $$\upsilon_{3}(V_{3}(n)) = \left\{ \begin{array}{l} {0,\mspace{166mu} V_{3}(n) \leq 0.1010367} \\ {1 + \frac{\lbrack V_{3}(n) - 0.2042250\rbrack}{0.1031883},\mspace{22mu} 0.1010367 < V_{3}(n) \leq 0.204225} \\ {1,\mspace{166mu} V_{3}(n) > 0.204225} \\ \end{array} \right.$$ Then, the pessimistic model of the nonresponse multivariate stratified sampling is as follows: $$\begin{array}{l} {{Max}\ (\mu - \upsilon)} \\ {{subject}\ {to}} \\ {\mu_{1}(V_{1}(n)) = 1 + \frac{\lbrack 0.8441696 - V_{1}(n)\rbrack}{0.8464674},\mu_{2}(V_{2}(n)) = 1 + \frac{\lbrack 1.663423 - V_{2}(n)\rbrack}{1.804234}} \\ {\upsilon_{1}(V_{1}(n)) = 1 + \frac{\lbrack V_{1}(n) - 1.690637\rbrack}{0.6763674},\upsilon_{2}(V_{2}(n)) = 1 + \frac{\lbrack V_{2}(n) - 3.467657\rbrack}{1.413134}} \\ {\mu_{3}(V_{3}(n)) = 1 + \frac{\lbrack 0.0917367 - V_{3}(n)\rbrack}{0.1124883},\upsilon_{3}(V_{3}(n)) = 1 + \frac{\lbrack V_{3}(n) - 0.2042250\rbrack}{0.1031883}} \\ {V_{1}(n) = \frac{0.4388203}{n_{1}} + \frac{2.663113}{n_{2}} + \frac{49.60277}{n_{3}} + \frac{2.66616}{n_{4}} + \frac{9.938173}{n_{5}} + \frac{0.002437891}{r_{1}}} \\ {\mspace{90mu} + \frac{0.08937845}{r_{2}} + \frac{1.206554}{r_{3}} + \frac{0.044436}{r_{4}} + \frac{0.2094497}{r_{5}}} \\ {V_{2}(n) = \frac{2.047828}{n_{1}} + \frac{70.97196}{n_{2}} + \frac{25.1802}{n_{3}} + \frac{11.2677}{n_{4}} + \frac{2.736015}{n_{5}} + \frac{0.03413047}{r_{1}}} \\ {\mspace{90mu} + \frac{2.381936}{r_{2}} + \frac{0.6124914}{r_{3}} + \frac{0.187795}{r_{4}} + \frac{0.05766226}{r_{5}}} \\ {V_{3}(n) = \frac{1.510273}{n_{1}} + \frac{0.7689739}{n_{2}} + \frac{0.4857774}{n_{3}} + \frac{0.36501}{n_{4}} + \frac{1.561138}{n_{5}} + \frac{0.02517122}{r_{1}}} \\ {\mspace{90mu} + \frac{0.02580803}{r_{2}} + \frac{0.01181621}{r_{3}} + \frac{0.0060835}{r_{4}} + \frac{0.03290141}{r_{5}}} \\ {4n_{1} + 4.9n_{2} + 5.9n_{3} + 7.75n_{4} + 8.92n_{5} + 6r_{1} + 7r_{2} + 9r_{3} + 11r_{4} + 12r_{5}} \\ {+ 100{\sum_{h = 1}^{5}\frac{n_{h}}{4}} + 100{\sum_{h = 1}^{5}\frac{r_{h}}{4}} \leq 10000} \\ {\upsilon \geq \upsilon_{j}(V_{j}(n)),\mu \leq \mu_{j}(V_{j}(n)),\mu_{j}(V_{j}(n)) + \upsilon_{j}(V_{j}(n)) \leq 1} \\ {\upsilon_{j}(V_{j}(n)) \in \lbrack 0,1\rbrack,\mu_{j}(V_{j}(n)) \in \lbrack 0,1\rbrack,\forall j = 1,2,3;} \\ {2 \leq r_{h} \leq {\hat{n}}_{h2},2 \leq n_{h} \leq N_{h}\& n_{h},r_{h} \in {integers}\ \forall h = 1,2,3,4,5.} \\ \end{array}$$ We can devise a compromise solution for the integer NLP problem by adopting a pessimistic approach to intuitionistic programming. <img src="info:doi/10.1371/journal.pone.0284784.e090" id="pone.0284784.e090g" /> n 1 = 44 , n 2 = 66 , n 3 = 86 , n 4 = 33 , n 5 = 49 , r 1 = 5 , r 2 = 12 , r 3 = 12 , r 4 = 4 , r 5 = 7 V 1 = 1.060225 , V 2 = 2.123498 , V 3 = 0.1089356 , Cost = 9999.63 , Trace value = 3.292658 # Conclusion This paper describes the integer NLP problem discussed by the optimistic and pessimistic methods of intuitionistic fuzzy programming and solved by Lingo 18 software. Tables and compare the practical utility of the proposed technique. shows the trace values obtained for the different allocations with respect to the total survey cost. It is observed that the model based on the pessimistic approach determines the minimum trace value compared with the model based on the optimistic approach allocation for the response case of multivariate SRSS, and the total cost is almost exhausted for this approach. Cochran’s allocation, the most commonly used allocation, is also used for comparative studies. It is observed that Cochran’s allocation gives the maximum value for trace and violates the cost constraint; hence, it gives an infeasible allocation, for example, 1 in multivariate case. demonstrates the allocation’s trace value and survey cost, with the optimistic approach model determining the minimum trace value, among others. The survey cost is fully exhausted by using allocation by a pessimistic approach for the nonresponse case of multivariate SRSS. For this example, Cochran’s allocation was also calculated and used for possible comparisons. The suggested technique works better for this example than the most commonly used Cochran allocation. Based on these examples, it can be concluded that the optimistic approach model is appropriate for determining the best compromise allocation for a multivariate SRSS. The study could be useful in marketing surveys of any new launch product and help the company how the company can improve the better quality and attractive features of their product so that they remain in the market competition. The excellent performance and underperformance products could be identified with the help of conducting such marketing surveys. In wildlife and agricultural-related surveys, the study could be helpful. National Planning policies related to surveys in this study could also be helpful. We would like to express our sincere thanks to all the reviewers for their valuable comments and suggestions that significantly enhanced the quality of the article. We also thank Editor-in-Chief for his great support throughout this review and publication process. 10.1371/journal.pone.0284784.r001 Decision Letter 0 Ghadiri Nejad Mazyar Academic Editor 2023 Mazyar Ghadiri Nejad This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 7 Feb 2023 PONE-D-23-01575Multi-objective Intuitionistic Fuzzy Programming under Pessimistic and Optimistic Applications in Multivariate Stratified Sample Allocation ProblemsPLOS ONE Dear Dr. Haq, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Mar 24 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Mazyar Ghadiri Nejad, Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please note that PLOS ONE has specific guidelines on code sharing for submissions in which author-generated code underpins the findings in the manuscript. In these cases, all author-generated code must be made available without restrictions upon publication of the work. Please review our guidelines at <https://journals.plos.org/plosone/s/materials-and-software-sharing#loc- sharing-code> and ensure that your code is shared in a way that follows best practice and facilitates reproducibility and reuse. 3\. Thank you for stating the following in your Competing Interests section:   "NO authors have competing interests: The authors have declared that no competing interests exist." Please complete your Competing Interests on the online submission form to state any Competing Interests. If you have no competing interests, please state "The authors have declared that no competing interests exist.", as detailed online in our guide for authors at [http://journals.plos.org/plosone/s/submit- now ](http://journals.plos.org/plosone/s/submit-now )  This information should be included in your cover letter; we will change the online submission form on your behalf. 4\. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. "Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. ​ Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: N/A \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors could provide good work. However, there are some concerns to be resolved. The abstract is loosely written. It is not as informative as expected. A standard abstract must present, without leaving any doubt, the objective of the paper precisely; source of data (which is not present in your abstract) and analytical approach used; key findings and any policy implication and recommendations The structure of the introduction is not in a standard form. first, we need to find the basics of your problems and then the main aspects and issues related to them. Do not pay attention to lots of details (especially the foundation of the problem) English presentation of the paper should be modified so as to be more readable. I suggest the authors read and consider the studies performed by Alireza Goli et al. and their groups on the application of metaheuristic algorithms and supply chain optimization. Authors should improve the material selection criteria and methods section considering the latest published work in reputed journal. The implications of the results should be discussed in more detail. The authors should provide managerial insights based on the output. Reviewer \#2: The paper deals with the challenge of allocating resources for a multivariate stratified sample survey. The goal is to estimate population means for a fixed cost and this problem is formulated as a mathematical programming problem. The paper presents solution procedures using fuzzy programming optimistic and pessimistic methods. The paper relevance and subject is interesting. However, it can be improved further. 1\. The writing of the paper needs to be reconsidered. In many cases the sentences are either grammatically incorrect or simply incomplete. For instance, in the abstract L3, the word “respectively” is used incorrectly. Or in the second sentence of the introduction, I guess “When” needs to be replace with “while” and also the last part of the sentence “helps to keep the variance of the stratified sample” is incomplete. 2\. What is “s” in the notation 3\. I suggest make this sentence as some bullet points, providing definitions for each symbol: “The hth stratum, and denote stratum size, the population mean, weight, and population mean square for the hth stratum”. 4\. Although the definitions and formulations in the introductions are known, proper references to the original studies should be provided. 5\. The literature review provide no story and background to the current work. What is it that the authors find defective and how the use of fuzzy programming will address the shortcoming in the previous studies? I suggest reorganizing the background review in a way that the reader in informed about the gap and then authors need to clearly justify their proposal and how it will add value to the field before proceeding to methodology. 6\. As mentioned earlier, the methodology needs to be rewritten in a clear way. From the first sentence “The multiobjective multivariate stratified sampling (MOMSS) model where the variances form the objectives and cost work as a constraint.” Which I find difficult to grasp what the authors are trying to deliver. Later authors define aspiration level, reservation levels … which are not used in the following models. It is not also clear how the preference information is elicited from the DM? a priori? Is it possible for the DM to provide such information before having any options or knowledge on the possibilities? Given the interesting nature of the topic but bad flow of the paper, I would suggest authors to be given a chance to make major revisions to make the paper ready for further review and consideration. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0284784.r002 Author response to Decision Letter 0 11 Mar 2023 Dear Editor We thank you for your time and efforts on our manuscript. Further, we take this opportunity to thank the anonymous reviewers for their valuable suggestions. The suggested major revisions have been efficiently carried out in the revised version of the manuscript. All the changes have been highlighted in yellow. Thanking you 10.1371/journal.pone.0284784.r003 Decision Letter 1 Ghadiri Nejad Mazyar Academic Editor 2023 Mazyar Ghadiri Nejad This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 10 Apr 2023 Multiobjective Intuitionistic Fuzzy Programming under Pessimistic and Optimistic Applications in Multivariate Stratified Sample Allocation Problems PONE-D-23-01575R1 Dear Ahteshamul Hag, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Mazyar Ghadiri Nejad, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Dear respected Authors, Considering the opinion of the respected reviewers, the manuscript is accepted to be published in the respected journal of PLOS ONE. Thanks for your efforts and hope to see another submission from you to this journal. Regards, Mazyar Ghadiri Nejad Reviewers' comments: **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#3: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#3: No \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Authors have improved the paper well based on my prevously- mentioned comments, So my recommendation is ACCEPT Reviewer \#3: (No Response) \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#3: No \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0284784.r004 Acceptance letter Ghadiri Nejad Mazyar Academic Editor 2023 Mazyar Ghadiri Nejad This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 13 Apr 2023 PONE-D-23-01575R1 Multiobjective intuitionistic fuzzy programming under pessimistic and optimistic applications in multivariate stratified sample allocation problems Dear Dr. Haq: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Assoc. Prof. Dr. Mazyar Ghadiri Nejad Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist. [^2]: Current address: Department of Mathematics & Statistics, Integral University, Lucknow, India

# Introduction *Campylobacter jejuni* is a Gram-negative spiral shaped bacterium, which is pervasive in mammals and birds. The chicken intestine is the natural reservoir that is frequently colonized by the pathogen. In recent years, *C. jejuni* has emerged as an abundant reported cause of bacterial diarrhoea in industrialized as well as developing countries with approximately 2.5 million estimated cases per year in the United States and more than 60,000 annual cases in Germany. Contaminated chicken meat, beef and milk products are common sources of transmission and human infection. The progress of the disease can vary from watery to bloody diarrhea including fever and abdominal cramps. In rare cases immunopathological sequelae such as Guillain-Barré syndrome might arise even months or years post infection. Adherence of *C. jejuni* to intestinal epithelial cells is essential for successful infection in human hosts. In the past, *in vitro* as well as *in vivo* studies revealed distinct proteins which are important for *C. jejuni* to adhere to its particular target cells. For instance, PEB1, the periplasmic component of an aspartate/glutamate ABC-transporter, mediates adherence and invasion of human epithelial cells and is important for the intestinal colonization of mice. Furthermore, the major outer membrane protein MOMP adheres to fibronectin and is involved in binding of *C. jejuni* to the membrane of INT407 cells. CadF, another outer membrane protein with an apparent molecular weight of 37 kDa, and FlpA have also been shown to connect to fibronectin. These interactions, in turn, result in the activation of integrin receptors to launch a host cell signal cascade leading to restructuring of the actin cytoskeleton mediating the uptake of *C. jejuni*. The surface-exposed lipoprotein JlpA is required for efficient adherence of the pathogen to HEp-2 epithelial cells and initiates the activation of NF-κB and p38 MAP kinase. Consequently, JlpA seems to be involved in the proinflammatory host cell response upon *C. jejuni* infection. Furthermore, CapA, an autotransporter protein of *C. jejuni,* has been shown to be associated with the adherence and invasion of epithelial cells by the pathogen and plays an important role in the colonization of the chicken gut. Recently, it was demonstrated that the bacterial outer membrane protein Cj0091 mediates adherence of *C. jejuni* to INT407 cells and contributes to the colonization of chickens as well. Taken together, all these adhesion factors contribute significantly to the interaction between host cell and bacterial pathogen to allow the subsequent process of cellular invasion. In addition to these proteins described to be involved in the process of adherence, also lipooligosaccharides (LOS) are important for pathogen-host cell interactions given that *C. jejuni* strains deficient in genes involved in LOS metabolism (*wlaRG, wlaTB* and *wlaTC*) exhibited diminished adherence properties to chicken embryo fibroblasts. Recently, our group generated a *C. jejuni* mutant deficient in sulphite:cytochrome c oxidoreductase (SOR) which exhibited a down- regulated transcription of genes involved legionaminic acid synthesis and possessed reduced adherence properties to Caco2 cells. Finally, a recently identified *C. jejuni* type VI secretion system (T6SS) was shown to be involved in cell adhesion. Following functional knockout of the T6SS-genes *hcp1* and *icmF1*, the adherence capacity of the respective mutants to T84 colon epithelial cells was reduced by approximately 50% as compared to the parental strain. In this report we describe the *C. jejuni* gene *cj0268c* which has been shown by us and others to be important for the capability of the pathogen to infect host cells. This protein with a molecular weight of 40.2 kDa and an isoelectric point of 8.93 possesses a putative transmembrane domain around amino acid 60 and a SPFH domain encompassing the amino acids 64 to 259. Proteins containing the stomatin/prohibitin/flotillin/HflK/C (SPFH) domain can be found in divergent species ranging from bacteria to mammals. The precise function of this domain, however, is still unclear even though mammal proteins containing SPFH domains are frequently found in lipid raft microdomains within several cellular membranes. In support of this, Hinderhofer *et al*. could categorize altogether 1090 SPFH domain-containing proteins from 497 different bacterial species encompassing all phyla into 12 subfamilies. However, despite the knowledge acquisition about the evolutionary development of prokaryotic SPFH proteins, the general biological function of this motif is still unclear. Here we demonstrate that Cj0268c is required for adhesion of *C. jejuni* to different host cells. Heterologous expression revealed its potential to alter the adhesion capacity of *E. coli*. The exact subcellular localization of Cj0268c was not known yet. Although analysis of the protein sequence using SignalIP 4.1 (Technical University of Denmark) revealed no evidence for the presence of a signal peptide, flow cytometry analysis after immunolabeling of Cj0268c in *E. coli* indicated the protein to reside in the periplasmic space with no exposure of the C-terminus at the bacterial surface. Furthermore, we determine the relevance of Cj0268c regarding motility, autoagglutination, the resistance of *C. jejuni* to bile salts and the stability of the bacterial cell to the nonionic surfactant Triton X-100. Finally, we investigate the affiliation of *cj0268c* to particular clonal groups of *C. jejuni*. # Materials and Methods ## Bacterial strains, media and culture conditions *C. jejuni* strain B2 initially isolated from a patient suffering from gastroenteritis, and strain NCTC 11168 were grown on Columbia agar supplemented with 5% defibrinated sheep blood at 42°C in microaero- and capnophilic conditions (85% N₂, 10% CO₂, 5% O₂). If required, appropriate antibiotic concentrations of kanamycin (50 µg ml⁻¹) or chloramphenicol (30 µg ml⁻¹) were added. Growth experiments were carried out at 42°C in Mueller-Hinton (MH) broth under microaero- and capnophilic conditions. *Escherichia coli* strain DH5α was grown on Luria bertani (LB) agar or broth at 37°C. When necessary, ampicillin (100 µg ml⁻¹) was added. Growth experiments were carried out at 42°C in Mueller-Hinton (MH) broth under the above mentioned conditions for 24 h. ## Generation of competent cells and electroporation 10 ml of LB-broth were inoculated with a single *E. coli* DH5α colony and incubated overnight at 37°C under shaking. Three ml of the overnight culture were grown in 100 ml LB-broth at 37°C to an OD_{(600 nm)} of 0.35–0.45. The culture was transferred into a 50 ml Falcon tube, placed on ice for 10 min and centrifuged for 15 min at 4000×g at 4°C. Then the cell pellet was gently resuspended in 30 ml ice cold TFB1 buffer and incubated on ice for 30 min. Cells were pelleted by centrifugation as described above. Then, the cell pellet was carefully dissolved in 2 ml ice-cold TFB2 buffer and incubated on ice for another 30 min. After incubation aliquots of 100 µl were stored at −80°C. *C. jejuni,* cells were harvested from Columbia blood agar plates and centrifuged at 5,000×g at 4°C for 10 minutes. After washing of the *C. jejuni* cells three times in 1 ml ice-cold wash buffer containing 272 mM sucrose and 15% glycerol at 4°C, the pellet was resuspended in 400 µl washing buffer and 100 µl aliquots were used for electroporation, respectively. For each transformation 0.5 to 3 µg of plasmid DNA were added to an aliquot of competent cells in an ice-cold electroporation cuvette. After incubation of the cuvette containing the mixture of bacteria and DNA, electroporation was performed at 2.5 kV, 25 µF and 200 Ω using the BTX Electro Cell Manipulator. While adding 500 µl of SOC medium to transform *E. coli*, in case of *C. jejuni*, the suspension was transferred onto a non selective Columbia blood agar plate and incubated overnight at 37°C under microaerophilic conditions. Finally, cells were transferred onto a selective plate and incubated at 42°C under microaerophilic conditions for additional 2–3 days. *E. coli* were plated directly on selective LB agar containing the appropriate antibiotic agent and incubated at 37°C overnight. ## Cultivation of cells Human colon carcinoma Caco2 cells were cultivated in Dulbecco minimal essential medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 1 x non-essential amino acids, 100 U ml⁻¹ penicillin, and 100 µg ml⁻¹ streptomycin. Primary chicken cecal (PCC) cells, kindly provided by Ingrid Hänel, were isolated as described elsewhere and maintained in Quantum 286 medium for epithelial cells (PAA Laboratories) supplemented with 5% chicken serum and 0.5% chick embryo extract. Both cell lines were incubated in a humidified atmosphere of 95% air and 5% CO₂ at 37°C. ## Isolation of nucleic acids Genomic DNA of *C. jejuni* was isolated with the QIAamp DNA Mini Kit (Qiagen) according to the instructions of the manufacturer. Plasmid DNA was prepared using the GeneElute Plasmid Miniprep Kit (Sigma) following the manufacturer's protocol. ## Insertional knockout of gene *cj0268c* Initially, using genomic DNA from strain NCTC 11168 as a template, a 1090 bp DNA fragment representing *cj0268c* was amplified with primers Cj0268cF and Cj0268cR. The obtained PCR product was *XbaI* digested and ligated to the *XbaI* restricted and dephosphorylated plasmid vector pBluescript II KS (Stratagene). For the subsequent knockout, inverse PCR with the primers Cj0268cinvF and Cj0268cinvR using the plasmid described above as a template was carried out. The PCR reaction containing 10 mM Tris-HCl pH 8.3, 50 mM KCl, 1.5 mM MgCl₂, all four dNTPs (each 0.2 mM) and 10 pmol of each primer was carried out using 1 U PfuUltra High-Fidelity DNA Polymerase (Stratagene) to obtain a blunt end PCR product. Initial incubation at 95°C for 1 min was followed by initial 10 cycles at 95°C for 30 s, 55°C for 30 s and 72°C for 5 min, followed by 25 cycles under the same conditions with the exception of a shifted annealing temperature to 58°C. The resulting PCR product with a length of 3752 bp comprised of the complete cloning vector pBluescript, the 413 5′-terminal nucleotides and the 378 3′-terminal nucleotides of *cj0268c*. After gel extraction of the PCR product using the QIAquick PCR Purification Kit (Qiagen), a phosphorylated blunt end kanamycin resistance cassette described previously was ligated with the obtained PCR product using Quick Ligase (New England Biolabs) following the instructions of the manufacturer to obtain plasmid pBcj0268c-kanR. ## Cloning of gene *cj0268c* into pRRC For functional complementation of the *C. jejuni cj0268c*-knockout strain, gene c*j0268c* which was PCR-amplified with primers Cj0268cF and Cj0268cR and *XbaI* restricted as described above was cloned into likewise digested and dephosphorylated *C. jejuni* expression vector pRRC to obtain pRRC-*cj0268c*. In order to provide *cj0268c* with a His-tag, the gene was amplified with primers Cj0268cF and Cj0268cHis, which are listed in. PCR was performed in a TRIO- Thermocycler (Biometra) with 10 ng of genomic DNA of *C. jejuni* as a template. The PCR mixture contained 1 U PfuUltra High-Fidelity DNA Polymerase (Stratagene), 1x inherent reaction buffer, dNTPs (each 0.2 mM) and 10 pmol of each primer. After initial incubation at 95°C for 3 min, 40 cycles at 95°C for 30 s, 55°C for 30 s and 72°C for 2 min were carried out with a final incubation at 72°C for 5 min. Afterwards, the PCR amplicon was *XbaI*-digested and cloned into pRRC as mentioned above. ## Adhesion and invasion assays Bacterial invasion assays were performed according to the publication of Everest *et al*.. In brief, Caco2 cells were grown to approximately 80% confluence in a 6 well plate, washed with PBS and inoculated with 400 µl *C. jejuni* suspension adjusted to an OD_{(600 nm)} of 0.5 which corresponds to a multiplicity of infection (MOI) of 100. To investigate invasion, the *C. jejuni* suspension was removed after two hours and the cells were washed three times with PBS before further incubation with culture medium supplemented with 100 µg ml⁻¹ gentamicin. For the release of intracellular bacteria, the cells were lysed with 1% Triton X-100 for 10 min and the number of viable bacteria was determined by counting the number of bacteria colony forming units (cfu) grown on Columbia blood agar plates after incubation for 48 h at 42°C under microaerophilic conditions. For investigating bacterial adhesion, 6 well plates containing Caco2 cells and inoculated with *C. jejuni* or *E. coli* bacteria were centrifuged at 600×g for 5 min to increase the association of the bacteria with the cells. After incubation for 30 min, the monolayers were washed with PBS, cells were lysed and subsequently, plating of the bacteria was performed as described above. Every experiment was repeated four times. ## Motility and autoagglutination assays Tests for altered motility or autoagglutination of the *C. jejuni* strains and mutants were carried out as described previously by Tareen *et al*., 2011 and Tareen and Dasti *et al*., 2010. ## Resistance of *C. jejuni* to bile salts Bacteria were harvested from overnight-incubated blood agar plates and adjusted to an OD_{600 nm} of 0.1 in Muller Hinton broth. Then, the bacterial cultures were supplemented with 0.09, 0.18, 0.37, 0.75, 1.5, and 3% of cholate and deoxycholate, respectively. After incubation for 24 h at 37°C in a microaerophilic atmosphere, OD_{600 nm} was determined. All tests were carried out in triplicate. ## Sensitivity of *C. jejuni* to Triton X-100 The sensitivity of *C. jejuni* to Triton X-100 was analyzed using bacterial colonies grown overnight which were harvested from blood agar plates. The bacteria were resuspended in distilled water and adjusted to an OD_{600 nm} of 0.1. Triton X-100 was added to a final concentration of 1.0%, and after incubation for one hour at room temperature the number of viable bacteria was determined by plating serial dilutions onto blood agar plates and subsequent incubation at 42°C for 48 h in a microaerophilic atmosphere. Every experiment was repeated three times. ## Screening of *C. jejuni* strains for presence of gene *cj0268c* Genomic DNA samples of 56 *C. jejuni* strains were analyzed by PCR with 10 pmol of primers 268cGMF and 268cGMR and 1 U *Taq*-Polymerase (Roche) in a TRIO- Thermocycler (Biometra) under following conditions. After an initial denaturation at 95°C for 1 min, 35 cycles at 95°C for 30 s, 55°C for 30 s and 72°C for 1 min were performed. Afterwards, samples were run on a 2% agarose gel for the detection of the resulting 146 bp PCR amplicon. ## Lysis of *E. coli* DH5α cells and immunoblot analysis *E. coli* from 5 ml overnight culture were centrifugated and resuspended in 500 µl 1x PBS. After addition of 100 µg lysozyme and NP40 to a final concentration of 1%, lysates were incubated on ice and were casually shaken for 10 min. Following the addition of 22 µl of 5 M NaCl, the lysates were centrifuged for 45 min at 16000×g. Then, the supernatants were subject of subsequent immunoblot assays. Aliquots of 20 µl were separated in a 15% SDS-PAGE gel and blotted onto a PVDF membrane (GE Healthcare) by a semidry transport system (Sartorius). After protein transfer, the membrane was blocked for 1 h with 5% milk powder in PBS containing 0.05% Tween 20. Subsequent incubation of the membrane with a 1∶3000 dilution of monoclonal mouse anti-His primary antibody (Qiagen) over night at 4°C was followed by labelling of the immune complexes with a 1∶3000 dilution of horseradish-peroxidase-conjugated anti mouse secondary antibody (Dianova) for 1 h at room temperature and visualization by ECL chemiluminescence. ## Permeabilization of *E. coli* DH5α cells In order to get access to the bacterial periplasm, *E. coli* cells were permeabilized. prior to antibody labelling and flow cytometry analysis. 5×10⁷ bacterial cells of each sample were washed in 1 x PBS and were resuspended in 180 µl 20% sucrose, 50 mM Tris-HCl, pH 8 containing 10 mM EDTA. After 10 min 20 µl lysozyme at a final concentration of 2 mg/ml in 20 mM Tris- HCl, pH 8, 2 mM EDTA, 1.2% NP40 were added, and the samples were incubated at 37°C for 20 min. Finally, cells were washed twice in 20% sucrose, 50 mM Tris- HCl, pH 8. ## Antibody labelling of *E. coli* cells and flow cytometric measurements Unspecific antibody binding was prevented by incubation of *E. coli* cells in 100 µl 1 x PBS containing 1% BSA for 30 min at 4°C. Cells were collected by centrifugation at 10.000 g for 2 min and incubated in 100 µl 1 x PBS, 1% BSA containing anti-penta His antibody (Qiagen) at a concentration of 10 µg/ml at 4°C for 30 min. After washing three times with 1 x PBS, 1% BSA, immune complexes were labelled with 50 µl R-phycoerythrin-conjugated goat F(ab′)₂ fragment anti-mouse IgG (1∶50, Dianova) for 30 min at 4°C. After having been washed, cells were fixed in 1% PFA in PBS. Subsequently, flow cytometric measurements were carried out using a FACSCalibur flow cytometer (Becton Dickinson). ## Statistical analysis The significance of differences (*P*-value less than 0.01) between mean values was calculated by the Mann-Whitney U test. # Results ## Knockout of *cj0268c* in *C. jejuni* reference strain NCTC 11168 and functional complementation In a previous report we detected gene *cj0268c* to contribute to the invasion of host cells in *C. jejuni* strain B2. Since strain B2 is highly invasive in human colon epithelial cells but fails to infect chicken cells or to colonize birds we wanted to confirm the invasion deficient phenotype in the *C. jejuni* reference strain NCTC 11168. This strain is capable of causing gastroenteritis in men as well as settling in livestock and, for this, enables us to test whether *cj0268c* has a defined function for colonization or infection of a particular host. After introduction of plasmid pBcj0268c-kanR to obtain NCTC 11168::*cj0268c* we restored the parental phenotype by transformation of the mutant strain with pRRC-*cj0268c* to obtain NCTC 11168::*cj0268c*-comp-*cj0268c*. A description of the genetic arrangement of *cj0268c* in strain NCTC 11168, the *cj0268c* knockout mutant, the complemented strain and the location of the primers is shown in. In order to independently confirm the invasion deficient phenotype of *C. jejuni* strain B2Δ*cj0268c* in reference strain NCTC 11168, we repeated the gentamicin protection assays on Caco2 cells with the wild type strain, the *cj0268c*-knockout mutant and the corresponding complemented strain. Thereby, the phenotype of strain B2Δ*cj0268c* could be approved and, hence, the invasion- deficient phenotype demonstrated to be due to the functional loss of *cj0268c* in both, the *C. jejuni* strains B2 as well as in NCTC 11168. If the number of recovered NCTC 11168 wild type colonies is defined as 100%, we detected a mean value of *cj0268c*-mutant colonies of only 62% (p\<0.0007). The percentage of obtained colonies from the complemented strain was 93% which was not significantly different from values of the parental NCTC 11168 strain. To address, whether the invasion-deficient phenotype due to the loss of *cj0268c* was restricted to human cells, we performed gentamicin protection assays with primary chicken cecal cells infected with strains NCTC11168, NCTC11168::*cj0268c*, and NCTC 11168::*cj0268c*-comp-*cj0268c*. In support of our data obtained with Caco2 cells, the invasion capacity of the *cj0268c*-mutant strain was significantly reduced compared to parental NCTC 11168 strain, but was completely restored in the complemented mutant. If the invasion capacity of wild type strain NCTC 11168 was defined as 100%, the percentage of colonies obtained from mutant strain NCTC 11168::*cj0268c was* only 62% (±2.36, P\<0.0007) but was reconstituted to 96% (±4.96) by the intact gene in the complemented strain NCTC 11168::*cj0268c*-comp-*cj0268c* compared to NCTC 11168. ## Adherence Given that *cj0268c* encodes a putative transmembrane protein, we next tested whether this protein is involved in the pathogen-host cell adherence process. Applying adhesion assays with wild type strain NCTC 11168, the *cj0268c*-mutant and the complemented mutant on Caco2 cells, we could decisively detect an adherence-deficient phenotype of NCTC 11168::*cj0268c* compared to NCTC 11168 which could be restored to wild type level in the *cj0268c*-complemented strain. When defining the number of colonies recovered from the parental strain NCTC 11168 as 100%, the mean value of corresponding colonies for the mutant NCTC 11168::*cj0268c* was 60.9% (±5.18, P\<0.0039) whereas the recovery rate of colonies for the complemented strain was 96.4% (±1.77) which represents the wild type adhesion level. To further phenotypically characterize the *cj0268c*-deficient mutant, we performed assays to test for altered motility or an affected capacity to autoagglutinate compared to the wild type strain and the complemented mutant. As shown in the mutant strain exhibited the same motility and autoagglutination properties as compared to NCTC 11168 wildtype or NCTC 11168::*cj0268c-*comp-*cj0268c* and, thereby, excluding a role of *cj0268c* regarding these characteristics. ## *E. coli* strain DH5α expressing Cj0268c possesses an increased adherence to Caco2 cells After cloning of *cj0268c* including a C-terminal His-tag in pRRC, we were able to detect recombinant Cj0268c expression also in *E. coli* DH5α even though in this vector the chloramphenicol resistance-mediating gene, as well as *cj0268c* is under control of the *C. jejuni* 16S promoter. Performing Western-Blot analysis with corresponding *E. coli* DH5α lysates and a monoclonal antibody against the His-tag, we determined a specific protein band of an approximate size of 41 kDa which represents the expected size of Cj0268c. To investigate if Cj0268c expressed in the heterologous context of *E. coli* intensifies its capability to interact with eukaryotic cells, we repeated the adherence assays with Caco2 cells. By defining the number of recovered *E. coli* DH5α colonies representing the parental strain as 100%, the relative recovery rate of live bacterial colonies from the Cj0268c-expressing *E. coli* was 201.7% (±6.44). Hence, we could confirm the adherence-mediating phenotype of Cj0268c, indicating that Cj0268c does not necessarily need to interact with other *C. jejuni* proteins, but by itself enhances adhesion properties in a heterologous host. ## Localization of Cj0268c In order to determine whether Cj0268c is localized at the bacterial surface we carried out flow cytometric measurements with non-permeabilized *E. coli* cells. After incubation of *E. coli* expressing a His-tagged version of Cj0268c with a monoclonal anti-His primary antibody followed by labelling with R-Phycoerythrin- conjugated secondary antibody, we could not detect any particular staining of the bacterial cell surface compared to *E. coli* bacteria that harboured plasmid pRRC without gene *cj0268c*. Hence, at least the His-tagged C-terminus of Cj0268c is not localized at the bacterial surface. In contrast, when we permeabilized the bacterial cell wall with EDTA and lysozyme to allow access of the antibodies to the proteins of the periplasm, we could clearly detect immunolabelling of the *E. coli* population expressing Cj0268c. Whereas anti-His labelling was also obtained after labelling of permeabilized *E. coli* which had been transformed with the empty vector, the mean fluorescence intensity of antibody-labelled cells was considerably higher for Cj0268c-expressing cells as compared to controls (824.6 FLI units as compared to 484.9;). Control staining with the secondary antibody only of both Cj0268c-positive and negative *E. coli* confirmed the specificity of the FACS analysis. Together, these results indicate that the protein Cj0268c resides in the periplasmic space. ## Resistance to bile salts and Triton X-100 Next we determined whether the presence of Cj0268c strengthens the resistance against bile salts and detergents. Strains NCTC11168, NCTC11168::*cj0268c*, and NCTC 11168::*cj0268c*-comp-*cj0268c* were incubated with different concentrations of cholate and deoxycholate ranging from 0.09 to 3%. Subsequent measurements of the optical density of the cultures revealed no differences between wildtype, *cj0268c*-mutant and its complemented mutant. Given that we were not able to detect any bacterial growth in the presence of bile salt concentrations exceeding 0.75%, further experiments were carried out only with cholate and deoxycholate concentrations up to 0.75%. However, the measurements of the optical densities in these experiments did not reveal any significant differences neither comparing the respective bacterial strains nor with respect to different bile salt concentrations tested (data not shown). Thus, these data suggest that Cj0268c does not exert a functional correlation with e.g. efflux pump systems such as CmeABC, for instance. To find out whether Cj0268c has any influence on the stability of the bacterial cell wall, we incubated the *cj0268c*-deficient strain, its complemented version and NCTC 11168 wildtype strain in the presence of the nonionic surfactant Triton X-100 which is commonly used as a detergent to permeabilize cellular membranes. Thereby, the *cj0268c* mutant strain was much more sensitive to Triton X-100 as compared to wild type and complemented mutant. When we incubated the respective strains in a final concentration of 1% Triton X-100 for 1 h, subsequent plating onto Columbia blood agar yielded significantly fewer CFU of the *cj0268c*-mutant strain as compared to wild type strain NCTC 11168 and the *cj0268c*-complemented version. When defining the number of wild type colonies obtained after plating of the fourth dilution as 100%, the relative abundance of *cj0268c*-mutant CFU yielded only a mean of 21.3% , indicating that Cj0268c significantly contributes to the bacterial cell stability. ## The *cj0268c* gene is ubiquitous in the *C. jejuni* population In recent studies multilocus sequence typing (MLST) analysis of defined genetic markers allowed the classification of different *C. jejuni* clonal groups. In addition, the combination of these genetic markers correlated, at least to some degree, with a corresponding animal source. Moreover, after expansion of the MLST analysis by the inclusion of further gene markers, an association with a higher prevalence of campylobacterioses in humans or livestock adaption could be assigned regarding to the distribution of these markers. We therefore studied next whether gene *cj0268c* fits into one of these clonal groups and, furthermore, could be associated with the *C. jejuni* settlement of particular animal groups or the severity of human campylobacteriosis. Altogether 56 isolates out of the 266 isolates used in the studies mentioned above from all clonal groups were screened for the presence of gene *cj0268c* by PCR. We were able to detect gene *cj0268c* in all isolates of *C. jejuni* indicating that *cj0268c* is not related to a distinct clonal group but rather seems to be ubiquitous in *C. jejuni* (not shown). Thus, a contribution of *cj0268c* to the *C. jejuni* settlement of specific animal groups or the clinical course of campylobacteriosis is unlikely. # Discussion Screening of our transposon-generated mutant library of *C. jejuni* strain B2 revealed altogether seven genes that mediate a diminished invasion capacity towards Caco2 cells as shown by gentamicin protection assays. Gene *cj0268c* which was further characterized here has been shown to be involved in host cell invasion earlier. In order to investigate this gene regarding its biological function in further detail, we inactivated *cj0268c* in *C. jejuni* reference strain NCTC 11168. This strain, in contrast to strain B2, is able to infect both, human as well as chicken cells. After generation of a corresponding *cj0268c*-complemented NCTC 11168 strain, we could verify this gene to belong to a number of factors which are important for the adhesion to host cells by the pathogen. Furthermore, we confirmed this role of Cj0268c by heterologous expression of *cj0268c* in *E. coli* strain DH5α. This result confirmed that Cj0268c possesses an adhesion mediating function alone and does not have to interact with other proteins of *C. jejuni*. On the other hand, an alteration of surface properties of *E. coli* after heterologous expression of Cj0268c leading to an indirect effect cannot be ruled out. Since we obtained equal CFUs of the *E.coli* populations independent of the expression of Cj0268c after incubation for 10 min in the presence of 1% Triton X-100, we could exclude alterations in bacterial stability caused by Cj0268c as the reason for different CFU numbers. However, since the adherence of *C. jejuni* to human and chicken cells depended exclusively on the presence or absence of Cj0268c irrespective of the specific host cell species, we determined Cj0268c as a protein for the mediation of adherence in general. The chicken intestine represents a natural habitat for *C. jejuni* colonization and, hence, resistance to bile salts is essential for the pathogen's survival in such a hostile milieu. One of the resistance mechanisms exerted by *C. jejuni* employs the multidrug efflux pump system CmeABC, consisting of an outer membrane protein CmeC, a drug transporter CmeB, localized in the inner membrane, and periplasmic CmeB to connect CmaA and CmeC. Since Cj0268c is a predicted transmembrane protein, interaction with CmeB to stabilize the CmeABC complex for instance was conceivable. However, after incubation of parental strain NCTC 11168 and its corresponding *cj0268*-mutant, the resistance to bile salts like cholate and deoxycholate exerted by the pathogen strains was virtually identical. Triton X-100 incubation of *C. jejuni* strains with mutated genes leading to an incomplete lipooligosaccharide metabolism had an inconsistent outcome reported so far. Whereas a mutation in the heptosyltransferase gene *waaF (cj1148)* decreased the stability of the *C. jejuni* cell wall, the resistance to Triton X-100 of the galactosysltransferase *cj1136-*deficient mutant was not altered compared to the parental strain. Nevertheless in the case of Cj0268c, incubation of the parental strain, the mutant and the complemented mutant with 1% Triton X-100 revealed that Cj0268c is important for maintaining integrity of the bacterial cell wall. Finally, although we could clearly demonstrate Cj0268c to possess an adherence-mediating function and, therefore, contributes to the invasion process, the corresponding gene seems to be ubiquitous and not to belong to distinct clonal *C. jejuni* groups which could be related to pathogenicity according to earlier findings of. The *in vitro* results presented here need to be further complemented by investigating the biological impact of the respective mutant strains in a suitable murine model mimicking human campylobacteriosis. We thank Ingrid Hänel (FLI, Jena, Germany) for providing PCC cells, and we thank the CASVAB, University of Balochistan, Pakistan and the Higher Education Commission, Pakistan. [^1]: Markus M. Heimesaat and Stefan Bereswill currently serve as editors for this journal. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: RL AMT UG. Performed the experiments: AMT AEZ CGKL. Analyzed the data: AMT RL MMH SB CGKL. Contributed reagents/materials/analysis tools: AMT RL AEZ UG. Wrote the paper: RL UG MMH SB.

# Introduction Policymakers need solid information on how labour market evaluates higher education graduates. Institutions also should collect and analyse relevant information about their graduates for the management of their programs. Since the salary and the chance of finding a job are important decision factors at the college attendance, university and program level public information about the career paths are also important to candidates of higher education. Although self-reported data can have validity problems, questionnaire based databases are useful to study education-occupation matches. Among these, the Reflex database is the most comprehensive information source in Europe. The analysis of this database showed that graduates working in the field of their study have higher income and satisfaction, so they are a happier members of the society. Administrative data can replace traditional questionnaires to offer much more objective information for evidence-based educational policy in decision-making. In Austria database of the whole state insurance system is accessible in anonymized form, which is also ready to career path analysis. With administrative data, we can also measure the added value of higher education institutes by combining information about persistence rates, graduation rates, and post-college earnings. The use of administrative data has a long tradition in Northern Europe. Finland recently connected administrative and survey data sources. Based on the register of Statistics of Finland some employers were suggested to be interviewed to study unemployment of young graduates and transition from higher education to work. The Swedish Ladok database was used to determine the influence of higher education institutions on labour market by regression analysis. The availability of extensive, longitudinal data made it possible to the evaluate the matching of the occupation and the level of the degree among engineering, teaching, nursing, business specialisations. In our study, we try to dig deeper by focusing on the more detailed program level by proposing a goal oriented graph mining tool to evaluate the matching of programs and occupations. In recent years, network-type models have been proven to be useful in understanding complex systems in different subject areas (e.g. sociology, economy, industry, and biology). Real life entities (e.g. people, universities, educational programs) can be characterised by numerous categorical properties (e.g. education can characterise people). Relationships between entities and values of a selected property can be modelled with a two-mode network (also known as a bipartite graph). The proposed network model is based on the integration of the databases of the National Tax Administration, the National Health Insurance Fund, and the data warehouse of the Hungarian higher education. This administrative dataset covers 15 thousand people graduated in 2009/2010 school year and worked in 2012 May. Based on the data of 7402 Bachelor students we defined a bipartite graph of 110 bachelor programs and 113 occupations encoded by the third level of International Standard Classification of Occupations (ISCO) code system. The nodes of the resulted network are connected by 7400 links that represent the employees who received their bachelor level in a given program and work in a given profession. (The data will be available on the website of the author: [www.abonyilab.com](http://www.abonyilab.com)). To demonstrate the background of the research, we present a brief analysis of gender pay gap and the spatial distribution of over-education. The analysis of the bipartite network shows that both the programs and the occupations follow a power law distribution which reflects there is a structure in the carrier paths. Our key idea is that we compare the weights of the edges with the expected number of edges of a random graph that has the same degrees as the studied network. This configuration model seems the most sophisticated reference because it takes into account the expected number of links by degrees of given program and occupation. To search patterns in education-occupation transition in different levels of details, we cluster the graph by looking for subgraphs whose vertices are more likely to be connected to one another than to the vertices outside the subgraph. To evaluate the consistency of the detected clusters we use a graph modularity based measure which assesses the quality of the clusters based on the number of edges of the configuration model. We elaborated a multi-resolution type analysis of the network by the step by step removal of the weak connections. The results highlight that the educational programs have a hierarchical structure. A large number of higher education programs can lead to a fragmented and inefficient education system. The results confirm that the extracted clusters can support decisions related to the monitoring and (re-)design of the program structure. # Methods ## Bipartite graph model of the education to work transition The vertices of the bipartite graph model of the education to work transition are divided into two disjoint sets, $\mathcal{U},\mathcal{V}$. The $\mathcal{U}$ represents the educational programs, and the $\mathcal{V}$ represents the sets of occupations. Every edge connects a program to an occupation. The edges are weighted, and the weights are representing the number of graduated students in a given program connected to a specific profession. The graph can be represented by an **A** adjacency matrix, where the *A*_*ij* element of the matrix represents how many graduates of the *i*-th bachelor program are working on the *j*th profession. By following this arrangement, the sum of the *i*-th row of represents the number of students graduated in the *i*-th program, while the sum of the *j*-th column represents the total number of employees having a given profession. These sums can be considered as the degree of the nodes, calculated as *k*_*i* = ∑_*j* *A*_*ij* and *k*_*j* = ∑_*i* *A*_*ij*, respectively. Not all nodes in a network have the same number of edges (same node degree). The probability that a node has \< *k* \> edges can be described by a distribution function *P*(*k*). The analysis of the degree distribution can show how the graduates are distributed among the programs and the occupations. ## Evaluation of the education-occupation match To decide which education programs and occupation pairs are relevant and which can be considered as a “noisy” individual case, we propose a measurement to evaluate the strengths of the connections. Our core idea is that we can compare the *A*_*ij* weight of the edge with the expected edge weight of a random graph that has the same degrees as the studied network. This configuration model, which is often referred as a random network with a pre-defined degree sequence, seems the most sophisticated application because it takes into account the expected number of links by degrees of given program and occupation. If the edges were randomly distributed, $\frac{k_{i}k_{j}}{L}$ would be the expected number of links between the *i*-th program and *j*-th occupation, where *L* represents the total number of links in the network, *L* = ∑_*i*,*j* *A*_*ij*, while *k*_*i* and *k*_*j* are the degrees of the program and occupation nodes, respectively. Since in the case of random matching $\frac{k_{i}k_{j}}{L}$ graduates of the *i*-th program would choose the *j*-th occupation, the difference between the actual and the expected number of graduates in the case of random arrangement can be calculated as: $$\begin{array}{r} {A_{ij} - \frac{k_{i}k_{j}}{L}\,} \\ \end{array}$$ which difference can be used as a measure of the strength of the education—occupation matchings. ## Simultaneous clustering the programs and the occupations In the previous session, we evaluated the connection of individual educational programs and occupations. To provide information about the whole structure of the network, we cluster the edges to obtain groups of similar programs and professions. To formalise this clustering problem, we utilise the modularity measure introduced by Newman and improved for bipartite graphs by Barber. A module of the network is a subgraph whose vertices are more likely to be connected to one another than to the vertices outside the subgraph. Modularity reflects the extent, relative to a random configuration network, to which edges are formed within modules instead of between modules: $$\begin{array}{r} {Q = \frac{1}{L}\sum\limits_{ij}\left( A_{ij} - \frac{k_{i}k_{j}}{L} \right)\delta\left( C_{i},C_{j} \right)\,} \\ \end{array}$$ where the Kronecker delta function *δ* is equal to one when nodes *i* and *j* are classified as being in the same module (i.e. they have the same label value) or zero otherwise. The modularity can be determined for each community of a network. A network with *n*_*c* communities, the following modularity value is used to determine *M*_*c* community modularity value. Each *C*_*c* community with *N*_*c* nodes are connected with by *L*_*c* links, *c* = 1, … *n*_*c*. $$\begin{array}{r} {M_{c} = \frac{1}{L}\sum\limits_{{(i,j)} \in C_{c}}\left( A_{ij} - \frac{k_{i}k_{j}}{L} \right)} \\ \end{array}$$ The *M*_*c* modularity value of a *c* cluster can be either positive, negative or zero. In the case of zero, the community has as many links as a random subgraph. If it is a positive value, then the *C*_*c* subgraph tend to be a community, while a negative *M*_*c* means it is not. We used the multi-level modularity optimization algorithm (so called Louvain algorithm) to find clusters in the programs-occupation bipartite graph. This algorithm uses an iterative procedure to assign each node to a module by maximising the modularity. The rows and the columns of the adjacency matrix of the bipartite graph can be reordered to visualise the similarities of the programs and relationships (see later in chapter Clustering and Visualization). ## Multi-resolution cluster analysis Since we would like to determine how the significant matchings are structured, we applied a method for cleaning the network by step by step removing of the weak connections. $$\begin{array}{r} {{\widetilde{A}}_{ij} = \begin{cases} {A_{ij},} & {if\, A_{ij} \geq \frac{k_{i}k_{j}}{L} \cdot \alpha} \\ {0,} & {if\, A_{ij} < \frac{k_{i}k_{j}}{L} \cdot \alpha} \\ \end{cases}} \\ \end{array}$$ As the describes the cleaning procedure has one 0 ≤ *α* threshold parameter. When *α* = 0, none of the edges are removed. It should be noted, that *α* can be considered as a minimum relative edge strength. After the pruning of network, all connections will have *α* times larger weight than weight we would expect based on the random configuration model: $$\begin{array}{r} {\frac{{\widetilde{A}}_{ij}}{\frac{k_{i}k_{j}}{L}} \geq \alpha\,\forall i,j} \\ \end{array}$$ It should be noted that this equation measures how the given edge contributes to the Louvain ratio used to measure the compactness of a module/cluster: $$\begin{array}{r} {LR_{C_{c}} = \frac{A_{C_{c}}}{P_{C_{c}}}\,\, A_{C_{c}} = \sum\limits_{{(i,j)} \in C_{c}}A_{ij}\,\, P_{C_{c}} = \sum\limits_{{(i,j)} \in C_{c}}\frac{k_{i}k_{j}}{L}\.} \\ \end{array}$$ As can be seen later in chapter Evaluation of Education—Occupation Matching, it is interesting to analyse how the step-by step increase of this parameter decreases the network density, what is the ratio of the non-significant edges, how characteristically structured the network. Since after this pruning we also applied modularity based clustering, the resulted method can be considered as a special multi-resolution analysis technique. Modularity optimization based community detection has a resolution limit, failing to detect communities smaller than a scale that depends on the total number of edges in the network and degree of interconnectedness of the communities. To handle this problem multi-resolution methods were introduced by adjusting the resolution of the algorithms by modifying the modularity function, weighting the contribution of the null model or adding self-loops to the nodes. These methods still have the intrinsic limitations that large communities may have been split before small communities become visible. Since the problem is that modularity based community detection algorithms join small fully connected subgraphs connected only by weak edges into larger groups, our methodology which gradually removes the less important connections by the increase of *α* can also be considered a graph-modification based multi- resolution approach that handles this problem. It should be noted, that we developed our algorithm as we were interested in how the statistically significant education-occupation matchings are structured. # Results and discussion ## Administrative data of the Hungarian career path tracking system The development of our methodology is motivated by the question, how the Hungarian carrier path tracking system can be used to support evidence based policy making and monitoring. The studied administrative government data were collected as the integration the databases of the Hungarian tax office and National Health Insurance Fund in 2014. The database was designed by using individual hash codes, so although it does not contain personal information. It allows the micro data level analysis of student career paths. The integrated database contains 70 variables about - personal data (date of birth, county of address, citizenship, gender) - occupational data in the year of 2012 (employment relationship, occupation, gross wage, etc.) - employer data (county of company headquarters, company size, company activity, etc.) - if the graduate runs her/his own enterprise the primary data of the company - educational data (institute, faculty, program where the graduate graduated) The dataset contains 29873 individual records. Among these only 15253 people have occupational data. It must be noted, that this integration was the first made by the related governmental organisations in Hungary, so probably this is the reason why only the half of the persons were correctly merged. shows contents of published database. The correctly identified 15253 people graduated 398 education programs delivered by 52 institutions and worked in 402 occupations encoded by the fourth level (unit groups) of the International Standard Classification of Occupations (ISCO) code system. In this work, we focus on just bachelor degree graduates. Among 15253 people 7402 has a bachelor degree in 45 institutions, 110 programs, and works in 372 occupations, and 113 third level occupation groups. This database is open for research purposes. The cleaned dataset used in this study is available on our website: [www.abonyilab.com](http://www.abonyilab.com) ## Measuring overeducation and gender pay gap Interesting point of the dataset that according to the main group of ISCO code we can determine that a given occupation requires higher education degree or not. shows how much percentage of the graduates has jobs that require higher education degree. The Shankey diagram of the BSc/BA graduates shows that who graduated in computer science and information technology, health science, engineering science works more likely in an occupation that requires higher education degree compared with graduates in sports science, arts and humanities, natural sciences, agricultural science. Working in a field that matches to the education has a positive effect on job performance and satisfaction. Results of Iriondo and Pérez-Amaral indicates that overeducated workers suffer a wage penalty since earnings depend mainly on the educational requirements of jobs. Three primary measures of education—job mismatch can be distinguished based on how the required education level is determined. The first method relies on the self-assessment, the second approach evaluates the “realised matches”, while the third “job analysis method” refers to a systematic evaluation of the “professional job analysts” who specify the required level of education for the job titles in an occupational classification . These last two methods require large scale and up to date administrative data-based studies, similar that we would like to deliver in our research. Groot and Maassen van den Brink conducted a meta-analysis of 25 studies on overeducation and found that the matching based methods show 13.1% of overeducation, while self-assessment based studies estimate the much higher percentage of overeducated employees, 28.6%. McGuiness and Sloane used the REFLEX dataset to study overeducation in the UK. When both education and skill mismatch variables were included in the model, overskilling reduced job satisfaction consistently for both sexes. In the UK 36% of the respondents felt as overeducated, which is quite high, compared to the 14% measured elsewhere in Europe, and 17% in Taiwan in 2008. A much more objective study has been performed in Spain where the Spanish Wage Structure Survey (WSS) dataset was used to examine the effects of educational mismatch on wages. Based on this employer-worker microdata 32-37% overeducation rates were calculated. Our database allows a more detailed analysis. Similarly to other countries, this dataset also shows 26-39% of overeducated employees. The spatial distribution of occupations of the graduates was also investigated. shows how much percentage of the bachelor graduates are working in jobs that require higher education. This figure well illustrates that the problem of over-education is mainly related to the economic development of the regions. The database also allows the sophisticated analysis of the gender pay gap. shows income gaps grouped by different education areas and shows incomes of different occupation categories according to the first level ISCO. ## Evaluation of the degree distributions The cumulative degree distributions of the bipartite network are shown in Figs and. The *k*_*i* weighted degrees of the five biggest programs are: Business Management: 874, Communication and Media Science: 469, Andragogy: 367, Tourism: 364, Finance and Accounting: 310. The *k*_*j* degrees of the top five occupations are: Business services agents: 614, financial and mathematical associate professionals: 419, Engineering professionals (excluding electrotechnology): 410, Legal professionals: 372, Sales and purchasing agents and brokers: 371. To evaluate the whole structure of the network we fitted power-law distribution, exponential, Poisson, log-normal distributions to degrees of nodes and occupations in R with the help of the *poweRlaw package*. As can be seen, there is a well defined linear region between *k*_*sat* and *k*_*cut*. The slope of this linear trend gives *γ* which is 2.00. There are less number of small degree nodes (occupations) then power-law fit would require therefore in *k* \< *k*_*sat* region data point are below the extrapolation of fitted line. Similarly, there are less number of high degree nodes or hubs then power- law fit would require thus in *k*_*cut* \< *k* region data point are also below the extrapolation of fitted line. The degree distribution of these scale free networks can be described by a power-law tail function, *P*(*k*) = *k*^−*γ*, and the *γ* parameter is one of the most important property of a graph. The question is that which model is closer to the empirical distribution. The p-values shown in suggest that we cannot reject the null hypothesis that the data follows power-law distribution. Power-law and log-normal distributions were compared using Vuong’s test statistic. The two sided p-value of comparison test shows that both distributions are equally close to the empirical distribution. When the nodes of a network are randomly connected, *γ* is bigger than three. 2 \< *γ* \< 3 relates to scale-free networks. Our analysis shows that the studied bipartite network has scale-free structure, which proves that graduates do not randomly choose an occupation, and they preferences can be studied by a more detailed analysis of the edges. ## Evaluation of education—Occupation matching The proposed graph configuration model based measure can be used for ranking the education—occupation matchings. Tables and show the strongest and the weakest educational program—occupation pairs. shows the distribution of the weak edges that will be removed at a given *α* threshold. ## Degree correlation and centrality measures Programs with a relatively small number of connections are not “spread”. In some of these programs, there are a lot of graduates, but they work in a few kind of occupations. These programs are the following: computer science, engineering (electrical engineer, mechanical engineer, civil engineer, chemical engineer, mechatronic engineer), special need teacher, nursing and patient care, laboratory and diagnostic imaging analyst, and economic science (finance and accounting, international management, management organisation). Conversely, only small number of agricultural, teacher training, liberal arts, light industrial engineering graduates work in several kinds of occupation. To get more information about the structure of our network, we calculated the degree correlation as the correlation of the node degree with the node degree of the neighbour nodes. The results show that our graph is disassortative, which means high-degree components (hubs) tend to connect to low-degree nodes, while low-degree nodes are connected to hubs. In practice, this means that graduates of programs on that few people graduated work in popular occupations. Computer science, nursing, and engineering programs are interesting in the context of eigenvector centrality as well. These programs are mainly connected to occupations that have few kind of “supplier”, so they are not so embedded in the graph. This group of programs have relatively high betweenness and low closeness centrality which means that they include many shortest path, but they are far from the centre of the graph. This phenomenon predicts that the related programs—occupations connections are strong and form subgraphs that have fewer connections to other parts of the graph. ## Clustering and visualization During clustering each program and occupation were assigned to one module, so each module contains a set of programs and professions. The result of clustering can be seen in. This figure shows the adjacency matrix with columns and rows ordered according to the result of the clustering. The resulting five clusters are marked by A-E letters are the diagonal blocks of this matrix. shows the Louvain ratio for every module. Module ‘A’ consists of engineering programs supplemented with design, technical trainer and germanistics programs. This cluster highlights that 25% of graduates of germanistics work in manufacturing and IT sector, indicating the strong presence of the German industry requiring advanced knowledge of German language. Module ‘B’ consists of economy, social, political, and language teacher programs weakly connected to sales, office workers, client information, journalist, brokers, marketing and PR professionals. Module ‘C’ contains management, financial economic, and agricultural programs, along with small programs such as physics, earth science, cultural anthropology. With them, financial, business, clerk, trade workers, keyboard operators, and service worker occupations are associated. It should be noted that there is a relatively strong connection between the ‘B’ and ‘C’ modules. Module ‘D’ connects medical, pedagogy, teacher, social work, dancer and arts type programs with health, teaching, child care workers, medical technicians, and personal care workers. In this module, there are the fewest number of occupations which do not require higher education diploma. Module ‘E’ collects programs with a small number of graduates. This module shows how agricultural, natural science, teacher, sport, art type of programs are connected to operators, technicians, workers, vocational education teacher, animal producer, crop grower, cooks, salesperson, services manager, and life science professionals. In the case of the A, D and E modules this ratio shows a larger difference from the null model compared to the B and C models, which indicates the stronger connection of programs and occupation in the A, D, and E modules. During our work, we tested several clustering algorithms, including the BRIM algorithm (bipartite, recursively induced modules) developed specifically for bipartite graphs. As shows, the first module contains mainly teaching, humanity and art programs. The second module exhibit business, economic, finance, HR, social work, nursing, medical programs. In this module, almost half of the occupations do not require higher education diploma, like cooks, hairdresser, personal services, cashier, personal care, food preparation assistant, elementary worker. The third module represents natural and technical sciences programs, like engineering, IT, physics, ecology, earth science connected to production, manufacturing, information managers, life science, engineering professionals. ## Application of multi-resolution cluster analysis Louvain modularity optimization algorithm was performed with different *α*.. shows the number of clusters resulted from Louvain algorithm in the function of *α*. We detected hierarchical structure since all communities found at a value of *a*₂ \> *a*₁ are sub-communities of the communities found at *a*₁. We measured the similarities of the nodes based on whether their share the same cluster in different resolutions. A hierarchical splitting occurs when the cluster of health type programs splits into nursing and medical professionals. Similarly, the cluster of pedagogy and social programs is divided into teaching and social professionals and the group of child care workers. The relationships of the clusterings resulted in different resolution level is visualised in. As can be seen, somewhat hierarchical splitting occurred in the first cluster. By increasing *α* IT engineer, business information, software engineering, germanistics programs with software developer, database professionals, information technology operations technicians occupations separated (see *y* = 1, *x* = 3) # Conclusion Administrative data based career path analysis can of support governmental policy making and program development of higher education institutes. To support the extraction of useful information from these databases we developed a graph- based data structure to represent the career path of higher education graduates. Education—occupation mismatch can be analysed based on the bipartite graph of bachelor programs and occupations encoded by International Standard Classification of Occupations (ISCO) code system. We modified the Newman modularity measure to evaluate the matching of the programs and the professions. Based on this measure the hidden structure of career paths can also be clustered and visualised. The proposed network model is applied on the integrated databases of the National Tax Administration, the National Health Insurance Fund, and the data warehouse of the Hungarian higher education. To demonstrate the information content of this administrative database, we presented a brief analysis of the gender pay gap and the spatial distribution of the over-education. Similarly to other countries, we showed 26-39% of overeducated employees. The transition of graduates from higher education to employment is affected by individual characteristics. However, graduates with well-defined qualification start working in a somewhat similar profession. Our graph model gives the opportunity to cluster the typical career paths and find outliers whose education and occupation does not match. The results illustrate that the proposed multi-respolution type community finding approach provides useful results, as it highlights the groups of programs that are strongly connected to groups of bachelor programs. The analysis of the clusters allows us the more sophisticated analysis of the performances of the programs in the labour market. For example, our method showed that significant proportion of graduates of Germanistics work as a system administrator with approximately 30% higher salary indicating the strong presence of German origin industry in Hungary. Such results can be useful for education policy experts and decision makers who can see the structure of the Bachelor programs from the objective viewpoint of the labour market. The resulted orderings and matching measures can support the policymakers to fine-tune the fragmented program structure of the Hungarian higher education. We found bachelor programs that are almost identical in their content, but they are different from the view of the labour market because graduates work in the different occupation. For example, pedagogy and andragogy use similar methods, but the graduates of andragogy work profession that is more related to communication and media science. (Probably this was one of the reasons why the andragogy Bachelor program has been closed in Hungary in 2017.) The results are also informative to students and applicants of the higher education who want to be prepared for a finding job with good expectations. The research has been supported by the National Research, Development and Innovation Office—NKFIH, through the project OTKA—116674 (Process mining and deep learning in the natural sciences and process development) [^1]: Laszlo Gadar is employed by a commercial company named "Innopod Solutions Ltd, Budapest, Hungary", but this commercial affiliation does not alter his adherence to PLOS ONE policies on sharing data, materials and results.

# Introduction The *Dinophysis* genus is an ecologically important group of dinoflagellates. *Dinophysis* spp. play dual roles in the marine ecosystems: as primary (photosynthetic) and secondary (heterotrophic) producers. Furthermore, many *Dinophysis* species are known to produce potent polyether toxins. For instance, *D. caudata* and *D. miles* have formed blooms and caused diarrhetic shellfish poisoning through accumulation of toxins in the green mussel. Therefore, the genus *Dinophysis* is important in microbial food webs and for its potential influence on public health. In addition, *Dinophysis* spp. have peculiar and unique morphologies that are not shared by any organisms outside the class of Dinophysiales, making this genus an interesting subject of evolutionary studies. However, until recently their phylogenetic position among dinoflagellates and their ecology such as trophic modes have remained poorly understood in most species due to the paucity of cultures or tools to study wild populations. The genus *Dinophysis* has an obscure phylogenetic position among dinoflagellates. Using rRNA gene (rDNA) small subunit (SSU) and mitochondrial genes encoding cytochrome B (*cob*) and cytochrome C oxidase subunit I (*cox*1) and its mRNA editing patterns, a natural population of *D. acuminata* was placed phylogenetically between Gonyaulacales and Prorocentrales. Recently, a sister kinship to *Phalacoma* was established for the genus *Dinophysis*. Dinophysioids have diverse trophic modes; some species are heterotrophic feeding on other algae, whilst others have intracellular and extracellular cyanobionts and probably acquire carbon fixed by these symbionts. In *Histioneis* and *Ornithocercus*, the cyanobionts resides on the cingular lists,, whereas *Amphisolenia*, and *Sinophysis canaliculata* cells host the cyanobionts intracellularly. Typical *Dinophysis* spp. have been found to contain a plastid of cryptophyte origin,, in most cases *Teleaulax*-derived, although whether such uniformity in plastid acquisition is likely in other species and whether the plastids are kleptoplasts or permanent plastids have been debated. Hackett *et al*. (2003) detected plastid rDNA sequences of a cryptophyte and a rhodophyte in *D. acuminata* and attributed the former to plastid and the latter to prey. Meanwhile, *D. mitra* was found to harbor plastids of haptophyte origin. The recent success in culturing *D. acuminata* has greatly facilitated physiological, phylogenetic and molecular studies of the genus. However, because the number of *Dinophysis* cultures is currently limited, work on many species still relies on natural populations. Work on natural populations not only broadens the range of species to be studied, but also can reveal in situ status of physiology and gene expression. A population of *D. acuminata* was isolated via flow cytometer from Narragansett Bay that enabled both the detection of mitochondrial mRNA editing in this species and its phylogenetic position based on nuclear rDNA SSU. More phylogenetic studies have been conducted for natural populations from Florida embayments and Indian Ocean. rDNA LSU and SSU have been used to determine the relationship between the genera *Phalacroma* and *Dinophysis*, although their resolving power has yet to be demonstrated in some species in the *Dinophysis* genus. For instance, a study showed that rDNA LSU failed to distinguish *D. miles* from *D. tripos*, and *D. odiosa*. To date, hardly any studies have been dedicated to *D. miles*, and the plastid type of this species remains undocumented. *D. miles* is recognized as variant *D. miles* var. *schroeteri* in Southeast Asia and *D. miles* var. *indica* in Indo- West Pacific, the latter widely distributed in the northeast area of South China Sea, such as Hainan island and Nansha islands waters. In this study, we have investigated the phylogenetic position and plastid types of *D. mil*es var. *indica* from South China Sea. # Materials and Methods ## Sample collection A phytoplankton sample was collected at 18°11.5′N, 119°27′E (latitude, longitude) near Sanya in the South China Sea with a 55-µm mesh plankton net in March, 2010. The towed sample was transferred into a 500-mL plastic container and preserved with neutral Lugol's solution. The sample was stored in the laboratory in the dark until analysis (within 3 months). ## Microscopic observations and cell sorting Microscopic examination of the preserved phytoplankton sample revealed an abundant population of *D. miles*. The abundance of this species and other phytoplankton in the sample was determined using Sedgwick-Rafter chamber. Identification of the species was carried-out according to Steidinger (1997) and Wood (1963). The abundance of this species in the natural environment was estimated by adjusting the cell concentration in the retrieved sample to the volume of water filtered in the net tow. Morphocytological features were examined both under Lugol's staining and after Lugol's stain was removed. To remove Lugol's stain, a subsample was centrifuged and supernatant discarded. The cell pellet was rinsed with 0.45-µm filtered seawater, followed by treatment with 10% (weight/volumn) sodium thiosulfate. DNA was stained using SYBR Green I (35149A, Molecular probes, Invitrogen Corporation, Carlsbad, CA, USA) at 1∶10000 dilution at room temperature for 30 min. DNA and pigment fluorescence was observed under an Olympus BX51 epifluorescence microscope. From the original Lugol's-preserved samples, colonies consisting of eight *D. miles* cells were isolated under the inverted microscope. The isolated cells were rinsed carefully with 0.45-µm filtered seawater for subsequent DNA extraction. ## DNA extraction, PCR, and gene sequencing Four eight-cell *D. miles* colonies were resuspended in 0.5 mL DNA lysis buffer (0.1 M EDTA pH 8.0, 1% SDS, 200 µg mL⁻¹ proteinase K) and incubated for 48 hours at 55°C. DNA extraction followed a previously reported protocol. Briefly, after incubation, NaCl was added to achieve 0.7 M, and CTAB was added to the final concentration of 1.7%. The lysate was then extracted in chloroform. After centrifugation, the supernatant was removed and DNA further purified using Zymo DNA Clean and Concentrator kit (Zymo Research Corp., Orange, CA). At last, DNA was eluted in 32 µl Tris-HCl solution so that each µl contained DNA from about 1 cell of *D. miles*. Using 1 µl of the extracted DNA as the template, PCR reactions were carried out using a pair of dinoflagellate-specific rDNA SSU primers, a pair of rDNA primers extended from internal transcribed spacer (ITS) to LSU regions, a pair of *cob* primers, a pair of *cox*1 primers, and a pair of plastid rDNA SSU primers. The sequences of the primers were as shown in. PCR cycles consisted of one initial cycle of denaturation at 94°C for 3 min followed by 35 cycles of at 94°C for 30 sec, 56°C for 30 sec, and 72°C for 45 sec, followed by 10 min at 72°C for the final extension. PCR products were resolved on an agarose gel electrophoretically and the specific DNA band was excised. DNA was recovered and purified using a Zymo DNA column and sequenced directly using BigDye sequencing kit. For the plastid rDNA SSU, direct sequencing of the PCR product indicated the presence of different sequences. Therefore, the purified PCR product was ligated, cloned, and multiple clones were sequenced on both strands of the DNA. ## Phylogenetic analyses DNA sequences were trimmed of primers and the two strands were merged. The assembled sequences were analyzed using Basic Local Search Tool (BLAST) against databases in GenBank to determine what organisms these rDNA sequences represented. Sequences showing significant similarity in BLAST to the sequences obtained in this study were retrieved from the databases. Phylogenies based on partial SSU, ITS1-5.8S-ITS2, partial LSU (D1-D2, 700-bp;), *cob* (334-bp), and *cox*1 (840-bp) regions were used to investigate the phylogenetic position of *D. miles*. Phylogenetic trees were also inferred from plastid rDNA SSU to analyze the plastid type in *D. miles*. These datasets were separately aligned using ClustalX. The alignments were run through ModelTest to select the most appropriate evolutionary model. The selected General Time Reversible (GTR) model with gamma distribution was employed for Maximum Likelihood analysis using PhyML3.0 aLRT. Categories of substitution rates were set at 4, and other parameters were estimated based on the datasets. The proportion of invariable sites and gamma shape parameter were 0.464 and 0.583, respectively for the SSU dataset, 0.127 and 1.296 for ITS, 0.185 and 0.689 for LSU, 0.098 and 1.130 for *cob*, 0.000 and 0.725 for *cox*1, and 0.214 and 0.360 for plastid SSU. ## Nucleotide sequence accession numbers The sequences obtained in this study were deposited in GenBank under accession numbers JN982970-JN982975. # Results ## Microscopic observations Microscopic examination confirmed that the isolated cells were morphologically identical to *D. miles* var. *indica*. The cells had two posterior projections that extended from the end of the hypotheca, which are characteristic of *D. miles* and *D. tripos*. In contrast to *D. tripos*, our sorted cells had slim cell bodies and the dorsal process was longer than that of *D. tripos*, plus the ends of the processes were smooth, which is typical of *D. miles*. The angle between the two projections was about 70°, matching that of *D. miles* var. *indica*. The cell concentration ranged from 28 to 34 cells L⁻¹. The size of *D. miles* cell was about 16–21 µm in width and 140–165 µm in length. Most of the cells were found in eight-cell colonies except two-cell pairs in some cases. The eight cells formed a ring by attaching to each other at the end of the dorsal process of the cell, i.e. the process opposite to the sulcal list. In the cells of *D. miles* that were examined under the microscope, 5–10 plastids-like entities (n = 10) were observed, which showed dark staining of starch deposit by Lugol's solution, indicating plastids likely of cryptophyte origin. After removal of Lugol's stain followed by DNA staining using SYBR Green I, DNA fluorescence and pigment autofluorescence were apparent under the epifluorescence microscope. ## Phylogentic position of *D. miles* based on nuclear rDNA and mitochondrial *cob* and *cox1* We obtained the nuclear-encoded ribosomal RNA sequence 2,824-bp (JN982970) from the sorted cells, composed of the partial sequence of SSU, ITS1, 5.8S, ITS2, and the partial sequence of LSU (D1–D2). Within the 2.824-kb sequence, the dinoflagellate SSU region spanned 1.59 kb (nucleotide positions 1–1593), the ITS1-5.8S-ITS2 region (abbreviated as ITS hereafter) 0.59 kb (positions 1557–2146), and the LSU region 0.68 kb (positions 2147–2824). The phylogenetic tree of SSU, ITS and LSU included 32, 40 and 36 sequences, respectively from Genbank, in addition to the sequences obtained in this study. The topologies of these trees inferred from the three datasets using Neighbor Joining (NJ) and Maximum Likelihood (ML) were similar and indicated clear separation of well- supported four genera, *Phalacroma*, *Histioneis*, *Ornithocercus* and *Dinophysis*. In all three sets of trees, the genus of *Dinophysis* (such as *D. acuminata* and *D. acuta*) was distinct from other species. However, resolution of *D. miles* from other *Dinophysis* species varied among the three genes. In the LSU tree, the South China Sea *D. miles* was identical to a sequence reported for *D. miles* from the Indian Ocean (FJ808688), but appeared to be identical also to *D. tripos* (FJ808692, AY040585) and *D. odiosa* (AY259241). Thus LSU was unable to resolve the three species. In the SSU tree, *D. miles* could not be separated from *D. caudata* (EU780644) and *D. norvegica* (AF239261, AB073119, AJ506974). In contrast, ITS phylogeny placed *D. miles* as a distinct lineage, well separated from *D. caudata* (EU780642, EU780643, EU780644), *D. tripos* (AJ304806, EU927484, AY040585), and other *Dinophysis* species. LSU and ITS results combined verified the morphological identification of the sorted cells as *D. miles*. Based on all the three sets of trees, *D. miles* appeared to be closely related to *D. tripos* and *D. caudata*. The alignment of *cob* consisted of the *D. miles* sequence obtained (JN982971) in the present study and 55 sequences from other dinoflagellates available in GenBank. The 913-bp *cob* sequence from *D. miles* var. *indica* differed by only 3 bp (0.33%) from that of *D. acuminata* (EU130568), the only *Dinophysis cob* sequence available in GenBank. The *cox*1 sequence obtained from *D. miles* var. *indica* (JN982972, 840-bp) contained the widely used DNA barcode region (∼650-bp). It was aligned with 46 homologous sequences from other dinoflagellates available in GenBank. The *cox*1 sequences from *D. miles* var. *indica* differed by only 3 or 4 bp (0.36% or 0.48%) from counterparts of *D. ovum* (AM931583, GU452507, GU452508), and also only 3 bp (0.36%) from a *D. acuminata* sequence (EU130566, mRNA sequence is EU130565), and 0 bp or only 1 bp (0.24%) from *D. tripos* sequences (EU927473, EU927472). *Cob* and *cox*1 molecular phylogenies showed that *Dinophysis* species formed strongly supported lineages. ## Phylotypes of the plastid Sequencing results revealed three types of plastid SSU rDNA sequences (JN982973–JN982975) from colonies of *D. miles* var. *indica*. BLAST analyses of the 423-bp sequences indicated that they belonged to different lineages. One (JN982974) was 96% identical to the plastid SSU of the cryptophytes *Teleaulax amphioxeia* (AY453067) and *Plagioselmis* sp. TUC-2 (AB164407), one (JN982973) 98% identical to that of the haptophyte *Phaeocystis antarctica* (DQ442654) and the plastid SSU of *D. mitra* (AB199888), and the other (JN982975) 100% identical to that of an uncultured cyanobacterium (DQ431889) and 91% identical to that of the cyanobionts of *Dinophysis* sp. (AY918886). Phylogenetic analyses also showed that these *D. miles* var. *indica* sequences clustered with the plastid SSU of cryptophytes, haptophytes and cyanophytes, respectively. Of these, the cryptophytes-type clade comprises cryptophytes and the majority of photosynthetic *Dinophysis* species; the haptophyte-type clade consists of haptophytes and several populations of *D. mitra*; the rhodophyte-type clade contains rhodophytes and *D. acuminata*; the cyanophyte-type clade is composed of cyanobacteria and *Dinophysis* sp.. While *D. acuminata* is represented in two (cryptophyte and rhodophyte) clades, only *D. miles* var. *indica* covers three clades. # Discussion Analyzing natural populations of a dinoflagellate species alleviates the barrier of lack of cultures to study the species. The culture-independent approach also is the only way to gain understanding on physiological and molecular genetic characteristics in the natural populations. As the first study dedicated to *D. miles*, we have sequenced SSU and ITS in *D. miles* var. *indica*, and analyzed *Dinophysis* phylogenies based on nuclear SSU-ITS-LSU and mitochondrial *cob* and *cox*1 to compare their performance in distinguishing different species within this genus. The sequences obtained and the results of phylogenetic analyses will be useful for future phylogenetic and DNA barcoding studies for this and related species. Further, analysis of plastid SSU on the natural population of *D. miles* reveals multiple plastids (and cyanobionts) associated with this species, a finding that would be difficult to obtain using laboratory cultures. Therefore, taking advantage of culture-independent molecular techniques, research on natural populations of dinoflagellates has the potential of yielding more information. This potentially can be applied to other protists that are amenable to single cell (colony) isolation, which is becoming increasingly feasible with the aid of flow cytometry. However, working directly on natural populations of protists is challenging because it is often difficult to isolate the target species from the plankton assemblage and it is prone to contamination by co-existing organisms. In our study, *D. miles* is relatively large in cell size, and hence relatively easy to isolate. Careful washing and microscopic examination further minimized the chance of contamination by other phytoplankton. ## Comparison of phylogenies based on the three regions in the nuclear rDNA sequences and mitochondrial *cob* and *cox1* Morphological observations augmented by molecular analyses indicate that the *Dinophysis* population we detected was *D. miles* var. *indica*. Molecular phylogenies indicate that nuclear SSU, ITS, LSU rDNA and mitochondrial *cob* and *cox*1 all have sufficient resolving power to discriminate genera in Dinophysiales. Our results showed that among these gene regions, the ITS region offered the best resolution between *D. miles* and other *Dinophysis* species. The phylogenies of the nuclear rDNA regions showed varying interspecific distances in the genus of *Dinophysis*. LSU fails to differentiate the morphologically similar species *D. miles*, *D. tripos*, as well as the morphologically more distinct *D. odiosa*, and SSU could not distinguish *D. miles* from *D. norvegica* and *D. caudata*. Handy *et al*. (2009) indicated that the nuclear-encoded ITS1 and ITS2 have undergone higher evolutionary rate than LSU and SSU rDNA regions based on a comparison of percent identity among *Histioneis* sp., *Ornithocercus magnificus*, and *Dinophysis* spp. relative to *Phalacroma rapa*. In the *cob* phylogenic tree, *D. miles* is closely related to, but different from, *D. acuminata* among other dinoflagellates. The sequence we obtained embraced a 334-bp region, which has been demonstrated to be a promising DNA barcoding marker for dinoflagellate species. This gene sequence exhibit only three nucleotide difference between *D. miles* and *D. acuminata*, two of which are located within the 334-bp region. The separation of these two species is consistent with the result based on rRNA genes, but the overall resolving power of this gene for *Dinophysis* species remains to be determined in further studies with broader taxon sampling. In the *cox*1 phylogenic tree, *D. miles* is well resolved from *D. acuminata* and *D. ovum* although their distances were short. *D. miles* and *D. ovum* only differed by 3 or 4 bp (0.36% or 0.48%). *D. miles* differed from a previously reported *D. acuminata* sequence (EU130566) by 3 bp (0.36%) yet from another (AM931582) by 9 bp (1.07%). These two reported *D. acuminata cox*1 sequences showed a difference of 93 bp (7.74%), which is unprecedented and highly unlikely for any dinoflagellates. Raho *et al.* (2008) based on their sequence of *D. acuminata* (AM931582) concluded that the *cox*1 region had higher resolving power than ITS. Our results show that this is not the case, casting question on the accuracy of that reported sequence. Careful comparison of AM931582 with EU130566 and counterpart sequences from other *Dinophysis* species showed that the apparent variable sites in AM931582 were mostly in the 3′ end, suggesting possibility of sequencing errors toward the end of read length. Alternatively, host of the AM931582 might have been a totally unrelated organism misindentified as *D. acuminata*. Furthermore, previously reported *cox*1 sequence from *D. tripos* (EU927473) was identical to the *D. miles* sequence (JN982971) obtained in this study. Unlikely, this gene would separate the two species so well. Because ITS as a non-coding region has higher variability than the coding regions SSU and LSU, it is expected to have greater resolving power for all eukaryotes. The usefulness of ITS in resolving dinoflagellate species has been demonstrated. Consistent with these findings, our results also showed that the ITS region separated *D. miles* from *D. tripos*, *D. acuminata*, and other *Dinophysis* species with strong bootstrap support, indicating its greater resolving power for *D. miles* and related species. In contrast, as shown above, the SSU, LSU, and the two mitochondrial genes, overall show lower, albeit varying, levels of resolving power between *Dinophysis* species. Therefore, ITS1-5.8S-ITS2 region seems to be the most effective region to distinguish *D. miles* from other *Dinophysis* species among these five gene loci. In addition, based on all the current phylogenies inferred from the five gene loci, *D. miles* is closely related to *D. tripos* and *D. caudata* and more distant from *D. acuminata*. ## “Plastid” consortium in *D. miles* In this study, we retrieved three different types of plastid SSU rDNA sequences from *D. miles* var. *indica*. Based on the phylogenetic analyses of the plastid genes, two plastid sequences are of crytophyte and haptophyte origin, the third sequence is closely related to cyanobacterial SSU. These different plastid SSU sequences are unlikely to be a result of contamination. First, microscopic examination of our net tow samples showed predominance of diatoms (*Chaetoceros*, *Rhizosolenia* and other genera); any cryptophytes, haptophytes, or cyanobacteria cells present in the study ocean area would have been mostly lost through the 55-µm mesh during the net tow. Second, our picked cell colonies were extensively rinsed in filtered seawater before DNA extraction. Furthermore, cryptophyte and haptophyte plastids have both been demonstrated to be plastids in *Dinophysis* spp. and cyanobacteria have been reported to associate with some dinophysioids. Our microscopic observation on some of the cells we isolated revealed the intracellular plastid stained intensely with iodide, indicative of starch storage, and phycoerythrin-like fluorescence, indicating presence of cryptophyte type of plastid or cyanobacteria inside *D. miles* var. *indica* cells. Therefore, the *D. miles* var. *indica* population in the South China Sea likely possesses a consortium of plastids and cyanobionts previously documented separately in different dinophysioids species. One of the plastid SSU sequences retrieved in our study is most closely related to that in *Proteomonas sulcata*. One the one hand, this agrees with the previous results that most of the *Dinophysis* species contain plastids originated from cryptophytes,; on the other hand, this distinguishes *D. miles* from most of *Dinophysis* spp. which have plastids originating from a different cryptophyte. The second plastid SSU sequence found from *D. miles* var. *indica* is of haptophyte origin, similar to *D. mitra* from Okkirai Bay, Japan. Intriguingly, the *D. mitra* population harbors plastids of different haptophyte lineages, including those closely related to *Phaeocystis* and *Chrysochromulina*, respectively, suggesting that these are kleptoplastids retained from prey algae, in contrast to the more controversial status of cryptophyte-derived plastids in other *Dinophysis* species. The haptophyte-type plastid of *D. miles* var. *indica* is most closely related to plastids of *Phaeocystis antarctica*. Interestingly, Gast *et al*. (2007) showed that a haptophyte alga closely related to *Phaeocystis antarctica* was grazed by a dinoflagellate in the Ross Sea, Antarctica, and its plastid was retained in the dinoflagellate cell for temporary photosynthesis. This suggests that grazing and retention of haptophyte plastids by dinoflagellates occur in both polar and tropical waters, and are likely a widespread phenomenon in dinoflagellates. The third plastid-like SSU sequence from *D. miles* var. *indica* belongs to the lineage of cyanobacteria. While cyanobacteria have been shown to be endosymbionts of some dinophysioid species, most cyanobacterial associations are believed to behave as extracellular symbionts (cyanobionts). Cyanobionts occur in three genera of Dinophysiaceae, *Citharistes*, *Histioneis*, and *Ornithocercus* and our finding extends that to the genus of *Dinophysis*,. It was thought that the lists that develop from extended cingulum and sulcus provide a habitat for the cyanobionts in some dinophysioids,. *Histioneis* and *Ornithocercus* possess prominent lists on the epicone or cingulum for the ectophytic cyanobionts to reside in. It was postulated that in *Phalacroma* and *Dinophysis* both the cingular and sulcal lists are not so elaborate and as a result no cyanobionts occur on them. It is unclear if the cyanobionts detected in *D. miles* are endosymbiotic or ectosymbiotic. Our microscopic observations showed that *D. miles* cells had a well-developed anterior cingular list, sulcal list and rib systems , suggesting that it is suited for cyanobionts to inhabit. Handy *et al.* (2009) showed, based on SSU phylogeny, that *Histioneis* and *Ornithocercus* cluster together and both have cyanobionts; in contrast, *Dinophysis* and *Phalacroma* were separated from those two genera and did not have cyanobionts. However, in our nuclear SSU, ITS, and LSU phylogenetic trees, *Dinophysis*, *Histioneis*, and *Ornithocercus* consistently clustered together, and the clade was distinct from *Phalacroma*. *Citharistes* was not included in our analyses due to the unavailability of SSU and ITS sequences and its phylogenetic relationship with those lineages could not be confirmed. Nevertheless, our nuclear rDNA phylogenetic analysis results consistently show that *Dinophysis* as well as *Histoneis* and *Orthithocercus* can host cyanobionts. It is noteworthy that our detected cyanobacterial sequence is 91% identical to recently reported cyanobionts of *Dinophysis* sp. cells. Three plastid-types suggest a possibility that *D. miles* has cryptic species that acquire different types of plastids. They can also be indication that *Dinophysis* nutritional physiology is more complicated than currently understood. The cryptophyte-type plastid seems to be the most common among *Dinophysis* spp., although whether it is a permanent or temporary (kleptoplastid) plastid is still being debated. The only exception is in *D. mitra*, if verified by further research. The different type of cryptophytes found in *D. miles* var. indica suggest that the cryptophyte plastid is probably not a permanent and universal plastid for the genus of *Dinophysis*. The failure to detect plastid-maintaining gene transcripts in *D. acuminata* further supports the case for kleptoplastidy. The more variable and spotty presence of haptophyte (*D. mitra*, *D. miles*), rhodophyte (*D. acuminata*), and cyanobacteria (*Dinophysis* sp., *D. miles*) most likely indicate the availability and the selection (if any) in the environment by the different *Dinophysis* species. This remains a question that can be answered only by systematic investigation on *Dinophysis* species and their sympatric phytoplankton assemblages in the natural environments. Further studies are also needed to determine whether all these photosynthetic entities are present in every single *D. miles* cell in the population, and whether they all are functional for photosynthesis and benefit the growth of the *D. miles* var. *indica* population. We thank Dr. Huan Zhang and Ms. Yunyun Zhuang from the Department of Marine Sciences, University of Connecticut, for their technical assistance in our work, and the two anonymous reviewers for their valuable comments that led to significant improvement of the manuscript. [^1]: Conceived and designed the experiments: DQ S. Lin. Performed the experiments: DQ. Analyzed the data: DQ S. Lin. Contributed reagents/materials/analysis tools: LH S. Lin. Wrote the paper: DQ S. Lin S. Liu LH. [^2]: The authors have declared that no competing interests exist.

# Introduction Asthma is a common disease in childhood. Twin studies have demonstrated a large contribution of genetic factors to the development of asthma. While the cumulative effect of genetic factors may be large, the individual contribution of each factor may be limited. Recently much progress has been made in the field of asthma genetics with the introduction of the genome wide association studies (GWAS). However, these GWAS use general definitions of (doctors diagnosed) asthma, and the specific effect of many candidate genes in relation to the development from wheeze to asthma in young children still needs to be defined. Asthma is characterized by chronic airway inflammation and airway (hyper-) responsiveness. Although asthma starts with wheeze, not all wheezing children will develop asthma. It is assumed that at a young age a dysfunction of the maturating immune system at a young age caused by genetic predisposition in combination with environmental triggers, such as environmental tobacco smoke and bacterial infections, can lead to asthma. Several asthma candidate genes can be functionally implicated in asthma onset and development. Amongst these are pro- inflammatory genes (*IL4*, *IL5*, *IL8*, *IL13*, *IL33*, *TNFα*), anti- inflammatory genes (*IL10*, *CC16*), genes involved in airway remodelling (*ADAM33*, *PLAUR* and possibly *ORMDL3/GSDMB*), genes involved in the epithelial barrier function (*PCDH1* and possibly *ORMDL3/GSDMB*), genes involved in leukocyte (*ICAM1*) or eosinophil activity (*ORMDL3/GSDMB*) and genes involved in immune modulation (*IL4R*, *LTC4*, *IL1RL1*). The linkage of genetic variants in asthma candidate genes to progression from early wheeze to persistent wheeze into childhood asthma is expected to result into an increased insight into the pathophysiology of asthma. We therefore aimed to link genetic variants in various asthma candidate genes to progression of early wheeze to persistent wheeze and childhood asthma. Next, we aimed to replicate our findings in an independent birth cohort study. # Methods ## The ADEM study ### Study population of the ADEM study The Asthma DEtection and Monitoring (ADEM) study is a long-term case-control study executed in the Netherlands and is registered at clinicaltrial.gov (NCT 00422747). The aim of this study is to develop a non-invasive instrument for an early asthma diagnosis in children and to study aetiological factors in relation to the early development of asthma. A total of 202 children with recurrent wheeze (≥2 episodes during life according to the International Study of Asthma and Allergies in Childhood (ISAAC) questionnaire) and 50 healthy controls (random selection of children without wheezing episodes during life) at two to four years of age were included. Exclusion criteria were mental retardation, cardiac anomalies, congenital malformations, other diseases of the lungs/airways, Crohn’s disease or rheumatic arthritis, and the inability to perform lung function measurements or exhaled breath collection. For the current analysis, only the children with recurrent wheeze were included. ### Asthma classification in the ADEM study At the age of six years a classification (transient wheezer or true asthmatic) was assessed by an experienced paediatrician in the field of respiratory medicine. This classification was based on symptoms, lung function (reversibility to a β₂-agonist and bronchial hyper-responsiveness) and medication use. In addition to this clinical classification, the classification was assessed by a computer-calculated algorithm as described previously. Bronchial hyper-responsiveness (a 20% fall in forced expiratory volume in one second (FEV₁) induced by a provocative concentration of histamine \<2 mg/ml) and reversibility measurements (bronchodilator reversibility ≥9%) were performed according to the European Respiratory Society guideline. Mismatching cases between the clinical and the computer classification were re-evaluated by the same paediatrician, who was blinded to his previous assessment. ### DNA isolation and genotyping in the ADEM study Saliva was collected by Oragene DNA self-collection kits (Oragene, Ottowa, Canada). If children were unable to produce sufficient saliva, buccal cells were obtained. DNA was isolated according to the manufacturer’s protocol. Participants were genotyped for 30 single nucleotide polymorphisms (SNPs) in 16 genes ( Candidate genes and selected SNPs). Candidate SNPs were selected based on previous association with childhood asthma, and a minor allele frequency of at least 10% ([www.hapmap.org](http://www.hapmap.org/) phase I, II & III or literature). In total, 27 of the SNPs were analysed by using a mass-spectroscopy based, high-throughput Mass ARRAY iPLEX platform (Sequenom Inc., Hamburg, Germany) using three multiplex genotyping reactions. Sequences were evaluated by ProxSNP and PreXTEND software ([http:www.realsnp.com](http://www.realsnp.com)). Sequenom Assay Designer 3.1 software was used to create the different multiplexes. Genotyping was performed according to the iPLEX method. Primer and probe information is provided in S2_Table.docx Primer sequences used for genotyping by Sequenom. Three of the selected SNPs could not be fitted into the multiplex reactions (rs1805010, rs2243250, rs3741240), and therefore these SNPs were determined by Taqman genotyping assays ID C_2769554_10, ID C_16176216_10 and ID C_25473445_10 (Applied Biosystems, California, USA). ## The KOALA Birth Cohort Study ### Study population of the KOALA Birth Cohort Study The KOALA study (the Child, Parent and Health: Lifestyle and, Genetic Constitution study) is a prospective study in the Netherlands with the goal of investigating early life risk factors for atopy and asthma. Pregnant healthy women (2,343 women with conventional lifestyle and 491 women with alternative lifestyles) were enrolled at 34 weeks of gestation. Parents were asked to take a buccal swab sample from the child for DNA isolation. From the group of 1,656 children with DNA available, 43 were excluded for congenital conditions like Down’s syndrome and cystic fibrosis or missing baseline data. In the remaining group, follow-up on respiratory complaints until age four years was complete in 1,364 (85%) children. For the main analysis, only the children with recurrent wheeze (≥4 episodes in one questionnaire or at least one episode in multiple questionnaires until four years of age according to the ISAAC questionnaire) and a definitive classification (asthma or transient wheeze) at six to seven years of age were included (n = 248). ### Asthma diagnosis in the KOALA Birth Cohort Study Asthma at age six to seven years was defined as ever physician diagnosed asthma with clinical symptoms and/or the use of asthma medication in the last 12 months, adapted from the ISAAC questionnaire. Clinical symptoms were defined as having had at least 1 episode of wheeze or dyspnoea in the last 12 months. The use of asthma medication was defined as regular use (everyday use during at least 2 months or use associated with physical activity) of short-acting β₂-agonists or the use of inhaled corticosteroids, and medication use according to the Dutch guidelines of treatment of bronchial asthma in children. ### DNA isolation and genotyping in the KOALA Birth Cohort Study Parents were asked to collect buccal swabs from their children. Genomic DNA was extracted from these swabs by standard methods. DNA was amplified by using REPLI-g UltraFast technology (Qiagen, Hilden, Germany). Participants were genotyped for five SNPs in three genes that had demonstrated an association with an asthma classification at age six in the ADEM study. Genotyping was performed by Competitive Allele-Specific PCR by using KASPar genotyping chemistry, under contract by LGC Genomics (LGC, Teddington, UK) with extensive quality control as described previously. ## Ethics statement The ADEM study protocol was approved by the Dutch Central Committee on Research Involving Human Subjects. The KOALA study was approved by the Ethical Committee of the University Hospital of Maastricht. All parents signed informed consent. ## Statistical analysis IBM SPSS version 20 was used for data analysis (SPSS inc., Chicago IL, USA). Differences in baseline characteristics were evaluated by chi-square test for categorical variables and independent t-tests for continuous parametric variables. The Hardy-Weinberg equilibrium was calculated for each SNP based on a chi-square test and defined deviant in case of a p≥0.05. Furthermore, Linkage disequilibrium (LD) was calculated with Haploview to demonstrate the relationship between SNPs. No attempt was made to analyse the association of haplotypes. Logistic regression with outcome transient wheezer or true asthmatic was performed for each individual SNP. Models were adjusted for sex and exposure to parental smoking and furry pets assessed at time of asthma diagnosis on basis of previous literature and remained in the model irrespective of their statistical significance. Both, adjusted and unadjusted results were displayed in the tables. SNPs were tested according to a co-dominant model as this model has been shown to be the most powerful model over the additive, recessive and dominant model to detect associations when the inheritance model is not known. In case the genotype of the two variant alleles was present in \<10% of the population of the ADEM study, a dominant model was applied. In case a significant association was observed, all models (co-dominant, dominant and recessive) were applied if the genotype of the two variant alleles was present in ≥10%. SNPs that demonstrated an association of p\<0.10 with the asthma classification in the ADEM study were replicated in the KOALA study and were considered statistically significant when p\<0.05. Finally, for SNPs that demonstrated successful replication, we evaluated their association with early wheeze (in order to distinguish it from progression from wheeze to asthma at age four. This was done by logistic regression (adjusted for sex and exposure to parental smoking and furry pets) in the children from the KOALA study with complete follow-up until age four years, with recurrent wheeze as the outcome. ## Power calculation Due to the specific selection of wheezing children, we anticipated to find large ORs. A group of 146 transient wheezers and 73 asthmatics is sufficient to detect associations with an OR of 2.2 in allele frequency in a dominant model, when assuming the presence of one or more variant alleles of at least 20% in transient wheezers with a power of 0.80 and an alpha of 0.10. # Results ## The ADEM study ### Population characteristics In four children a diagnosis at six years of age could not be assessed due to personal constraints of the parents such as lack of time and interest, leaving 198 children in the current analysis. At the age of six years, 122 children were classified as ‘transient wheezer’ and 76 as ‘true asthmatic’. Characteristics are displayed in. At the age of six, atopy was higher in the asthmatic group compared to the transient wheeze group. ### Association of genetic variants with childhood asthma DNA extraction was successful for all children. All SNPs had a high call rate (92–100%, Candidate genes and selected SNPs). No deviation from Hardy-Weinberg equilibrium was observed (p≥0.05) with the exception of *IL1RL1* rs1861245 and *TLR9* rs5743836 ( Candidate genes and selected SNPs). Three LD blocks were identified (block 1: R² = 0.65 for *ADAM33* rs528557 and rs511898; block 2: R² = 0.56 for *IL4R* rs1805011 and rs1805015, R² = 0.45 for *IL4R* rs1805011 and rs1801275, R² = 0.69 for *IL4R* rs1805015 and rs1801275; block 3: R² = 0.13 for *TLR9* rs187084 and rs5743836). The TT-genotype of *ADAM33* rs511898 (p = 0.03), the CG/GG-genotype of *ADAM33* rs528557 (p = 0.08) and the TT-genotype of *ORMDL3/GSDMB* rs7216389 (p = 0.08) were negatively associated with childhood asthma. The CT/TT-genotype of *IL4* rs2070874 (p = 0.07) and the CT/TT-genotype of rs2243250 (p = 0.06) were positively associated with childhood asthma. For *ADAM33* rs511898 and *ORMDL3*/*GSDMB* rs7216389 results of the recessive and dominant model are presented in results of the additional model analysis of significant genetic variants in the ADEM study. For *ADAM33* rs528557, *IL4* rs2070874 and *IL4* rs2243250 no alternative models were calculated as the genotype of the two variant alleles was present in \<10% of the population. None of the other tested genetic variants demonstrated an association with childhood asthma ( results for analysis of genetic variants in the ADEM study). Odds ratios with 95% confidence intervals (horizontal bars) from logistic regression analysis for both the ADEM study and the KOALA study adjusted for sex, exposure to parental smoking and furry pets for those SNPs that demonstrated a significant association with asthma based on a p\<0.10 in the ADEM study. Abbreviations: ref; reference category. ## The KOALA Birth Cohort Study ### Population characteristics At four years of age, for 1,364 children DNA was available and wheeze classification known (recurrent wheeze versus no recurrent wheeze). Children could only be defined as a child without recurrent wheeze in case all questionnaires until the age of four years were available (n = 1,079). Of the children with recurrent wheeze (n = 285), a definitive classification (asthma or transient wheeze) at the age of six to seven years could not be assessed in 37 children due to missing data. Consequently, a definitive classification (asthma or transient wheeze) was available in 248 children with recurrent wheeze (191 children with transient wheeze and 57 children with asthma). Eczema was significantly more frequent and exposure to furry pets was significantly less frequent in asthmatics compared to transient wheezers at six years of age. ### Replication of associated genetic variants with childhood asthma All SNPs had a high call-rate (93–96%). No deviation from Hardy-Weinberg equilibrium was observed (p≥0.05). Furthermore, LD was calculated (R² = 0.63 for *ADAM33* rs511898 and rs528557, R² = 0.00 for *IL4* rs2070874 and rs2243250). In the children with recurrent wheeze the CG/GG- genotype of *ADAM33* rs528557 was significantly negatively associated with subsequent childhood asthma when compared to the asthma group (OR (95%CI): 0.50 (0.26–0.97) p = 0.04,). No alternative models were displayed as the genotype of the two variant alleles was present in \<10% of the population. When the analysis was restricted to participants with a conventional lifestyle (n = 199), the association did not change (OR (95%CI): 0.46 (0.22–0.94), p = 0.03). *ADAM33* rs511898, *ORMDL3/GSDMB* rs7216389, *IL4* rs2070874 and *IL4* rs2243250 polymorphisms were not associated with childhood asthma in the KOALA study at a 0.05 significance level. ### Association between ADAM33 rs528557 and recurrent wheeze at age four For the replicated gene variant *ADAM33* rs528557 we assessed the role in the presence of recurrent wheeze in the 1,364 children with complete follow-up until age four years, 285 children with recurrent wheeze and 1,079 without recurrent wheeze. The CG/GG-genotype of *ADAM33* rs528557 was not associated with recurrent wheeze at age four (OR (95%CI): 1.03 (0.77–1.39), p = 0.82). # Discussion Multiple genetic variants in asthma candidate genes were assessed to determine their relationship with the presence of asthma at six years of age in a group of 202 children with recurrent wheeze at preschool age. We demonstrated association of *ADAM33*, *IL4* and *ORMDL3/GSDMB* gene polymorphisms with childhood asthma. In an independent birth cohort study we were able to replicate the significant negative association of the CG/GG-genotype of *ADAM33* rs528557 with childhood asthma at age six. Since no association was found for this SNP when assessing wheeze at age four, we demonstrated that *ADAM33* is mainly involved in the progression of wheeze into childhood asthma rather than being involved in the presence of recurrent wheeze. The other associations could not be replicated. *ADAM33* was first reported as a susceptibility gene for asthma and bronchial hyper-responsiveness through genome wide linkage analysis identifying a candidate region on chromosome 20p. *ADAM33* consists of 22 exons which can generate a protein with eight different functional domains. The gene is highly polymorphic, containing more than 70 SNPs with extensive LD. Some of the disease-related SNPs encode amino acid changes. Other SNPs are located in the non-coding regions and may affect proliferation of (myo)fibroblasts and smooth muscle and/or inflammation of the airways by alternative splicing and splicing efficiency of messenger RNA turnover or their association is based on linkage with other SNPs.\[,–\] In previous studies, polymorphisms in *ADAM33* have been related to childhood asthma and impaired lung function in early life in one study. In the present study we found a negative association of the *ADAM33* rs528557 and rs511898 polymorphism with progression of wheeze into childhood asthma. The replication of the negative association of the CG/GG- genotype of rs528557 with childhood asthma in an independent birth cohort study (KOALA study) confirms the relationship of this gene with childhood asthma in (Caucasian) children with recurrent wheeze. Contrary to our findings, the study by van Eerdewegh et al. revealed a positive association of this SNP with asthma in a genome wide scan in 460 Caucasian asthma affected sib-pair families. As the rs528557 polymorphism does not lead to an amino acid change, this conflicting observation might be based on variability of LD between the studied populations. Another explanation might be the pre-selection we applied on children with preschool wheeze resulting in a different stage of asthma development. This was further emphasised by our demonstration that the *ADAM33* rs528557 CG/GG-genotype was not associated with recurrent wheeze at four years of age. Consequently, this SNP is associated with the progression of wheeze into childhood asthma rather than with recurrent preschool wheeze. *ORMDL3/GSDMB* is mapped to a locus on chromosome 17q12–21, which was first identified in association with childhood asthma through genome-wide analysis. Since then, other studies have confirmed an association of *ORMDL3/GSDMB* with childhood asthma, making it the strongest replicated gene for childhood asthma.\[–\] Its function is still unknown, but it has been suggested that it might have a role in airway remodelling, the epithelial barrier function, or eosinophil trafficking.\[–\] In the present study, the CC- genotype of rs7216389 in *ORMDL3/GSDMB* was demonstrated to be borderline significantly associated with childhood asthma. However, replication failed in the independent birth cohort study. In contrast to our findings, previous studies identified the T-allele of rs7216389 in *ORMDL3/GSDMB* as the childhood asthma risk allele. Exposure to environmental factors such as tobacco smoke and domestic furry pets, have been demonstrated to modify the relationship between polymorphisms in *ORMDL3/GSDMB* and childhood asthma. Remarkable was the difference in prevalence of parental smoking between the ADEM and the KOALA study even though the same definition for passive smoking was used. This is probably due to the different recruitment strategies of the studies. In our analysis correction for sex, exposure of parental smoking and furry pets did not influence our findings. However, we did not assess effect modification in the current study. Furthermore, it might be that unknown environmental influences affect the relationship of *ORMDL3/GSDMB* with asthma, which may explain the difference between our findings and those of others. Moreover, the small sample size of our study might have led to spurious findings due to limited power. *IL4* maps to a cytokine cluster on chromosome 5q31-q33.\[–\] It is a pro- inflammatory cytokine that is involved in a number of immunoregulatory pathways such as the induction of IgE synthesis by B-cells and differentiation of T-helper-type-2 lymphocytes.\[–\] *IL4* has been linked to asthma phenotypes and atopy in several studies including childhood populations. In accordance with these findings, we found associations of two SNPs in *IL4* with childhood asthma. This strengthens the suggestion, brought out by previous studies, that children with CT/TT-genotypes for rs2070874 and rs2243250 run an increased risk of developing asthma. Unfortunately, we were unable to replicate these findings in the independent birth cohort study, possibly due to low power. As seen in the present study, many asthma candidate gene studies are confronted with failure of replication or even opposite findings in independent studies. There are multiple causes for failure of replication. Firstly, asthma is caused by different polymorphisms that do not need to be universal, leading to genetic heterogeneity. In addition, the gene effects are small and they may be subject to ethnic diversity and variability in LD between populations, which can lead to population specific results. However, the populations of the ADEM study and the KOALA study have, based on region, ethnicity, and genetic origin, similar population characteristics. Furthermore, methodological differences between the studies can influence findings. For example, the definition of asthma used, varies between studies, resulting in different phenotypes. As the asthma classification in the KOALA study differed from the ADEM study, this might be the cause of replication failure in four of the five childhood asthma associated genetic variants. In the ADEM study, the definition of asthma was based on symptoms, lung function features and asthma medication with a high rate of concordance between a doctor diagnosis of asthma and the diagnosis by means of a computer algorithm. In the KOALA study the diagnosis of asthma was based on clinical symptoms and the use of asthma medication which is a universal accepted definition in this type of studies. In addition, it is generally known that environmental influences can affect or even change the direction of underlying associations. Therefore, different environmental conditions between studied populations can be responsible for failure of replication. In our analysis, we corrected for passive smoking and exposure to furry pets. We cannot fully exclude an influence by other (unknown) environmental factors. Moreover, we concentrated on candidate polymorphisms instead of a complete assessment of all genetic variants due to power restrictions. Although no complete coverage of the variation of each candidate gene was guaranteed as would be the case for tagSNPs, the selected SNPs have previously been proven to be important in association with childhood asthma. Consequently, testing these specific hypotheses, the a priori chance of finding contributing SNPs is large. Thus, we choose not to correct for multiple testing these specific hypotheses. Another reason to ignore multiple testing correction is replication in an independent study which reduces the chance of finding spurious associations. Finally, observed associations can also be caused solely by LD. Naturally, a significant finding based on chance can also be a cause of replication failure. In contrast to mentioned limitations, our study has several strengths. The design of the ADEM study enabled us to follow a large group of children with recurrent wheeze at preschool age until the asthma classification at six years of age. Furthermore, our definition of asthma was based on a clinical assessment and a computer-algorithm with re-assessment of inconclusive cases. This is expected to result in a highly accurate classification. Furthermore, replication of found associations was assessed in the independent KOALA study. A limitation of our study might be that we did not correct for multiple testing. However, the use of an independent birth cohort for replication reduced the likelihood of finding associations based on chance. # Conclusions In conclusion, we assessed 30 genetic variants in 16 asthma candidate genes in relationship to childhood asthma in a cohort of 202 children with recurrent wheeze. Polymorphisms in *ADAM33* (rs511898 and rs528557) and *ORMDL3/GSDMB* (rs7216389) were negatively associated and polymorphisms in *IL4* (rs2070874 and rs2243250) were positively associated with childhood asthma. In an independent birth cohort we were able to confirm the negative association of *ADAM33* rs528557 CG/GG-genotype with progression of recurrent wheeze into childhood asthma rather than with the presence of wheeze. # Supporting Information We thank; Patrick van Gorp for his skilled assistance with the laboratory procedures; medical students Imke Duijf, Kiki Vangangelt, Nedim Dzino, Brenda Thönissen, Esther Kalicharan and Emily Cohen for their outstanding assistance during the measurements; and all parents and children who participated in our study. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: ED QJ CS GK. Performed the experiments: EK KK GE MM. Analyzed the data: JP EK CT. Contributed reagents/materials/analysis tools: EK GE GK JP MM. Wrote the paper: EK JP QJ KK CT MM CS GE GK ED.

# Introduction A peripartum hysterectomy (PH) is a procedure performed at the time of delivery or in the immediate postpartum period as a life-saving measure in response to severe postpartum hemorrhage that does not respond to any other interventions. However, PH is one of the most severe complications in obstetrics and results in significant maternal mortality and morbidity. Moreover, PH results in a permanent loss of future childbearing opportunities. Its reported rate varies from 0.24 per 1000 deliveries to 5.09 per 1000 deliveries. Moreover, the results from studies that evaluated trends in the number of peripartum hysterectomies performed over time are mixed,. These discrepancies between studies may be due to the fact that these studies were conducted in single institutions and the sample sizes were therefore small. Furthermore, these data are not appropriate to obtain reliable nationwide rate estimates. In fact, only a handful of studies have examined PH rates in total populations of a nation or region. An alternative to PH to manage postpartum hemorrhage is selective uterine artery embolization (UAE). UAE is a reliable, safe, and minimally invasive procedure that has consistently been demonstrated to have success rates of over 90% with regard to achieving hemostasis. One of the major advantages of UAE in the treatment of postpartum hemorrhage is its potential to avoid hysterectomy and thereby preserve a woman's future childbearing options. However, few studies have evaluated trends in the UAE rate and its association with the PH rate. Our aims in this study were to describe nationwide trends in PH and UAE in Korea and identify the risk factors of PH. # Materials and Methods Data were collected from the Korea National Health Insurance Claims Database of the Health Insurance Review & Assessment Service (HIRA) for the period 2005 to 2008. Under the National Health Insurance System, Koreans are entitled to medical coverage as either an employee or a member of a community. Healthcare providers are required under the health insurance policies of HIRA to provide a review of the medical costs incurred. Accordingly, the HIRA database contains information on claims for approximately 50 million Koreans. The study protocol was approved by the institutional review boards of the Health Insurance Review & Assessment Service (IRB No. HIRA-1587)(10/November/2010). All data were de- identified by HIRA. The diagnosis and procedure codes from the International Classification of Diseases, 10th revision, were used to identify all women who gave birth during the study period and women who underwent PH or UAE. Cases of PH included women who underwent a vaginal or cesarean delivery in combination with an abdominal hysterectomy (either a total or subtotal abdominal hysterectomy). Cases of UAE were defined as UAE and delivery occurring during the same hospitalization event. Women with a concomitant diagnosis of a malignancy were excluded from the analysis. To identify the risk factors for PH and UAE, demographic characteristics, namely age, multiple pregnancy (defined as twins or higher-order gestations), parity (primiparous or multiparous), placenta previa, and obstetric procedures used for delivery (cesarean delivery, instrumental delivery, induction of labor) were obtained. The rates of PH and UAE were calculated per 1000 deliveries. The rates of several factors related to PH were calculated per 1000 deliveries. The rates of PH and UAE related to each factor were calculated per 1000 deliveries. Data from each year were evaluated individually and then compared to identify emerging trends. Trends over time were assessed by entering year as a single term with equally spaced category scores into logistic regression models. Multivariate logistic regression analysis was carried out with PH and UAE as the final outcome. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated using logistic regression. A *P*-value of less than 0.05 was considered statistically significant. Statistical analyses were performed using SPSS software, version 12.0 (SPSS Inc., Chicago, IL, USA). # Results For the study period of 2005–2008, a total of 1785,178 deliveries were recorded and a total of 2636 PH were performed in Korea, corresponding to a rate of 1.48 per 1000 deliveries. A total of 1239 UAE were performed, translating to a rate of 0.69 per 1000 deliveries. shows trends in PH and UAE rates during the study period. The PH rate decreased slightly but significantly from 1.57 per 1000 deliveries in 2005 to 1.33 per 1000 deliveries in 2008 (*P*\<0.001). There was a noticeable and significant increase in the UAE rate from 0.38 per 1000 deliveries in 2005 to 0.98 per 1000 deliveries in 2008 (*P*\<0.001). The case numbers and rates of risk factors for PH and UAE and their trends are shown in. The rates of cesarean delivery, multiparity, and instrumental delivery decreased during the study period. However, the rates of multiple pregnancy, induction of labor, and placenta previa increased. shows case numbers and rates of PH for each risk factor and their trends during the study period. The rate of PH for instrumental delivery increased, but the rate of PH for other factors did not change significantly during the study period. The rate of UAE for each risk factor increased during study period although this was not statistically significant. Multivariate-adjusted ORs for PH are shown in. Age, multiparity, multiple pregnancy, cesarean delivery, induction of labor, instrumental delivery, and placental previa were associated with an increased risk of PH. The highest risk was noted in women with placenta previa (OR 22.710, 95% CI 20.547%–25.100%). Mutlitvariate-adjusted ORs for UAE are shown in. Age, multiple pregnancy, cesarean delivery, induction of labor, instrumental delivery, and placental previa were associated with an increased risk of UAE, but multiparity was associated with a decreased risk of UAE. # Discussion To the best of our knowledge, this study is the first to report nationwide trends in PH and UAE rates. The PH rate decreased substantially from 2005 to 2008 in Korea. Although the reason for this downward trend is unclear, there are several possible explanations. Multiple factors are likely to affect the trend in the PH rate, and the observed trend may reflect complex changes in the rates of risk factors. In this study, significant risk factors for PH were age, cesarean delivery, multiparity, multiple pregnancy, induction of delivery, instrumental delivery, and placenta previa, consistent with the results from previous studies,. However, trends in the rates of individual risk factors varied during the study period. The rate of multiparity decreased and the rate of cesarean delivery, which is now cited by the majority of modern reviews as the major risk factor for PH, and which was the most common factor for PH among various risk factor in this study, also decreased in contrast to the trend reported by other studies. However, in cases of abnormal placentation such as placenta previa, which was a significant factor for PH with the highest OR in this study, consistent with other studies, its rate increased without significant changes in the PH rate. Rates of other risk factors showed increases or decreases in our study, but their effects on trends in the PH rate were likely minimal because of the small number of cases. Therefore, the downward trend in PH rate observed in our study might reflect changes in the rates of the risk factors described above, especially the decreased rate of cesarean delivery. Furthermore, increased use of UAE may also lead to a decrease in the PH rate, as UAE has been demonstrated to have a success rate of over 90% for the treatment of postpartum hemorrhage. We hypothesized that given the same indications; UAE would be preferred over PH, resulting in a decrease in the PH rate and an increase in the UAE rate over time. In particular, the PH rate for each risk factor should decrease over time. It is interesting to note that even though the rate of PH decreased slightly, the PH rate for each risk factor did not change significantly despite the significant and dramatic increase in the UAE rate. These results indicate that the decrease in PH rate demonstrated may be due to a decreased rate of significant risk factors for PH including cesarean delivery, rather than increased use of UAE. The immediate availability of UAE remains a challenge, especially in community hospitals in rural areas or smaller community hospitals, because UAE procedures are performed by specially trained interventional radiologists and require a well-equipped radiology suite. Otherwise, in this study, the If the resources are available, it is likely that UAE will increasingly be used with a low threshold for a prompt aggressive response. It may partially explain the results that the rate of UAE for each risk factor increased although this was not statistically significant and the UAE rate increased sharply but the PH rate decreased slightly. Moreover, UAE is usually performed when the mother is hemodynamically stable. For example, if there is a massive hemorrhage during caesarean section the patient would be considered hemodynamically unstable and would not be suitable for UAE since UAE procedures require some time for preparation; therefore, in some obstetric situations with the potential for massive hemorrhage, including placenta percreta, planned PH is the preferred delivery strategy. Therefore, although the UAE rate increased rapidly during study period, its effects on the decrease in the PH rate were minimal. However, decreases in the PH rate caused by other factors not examined in this study may have contributed to the observed PH rate. Further studies are required to evaluate the exact effects of UAE on PH rate. Studies of trends have provided a mixed picture with increase (8,10), decrease (7,9), or no change (4).In our study, the trend in the PH rate was evaluated only for the most recent 4 years. Moreover, our data collection methods were different to those used in previous studies. Therefore, direct comparison of our results with those of previous studies is not possible. In our study, the overall PH rate was 1.48 per 1000 deliveries, comparable with some studies, but higher than others. This may be due to the high cesarean section rate (36.45%) demonstrated in this study, as the rate of cesarean sections is tightly linked to the PH rate. The high PH rate in our study may also be due to our study design, such as our definition of the time period for PH. Several limitations should be kept in mind when interpreting our findings. First, this study was based on insurance claim data in the Korea National Health Insurance Claims Database, which is a database that was designed for cost claim issues, not research. Thus, the main limitation remains the validity of the data in this database. However, KNHI data has been validated in a previous study. Another limitation of our study is that we were not able to identify the standard hemodynamic parameters or the indications for PH and UAE from each hospital, and the specific individual characteristics of maternities, because the data is based on insurance claims and we did not perform comprehensive chart reviews. Limitations of this study also include the short study period (years 2005–2008), the data collected from the Korea National Health Insurance Claims Database of the Health Insurance Review & Assessment Service (HIRA) are strictly regulated since 2005 in terms of accuracy and reliability hence we decided to use data from 2005. Secondly, since the data were obtained upon claims, at least one year was required to receive the data for year 2008. Therefore, at the time we designed our study, data up to year 2008 was the most recent available and accurate data. Nevertheless, we included all deliveries and PH procedures performed in Korea in our study. Therefore, our results are unlikely to have been influenced by the type of hospital or the characteristics of the individual patients and physicians. The multi-center nature of our study can explain the discrepancy between our study and another study that reported a high rate of PH in Korea (3.25 per 1000 deliveries for Korean women), as this latter study was performed in a single referral hospital, which is more likely to handle a greater proportion of complicated deliveries. In conclusion, during the period from 2005–2008 in Korea, the PH rate decreased slightly, but the UAE rate increased sharply. Further studies are needed to evaluate the long-term trends in the PH rate and the effects of UAE on the PH rate. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: GJC HRH LYK CEL SCH MJO HJK. Performed the experiments: GJC HRH LYK CEL SCH MJO HJK. Analyzed the data: GJC HRH LYK CEL SCH MJO HJK. Contributed reagents/materials/analysis tools: GJC HRH LYK CEL SCH MJO HJK. Wrote the paper: GJC HRH LYK.

# Introduction Bacteremia is the presence of viable bacteria in the bloodstream and is the consequence of several clinical conditions, such as trauma, burn injury, abdominal surgery, and catheterization. The spread of bacteria to the bloodstream leads to a hyperactive inflammatory immune response and subsequent production of excessive inflammatory cytokines, resulting in a systemic inflammatory response syndrome and multiple organ dysfunctions. *Klebsiella pneumoniae* (*K. pneumoniae*) is a facultative anaerobic gram-negative *bacilli* bacterium and, after *Escherichia coli*, is the second most common cause of community- and hospital-acquired bacteria. Incidence and mortality rates associated with bacteremia are 7.1 in 100,000 per year and 1.3 in 100,000 per year, respectively. Nutritional support is important in the management of patients with bacteremia. Previous studies have shown that glutamine treatment decreases the incidence of gram-negative bacteremia and a choline-rich diet improves the survival from endotoxin shock in a rat model. Bacteremia is expected to generate measurable changes in metabolic levels. Therefore it is possible to monitor dynamic metabolic changes associated with bacteremia and identify metabolites related to the event. Developing an in-depth and systematic study of changes associated with bacteremia could provide a comprehensive view on the host metabolic response to bacteremia and open a window for nutritional intervention against the disease. Metabonomics involves multivariate statistical analyses on spectroscopic fingerprints of biofluids generated from nuclear magnetic resonance (NMR) spectroscopy or mass spectrometry. This is an emerging field of post-genomic science, which has been established as an extremely powerful analytical tool and has widespread applications in diverse research areas including genetics, toxicology, metabolic regulation, and infectious diseases. In the current study, we employ ¹H NMR spectroscopy in conjunction with multivariate data analysis to investigate metabolic changes in response to *K. pneumoniae in vivo.* The aim of the investigation is to uncover the mechanisms of *K. pneumoniae* infection at the metabolic level and to exploit the potential of metabonomics as a guidance tool for the management of bacteremia, which could be important for the improvement of disease survival. # Materials and Methods ## Bacteria *K. pneumoniae*, isolated from mesenteric lymph node in rat that suffered from intestinal ischemia and reperfusion injury, was cultured with Luria-Bertani broth (Oxoid Limited, Basingstoke, Hampshire, England) for 16 h to stationary phase, producing a concentration of 4×10¹⁰ colony forming units per mL (CFU/mL). Counting of bacteria was conducted by culturing diluted bacteria on Luria-Bertani agar plates and colonies were counted after 24 hours. Bacterial suspensions were centrifuged at 6000g for 10 min, washed twice and re-suspended in sterile saline solution for infection experiments. ## Chemicals Sodium chloride, K₂HPO₄·3H₂O, and NaH₂PO₄·2H₂O (analytical grade) were obtained from Guoyao Chemical Co. Ltd. (Shanghai, China). Sodium 3-trimethylsilyl \[2,2,3,3-d₄\] propionate (TSP-d₄) and D₂O (99.9% in D) were purchased from Cambridge Isotope Laboratories (Miami, FL). ## Ethics Statement Animal experimental procedures were performed according to the National Guidelines for Experimental Animal Welfare (Ministry of Science and Technology of People’s Republic of China, 2006) and approved by the Animal Welfare Committee of Wuhan Institute of Physics and Mathematics, Chinese Academy of Sciences, with permission from China Hubei Provincial Science and Technology Department. All surgery was performed under isoflurane anesthesia, and all efforts were made to minimize suffering. ## Animal Experiments and Sample Collection All animals used in this investigation are female Sprague Dawley (SD) rats (120–150 g, 5 weeks old, No. hnaslkj20101332) that were purchased from Hunan Slac Jingda Laboratory Animal Co. Ltd. (Changsha, China), and housed in groups of four at a certified local animal experimental laboratory (No. 00018445) with a 12 h light/dark cycle at a constant temperature of 23±1°C. Animals were allowed to have access to food and water *ad libitum*. A preliminary experiment was conducted to certify the dosage and duration of infection; the result suggested that 0.3 mL of 4×10¹⁰ CFU/mL *K. pneumoniae* was the maximum level that could be intravenously injected without causing mortality. In order to follow the infection and recovery processes clinically, 24 SD rats were injected with 0.3 mL of *K. pneumoniae* (4×10¹⁰ CFU/mL) via the tail and 4 rats were sacrificed at each of the following time points: 4 h, 8 h, 1 day, 2 day, 3 day and 7 day postinfection. Another 8 rats were kept as controls and injected with 0.3 mL of saline solution; they were sacrificed at 4 h after injection. A total of 0.5 mL of whole blood was collected and cultured to measure bacterial burden, plasma samples were also collected in tubes containing ethylene diamine tetra-acetic acid for a white blood cell count as well as C-reactive protein and procalcitonin assays. A separate animal experiment was conducted for the metabonomics investigation. A total of 24 SD rats were randomly divided into two groups after two weeks of acclimatization. They were subjected to treatments for 14 days: a control group (n = 12) and infection group (n = 12) were intravenously injected with 0.3 ml of sterile saline (0.9% sodium chloride) and 0.3 ml of *K. pneumoniae* (4×10¹⁰ CFU/mL) via the tail respectively. Blood and urine samples were collected at 9 time points: before the injection (hour 0), and at 4 h, 8 h, 24 h, 48 h, 3 d, 7 d, 10 d and 14 d postinfection. Urine samples were collected by placing rats individually into empty cages covered with a disposable plastic wrap. Urine was immediately transferred into 1.5 mL Eppendorf tubes, and snapped frozen in liquid nitrogen as soon as rats released a few drops of urine. Between 50 and 60 µL of blood was collected into 0.5 mL Eppendorf tubes containing 10 µL sodium heparin from the tail of the rats by cutting off its tip. Plasma was obtained by centrifugation (Microcentrifuge Hettich MIKRO22 Zentrifugen, Germany) at 4000 g for 10 min. The plasma was then transferred into 0.5 mL Eppendorf tubes, and snapped frozen in liquid nitrogen. Plasma and urine samples were stored in a freezer at −80°C for later analysis. At the end of experimental period (day 15), all animals were sacrificed by cervical dislocation under isoflurane anesthesia after 12 h fasting. No further samples were collected. ## Sample Preparation for NMR Spectroscopy Plasma samples were prepared by mixing 30 µL plasma with 30 µL saline solution containing 100% D₂O for the magnetic field lock and the 60 µL sample was transferred into 1.7 mm micro NMR tubes. ¹H NMR spectra of plasma were recorded at 298 K on a Bruker Avance II 500 MHz NMR spectrometer (Bruker, Germany), equipped with a Bruker 5 mm BBI probe with inverse detection, operating at 500.13 MHz proton frequency. A one-dimensional ¹H NMR spectra with water presaturation were acquired with Carr-Purcell-Meiboom-Gill (CPMG) pulse sequence \[recycle delay −90°-(τ-180°-τ)_n-acquisition\] to attenuate NMR signals from macromolecules. A total transverse relaxation delay (2nτ) of 70 ms was used. 90° pulse was set to about 10.0 µs and 256 transients were collected into 32 K data points for each spectrum with a spectral width of 20 ppm. An anomeric proton signal of α-glucose (δ 5.233) was used as a chemical shift reference. A total of 550 µL urine sample was mixed with 55 µL phosphate buffer (K₂HPO₄/NaH₂PO₄, 1.5 M, pH 7.4, 100% D₂O) containing 0.05% TSP-d₄ for chemical shift calibration and 0.1% of NaN₃ for prevention of bacterial contamination. After centrifugation at 12000 *g* for 10 min, the supernatant was transferred into 5 mm NMR tubes for NMR analysis. ¹H NMR spectra of urine were acquired at 298 K on a Bruker Avance 600 MHz NMR spectrometer equipped with a 5 mm TCI cryogenic probe, with inverse detection using a water presaturation pulse sequence \[recycle delay-90°-t₁-90°-t_m-90°-acquisition\]. The recycle delay was set to 2 s, t₁ to 3 µs and mixing time (t_m) to 80 ms. A total of 64 transients for urine spectra were collected. The spectra were referenced to TSP-d₄ at δ 0.00. For spectral assignment purposes, a series of two-dimensional NMR spectra were acquired on selected plasma and urine samples, which include ¹H-¹H correlation spectroscopy, ¹H-¹H total correlation spectroscopy, ¹H-¹³C heteronuclear single quantum correlation spectroscopy, and ¹H-¹³C heteronuclear multiple bond correlation spectroscopy. The standard parameters used for these spectral acquisitions have previously been reported. ## NMR Data Processing and Multivariate Data Analysis All free induction decays were multiplied by an exponential function with a 1 Hz line broadening factor prior to Fourier transformation and all the ¹H NMR spectra were corrected manually for phase and baseline distortions. The spectral region δ 0.5–9.5 was integrated into regions with an equal width of 0.004 ppm (2 Hz) using an AMIX software package (V2.1, Bruker Biospin, Germany). Regions distorted by imperfect water saturation were discarded together with the regions containing urea signals. These regions are δ 4.5–5.0 for plasma and δ 4.4–6.2 for urine. Each bucketed region was then normalized by probabilistic quotient normalization prior to statistical data analysis. Multivariate data analysis was carried out with the SIMCA-P⁺ software (version 11.0, Umetrics, Sweden). Principal component analysis (PCA) was initially carried out on mean-centered NMR data to generate an overview. Projection to latent structure with discriminant analysis (PLS-DA) and orthogonal projection to latent structure with discriminant analysis (O-PLS-DA) were subsequently conducted with the data scaled to unit variance. The quality of the models was assessed by model parameters; Q², indicated the predictability of the model and R² denoted the interpretability of the model. A 7-fold cross-validation method, permutation test and ANOVA of the cross-validated residuals (CV-ANOVA) test were used to validate the models. The loadings that indicated altered metabolites after the infection were back- transformed and plotted with a color-coded correlation coefficient for each data point using an in-house developed Matlab script (MATLAB 7.1, the Mathworks Inc., Natwick, USA); this facilitated the interpretation of the results. The color- coded correlation coefficient indicates the importance of the metabolite in contributing to the class separation; a “hot” color (e.g. red) being more important than a “cold” color (e.g. blue). The number of animals used was 12; according to Pearson linear correlation coefficients, a correlation coefficient \|r\| greater than 0.553 was considered to be significant at p\<0.05. # Results ## Bacteremia and Clinical Biochemistry In order to establish bacteremia and monitor the development and recovery of bacteremia, bacterial burden, white blood cell count, C-reactive protein and procalcitonin levels in blood were measured at each time point. Bacteria were detected at 4 h postinfection, reached its highest levels at 8 h and diminished after 2 days postinfection. The level of procalcitonin followed a similar trend to the bacterial load, although the level of procalcitonin was highest at 1 day postinfection. The white blood cell count was significantly reduced at 4 h postinfection and increased at 7 days postinfection, whilst the level of C-reactive protein was significantly increased at 1 day postinfection. ## Metabolites Assignments with ¹H NMR Spectroscopy Typical ¹H NMR spectra of blood plasma and urine obtained from both control and *K. pneumoniae* infected rats at 8 hours after treatment were shown in. The metabolite resonances were assigned according to literature and 2D NMR spectra. Plasma spectra displayed signals from lipoproteins, unsaturated fatty acid (UFA), poly unsaturated fatty acid (PUFA), ω-3 fatty acid, triglyceride (TG), *N*-acetyl glycoprotein (NAG), *O*-acetyl glycoprotein (OAG), glucose, amino acids, dihydrothymine, carboxylic acids, such as lactate and _D-3-hydroxybutyrate (3-HB), and choline metabolites. Urine spectra were comprised of tricarboxylic acid (TCA) intermediate metabolites (citrate, 2-oxoglutarate, succinate, fumarate, malate), alanine, taurine, hypotaurine, dimethylglycine (DMG), dimethylamine (DMA), creatinine, pantothenic acid, 4-cresol glucuronide (4-CG), 2,3-dihydroxybutyrate, 4-deoxyerythronate, trimethylamine *N*-oxide (TMAO), 1-methylnicotimamide, and gut microbial-host co-metabolites (hippurate, indoxyl sulfate, and phenylacetylglycine). The detailed NMR assignment can be found in. To extract the detailed information about *K. pneumoniae-*infected metabolic alterations, multivariate data analysis of these NMR profiles was performed. ## Infection Progression In order to characterize the evolution of the infection through time, PCA was conducted on the NMR data of urine and plasma separately from control and infected rats at all time points. The PCA trajectory plots illustrated the time dependence of the alterations of the plasma and urinary metabolic profiles induced by *K. pneumoniae* infection. Clearly, the global metabolic responses from the profiles of plasma showed a rapid metabolic shift at 8 h postinfection and a speedy recovery through time; this is in contrast to the trajectory of urine profiles, where gradual recovery appears to be made. Cross-validated PLS- DA pair wise comparisons between spectra obtained from the control group and infection group were constructed and validated by a permutation test; it suggested that metabolic disturbances in plasma were diminished at day 10 postinfection while metabolic deviations in urine could still be observed even at day 14 postinfection. ## Metabolic Changes in Plasma Samples To identify the metabolites altered after the infection, the O-PLS-DA models comparing the control group and infection group were constructed for plasma profiles. CV-ANOVA validated model parameters (R², Q² and p values) are listed in. For illustrative purpose, we only showed the cross- validated scores plot and corresponding coefficient plot generated from the model constructed for 8 h after infection. The time dependence of metabolic alterations was displayed in. Compared with the control rats, *K. pneumoniae-*infected rats produce significantly higher levels of lipoproteins, TG, UFA, PUFA, ω-3 fatty acid, 3-HB, lactate, NAG and creatine, and lower levels of glucose and membrane related metabolites such as choline, phosphorylcholine (PC), and glycerophosphocholine (GPC) in plasma. ## Metabolic Changes in Urine Samples Similar analysis was performed for urinary profiles and CV-ANOVA validated model parameters (R², Q² and p values) are also listed in. The cross-validated scores plot and corresponding coefficient plot generated for urine profiles at 8 h after infection is displayed in. The time dependence of urinary metabolic alterations was displayed in. A range of urinary metabolites were also altered after *K. pneumoniae-*infection. The levels of creatine were elevated markedly at 24h post infection and leveled off at 3 days post infection. The levels of taurine, citrate, 2-oxoglutarate, 2,3-dihydroxybutyrate, 4-deoxyerythronate and hypotaurine altered concurrently; these displayed an initial increase to the maximum level at 8 h postinfection and decrease at 2 days postinfection. In contrast, the levels of hippurate, DMG, DMA, *N*-methylnicotinate, formate and indoxyl sulfate and pantothenic acid were reduced at the early stage of infection and gradually increased at the later stage of the infection. Unlike metabolites in plasma, full recovery of urinary metabolites was not achieved after 14 days postinfection. # Discussion Bacteremia is caused by bacterial infection in the blood and can rapidly spread to other parts of the body, causing multiple organ failure. In order to understand metabolic perturbation associated with bacteremia and thus provide a useful nutritional guide for patients with bacteremia, we employed a rat model to investigate metabolic modification induced by *K. pneumoniae* infection, using a metabonomic strategy. ## Infection Progression PCA trajectory of plasma profiles illustrated relocations between 4 h and 24 h postinfection with maximum deviations at 8 h postinfection, which matched perfectly with the bacterial burden in bloodstream which displays plasma profile as a better indication for bacteremia than other immunological response parameters (such as white blood cells, C-reactive protein and procalcitonin). Inspection of concurrently altered metabolites suggested that sharp elevations in the levels of ω-3 fatty acid, UFA, PUFA, TG, lactate and NAG in plasma at 8 h postinfection contributed to the maximum deviations in metabolic space observed at 8 h postinfection. Given that bacterial cultures in blood stream generally takes 24 h, blood tests for aforementioned metabolites could be a valuable and early indicator of bacteremia. ## Energy Metabolism We observed a marked reduction in the levels of glucose in plasma of infected rats. This suggested that stimulated glycolysis is associated with bacteremia; concurrent elevation in the levels of lactate and pyruvate support this notion. The raised urinary level of TCA cycle intermediates in the infected group, such as 2-oxoglutarate and citrate, suggested that the stimulated glycolysis facilitates the rate of the TCA cycle. Previous metabolic investigation of *Trypanosoma brucei brucei* infection in mice also observed stimulated glycolysis. The stimulated glycolysis and TCA cycle reflect the high energy expenditure that is required to fight the infectious process. This is consistent with a previous report stating that bacteremia is accompanied by a decline of mean arterial blood pressure, hypothermia, leucopenia, and hypoglycemia. Disturbed hepatic glycogen mobilization is likely to partially result in hypoglycemia because excessive burdens bacteria and endotoxin could directly lead to liver injury, which is caused by the liver macrophage acting as a filter to remove bacteria from the bloodstream. Administrating glucose to patients with bacteremia could potentially supply the extra energy required to fight the infection, which could reduce the bacteriemia-associated mortality; this has been previously suggested. One of the most prominent findings in the current study was the increase of TG and lipoproteins in plasma after *K. pneumoniae* infection. TGs played an important role in metabolism as an energy source. TGs are constituents of lipoproteins, which deliver the fatty acids to and from adipocytes. When the body requires fatty acids as an energy source, the hormone glucagon signals hormone-sensitive lipase to break down TGs to release free fatty acids. Previous studies have demonstrated that infection and inflammation induce marked changes in lipid and lipoprotein metabolism, including increased serum fatty acids and TGs, increased hepatic TG production and very-low-density lipoprotein secretion and increased adipose tissue lipolysis. Our observation of a significant increase in the levels of TG and lipoprotein in plasma after *K. pneumoniae* infection suggest that bacteria provoke a dramatic response in the host. Our findings are in good agreement with previously observed results of patients with cholera and patients experiencing polymicrobial infection. In addition, we observed increased levels of ketone bodies (such as 3-HB in plasma) from the infected rats, suggesting that existence of *K. pneumoniae* in the bloodstream promotes the β-oxidation of fatty acids in mitochondria. The metabolic profiles of the plasma showed a strong increase in the β-oxidation and a drop in glucose concentration which could mirror the high demand of the body for energy in response to bacterial infection. It is known that the oxidation of fatty acid produces more energy per molecule than glycolysis, therefore ATP generated from fatty acid oxidation is an important energy source required by the liver, lung and kidney to function during severe sepsis. The inability to generate energy via fatty acid oxidation might contribute to the development of multiple organ failure. Furthermore, elevation of creatinine was associated with bacteremia. Creatinine and creatine are inter-convertible metabolites. Creatine is generated from the break down of creatine phosphate, an energy reserve in skeletal muscle; ATP is released when there is a high energy demand. The level of creatine in plasma increases in critically ill patients due to the intracellular breakdown of creatine phosphate to creatine and inorganic phosphates, which restores the dwindling supply of ATP. ## Anti-endotoxin, Anti-inflammatory and Anti-oxidization Responses A sharp rise of lipoproteins in the plasma of rats that have been challenged with bacteria could be one of the anti-endotoxin responses by the host. There is substantial evidence showing that triglyceride-rich lipoproteins can bind and neutralize lipopolysaccharide (LPS), a major component of the cell wall of gram- negative bacteria. Lipoprotein-lipopolysaccharide complexes can ameliorate the effects of the host immune defense to bacterial infection. Hence detoxification by lipoproteins prevents endotoxin from initiating an inflammatory response. New evidence shows that a high-fat diet results in increased plasma triacylglycerol and apolipoprotein B levels, and can significantly decrease endotoxemia and bacterial translocation after hemorrhage. Our observation of marked elevation of lipoproteins is consistent with the anti-endotoxin function of lipoproteins. In addition, increased levels of poly unsaturated fatty acid (PUFA) and ω-3 fatty acid were observed simultaneously in plasma of infected rats. Other studies have shown an increase in the concentrations of PUFA (such as linolenic acid, docosapentaenoic acid and docosahexaenoic acid) in plasma of septic rats. PUFA, principally classified as ω-6 fatty acids and ω-3 fatty acids, have roles in regulating inflammatory responses. The exact roles of ω-6 fatty acids are still unclear. For example, eicosanoids, including prostaglandins, thromboxanes, leukotrienes and other oxidised derivatives, are key mediators and regulators of inflammation; they are mainly synthesized from arachidonica acid, a 20 carbon ω-6 fatty acid, whilst lipoxins, derivatives of ω-6 fatty acids, play important roles in anti-inflammatory processes. ω-3 fatty acids (such as docosahexaenoic acid and eicosapentaenoic acid) was reported to decrease the production of inflammatory eicosanoids (prostaglandin E₂, thromboxane B₂, leukotriene B₄), cytokines, and reactive oxygen species and the expression of adhesion molecules. Although further investigation is needed to certify the levels and roles of ω-6 fatty acids, the current observed marked increase in ω-3 fatty acid implicates anti-inflammatory effect, particularly at 8 h postinfection. In addition, the NAG in rat plasma is known to represent “acute-phase” glycoprotein in animals under inflammatory conditions and may be useful in the diagnosis and prognosis of acute and chronic inflammatory disorders. Hence from a metabolism point of view, the observation of elevated levels of NAG and ω-3 fatty acid was in concurrence with the inflammatory response. Anti-inflammatory responses of the host are also manifested in the increased levels of procalcitonin and concurrently the reduced levels of membrane metabolite, phosphocholine. One of the mechanisms of eliminating bacteria is binding C-reactive protein to phosphocholine on the surface of bacteria. The binding may not be specific to phosphocholine on the surface of bacteria as Bach et al has demonstrated by the binding between C-reactive protein isolated from rabbit with phosphocholine *in vitro*. The interactions between C-reactive protein and phosphocholine could in turn explain the reduced levels of phosphocholine observed in the infected rats and the inconsistence between the levels of C-reactive protein and bacterial load. As mentioned previously that bacteremia-induced β-oxidation of lipid, free radicals generated from this β-oxidation would no doubt promote anti-oxidative response from the host. Indeed, here we have observed elevation in the levels of urinary hypotuarine at 8 hours postinfection and its alteration followed the same trend as the levels of fatty acids. Promotion of lipid oxidation was previously observed in mice infected with *Trypanosoma brucei brucei*. Hypotaurine is an intermediate of taurine biosynthesis, and has been implicated in a wide array of physiological phenomena including membrane stabilization antioxidant, and the regulation of the pro-inflammatory and immune response. In addition, reduction in the levels of *N*-methylnicotinate is associated with bacteremia. *N*-methylnicotinate is the methylated metabolite of niacin (vitamin B₃) and can be generated during the conversion of *S*-adenosyl- methionine to *S*-adenosyl-homocysteine during cysteine biosynthesis (which an important substrate for glutathione synthesis). Hence a depleted level of *N*-methylnicotinate represents an anti-oxidation response of the host. Interestingly, dietary choline participates in the anti-oxidative processes by enhancing the *S*-adenosyl-methionine to *S*-adenosyl-homocysteine ratio, and regulating the activities of methyltransferases as well as promoting the formation of glutathione. This results in an attenuated inflammatory response, reduced tissue injury and mortality in the rat. Our results suggest that choline supplementation during sepsis could be beneficial to patients. ## Disturbance of Gut Microbes In our current investigation, decreased levels of choline in plasma, and concurrent decreased levels of DMA and DMG in urine were observed in rats treated with *K. pneumoniae*. Previous research has demonstrated that urinary DMA and DMG are produced via the action of gut microbiota on choline. Therefore, it is plausible to suggest that bacteremia causes a disturbance to gut microbiota. The changes in gut microbial co-metabolites, such as hippurate and indoxyl sulfate further validated our suggestion. Hippurate, generated in the liver, originates from bacterial action upon plant phenols to produce benzoate, which becomes conjugated with glycine. Indoxyl sulfate is the metabolite of tryptophan under the role of a subset of microbiota that has tryptophanase activity. Alterations in the level of hippurate and indoxyl sulfate were previously reported as a consequence of the perturbation in gut microbiota. However, since no previous report has shown the association between the changes of gut microbiota and bacteremia, further microbiological studies are warranted to ascertain this association. ## Conclusions In summary, we have characterized time dependence of plasma and urinary metabolic alterations in response to *K. pneumoniae* infection using the metabonomic strategy, indicative of global changes in metabolic regulation. We have shown that metabolic profiles of plasma could be a better indication of bacteremia. *K. pneumoniae* bacteremia disrupts energy metabolism, which is manifested by stimulated glycolysis, TCA cycle and oxidation of lipid and creatine phosphate. In addition, *K. pneumoniae* bacteremia induced focused anti-endotoxin, anti-inflammatory and anti-oxidization responses. Further investigation is needed to validate the disruption of the gut microbiota balance. Our results indicated that infection by *K. pneumoniae* caused altered metabolites that act as a guide for clinical nutrition intake in human conditions of bacteremia. An integrated NMR analysis of plasma and urine provided a holistic method for elucidating metabolic cross-talk between the host and the bacteria *in vivo* during the progress of the infection. Hence, a global metabolic profiling strategy based on ¹H NMR spectroscopy in conjunction with multivariate data analysis can be utilized for the development of novel, valid, and rapid methods for disease management. # Supporting Information We would like to thank Dr L. Caetano M. Antunes, The University of British Columbia, for his comments in the peer review process, which improved our manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: FD BW YW. Performed the experiments: FD BW LZ. Analyzed the data: FD YW. Contributed reagents/materials/analysis tools: BW JL HT YW. Wrote the paper: FD BW JL HT YW.

# Introduction Loquat (*Eriobotrya japonica* Lindl) originated in China and has been cultivated for 2100 years. Owing to its economic and ecological attributes, loquat is an important perennial fruit crop species and is cultivated largely between the N 35° and S 35° latitudes worldwide. Loquat blossoms in late autumn or early winter, and young fruits are vulnerable to freezing injury. In 2016, 90% of the loquat planting area in China experienced freezing, with almost no material harvested. Freezing injury has severely jeopardized the economic benefits of farmers and has become a major restricting factor for sustainable development in many production areas worldwide. Current research on loquat has mainly focused on cell genetics, physiology and biochemistry, molecular markers, molecular clones, etc. Several transcriptome studies in loquat focused on flower bud differentiation, fruit development and ripening, and postharvest storage, research on transcriptome in cold stress of loquat is limited, little is known about its cold tolerance mechanisms. Previous studies have been performed using second generation sequencing technology, and many unigenes have been obtained, however, transcriptomic sequences using second generation sequencing technology may be misassembled without a high-quality genome sequence or full-length (FL) transcriptomic sequences available as a reference. To date, FL transcriptomic data are scarce. In addition, Loquat is a non-model plant species with high heterozygosity, and a loquat reference genome is still lacking, which has limited molecular biological research of this species. In recent years, third-generation sequencing technology has been successfully applied to functional genomics research of sweet potato, Populus, sorghum, corn, and cotton, among others. Compared with second-generation sequencing technology,third-generation sequencing technology not only has advantages that include handling a large volume of data and the ability to read long sequences and FL gene transcripts, but it is also greatly more accurate in terms of gene functional annotation without sequence splicing and assembly. In the present study, The FL transcriptome of embryos of young loquat fruit under low-temperature stress was obtained by single-molecule real-time (SMRT) sequencing. This work will facilitate future research on identifying functional genes and analysing molecular mechanisms related to the cold stress response of loquat. # Materials and methods ## Plant materials and treatments Two-year-old grafted Ninghaibai loquat plants that were growing in pots and that had already produced fruit (with a diameter of approximately 1.5 cm) were used as the experimental materials, and the growth status of the plants was as uniform as possible. The plants were subjected to three different temperatures, 1°C, -1°C, and -3°C, for 12 h or 24 h separately after being subjected to a gradient of cooling at a rate of 4°C/h. The treatments were applied in a low- temperature plant incubator with 60% relative humidity, a 3000 lx light intensity, and a 12-h/12-h light/dark cycle. The plants were then removed and incubated at room temperature for 6 h to recover, after which the embryos of young loquat fruit were collected, immediately frozen in liquid nitrogen and stored at -80°C. Plants that had been growing at room temperature were used as controls. Each treatment involved three biological replications. A total of 21 samples of embryos of young loquat fruit (three biological replicates for treatments of 1°C, -1°C, and -3°C, for 12 h or 24 h, including the control group) were collected for the following experiments. ## RNA extraction and quantification Total RNA was extracted with the RNAprep Pure Plant Kit (TIANGEN, Cat. No. DP441) following the manufacturer’s protocol. The samples were quantified as follows. The purity and concentration of RNA were first measured using a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Rockland, DE, USA) according to their OD260/280 value, after which the RNA integrity was assessed using an RNA Nano 6000 Assay Kit in conjunction with an Agilent Bioanalyzer 2100 system (Agilent Technologies, CA, USA). The RNA degradation and contamination were measured on 1% agarose gels. Only total RNAs with a RIN score ≥8.0 were used to construct cDNA libraries for SMRT or Illumina sequencing. ## PacBio Iso-Seq library preparation and sequencing After the RNA quality was verified, libraries were constructed. mRNA was purifed from 3μg of mixed total RNA of 21 samples of embryos of young loquat fruit for SMRT library preparation and sequencing. The instruments used include a SMARTer^™ PCR cDNA Synthesis Kit (Clontech, CA, USA)and BluePippin^® Size Selection System (Sage Science, Beverly, MA, USA). The SMARTer^™ PCR cDNA Synthesis Kit (Clontech, CA, USA) was used for synthesizing FL cDNA, the generated cDNAs were then reamplified via PCR. The remaining overhangs were converted to blunt ends by exonuclease/polymerase activities. After adenylation of the 3′ ends of the DNA fragments, NEBNext Adaptors with a hairpin loop structure were ligated in preparation for hybridization. The BluePippin^® Size Selection System was used for size selection(1–2 kb, 2–3 kb and 3–6 kb) to bulid 3 libraries. The quality of the libraries was assessed using an Agilent Bioanalyzer 2100 system, and SMRT sequencing was performed using a Pacific Biosciences real-time sequencer in conjunction with C2 sequencing reagent. ## Illumina transcriptome library preparation and sequencing 21 second-generation-sequencing cDNA libraries of embryos of young loquat fruit (three biological replicates for treatments of 1°C, -1°C, and -3°C, for 12 h or 24 h, including the control group) were constructed respectively using a NEBNext^® Ultra^™ RNA Library Prep Kit for Illumina^® (NEB, Beverly, MA, USA) according to the manufacturer’s protocol. Briefly, mRNA was purified from 5μg of total RNA using poly-T oligo- attached magnetic beads. Fragmentation was carried out using divalent cations under high temperature in NEBNext First Strand Synthesis Reaction Buffer (5X). First-strand cDNA was synthesized using random hexamer primers and M-MuLV Reverse Transcriptase (RNase H-). Second-strand cDNA synthesis was subsequently performed using DNA polymerase I and RNase H. The remaining overhangs were converted to blunt ends via exonuclease/polymerase activities. After poly- adenylation of the 3’ ends of the DNA fragments, NEBNext adaptors with hairpin loop structures were ligated in preparation for hybridization. An AMPure XP system (Beckman Coulter, Beverly, USA) was used to select cDNA fragments that were 200–250 bp in length. Afterward, 3 μl of USER Enzyme (NEB, USA) together with size-selected, adaptor-ligated cDNA was incubated at 37°C for 15 min and again at 95°C for 5 min. PCR was then performed with Phusion High-Fidelity DNA Polymerase, universal PCR primers, and Index (X) Primer. The PCR products were ultimately purified (AMPure XP system), and the library quality was assessed using the Agilent 2100 system. The qualified libraries were pair-end sequenced on an Illumina HiSeq 2500 (Illumina, San Diego, CA, USA) system. ## Error correction and quality control of SMRT reads Raw data (raw reads) in fastq format were first processed using in-house Perl scripts. Raw SMRT sequencing reads were processed by removing polymerase reads that were \<50 bp and had a accuracy \<0.8, resulting in subreads. The joined subreads were disconnected, and joint sequences that were \<50 bp were removed, resulting in clean data. The obtained clean reads were processed into error- corrected reads of inserts (ROIs) with parameters including full passes ≥0 and a sequence accuracy ≥0.8. Then, full-length, non-chimeric (FLNC) transcripts were determined by searching for poly-A tail signals and the 5’ and 3’ cDNA primers within the ROIs. Iterative clustering for error correction (ICE) was used to obtain consensus isoforms, and FL consensus sequences from ICE were polished using Quiver. High-quality FL transcripts were classified as those with a post- correction accuracy criterion surpassing 99%. Any redundancy in high-quality, FL transcripts was removed by CD-HIT, and the integrity of the transcriptome was evaluated without redundancy by BUSCO. ## Alternative splicing (AS) detection We subjected Iso-Seq^™ data directly to an all-vs-all BLAST analysis, with high identity settings. The BLAST alignments that met all the criteria were considered products of candidate AS events. There should be two high-scoring segment pairs (HSPs) in the alignment: two HSPs had the same forward/reverse direction, and within the same alignment, one sequence should be continuous, or with a small "overlap" size (smaller than 5 bp); the other sequence should be distinct to show an "AS gap", and the continuous sequence should align to the distinct sequence almost completely. The AS gap should be larger than 100 bp and at least 100 bp away from the 3'/5' end. ## Simple sequence repeat (SSR) detection Transcripts \>500 bp were selected for SSR analysis using the MIcroSAtellite identification tool (MISA; <http://pgrc.ipk-gatersleben.de/misa/http://pgrc.ipk- gatersleben.de/misa/>). MISA was used to identify seven SSR types, namely, mononucleotide, dinucleotide, trinucleotide, tetranucleotide, pentanucleotide, hexanucleotide, and compound SSRs, by analysing transcript sequences. ## Prediction of coding DNA sequences (CDSs) The CDSs and corresponding amino acid sequences within the transcript sequences were predicted using TransDecoder (<https://github.com/TransDecoder/TransDecoder/releases>). TransDecoder was used to identify candidate protein-coding regions based on the open reading frame (ORF) length, log-likelihood score, nucleotide composition, and (optional) Pfam domain content. ## Long non-coding RNA (lncRNA) prediction Putative protein-coding RNAs were filtered and removed using the following minimum length and exon number thresholds. Transcripts that were longer than 200 nt and that had more than two exons were selected as lncRNA candidates and further screened using the Coding Potential Calculator (CPC)/Coding-Non-Coding Index (CNCI/Coding Potential Assessment Tool (CPAT)/Pfam database, which has the power to distinguish protein-coding genes from non-coding genes. ## Functional annotation of transcripts and analysis of transcription factors (TFs) The non-redundant transcript sequences obtained were mapped to eight different databases to obtain annotation information associated with the transcripts. These databases included the non-redundant (NR), Swiss-Prot, Gene Ontology (GO; <http://www.geneontology.org>), Clusters of Orthologous Groups of proteins (COG; <http://www.ncbi.nlm.nih.gov/COG>), euKaryotic Orthologous Groups (KOG), Pfam (<http://pfam.janelia.org/>), evolutionary genealogy of genes: Non-supervised Orthologous Groups (eggNOG; <http://eggnog.embl.de>), and Kyoto Encyclopaedia of Genes and Genomes (KEGG, <http://www.genome.ad.jp/kegg/>) databases. Finally, TFs were predicted using iTAK predictive software. # Results ## SMRT- and Illumina-based RNA sequencing and error correction A total of 13.41 Gb of clean data were generated via Pacific Biosciences SMRT sequencing technology. Based on the conditions of full passes ≥0 and a quality \>0.8, a total of 215,636 reads of inserts (ROIs)were obtained, and 121,654 full-length non-chimeric (FLNC) sequences were identified. In total, 76,586 consensus isoforms were obtained by iterative clustering for error correction(ICE). After error correction with second-generation sequencing short reads was performed, a total of 38,435 non-redundant transcripts with an average length of 2607bp were obtained, including 12,520 high-quality transcripts. All the raw data were deposited in the NCBI Sequence Read Archive (SRA) under accession number PRJNA623262 and are available at <https://www.ncbi.nlm.nih.gov/bioproject/PRJNA623262>. ## Predictions of CDSs, lncRNAs, and SSRs A total of 1,295 alternative splicing (AS) sequences were obtained. There were 37,230 ORFs that included 27,905 CDSs identified by TransDecoder, the distribution of the coding sequence lengths of complete ORFs is shown in. Four computational approaches (CPC analysis, CNCI analysis, Pfam protein domain analysis, CPAT analysis) were used to screen the transcripts that encode coding proteins, and 407 lncRNAs were predicted. Transcripts that were \>500 bp were selected for SSR analysis using MISA. In total, 24,832 SSRs were identified, including 5,317 sequences containing more than 1 SSR and 3,536 SSRs present in compound formation. Moreover, SSRs consisting of one to six (mono-, di-, tri-, tetra-, penta-, and hexa-nucleotides) tandem repeats were identified, Mono- nucleotid repeats (12,230) were the most abundant, followed by di-nucleotid repeats (8857), tri-nucleotid repeats (3327), tetra-nucleotide repeats (254), hexa-nucleotide repeats (95) and penta-nucleotide repeats (69). ## Transcript functional annotation and sorting of transcription factors In total, 37,993 transcripts were annotated in eight databases. Among these transcripts, 37,908 were annotated in the NCBI NR database, 16,261 were annotated in the COG database, 22,732 in the GO database, 16,507 in the KEGG database, 24,787 in the KOG database, 31,494 in the Pfam database, 28,599 in the Swiss-Prot database, and 37,074 in the eggNOG database. NR contains protein data from the Swiss-Prot, Protein Information Resource, Protein Research Foundation, Protein Data Bank, GenBank, and RefSeq databases;it is a non-redundant protein database housed within the NCBI. The non-redundant transcripts were compared to those in the NR database,the results showed that 46.22% of sequences were aligned to Pyrus x, followed by *Malus domestica*(45.40%), only 0.35% of sequences were aligned to loquat itself. GOanalysis indicated that 22,732 transcripts enriched in the pathways related to biological processes, cellular components, and molecular functions. A large number of transcripts in ‘‘cellular components” were mainly involved in cell part, cell, organelle, membrance, membrane part, and macromolecular complex. The category ‘‘molecular functions” mainly consisted of transcripts involved in catalytic activity, binding and transporter activity. The category ‘‘biological process” mainly consisted of transcripts involved in metabolic process, cellular process, single-organism process,biological regulation, localization, responses to stimulus, and cellular component organization or biogenesis. In the COG database, we found that the R function (general function prediction only) had the largest number, followed by the K function (transcription), L function (replication, recombination, and repair), and T function (signal transduction mechanisms). Transcription factors (TFs) play a very important role in the biological processes of plants, A total of 5,322 TFs were predicted by iTAK software, and the numbers of TFs enriched were as follows: RLK-pelle_DLSV (315), C3H (146), SNF2 (136), bHLH (127), and RLK-pelle_LRR-XI-1 (117). # Discussion The loquat genome has yet to be sequenced, research on the physiology and genetics mechanisms of this species has been restricted. Second generation sequencing technology is incapable of assembling full-length transcripts because of the shortness of sequencing reads. AS sites cannot be accurately detected, and the prediction accuracy is lower than 50%. Moreover, fusion genes and gene families cannot be accurately detected. Thus, we can improve the accuracy of transcriptomic data and the prediction accuracy of AS by combining third- generation FL transcriptomic data with second-generation transcriptomic data. Third-generation combined with second-generation sequencing has been widely used to analyze rare transcripts, mining functional genes, analysing different genes in different tissues and at different developmental stages, and analysing the regulatory activity of TFs. To study plants for which a reference genome is not available, the most direct and effective use of ‘omics’ involves transcriptome and digital gene expression profile analysis, but obtaining high- quality reference genomes of genetically complex organisms remains costly and is technically challenging. In this study, a total of 13.41 Gb of raw data were obtained by SMRT sequencing, and after clustering analysis, non-FL sequence correction and the removal of redundant sequences, 38,435 transcripts with an average length of 2607 bp were obtained, which is far superior to previous studies of the loquat transcriptome using only the second-generation sequencing technique. For example,Song obtained48,838 transcripts with an average length of 790 bp, and Xu obtained 87,379 transcripts with an average length of 710 bp. Thus, Our findings indicated that SMRT sequencing is an effective route for obtaining reliable full-length transcript sequence information in plants. LncRNAs are a class of non-coding RNA with a length longer than 200 nucleotides. Currently, many studies have been conducted to examine lncRNAs in animals, while research on lncRNAs in plants mainly focuses on a few model plants such as Arabidopsis thaliana, rice, and tomato. In recent years, with the development of high-throughput sequencing technology, an increasing number of studies have focused on lncRNAs in plants, which have been found to play a regulatory role in plant flowering, reproductive development, photomorphogenesis, response to biotic and abiotic stresses, and in other biological processes. In the present study, 407 lncRNAs were predicted from the non-redundant transcripts. These newly identifed lncRNAs will be helpful for loquat research in several aspects, and the function of lncRNAs in response to low temperature stress of loquat requires further study. Full-length sequence transcripts are crucial for genome annotation and gene function research. However, most methods for obtaining full-length transcripts are expensive, time-consuming and inefficient. To date, no full-length sequence transcripts in loquat have been reported. In this study, 38,435 transcripts were obtained using the PacBio SMRT sequencing platform. Based on these transcripts, 37,230 ORFs were predicted, of which 27,905 had a complete CDS, and 37,993 transcripts were annotated into 8 databases including NR, eggNOG, Swiss-Prot, GO, COG, KOG, Pfam and KEGG. 37,908 transcripts annotated to the NR database, 46.22% of the sequences were aligned to Pyrus x and 45.40% to Malus domestica, whereas loquat itself had a best match percentage of 0.35%. These results may be due to the lack of transcript data related to loquat in the current NR database, reflecting the urgent need to improve the genetic database for this genus. The rational classification of protein coding is critical to maximize the use of transcripts for functional research. The results of the COG analysis showed that the R function (general function prediction only) constituted the greatest proportion, followed by the K function (transcription), L function (replication, recombination and repair) and T function (signal transduction mechanisms),which was similar to the results reported by Gong. This result indicated that the gene expression of loquat under low-temperature stress is related to the above functions and suggested that the use of transcriptome sequencing technology is an effective method for the study of functional genes. The results of this study provide a new reference for loquat transcription. However, analysis of the loquat transcriptome was not comprehensive, and gene expression and metabolic pathways associated with the mechanism underlying the cold stress response of loquat require further analysis. # Conclusion This is the first study to perform SMRT sequencing of the FL transcriptome of embryos of young loquat fruit of plants under low-temperature stress. A total of 38,435 transcripts were obtained, 407 lncRNAs were predicted, 24,832 SSRs and 27,905 coding sequences were identified, and 37,993 transcripts were annotated for subsequent analysis. The number and average length of the transcripts were much better than those of previous studies in the loquat transcriptome using only the second-generation sequencing technique. SMRT sequencing is a useful and effective tool for acquiring reliable full-length transcripts of loquat. This work will facilitate research on the functional identification of genes and elucidation of the molecular mechanism underlying the cold stress response in loquat. We would like to thank Biomarker Technologies Co., Ltd., for technical assistance with the RNA-Seq analysis. Professor Kevin is gratefully acknowledged for critical comments on the manuscript. [^1]: NO authors have competing interests.

# Introduction The sweet potato leaf folder, *Brachmia macroscopa* Meyrick (Lepidoptera: Gelechiidae), is one of the most destructive pests on *Dioscoreae sculenta*, *Ipomoea aquatic*, *Calystegia sepium*, *C*. *japonica* and many other crops belonging in the Convolvulaceae. It is widely distributed in Europe, Russia, Caucasus, the Transcaucasian region, West Kazakhstan, central Asia, Korea, Japan, China, and northern India. Larval damage to the host plant leaves results in complete loss of the mesophyll layer, leaving only the transparent leaf epidermis. The extent of damage may be severe enough to disrupt the host’s ability to photosynthesize, causing withering of the affected leaves, and in severe cases ultimately lead to death of the host plant. Serious infestations may cause high host mortality and result in serious economic loss. The biological responses of *B*. *macroscopa* larval, pupal and adult stages to different temperatures as well as a brief discussion of some of the pests’ thermal requirements have previously been reported. To date, however, none of the previous studies have included quantitative life table data such as the many useful parameters that are obtainable from calculating two-sex life table parameters, including net reproductive rate (*R*_*0*), gross reproductive rate (*GRR*), intrinsic rate of increase (*rm*), finite rete of increase (*λ*), and mean generation time (*T*), etc. Life tables are widely accepted as a powerful and necessary tool for analyzing and understanding the effects abiotic and biotic factors such as temperature have on the growth, survival, reproduction, and intrinsic rate of increase of insect populations. Different methods for analyzing life table data needed in studies on population ecology have been reported. Life tables have also been widely adopted in ecological studies involving insect populations, including insect mass-rearing techniques, timing of pest control procedures, studies on host preference and fitness of insects, as well as ecological studies on the effects that environmental variables, including temperature, have on populations dynamics. Population studies based on life tables other than the age-stage, two- sex life table can result in the incorporation of errors since they do not take the variation in developmental rates among individuals and between sexes into consideration. The age-stage, two-sex life table, on the other hand, was designed to allow incorporating variations in pre-adult development time, which, in turn, produces precise survival and fecundity curves. Research on the demography of an insect pest species under different temperature conditions has often been regarded as a basis for developing an eco-friendly management strategy. The objective of this study was to gather additional data on the biological and ecological properties of *B*. *macroscopa*, by conducting a study of its demographics using the results obtained from the age-stage, two-sex life table parameters under different temperature conditions. The ultimate goal in this study would be to provide key data needed to make appropriate, efficient, and strategic decisions in devising an effective control program for the species. # Materials and methods ## Insect culture Larvae, pupae and adults of *B*. *macroscopa* were collected from experimental fields belonging to Hunan Agricultural University (Changsha, Hunan, China; 28°110’, 113°40’). Insects used in each of the temperature treatments had been bred in the laboratory for at least two generations. Larvae were provided with freshly cut sweet potato leaves as a food source. The leaves were changed daily with freshly cut leaves to maintain hygienic conditions and to provide fresh leaves for the larvae until they reached the pre-pupal stage. Pupae and adults were kept in an insect rearing cage containing a sweet potato and allowed to mate. The sweet potato plants were covered with a mesh net to serve as an oviposition substrate. Adult nutrients consisted of a cotton ball saturated with a 30% honey solution which was replaced daily. ## Developmental time In order to obtain eggs of uniform age, 20 pairs of female and male *B*. *macroscopa* were placed in a separate rearing container that was covered with mesh net to provide ventilation. A potted sweet potato plant, covered with mesh net as before, had been placed inside the cage to serve as the oviposition substrate. One hundred and fifty eggs were collected daily and used to initiate every temperature treatment, and separate cultures were maintained at different constant temperatures of 21, 24, 27, 30, 33, and 36°C, in artificial climate chambers at 75±20% RH and a photoperiod of 10L: 14D. Fresh sweet potato leaves were provided daily for the neonate larvae. Since each larva was considered as a replicate, they were individually transferred, using a fine brush, to separate plastic petri dishes (9 cm in diameter and 1.5 cm in height). To maintain freshness, the leaf petioles of the detached leaves were wrapped in water-soaked cotton. New pupae were placed separately in glass tubes (2 cm in diameter and 10 cm in height) containing a moistened cotton ball at the base of the tube, and the tubes covered with gauze. Daily records were kept for each individual, including recorded data for each larval molt, pupation and adult emergence, as well as noting any mortality that occurred. The entire larval period, prepupal period and pupal period, i.e., the developmental time from egg to adult emergence, was defined as the length of the pre-adult stage. In addition, the mortality of all stage was also exhibited. The experiment was continued until the death of all individual insects. ## Oviposition period, fecundity and longevity When the pupae emerged as adults, individual males and females that had emerged during the same 24 hr period were paired and the pair transferred to a plastic oviposition container (13 cm in diameter and 17 cm in height). Each of the oviposition chambers contained a small cotton wick soaked with 20% honey solution for adult nutrition, along with a small sweet potato plant grown in a disposable cup to allow for oviposition. The potted plants were replaced daily and the number of eggs produced during that 24 hr period were recorded. Other parameters including the adult pre-oviposition period (APOP; the period of time between the emergence of an adult female and the on set of her first oviposition), the total pre-oviposition period (TPOP; the time interval from birth to the beginning of oviposition), the oviposition period, daily fecundity, and total fecundity (total number of eggs produced during an individual’s reproductive period) were obtained using the experimental data. ## Life table parameters The age-stage, two-sex life table approach was used to analyze the raw life- history data for *B*. *macroscopa*. The age-stage -specific survival rate (*s*_*xj*) (= the probability that an individuals would survive to age *x*_*j* and stage *j*) \[the first stage is the egg stage, the second stage is the 1st larval stage, the third stage is the 2nd larval stage, the fourth stage is the 3rd larval stage, the fifth stage is the 4th larval stage, the sixth stage is the 5th-6th larval stage (since the *B*. *macroscopa* fed on different temperatures had 5th to 7th stages, we grouped the 5th, 6th and 7th larval stages as the sixth stage), the seventh stage is the pupal stage, the eighth stage is the adult stage\] was evaluated. The age-stage-specific fecundity (*f*_*xj*) (daily number of eggs laid by individual at age *x* and stage *j*), the age-specific fecundity (*m*_*x*) \[(daily number of eggs produced per female (this number is divided by all individuals’ males and females)\], and the age specific survival rate (*l*_*x*) (the probability that the newly oviposited egg will survive to age *x*), were calculated. The intrinsic rate of increase (*r*) was calculated using the Eule- Lotka equation as ${\sum\limits_{x = 0}^{\infty}e^{- r_{m}(x + 1)}}l_{x}m_{x} = 1$. The gross reproductive rate (*GRR*) was calculated as $\mathit{GRR} = \sum m_{x}$. The finite rate of increase (*λ*) was calculated as $e^{r_{m}}$. The net reproductive rate (*R*_*0*) was measured as $\sum\limits_{x = 0}^{\infty}{l_{x}m_{x}}$. The mean generation time (*T*) -defined as the time that the population could increase to *R*_*0*-fold of its population size as it approaches the stable stage distribution, was calculated as *T* = (ln*R*_*0*)/*r*_*m*. Finally, the means, standard errors and variances of the population parameters were estimated via the bootstrap technique, which is contained in the TWOSEX-MSChart program. Sigma plot 12.5 was used to create the graphs. ## Data analysis The raw life-history data obtained for *B*. *macroscopa* in each of the temperature treatments were entered separately into Microsoft Excel 2013. The computer program, TWOSEX-MSChart for the age-stage two-sex life table analysis in VISUAL BASIC (version 6, service pack 6) for the Windows system available at [http://140.120.197.173/Ecology/ Download/TWOSEX- MSChart.zip](http://140.120.197.173/Ecology/ Download/TWOSEX-MSChart.zip) (National Chung Hsing University) and at <http://nhsbig.inhs.uiuc.edu/wes/chi.html> (Illinois Natural History Survey), was used for analysis of the raw data. The computer program greatly simplifies the otherwise lengthily procedure required for analysis of the large amount of data generated in life table studies. One-way analysis of variance (ANOVA) was adopted to analyze the duration data. The LSD test was used to detect significant differences in the statistical analysis among means at *P = 0*.*05*. Nonlinear regression was used to analyze the development rates with the temperatures (using SPSS 19.0) # Results ## Development time, adult longevity and lifespan In the present study, the mean development rate of the larval stage fit the nonlinear equation: y = -0.362 + 0.136lg *x* with a coefficient of determination (*R*²) of 0.887. Eggs in the 36°C treatment were unable to successfully hatch, therefore, the mean duration of the total pre-adult stages of *B*. *macroscopa* at different temperatures are only listed for temperatures between 21–33°C. Significant differences (*P\<0*.*05*) occurred in both the larval and total immature periods in the five different temperature treatments, with the shortest developmental time of the two periods being at 30°C and the longest at 21°C. The longest pupal period also occurred at 21°C while the shortest was in the 33°C treatment, although there was no significant difference (*P\<0*.*05*) noted at the 30°C treatment. At 21°C, larvae required a seventh instar to complete their development, but needed only six instars in the24°C controls, and five when temperatures were between 27–33°C. shows the adult longevity (from emergence to death of the males and females separately), the entire lifespan (from egg to adults’ death) and the mortality rates of *B*. *macroscopa* immature stages. The four different temperature settings (21, 24, 27, 30°C) used as rearing temperatures had a significant effect (*P\<0*.*05*) on both the female and male longevities. However, neither the females nor the males had a statistical significance in longevity between the two temperature extremes, 33°C and 21°C. Each of the five temperature regimens had a significant impact (*P\<0*.*05*) on the female lifespan, whereas there was no notable difference between male lifespans at 24°C and 27°C (*P\>0*.*05*). The highest (31.43%) and lowest (15.33%) mortality rates occurred at 24°C and 27°C, respectively. ## Oviposition period and fecundity shows the APOP, TPOP, oviposition period and female fecundity values of *B*. *macroscopa* under different temperature conditions. Although the longest APOP value (3.81±0.29c) occurred at33°C the reverse was true in the TPOP values, where the shortest value (18.91±0.31c) occurred at 33°C. Females rearedat21°C had the highest fecundity (376.02±12.53a), whereas those from the 33°C treatment had the lowest (23.24±3.60c). The length of the oviposition period, in general, decreased from 21°C to 33°C. ## Life table analysis The age-stage specific survival rate (*s*_*xj*) values for insects cultured at various temperatures are shown in. The lowest survival rate for the egg and larva stage was at 33°C, while the 27°C group had the highest survival rate for the egg and female pupal stage. The two remaining stages (larva and male pupa) exhibited the highest survival rate in the 30°C treatment. lists the age-specific survival rate, age-stage-specific fecundity (*f*_*x8*) and age-specific fecundity (*m*_*x*) values of *B*. *macroscopa* reared at the five different temperatures. Since only female adults (the tenth stage) produce offspring, there is only a single curve for the female age-stage- specific fecundity (*f*_*x8*). The peak age stage-specific fecundities observed at the 21, 24, 27, 30, and 33°C temperatures occurred at 42, 29, 25, 20, and 25 days of age respectively, with corresponding values of 29.02, 37.44, 15.93, 21.12 and 2.49 eggs/female/day. The age-specific fecundities in the five temperature treatments ranging from the highest to lowest fecundity values were found at24, 27, 30, 21, and 33°C. shows the values for the age-stage, two-sex life table parameters. The highest values in net reproductive rate (*R*_*0*) and finite rate of increase (*λ*) were found in the 27°C group, while the lowest R₀, gross reproductive rate (*GRR*), intrinsic rate of increase (*r*_*m*), and λ values for *B*. *macroscopa* occurred in individuals reared at 33°C. The longest and shortest mean generation times (*T*) for *B*. *macroscopa* occurred in the 21°C and 30°C treatments, respectively. # Discussion A complete understanding of the population dynamics of the target species is essential to developing viable integrated control programs. Life tables offer an effective means of tracking changes in a population’s growth as well as many other important parameters. In population ecology studies involving insects, life tables offer the most integrated information on their growth, survival, reproduction and population growth parameters. Many life table parameters including development, mortality, reproduction and other elements can be temperature-dependent. Temperature is considered to be the most crucial abiotic factor affecting the establishment and growth of pest populations. Temperature has proven to be a vital factor in an insect’s development and survival, with relatively minor variations having disproportional effects on the growth rate of one or more of their stages. Understanding the relationship between temperature variations and their effect on population dynamics would be invaluable in predicting the behavior of a pest on its host crop cultivars. In order to represent temperature-dependent developmental rates, using linear or non-linear regression equations are effective means for demonstrating the relationship of development rate on temperature changes. In this study, we used 21, 24, 27, 30, 33 and 36°C to examine the effects that various temperatures have on the development of *B*. *macroscopa*. The 36°C temperature was apparently lethal to *B*. *macroscopa* -based on our observation that eggs never hatched at this temperature. Our results verified the sensitivity of this moth to high laboratory temperatures. The female adults were capable of reproducing at all other temperatures (21–33°C), even when the 33°C group had an amazing low fecundity compared to the remaining four groups. Similarly, *Spodoptera exigua* Hübner was reported to experience suppression of growth and development at 32°C, whereas it could successfully complete its development from 20–29°C. The developmental and metabolic rates apparently increase as the temperature is raised, however, as the temperature approaches the upper lethal limit, the metabolic rate decreases after the developmental rate. The fecundity of *B*. *macroscopa* was a vital parameter that was directly related to variations in rearing temperatures. This phenomenon has been recorded in various other insect species, where increases in culturing temperatures often result in noticeable decreases in female productivity. There were distinct variations in the number of larval instars reared at diverse temperatures. The numbers of larval instars reached seven in the 21°C cohort, decreased to six at 24°C, and at higher temperatures (27–33°C) had only five instars. The discrepancies in the number of instars were likely related to the rearing conditions. Since the immature stages are hypersensitive to abiotic and biotic elements in their environment, there is constant pressure on them to complete their immature stages as rapidly as possible to reach the adult stage and onset of maturity. In our study, although the developmental time in the 21°C rearing environment was significantly longer than at other temperatures, the larvae had more opportunity to prey. Which in itself may increase their effectiveness as a biological control in the Integrated Pest Management (IPM) program. Nevertheless, the shorter developmental durations experienced at higher temperatures may sharply accelerate the immature stages, causing an early onset of maturity, which, in turn, would lead to increases in progeny and larger populations. The small size of the eggs produced by *B*. *macroscopa*, making them difficult to observe in the field, coupled with the fact that the immature stages of this leaf folding pest develop entirely in the revolute leaves, makes direct observation of the moths’ development challenging. Establishing age-specific life tables based on statistics gained from laboratory studies would be of fundamental importance in creating integrated pest management programs for this important pest. In most insect populations, variations in developmental rates between individuals and among the sexes commonly occur. The stage distinction would not be detected if traditional age-specific female life tables had been used. Realizing that the susceptibility of insects to varied environmental factors, pesticides, natural enemies and so on, often differs depending on their developmental stage, information regarding the population stage structure is critical to effective pest management. This variation is not accounted for in traditional life table analysis and usually results in erroneous data conclusions. Development of lab generated age-specific life tables can also play an important role in developing pest-resistant cultivars—one of the most environmentally friendly approaches of integrated pest management in reducing the negative effects produced by phytophagous insect pests. In this research, the bootstrap technique was used to estimate the means and variance data of the population parameters. In our results, the highest *R*_*0* value for *B*. *macroscopa* (147.60) occurred at the27°C temperature group. Among the various population parameters, the intrinsic rate of increase (*r*_*m*) is a critical demographic element for determining levels of environment resistance to insects. Comparisons of the values within the *R*_*0* and *r*_*m* parameters invariably available for considerable insight beyond that from the independent analysis of individual life-history parameters. Conversely, the decreased values obtained in the 33°C treatment revealed that this temperature was near the maximum developmental threshold for the pest. Consequently, *B*. *macroscopa* would be most likely to reach maximum population levels at 27°C. The study establish baseline constant-temperature age-stage two-sex life tables for the developmental and survival parameters of the *B*. *macroscopa*, which could be available to model development in the wild and estimate potential distribution limits. Knowledge of the temperature-dependent growth and development of *B*. *macroscopa* could be used to forecast the peak periods as well as reduced activity of this pest, which could inform control programs. In addition. key bioclimatic parameters such as the reproductive growth under different temperatures could be used to optimize the production methods in mass rearing for parasitoids. Further studies are needed to identify the phytochemicals responsible for retarding development at the resistant temperatures in order to design a more effective IPM program for *B*. *macroscopa* in the field. # Supporting information The authors would like to thank Prof. Jian-Hong Li (Huazhong Agricultural University) for providing guidance in research, and Prof. Hsin Chi (National Chung Hsing University) for his kind help in data analysis. We thank Dr. Cecil L. Smith (University of Georgia, USA) for editing and revising the English Language. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** GHH LM XW. **Data curation:** GHH LM. **Formal analysis:** LM. **Funding acquisition:** GHH. **Investigation:** YL MZS LM. **Methodology:** GHH LM XW. **Project administration:** GHH. **Resources:** YL MZS LM. **Software:** GHH LM XW. **Supervision:** GHH LM XW. **Validation:** GHH LM XW. **Visualization:** GHH LM XW. **Writing – original draft:** GHH LM XW. **Writing – review & editing:** GHH LM XW.

# Introduction The Indian economy is heavily reliant on the agricultural sector. More than 70% of rural households are dependent on agriculture. Given that it accounts for 20% of the country’s GDP and employs more than 60% of the workforce, India’s agricultural industry plays a crucial role in the nation’s economy. Over the past few years, Indian agriculture has experienced significant growth. Processing, floriculture, seed production, mushroom cultivation, nursery preparation, post-harvest production etc., all fall within the horticultural umbrella, making it an essential and crucial component of stabilising farmers’ income. The scenario of horticultural crops in India has become very encouraging. During 2020, according to the National Horticulture Board (NHB), the production of horticultural crops in India was 320.77 million tonnes from an area of 26.46 million hectares. It is further expected to grow at the rate of 1.8% in financial year 2021 with the area expected at 27.17 million hectares and production 326.58 million tonnes. Bananas are widely grown in the tropics and subtropics. It is the world’s seventh most traded agricultural product, behind wheat, maize, soybeans, rice, barley, and sugar. Annually, 14.2 million tonnes of banana are produced in India, which is more than any other country in the world. After the mango, the banana is India’s second most popular fruit. Tamil Nadu, Andhra Pradesh, Maharashtra, and Karnataka contribute to a significant percentage of banana production in India. Bananas have long been revered for their many cultural, medical and dietary benefits. It is great source of carbohydrates (22.84 g/100 g), provides around 370 kJ of energy per 100 grams and also considered as one of the best sources of potassium (358 mg/100 g) that provides 8 percent of the recommended daily intake. India is the world’s largest producer and exporter of bananas with an estimated 884 thousand hectares under cultivation and production of about 30 million tonnes (MT), contributing around 26% to the global banana basket. India holds a significant position among the top five banana-producing countries, accounting for approximately 50% of the total share (FAOSTAT 2019). Additionally, Gujarat, as the second-ranked state in banana production within India, contributes approximately 24% of the total share among the top five Indian states, as reported by the National Horticulture Board (NHB). In Gujarat, the area under production is 69.537 thousand hectares with production 4627.52 metric tonnes (Directorate of Horticulture, 2021). According to market arrival statistics, the Rajpipla market (Narmada) accounts for 78% of Gujarat’s overall arrival, making it the state’s primary market for determining banana prices(Agriculture marketing website 2020). Naturally, anticipating what will happen next, rather than why something happened in the past, is one of the primary purposes of time series analysis. One can use a forecasting model based on the experience and external information. Auto Regressive Integrated Moving Average (ARIMA) is most widely used time-series models amongst the base statistical models. Recently, the Artificial Neural Network (ANN) model has gained much of attention as a potential replacement of traditional models for estimation and forecasting in economics and finances. Future price predictions are exceedingly difficult to make. There is a substantial body of literature on various techniques and predictors that can be added to those techniques in order to achieve higher accuracy. Farmers will be able to utilise this information to determine the best time to sell their crops. As a result, the frequency and accessibility of time- and location-based arbitrage should reduce the volatility of prices.One of the biggest obstacles to making reliable banana price predictions is the seasonality of banana price series. Given the complexity of the price series, several models have been established for capturing price behaviour, but there is no consensus between researchers as to which model is best. Numerous linear and nonlinear methods have been developed within the time series framework, including the Autoregressive Integrated Moving Average (ARIMA) model, the Seasonal ARIMA (SARIMA) model and the Generalized Autoregressive Conditional Heteroscedastic (GARCH) model. Multiple previous studies have attempted to predict agricultural commodity prices. The SARIMA model is supposedly superior to other price forecasting algorithms when it comes to predicting onion prices in Mumbai’s marketplaces. Application of the ARIMA model for forecasting agricultural productivity in India can be found in. Application of the SARIMA model for forecasting meat exports from India. Price volatility for agricultural commodities in India has been extensively studied in. Spot electricity price forecasting in Indian electricity market using autoregressive-GARCH models can be seen in. Machine Learning (ML) methods that have recently emerged under the data science has become dominant approach of modelling. Time series forecasting, in the fields of finance and economics, have greatly benefitted by its application and also has been applied to forecast the area, production and productivity of agricultural commodities like citrus, banana and mango. Several empirical studies have demonstrated that when forecasting various assets, ML algorithms outperform time series models. Comparison of efficacy of statistical models and machine learning techniques can be found in. It is reported that ANNs works well over the classical statistical methods such as linear regression and Box- Jenkins approaches. For both price and yield forecasting in agriculture, neural networks were shown to be more accurate than statistical techniques. Superiority of neural network in price forecasting, in percent losses of pods by pod borer and pod fly in pigeon pea, were discussed. While application of RNN model found in forecasting prices of arecanuts in Kerala and in agricultural product prices prediction. The primary goal of this study is to forecast the price of banana in Gujarat, India using ARIMA, SARIMA, ARCH, GARCH, ANN and RNN by using secondary data obtained from the Agricultural Marketing website, focusing on banana prices in the major market of Gujarat covering the period from January 2009 to December 2019. The research aims to identify the forecasting model that exhibits the highest level of prediction accuracy, assessed through performance metrics such as RMSE, MAPE, SMAPE, MASE and MAD. # Materials and methods The sample consisted of 132 observations (i.e., monthly data for 11 years). The data represents the modal prices of banana. One important characteristic of the dataset used in this investigation is that it spans over 11 years, which is helpful in analysing long-term trends and patterns in the market. However, the dataset has its own limitations of having missing observations at long length apart from potential errors in the data collection. Following analytical models were utilized in the present study: ## Statistical models ### ARIMA—Auto Regressive Integrated Moving Average model George Edward Pelham Box and Gwilym Meirion Jenkins proposed the ARIMA (p, d, q) model in the 1970s which is called Box-Jenkins model. In the ARMA model, a variable’s projected future value is a linear mixture of its previous lag value and error: $$y_{t} = \varnothing_{1}y_{t - 1} + \varnothing_{2}y_{t - 2} + \ldots + \varnothing_{p}y_{t - p} + \varepsilon_{t} - \varnothing_{1}\varepsilon_{t - 1} - \varnothing_{2}\varepsilon_{t - 2}\ldots - \varnothing_{p}\varepsilon_{t - p}$$ where, *y*_*t* is the actual value at t, {*ε*_*t*} is the white noise sequence, *p and q* are integers which are called autoregressive and moving average, respectively. When dealing with a non-stationary time series, difference can be made to make it stationary series. ### SARIMA—Seasonal Auto Regressive Integrated Moving Average model Box-Jenkins generalized this model to deal with seasonality. This model uses seasonal differencing to remove non-stationarity. SARIMA (p, d, q) \* (P, D, Q)^S is presented below in terms of Lag polynomial: $$\Phi_{P}(L^{S})\varphi_{P}(L){(1 - L)}^{d}\left( {1 - L^{S}} \right)^{D}y_{t} = \theta(L^{S})\theta_{q}(L)\varepsilon_{t}$$ Here *L*, *L*^*S* are non-seasonal and seasonal differencing, *φ*_*P* *(L) & θ*_*q* *(L)* are parameters of *“p”* and *“q”* for non-seasonal lag value, *Φ*_*P**(L*^*S**) and Θ(L*^*S**)* are parameters of *“P” and “Q”* for seasonal lag while *y*_*t* and *ε*_*t* are original time series and error at given time t. ### ARCH—Autoregressive Conditional Heteroscedasticity model Autoregressive conditional heteroscedasticity (ARCH) models predict conditional variances. The ARCH model comprises two primary components: the mean equation and the variance equation. The mean equation specifies the conditional mean of the series and typically incorporates autoregressive or moving average terms. On the other hand, the variance equation models the conditional variance of the series. Engle (1982) first suggested ARCH models, later generalised as GARCH. ### Generalized Autoregressive Conditional Heteroscedasticity (GARH) models GARCH models offers enhanced capabilities in capturing the intricate dynamics of volatility. Then GARCH model (p, q) by Bera, A. K. & Higgins, M. L. (1993) is: $$\sigma_{t}^{2} = \alpha_{{^\circ}} + \alpha_{1}e_{t - 1}^{2} + \cdots + \alpha_{q}e_{t - q}^{2} + \beta_{1}\sigma_{t - 1}^{2} + \cdots + \beta_{p}\sigma_{t - p}^{2}$$ Where α° \> 0, term is the constant in the model, which represents the long-run average of the conditional variance while $$\alpha_{i} \geq 0\ for\ i = 1,\ldots.,q;\ \beta_{i} \geq 0\ for\ i = 1,\ldots.,p$$ are imposed to guarantee that the conditional variance is non-negative. ### Exponential Generalized Autoregressive Conditional Heteroscedasticity (EGARCH) model The exponential GARCH can be written as follows: $$\varepsilon_{t} = \sigma_{t}z_{t}$$ $$ln\sigma^{2} = \omega + {\sum_{i = 1}^{p}{\alpha_{i}\varepsilon_{t - i}^{2} + {\sum_{j = 1}^{q}{\beta_{j}ln\sigma_{t - j}^{2}}}}}$$ Because of the log variance, this model differs from the GARCH structure which captures asymmetrical consequences of shocks. In the financial literature, the following specification has also been used (Ali 2013). <img src="info:doi/10.1371/journal.pone.0275702.e007" id="pone.0275702.e007g" /> ε t = σ t z t <img src="info:doi/10.1371/journal.pone.0275702.e008" id="pone.0275702.e008g" /> l n σ 2 = ω \+ α i ε t − i 2 \+ ∑ j = 1 q λ j ln ( σ t − j 2 ) \+ ∑ i = 1 p γ i ( \| ε t − i \| σ t − i − 2 π ) Where *z*_*t* represents the standardized residual at time t, *α*_*i*, *λ*_*j*, *γ*_*t* are coefficients of lagged squared residuals, lagged logarithms of the squared conditional standard deviations and standardized residuals. ## Deep learning artificial intelligence ### Artificial Neural Network (ANNs) ANNs are the cutting-edge machine-learning algorithm. Features of ANN are its inherently non-linearity, data-driven and self-adaptive approach and universal approximate function. The output of the ANN model is computed using the following mathematical expression: $$y_{t} = a_{{^\circ}} + {\sum_{j = 1}^{q}{a_{j}g\left( \beta_{{^\circ}j} + {\sum_{i = 1}^{p}{\beta_{ij}y_{t - i}}}n \right) + \varepsilon_{t,}}}$$ Here *y*_*t−i* (i = 1, 2,…, p) are the *p* inputs and *y*_*t* is the output. The integers *p*, *q* are the number of input and hidden nodes respectively. *a*_*j* (j = 0,1,2,………q) and *β*_*ij* (i = 0,1,2,………q) are the connection weights and *ε*_*t* is the random shock. *a*_*j* and *β*_°*j* are the bias terms and *g(x)* is the nonlinear activation function. The architecture of feed forward neural network is described in. ### Recurrent Neural Networks (RNNs) The RNN’s goal is to process sequences of data. Unlike the traditional neural network architecture, which connects the input layer to the hidden layer to the output layer in a fully connected manner with no inter-layer connections, this ordinary neural network is inadequate for many problems. RNN, on the other hand, is referred to as a recurrent neural network since the current output of a sequence is influenced by the previous output as given in. This implies that the hidden layer nodes are no longer isolated but rather interconnected and the input to the hidden layer includes not only the input layer’s output but also the previous hidden layer’s output. RNNs are a type of artificial neural network that is designed to work with time series data that contains sequences. It has the concept of "memory," which stores the information of previous inputs in order to get the next output. A feedback loop is present in a basic RNN as illustrated in. The equation of RNN can be expressed as: $$h_{t} = f\left( W_{xh}X_{t} + W_{hh}H_{t - 1} \right.$$ $$O_{t} = f\left( W_{hy}S_{t} \right)$$ *X*_*t* is input at time t, *H*_*t*−1 is state at time *t-1*, *W*_*xh*, *W*_*hh* *and W*_*hy* are weight matrices for input, hidden and output layers. ### Accuracy measures All the models were evaluated based on the Root Mean Square Error (RMSE), Mean Absolute Percent Error (MAPE), Mean Absolute Scaled Error (MASE), Symmetric Mean Absolute Percentage Error (SMAPE) and Mean Directional Accuracy (MDA) having details mentioned in. # Result and discussion ## Result ### Data pre-processing Data pre-processing and statistical investigation for prices of banana in Gujarat was done before using it for forecasting models. The dataset of banana prices in Gujarat has mean Rs. 806.76/quintal while 2nd quartile Rs. 800.00/quintal, reflects slight deviation from non-normality. Also, banana price ranges from minimum of 183.33 (Rs. /quintal) to maximum of 1533.33 (Rs. /quintal) indicating high variation exist among the dataset. Furthermore, data exhibit high degree of asymmetry and is characterized by a leptokurtic shape. Time series price movement over the period of January, 2009 to December, 2019, shown in, signifying high prices mostly occur in the month of Feb–April (colour palette 2 to 4) every year, which reflects seasonal impact. The first subplot of reveals the original time series graph depicting the fluctuations in banana prices in Gujarat. It highlights a significant price surge from 2009 to 2011, followed by a decline between 2018 and 2019. These fluctuations introduce non-linearity into the dataset during that specific period. Further, the time series data of banana prices is decomposed into trend, seasonal and residual components using additive seasonal decomposition techniques in the subplot 2^nd, 3^rd and 4^th of. The 2^nd subplot is basically reflecting the long-time trend of banana prices with some shift in the initial as well at the end year and that is because of a lot of missing data in the initial year while the distorted price in 2019, as mentioned on the Agmarknet website. Additionally, the 3rd subplot reveals a noticeable pattern of seasonality, specifically in the months of February through April, as well as during September to October. Last residual subplot is basically showing variations in the prices of banana which is quite high for the period of 2009–2011 and 2018–19. is reflecting the strength of relationships among multiple lag prices of banana. The upper triangle of pair plot is indicating that previous four month’s prices are highly correlated with current prices indicating serial correlation and non- randomness among the price with its multiple lags. In the diagonal plot, kernel density estimates (KDE, univariate) are plotted for current price as well for previous month (Lag 1 to Lag 4) indicating that distribution of data is not normal while in the lower triangle of pair plot, kernel density estimates (KDE, Multivariate) of prices are represented as a contour plot depicting the relationship of the non-normal probability distribution between those two variables. The plots of Auto correlation function (ACF) and Partial Auto correlation function (PACF) were generated in, for determining the stationarity of time series data of banana prices in Gujarat, which confirms that ACF of the price lag reflect a slow decline, indicating the presence of non-stationarity. This is supported by the non-significant p value of the Augmented Dickey Fuller (ADF) test statistics in indicating acceptance of null hypothesis. Therefore, differencing is required to make it stationary which becomes stationary after one differencing. The linearity of fitted model of time series data was judged based on Lagrange’s Multiplier test. displays the test statistic values at various lags, indicating a high level of significance at a 5% level of significance. This implies a robust presence of volatility impact on the banana price series. This description suggests that data exhibits non-normality, strong autocorrelation among lagged prices, seasonality, a large dispersion potentially caused by shifts, non-stationarity, slight skewness and leptokurtic distribution. These characteristics must be considered during the development of any models. ### Model building shows the train-test split of the datasets as well as the period of forecasting for Banana. The model’s robustness was assessed by developing it on training data points, as specified in. Subsequently, the performance of model was examined on testing data points. The forecasting period for the model was from January 2020 to December 2020, considering data up to the year 2019. The generated forecast for this period was compared against the actual modal prices that were made available on the website during the completion of the study. The simulation procedure was used to assess the reliability of the model. ### Model performance The forecasting model like ARIMA, SARIMA, G/ARCH, ANN and RNN as described in methodology sections have been fitted for the training data under consideration. Selection of different model/architecture were solely based on lowest accuracy measures like MSE, RMSE, MAPE, SMAPE and MASE generated on testing data set was considered and are mentioned in. It shows that among the five models examined, the Recurrent Neural Network (RNN) model has the lowest absolute percentage error (MAPE) of 9.58, indicating its superior suitability for the problem. Additionally, the RNN model exhibits an impressive Mean Absolute Scaled Error (MASE) of 0.12, which highlights its effectiveness in improving accuracy by almost 10 times when compared to the Naïve model. In contrast, the ARIMA, SARIMA and ARCH models demonstrate poor fit, with their respective MASE values exceeding 1. The RNN model outperforms its counterparts and provides the most accurate predictions for the test dataset. In, the predicted and observed values of the test data set were plotted for each model. The graph depicted in provides insight into the predictive capability of RNN model over other models on test data. It presents a polar chart illustrating the scaled accuracy measures of five models on test data of banana prices in Gujarat. The RNN model surpasses remaining five models across various error metrics, including MSE, RMSE, MAPE, SMAPE and MASE. Moreover, the fitted models were leveraged to make forecast for the upcoming 12-month period for the year 2020, encompassing the months that have been adversely impacted by the COVID pandemic. The actual reported value for the given time interval was compared with the projected value derived from a range of time series models. In situations where other methodologies faltered, recurrent neural networks (RNNs) outperformed others in predicting the prices of bananas. presents a comprehensive breakdown of the accuracy measures for each model during the forecast period. It is evident that RMSE, MAPE, SMAPE and MASE decrease significantly as the models move from ARIMA to RNN, over the forecasting time span. Furthermore, mean directional accuracy (MAD) demonstrates that RNN outperforms other models in predicting the direction accuracy. presents a comparison of the forecasting performance of various models over the period of January 2020 to December 2020. It shows how closely each model’s forecast aligns with its corresponding actual reported value. The figure serves as a reliability check of model during forecast period from January to December 2020 which is basically the COVID 19 period. Additionally, a polar chart, displayed in, shows scaled measures for model accuracy, revealing negligible error and hence better accuracy for RNN across all five measures. Furthermore, a paired t-test was performed to determine the statistical significance of the difference between the forecasted price generated by the model and the actual prices during the forecast period. shows that there was no significant difference between the forecasts of the RNN, while the ARIMA and SARIMA showed significant differences, indicating their poor fit. ### Model fit parameters This section describes the parameters and hyperparameters of the RNN model, which was identified as the most suitable model among all other models *i*.*e*. ARIMA, SARIMA, ARCH and ANN. presents the details of the optimal RNN model’s architecture and its parameters. The proposed RNN architecture as mentioned in, aims to address overfitting through a combination of parameter choices and regularization techniques. The model comprises an input layer with 3 input values (previous lag value of price), a hidden layer with 10 neurons and an output layer with a single neuron. The ReLU activation function is employed in the hidden layer, promoting non- linearity and feature representation while a linear activation function for output layer is used to allow direct prediction of continuous values. To prevent overfitting, L1 regularization is applied with a coefficient of 0.001, encouraging sparse weight values and reducing dependence on specific features. Additionally, dropout is implemented with a rate of 0.2, randomly deactivating hidden units during training to reduce dependence on individual neurons. # Discussion The findings indicate that the ARCH/GARCH statistical models performed better than the ARIMA and SARIMA models. One suggestion put forth by certain academic experts is to combine the ARIMA model with the GARCH model to overcome the limitations of linear models. However, even with this approach, there remain various factors contributing to volatility that cannot be effectively addressed by these conventional models alone. In such cases, the utilization of advanced machine learning models could prove invaluable in capturing and addressing the full spectrum of variables affecting volatility. Nonetheless, statistical models ARIMA/SARIMA and ARCH/GARCH were unable to compete with the neural network models, which produced more accurate results. This discovery is consistent with the study conducted by, in which they concluded that the artificial neural network (ANN) model is a significant substitute for theoretical models in anticipating the rainfall-runoff dataset. Additionally, out of two neural network architecture, RNN performed better than ANN. Non-linearity as well as time sequence data, give Recurrent Neural Network, more power. The same is supported by the finding in where accuracy of deep- learning RNN methods are better and more accurate than ANN while simulating the streamflow of reservoir. However, some previous studies have found contradictory results. For instance, a study by found ARIMA outperformed ANN in predicting stock prices while in GARCH models outperformed deep learning models in predicting the volatility of the Shanghai Composite Index. Thus, methods like S/ARIMA, G/ARCH and ANN have proven to be valuable initial approaches, it is crucial to acknowledge their inherent limitations due to their data-specific nature and supplement them with more comprehensive techniques having capabilities to deal with dynamic data when necessary. # Conclusion After analysing various models to anticipate banana prices in Gujarat for the COVID period from January 2020 to December 2020, our study found that the Recurrent Neural Network (RNN) outperformed all other models in terms of RMSE, MAPE, SMAPE, MASE and MAD values, showing its ability to handle unexpected events and their impact on future prices. Our results suggest that the RNN model can aid policymakers in improving their decision-making processes, leading to increased profitability. The practical applications of this research findings include the development of tools and applications that farmers, traders and end- users can use to access the forecasted prices of bananas in Gujarat. Policymakers can also use the results to address the challenges that farmers and traders face in the market. However, our study also has limitations like it only considers its own price lag value and haven’t incorporate other essential factors such as weather, commodity arrival, demand & supply, government export-import policies and commodity diversity into the forecast. As a result, future research should explore different deep learning architectures which incorporate these variables to improve prediction accuracy. Addressing these limitations would lead to more comprehensive and nuanced conclusions in this study, along with practical applications for farmers, traders, and end-users in the horticultural commodity market. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Three randomized controlled trials in Kisumu, Kenya, Rakai, Uganda, and Orange Farm, South Africa showed that medical male circumcision reduces the risk of HIV-acquisition by 60%., Male circumcision is now a recommended component of a comprehensive prevention package for men in communities with high HIV- prevalence, low circumcision rates, and where heterosexual sex is the primary mode of transmission. Circumcision involves the surgical excision of the prepuce which results in a circumferential, full-thickness cutaneous, incisional wound near the junction of the glans and the shaft of the penis. Wound healing is initiated via primary intention, wherein the edges of the wound are pulled together, closed, and sutured at the time of surgery. While voluntary medical male circumcision (VMMC) has been shown to be protective in preventing HIV- acquisition, there is concern that the procedure may place men and their partners at higher risk of HIV-acquisition in the period immediately following the circumcision before the wound is healed. During the inflammatory and early proliferative stages of healing, the wound is open to the environment and the integrity of the dermis is compromised. If sexual activity is initiated during these early phases, HIV-negative men may be more susceptible to acquiring HIV as the wound provides an easy portal of entry in an area with a potentially high- concentration of HIV-target cells., Similarly, HIV-positive men with unhealed wounds may have high viral shedding, placing their sex partners at high risk of infection. To allow for complete healing, the World Health Organization recommends 42-days of post-circumcision abstinence.. To date, no published study has systematically investigated the length of time required for the wound to fully heal following circumcision. The Kisumu trial reported complete healing in 98.7% of men returning for their 30-day follow-up visit, and Rakai reported 88.4% of 18–24 year old men and 86.1% of 15–49 year old individuals with complete healing at four weeks post surgery. Neither of these trials used a stringent follow-up protocol for assessment of wounds; the timing of observations of wounds classified as either 30-days or 4-weeks after surgery varied widely, making true assessment of proportions healed at specific time points difficult. There is concern that the 42-day post-circumcision abstinence period may discourage men, particularly those in committed relationships, from seeking circumcision services. Studies of Kenyan men having undergone VMMC have revealed that between 30.9% and 37.7% of men resumed sex before the recommended six weeks abstinence period had elapsed., Higher proportions of married than unmarried men have been shown to resume sex early, prior to the completion of the recommended post-surgery abstinence period., Men’s resumption of sex earlier than recommended may suggest that the imposed abstinence period could be a barrier to uptake of circumcision services, especially among married men or those with a steady sexual partner. Conversely, because we do not yet know the true proportion of men healed by 42 days post- surgery, the recommended abstinence period may be insufficient to protect men and their partners from increased exposure to HIV. In a study of HIV- transmission among serodiscordant couples, Wawer et al reported nearly a three- fold increase in HIV-incidence among female partners of men who resumed sex before certification of healing (RR 2.92 95% CI 1.01–8.46). Additionally, Kigozi et al report a significant difference in healing rates between HIV-positive (92.7%) and HIV-negative (95.8%) men at 6-weeks following surgery (p\<0.007). Because of the potential risk to seronegative partners, the possibility that HIV-positive men heal slower than HIV-negative men, and that UNAIDS recommends HIV-positive men be provided VMMC on request if there are no contraindications, we have included HIV-positive men in this study. This prospective cohort study was undertaken to evaluate post-circumcision wound healing in HIV-positive and HIV-negative men. We examined demographic, behavioral, biological, and surgical factors that influence healing progression with the goal of determining whether the recommended 42-day post-circumcision abstinence period, currently recommended by the World Health Organization and employed by the Kenyan Ministry of Health, is optimal.. # Methods ## Participants The study was conducted at the Universities of Nairobi, Illinois, and Manitoba (UNIM) Clinic within Lumumba Health Center in Kisumu, Kenya. Participants were selected from men seeking voluntary medical male circumcision (VMMC). These men received a study flier that expressed the goals of the Wound Healing Study (WHS) and were told that their eligibility for VMMC would not be affected if they chose not to participate in the WHS. Men expressing interest in the WHS then underwent voluntary HIV testing and counseling (VCT) and were screened for circumcision status. Men aged 18–35 years were eligible for enrollment if they were uncircumcised, had no contraindications for circumcision, were a resident of Kisumu, and intended to stay in Kisumu for a period of 3-months following VMMC. Consistent with Kenyan VMMC guidelines, HIV-positive men with signs of advanced HIV- infection (WHO Stage 3 or greater) were excluded. Because age may be associated with time to healing and HIV-status, we matched HIV-positive individuals to an HIV-negative individual of the same age +/−2 years. Because most men in VMMC programs are HIV-negative, additional HIV-negative men were enrolled to increase precision in this group. Analysis of all the HIV-negative men was performed separately. This study was approved by the Institutional Review Board at the University of Illinois at Chicago, USA and the Ethics and Research Committee at Kenyatta National Hospital, Kenya. All participants provided written informed consent. ## Procedures After giving informed consent, enrolled men underwent the same pre-operative assessment that is in place for Kenya MOH circumcision programs. Additionally, men were tested for HSV-2 serostatus, random blood sugar level, hemoglobin level, serum albumin level, and were given an extensive interview on behavioral and demographic characteristics. HIV-positive men also had blood drawn for CD4+ T cell count. Men who completed the enrollment interview and passed the pre- operative assessment, which included a physical examination, were then circumcised by one of four clinicians (2 clinical officers and 2 nurses). In order to control for surgeon experience, a possibly influential covariate with time to healing, each of the study surgeons was hired because he had previously performed several hundred circumcisions either under the Kenya MOH program or as part of previous studies at the UNIM clinic. All surgeries were performed using the forceps-guided method for male circumcision.. Following circumcision, men were instructed to return on a specific date every 7-days from the date of operation for seven weeks and then a final visit at 12 weeks. At each follow-up visit the wound was assessed for progression through the healing process, photographs were taken to document healing and adverse events, and the participant was interviewed to collect data on post-operative behavior and satisfaction with the healing process. Visits were eligible to be included in analysis if they occurred on the scheduled day +/−2 days. If a participant missed his scheduled follow-up visit, a trace was implemented to locate the individual and ensure his attendance the following day. In this way, we were able to complete 97.1% (2510/2584) of scheduled visits. The primary outcome of complete wound healing was assessed by clinicians using the following operational definition: healthy scar formation with no scab or opening along the incision line. Prior to beginning the study and to ensure adherence to the definition of a completely healed wound, our clinicians underwent extensive training and pilot testing where individuals with wounds of varying post- operative age were assessed by each clinician and discussed as a group. ## Statistical Analysis A participant could not be certified as healed until he was examined by one of the study clinicians. The number of post-operative days elapsed at the time of certified complete healing was recorded for each individual. Clinic visits proceeded after certified complete healing and as such it was possible for a participant to be certified as completely healed at one visit and not be certified as completely healed at the next visit. This scenario occurred in 11 of 323 individuals. For analysis purposes, an individual could not be considered completely healed until he was certified healed at all subsequent visits. Censored individuals were those that were either lost-to-follow-up before certified healing or that completed the full 12 weeks of follow-up without accomplishing complete healing, which occurred in just one individual. We performed two separate series of analyses. First, we examined the relationship between HIV-status and time to complete healing in an age-matched cohort of 108 HIV-positive and 108 HIV-negative individuals. Second, we examined time to healing in 215 HIV-negative men. Our sample size of 108 pairs was sufficient to detect a 12% or greater difference in the proportion of completely healed men between the two groups at alpha = 0.05 and power = 80%. Kaplan-Meier methods were used to assess time to complete wound healing in the matched cohort stratified by HIV-status. Cox proportional hazard models were used in both series to examine the relationship between *a priori* identified potentially important covariates and the hazard of healing at any time point. These covariates fall into four categories: demographic, biological, behavioral, and surgical. Demographic characteristics considered were age and marital status. Biological characteristics were: HIV-status, HSV2 status, baseline random blood sugar, hemoglobin, serum albumin, and the presence of a post-operative infection in the first 3 weeks of follow-up. Behavioral characteristics were: baseline alcohol consumption (days/week), physical activity in the first week following surgery (riding a bicycle, digging, and walking long distances), and time of onset of sexual activity. Surgical characteristics were: amount of dermis exposed at week 1 (in total mm), evidence of tight sutures at week 1, surgical time as a proxy for difficult surgery, and surgeon cadre (clinical officer vs. nurse counselor). We investigated each of the covariates individually in separate univariate models to explore the reduction in −2 log likelihoods using a one-degree of freedom chi-square test. Covariates significantly reducing the model deviance at the p\<0.05 level were then included in a multivariate Cox Proportional Hazard model. Covariates failing to meet this criterion were dropped from further analysis with the exception of the primary exposures of interest: HIV-status for matched analysis and age for HIV-negative analysis. We applied this strategy with both the age-matched analysis of HIV-positive and HIV-negative men and with the analysis of HIV-negative men only. In each analysis, we modeled the hazard of healing and thus hazard ratios less than 1.0 are indicative of delayed healing. Analyses were performed using SAS version 9.2 (SAS Institute, Cary, NC). # Results ## Participant Characteristics A total of 215 HIV-negative and 108 HIV-positive men aged 18–35 years (median 26 years, IQR 23–30) were enrolled. Nearly half of all men (46%) were married; 58% of HIV-positive men were married. HSV-2 serology was positive in 75% of HIV- positive men and 36% of HIV-negative men. Almost all men (95%) had random blood sugar levels in the normal range (70–139 mg/dL); 91% of men had normal hemoglobin levels (≥13.0 g/dl); and 94% of all men had serum albumin levels in the normal range (3.40–5.40 g/dL). Among HIV-positive men, 46/108 (43%) had CD4 counts below 350 cells/µL. Each of the 323 participants was scheduled for weekly post-circumcision visits till week 7 and at week 12 giving a total of 2,584 expected visits, of which 2,510 (97.1%) were completed. Baseline characteristics are provided in. ## Time to Complete Healing provides the results of the Kaplan-Meier analysis for the age-matched cohort. No statistically significant difference in time to complete healing by HIV-status was observed (log-rank test = 0.69, p = 0.41). Mean time to complete healing was 33 days for HIV-positive individuals, 31 days for HIV-negative matches, and 31 days for all HIV-negative individuals. One HIV-negative individual failed to accomplish complete healing by the end of the follow-up period at 12 weeks due to an adverse event (sub-dermal hematoma) in the first post-operative week. This was the only severe adverse event observed in the 323 individuals (0.3%). By the week 4 visit, 64.7% of all men were completely healed. This figure rose to 83.1% at week 5, 94.1% at week 6, and 96.6% at week 7. The largest disparity in complete healing between HIV-negative and HIV-positive men occurred at week 4 when 70.4% of HIV-negative and 59.3% of HIV-positive men were healed (p = 0.09). By week 6 cumulative percentage of healed individuals reached 93.3% in HIV- positive men, 92.6% in age-matched HIV-negative men, and 94.4% in all HIV- negative individuals. Mean time to complete healing was 34.5 days in HIV- positive men with baseline CD4 counts less than or equal to 350 cells/µL and 31.9 days in men with baseline CD4 counts greater than 350 cells/µL (p = 0.20). Median time to complete healing was 28 days for both groups. Among ARV-naïve HIV-positive men, 59.7% (40/67) had baseline CD4 counts greater than 350 cells/µL. Mean time to complete healing was 31.1 days in this group versus 37.1 days in ARV-naïve men with baseline CD4 counts less than or equal to 350 cells/µL (p = 0.04). There was no difference at week 6 between those with CD4 counts above versus those below 350 cells/µL (94.7% vs 88.9%; p = 0.64). ## HIV-positive Versus Age-matched HIV-negative Men In the age-matched cohort of HIV-positive and HIV-negative men, mean time to complete healing was 45.9 days in individuals presenting with an early post- operative infection versus 30.9 days in men who did not (p\<0.001). Under univariate analysis early post-operative infection resulted in a significantly reduced rate of complete healing (HR 0.53, 95% CI 0.32–0.89). At six-weeks post- circumcision, 72.2% of men with early post-operative infections were healed versus 94.9% of men without infections (p\<0.001). Men presenting with evidence of tight sutures at week 1 also had reduced rates of healing (HR 0.65 95% CI 0.42–0.996), but at six-weeks post-circumcision the difference in proportions healed between men without and with evidence of tight sutures was non- significant (94.2% vs. 83.3% p = 0.07). This variable, along with early post- operative infection, was included in multivariate modeling. HIV-status, although not significantly associated with time to healing, was also included in the multivariate model. In the multivariate Cox proportional hazard model, early post-operative infection was associated with a reduction in the rate of healing (HR 0.52 95% CI 0.31–0.87) and men with evidence of tight sutures also experienced reduced rates of healing (HR 0.63 95% CI 0.41–0.98). ## HIV-negative Men Exploration of age in HIV-negative men as an explanatory factor for either accelerated or reduced rates of healing yielded non-significant results (HR 0.91 95% CI 0.69–1.19). As with the age-matched analysis, early post-operative infection was associated with reduced rates of healing (HR 0.48 95% CI 0.23–1.00). Mean time to healing was 30.2 days in men with no infection and 46.7 days in men presenting with an infection (p = 0.005). At six-weeks post- circumcision, 95.6% of men without an infection were certified healed versus 66.7% of those with early post-operative infections (p = 0.009). In multivariate Cox proportional hazard models early post-operative infections maintained a significant reduction in healing rates while adjusting for age (HR 0.48 95% CI 0.23–1.00). Unlike the age-matched analysis, tight suturing was not associated with delays in healing (HR 0.80 95% CI 0.53–1.22). # Discussion We found that 94% of men are completely healed within the 42 day time period recommended by WHO for post-surgical abstinence. Our finding of 64.7% healed at week 4 is lower than the proportion healed at 30-days in the Kisumu trial (98.7%) and week 4 in the Rakai trial (86.1%). Neither of these trials used a stringent follow-up protocol for assessment of wounds; the timing of observations classified as 30-days or 4-weeks post-surgery varied widely. In the Kisumu trial, 5.3% of wounds were assessed at or before 27 post-operative days, 89.7% were assessed at between 28 and 34 days, and the remaining 5.0% were assessed after at least 35 days. Similarly, in the Rakai trial, among all participants, 25.5% of men were seen at or before 27 days, 69.5% at 28–34 days, and 5.1% after 35 days. In the Kisumu trial, suturing was performed using 4.0 sutures, which may provide better apposition of the wound edges compared with the 3.0 sutures used in this study and in the Kenyan national VMMC program. We also used a slightly different definition of complete wound healing from the trials. In the Kisumu trial, a completely healed wound was defined as a wound without a scab or open wound and no evidence of swelling or redness. The Rakai definition emphasized the formation of an intact, healthy scar with no residual exudate or scab formation and the complete absorption of all sutures. Our definition, healthy scar formation with no scab or opening along the incision line, was formulated by the clinical team. Color was eliminated from the definition as we found wide variability in the color of healthy, fully healed tissue during pilot testing. Having all the sutures absorbed was unnecessary language, since this occurs early in the post-circumcision period. Therefore, differences between our findings and those reported for trial participants may be a function of different definitions of complete healing, the requirement in our study that men were seen in seven-day intervals (+/−2 days), and the requirement that a man be considered healed only after being recorded as healed at all subsequent follow-up visits. Kigozi et al report that 92.7% of HIV- positive men and 95.8% of HIV-negative men were healed at 6-weeks following surgery (p\<0.007), results that closely approximate our findings, although our rates of 93.3% in HIV-positive and 94.4% in HIV-negative men were not significantly different. In assessing whether a change in the WHO guidelines is recommended, the results from this wound healing study suggest that the 42-days of recommended post- circumcision abstinence is reasonable and should be maintained. While most men are healed at week 4 and week 5, there is significant healing that takes place between 28 and 42 days; nearly thirty-percent of men healed in this period (29.6%). Additionally, since 94% of men are healed by 42-days, extending the abstinence recommendation to 7 weeks or beyond may present an additional barrier to uptake of circumcision with limited returns in the form of reduced HIV- incidence. Rather than extend the abstinence period, it would be prudent to reinforce counseling for clients and to stress consistent condom use when sex is resumed. Our results are specific to the forceps-guided method of surgical circumcision. Recent attention to the use of devises for circumcision may require that we consider the WHO guidelines for post-circumcision abstinence in the context of these devices. A recent randomized controlled trial of the Shang Ring versus surgical circumcision in Kenya and Zambia reported 76% of men undergoing the Shang Ring procedure were healed at 42-days post-circumcision, compared to 85% of men undergoing surgical circumcision. With nearly a quarter of the Shang Ring population unhealed at 42-days, a revision to the WHO recommended post- circumcision abstinence period may be necessary. In a study of the Prepex device for male circumcision in Rwanda, the authors report that “the number of days required for complete healing” following device removal was 31 days, or approximately 38 days from device placement. Assuming that the authors are reporting median time to healing, circumcision with Prepex may require a longer healing period than either the forceps-guided method or the Shang Ring. This may be expected as circumcision with the Prepex device requires healing by secondary intention. Unfortunately, the authors fail to report healing rates at 42-days post-circumcision. We found that HIV-status was not associated with a significantly longer healing period in this population. However, within ARV-naïve HIV-positive men, those with CD4 counts of less than 350 cells/µL experienced slower healing than those with counts above 350 cells/µL. These findings suggest that men of known HIV- positive status who are not on ARVs should be counseled to refrain from sex for longer than the 42 day recommended period and, as with all men, counseled to use a condom with resumption of sex. All other patient-centered covariates were not significantly related to time to healing. Variables associated with surgery proved most influential; namely, evidence of tight suturing and post-operative infection, the later being related perhaps to poor hygiene and post-operative wound care or contamination at the time of surgery. In analysis of HIV-negative men the variable most associated with hazard of delayed healing was post-operative infection. We failed to account for hygiene covariates during the course of the study. Carriage of the penis following surgery, where the dorsal surface is held against the lower abdomen to avoid postural edema around the wound, and post- surgical washing may influence the wound’s ability to heal. Investigating these covariates will be helpful for further understanding of post-circumcision healing. We introduced selection bias to the study by age-matching HIV-negative men to enrolled HIV-positive men. HIV-status is linked to age in this population. As a result, we have an HIV-negative study population that is older than what is reflected in men seeking circumcision under the Kenya National VMMC Program. Although our clinicians are instructed to follow a strict protocol with regards to assessment of each wound, there exists the possibility that the number of completed weeks of follow-up influenced the clinicians’ assessment of each wound. We have accounted for this possibility by requiring that each wound be certified healed at all subsequent study visits. Additionally, because the clinicians conducting the wound assessments were the same as those performing the surgery, it is possible that they may be biased towards reporting earlier healing. However, each participant could see any one of the four clinicians at each visit. Also, it was not feasible to employ additional clinicians for independent follow-up assessments. We required our clinicians to attend an extensive training program and adherence to the definition of complete wound healing was piloted repeatedly before onset of the study. Infection in the first three weeks post-circumcision and evidence of tight suturing at the first follow-up visit were associated with delayed healing. When these events are observed after surgery, clients should be counseled that they may need to remain abstinent beyond 42-days and that they should use condoms upon resumption of sex. Additionally, HIV-positive men who are ARV-naïve might be counseled to refrain from sex longer than 42-days. As with counseling for all men undergoing circumcision, emphasis should be placed on condom use at resumption of sex. This is the first study to assess systematically wound progression following voluntary medical male circumcision by forceps-guided method. Based on our results, the WHO recommendation for 42-days post-circumcision abstinence should be maintained and is appropriate for HIV-positive and HIV-negative men since we found no difference in time to complete wound healing between these groups. We have identified factors associated with delayed wound healing - post-operative infection, tight sutures, and low CD4 counts in ARV-naive HIV-positive patients - that should be taken into account in training for and implementing VMMC programs. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JHR EOJ RCB. Performed the experiments: JHR EOJ RCB. Analyzed the data: JHR. Wrote the paper: JHR EOJ WJ RCB.

# Introduction Between the late 1970s and 2000, the western distinct population segment (WDPS) of Steller sea lions (*Eumetopias jubatus*) declined by more than 80% in the Aleutian Islands and Gulf of Alaska (GOA) and was listed as endangered in 1997. The designation led to years of unprecedented federal funding for studies aimed at determining the cause(s) of the decline and the reason(s) for a lack of recovery. The impetus derived in major part from two related factors: 1) the importance of walleye pollock (*Theragra chalcogramma*) to the nutritional status of the animals—pollock is widely consumed by Steller sea lions, and 2) pollock is the target of the largest single-species commercial fishery in the world, with an exvessel value in the order of half a billion dollars U.S. Yet despite the massive monetary expenditures and scientific effort, no consensus of opinion has emerged about the cause of the decline. However, two general classes of hypotheses have been proposed: top-down forcing, primarily by predation, ; and bottom-up forcing through changes in prey resources due to climate change or competition with commercial fisheries. The reproductive rate (natality—the number of young produced per reproductively mature female) of animals is an important life history characteristic and can be an indicator of nutritional status. In the context of bottom-up control of population dynamics, reduced natality of Steller sea lions and low juvenile and adult survival, due to poor nutrition are believed by some to have been the causes of the population collapse. Since 2000, some parts of the WDPS have experienced modest increases in abundance. Inferential population dynamic models based on census counts of Steller sea lions indicate that the recent small increases are related to improved juvenile and adult survival, but that natality continued to deteriorate during the 1990s and 2000s. Natality in the Central GOA was estimated to be 67% during the 1970s, 55% in the 1980s, and just 43% in the 2000s. Estimates of natality in the 1970s and 1980s were obtained by collecting females during early (October – November) and late (April – May) gestation and determining the proportion that were reproductively mature with a developing fetus. Sources of error in natality calculations using those methods would have included variation in the status of females that were collected early and late, variation in abortion rates during the last month of gestation after late collections occurred, and potential violation of the assumption of random sample collection (e.g., bias towards collecting younger, more naïve, or bigger, more easily observed animals). Now that sacrificing endangered Steller sea lions for science is no longer acceptable or permissible, broad-scale census counts of non-pups and pups, and estimates of the proportions of non-pups that are adult females and juveniles, have provided the primary data for estimating natality in the WDPS. Those data and estimation procedures, however, may not be appropriate for an accurate assessment of natality when compared with earlier studies because they contain different sources of error, such as variations in the proportion of animals hauled out between censuses, the proportion of pups that have died and/or washed away prior to the censuses, the number of pups that have not yet been born at the time of the censuses, and proper determination of which animals are reproductively mature females. Furthermore, a shifting ocean climate, may have caused systematic changes in sightability of these animals over time that led to an illusion of declining natality. In this study, we emulated the earlier studies of natality in Steller sea lions, without some of the potential biases by tracking known individuals over time (7 yrs). This obviated the need to estimate proportions of females hauled out on the rookery or the proportion of pups that had died prior to surveys, as need to be estimated from census counts, because both were fully accounted for by virtually continuous observations. Thus, the findings of our study are based on direct observations and are more directly comparable to the estimate of natality in the 1970s in the GOA, and they contrast with recently hypothesized estimates from an inferential model in that they do not indicate a difference in natality between current levels and those in the 1970s. We will discuss the likely reasons for the incongruity in light of methodological considerations and changing ocean climate regimes, and how it affects our perception of the status of the population in the GOA and controls on their abundance. # Results One hundred and fifty one female Steller sea lions met the criteria for maturity and repeat sightability for at least two years. Females of known age (n = 6) gave birth for the first time at 5.3 yr (range: 4–6 yr). Results of the GOF test indicated an insignificant degree of overdispersion to the data (ĉ = 1.10; χ² = 51.67; d.f. = 47; *P* = 0.296). Nevertheless, to be conservative, the ĉ value was applied to tested models to inflate variances of estimated parameters. The most parsimonious model included both survival and state transitions as dependent on whether or not the female produced a pup in the previous year; however, the next best model, which did not include a difference in survival between states, was virtually identical (LRT: χ² = 2.16; d.f. = 1; *P* = 0.142;). Together, the likelihood associated with these two models was 89%. Sighting probabilities were appropriately estimated at 0.999 for females giving birth and at 0.843 for those not giving birth. Natality, estimated from results of the most parsimonious model, was 69.2% (±2.5%, SE;) for all years combined and was fractionally higher when calculated from the next best model that expressed no difference in survival. Females giving birth had a higher probability of surviving to the following year (0.851) than females that did not give birth (0.777;) but the nearly equivalent, second-best model indicated no difference in survival at 0.828 (±0.021). Also, females that gave birth in year *i* were more likely to give birth in year *i*+1 (ψ^bb = 0.760) than females that did not give birth in year *i* (ψ^nb = 0.584) with no overlap in confidence intervals. Results were similar from the second-best model indicating significant differences between these transitions. # Discussion Natality is not the only life history trait that can be influential in driving dynamics of populations and that is susceptible to effects of prey limitation under bottom-up forcing scenarios for pinnipeds in decline. However, we made no measurements of other factors such as juvenile survival and recruitment. The focus of this study was on natality which is a critical element of special concern for Steller sea lions in Alaska. Female Steller sea lions reach sexual maturity with their first ovulation at an average age of 4.6 y and nearly all females that are mature become pregnant each year. At Chiswell Island, known-age females (n = 6) produced their first pups at an average age of 5.3 y, indicating they were ovulating at 4.3 y, although we cannot necessarily assume that was their first ovulation. Yet, it is apparent from the data presented here that age at first reproduction was similar to that in the 1970s and justifies choosing females ≥5 years of age as part of this analysis for direct comparisons with earlier work on natality rates. The estimate of natality found in this study (69%) was similar to natality in the 1970s (67%), prior to the population decline in this region. However, our value may be slightly underestimated because of the possible inclusion of older, post-reproductive animals. There is some evidence that Steller sea lions become reproductively senescent at more than 20 years of age and previously calculated natality for the 1970s did not include elderly, non-pregnant females because of potential biases. At least two adult females of unknown age were included in our study and may have been post-reproductive, as they never gave birth over the 4+ years they were observed. Future studies of known-age individuals should help to clarify the extent of senescence in this species. In contrast, natality of Steller sea lions estimated in our study is substantially higher than the recently published estimate for the 2000s of 43% which was inferred from a population dynamic model. Our estimate of natality at Chiswell Island may be considered normal and indicative of a stable or increasing population, whereas the inferential model estimate suggests a population that is still under stress, nutritional or otherwise. Notwithstanding variation in survival, natality rates of 60% to 75% have been generally associated with stable or increasing populations of pinnipeds, whereas rates of 55% or lower have been associated with declining populations and related to the adverse effects of density dependant factors or food stress. There are at least two possible reasons for the large discrepancy between the two estimates which we will examine briefly in turn. One is that the population status and natality trends in Kenai Fjords are not representative of the greater GOA. The other is that the methods for calculating natality were very different between the two studies making comparisons difficult; i.e. ours is a direct estimate, whereas the other is a hypothetical value based on an inferential model. The inferential population dynamic model was based on data from only the central GOA, and was assumed to represent a major portion of the WDPS of Steller sea lions. Population trajectories within the WDPS vary widely with location and we do not assume that our estimate of natality in Kenai Fjords is necessarily representative of natality throughout the entire western range of these animals. However, the evidence presented below suggests that our findings may be representative of the eastern and central GOA (Chiswell Island lies in the transition zone between these somewhat arbitrary regions;). Steller sea lions in the eastern GOA, which includes Chiswell Island, have experienced a 35% increase in their population over the period 2004–2008, while those in the central GOA increased by only 10% over that period, although it is argued that the large growth in numbers in the eastern GOA was due to a seasonal influx of animals from southeastern Alaska (the eastern distinct population segment). The increasing population trends of resident animals in the eastern GOA, therefore, are more equivalent to those in the central GOA. Furthermore, the ratio of adults and juveniles to pups counted in aerial censuses in the 2000s at Chiswell Island (median  = 1.64) is the same as at other rookeries in both the central and eastern GOA (median  = 1.71; Mann-Whitney U = 46.00, P = 0.296; based on data). An additional similarity between the Chiswell Island rookery and other GOA rookeries is that measurements of maternal care are excellent at Chiswell Island and are comparable to maternal care at other rookeries in the central GOA, suggesting prey is readily available across this broad area. With similar trends in behavior, population trajectories, and observed ratios of age classes throughout these regions, we find no reason to suspect that natality of sea lions in this study is unusually high compared to sea lions elsewhere in the GOA. The previously published estimates of natality were based on long-term census counts of adults, juveniles, and pups across a broad range of the GOA. However, the data suffer from several biases including, but not limited to, confounding influences of neonatal mortality, temporary immigration, and changing female sightability. In order to estimate natality from counts of adult and pup Steller sea lions, and make them comparable to the pre-decline estimate, accurately determining neonatal survival is necessary to account for pups lost prior to the time of census. Overestimates of pup survival would reduce the calculated number of pups actually born, thereby reducing apparent natality. The authors of the population dynamic model applied a survival correction to their life history matrices of 0.949 to account for neonatal mortality and assumed that it was constant over time. That correction factor was derived from counts of live and dead pups found on rookeries in the 1970s. However, the leading cause of mortality in young pups results from being washed away in high surf conditions with survival to three weeks of age estimated to be much lower in dedicated studies (0.679 and 0.896). Although there is currently no evidence that there was a significant change in Steller sea lion pup mortality over time, there can be significant variation in pup mortality between years, and a high estimate of survival, assumed to be constant, would have decreased all estimates of natality over the periods studied. Immigration, whether temporary or permanent, of animals from the growing southeastern Alaska population in recent years will also skew census-based estimates of natality lower because animals that are not part of the breeding population may be counted as breeders. It is not known how much temporary immigration might have affected estimates of declining natality in the 2000s but some effect is probable. It is also likely that female sightability in the GOA has changed systematically between ocean climate regimes in recent decades causing the appearance of reduced natality based on estimates of the proportion of animals hauled out. That is, if more females were hauled out during surveys in recent years compared to earlier years, then there would have been an appearance of reduced natality based on relative proportions of females to pups. The authors of the population dynamic model assumed female sightability was constant over time and suggested that an increase of about 40% in the number of females observed would be necessary to counter the estimated decline in natality. There is no direct evidence of a long-term, increasing trend in female sightability but some compelling indirect evidence is explained as follows. The availability of some important prey for Steller sea lions was probably reduced during the 1980s. Therefore, females would have spent more time foraging at sea to adjust for prey deficiency during that time. Such behavioral changes associated with food limitation in otariids make sense and have been observed in other studies. As sea lion populations continued to decline through the 1990s, changing ocean climate regimes probably led to improved forage availability, which resulted in a systematic reduction in foraging durations by adult females. Shorter foraging periods would effectively cause an apparent decline in natality rates when simply counting ratios of adults to pups because more adults would be counted in relation to the number of pups in later years. There is no good information on perinatal periods (the time females spend on shore between giving birth and their next foraging trip to sea) in the 1980s, but perinatal periods are also greatly affected by nutritional limitation in the same way as foraging trip durations and would therefore exacerbate the effect of female sightability across changing prey regimes. Hence, it may not be possible to accurately determine changes in pinniped vital rates based on census data without a complete understanding of how environmental factors affect sightability of different age-classes throughout a region over extended periods. Further evidence of a healthy population in our study is indicated by lack of a cost of reproduction. Female reproduction is normally believed to carry costs in terms of a reduced likelihood of survival and/or future reproductive potential. This effect has been shown in some studies of pinnipeds, including Steller sea lions during the 1980s when pregnancy was negatively correlated with lactation status. We found the opposite effect in this study with reproduction being positively correlated with survival (though not significantly so) and future reproduction, suggesting variation in overall fitness between individuals rather than a reproductive effect on fitness. Similar findings were reported for subantarctic fur seals (*Arctocephalus tropicalis)*, and strong evidence for an effect of individual quality on reproductive success has been seen in other large mammals. Such variable reproductive strategies between fit and unfit individuals are most evident when resources are plentiful. Alternatively, when food resources are more limited, reproductive costs on future reproduction are more evident. This provides additional evidence that lowered fitness and associated costs of reproduction in Steller sea lions during the 1980s were consistent with resource limitation, whereas the findings in this study of no cost of reproduction during the 2000s suggest that resources are more plentiful. In recent decades, most researchers agree that prey limitation is not a problem for Steller sea lions in the GOA. This study found that natality in Steller sea lions at Chiswell Island is at a level similar to that before the population decline, and evidence presented above suggests that the animals in Kenai Fjords could be representative of those across the eastern and central GOA, but not necessarily further afield. Population losses during the 1980s are thought by some researchers to have been caused by nutritional stress resulting from the ocean climate regime shift in the mid-1970s, although others disagree. Indeed, there is good evidence that juvenile survival and recruitment was reduced by predation, and/or food limitation. Nevertheless, it is plausible that decreased natality in the 1980s compared to the 1970s was caused by nutritional limitation during that period and that it may explain some of the population decline. In more recent years, studies of juvenile health and maternal care provide no evidence of nutritional limitation in this species. Disease, parasitism, and contaminants could adversely affect reproduction, but research has not shown significant trends over time or major problems in the current decade. Other explanations for population losses, such as predation and fisheries related mortalities, could play a major role in adult and juvenile survival but probably have less of an effect on natality, although exposure to predation risk does increase levels of stress in female Steller sea lions and can decrease natality in other species of large mammals. Given the evidence presented here, we suggest that the apparent long-term decline in GOA Steller sea lion natality as inferred elsewhere is probably due to an artifact of increasing female sightability as resources became more abundant from the 1980s to present. Other factors such as neonatal mortality and immigration may have also affected those inferential estimates of natality. Direct estimates during the 1980s provided sufficient evidence of a reduction in natality only during that time period, but those findings do not necessarily exclude the additional, possibly more important role of top-down effects of predation in the collapse of WDPS Steller sea lions. Our result suggests that natality of Steller sea lions in the 2000s is similar to that before the population decline (1970s) and is consistent with natality found in stable or increasing pinniped populations. The contrasting results presented here and by the authors of the population dynamic model have major implications on our understanding of factors at play in the GOA ecosystem that affect Steller sea lion populations, and by association populations of harbor seals (*Phoca vitulina*) and sea otters (*Enhydra lutris*) that also collapsed during the same era in the same region. There is no evidence of a nutritional mechanism that might have driven natality of Steller sea lions down to such a low level subsequent to the late 1980s as inferred by the population dynamic model, and low natality is opposite that which would be expected in an otherwise healthy population of animals. The different estimates also have important implications on management strategies that have been, and might be, enacted to help Steller sea lion populations recover. Resolving uncertainties that have arisen from the two approaches to estimate natality, i.e., whether there is a systematic difference between them, could easily be put to a direct test by applying them simultaneously at several rookeries in the GOA. Until then, attempts to explain the lack of recovery of the WDPS in the GOA should more fully explore alternative hypotheses to nutritional limitation, such as high predation mortality of juveniles as suggested by recent findings of Horning and Mellish. # Materials and Methods ## Ethics Statement This study meets all ethical standards based on an approved Animal Care and Use Committee permit and National Marine Fisheries Service permits to conduct research on Endangered Steller sea lions. ## Study Site and Observational Methods This study was conducted at the Steller sea lion rookery on Chiswell Island and nearby haulouts in Kenai Fjords which lie within the range of the endangered WDPS. The pattern and magnitude of population decline at the rookery were similar to other rookeries in the central GOA—that is, abundance fell by 90% from 1,459 adults and 564 pups in 1956 to approximately 90 adults and 50–80 pups in the 2000s. Beginning in 1999, up to six remotely operated video cameras were used to monitor Steller sea lions (see for details). Video images, which provided complete spatial coverage of the Chiswell Island rookery, were viewable and controllable in real-time from the Alaska SeaLife Center 65 km away. Cameras were also installed and monitored at nearby haulouts beginning in 2000. Most adult Steller sea lions can be individually identified by unique scars, fungal patches, and/or flipper patterns and longitudinal studies have been successfully conducted on animals identified by such means. During the course of this study, female sea lions with unique markings were tracked and digital photos of those animals and their distinguishing marks were taken on a regular basis from all remotely-monitored sites in Kenai Fjords. A few breeding females were identified by flipper tags (n = 4) or brands (n = 2), and age was known only for those animals. Females that did not have at least two distinguishing marks and could not be reliably resighted from one year to the next were not used in this analysis. Although pictures and data for some females were collected as early as 1999, they were not considered during1999–2002 in the analysis of natality rates because of more focused sighting effort on those giving birth over those that did not in those years. All females with unique markings (an average of 68.9%±4.8% SE of the Chiswell Island female population in each year) were non-preferentially identified and tracked from 2003 onward whether or not they gave birth. Observations each year began with the arrival of the first female on the rookery in mid- to late-May and included full census counts of all sea lions by age- class (male/female adult, juvenile, yearling, and pup) on the rookery throughout the breeding season. Census counts were made at approximately 1100 h and 1900 h, and hour-long scan sampling for identifiable females and their pups was done four to ten times daily from 0600 h to 2200 h; earlier and later hours were added around the summer solstice when light levels were sufficient for viewing sea lions. After 10 August, observations were recorded from approximately sunrise to sunset as diminishing daylight allowed. Events such as births and deaths were opportunistically recorded as they occurred or within 4 hr of their known occurrence. Births that happened overnight were recorded the following morning as having occurred at the half-way point of non-observation hours. Steller sea lion mothers in the WDPS will normally remain with their newborn pups for 8 to 12 days following parturition. Given the duration and detail of observations in this study (frequent scans and complete spatial coverage of the rookery), it was highly unlikely that any births went unnoticed. Identified females were considered for this analysis if they were present on the Chiswell Island rookery during the pupping and breeding season from 15 June until 15 July. Females that gave birth earlier still had a definitive presence on the rookery during that time. That time period also included females that were present to copulate, and hence had a presumed intention to breed at this rookery, but excluded some females that hauled out briefly on Chiswell Island before leaving to potentially pup elsewhere. Typically, females that give birth to stillborns should not be considered productive. However, all recent estimates of natality in Steller sea lions are compared to natality in the presumed healthy population during the 1970s and declining population in the 1980s. Those earlier estimates were based on late- term pregnancies and could not account for stillbirths. Therefore, full-term stillbirths were included as births in this study to make the data comparable to those earlier standards, but this probably had little effect on the estimates of natality because fewer than 2% of pups born at Chiswell Island were stillborn. Furthermore, the published standards for natality were only considered for reproductively mature females whose status was known by examination of ovaries. It was not possible to verify reproductive maturity in this study even when age was known. To reduce the chance of including pre-reproductive animals in our dataset, the first year of sighting of each apparently mature female of unknown age was removed whether or not she gave birth. Those that gave birth in their first year of observation were removed to avoid bias toward more fecund animals. Females of known age were included in this study beginning at 5 years of age to be consistent with the average age of sexual maturity at 4.6 yr, which would indicate that age of first pupping would be at about 5.6 yr. In order to decrease sample bias toward more fecund females that may spend proportionally more time at a rookery, nearby haulouts were also monitored during the pupping season to account for females that may have spent more time at those locations. Females at haulouts were included in the analyses if they met the abovementioned sighting and maturity criteria unless they were accompanied by juveniles that were known to be born elsewhere (i.e., not at the Chiswell Island rookery). Many of the animals in the Chiswell Island population that were not giving birth on the rookery in any given year spent the summer elsewhere, presumably outside of the study area. Females that returned to the study area later in the year without a pup were classified as not giving birth in that particular year because of known breeding-site fidelity in this species and to ensure a conservative approach to estimated natality. ## Data Analysis Multi-state models were constructed using the logit link function in Program MARK with the following parameters being estimated over 7 years of observation (2003–2009): S*_i*^x = probability that a female in state x in year *i* survives until *i* +1. P*_i*^x = probability that a female is resighted in year *i* in state x, given that it is present in the study area in year *i*. ψ*_i*^xy = probability that a female in state x in year *i* is in state y at *i*+1, given that she survived from year *i* to *i* + 1. States were recorded as “B”–observed birth or with pup, “N”–observed but did not give birth or not seen with pup, and “0”–not observed. Calculation of the proportion of females that were productive (natality) was performed using equation 2 in Nichols et al. with corresponding estimates of variance. The Cormack-Jolly-Seber (CJS) modeling approach was chosen over models that account explicitly for temporary emigration because CJS models required fewer assumptions and constraints in addition to providing sufficient parameter estimates for animals that show breeding site fidelity as Steller sea lions do, especially after breeding has been established. Sighting probabilities for the two strata (B and N) were retained in tested models to express breeding-site fidelity and differences in the ability to detect those states. We compared models with Akaike's Information Criteria (AIC), corrected for small sample bias (AIC_c) with additional comparisons for nested models using likelihood ratio tests (LRT). The general, fully time and state dependant model was initially tested for goodness-of-fit (GOF) with program U-CARE 2.2 and the estimated overdispersion coefficient (ĉ) was used to adjust model results and convert AIC_c values to quasi-AIC_c (QAIC_c) values. QAIC_c weights, calculated from model differences in QAIC_c values (ΔQAIC_c), indicated relative support for the various models. Finally, we examined indicators of potential costs to giving birth in regard to survival and state transitions. Cost is suggested if birthing in one year is associated with a significant reduction in survival probability for the following year. Birthing in one year may also cause a reduction in the probability of birthing in the next year, as was indicated for Steller sea lions in the 1980s. An effect of birthing on subsequent birthing is suggested if transitions from not birthing to birthing (ψ^nb) were greater than transitions from birthing to birthing (ψ^bb). Data collection for this study was accomplished by many hard-working technicians and interns at the ASLC including Melinda Fowler, Karin Harris, Carlene Miller, Kimberly Smelker, Emily Teate, et al. We thank them and I. Boyd, J. Estes, D. Hennen, M. Horning, J. Nichols, G. van Vliet, B. Wilson, and two anonymous reviewers for discussions, commentaries, and much helpful advice on this material. The Chiswell Island archipelago is part of the Alaska Maritime National Wildlife Refuge. Steller sea lion research conducted on this and nearby islands was granted under U.S. Fish and Wildlife Service Special Use Permit No. 74500–03-045 and National Marine Fisheries Service, Office of Protected Resources Permit Nos. 782-1532-00, 881-1668-00–05, and 881-1890-02. [^1]: Conceived and designed the experiments: JMM. Performed the experiments: PP. Analyzed the data: JMM. Wrote the paper: JMM AMS. [^2]: The authors have declared that no competing interests exist.

# Introduction During a pandemic, such as the one of the coronavirus disease 2019 COVID-19, management of patient flow and hospital resources are pushed to their limits: Hospital Emergency Treatment Facilities through intermediate care unit (IMU) and intensive care unit (ICU) are or are going to be be severely strained. Healthcare professionals are often times forced to make difficult decisions in patient care and resource allocation. Patient profiles might be out of the ordinary routine of the hospital and workflow must be different. End-to-end on demand visibility with identification of real constraints is needed for the senior management. A manager may have simple but essential questions such as: how many beds do I need on the floor, how many beds are available in the critical care unit, how much supplies should be ordered to take care of our patients and protect our staff from infection, how long will the facility have to work at maximum capacity, is there enough staff to hold this workload long enough, are we doing well with patient outcomes, etc. Multiple governmental and private agencies have focused on creating dashboards for easy access and understanding of global pandemic data. These applications are great to give the general assessment of the pandemic but do not allow a projection of detailed information at the local hospital scale that is necessary to optimize the management of patient workflow. There is a significant amount of literature about mathematical models in epidemiology that provide a rigorous framework to make predictions on the number of people who are going to be symptomatic enough to require hospitalization. This approach has been quickly applied to COVID-19 with success. In the case of the COVID-19 pandemic, it is particularly difficult because large bodies of infected people are asymptomatic. Consequently, the basic reproduction number *R*₀ factor of COVID-19 is still under active debate. On the hospital workflow side, while there is a large amount of work on this topic, one of the difficulties is to asses the death rate of patients hospitalized at the beginning of the pandemic because the Length of stay (LOS) is rather long and the disease is still not well understood. Every hospital has to adapt to the new crisis as it arrives, so clinical practice may vary greatly from one institution to another. A number of guidelines and great reports have been quickly edited to support the heath community, but it takes time to standardize the healthcare process. Our goal in the paper was to come up with a simple and robust mathematical framework that is easy to use and that supports the management of the patient workflow during a pandemic. Such a model should operate on a relatively limited data set that reports daily on the number of patients admitted for hospitalization, patient output of the facility (such as number of patients healed per day or number of deaths per day), and at minimum the number of patients staying in the most critical unit of the facility—the ICU. The model can then be customized to the local hospital system with an optimization technique to achieve calibration after a few weeks of data acquisition. Much more can be done with the patient electronic records that detail patient comorbidities and chronic conditions, provided that the disease of the pandemic is well understood. We have used a Markov process description of the workflow’s graph with probability governing the patient transition from one care unit to another, as well as a simple statistical model of patient LOS at each stage. We will show that with a minimum number of parameters used to fit on the time series listed above for a period of a few weeks, one may start to assemble the information needed to assist the senior management in getting answers and identifying real constraints to reduce speculation or misallocation of resources. This work is our first iteration to achieve a very ambitious goal: as data becomes available, the quality and level of detail of modeling should keep improving to achieve better results. It is our hope that such an effort, among many others, will once again prove how much digital health can benefit from computational science to improve patient care. The paper is organized as follows: Section 2 describes our method to construct the model and details the choices we made to work with the data set on hand; Section 3 gives the main results and solution to our initial goal in supporting management; Section 4 discusses the benefit and limitation of our method and concludes with further potential development. # Materials and methods Because of the sparsity of data available to construct a predictive model during a pandemic crisis, we are going to use a very simple model that reproduces the workflow of. Let’s start with a brief description of standard patient workflow—see —with respect to disease progression—see. The patient moves from one care unit to another according to his/her condition. The first two steps are registration and diagnostics, which in principle should be a relatively quick process. For the patients who stay in the hospital because their health condition justifies a longer stay, they are first put in a ward unit for further assessment and treatment. This step is where a number of medical imaging steps start involving either a chest CT scan in the imaging center or a chest X-ray with a mobile unit. Meanwhile, significant biological lab work starts to grade the patient’s condition more precisely and continues during the patient’s stay. These resources, i.e. imaging and lab work, are typically shared by all patients in the hospital and therefore may slow down the process. For simplicity and in the absence of adequate data set for validation, we neglect these constraints. Some of the patients who receive medical attention do well with conservative management only and can be discharged home after a few days. But for others, their health condition may deteriorate and those patients will need to be moved to the IMU for higher level of care and/or to transfer to the ICU for ongoing monitoring and mechanical ventilation. The IMU and ICU require extensive supplies and resources. It is often mentioned that the number of available ventilators is critical to ICU functions. However, it is not the only limiting factor: patients under mechanical ventilation need sedation and might be connected to a number of additional systems to deal with organ failures. Once again for simplicity and because of the lack of input data, our model will not take into account these bottlenecks. There are no technical difficulties required to add those constraints in the mathematical model with our bottom up description of the workflow as in. Additional steps can be recovery for patient being well or unfortunately palliative care when the patient is not responsive to treatment. There are many exceptions and singularities to these standard paths: for example, a patient may go directly from admission to the ICU when their condition is too unstable. In some hospitals, the floor might be shared by patients who are recovering from COVID-19 and palliative care patients. Despite this, we will separate these functional units in our model to clarify the workflow process according to what each patient stage requires in terms of resources and time to deliver adequate care. To summarize, a simple workflow graph is created and the main requirement is to know (i) the probability that a patient goes from one care unit to another and (ii) a statistical estimate of how long the patient should stay in each care unit before moving on. Our model follows a Markov process for (i): there is a probability associated with each branch of the graph summarized in. With respect to (ii), we use a lognormal distribution that can be reconstructed from the parameters listed in Tables and. This simple framework allows us to construct a generic model that dynamically computes the number of patients for each care unit as a function of the number of patients showing up in the ER. In particular, we get a time series of the number of patients staying in the ICU, which is the most critical care unit in term of resource allocation, as well as the number of patient outputs, such as the number of patient healed and discharged per day, or the number of death(s). These time series can be fitted to existing data the hospital obtains during a period of a few weeks prior to retrieving the performance parameters of. Once the model is calibrated, it can be used to extrapolate the load of each care unit in the next few days and anticipate the need of staff and supplies—see. This discrete model is stochastic, so one needs to run many simulations to build a statistical estimate of such quantities. It is appropriate to retrieve the unknown parameters of the model using a form of stochastic optimization method, such as genetic algorithm, since the model workflow process, like the one in the hospital, is discrete, noisy, and nonlinear. Let us describe the data set we are using to construct our model. The French government has kindly decided to release the records of most public hospitals around the country during the COVID19 crisis. From this excel file, we can easily recover the number of patients staying in hospitals, the number of patients in ICU, the number of patients healed and discharged, and the number of patients dying in a medical institution. Those numbers are updated daily and go back to March 18, 2020. We will extensively use this French Data Set (FDS) to identify the missing parameters of our model. The number of parameters of our model is relatively large: about one parameter for each branch of the graph minus the number of nodes for (i) and two parameters for the log distribution of (ii) in each care unit. To avoid over- fitting, one should come up with a strategy that lowers the number of unknown based either on literature or hypothesis that can be validated otherwise. We are going to describe thereafter the rationale for our choices to the best of our knowledge and further discuss some of the limitations of our model in Section 4. First of all, a lognormal distribution of the duration of each step of the process might be justified as follows. Biological process, such as incubation and recovery, are often described as such. First, the patient’s condition is indeed dominated by his/her biological time. Second, medical procedures with their associated time lag and delay are also often best described as lognormal processes with a long tail. This is not in contradiction with the fact that patient LOS in the hospital may not ideally be described by a simple exponential distribution or similar. Overall, LOS adds up the time distribution of each step in a Markov process and might be described at the convolution of the probability distribution of each step. Now let’s review the parameters of that gives the probability transition from one unit to another, in order to rationalize the construction of our generic model. One can first list the following constraints assuming that all possible paths are exhaustively listed in the workflow of, so we have: $$\begin{array}{r} {\alpha_{2} + \alpha_{3} + \alpha_{4} = 1,\;\alpha_{5} + \alpha_{6} + \alpha_{7} = 1,\;\alpha_{8} + \alpha_{9} = 1,\;\alpha_{10} + \alpha_{11} = 1,\;\alpha_{12} + \alpha_{13} = 1.} \\ \end{array}$$ Overall, the death rate and recovery rate of patients who are staying in the hospital should be within an acceptable limit. Technically, the death rate of hospitalized patients is: $$\begin{array}{r} {\beta_{d} = \alpha_{2}\alpha_{6}\alpha_{8} + \alpha_{2}\alpha_{6}\alpha_{9}\alpha_{10} + \alpha_{2}\alpha_{5} + \alpha_{4}\alpha_{8} + \alpha_{4}\alpha_{9}\alpha_{10}.} \\ \end{array}$$ Similarly, the recovery rate of hospitalized patients *β*_*h* = 1 − *β*_*d* is: $$\begin{array}{r} {\beta_{h} = \alpha_{3} + \alpha_{2}\alpha_{6}\alpha_{9}\alpha_{11} + \alpha_{2}\alpha_{7} + \alpha_{4}\alpha_{9}\alpha_{11}} \\ \end{array}$$ *β*_*d* is difficult to asses with a pandemic that just started. As a matter of fact, most infected patients are still in the hospital and their outcome may not be clear. We look thereafter for some lower and upper bounds of *β*_*d* that limits our search. According to the Intensive Care National Audit &*amp*; Research Center (ICNARC) report of March 27, 2020, we may conclude that a lower bound to overall death of patients admitted into ICU units is about 10%—as a matter of fact most patients were still in the hospital at this early stage. On the other hand, and based on a very small case series in the Seattle region, the death rate was up to 50%, but in this study most patients had chronic medical conditions. These two publications illustrate the difficulty of recovering the true rate of death for today’s large population whom have been hospitalized, mostly due to the heterogeneity of the population they encompass. In France, as of April 17, 2020, the number of deaths in hospitals was 11,842 patients and recovered was 35,983 patients. Assuming that the proportion of death versus recovery will be about the same for the patients who are still ill, the death rate of hospitalized patients should be around 25%. Finally, according to, an early estimate of the death rate for hospitalized patients in Wuhan, China based on a case series of 191 patients was 54/191 = 28%. We restrict ourselves to the model matching the FDS to a \[10%, 40%\] death rate interval, that is: $$\begin{array}{r} {0.1 \leq \beta_{d} = \alpha_{2}\alpha_{6}\alpha_{8} + \alpha_{2}\alpha_{6}\alpha_{9}\alpha_{10} + \alpha_{2}\alpha_{5} + \alpha_{4}\alpha_{8} + \alpha_{4}\alpha_{9}\alpha_{10} \leq 0.4.} \\ \end{array}$$ According to several reports including the ICNARC one mentioned above, it is expected that the number of patients dying in ICU is about 50%. provides much further details on the probability of survival of patients with ARDS under mechanical ventilator as a function of the day of the start. It shows that about 25% of the patients in ICU die during the first few days from severe complications. We will introduce an artificial two phases ICU decomposition of the patient stay in the ICU to bypass the limitation of a single lognormal distribution that may not represent an adequate model of LOS in this unit according to clinical studies: a short phase one with mortality driven by *α*₈ and a longer phase 2 with mortality driven by *α*₁₀. Consequently, we will assume that: $$\begin{array}{r} {\alpha_{8} \in \left( 0.1,0.3 \right),\;\text{and}\;\alpha_{10} \in \left( 0.4,0.6 \right).} \\ \end{array}$$ There are also few parameters in that should have near to no limited effect on statistics when matching our model to FDS. FDS is based on hospitalized patients, so *α*₁ cannot be recovered from this data set. According to FDS, about 30% of patients who show up at the emergency room (ER) are returning home. We will choose *α*₁ = 0.3. According to Dr. M. Mueller, 25% of the patients who are not responsive to treatment may leave palliative care alive and are discharged home. This may vary depending on each country or hospital policy. Because patients with COVID-19 in palliative care are still very contagious, we will assume they stay in the hospital until the end. We will choose *α*₁₂ = 0., for all our calculations. To sum up, our model essentially needs the calibration of 6 parameters, namely $$\begin{array}{r} {\overset{\rightarrow}{A} = \left( \alpha_{2},\alpha_{3},\alpha_{5},\alpha_{6},\alpha_{8},\alpha_{10} \right)} \\ \end{array}$$ under the set of constraints. Let us denote *F*^*admission*(*jd*) the number of patients admitted per day *jd* ∈ 1..*N* in the hospital who have a positive diagnosis and must stay in the hospital. We will use this time series as an input to our model simulation in order to calibrate the model against the FDS. Let us denote $\left( F_{s}^{ICU}(jd),F_{s}^{healing}(jd),F_{s}^{death}(jd) \right)$ the time series of patients in the ICU, patients healed and discharged home, and patients who died per day *jd* ∈ 1..*N* obtained from the simulation output. We will denote respectively $\left( F_{fds}^{ICU}(jd),F_{fds}^{healing}(jd),F_{fds}^{death}(jd) \right)$ the times series extracted from the FDS. We find $\overset{\rightarrow}{A}$ as the solution of the minimization problem of the weighted norm: $$\begin{array}{r} {min_{\overset{\rightarrow}{A}}(\gamma_{1}{||}\mathcal{F}_{s}^{ICU} - F_{fds}^{ICU}{||} + \gamma_{2}{||}\mathcal{F}_{s}^{healing} - F_{fds}^{healing}{||} + \gamma_{3}{||}\mathcal{F}_{s}^{death} - F_{fds}^{death}||)} \\ \end{array}$$ where $\mathcal{F}_{s}$ is the mean of a large number of runs of the model. This number of runs is set large enough to let the solution of the optimization problem be independent of it. As mentioned above, we will use a genetic algorithm to solve that minimization problem. The weight factor (*γ*₁, *γ*₂, *γ*₃) in can be set equal or unequal to favor the quality of the fitting for one of the variables, such as the number of patients in the ICU that is critical to management. Tables and give the time window we used for each transient stage. We construct a lognormal distribution of duration for the patient stay in such a way that about 90% of the patients’ stay will be within a coarse approximation \[*P*, *Q*\] listed in these tables. The choice of the parameters in Tables and might be easier to come up with. One of the most remarkable features is that patients with COVID-19 who stay in the ICU can be longer than usual. The LOS in palliative care was set according to Dr. M. Mueller’s data. We have used extensively, as well as the feedback from clinicians in the field to estimate the interval of variation for the parameters \[*P*, *Q*\] the best we could. We used a fairly large interval since it can be observed that the standard deviation for LOS in each care unit is large as described in this report from the Imperial College London COVID-19 Response Team. One may fine tune the interval value \[*P*, *Q*\] if needed in the fitting process of the model to the data set of time series available. To distinguish those unknown parameters that are important from those who are less significant, we run linear sensitivity analysis for each of our results. This method is used to confirm that the time window parameters of Tables and have a secondary effect on the quality of the model fitting. Finally, we derive from our model some predictions on staffing and supplies for the next week or so, as well as the load foreseen for each care unit. The nature of the stochastic simulation automatically gives an uncertainty estimate on these predictions that increases as time grows. To compute supplies such as personal protection kits, we can use some adaptation of the reference of the CDC web site that was constructed for Ebola. Our software can then be used to feed the stock management scheme implemented by CDC for COVID-19. In this paper, we use a growth estimate of two personal protective equipment (PPE) per shift and per staff member for simplicity. Similarly, we have listed in a gross approximation of the number of nurses and staff per bed site in each unit. Those figures are depending on the crisis situation and might differ depending on the country. In order to take into account the fact that staff and supplies are limited and require hard management choices during a pandemic crisis, we tested the model further against the scenario of a shortage on nurses who are essential in intensive care units. To introduce a risk factor due to the shortage of nurses, we have extrapolated from and, a scenario where a 40% shortage of nurses results in: - Time spent on floor increases by 15% - Chances of transition from floor to Recovery or floor to IMU decreases by 15% - Chance of dying in ICU increases by 15% To get a continuous approximation, we assume that the shortage of nurses has a linear effect, and use linear interpolation for shortages from 0% to 40% maximum. This is certainly a gross approximation, but we felt that it was important to bring awareness to those effects with a simulation tool. We will present in the next section our results. # Results Let us first report on the model fitting with the FDS. We sum up the number of admissions, patients in ICU, number of recoveries and deaths for the whole country of France in order to get a robust data set that averages the noise of the data. We calibrated the model to this largest data set that covers the period 3/18/20 to 4/11/20 and found a death rate of about 25%. This result is in agreement with the estimate we did in Section 2, as of April 17, 2020. The sensitivity analysis on the alpha unknown vector $\overset{\rightarrow}{A}$ is reported in. Figs and show the results with the parameters as listed in Tables to. All numbers have been scaled by a factor to represent an average hospital size. We observed that the number of patient admissions is not a smooth curve. Typically, Sunday’s have less activity with less patients discharged than weekdays. However, the model fitting seems adequate and robust to a small variation of parameters. The logic on the influence of parameters is simple, *α*₂ being the one who is most important for all output. Each of the six parameters seems to have some significant influence for at least one of the three outputs. This is not completely surprising because the construction of the model was subject to a number of experimentations and trial errors by way of simulations before getting what we found: a model with low complexity that makes the fitting feasible. The curve of the number of patients under mechanical ventilation is smooth as expected because this care unit has by far the longest LOS. Consequently, the system has a lot of delays and the relatively small number of patients per day who have been healed or died is still relatively small. Similarly, the number of patients who are mechanically ventilated does not directly reflect the number of admissions at the early stages of the pandemic in the FDS. For some unknown reasons to us, the model overestimates the number of patient deaths in the last few days. Those numbers might be small however and more sensitive to singular events or simply involve delay in reporting. It might also be the accumulation effect of the small lag differences of the prediction of patients under mechanical ventilation and the reality. In order to compare the results obtained with designated subset of the FDS that corresponds to the hospital in Paris and the hospitals in Alsace, we used the simulation with the exact same set of parameters found for the data set with the whole country. Alsace has been the busiest cluster at the beginning of the pandemic, followed later on by Paris and Ile de France. The results for Alsace are reported in Figs and. We observed a fairly large difference of the model’s prediction on the number of patients under mechanical ventilation. It seems that at the peak of the pandemic in Alsace, the number of patients under mechanical ventilation was less in reality than in the model. One possible factor would be the shortage of available beds in the ICU. On the other hand, the number of deaths did not go higher than significantly expected. A better explanation might be the fact that a fairly large number of patients in critical condition were transferred to hospitals in different parts of the country or neighboring countries: according to local newspaper more than 110 patients from Alsace have been transferred. This seems coherent with our results: the scaling factor for the Alsace data set to get a maximum hospitalization rate of about 50 patients per day is 6; the overshoot on the ICU prediction is about 20 in ; The total maximum overshoot is therefore about 120; considering that the average LOS in ICU is roughly 12 days, our model still seems to give an adequate approximation. But unfortunately, we do not have enough information to add this new patient path in the workflow of. This phenomena is less present in the results for the data set with Paris but are still there—see Figs and. One can indeed refine the parameter fitting to be specific for Alsace and Paris in order to reflect that the clinical decision process in the workflow, i.e parameters of Tables to, might be sensitive to how much the local system is under stress, but we should then take into account those number of transferred patients that are not negligible. Next, let us describe the use of our model to assist daily management in the hospital during the pandemic. One key factor is to anticipate the load of each care unit and required resources, either to match the increase in number of patients or to reallocate resources to other patients who have seen their surgery postponed. We choose a hypothetical scenario that might occur if confinement conditions to contain the pandemic are lifted too early. We assume that the hospital has a nominal low flux of patients from week 1 to 7, and a recurrence with a daily 20% increase of new patients coming in occurs in week 8. shows the dynamic of the load of each care unit, in particular the large delay in the number of patients in the ICU that becomes saturated the latest. The black curves are a simulation of the previous week’s load (week 7), while red curves are the prediction for the following week (week 8). The upper thin red curve shows the deviation up one standard deviation to give a sense of the uncertainty of the estimate. This uncertainty grows as the time of the prediction gets further away. The cyan curve shows an hypothetical capacity of each care unit: the floor gets saturated first and needs new beds after a few days. As an illustration of the capability of the model, Figs and provide an estimate of the growth of resources needed to face the new patient wave. A number of decisions should be made in regards to patient care. compares the patient output with or without shortage of nurses. Those results are speculative since it is difficult to quantify the risk for patients beyond the nice publication results of and. It is our hope that data accumulated during crises such as the present episode of COVID-19 will give the mathematical modeling the base to do this estimate rigorously in future work. Finally taking a step further, we have looked at the prediction of the model to check either the effect of confinement policy on pandemics or new pandemics with an arbitrary rate of transmission of the disease. There are many epidemiological mathematical models available, even for the present crisis, see and. It might be difficult to assess the basic reproduction number *R*₀ factor, which is under active debate. It is probably even more difficult to assess the exact impact of global confinement or targeted confinement on those parameters that characterized the pandemic model. We should however be able to use our model to test if the effect on the most critical resource, such as ICU beds and delay in care, are linearly or nonlinearly related to those parameters. Let us use the most simplistic ordinary differential equation epidemiology model: $$\begin{array}{r} {\overset{˙}{S} = - \rho_{1}S\; I,} \\ \end{array}$$ $$\begin{array}{r} {\overset{˙}{I} = \rho_{1}S\; I - \rho_{2}\; I.} \\ \end{array}$$ where S is the population of susceptible individuals, and I is the population of infected individuals. *ρ*₁ is the transmission rate of the disease, and *ρ*₂ represents the addition of the rate of recovery over infection and the disease induced death. The function *I*(*t*) is used as the input of our workflow model, and represents the number of patients admitted to the hospital. We test the influence of the transmission rate on the number of ICU beds over a 16-week period. shows that the maximum number of ICU beds required during the epidemic is significantly higher when the transmission rate increases from 0.0015 to 0.0025, while the average number of ICU beds stays about the same. This clearly shows the nonlinear nature of the ICU load management problem and the benefit of a confinement method that lowers the transmission rate during a pandemic. Gathering more recent data coming from the french registry for 4 weeks after April 17, 2020, additional simulation were run to validate the model and its capability to predict. The simulation were done with the same parameter setting to predict ICU stays, output on patient healed, and patient death on these additional 4 weeks using the French database of patient inflow. Figs and show the predicted value on the right of the vertical blue line. The ICU occupation gets underestimate by 6% after one week, 15% after two weeks, and 40% after a month. Combined with the fact that the inflow of patients is also model dependent and that the data in this period was presenting some errors or missing values, it seems that our model might be useful for a one-week prediction interval. # Discussion and conclusion In this work, we have developed a simple computational model to mimic the workflow of an average hospital during a pandemic crisis, such as COVID-19 where patient admission goes up to 50 patients per day. This is a significant load for any hospital system because all patients suffer from the same disease and cannot be triaged using the existing departmental structure. The hospital system needs to recruit resources quickly enough to deliver quality patient care while keeping the staff safe from infection. There are many ways of developing such a mathematical model. We chose a Markov process that can augment a workflow graph provided by the clinicians and used a simple statistical model for the LOS of the patient at each stage corresponding to a graph node. A number of variations in the model construction are available: for example, changing the probability distribution of LOS for specific stages with a more sophisticated model than lognormal or decomposing the graph nodes into subgraphs of the workflow with more details. In particular, the ICU supports different paths of medical care depending on patient conditions. Because of the sparsity of data on hand, we kept the model as simple as possible and we were able to fit the French Data Set with good accuracy. Using this approach, we could: - recover important parameters that are characteristics of the workflow such as the probability for a patient to transition from one unit to another, and important patient outcomes such as healing rate or death rate. - on a pragmatic side, we use the model to assist the senior manager in answering his/her questions as listed in our introduction: how many beds do I need on the floor, how is this affecting patient outcomes, do we need to transfer patients to a different facility, etc.? Interestingly enough the protocol to handle COVID-19 patients has changed a lot since the beginning of the pandemic: by using a different set of drugs, patients might recover quickly and only need oxygen assistance instead of a full intubation. From the validation of our system on 4 more weeks of data, we believe that our model is agile enough and can be calibrated again on a second wave of COVID-19 data set to automatically handle that evolution. There are a number of limitations to our approach. The smaller the hospital, the less predictable the outcome will be. With time, the characteristics of the population of patients who show up to the ER may change and the pandemic management by the governing organizations would evolve. One can think, for example, that systematic testing would provide early diagnostics and impact the performance of the health system as shown by the statistics of countries who were early adopters of that strategy. Due to the heterogeneity of the patient population and disease patterns that depend heavily on patient characteristics, our next step in improving this model would be to include patients’ medical history listed in the electronic medical record. Above all, any model of workflow especially during a pandemic should be aware of the *Human Factor*. Staff can get sick or burnout during a pandemic and there should be a number of strategies to compute that risk and enter this into the constraints imposed on the health care system. Further, human behavior and decision process changes under stress: it can be for economical or psychological reasons. The future of computational models in digital health during a pandemic crisis should extensively include sociological and economical modeling components in the matter. We would like to thank Patrick Doolan for sharing his view with us on management and risk evaluation from his great experience acquired from the energy sector. [^1]: MG funded a start-up called ORintelligence, MG and GJ are respectively President and CTO of this company but not employees. The funder provided support in the form of salary for SF and to acquire research materials but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction Temperate grasslands, which cover 1.25 × 10⁹ ha globally, are important sinks of SOC, containing approximately 12% of the global SOC pool. Changes in grassland management (e.g., stocking rate, fertilisation) are frequent in temperate conditions affecting SOC dynamics. Grasslands ecosystems under temperate moist conditions are subject to processes that may differ from arable systems in regards with SOC sequestration. In particular, below-ground plant residues in grasslands provide important C inputs for soil C sequestration: Grassland species allocate more C below-ground than cereals and below-ground C has longer residence time than above-ground C. Moreover, rhizodeposition is an important source of C inputs, which is rarely quantified and still remains the most uncertain component of soil C fluxes in terrestrial ecosystems. Furthermore, grazing and wheeling by vehicles can cause damage soil and vegetation structure by trampling and poaching, both affecting plant production, and the potential amount of C inputs causing soil C loss. Under temperate moist conditions, precipitations are high and winters are relatively mild with a relatively long growing season, susceptible to poaching. Poaching is a common soil damage problem of livestock treading which has not been extensively simulated in grazing ecosystems. Also, temperate moist climatic conditions imply that soils are wet-saturated during certain wet periods in which decomposition of organic matter is limited. Therefore, improving the methods to estimate SOC stock changes in managed grasslands is key to obtain reliable estimates of SOC and determine the real contribution of livestock to the net global greenhouse gas emissions. Recent research in temperate grasslands has shown that grasslands can act either as C sink or source depending on how animals, vegetation, soil, climate, and management practices interact with each other. To study the long-term responses of SOC changes in grasslands, we can use both data from long-term field trials and simulation models. Models allows to obtain complementary information to trials, for example, hypothesis forming or/and to predict long-term responses of grasslands to climate change and management. For strategic studies, e.g. assessing potential of grasslands to sequester SOC in a region, simple soil models, e.g. RothC, ICBM, C-Tool and Yasso07 are most useful as they require a limited and easily available input data. The RothC model, originally developed for arable soils, is one of the models that has been most widely validated and effectively used for different cropland and grassland systems at different spatial scales. In general, RothC showed a good performance under grassland ecosystems. Studies using RothC for grassland ecosystems have required specific initialization using information from long term grassland experiments. On the other hand, there are also several limitations to RothC particularly under managed moist grasslands. For instance, RothC presented a limitation considering management. Despite the number of possible interactions in grassland systems, e.g. between plant, soil and animals, RothC simplified the effects of different management affecting some of these processes on grasslands and indirectly simulates grazing activity by altering the amount of total plant C inputs. As for animals C inputs, RothC offers default quality values for C inputs from grazing animals or manure applications. Moreover, the model does not consider the trampling effect on soil physical conditions related to grazing. Besides, under temperate moist climatic conditions, RothC model is unable to adequately predict C dynamics in waterlogged soils, which imply oxygen limitation and thus a decline in decomposition rate. Furthermore, as a general limitation, regarding plant residues, RothC does not differentiate between above- and below-ground C inputs. Considering the model limitations, we aimed to introduce modifications to RothC and assess the ability of the proposed modifications to predict the measured SOC stocks from intensive grassland sites under moist climatic conditions. To adapt RothC to temperate moist managed grassland, we hypothesized that the aforementioned factors (i) could be easily implemented in RothC, (ii) significantly affect SOC changes and (iii) could improve RothC predictions of SOC changes. To evaluate the suggested modifications, we assessed the model performance against published experiments through a stepwise approach, as well as its sensitivity to the main modifications. # Materials and methods ## RothC model overview The RothC -26.3 model divides the SOC into five fractions, four of them are active and one is inert (i.e., inert organic matter, IOM). The active pools are: decomposable plant material (DPM), resistant plant material (RPM), microbial biomass (BIO) and humified organic matter (HUM). The decomposition of each pool (except IOM) is governed by first-order kinetics, characterized by its own turnover rate constant and modified by environmental factors related to air temperature, soil moisture and vegetation cover, which are the main input parameters to run the model. Incoming plant C is split between DPM and RPM, depending on the DPM: RPM ratio of the particular incoming plant material or organic residue. Both of them decompose to produce BIO, HUM and evolved CO₂. The proportion that goes to CO₂ and to BIO + HUM is determined by the clay content of the soil which is another input to the model. The model uses a monthly time step to calculate total SOC and its different pools changes on years to centuries time scale. ## RothC tested modifications Four modifications were proposed and tested in this study to the RothC excel version (*“Rothc_single_layer_4\_active_pools_Feb_2013”*): (i) extensions of soil water content function up to saturation; (ii) enlargement of C input pools to account for the diversity of applied exogenous organic matter (EOM) from ruminant excreta; (iii) affinition of plant residue components and quality variability; and (iv) the trampling/poaching effect of grazing animals. ### Soil water saturation in RothC RothC contains a minimum rate modifying factor of 0.2, when soil moisture is at minimum moisture capacity (i.e., at the extreme of water limitation). However, no correction is applied under water saturation and when soil is oxygen limited. In order to represent the reduction in the decomposition rate above field capacity, the rate modifying factor for moisture was assumed to follow a linear decline trend until a minimum rate of 0.2 (20%), at saturation conditions, as suggested by in the ECOSSE model. ECOSSE soil moisture function was derived from SUNDIAL and RothC models under anaerobic and aerobic conditions and based on Rothamsted field experiment. The model was designed for use across a range of land uses, and water contents are included up to saturation. It was evaluated under European conditions and showed a good performance. Soil water contents at saturation and field capacity conditions are estimated by considering soil properties related to soil texture as described by. The conversion from soil water content to soil moisture deficit (SMD_i, mm) used in RothC referred to (S1 Appendix). ### Exogenous organic matter diversity (EOM) Exogenous organic matter partition for the RothC model was estimated by, based on an indicator of potential residual organic C in soils (IROC), which is derived from Van Soest fractions and the proportion of EOM mineralized during 3 days of incubation. Similarly, improved the prediction of SOC stocks in amended soils by fitting the RothC partitioning pools of different EOM to the respiratory curves. Such adjustment of the partition of EOM into RPM, DPM and HUM entry pools of RothC provided a successful fit and had been reproduced in other studies. However, the above-mentioned studies have summed up all the different animal excreta into one category and did not distinguish excretions from different animal types (e.g., ruminants, pigs…). In order to capture the specific characteristics of ruminant excreta, we developed a methodology based on as illustrated in S1 Fig in. In this study proposed a partition of the C inputs from excreta into RothC pools based on the relationship between lignin content (Van Soest fractions) and anaerobic biodegradability, estimated as follows (Eq): $$B = 0.905 \times {\mathit{\exp}\left( - 0.055 \times lig(\%) \right)}$$ Where B is biodegradability and Lig is lignin content as % of Volatile Solids (VS). The Van Soest fractions are then partitioned into the pools of RothC based on its degradability, represented by the parameter B (i.e, lignin, holocellulose and solubles). A fraction of lignin is allocated into the HUM pool, representing the most resistant material. The rest of the lignin and the most resistant fraction of holocellulose and solubles are assigned to the RPM, while the most labile fraction of holocellulose and solubles are allocated to DPM. This is expressed as VS %, following the equations. <img src="info:doi/10.1371/journal.pone.0256219.e002" id="pone.0256219.e002g" /> H U M = L i g × ( 1 − B ) <img src="info:doi/10.1371/journal.pone.0256219.e003" id="pone.0256219.e003g" /> R P M = l i g × B \+ ( H o l o c e l l u l o s e \+ S o l u b l e s ) × ( 1 − B ) <img src="info:doi/10.1371/journal.pone.0256219.e004" id="pone.0256219.e004g" /> D P M = ( H o l o c e l l u l o s e \+ S o l u b l e s ) × B The Van Soest fractions were derived from literature review for animal excreta of ruminants. As a result of this review, we identified large variability in animal excreta’s fractions (lignin: S2 Fig in and soluble: S3 Fig). These differences were associated to a diverse array of factors and especially those in relation with the animal diet composition (e.g., high concentrate diet generally would imply lower lignin content in the ruminant´s excreta). For the ruminant excreta quality to the RothC entry pools, we used as input to the above questions an average value for the different fractions considered (data from S2 and S3 Figs). Additionally, in a separate exercise, we evaluated how the impact of uncertainties of these fraction values could lead to uncertainties on the SOC results. For this exercise, both extreme values (i.e., maximum and minimum) were assessed using a sensitivity analysis (See Sensitivity analysis section). ### Plant residue: Components and quality The RothC model does not distinguish between above- and below-ground plant residues. We hypothesise that accounting for month-to-month changes in plant residue quality may improve RothC predictions under wet conditions, while not adding too much complexity to the modelling approach. Regarding plant C inputs distribution, RothC is known to be relatively insensitive to the distribution of C inputs through the year. Model users generally use above-ground residues as surrogate for total plant C inputs and do account less for root inputs in RothC. Here we separated the plant residue C inputs into three components (i.e., above-ground residues, below- ground residues and rhizodeposits). The structure of C input derived from plant residues in RothC modified model is as illustrated in S4 Fig in. Parting from above-ground biomass, we used root to shoot (R:S) ratio to estimate below-ground biomass (when its value is not available). We assumed N fertilisation as the main driver for R:S ratio in grasslands as many studies have proved the strong dependence of the latter on N inputs. We referred therefore to equation for RothC C input parameterisation under temperate grasslands in order to consider the fertilisation effect on the R:S ratio: $$R:S = 4.7375e^{- 0.0043.\ N\ input}$$ Where R:S is the Root: Shoot ratio and N input is nitrogen fertilisation expressed in kg N ha^-1 year^-1. Unlike in annual croplands, in perennial grassland ecosystems, below-ground C biomass does not correspond to the below-ground residue. Instead, below-ground residues correspond to 50% of the total below-ground C biomass since the average annual root turnover of grasslands has been estimated to be 50% in the temperate zone. Regarding rhizodeposition estimation, we referred to an extensive literature review in which net rhizodeposition-to-root-ratio from grasslands was estimated to be 0.5. We assumed a C concentration of 45% of the plant biomass. Plant residue quality (biochemical composition), as one of the main drivers of decomposition, is represented in the RothC model by the DPM:RPM ratio (i.e., ratio of rapidly and slowly decomposing pools), which can be obtained by optimization to obtain the best fit according to different land use types. For instance, for most agricultural crops and improved grasslands, RothC uses a DPM: RPM ratio of 1.44 (i.e. 59% of the plant material as DPM and 41% as RPM). For unimproved grasslands and scrubs (including Savannas) a ratio of 0.67 is used. Plant residue quality is variable in time and depends on several factors (e.g., maturity stage, climate variables and nitrogen fertilisation). In order to fit the DPM: RPM ratio to the specific conditions of temperate grasslands, including its variability over the year, we used the stepwise chemical digestion (SCD) method, already used by. For simplicity, we assumed that the DPM pool could be approximated to the C in the Neutral Detergent Soluble (NDS), and the RPM pool as the C in the Neutral Detergent Fiber (NDF) (i.e., holocellulose and lignin fractions). Regarding below-ground plant material quality, the quantity of lignin itself is the main potential driver of differential degradation between both above- and below-ground plant components. Therefore, we added up the difference of lignin percentage of \~ 8% (between above- and below-ground parts) to get the below- ground RPM pool, referring to. The DPM pool is then derived by subtraction according to the equation: $$DPM(\%) = 100 - RPM(\%)$$ Finally, we assumed that the C inputs derived from rhizodeposition are transferred to DPM of the RothC because of the expected rapid decomposition of this labile substance by rhizosphere microorganisms. ### Animal trampling effect: Poaching We hypothesise that accounting for animal trampling may improve RothC predictions, while not adding too much complexity to the modelling approach. The trampling effect generally depend on stocking density, soil moisture content, soil texture, and the presence/absence of a protective vegetation cover. Apart from the stocking rate, the remaining factors were reflected in the RothC default model. In this context, we developed a simple poaching modification based on available data obtained from temperate moist grassland studies, considering that our modelling should be validated apart. The main objective of introducing the poaching effect was to predict the level of soil damage and its impact on plant C inputs as a function of soil moisture, soil compaction and degradation under grazing conditions (e.g. under different stocking rates) (S5 Fig). Soil moisture is estimated in RothC using the Soil Moisture Deficit (SMD) value that considers rainwater inputs and soil texture properties (i.e., clay content). According to, we used SMD as a proxy for soil moisture to predict when soil water conditions are likely to lead to hoof damage. For simplification reasons, we assumed water saturation conditions from SMD = -10 mm onwards (according to the soil moisture modification), as a condition of poaching occurrence as in. Livestock density has an effect on the level and extent of treading damage illustrated by hoof print depth and soil surface deformation. Depending on the soil surface deformation of a treading event, the pasture production is reduced and thus its plant C input (S4 and S5 Figs). The main equations related to the conceptual diagram of poaching modification are described in S1 Appendix in. As the poaching effect in temperate grazing systems seems to cause only short- term reduction in pasture plant production but there is a relatively fast recovery to these damages, we considered that plant C input reduction due to poaching effects would only occur in months when soil was prone to poaching. ## Study sites and input datasets ### Study sites description In order to validate the proposed modifications, we used data from four studies of European managed grasslands having temperate conditions and being characterized by precipitations \> 1000 mm, and grass and clover mixture. The grassland sites (Laqueuille intensive grazing grassland, Oensingen intensive cutting grassland, Easter Bush intensive grazing grassland and Solohead dairy research farm) were mainly selected from the FLUXNET program (<http://www.fluxnet.ornl.gov/>;. Information on geographic and climatic characteristics, soil properties and management of the different sites are listed in the. More details are provided in S2 Appendix in. ### Input data for the model and main assumptions Plant carbon inputs in the different sites were estimated depending on the available data using the method described in the section “Plant residues: Components and quality”. For the Laqueuille site, average above-ground C residue was obtained from available measured data and it represented 20% of above-ground C standing biomass. We used the R:S ratio to estimate below-ground biomass from average measured above-ground standing biomass. Below-ground C residues were assumed to be 50% of the below-ground C biomass. For the Oensingen site, average above- and below-ground C biomass were obtained from. We used the same assumption as for cutting grasslands, assuming that 30% of the above-ground biomass is not harvested, and that only 50% of that fraction is turned over annually and becomes available for soil organic matter formation. To estimate below-ground C residue, we used the same assumption as commented for Laqueuille site. The same assumptions were considered for the grazing Easter Bush site. From the average measured above-ground biomass we assumed only 20% as residues as in the Laqueuille grazing site and the same hypothesis for the below-ground C residue as in the other previous sites. For Solohead dairy research farm, we used as input the available measured data of above- and below-ground C biomass and used the same assumption for above- and below-ground C residues as all the previous sites. Finally, for the rhizodeposition as commented previously, we used an annual net rhizodeposition-to-root ratio of 0.5. The proportions of plant C input added to the soil in each month were set as the pattern of inputs typical of European grasslands suggested by. Referring to plant residue quality we ascribed RPM and DPM pools related to NDF and NDS, respectively for each plant residue component (as described in the sub-section “Plant residues: components and quality”). The C animal excreta in Laqueuille grazing grassland were derived from referring to the C intake grass-based rations, as the management is a continuous grazing from May to end of October without additional feed supply. Therefore, we estimated the C animal excreta as 32% of the measured C intake using average values for the simulation period 2004–2012. Annual C derived from cattle slurry in Oensingen site were estimated from as an average of the provided years. Carbon input from grazing animal excreta was estimated the same as in Laqueuille site, while annual C input derived from organic fertilisation for Easter Bush was deduced from during the period 2004–2010 as 0.32 Mg C ha^-1yr^-1. The same method was used to estimate annual total N fertilisation and annual stocking rate of this site. For Solohead dairy research farm, C input derived from animal excreta were calculated the same as in Laqueuille site and all other input data were estimated as average annual from the same study. The different input data to the model regarding management, soil properties to estimate soil water content at saturation and field capacity conditions, as well as grass type to characterise the plant residue quality for the different study sites are illustrated in. ### Model initialisation and running For RothC initialisation, since radiocarbon measurements are costly and rarely performed routinely, we used the pedotransfer functions established by to estimate all active C pools from initial provided measured SOC stocks. The initial IOM pool, according to these pedotransfer functions was set to match the equation proposed by: $$IOM = 0.049\ {TOC}^{1.139}$$ We modelled SOC dynamics from the different study sites using a stepwise approach. First, we used the default RothC version (RothC_0) and, subsequently we progressively added the different modifications tested : (i) ruminants excreta (RothC_1 modification); (ii) plant residue components and its characteristics (RothC_2 modification); (iii) saturation conditions for soil water function (RothC_3 modification) and (iv) soil poaching (RothC_4 modification). Soil organic carbon stocks were simulated at 20 cm depth for Laqueuille and Oensingen and at 30 cm depth at Easter Bush and Solohead dairy farm. ## Model evaluation We used different performance indices and threshold criteria based on (S1 Table). The ability of each modification to improve SOC dynamics simulation was evaluated using the root mean square error (RMSE), mean difference of simulations and observations (BIAS) and the model efficiency (EF) (S1 Table). ### Sensitivity analysis Several studies have indicated that the RothC model is most sensitive to C inputs. In our study, analyses were performed to test the sensitivity effect on SOC changes of the different modifications (other than C inputs) implemented in the model, using RothC_4. Model sensitivity was expressed as an index, which considered different input values related to the modifications (i.e., plant residues quality, ruminant excreta quality and soil moisture up to saturation) from minimum to maximum and then the output values were analysed according to the following index. <img src="info:doi/10.1371/journal.pone.0256219.e008" id="pone.0256219.e008g" /> S e n s i t i v i t y i n d e x = m a x ( P i ) − m i n ( P i ) m a x ( P i ) Where max (Pi) is the maximum output value and min (Pi) is the minimum output value. We used NDF as a proxy for RPM in relation with plant residues quality , assuming that NDF varies from 30 to 70% as minimum and maximum values based on 15 papers (S2 Table). We used the lignin fractions (% VS) as a proxy for EOM in relation with ruminant excreta quality assuming minimum and maximum values from literature values shown in. Similarly, for soil moisture variation, we tested minimum (0.2) and maximum values (1) of the rate modifying factor for moisture. # Results and discussion ## Measured versus simulated SOC stocks All four sites showed, in general, a similar pattern of annual SOC stocks with the RothC default version (i.e., RothC_0) as well as with the four modified versions. In all four sites, the lowest simulated SOC stocks were observed in the default model version (RothC_0). RothC_0, for Laqueuille, Oensingen and Solohead sites, simulated that SOC was reduced during the time of the experiment, which was the opposite trend that measurements showed. For example, in the Laqueuille intensive grassland, SOC stocks predicted by the RothC_0 version decreased from 114 to 102 Mg C ha^-1 whereas measured values increased from 114 to 121 Mg C ha^-1. By implementing changes to account for ruminant excreta quality (RothC_1) on the study sites, the model resulted in a slight increase in SOC in time. Moreover, this SOC increase was lower than that simulated by RothC_2. Changes in the modification of plant residues (RothC_2) resulted in greater SOC increased values in time when compared with the previous modification (RothC_1). The lower effect of the simulation of animal excreta characteristics in RothC_1 could be explained by the higher quantity of plant residues while adding the rhizodeposition component together with above- and below-ground components in RothC_2. By introducing the soil moisture modification in RothC (RothC_3), the model simulated an increase in SOC stocks which, trend-wise, differs from the RothC_0 model, but coincides with measured data. For example, SOC stocks at the end of the simulation period in 2011 reached 88.38 Mg C ha^-1 (RothC_3) compared to 83.7 Mg C ha^-1 (RothC_0) in the Easter Bush intensive grazing grassland. Soil moisture modification at saturation reduces decomposition rates values for very wet soils conditions. In fact, the 4 sites used in our study have soil water saturation during many months of the year (with an average of 8 months). Including the poaching effect (RothC_4), resulted in slightly reduced SOC stocks compared with RothC_3, specially for the Solohead site. This reduction in SOC stocks in RothC_4 compared with the RothC_3 version could be explained by the reduction in plant C inputs due to poaching that typically occurs at saturation conditions. In general, the highest predicted SOC stocks values and the closest to the measured values at the end of the simulation period resulted after RothC_3 and RothC_4 simulations. For Laqueuille grassland intensive site, RothC_3 and RothC_4 were able to match the general trend of SOC increase (between 2004 and 2012) and the SOC stocks at the end of the simulation period, but not the change of SOC stocks corresponding to the year 2008. However, SOC simulation for Solohead research farm, using RothC_3 and RothC_4 modified versions were within the range of measured data of SOC stocks. ## Model performance In general, the RothC default version (RothC_0) showed a good performance with an EF value of 78%. However, the different cumulated modifications presented enhanced the predicting performance of RothC for these specific sites. In particular, simulated SOC stocks using the RothC_3 and RothC_4 versions almost matched measured values achieving model efficiencies of 99% and 98%. Therefore, these two modifications accurately predicted SOC changes. The negative bias (reaching -18.8 in Laqueuille site) and the higher RMSE values obtained in RothC_0 compared with the different RothC modified versions indicated that the default RothC version underestimated SOC stocks, especially in the Laqueuille and Solohead sites, which presented the highest SOC content. This confirmed the fact that the RothC model is unable to adequately predict soil C dynamics in organic or waterlogged soils. In this context, adding the modification of the soil moisture function in RothC_3 reduced the bias and the RMSE and improved the general trend of SOC stocks compared with the default version RothC_0 in all simulated sites. RothC_0 assumes high decomposition rates with high soil moisture, but it does not consider the cessation of the decomposition process which occurs in high wet soils close to saturation conditions, frequent in temperate moist grasslands. The inclusion of the ruminant excreta quality in the model only slightly improved the SOC predictions in RothC_1 compared to RothC_0. In this context, emphasised the importance of modifying the quality of residues to improve the model performance, concluding that the adjustment of DPM:RPM ratio led to better model performance than the use of default DPM: RPM values provided by the model. Comparing RothC_1 and RothC_2 versions, it could be deduced that integrating quantity and quality distinction of plant residue in RothC_2, as a primary source of SOC, improved SOC predictions. Adding the modification of plant residues in terms of quantity and quality contributed to improve SOC simulation compared to the modification of specifying animal excreta quality. The improvement showed by plant residues modification, particularly in Solohead Research farm, could be explained by the higher sensitivity of the model to C inputs quantity compared to C inputs quality and the importance of including rhizodeposition together with above- and below-ground components in plant C input quantification. Indeed, as a fundamental source of C inputs, rhizodepostion was recommended to be added to the different plant residue components in SOC models, particularly RothC. The poaching effect is assumed to reduce plant productivity and the potential amount of C inputs to the soil and thus causing SOC loss. Consequently, the poaching modification included in the RothC_4 version predicted reductions in SOC stocks compared to the RothC_3 version. The reduction in SOC stocks is explained by the lower C inputs during the months when grazing occurs under saturation conditions. Only in the case of the Easter Bush site, the poaching modification contributed to improve SOC predictions in the RothC_4 version. A possible explanation to this improvement in the SOC predictions is that the soil in Easter Bush site is poorly drained and grazing by ruminants occurs all year round and thereby highly susceptible to poaching. In the same context, enhanced the original PASIM grassland constructing a simple and empirical model of the detrimental impact on vegetation of trampling by grazing animals by removing at each time step a fixed proportion of the above- ground biomass. However, it is important to point out the complexity of the poaching effect, as it induces more impacts other than the detrimental vegetation impact which are beyond the scope of our study. In this context, pointed out the inconsistency and limitation of the studies dealing with the grazing effect on SOC. Therefore, more robust experiments are needed in order to define the severity of the poaching effect according to soil moisture saturation, livestock density and soil type. Therefore, particularly, RothC_3 showed the best agreement, as the effect of the poaching modification added in RothC_4 is minimal and uncertain. In this sense, the poaching modification could be of major importance under heavy stocking rates or overgrazing management associated to SOC loss. Testing the model performance based on each of the individual modifications for the different sites allowed improving our understanding of its impact to the model. Soil moisture up to saturation conditions in the soil water function of RothC showed the best performance compared with the other modifications. The modification of RothC water function at saturation conditions fit to the temperate moist climatic conditions, since the different study sites showed saturation conditions most of the year. However, the poaching effect alone contributed to reduce SOC stocks and thus the model performance, since the poaching effect is related to water saturation conditions. The enhancement in the model performance showed by the quality of ruminant excreta depends on its quantity. Indeed, the BIAS reduction with ruminant excreta quality modification compared with the default version (Tables) was more important in the grassland sites with major ruminant excreta application (e.g., Solohead research farm). However, the plant residue modification showed a higher improvement compared with the ruminant excreta quality as it implies an increase in C inputs with the inclusion of the rhizodeposition component. However, testing the model based on the combined effect of soil moisture up to saturation and poaching effect showed a great performance compared with the effect of excreta and plant residues for the different sites with a RMSE of 5.96 compared with 8.66. The modifications of soil moisture up to saturation and poaching effect reduced the BIAS compared with animal excreta and plant residue modifications for the different study sites, except for the Solohead research farm. This could be explained by the fact that the latter received higher C inputs derived from animal excreta and plant residues and lower water saturation conditions compared with the other sites. The modifications of soil moisture up to saturation and plant residues presented the best performance among all sites. Particularly, the plant residues modification implied an accounting for rhizodeposition component and thus a significant increase in C inputs compared with the minimum proportion of plant residues reduction induced by the poaching effect of grazing animals at saturation conditions. Therefore, the model modification with the greatest positive impact was soil moisture up to saturation (Tables). However, the impact of plant residues and ruminant excreta modifications depends on the C input quantity (Tables and). The poaching effect could not be considered without taking into account the soil moisture saturation modification, as it showed a lower performance than the default model version (Tables). ## Sensitivity analysis A sensitivity analysis of RothC_4 was performed to assess the robustness of the modifications (plant residues quality, ruminant excreta quality and soil moisture up to saturation) made in the different model versions presented. In general, RothC_4 seems to be more sensitive to C input quantity than to quality and to soil moisture function, particularly at saturation conditions. The sensitivity analysis performed for resistant plant residues pool with the RothC_4 version showed a sensitivity index varying between 0.8% for the Easter Bush site and 2.6% for Oensingen and Solohead research farm. Although the model was not very sensitive to the quality of plant residues, adding this modification enhanced the results depending on the quantity of plant residues. In this context, according to other studies, specifying plant C input quality depending on residues partitioning instead of using the default RothC ratio for DPM and RPM should enable more reliable modelling of SOM dynamics. In order to ensure the sensitivity of the model to the plant C inputs in terms of quantity, we assessed its sensitivity to the R:S ratio based on our extensive literature review for temperate grassland species (S3 Table in). The sensitivity shown by the model to plant residues was higher than the sensitivity to the plant residues quality (S4 Table). In relation to the sensitivity of the RothC_4 version to the animal excreta quality, the values of sensitivity index obtained for the different experiments were in general low (between 1.1% and 3%). So, the use of average value for the different animal excreta fractions does not really impact the results, as we implemented in EOM modification. As for plant residues, the greatest value for the Solohead research farm could respond to the higher C inputs derived from animal excreta that received Solohead research farm as compared to the other sites. In order to focus on RothC_4 sensitivity to animal excreta quality with relation to its quantity, we assumed an annual C input derived from animal excreta of about 2.5 t C ha^-1 distributed between March and June for the remaining sites that receive smaller amount of organic fertiliser. As animal excreta quality in the RothC model is connected to its quantity, the sensitivity index of animal excreta quality increased as its quantity increased (S4 Table). In this context, according to, RothC displayed a moderate sensitivity to variations in animal excreta quality, more specifically the ratio between decomposable and resistant pools. Sensitivity index regarding soil moisture modification was higher compared with the other modifications reaching, for example 12.8% in the Laqueuille site. The variation in the sensitivity among the different study sites depend on their soil properties. Therefore, the modified model is sensitive to the rate modifying factor for soil moisture up to saturation under temperate moist climate conditions. In this context, concluded that reliable prediction of carbon turnover requires that the soil moisture together with the soil temperature reduction functions of the model need to be valid for the environmental conditions. ## Sources of uncertainty and research needs Although RothC_3 and RothC_4 simulations performed well in simulating SOC changes for the selected sites, there were limitations related to the uncertainty of, both, model inputs and modifications, and to the limitation of the data used for validation. Regarding model inputs, uncertainty was mainly related to the lack of detailed measured data of C inputs derived from plant and/or animal origin. In this study, we used the average of available measured values (details can be found in the section “Input data for the model and main assumptions”). However, measured C inputs is not always available, so its value could be supplied via linkage with another model, considering the grazing effect (case of plant residues). It is important to point out that previous studies running RothC in grassland ecosystems overestimated C inputs and there is a lack of detailed information on how plant residues were estimated and/or assumptions regarding their conversion to C inputs. In particular, the estimation of below-ground C inputs is another major source of uncertainty for SOC modelling. Indeed, belowground C inputs depend on multiple factors, including site-specific agronomic practices and the response of plant genotypes to them, and direct measurements of belowground C inputs is a challenging issue. For instance, if we estimate R:S ratio according to Eq with the measured values in Oensingen site, we found close values of 1.9 and 1.5, respectively. However, for the Solohead research farm, the values were more different with a measured R:S ratio of 0.88 compared to an estimated value of 2.1. Moreover, the use of pedotransfer equations for initialising SOC pools, as an alternative for soil physical fractionation, may represent another source of uncertainty. Indeed, although the reliability of pedotranfer equations, they could reveal some errors which are in the range of measurement error for SOC. Regarding model modifications, a linear decline in the rate modifying factor for soil moisture was assumed under saturation conditions, like in the ECOSSE model, as there was not sufficient evidence to suggest a more refined relationship as indicated by. However, the effect of soil moisture on SOC dynamics is complex and non-linear, interacting with temperature effect. Improvements could be achieved by using a more refined function based on robust field experiments in order to better represent the different grassland sites. Furthermore, our estimations of animal excreta quality, based on literature review, are not conclusive and further refinements based on experiments could be made as, for example, to account for animal intake quality to predict its excreta quality. Regarding the poaching effect modification, based on the literature review we made, the number of long-term experiments under temperate moist region is limited. Moreover, due to the complexity of the soil damage (i.e., poaching) in which many factors could be involved (i.e., soil, animal, plant), it is difficult to generalise our findings. The lack of usable, mechanistic simulation models of soil deformation under hooves and wheels is partly due to the lack of appropriate conceptual understanding and theory of the complex soil mechanical processes involved as well as the shortage of good and relevant experimental data. Our equations and values suggested for the different modifications are representative for the conditions of moist temperate intensive grasslands and other site-specific equations, that are tailored to the objective study site, could be used. In our study, simulations of the different modifications were compared to measured data of different study sites. However, field measurements also have deviations, which are source of uncertainty as they are used as the scale to evaluate model performance. As future improvements, measurement of the different input data to the model (e.g., plant residues) would maximise the accuracy of estimations. However, this technique involves time, cost and labour. As an alternative non-destructive method, combining the process-based model RothC with machine learning techniques can successfully help infer additional information from incomplete data sets. For instance, the machine learning algorithms based on remote sensing data, such as the ArtificialNeural Network as a powerful empirical modelling, could improve the estimation of above-ground biomass with higher accuracy. For future work, our modifications could be reproduced and/or refined to improve assessments of SOC changes in managed grasslands under temperate climatic conditions not only at a plot level but also at regional level since grassland systems continue to be understudied at broader scales. # Conclusions This study adapted the RothC model to managed grasslands under temperate moist conditions. The proposed modifications to the model considered the incorporation of distinction for plant residues components (i.e., above- and below-ground residues and rhizodeposition) in terms of quantity and quality and distinction for ruminant excreta quality, the extension of soil moisture up to saturation conditions and, finally, the introduction of the livestock damaging effect (i.e., poaching) on plant residues under water saturation conditions. The moisture response modification and the partition of C inputs derived from plant residues components improved the model predictability, but plant residues and ruminant excreta quality modifications improved the model predictability at a lesser extent. Finally, poaching simulation did not improve the model, since it results in complex and multi-factorial effects in these temperate grasslands. These modifications do not impair the performance of the model under temperate conditions. Indeed, they represent a broadening in the capability of the RothC model to simulate managed grassland under temperate moist conditions. It must be kept in mind that although there was good agreement between results from modified model and measured data from different studies, validating against more sites would greatly improve model confidence. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction In the pathogenesis of acute myocardial infarction (MI) common risk factors for atherosclerosis as well as hemostatic factors are important determinants. Thrombin is a central enzyme in the blood coagulation cascade, regulating platelet activity and fibrin clot formation. In addition, thrombin appears to modulate atherosclerosis formation and plaque phenotype, at least in experimental studies. Clinically, markers of thrombin generation have been linked to risk of recurrent arterial thrombosis, but the evidence is still weak. We recently showed that the level of thrombin generation in plasma was elevated during and following a first MI. Contra-intuitively, endogenous thrombin potential (ETP), one of the main parameters derived from thrombin generation analysis, tended to be lower in those who had a subsequent recurrent arterial vascular event, which in conjunction with an elevated level of D-dimer yielded a significant risk for recurrence of 5.8. These analyses were however confined to plasma samples collected during the acute phase of AMI. Since *in vitro* thrombin generation analysis is typically aimed at establishing the *potential* to generate thrombin in a given plasma aliquot, rather than the actual amount of thrombin formed *in vivo* in the acute phase, we set out to compare thrombin generation in plasma samples collected in individuals with previous myocardial infarction and matched controls, collected in the Glasgow Myocardial Infarction Study (GLAMIS). Since, in patients with stable coronary artery disease we expected only modest differences in clotting potential we chose to stimulate plasma with low levels of tissue factor, ie the commonly used 1 pM as well as a slightly higher 2 pM tissue factor concentration. We calculated odds ratios for myocardial infarction based on *in vitro* thrombin generation data, in order to assess the importance of thrombin formation in plasma, in the absence of platelets, as a risk factor for coronary artery disease. # Methods As described in previous papers, GLAMIS was established to investigate associations of plasma haemostatic and inflammatory variables with previous myocardial infarction (MI) and conventional risk factors in a case-control study. The aim was to recruit all men and women with MI in the North Glasgow MONICA study diagnosed by W.H.O. - MONICA criteria from July 1994, between 3 and 9 months after the event, minimizing acute phase influences. Cases were patients with MI in this population survey who were still alive and willing to give consent (75% response rate). Controls were selected from a random sample of the same north Glasgow population, obtained from general practice registers, and frequency matched for sex and age (within 1 year), who had no history or electrocardiogram evidence of MI. Written informed consent was obtained from all participants, and the study was approved by the local Research Ethics Committee: Greater Glasgow Health Board Research Ethics Committee. Participants completed a general health questionnaire and blood pressure was recorded. A forearm venous sample was taken after a full overnight fast and care was taken to have properly collected blood samples in all individuals. Lipid assays were measured as previously described. Venous blood was anticoagulated with trisodium citrate (3.2%(w/v)) and centrifuged at 2000×g for 10 min at room temperature within 2 h of sampling, and aliquots were stored at −70°C until assay. For these analyses only previously unthawed plasma samples were used. ## Thrombin Generation Assay One of the methods developed for studying *in vitro* thrombin generation in plasma is the Calibrated Automated Thrombogram (CAT: Thrombinoscope BV, Maastricht, The Netherlands). This assay is based on the principle that rather than assessing the coagulation status at a single timepoint, the potential to clotting could reveal the actual risk of clotting under pathologic conditions, eg plaque rupture. To this end plasma is usually triggered with small amounts of tissue factor (TF) and the thrombin generation curve is measured in time. By adding thrombomodulin also the influence of the natural anticoagulant protein C system can be addressed. Briefly, *in vitro* thrombin generation in platelet- poor plasma was measured by means of the CAT method (Thrombinoscope BV, Maastricht, the Netherlands), which employs a low affinity fluorogenic substrate for thrombin to continuously monitor thrombin activity in plasma. Measurements were conducted in 80 µL platelet-poor plasma in a total volume of 120 µL (20 µL fluorogenic substrate, calcium and 20 µL solution of TF and phospholipids). Final concentrations of TF were 1 (PPP Reagent Low) and 2 pM TF (kind gift of Thrombinoscope BV) in the presence of 4 µM phospholipids (phosphatidylserine/phosphatidylcholine/phosphatidylethanolamine vesicles in HEPES-buffered saline). Since we anticipated that differences in the thrombin generating potential would be modest between cases and controls under non-acute conditions, we chose two low stimuli of plasma: 1 and 2 pM of TF. Although this would theoretically pose a risk of being susceptible to contact activation, based on our previous data we did not expect much contact activity influence at TF concentrations \>1 pM. In order to have a trigger that would be a little more robust we also tested a 2 pM TF concentration. To assess the contribution of the protein C pathway on *in* vitro thrombin generation, all measurements were conducted in the absence and presence of thrombomodulin (TM), titrated to obtain 50% reduction in ETP of normal pooled platelet poor plasma. In order to correct for inner-filter effects and substrate consumption, each thrombin generation measurement was calibrated against the fluorescence curve obtained in the same plasma to which a fixed amount of thrombin-α2-macroglobulin complex was added (Thrombin Calibrator, Thrombinoscope BV). Fluorescence was read in a Fluoroskan Ascent reader (Thermo Labsystems OY, Helsinki, Finland) equipped with a 390/460 nm filter set and thrombin generation curves were calculated with Thrombinoscope software (Thrombinoscope BV). Three parameters were derived from the thrombin generation curves: lag time (initiation phase of coagulation), endogenous thrombin potential (ETP; area under the thrombin generation curve), and peak height. Within and between run coefficients of variations for *in vitro* thrombin generation were below 10% for all parameters derives, as described previously. ## Statistical Analysis Thrombin generation data for lag time, normalized ETP (nETP), normalized peak height (nPeak Height) and reduced ETP were not normally distributed and are thus presented as median \[95% CI\]. Differences between cases and controls were calculated using the Mann-Whitney U test. Thrombin generation parameters were categorized based on the cut-off values of the highest 10% in the control group. Logistic regression was used to obtain odds ratios (ORs) and corresponding 95% confidence intervals (95% CI) as a measure of relative risk with the lowest category as a reference category. Analyses were adjusted for the stratification factors: hypertension (defined as diastolic blood pressure \>90 mmHg and/or systolic blood pressure \>140 mmHg), BMI, hyperlipidemia (defined as total cholesterol \>6 mmol/L), diabetes mellitus, (history of) smoking, and gender. All statistical analyses were performed using SPSS version 19. P\<0.05 was considered statistically significant. # Results The present analysis includes 171 cases and 185 controls. Baseline characteristics are indicated in, showing significantly more unfavorable risk factors and biochemical parameters in the cases compared to the controls, except for both systolic and diastolic blood pressure, which were higher in the control group. With regard to medication, we excluded any users of oral anticoagulants, because of its known inhibitory effect on thrombin generation. Aspirin and statins were used in 95% and 18% of cases and in 12 and 2% of controls, respectively. ## Thrombin Generation and show thrombin generation parameters for cases and controls. The ETP and peak height were normalized against pooled plasma from healthy donors in order to correct for between run variance, as reported previously. Although not significant, thrombin generation triggered with 1 pM TF was delayed in cases as indicated by a prolongation of the lag time from 5.6 min for controls to almost 6.0 min (p\>0.05) for cases, whereas the overall potential to generate thrombin was significantly increased in cases. Both the normalized ETP (180% \[145–206\]) and peak height (245% \[170–324\]) were enhanced in cases compared to controls (ETP: 168% \[143–189\], peak height: 213% \[166–279\], p\<0.05). Prolongation of the lag time (3.61 min \[3.00–4.00\] vs. 3.33 min \[2.94–3.67\], p\<0.05) for cases was confirmed by thrombin generation triggered with 2 pM TF. At this higher TF trigger, the normalized ETP and peak height were both higher in cases than in controls but the difference was not statistically significant. At both TF triggers, the addition of thrombomodulin did not reveal any differences in reduction of the ETP between cases and controls. Thrombin generation parameters showed statistically significant correlations with some of the classical risk factors for myocardial infarction, including BMI, cholesterol and triglycerides. However, almost all correlations were weak (Rs\<0.35) and not always consistent between the 1 and 2 pM TF trigger (HDL). All risk factors (age, BMI, hypercholesterolemia, hypertension, diabetes and gender) were included in the multivariate analysis. shows the odds ratios (ORs) for myocardial infarction for the different thrombin generation parameters, triggered with both 1 and 2 pM tissue factor. Unadjusted ORs for myocardial infarction with increasing thrombin generation parameters at 1 pM of TF trigger were 2.7 (95% CI 1.4–4.9) for the normalized ETP and 2.5 (95% CI 1.3–4.6) for the normalized peak height for cases in the highest 10^th percentile. ORs for the lag time and ETP reduction in the presence of TM were not statistically significant. Analysis of thrombin generation parameters obtained with the 2 pM tissue factor trigger revealed essentially similar ORs as those for 1 pM TF. Adjustment for putative confounders (age, gender, BMI, hypertension, smoking, diabetes, hypercholesterolemia) attenuated the OR for the normalized ETP to 2.4 (95% CI 1.3–4.5) and enhanced the OR for the normalized peak height to 2.6 (95% CI 1.3–5.0), both assessed with the 1 pM TF trigger. At 2 pM TF, adjusted ORs for myocardial infarction for the normalized ETP and peak height were comparable to the 1 pM condition showing increased OR’s for normalized peak and ETP in cases versus controls. The use of aspirin or statins did not significantly affect any of the thrombin generation variables (data not shown). # Discussion The present data show that in patients with previous myocardial infarction *in vitro* thrombin generation in plasma is altered as compared to control individuals, showing a prolonged lag time (at the 2 pM tissue factor trigger only) and an increased normalized ETP and peak height (albeit at the 1 pM TF concentration only). Moreover, those cases with the highest levels of thrombin generation (normalized ETP and peak height values) also have an increased risk of MI, with adjusted OR’s of between 2 and 2.6, for the different test conditions. Taken together, these data suggest that the sustained increased thrombin generation potential is a pathophysiologically relevant characteristic and indicator of a hypercoagulable state in plasma from patients with coronary disease. The prolonged lag time in cases of the present study is in accordance with the observation in the MARK study, where lagtime was prolonged in the acute phase but also at later timepoints. This prolongation may be due to release of TFPI from perturbed endothelium such as demonstrated in a recent case control study in young women with manifestations of arterial thrombosis (AMI or stroke). However, also other changes in the balance between pro- and anticoagulant proteins may be involved in the alterations in thrombin generation profile. Although there appears to be an effect of common cardiovascular risk factors including BMI, cholesterol and triglycerides on the plasma phenotype, such effects were modest at best in the present analysis. The effect of BMI was recently found to be due to total body fat in an analysis of the Hoorn study and probably mediated by inflammation. The effect of cholesterol and triglyceride has also been postulated to influence coagulation in earlier studies. Obviously, there is also the possibility of unrecognized confounding. The use of medication that might affect thrombin generation, including aspirin and statins did not affect any of the thrombin generation variables. For aspirin an effect in platelet poor plasma is not to be expected, although one recent report indicates a trend for accelerated thrombin formation in long-term aspirin users. The use of statins might have attenuated thrombin generation in platelet poor plasma, as reported in several studies. The fact that we did not observe any effects may have been due to the relatively low percentage of statin users in the present study. The present data provide circumstantial support for a causal association between hypercoagulability and arterial vascular disease, as can be postulated on the basis of extensive experimental data. Increased thrombin production may add to the risk of thrombosis in conditions, such as atherosclerosis, when the anticoagulant reserve may be limited and erosion or rupture of the atherosclerotic plaque cannot be properly compensated, resulting in atherothrombosis. It is uncertain which mechanisms link cardiovascular risk factors to thrombin generation, but inflammation driven tissue factor production within the vasculature may be a common mechanism. Inflammatory activity can trigger hypercoagulability in various ways, involving cellular and plasma compartments. In our analysis we only observe effects in platelet poor plasma, which however is the product of interactions with vessel wall and other cellular mediated reactions. It is attractive to consider the plasma coagulation proteome, studied in the *in vitro* thrombin generation assay, as a product of the bidirectional interplay between clotting, vessel wall and inflammatory systems. This may explain occasional “paradoxical” outcomes as prolonged lag time or reduced ETP and peak height in patients in the *acute* phase of coronary artery disease, where counteracting forces (including TFPI) may delay but not impair thrombin generation in plasma. Several comments regarding the study design and interpretation of outcomes should be made. First, MI was defined according to the WHO-MONICA criteria from 1994, which may have resulted to greater heterogeneity in the case selection than would have occurred according to current criteria. Second, inherent to the study design, where blood sampling in the cases took place three to nine months after the MI no firm conclusions can be drawn on causality. The possibility of reverse causation should indeed be considered. Besides, the diagnostic value of *in vitro* thrombin generation analysis remains to be established. Given the substantial overlap in thrombin generation values between cases and controls, it is unlikely that thrombin generation analysis will provide a diagnostic or prognostic tool for coronary disease. The main advantage of this and similar studies is to obtain better insight in the contribution of plasmatic hypercoagulability to cardiovascular disease, which may ultimately help to identify patients at high risk of fibrin clot formation. As expected in this case-control design, the case group shows more unfavorable risk factors for cardiovascular disease than the control group, although blood pressure was lower in the cases. This probably reflects blood pressure lowering therapy as secondary prevention after the MI. A limitation of the study is the significant time of storage of plasma samples until assay, a potential problem in all studies relying on stored plasma material. However, any measurement error incurred by long-term storage would lead to underestimation of the case-control differences. While limited literature, suggests there may be some loss in factor VII and factor VIII activities in stored samples over time there is no good reason to expect that, other than a systemic effect on all plasma determinations, this would alter the differences between cases and controls. Nevertheless, the potential of storage associated effects on the laboratory data cannot be excluded. A confounding effect of contact activation at these rather low TF triggers cannot be fully excluded, however, one would have expected a shortening of lag time in case of a significant contribution of contact driven thrombin generation. These test conditions were chosen in order to make the test sufficiently sensitive to the expected, modest differences in thrombin generation that might have escaped detection at the most common 5 pM TF concentration. The downside may be that the TG analysis becomes more sensitive to artificial contact activation. In spite of a certain uncertainty about contact activation interference, there is no good reason to suspect that a random effect of contact activation would contribute to a systematic difference between values in cases and controls. Therefore, we conclude that the observed differences are patient related in cause and not due to pre-analytical artifacts. In conclusion, we report sustained increases in thrombin generation three to nine months after a MI. Whether this hypercoagulable state indicates specific patients at risk of arterial thrombosis, due to increased plasma based clotting activity, posing a risk factor for recurrent cardiovascular disease, should be the topic of further research. The authors whish to thank Diane Fens, Patricia Pluijmen and Rene van Oerle for laboratory assistance. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MS MW GL. Performed the experiments: HS AR RvO. Analyzed the data: MS AD MW HtC GL. Contributed reagents/materials/analysis tools: HS HtC GL. Wrote the paper: MS HS MW HtC GL.

# Introduction As a species-rich superfamily within the Lepidoptera, the Pyraloidea comprises more than 15,576 species with a world wide geographical distribution. The Pyraloidea are of particular interest because it contains a large number of notorious pest of commercial crops, forests and ornamental plants, stored foodstuffs with significant economic importance. They are of further interest because of their diverse life history adaptations including larvae with phytophagous, detritivorous, coprophagous, parasitic habits, and even aquatic habitats, has prompted the idea that pyraloids could be an ideal model for biodiversity. An efficient taxonomy, management, and pest control of these important moths rely on a sound and comprehensive classification and phylogeny. Initially, apart from Crambidae and Pyralidae, several other families, including the Pterophoridae, Thyrididae, Hyblaeidae, Alucitidae, and Tineodidae, were historically recognized within the Pyraloidea. With a better understanding of moth and butterfly phylogenies, the current consensus view holds that most of the families formerly included in the Pyraloidea should belong to their own superfamily, and there is strong evidence from molecular studies for the sister relationship of Pyraloidea and Macroheterocera. Additionally, the monophyly for Pyraloidea and two members, the Pyralidae and Crambidae, is supported by both morphological and molecular analyses. And the relationships among subfamilies for Pyralidae and Crambidae has also been investigated deeply based on nuclear gene data. However, due to limited samplings only a few mito-phylogenetic analyses have been involved in subfamily-level relationships. The mitochondrial genome (mitogenome) is a common and practical system for comparative genomic and phylogenetic research. Recently, owing to the improvement of polymerase chain reaction (PCR) and sequencing technology, in particular the application of next-generation sequencing, mitogenome data have soared in many animal lineages, especially insects. To date, over 380 lepidopteran mitogenomes have been determined, while only 30 pyraloids mitogenomes representing three subfamilies of Pyralidae and eight subfamilies of Crambidae are now available ([www.ncbi.nlm.nih.gov/nuccore/](http://www.ncbi.nlm.nih.gov/nuccore/)). A comparative study of mitogenome evolution and phylogeny of Pyraloidea highlights the requirement for broadening taxon sampling in further studies. In the present study, we report five complete and one nearly complete pyraloids mitogenomes including the first representatives from the Pyralinae (Pyralidae) and Glaphyriinae (Crambidae). In addition, we summarize the evolutionary pattern of pyraloids mitogenomic features including base composition, codon usages, secondonary structures of transfer RNA (tRNA) and ribosomal RNA (rRNA) genes, and control region. Furthermore, the molecular phylogeny of the Pyraloidea was reconstructed using multiple mitogenomic data, which confirms for the first time the phylogenetic position of Pyralinae and Glaphyriinae. # Materials and methods ## Specimen collection and DNA extraction All specimens were collected in August 2014 by a light trap at Xunyangba (33.33°N, 108.33°E), Ningshan County, Shaanxi Province, China, preserved in 95% ethanol and stored at -20°C. All these specimens were identified by Ping You. For each species, Voucher specimens have been deposited in the Insect Collection (Accession Number SNU-Lep-20140017-22), College of Life Sciences, Shaanxi Normal University, Xi’an, China 710062. The total DNA was extracted using a TIANamp Micro DNA Kit (Tiangen Biotech, Beijing, China), according to the manufacturer's protocol. ## PCR amplification and sequencing Six mitogenomes were amplified with overlapping fragments using conserved primers. PCRs were performed using FastPfu Fly DNA Polymerase (TransGen Biotech, Beijing, China) as previously described. After purification with PCR Purification Kit (Sangon Biotech, Shanghai, China), all PCR products were sequenced directly with a primer-walking strategy. ## Genome annotation and sequence analysis Contiguous sequences were assembled using Staden Package v1.7.0. PCGs and rRNA genes were identified based on homologous regions of published Pyraloidea mitogenomes using the Clustal X in MEGA 5. The tRNAscan-SE was used to predict tRNA genes and their secondary structures. Secondary structures of the two rRNA genes were predicted according to the models for *Paracymoriza prodigalis*. The base composition and codon usage were calculated using MEGA 5. ## Phylogenetic analysis Mitogenomic phylogeny of Pyraloidea was reconstructed based on four datasets (PCG123: 13 PCGs including all codon positions; PCG123R: 2 rRNAs, 22 tRNAs and 13PCGs including all codon positions; PCG12: 13 PCGs without third codon positions; PCG12R: 2 rRNAs, 22 tRNAs and 13PCGs without third codon positions) using Bayesian inference (BI) and maximum likelihood (ML) methods. Three species from Thyrididae (*Pyrinioides aurea*, KT337662), Alucitidae (*Alucita montana*, KJ508059) and Pterophoridae (*Emmelina monodactyla*, KJ508063) were selected as outgroups. Each of 37 mitochondrial gene sequences was aligned with Clustal. Considering partitioning strategy of previous studies, a similar partitioning scheme (tRNA genes, rRNA genes, and each codon site of PCGs) was employed for phylogenetic analysis. The optimal model (GTR+I+Γ) for each partition was selected using Akaike information criterion in jModelTest. The BI analyses were implemented in MrBayes 3.1.2 with four MCMC chains running for five million generations. Each set was sampled every 200 generations with a burn-in of the first 25% of steps. The ML analyses were performed using RAxML 7.0.3 with 1000 bootstrap replicates. # Results and discussion ## General features of Pyraloidea mitogenomes Full or partial mitogenomes of the six pyraloid moths (*Endotricha consocia*, *Hypsopygia regina*, *Orybina plangonalis*, *Evergestis junctalis*, *Tyspanodes striata*, *Maruca vitrata*) were generated and deposited in GenBank. In addition, 38 complete or nearly complete mitogenomes of the Pyraloidea were integrated into a combined dataset for conducting comparative analyses. As reported in most metazoan mitogenomes, all the pyraloid mitogenomes contained 37 mitochondrial genes including 13 protein-coding genes (PCGs), 22 tRNA and 2 rRNA genes, and a putative control region (namely A+T-rich region for insects). The length of five newly sequenced complete mitogenomes fell into the range of previously reported pyraloid mitogenomes (from 14,960-bp *Glyphodes pyloalis* to 15,594-bp *Ephestia kuehniella*). The gene order of pyraloid mitogenomes was highly conserved and identical with the typical gene order of Ditrysia mitogenomes. Compared with the predicted ancestral gene order for insects, however, a gene rearrangement occurred in the tRNA cluster (trnI-trnQ-trnM) among most Ditrysia mitogenomes, which is also proved as a synapomorphy for all the Ditrysia lineages. Of this rearrangement event, the tandem duplication and random loss model appears to be the most reasonable mechanism with the following events: firstly, the tRNA cluster (ancestral gene order trnI-trnQ-trnM) duplicated followed by the random deletion of the supernumerary genes including trnI, trnQ (the first copy) and trnM (the second copy). The base composition is a common genome-level character for exploring mitogenome evolution. We assessed this feature of pyraloid mitogenomes by calculating A+T content, AT-skew, and GC-skew. In general, lepidopteran mitogenomes exhibit a strong bias to A and T and the negative GC-skew. And our analyses confirmed that the base composition of pyraloid mitogenomes is similar to the typical trend of lepidopteran mitogenomes. Within the Pyraloidea, Crambidae and Pyralidae there was no significant difference in A+T content and GC-skew, while the AT-skew presented a distinct tendency. All the Pyralidae species demonstrated the negative AT-skew (\< -0.035, except for *Lista haraldusalis* -0.007), while Crambidae species showed a higher AT-skew (\> -0.0249) than that of Pyralidae. Additionally, comparative analyses of A+T content and strand asymmetry at the subfamily level revealed that A+T content and AT-/GC-skew largely overlapped, suggesting that base composition evolved under an identical pattern among subfamilies from the same family. Overall, the relatively consistent patterns of base composition for the Pyraloidea mitogenomes not only reflects similar substitution pressures but appears to result from the conserved genome organization. ## Protein-coding genes and codon usage The total size of 13 PCGs in Pyraloidea mitogenome was intermediate, ranging from 11,134 bp (*Ephestia kuehniella*) to 11,230 bp (*Chilo suppressalis*). And the average A+T content of PCGs was also similar to other moths, ranging from 75.1% (*Scirpophaga incertulas*) to 81.1% (*Paracymoriza distinctalis*), while the A+T content largely differed among three codon positions. The third codon position showed a far higher A+T content than that of the first and second codon positions, which is similar to other moths. Overall, all of these features of PCGs make a major contribution to the strong AT bias of whole genome. Most of the PCGs in Pyraloidea mitogenomes possessed canonical start codons (ATN) with the exception of COI, which used CGA as a start codon. Apart from COI in most species, there were several genes with other start codons: GTG for ND3 (*Pycnarmon lactiferalis*) and TTG for ND1 (*Galleria mellonella*). Estimated among 13 PCGs, start codons of ATP6, COIII, ND4, and ND4L were the most conserved, and in contrast, COII, ND3, and ND1 held diverse start codons. Compared with start codons, the stop codon in Pyraloidea mitogenomes was more straightforward. ND2, ATP8, ATP6, COIII, ND6, and CYTB mainly used TAA as the stop codon, and the incomplete stop codon T or TA chiefly occurred in COI, COII, ND5, and ND4. Another typical stop codon TAG scattered among COII, ND3, ND4, ND4L, and ND1. The relative synonymous codon usage of five newly sequenced species is shown in. It is obvious that the usage of codons with high A/T bias is more frequent than that with G/C bias. The most frequently used codons in Pyraloidea mitogenome was identical, namely UUU, UUA, AUU (Ile), AUA (Met), UAU (Tyr) and AAU (Asn). Missing codons were constantly presented in most Pyraloidea mitogenomes with different degrees, ranged from 1 codon (*Elophila interruptalis*, *Chilo suppressalis*) to 7 codons (*Spoladea recurvalis*, *Nomophila noctuella*, *Loxostege sticticalis*, *Cnaphalocrocis medinalis*, *Parapoynx crisonalis*, *Paracymoriza distinctalis*). Generally, these missing codons showed high G/C content. Codons that were most commonly missed were those coding for the amino acid Ser, Leu, Arg. Additionally, we evaluated the lineage-specificity of missing codons, but no remarkable signal or pattern of evolution were found. On the whole, the codon usage fully reflected the A/T-preference base composition for Pyraloidea mitogenomes. ## Transfer RNA and ribosomal RNA genes A common set of 22 tRNA genes were found in all the complete mitogenomes of Pyraloidea. These tRNA genes illustrated a consistent length and base composition among pyralids. Although most tRNAs could be folded into the typical clover-leaf structure, the exception (trnS^AGN) existed widely in Pyraloidea mitogenomes. For trnS^AGN, the dihydrouridine (DHU) stem was replaced by an unstable loop, which has been observed in many insect mitogenomes. Additionally, we identified mismatched base pairs in different stems of tRNA, a general feature for animal mitogenomes, and meanwhile these mismatched nucleotides might be modified during post-transcriptional processing. In order to explore the evolutionary pattern of tRNA in Pyraloidea, we calculated the percentage of identical nucleotides. As shown in, the Crambidae and Pyralidae show similar levels of nucleotide conservation. Comprehensive analyses combining the base composition and gene arrangement revealed that the J-strand tRNAs were more conserved than N-strand tRNAs, but the identical degree did not closely link to A+T content or absolute location of mitogenome, which has been reported in other insect lineages. Furthermore, shows that the acceptor and anticodon stem were more conserved than DHU and TψC stem, and anticodon loop also presented the highest nucleotide similarity. Among Pyraloidea mitogenomes, the average size of rrnL (\~780 bp) and rrnS (\~1361 bp) is comparable to those of other moths, however a number of unique insertion sequences were identified in both of the two rRNA genes for a few species. To confirm an accurate position of these insert fragments, we predicted the secondary structures of rrnL and rrnS. As observed in other insect rRNAs, rrnL and rrnS contain five domains (46 helices) and three domains (27 helices), respectively. According to and Figs, the insert fragments are mainly in domain II of rrnS and domain III of rrnL. The largest inserting-sequences were found in *Orybina plangonalis*, which included several short repeated sequences. In fact, these sequences were excluded from the stable stem region of secondary structure, so the inserted or deleted sequences of hypervariable regions did not significantly influence the function of rRNAs. In contrast, the conserved regions show a high similarity in both sequences and secondary structure. In rrnL, helices H579, H1925, and H2547 were the most conserved and stable, and helices H944, H984 and H1399 lacked variation in rrnS. It appears that rrnL evolved under a more conserved pattern than rrnS. ## Non-coding regions In most cases, the animal mitogenome is compact and economic. The largest non- coding region is generally considered to be the control region (CR). Even though the CR of Pyraloidea mitogenomes is small in size (\~339.8 bp average) without any large tandem repeat sequences, it remains the longest non-coding region. The CR regulates the replication and transcription of mitogenome, and many conserved blocks (CBS) were considered to play a key role in the function of CR. In insect mitogenomes, the CBS are diverse in different lineages. In lepidopteran mitogenomes, four conserved elements have been found in nearly all the Ditrysia species, i.e. ATAGA (CBS-1) followed by a large poly-T stretch, microsatellite structures (AT)n, and a poly-A stretch, though only three of them were identified in Pyraloidea mitogenomes, which lacked the poly-A stretch. However, comparative analyses of Pyraloidea CRs indentified two other CBS: A(T)TTTA (CBS-2) and ACCRT (CBS-3). The CBS-2 was located upstream of (AT)n, while the CBS-3 occurred at the 3’ end of the CR. It should be emphasized that a clear function of these conserved elements is uncertain and thus should be included in future studies. In addition to CRs, two other intergenic gaps (trnS-ND1 and tnrQ-ND2) existed in all Pyraloidea mitogenomes. The first gap (trnS-ND1) contained a conserved motif ATACTAW, which is involved in regulatory functions as the binding site of the transcription termination factor (DmTTF). Alignment of the second gap (trnQ-ND2) and ND2 gene showed the relatively high sequence similarity suggesting that this gap may be the debris of duplicate ND2 genes and this duplicate event should occur before the divergence of Ditrysia. ## Phylogenetic analyses In spite of the Pyraloidea having a large number of important pests, the molecular phylogeny of the group is still ambiguous, especially for the Crambidae. Adding mitogenomes from new subfamilies and genera provide more data to investigate the phylogenetic relationships of the Pyraloidea. The inferred phylogenetic trees based on four datasets using Bayesian inference (BI) and maximum likelihood (ML) methods showed a similar topology, which is consistent with other researches basically. Monophyly of the families (Pyralidae and Crambidae) and subfamilies is well-supported, as is suggested by the morphological characters. Comparative analyses of the trees from four datasets revealed that rRNA genes could contribute to improving the node support values, and the third codon positions of PCGs also provided phylogenetic information. Thus, the PCG123R dataset is more appropriate for reconstructing the molecular phylogeny of the Pyraloidea. The Pyralidae is a relatively robust group, which contains five subfamilies. A previous nuclear genes study confirmed the phylogenetic relationships at the subfamily level, but this differs from some morphological studies. In this study, we validated a stable molecular topology: (((Pyralinae + Epipaschiinae) + Phycitinae) + Galleriinae) with high support values. The main difference between morphological and molecular results is the phylogenetic position of the Pyralinae. The Crambidae was divided into two large lineages: PS clade (Pyraustinae and Spilomelinae) and non-PS clade (the other subfamilies), and supported by our results. The placement *S*. *incertulas* in this paper is the same as in and; both studies placed *Scirpophaga* as sister group to the Crambinae. It differs from the nuclear gene-based hypothesis, where *Scirpophaga* is either sister group to the Midilinae or *Rupela* + Acentropinae. Unlike the nuclear-gene based study. Neither Midilinae or *Rupela* were included in this study and may account for the results.This significant difference could also be explained by long-branch attraction, and could be corrected by increasing the sampling number for Schoenobiinae and by combining both the nuclear and mitogenomic data. Overall, most lineages inferred by mitogenomic data confirmed the current view of Pyraloidea phylogeny. # Supporting information The authors would like to thank Fei Ye (College of Life Sciences, Sun Yat-sen University) for collecting the specimens and Dr. David K. Cone (Department of Biology, Saint Mary’s University, Canada) for editorial comments on an early draft of the manuscript and revised the final version. This study was supported by grants from the National Natural Science Foundation of China (31372158) and the Natural Science Foundation of Shaanxi Province (2017JM3014). [^1]: The authors have declared that no competing interests exist.

# 1. Introduction Highly pathogenic avian influenzas have become a major threat to human and livestock health in the last two decades. The H5N1panzootic (2004 ongoing) has been one the most geographically widespread and costly, resulting in the loss of hundreds of millions of poultry in 68 countries and over 450 human deaths worldwide—a mortality rate of 60 percent. For H5N1, and other H5 subtypes, most countries reporting poultry outbreaks also report evidence of the disease in wild bird populations, and the mechanisms for the spread of H5N1 have been identified as a combination of wild bird transmission and the live poultry trade. In this paper we reconsider the role of the poultry trade in the spread of H5N1. The background to the study is the somewhat conflicting results of earlier studies. Using phylogenetic relationships between virus isolates at the peak of the first H5N1 panzootic, Kilpatrick et al. identified pathways for 52 introductions in Asia, Europe, and Africa. They concluded that of 21 H5N1 introductions in Asia, 9 were due to the poultry trade and 3 were due to migrating birds; that of 23 introductions in Europe none were due to the poultry trade, and 20 were due to migratory birds; and that of 8 introductions in Africa, 2 were due to the poultry trade and 3 were due to migrating birds. A later study of the same panzootic through 2006 found that except for introductions into northern and southern Vietnam, which were due to the poultry trade, all remaining introductions were best explained by wild bird migration. The authors argued that the spatial and temporal regularity of H5N1 introductions could only be explained by wild bird migration along four flyways: the East Asia flyway connecting the Far East of Russia, eastern China and Southeast Asia; the Black Sea-Mediterranean Sea flyway connecting Eastern Europe, the Arabian Peninsula and the Nile River Valley; the East Africa-West Asia flyway connecting western/central Siberia and Central Asia and Africa; and the Central Asia flyway connecting Siberia and northern China to the southern part of North Asia and Southwest Asia. There is evidence for the transmission of a number of infectious zoonotic and epizootic diseases through commercially traded animals and animal products. It has been shown that transmission risk increases with the volume of trade, and decreases with both the distance between source and sink areas and the biosecurity measures applied in both places. A number of studies have identified a strongly positive relationship between the opening of new markets and the introduction of a range of animal diseases, and between growing trade volumes and the probability that those diseases will establish and spread. In the case of H5N1, phylogenetic studies have shown significant phylogenetic clustering in Southeast Asia, consistent with the high frequency circulation among the countries of that region resulting from the growth of intra-regional trade. The biosecurity measures applied in international trade are, in principle, regulated by the Agreement on the Application of Sanitary and Phytosanitary Measures (SPS Agreement) and the standard-setting bodies supporting that agreement (the Codex Alimentarius Commission for food products, the International Plant Protection Convention for plant products, and the World Organization for Animal Health (OIE) for animal products). Since these are not enforcement bodies, however, biosecurity standards are, in practice, regulated by bilateral and multilateral regional trade agreements. In this study, we consider transmission mechanisms in the three regions of the world where H5N1 panzootics have been most intense (as indicated by numbers of outbreaks): Europe, West Africa, and Southeast Asia. These regions are associated with three regional trade blocs: the European Union (EU), the Economic Community of West Africa (ECOWAS), and the Association of Southeast Asian Nations (ASEAN). These regions are significantly different from one another in socio-economic conditions, and particularly in the biosecurity measures applied to intra-regional trade. Using data on the 12,944 H5N1 outbreaks occurring in these regions between 2004 and 2016, we estimate disease risk as a function of a number of risk factors, including live poultry trade volumes and various measures of trade biosecurity. # 2. Materials and methods We follow Kilpatrick, Chmura (4) in considering the relative risks posed by migratory birds and trade in the spread of H5N1, but with an additional decade of data on both outbreaks and risk factors. Our primary interest is in the role of live poultry imports as a source of trade-related avian influenza risk at the regional level. We note that other poultry products, such as packaged meat and eggs, do pose a risk, but it is significantly lower. Although avian influenza can persist in frozen meat, contact with that meat is unlikely to cause infection. Furthermore, since HPAIs are lethal to egg embryos, eggs are not a potential source of transmission. The data comprise an unbalanced panel covering 53 countries over 13 years; the lack of balance is due to the fact that membership of the EU changed over the timeframe. The response variable in all models estimated was a log transformation of the number of H5N1 poultry outbreaks in a given country in a given year, obtained from the Emergency Prevention System for Animal Health (EMPRES), a joint project of the FAO and OIE. The log transformation was applied to account for the wide disparities in the numbers of the outbreaks across countries. In 2010, for example, Indonesia recorded 1206 outbreaks while Romania, the only EU country to be infected that year, had only 2. In addition to reflecting the differing directions and intensities of risk factors, this also reflects differences in reporting conventions for H5N1 at the international level. A series of outbreaks may be reported separately in one country, but be treated as a single event in another. Data on trade in live poultry were obtained from the United Nations’ Comtrade Database ([comtrade.un.org](http://www.comtrade.un.org/)) and resourcetrade.earth, a project of the Royal Institute of International Affairs ([www.chathamhouse.org](http://www.chathamhouse.org/)). These report the total imports of live poultry into a given country in a given year by weight (kg). The data on trade in live poultry did not distinguish between different types of domestic birds, such as chickens, duck, and geese, but grouped them under a single commodity category of “live poultry.” With respect to wild bird migration as a pathway for H5N1 spread, we used the density of wild bird habitat as a proxy for the presence and scale of migratory bird populations, and the likelihood that wild and domestic birds will mix. Lakes, wetlands, and (irrigated) agricultural areas have been consistently identified as wintering and breeding grounds for migratory birds, and as places where wild birds may come into contact with free-ranging poultry. The indicator for wild bird habitat used in this study was the set of “Important Bird and Biodiversity Areas” (IBAs) for “migratory and congregatory waterbirds” identified by BirdLife International ([datazone.birdlife.org](http://www.datazone.birdlife.org/)). In their 2006 analysis of H5N1 spread, Kilpatrick, Chmura (4) also identified IBAs as a proxy for migratory birds and the infection risks they pose. Country-level statistics on socioeconomic and agro-ecological conditions were taken from the United Nations’ Food and Agriculture Organization ([www.fao.org/faostat/en/](http://www.fao.org/faostat/en/)) and the World Bank ([data.worldbank.org](http://www.data.worldbank.org/)). Agricultural land cover was reported as a percentage of total land area of the country. Per-capita GDP was reported in purchasing power parity terms as current international dollars. Data for 2016 for these two variables were missing for certain countries. In these cases, the gaps were filled by extrapolating the missing data as a linear trend of the preceding 11 years. We assume that agricultural land–where free- ranging chickens, ducks, and geese are commonly raised in all three regions–also acts as a relevant proxy for susceptible poultry. Data on the biosecurity measures targeting avian influenza undertaken by each country were obtained from the World Organisation for Animal Health (OIE) ([www.oie.int](http://www.oie.int/)). These report a standardized series of biosecurity controls targeting wildlife and livestock diseases, including those related to surveillance, vaccination, border checks, and management of wild disease reservoirs, and whether or not a given country undertook them in a given year. We chose a subset of these biosecurity measures we considered most relevant to H5N1 avian influenza risks for inclusion in our model. Additionally, in any given year, there were 1 to 4 countries that did not provide a report of biosecurity measures to the OIE; we assumed that this indicates an absence of action, and the dataset records these cases as zeroes. Our modeling approach relied on generalized linear models (GLM) to analyze a panel of data on disease outbreaks and associated risk factors. In this we follow others who have sought to predict the spread of H5N1 at both national and international levels or H7N9. GLMs are well suited to epidemiological studies because of their flexibility regarding data type and the distribution of response variables, their simplicity of application, and their frequency of use. Our identification strategy involved the selection of three specifications for each of two estimators. We adopted both random and fixed effects estimators. Hausman tests conducted at the all-regions level favored a random effects estimator, as the *p-*value exceeded the 5% threshold below which fixed-effects regression is conventionally considered necessary. Some factors that influence the likelihood and number of outbreaks in a given country or region are not likely to change significantly over the course of several years, or even a decade. In our dataset, for example, the amount of land covered by wild bird habitat is time-variant, while agricultural land and even per-capita GDP for many countries experienced relatively modest variations over the timeframe of the study. In this case, and as the Hausman diagnostics indicate, a random effects estimator is more appropriate. Nevertheless, since we wished to control for time-invariant characteristics of regions and countries we also implemented fixed effects estimators at both the aggregate and trading bloc levels, implicitly assuming no changes in the trade or biosecurity environment at the bloc level that we are unable to control for. Our first specification (Model 1) included a number of factors related to disease risk but excluded both live poultry imports and biosecurity measures. Included predictors were land area, human population, per-capita GDP in purchasing power terms, agricultural area, wild bird habitat area, and the live chicken population. Our second specification (Model 2) added intra-regional trade bloc and extra-bloc imports of live poultry. Our third specification (Model 3) added four main biosecurity measures: border precautions, general surveillance, vaccination prohibition, and wild disease reservoir management. All are categories of OIE-reported biosecurity measures taken against avian influenza. The general forms of the estimated random and fixed effects models were: $$y_{it} = \beta_{0} + \sum\limits_{j = 1}^{6}X_{jit}\beta_{j} + \sum\limits_{k = 1}^{2}Z_{kit}\beta_{k} + \sum\limits_{s = 1}^{4}U_{sit}\beta_{s} + \delta_{1}ecowas + \delta_{2}asean + u_{i} + \varepsilon_{it},$$ $$y_{it} = \beta_{0} + \sum\limits_{j = 1}^{6}X_{jit}\beta_{j} + \sum\limits_{k = 1}^{2}Z_{kit}\beta_{k} + \sum\limits_{s = 1}^{4}U_{sit}\beta_{s} + u_{i},$$ where *y*_*it* denotes the number of poultry outbreaks in country *i* in year *t*, *X* includes the predictors for Model 1, *Z* includes the additional predictors for Model 2, *U* includes the additional predictors for Model 3, *ecowas and asean* are dummy variables for the two titular regional trade blocs (the EU is the reference group), and *u*_*it* and *ε_it* are the “between” and “within” errors respectively. To account for heteroskedasticity, we used robust standard errors. Finally, since the data used in this analysis are reported annually, and H5N1 has been a conspicuous and fast-moving epidemic (meaning the effects of an outbreak are unlikely to persist over a long period of time) among poultry, we did not use a lag structure in our statistical analysis. Therefore, we assumed that the factors driving an outbreak in a given year are contemporaneous with it (e.g., an outbreak that occurred in 2012 were modelled using trade volumes from 2012). We were also constrained by data availability in our use of annual increments: although monthly data exist for outbreaks, they do not for important predictor variables such as per-capita GDP, human and poultry populations, the volume of live poultry traded, and biosecurity. # 3. Results Regressions results from all models, including both random and fixed effects, are reported in Tables. At the all-regions level, the results for the random- and fixed-effects models were very similar, with the same set of predictor variables being statistically significant (i.e., *p-*values below the 5% or 10%) and the same direction of impact on the response variable. This set of predictors was human population (positive direction), per-capita GDP (negative direction), intra-trade bloc live poultry imports (negative direction), extra- trade bloc live poultry imports (positive direction), and the biosecurity measure of surveillance (negative direction). Additionally, although the coefficient values for the same predictor differed between the two estimators, all pairs were within the same order of magnitude. The only exception to this was migratory waterbird habitat variable—the percent of land area covered by IBAs for migratory and congregatory waterbirds. This was statistically significant and negative (i.e., had a mitigating impact on H5N1 poultry outbreaks) for the fixed-effects model but was not significant for the random-effects model. The overall R-squared for the random-effects model was significantly higher than that for the fixed-effects model (0.451 vs. 0.0181). The “between R-squared” value was particularly high (0.682) in the random effects model, signaling the importance of variation among countries (as opposed to “within R-squared,” which measures the variation within countries over time). As we had expected, we found significant differences across trade regions. In the random-effects model, ECOWAS diverged from all-regions conditions and from ASEAN with respect to per-capita GDP and extra-bloc imports: while the two predictors were, respectively, risk-decreasing and risk-increasing at the all- regions level and in ASEAN, they had the opposite impacts in ECOWAS. Furthermore, ECOWAS differed from the all-regions level and from the EU in terms of intra-bloc imports: while this was risk-decreasing for the former two, it was risk-increasing for ECOWAS. Finally, there were predictors that were statistically insignificant at the all- regions level but had a significant effect within different regions. For ASEAN, agricultural land cover was a mitigating factor for outbreaks while wild disease reservoir management showed a strong positive relation with outbreaks. For ECOWAS, wild waterbird habitats and border precautions had a mitigating effect on outbreaks while vaccination prohibition and wild reservoir management had a positive effect. In the EU, the population of live chickens had a strong negative relation with outbreaks, while vaccination prohibition, similar to the case with ECOWAS, was positively related. # 4. Discussion Following Liang, Xu (5), there is a perception that the long distance transmission of highly pathogenic avian influenza H5N1 was largely due to wild bird migration, with the live poultry trade playing a minor and more localized role in some cases. Our concern here has been to identify the nature of the risk posed by the live poultry trade in different regions of the world, and the conditions affecting that risk. Our measure of development status, per-capita GDP, is simultaneously a proxy for modernization, biosecurity, consumption, and value-at-risk. As a proxy for modernization, it reflects risk-reducing differences in production methods. Industrial livestock production methods typically include on-farm biosecurity measures that protect poultry from contact with disease-carrying wild birds. Unlike traditional methods of free-range or “backyard” husbandry, factory production minimizes the likelihood of poultry intermingling with wild birds or being exposed to environmental pathogen pollution. For all its epidemiological, ecological, and ethical problems, industrial livestock production allows for more timely and widespread disease surveillance and vaccination, and for greater compliance with animal health regulations. At the same time, per-capita GDP growth is also associated with risk-increasing changes in meat consumption, and hence poultry production. Indeed, the highest income elasticity of demand for meat and fish has been found in the poorest households and the poorest countries. In developing countries, 71% of the additions to meat consumption are from pork and poultry, with poultry dominating pork. Absent changes in on-farm biosecurity, increased production implies increased risk. Across all regions, the net effect of income growth is to reduce risk, dominating risk-increasing changes. In the ECOWAS region—the lowest income region—the effect is the opposite. The risk-increasing effects of income growth dominate the risk reducing effects. Amongst the landscape variables—land area, the proportion in agriculture, and the proportion in IBAs—our results reveal no uniform relation to H5N1 outbreaks. At the all-regions level we found a weakly negative relation between outbreaks and the proportion of the land area in IBAs. This was driven by the European Union, which includes the highest proportion of land area in IBAs, but also the most industrialized forms of poultry production. The degree to which poultry production is industrialized also shows up in the coefficients on poultry numbers, which are negative and significant only for the EU. While spatial heterogeneity at the landscape scale is important in terms of avian ecology, we were unable to take explicit account of these more detailed considerations in a country-scale analysis. The impacts of regional differences in biophysical conditions that are not directly controlled for are, however, included in bloc- level fixed effects. Our primary concern is with the role of the live poultry trade, and how that differs between regions. Across all regions we find that live poultry imports into a trade bloc are risk increasing. This is consistent with past studies that have shown that extra-bloc live poultry imports may be a significant source of additional avian influenza risk where they do not meet bloc sanitary and phytosanitary standards. The EU’s common market and the ASEAN free trade regime in particular have long-standing and standardized protocols, in accordance with the World Trade Organization’s Agreement on the Application of Sanitary and Phytosanitary Measures. But the two blocs have quite different exposures to external risk. A study of highly pathogenic avian influenza introductions to Vietnam, for example, found that extra-ASEAN imports of live poultry increased the risk of introduction. This is also what our study finds for the ASEAN region. We do not see an equivalent effect for the EU, reflecting differences in both import volumes and the biosecurity measures applied to imports. The EU imports less and applies stricter biosecurity measures to those imports. The ECOWAS story is different. Extra-bloc live poultry imports are risk reducing, not risk increasing. It is likely that imports from outside the bloc reduce avian influenza risk in the region in part because they meet biosecurity standards that are more stringent than the standards applied in the region. The effects of intra-bloc trade in live poultry mirror the effects of extra-bloc trade. In the EU and ASEAN, intra-bloc trade is risk reducing (Tables). This may reflect a “substitution effect” in which imports of safer intra-bloc poultry crowds out riskier extra-bloc imports. Other studies have come to similar conclusions. EU-derived live poultry imports to Spain, for example, were found to pose no threat of avian influenza introduction. Once again, ECOWAS is the exception. Extra-ECOWAS imports of live poultry are risk reducing while intra- bloc imports are risk increasing. This is likely due to poor internal biosecurity, such as lax standards and inconsistent execution of inspections. Regulatory standards within the ECOWAS trade bloc have been weak for the whole of the study period. While harmonized sanitary and phytosanitary standards for the 15 member states of ECOWAS were in principle adopted in 2010, most ECOWAS states had yet to submit legislation for international certification by 2017. Failure to adopt and enforce unified standards may be partly due to income constraints in ECOWAS countries. In PPP terms, the bloc’s per-capita GDP in 2016 was less than half that of ASEAN and approximately 1/8^th that of the EU, meaning it had less resources available for biosecurity policies and institutions. Political instability may be another important obstacle: a number of ECOWAS member states, including Nigeria, Niger, Sierra Leone, Mali, Liberia, and Cote d’Ivoire have suffered from civil wars and armed insurgencies over the past two decades. Such fraught geopolitical conditions are not conducive to the establishment and enforcement of cross-border regulations. It goes without saying, though, that certification of sanitary and phytosanitary legislation in ECOWAS states, and the establishment of enforcement agencies to bring states into compliance with the SPS Agreement and Codex Alimentarius is a necessary condition of improving regional trade-related biosecurity. In terms of biosecurity measures more specifically, we did not have direct measures of on-farm biosecurity (but conjecture that biosecurity is increasing in per-capita GDP), but we did have measures of four biosecurity policies at the national level. These include: (1) border precautions (measures applied at airports, ports, railway stations or road check-points open to international movement of animal, animal products and other related commodities, where import inspections are performed to prevent the introduction of the disease, infection or infestation); (2) general surveillance (surveillance not targeted at a specific disease, infection or infestation); (3) prohibition of vaccination (prohibition of the use of a vaccine to control or prevent the infection or infestation); and (4) management of wildlife reservoirs (measures to reduce the potential for wildlife to transmit the disease to domestic animals and human beings). The management of wild disease reservoirs differs widely across countries, but techniques include vaccination, treatment of infections with drugs, isolation of infected populations, population translocation, reproduction reduction, culling, and control (draining, flooding, or burning) of wild disease reservoir habitat. Of these measures, only general surveillance was significant at the all-regions level, while at the bloc level the effects of the different measures were frequently ambiguous. In the EU, for example, only the prohibition of vaccination was significant, and then in positive relation to outbreaks. For poultry, vaccination may be prohibited because the practice makes it difficult to distinguish infected from vaccinated flocks. This makes it a concomitant of policies centered on livestock culling as the primary response to outbreak risk. No other biosecurity policy was found to have a statistically significant relation to outbreaks in the region. The same set of policies had opposite effects in ASEAN and ECOWAS. The prohibition of vaccination and the management of wild reservoirs were positively related to outbreaks in ECOWAS but negatively related to outbreaks in ASEAN, while border protection measures were negatively related to outbreaks in ECOWAS but positively related to outbreaks in ASEAN. This may reflect regional disparities in the quality of implementation not captured in the data. But it may also reflect the greater importance of trade in the transmission of the disease in ECOWAS. In their survey of the international spread of H5N1 in the early years of the global epidemic, Kilpatrick, Chmura (4) found that transmission into Europe was by wild birds, that transmission into Southeast Asia was by the poultry trade, and transmission into Africa by a balance of both. Our results suggest that after introduction, inter-country spread had differing dynamics in each region. While intra-bloc trade facilitated H5N1 spread among West African countries, it did not in either Europe or Southeast Asia. In these areas, greater risk was posed by out-of-region live poultry imports. # 5. Conclusion In recent decades, avian influenzas have emerged as a major threat to human and animal health across the world. In particular, HPAI H5N1, which was first isolated in 1996, has been the most widespread and among the most devastating in terms of livestock and human mortality. It has inflicted severe losses to poultry stocks and caused hundreds of human deaths. Even today, as other avian influenzas have become epidemic, H5N1 remains in circulation among wildlife and livestock. Identifying and quantifying the mechanisms of its international spread can help lay the groundwork for prediction and mitigation. It may also provide an instructive framework for the management of other avian influenzas. In this study, we considered the risk posed by the international trade in live poultry and the effects of associated biosecurity measures. Differing agro- ecological and socioeconomic conditions across the trade regions were shown to influence epidemic dynamics in different ways, with certain factors being risk- enhancing or risk-decreasing in one region but having the opposite effect, or no significant effect, in another. In policy terms, there is no one-size-fits-all solution to mitigating avian influenza spread. The particular conditions, including those related to the trade agreements and associated regulatory standards, of a given region need to be carefully considered. But overall, biosecurity measures are potentially effective at controlling H5N1 risks, and should be undertaken as a means to forestall spread–in general, mitigation of epidemics is significantly more cost-efficient than suppression. On-farm and other forms of domestic biosecurity may be more important than trade-related measures, but where the protection of trade pathways is weak, the risk of avian influenza spread is clearly higher. # Supporting information We would like to thank Ann Kinzig, Jim Collins, Ben Minteer, and Peter Daszak for their insightful comments and discussions on the research presented here. [^1]: The authors have declared that no competing interests exist.

# Introduction Bioluminescence is well established as a highly sensitive technology for probing biological systems. In particular, firefly luciferases and their substrate, D-luciferin (LH2), have been used for a broad range of biological applications in life science research for the past 4 decades. Unlike fluorescence, which requires an external light source, bioluminescence relies on light-emitting chemical reactions catalyzed by luciferase enzymes. The absence of photon production by the enzyme or substrate themselves results in very low intrinsic background signal enabling a broad range of assay formats and applications with exceptional sensitivity and dynamic range. Recently, the increasing commercial demand for cost-efficient and sensitive detection of chemical or biological contaminants has driven interest in new formats of luciferin-based bioluminescence assays due to their utility, simplicity, and versatility. Within commercial settings, ATP quantification assays using bioluminescence have been widely recognized as a fast and cost-effective means for sensitive detection of live microbes from different sample types (e.g., water, crude oil, food). These assays minimally comprise firefly luciferase and LH2 in an optimized buffer which, upon the addition of ATP present in samples containing live cells, have all the necessary components for rapid generation of a bright, stable, bioluminescence output signal. To date, a major challenge in formulating ATP detection systems for use beyond laboratory settings has been the differing stabilities of enzymes and substrates, resulting in a requirement for separate compositions that are reconstituted and combined at the time of testing. While this strategy helps mitigate stability losses in the system as components with lower stability can be lyophilized or refrigerated, it also increases the complexity of the process and can introduce variability to the assay. In order expand the utility of ATP detection assays to more industrial settings where simplicity and cost-effectiveness are critical, there is a need for both highly thermostable luciferase enzymes and luciferin substrates that can be formulated together in a homogenous, liquid ATP detection system that would enable long- term storage and handling at ambient temperatures without loss of performance. Progress in improving the thermostability of the firefly ATP detection system up to now has primarily been achieved through formulation studies and protein engineering to increase the stability of the luciferase enzyme component. Notably, a thermostable firefly luciferase mutant, trademarked as Ultra-Glo™, has been developed via directed evolution to withstand variations in assay conditions. These include elevated temperature and high concentrations of ionic detergents and reducing agents. Together, these properties enable robust performance under reaction conditions that allow simultaneous cell lysis and bioluminescent detection that could not be achieved with the native luciferase enzyme. The relative stability of luciferin substrates in solution phase, on the other hand, has remained a major challenge to the performance of firefly-based ATP detection systems. This is because ambient storage of LH2 in solution for extended time periods leads to its irreversible decomposition into dehydroluciferin (L), which results in significant loss of ATP detection system performance. Mechanistically, dehydroluciferin is a potent inhibitor of firefly luciferases and its accumulation in reagent medium, even at low concentration, compromises assay performance over time. While lyophilization and temperature control strategies have been successful in mitigating formation of dehydroluciferin for laboratory-based applications, they are not able to completely eliminate its accumulation under ambient storage in the homogeneous liquid formats that are desirable for industrial settings. Here we report a general strategy for improving the solution thermostability of luciferin through substitution of alkyl groups at the 5-position, which prevents oxidative decomposition into dehydroluciferin. By preparing and evaluating a focused library of 5,5-dialkylluciferins, we identified analogs with significantly enhanced thermostability over LH2. Though they were significantly more stable, the analogs are poor substrates for firefly luciferases and require enzymes specifically engineered to utilize them efficiently. To accomplish this, we used directed evolution to engineer variants of Ultra-Glo™ luciferase that could efficiently utilize 5,5-dialkylluciferins as bioluminescence-generating substrates. The results of our studies establish a novel luciferase/luciferin pair that serves as a foundation for next-generation ATP detection systems with high performance and improved reagent stability for industrial applications. # Materials and methods ## General materials Reactions were performed using commercially obtained solvents. Unless otherwise stated, all commercially obtained reagents were used as received. Flash column chromatography was performed using pre-packaged RediSep®Rf columns on a CombiFlash Rf system (Teledyne ISCO Inc.). ¹H and ¹³C NMR spectra were recorded on a Bruker Avance III HD 400 (at 400 MHz and 100 MHz respectively) and are reported relative to internal DMSO-d6 (¹H, δ = 2.50), CD₃OD (¹H, δ = 3.31) and DMSO-d6 (13C, δ = 39.5), CD₃OD (13C, δ = 49.0). Data for ¹H NMR spectra are reported as follows: chemical shift (δ ppm) (multiplicity, coupling constant (Hz), integration). Multiplicity and qualifier abbreviations are as follows: s = singlet, d = doublet, t = triplet, q = quartet, m = multiplet, br = broad, app = apparent. High resolution mass spectra were obtained on AB Sciex TripleTOF 5600+. Preparative HPLC was performed with Waters 2535 Quaternary Gradient Module utilizing XBridge PREP C18 Column 5 μm (30 mm X 250 mm). Analytical HPLC was performed with an Agilent 1100 Series HPLC utilizing Phenomenex Synergi™ 2.5 μm MAX-RP 100 Å columns (4.6 mm x 50 mm). 2-cyano-6-hydroxy-benzothiazole (CBT) derivatives (I) were commercially available from Promega Biosciences. β,β-Disubstituted cysteine analogs (II a-h) were either commercially available (II-a from Alfa Aesar) or synthesized via protocols described previously. Ultra- Glo™ luciferase was obtained from Promega Corp. ## Synthesis of 5,5-dialkyllucierins General procedure: Hydrochloric salts of β,β-disubstituted cysteine analogs (II, 0.15 mmol, 1.5 equiv) dissolved in H₂O (1 mL) under N₂ were neutralized with 1 N NaOH (aq, 0.30 mmol, 300 μL, 3.0 equiv). To a solution of 2-cyano-6-hydroxy-benzothiazole (I, 0.1 mmol, 1.0 equiv) in DMF (2 mL) at RT under N₂, was added the neutralized solution of β,β-disubstituted cysteine analogs. The solution was then stirred under N₂ for 30 min. LC-MS indicated complete consumption of benzothiazole starting material (I). Luciferins (III) were obtained via preparative HPLC (mobile phase A: 10 mM NH₄OAc aqueous solution; mobile phase B: CH₃CN; gradient condition: 5% B to 95% B over 30 minutes). ### 2-(6-Hydroxybenzo\[d\]thiazol-2-yl)-5,5-dimethyl-4,5-dihydrothiazole-4-carboxylic acid (III-a) The product was isolated as white amorphous solid. ¹H NMR (400 MHz, DMSO-*d6*) δ 7.90 (dd, *J* = 9.0, 0.8 Hz, 1H), 7.41 (s, 1H), 7.13 (dd, *J* = 9.0, 0.8 Hz, 1H), 4.79 (s, 1H), 1.60 (s, 3H), 1.47 (s, 3H).The data is consistent with reported values. ### 5-(6-Hydroxybenzo\[d\]thiazol-2-yl)-4-thia-6-azaspiro\[2.4\]hept-5-ene-7-carboxylic acid (III-b) The product was isolated as off white amorphous solid. ¹H NMR (400 MHz, DMSO-*d*₆) δ 13.06 (br s, 1H), 10.23 (br s, 1H), 7.95 (d, *J* = 9.0 Hz, 1H), 7.46 (s, 1H), 7.06 (d, *J* = 9.0 Hz, 1H), 4.98 (s, 1H), 1.42–0.87 (m, 4H); ¹³C NMR (100 MHz, DMSO-*d*₆) δ 169.2, 165.8, 157.4, 157.2, 146.2, 137.1, 124.9, 117.2, 106.9, 81.7, 40.2, 40.2, 39.9, 39.7, 39.5, 39.3, 39.1, 38.9, 33.7, 17.1, 8.0; HRMS (ESI+) calc’d for C₁₃H₁₁N₂O₃S₂⁺ \[M+H\]⁺ 307.0206, found 307.0209. ### 6-(6-Hydroxybenzo\[d\]thiazol-2-yl)-5-thia-7-azaspiro\[3.4\]oct-6-ene-8-carboxylic acid (III-c) The product was isolated as off white amorphous solid. ¹H NMR (400 MHz, DMSO-*d*₆) δ 10.21 (s, 1H), 7.94 (d, *J* = 9.0 Hz, 1H), 7.44 (d, *J* = 2.4 Hz, 1H), 7.06 (dd, *J* = 9.0, 2.4 Hz, 1H), 5.25 (s, 1H), 2.60–2.30 (m, 4H), 2.05–1.73 (m, 2H); ¹³C NMR (101 MHz, DMSO) δ 169.9, 165.0, 162.8, 157.9, 157.6, 146.7, 137.6, 125.3, 117.6, 107.3, 85.9, 63.1, 36.3, 32.8, 31.3, 17.1; HRMS (ESI+) calc’d for C₁₄H₁₃N₂O₃S₂⁺ \[M+H\]⁺ 321.0362, found 321.0366. ### 2-(6-Hydroxybenzo\[d\]thiazol-2-yl)-1-thia-3-azaspiro\[4.4\]non-2-ene-4-carboxylic acid (III-d) The product was isolated as white amorphous powder. ¹H NMR (400 MHz, DMSO-*d*₆) δ 10.21 (s, 1H), 7.92 (dd, *J* = 8.8, 1.5 Hz, 1H), 7.44 (s, 1H), 7.05 (d, *J* = 8.8, 1.5 Hz, 1H), 5.15 (s, 1H), 2.14 (d, *J* = 6.9 Hz, 2H), 1.93 (t, *J* = 6.4 Hz, 2H), 1.85–1.63 (m, 4H); ¹³C NMR (100 MHz, DMSO-d6) δ 170.1, 164.9, 157.76, 157.79, 146.7, 137.6, 125.3, 117.6, 107.3, 83.8, 71.4, 41.3, 36.5, 24.2, 23.4; HRMS (ESI+) calc’d for C₁₅H₁₅N₂O₃S₂⁺ \[M+H\]⁺ 335.0519 found 335.0526. ### 2-(6-Hydroxybenzo\[d\]thiazol-2-yl)-1-thia-3-azaspiro\[4.5\]dec-2-ene-4-carboxylic acid (III-e) The product was isolated as white amorphous powder. ¹H NMR (400 MHz, DMSO-*d*₆) δ 10.22 (brs, 1H), 7.93 (d, *J* = 8.9 Hz, 1H), 7.44 (d, *J* = 2.4 Hz, 1H), 7.06 (dd, *J* = 8.9, 2.4 Hz, 1H), 4.92 (s, 1H), 2.10–1.49 (m, 8H), 1.46–1.15 (m, 2H); ¹³C NMR (100 MHz, DMSO-*d*₆) δ 169.9, 164.0, 157.8 (2C), 146.7, 137.6, 125.3, 117.6, 107.3, 86.3, 68.1, 38.9, 34.4, 25.8, 25.3, 24.9; HRMS (ESI+) calc’d for C₁₆H₁₇N₂O₃S₂⁺ \[M+H\]⁺ 349.0675 found 349.0677. ### 5,5-Diethyl-2-(6-hydroxybenzo\[d\]thiazol-2-yl)-4,5-dihydrothiazole-4-carboxylic acid (III-f) The product was isolated as off white amorphous powder. ¹H NMR (400 MHz, DMSO-*d*₆) δ 7.94 (d, *J* = 8.9 Hz, 1H), 7.44 (d, *J* = 2.4 Hz, 1H), 7.06 (dd, *J* = 8.9, 2.4 Hz, 1H), 5.04 (s, 1H), 2.09–1.94 (m, 1H), 1.90 (dt, *J* = 14.5, 7.4 Hz, 1H), 1.83–1.70 (m, 2H), 1.05–0.85 (m, 6H). ¹³C NMR (101 MHz, DMSO) δ 170.2, 163.6, 157.9, 157.7, 146.7, 137.6, 125.3, 117.6, 107.3, 83.0, 70.7, 31.2, 29.7, 10.8, 10.6. HRMS (ESI+) calc’d for C₁₅H₁₇N₂O₃S₂⁺ \[M+H\]⁺ 337.0675, found 337.0679. ### 2-(6-Hydroxybenzo\[d\]thiazol-2-yl)-8-oxa-1-thia-3-azaspiro\[4.5\]dec-2-ene-4-carboxylic acid (III-g) The product was isolated as white amorphous powder. ¹H NMR (400 MHz, DMSO-*d6*) δ 7.90 (d, *J* = 9.1 Hz, 1H), 7.40 (s, 1H), 7.09 (d, *J* = 9.1 Hz, 1H), 4.96 (s, 1H), 4.15–3.85 (m, 2H), 3.62–3.42 (m, 2H), 2.50–2.20 (m, 2H), 2.06–1.92 (m, 2H); ¹³C NMR (100 MHz, DMSO-*d*₆) δ 171.4, 166.3, 159.1, 158.8, 148.1, 139.2, 125.9, 118.3, 107.4, 87.1, 68.4, 67.3, 65.8, 39.5, 35.9; HRMS (ESI+) calc’d for C₁₅H₁₅N₂O₄S₂⁺ \[M+H\]⁺ 351.0468 found 351.0472. ### 2-(6-Hydroxybenzo\[d\]thiazol-2-yl)-8-methyl-1-thia-3,8-diazaspiro\[4.5\]dec-2-ene-4-carboxy-lic acid (III-h) The product was isolated as off white amorphous powder. ¹H NMR (400 MHz, DMF-*d*₇ with 3 drops of 0.1 N NaOH aqueous solution) δ 7.93 (d, *J* = 8.9 Hz, 1H), 7.48 (s, 1H), 7.09 (d, *J* = 8.9 Hz, 1H), 4.79 (s, 1H), 3.05–3.00 (m, 2H), 2.37–2.14 (m, 7H), 2.01–1.95 (m, 2H). ¹³C NMR (100 MHz, DMF-*d*₇ with 3 drops of 0.1 N NaOH aqueous solution) δ 172.0, 161.8, 159.2, 157.4, 146.8, 137.7, 124.9, 117.3, 106.9, 89.8, 65.5, 54.4, 45.1, 38.7, 34.0, 33.5. HRMS (ESI+) calc’d for C₁₅H₁₅N₂O₃S₂⁺ \[M+H\]⁺ 335.0519 found 335.0526. ## HPLC-based thermostability profiling Representative thermal stability profiling of luciferins: Luciferin stock solutions (\[LH2\]_final = 1.0 mM) in Bright-Glo™ assay buffer, with or without Ultra-Glo™ luciferase (\[enzyme\]_final = 0.1 mg/mL; Promega) were incubated at 37 or 60°C. Aliquots (20 μL) were taken out at various time points, diluted with H₂O (180 μL), and analyzed by RP- HPLC. The percentages of the components were calculated based on UV absorbance at 325 nm. ## Structural modeling of Ultra-Glo™ luciferase with 5,5-dialkyluciferins A homology model for Ultra-Glo™ luciferase was created from structural templates of firefly luciferases in closed conformation from *Luciola cruciata* (PDB ID: 2D1S) and *Photinus pyralis* (PDB ID: 4G36, chain-B) using Discovery Studio software version 19.1.0.18287 (Dassault Systèmes Biovia Corp). Luciferin analog **III-a** (D and L forms, trans isomers only) were superimposed in the active site of the models based on the positioning of the high-energy intermediate analogue, 5′-*O*-\[*N*-(dehydroluciferyl)-sulfamoyl\]adenosine (DLSA), present in 2D1S. ## Plasmid construction Plasmids for expression of Ultra-Glo™ luciferase and mutants were constructed using one-step Gibson assembly kit (SGI) to introduce genes into pF1A plasmid (Promega), resulting in the addition of a C-terminal 8xHis tag to the expressed protein. Constructs were transformed into *Escherichia coli* (*E*. *coli*) KRX chemical competent cells using manufacturers protocols (Promega). Following transformation, individual colony isolates were picked for purification of plasmid DNA using a Wizard SV Miniprep Kit (Promega) and the identity of the cloning region verified by DNA sequencing. All oligonucleotides were purchased from IDT. ## Mutagenesis and library preparation Individual point mutations were introduced into the Ultra-Glo™ luciferase gene on plasmids first by PCR amplification of the gene with primers encoding the mutation(s) at the site of interest. Amplification products were purified directly from the PCR using a ReliaPrep PCR cleanup kit (Promega). The purified amplicon was then used as a megaprimer to introduce the mutation into the desired plasmid using PCR with Phusion DNA Polymerase (NEB) by combining 20 ng megaprimer with 10 ng unmodified pF1A:Ultra-Glo™. Megaprimer PCR products were used to directly transform *E*. *coli* KRX competent cells, following by plasmid isolation and DNA sequence verification of the target gene. Site-saturation libraries were constructed similar to point mutations except that codons at the mutated site were generated using primers encoded a randomized NNK codon. Error-prone PCR libraries were created using the Diversify PCR Random Mutagenesis Kit (TaKaRa), following the manufacturer protocol. Randomly mutated PCR amplicons were purified and subcloned into pF1A using Gibson assembly and transformed into *E*. *coli* KRX competent cells. At least ten colonies for each library were selected for DNA sequencing to determine mutation frequency and confirm successful cloning. Libraries with 1–6 mutations/gene were selected for directed evolution screening. ## Directed evolution screening of Ultra-Glo™ luciferase mutants in bacterial lysates Automated picking of at least 5,000 *E*. *coli* colonies containing individual Ultra-Glo™ luciferase variants per round of screening was performed using a QPix 400 microbial colony picker (Molecular Devices) and used to inoculate LB media plus antibiotic in 96-well plates. Plates were incubated for 16 hours with 400 rpm shaking at 37 ⁰C to allow for outgrowth of the *E*. *coli* culture. A 1:20 dilution of overnight cultures was performed into autoinduction media (LB plus antibiotics, 0.15% Glucose, and 0.1% Rhamnose) in 96-well plates and incubated with 400 rpm shaking at 37 ⁰C for 16 hours to induce protein expression. Cell lysis and luminescence assays were performed in single reactions by mixing induced cultures 1:1 with Detection Reagent Buffer (Promega) containing 1 mM luciferin substrate and 1 mM ATP in a 100 μL total volume. Reactions were incubated for 3 minutes and luminescence measured using a Clariostar plate reader (BMG Biotech). Screening for enzyme variants that improved thermostability was performed by first lysing cells in Detection Reagent Buffer and incubating at 50 ⁰C for 10 min prior to the addition of 1 mM substrate and 1 mM ATP and subsequently measuring the remaining luminescence activity in the culture after 3 minutes. The library size of unique amino acid sequences resulting from introduction of single point mutations in Ultra-Glo™ by error- prone PCR, assuming retention of the start codon and absence of premature stop codons, is 3,184 mutations. By screening at least 5,000 individual clones during each round of directed evolution screening we ensured a minimum of 1.5-fold coverage of all single point mutations in the library. Depending on the selection criteria, this screening coverage typically resulted in 1–2% of clones being identified as hits which were then retested for confirmation and moved into more detailed biochemical characterization experiments as purified proteins. ## Protein purification Ultra-Glo™ luciferase mutants were purified from 1.5 ml LB autoinduction cultures of *E*. *coli* KRX cells after overnight incubation at 37 ⁰C and 250 rpm shaking. Cells were lysed and His-tagged protein was purified away from cell debris using HisLink Protein Purification Kit (Promega) following the manufacturers protocol. Eluted protein was dialyzed at 4 ⁰C for 16 hours against Storage Buffer (50 mM Hepes pH 7.5, 200 mM NaCl, 1 mM EDTA) and stored at -20 ⁰C. ## K_m and V_max determination for luciferin substrates Purified enzyme was diluted to a concentration of 5 μg/ml in Bright- Glo^TM buffer (Promega) containing 1 mM ATP and combined in individual reactions with a 10-fold dilution series of substrate ranging from 1pM– 1 mM. Reactions were incubated for 3 minutes at 25 ⁰C prior to measuring luminescence on a GloMax Multi+ Plate Reader. K_m and V_max were determined using Michaelis-Menten kinetics analysis of activity curves using GraphPad Prism software version 8.4.0. ## Luciferin competition assay Purified Ultra-Glo™ enzyme was diluted to a concentration of 0.01 mg/ml in 1xTBS + 1% PRIONEX. Separately, a reaction master mix containing 200 nM LH2 + 1 mM ATP was prepared in Bright-Glo^TM buffer. The diluted enzyme and reaction master mix were mixed 1:1 in a final volume of 100 μL. Reactions were incubated for 3 minutes at 25 ⁰C prior to measuring luminescence on a GloMax Multi+ Plate Reader. Following initial bioluminescence measurement, 5 μL of 20 mM 5,5-dialkylluciferin in DMSO was spiked into the reactions for a final concentration of 1 mM. Reactions were again incubated at 25 ⁰C for 3 minutes followed by luminescence measurement to determine loss in signal relative to uninhibited samples. ## K_m and V_max determination for ATP Purified enzyme was diluted to a concentration of 100 μg/ml in water and combined 1:1 with 100 μM substrate in Buffer A (100 mM MES pH 6.5, 5 mM MgCl2, 0.2% Tergitol, 0.002% Sodium Azide) in a total reaction volume of 100 μl. The mixture was incubated for 60 min at 25 ⁰C to eliminate any background luminescence. Reactions were initiated in individual reactions by spiking in a 50 μl volume of serially-diluted ATP ranging from 1 fM– 1 mM and reading luminescence continuously every 3 seconds for 2 minutes on a GloMax Discover Plate Reader (Promega). The peak luminescence value of each reaction across the reading time was used to calculate the K_m and V_max for ATP using Michaelis-Menten kinetics analysis of activity curves in GraphPad Prism software version 8.4.0. ## Luminescence-based thermostability profiling Thermal stability of proteins was determined using Differential Scanning Fluorimetry (DSF) by combining purified protein diluted to a concentration of 0.3 mg/ml in Storage Buffer and adding a final concentration of 5X Sypro Orange dye (Molecular Probes) in a 30 μL reaction volume. Reactions were transferred to Stratagene Mx3000p thermal cycler in a 96-well PCR plate and incubated at 30 ⁰C for 2 minutes prior to heating at a rate of 1°C/min to a final temperature of 100 ⁰C. Fluorescence measurements were taken at one-minute intervals using FAM excitation and ROX emission filters throughout the heating protocol. The transition melting temperature (T_m) of proteins were calculated from the peak of the first derivative of the fluorescence change over time. ## Luminescence spectra measurements Assays for measuring luminescence emission spectra were prepared by combining purified Ultra-Glo™ luciferase or mutants at a concentration of 5 μg/ml with 1 mM luciferin substrate and 1 mM ATP in Bright-Glo^TM buffer. The emission wavelength spectra of reactions were measured in white polypropylene 96-well plates at 25°C using a Tecan M1000Pro with 10 nm bandwidth, 90 gain, and 2 nm step settings. # Results and discussion ## Solution thermostability of luciferin is improved with modification at 5-positions With the goal of improving ATP detection reagent stability in a liquid format at ambient temperature, we decided to explore structure optimization of luciferin (LH2) to determine its impact on formation of dehydroluciferin (L), a major contributing factor to reagent instability. It was hypothesized that replacement of H atoms at 5 positions in the luciferin structure with alkyl substituents could potentially block the oxidation of the thiazoline ring, preventing inhibitory dehydroluciferin accumulation, thereby improving overall assay stability and performance. To test the hypothesis, 5,5-dimethylluciferin (**III-a**), an extensively studied luciferin analog, was compared to unmodified LH2 in an accelerated thermostability study where both compounds were incubated in solution at 60°C for 150 hours. Remarkably, we observed no degradation for **III-a** over the course of the study, whereas \>50% of LH2 decomposed into dehydroluciferin under identical conditions. Given that the presence of only 3% dehydroluciferin is sufficient to reduce assay performance by 20% (S1 Fig), its complete elimination would be predicted to have a significant impact on liquid detection reagent stability. This confirmed our hypothesis that modification of luciferin could improve its thermostability and suggested a general mechanism where two alkyl substituents at the 5 position, are capable of blocking degradation. Therefore, we synthesized and tested other 5,5-dialkylluciferins to identify candidates that are both thermostable and might be suitable as bioluminescent substrates for development into improved liquid detection reagents. ## Synthesis of 5,5-dialkylluciferins Encouraged by the improved thermostability of 5,5-dimethylluciferin (**III-a**), we set out to explore the 5 position of LH2. A focused set of 5,5-dialkylluciferin analogs were prepared to explore tolerance of steric hindrance and heteroatom substitution. In general, racemic β,β-dialkyl-cysteines (**II-a-h**) were first synthesized via reported procedures to pre-install the desired alkyl substitutions. Subsequently, racemic 5,5-dialkylluciferins (**III- a-h**) were synthesized via condensation between 6-hydroxy-2-cyanobenzothiazole (**I**) and racemic β,β-dialkyl-cysteines (**II-a-h**) following standard luciferin synthesis procedures. The synthetic process provided convenient access to 5,5-dialkylluciferin analogs with focused structural diversity, allowing us to systematically interrogate analog thermostability, enzyme-substrate interactions, and bioluminescence performance. ## Biochemical evaluation and HPLC-based stability profiling of 5,5-dialkylluciferins With these novel luciferin analogs in hand, we proceeded to perform biochemical testing to identify lead analogs using a firefly luciferase mutant, Ultra-Glo™, chosen because of its high thermostability and tolerance to chemical additives that are commonly found in homogenous liquid detection reagents. We were careful to identify reaction conditions that would establish a consistent baseline of background luminescence in our assays, since small amounts of activity from the 5,5-dialkylluciferins could easily be hidden due to fluctuations caused by buffer conditions, reagent concentrations, and reaction formulation. Under the conditions tested; however, none of the luciferin analogs showed significant luminescence above background levels. This result is consistent with another report that **III-a** is not a substrate for firefly luciferase. Although the luciferin analogs did not appear to be substrates for Ultra-Glo™ luciferase, during our testing we noticed that the established background luminescence signal of several reactions in Bright-Glo^TM buffer was suppressed in the presence of luciferin analogs such as **III-b** and **III-c**. Given that the background signal of our reactions required the presence of Ultra-Glo™ luciferase, we were intrigued by the observation that the 5,5-dialkylluciferins could inhibit the enzyme-dependent background luminescence of our research enzyme preparations. This suggested that the analogs could interact with the enzyme and that a relative measurement of their inhibition could help us evaluate substrate-enzyme interactions and subsequent potential as substrates for catalysis. We also drew support of their potential as substrates from previous report of 5,5-dialkylluciferin-AMP intermediates being substrates for *Photinus pyralis* and Click Beetle Green luciferases. To gain more insights into substrate-enzyme interactions with this strategy, a bioluminescence-based luciferin competition assay was designed to indirectly assess binding of the luciferin analogs in the active site. We reasoned that analogs that could efficiently bind to the active site would be more successful in competing with LH2 and result in reduced light output in the assay. Indeed, we observed that the relative light output of Ultra-Glo™ luciferase (5 ug/mL) with 100 nM LH2 was reduced in the presence of 1 mM of analogs **III-a-f**. There was also an inverse correlation between the size of the substituents at 5 positions and the percent of light output inhibition, indicating smaller substituents are better accommodated in the active site. In particular, we noticed analogs **III-a** and **III-b** showed significant inhibition, close to the background level of the assay, suggesting they had the strongest interactions with the enzyme. Based on this result, **III-a** and **III-b** were chosen as lead analogs for further thermostability characterization and subsequent evaluation as enzyme substrates. HPLC-based thermostability profiling was conducted to confirm stability of **III-a** and **III-b** in relevant assay buffer and investigate the decomposition products. Luciferin analogs **III-a** and **III-b** were reconstituted at 1 mM in Bright-Glo^TM buffer and incubated at 37 ⁰C; taking aliquots at various time points for analysis by RP-HPLC. Over the 2-month period of testing, 23% decomposition of LH2 to dehydroluciferin was observed, whereas no measurable decomposition could be measured for luciferin **III-a**. Some decomposition (5%) was observed for luciferin **III-b** bearing a spirocyclopropyl at the 5 position. We further investigated the decomposition of **III-b** to evaluate the impact of its individual products on bioluminescence assay performance. Decomposition products of **III-b** were enriched using a modified reaction condition (S2 Fig), and the identities were confirmed by ¹H NMR and HRMS (in). Decomposition product **III-b-s1** is likely to be formed via hydrolysis of the thiazoline ring due to increased ring strain induced by the spirocyclopropane. Decomposition product **III-b-s2** is a dehydroluciferin-like analog, which is possibly formed via an oxidative radical ring-opening process involving oxygen. Testing of both decomposition products, **III-b-s1** and **III-b-s2,** in biochemical assays showed only weak to moderate inhibition activities against Ultra-Glo™ mutants that make good utilization of **III-b** (S2 Fig), indicating that although III-b exhibits some decomposition in solution the products accumulate more slowly and are less inhibitory than those originating from LH2. Nevertheless, the improved thermostability properties of lead analogs **III-a** and **III-b** over LH2 demonstrated their utility as candidates for liquid-based detection reagents and, when taken together with their strong enzyme interaction potential, highlighted the need to identify luciferase variants that could use them as substrates efficiently. ## Improvement of substrate utilization through structure-guided mutagenesis of Ultra-Glo™ luciferase Although Ultra-Glo™ luciferase does not utilize the 5,5-dialkylluciferins we tested as substrates, our lead 5,5-dialkylluciferins did demonstrate the ability to compete for its native LH2 substrate, suggesting they could bind to the active site in an orientation potentially suitable for catalysis. Therefore, we sought to explore the evolvability of Ultra-Glo™ luciferase toward utilizing them as substrates and enable evaluation of our lead analogs in complete detection reagent formulations in real-time studies. We initially targeted mutagenesis of residues in the active site that would be predicted to influence substrate binding and/or specificity using a three-dimensional homology model of Ultra-Glo™ luciferase based on the structures of the related *Luciola cruciata* (PDB ID: 2D1S) and *Photinus pyralis* (PDB ID:4G36) luciferases in closed- conformation complexes with the high-energy intermediate analogue, 5′-*O*-\[*N*-(dehydroluciferyl)-sulfamoyl\]adenosine (DLSA). Using the positioning of the DLSA inhibitor in the native structure as a guide, we superimposed **III-a** in the active site and identified residues in proximity to the substrate that could be determinants in changing the specificity of Ultra-Glo™ for the 5,5-dialkylluciferins. From this model, 24 amino acid residues within 5 Å of the active site substrate were identified to serve as candidates for mutagenesis. Each of these positions in Ultra-Glo™ was saturated with amino acid substitutions in individual mutant libraries and screened in *E*. *coli* cell lysates for activity against LH2 and **III-b**, followed by secondary screening of hits in smaller batches against **III-a**. A single mutation, H244W, was identified that showed large 17-fold and 88-fold improvements in specific activity with **III-a** and **III-b**, respectively (M2). Examining this residue in our homology model showed that this substitution makes additional hydrophobic interactions between 5,5-dialkyl groups and the larger indole sidechain of tryptophan. The identification of the H244W mutant provided enough of an improvement in brightness to enable additional screening for Ultra-Glo™ luciferase mutations with **III-b**, while **III-a** was included in secondary assays throughout the process to monitor for general improvements. We next explored changes by incorporating a set of characterized mutations from an in-house repository of Ultra-Glo™ luciferase variants identified during other high-throughput screens of the enzyme that impact substrate utilization. Testing of these additional mutations (M1 and M3, S1 Table in) with our lead luciferin analogs followed by combinatorial optimization of the best hits resulted in the addition of mutations T344A and I396K to the H244W template which further improved its activity 225-fold with **III-b** over the starting Ultra-Glo™ template (M4). These mutations are not in close proximity to the active site of our homology model, suggesting they exert their influence on activity through other structural mechanisms. Together, the striking improvements in Ultra-Glo™ from these rationally-guided point mutations provided evidence that the enzyme could be further evolved to efficiently utilize our lead 5,5-dialkyl luciferin analogs and justified a larger-scale, randomized directed evolution effort to identify additional mutations. ## Improvement of substrate utilization through directed evolution of Ultra-Glo™ luciferase mutants Directed evolution of Ultra-Glo™ luciferase was carried out through successive rounds of introducing diversity through mutagenesis, activity screening, and combinatorial optimization of hits to yield variants with improved brightness and kinetic properties with our lead luciferin analogs. We employed multiple mutagenesis strategies of the full-length gene. We combined random mutagenesis by error-prone PCR of both the existing and a codon-diversified template as well as combinatorial mutagenesis by DNA shuffling. For both strategies, individual variants of Ultra-Glo™ luciferase were overexpressed in *E*. *coli* cultures and tested for activity in cell lysates with the addition of 1 mM ATP and 1 mM of either **III-a** or **III-b**. Variants with 2-fold or greater improvements were selected for secondary screening to further characterize their kinetic profile (K_m and V_max) with each substrate and their DNA sequenced to identify mutations. The brightest combinations of mutations in a single gene were used as the template for the next round of evolution. Following the initial brightness improvements realized with the triple H244W+T344A+I396K mutant, successive rounds of directed evolution showed continual improvements in brightness, eventually leading to a specific activity level of M6 with **III-b** comparable to that of Ultra-Glo™ and LH₂. Overall, our directed evolution screening was successful in both broadening the substrate specificity of Ultra-Glo™ to include multiple 5,5-dialkylluciferins and identifying variants that could efficiently utilize them for bright luminescence output. Achieving a level of bioluminescence performance close to that of Ultra-Glo™:LH2 is especially notable, considering our lead 5,5-dialkylluciferin analogs were actually inhibitors of our starting Ultra-Glo™ luciferase template prior to evolution. Two Ultra-Glo™ luciferase mutants, M6 and M7, resulting from our evolution strategy were chosen for more detailed biochemical characterization since they had the most significant brightness improvements with **III-a** and **III-b** while retaining relatively high thermostability. Comparison of V_max values between Ultra-Glo™ luciferase and the two mutant enzymes showed that M6 paired with **III-a** and **III-b** had V_max values of 2.8 x 10⁶ and 2.7 x 10⁷ RLU, respectively; the latter of which is about 1.5-fold brighter than the specific activity of Ultra-Glo™ with LH₂. M7, a variant evolved concurrently to improve the enzyme thermostability over M6, retains the preference for utilization of **III-a** and **III-b** although with slightly lower V_max values of 2.4 x 10⁶ and 3.5 x 10⁶ for the two substrates. Comparison of other kinetic properties for mutants M6 and M7, summarized in, showed both mutant enzymes exhibit a red shift of \~40 nm in their peak luminescence signal independent of the substrate tested and have higher Km for the 5,5-dialkylluciferins than LH2. Since we observed that many of the mutations selected during directed evolution with **III-b** also led to improvement utilization of **III-a** by Ultra-Glo™ mutants, we were curious if the other 5,5-dialkyl luciferins in our original panel would have comparable improvements as well. For unmutated Ultra-Glo™ luciferase, little or no luminescence could be observed across the panel of luciferin analogs we tested (S3 Fig). However, the evolved mutants M6 and M7 were able to utilize multiple analogs including **III-a**, **III-b**, **III-c**, and **III-d** (S3 Fig). The modifications in this subset are comprised of 5,5-dimethyl and a series that included 5,5-cyclopropyl, 5,5-cyclobutyl, and 5,5-cyclopentyl ring substitutions. These results demonstrate that the improvements to Ultra-Glo™ luciferase through directed evolution with **III-b** partially translated to other 5,5-dialkyl luciferin analogs and highlight the structural plasticity accommodated by the chemical mechanism of bioluminescence in the firefly-based system. ## Thermostability profiling of Ultra-Glo™ mutants While our evolution program was primarily focused on enhancing the utilization of 5,5-dialkyl luciferins with Ultra-Glo™ luciferase in order to evaluate their potential as substrates for single-liquid ATP detection systems, we also evaluated in parallel the thermostability of our mutant enzymes relative to the highly-stable Ultra-Glo™ parental template. Using differential scanning fluorimetry (DSF), we calculated the transition melting temperature (T_m) of the purified enzymes as they were exposed to an increasing temperature gradient. We observed that our brightest mutant, M6, had a 10°C reduction in T_m relative to Ultra-Glo™ luciferase, indicating that among the combination of amino acid substitutions selected for 5,5-dialkylluciferin utilization included some detrimental to the enzyme’s thermostability. Based on DSF data from enzymes that were intermediates in the evolution process, we were able to identify several mutations in M6 that were responsible for the decrease in stability. After reverting several of these mutations to their starting amino acid in Ultra-Glo™ and further directed evolution screening to identify additional stabilizing mutations, we were able to increase the T_m to 75 ⁰C in mutant M7, albeit at the expense of its V_max with our lead analogs. With its high thermostability and brightness with 5,5-dialkyl luciferins, mutant M7 had an appealing set of properties that would enable initial testing of our highly stable luciferin analogs as components of a homogenous, liquid ATP detection system. ## Evaluating 5,5-dialkylluciferins for use in single-liquid ATP detection systems The firefly luciferase-based ATP detection system has been widely adopted for monitoring cell health in life science research, and as first-line detection of microbial pathogens in industrial settings. Currently, single-liquid ATP detection reagents require storage under at least refrigeration conditions (4 ⁰C) to maintain reasonable stability (T_80, timing maintaining 80% initial activity, greater than 30 days). Therefore, improvements in the stability of single-liquid ATP detection reagents are highly desired to allow storage at ambient temperature, enabling accurate and reliable detection of ATP in industrial settings in simplified formats without the need for dedicated laboratory equipment or personnel. To evaluate the impact of enzyme and substrate stability in a relevant assay context, we decided to use bioluminescence-based, single-liquid ATP detection system as the testing model to compare the current Ultra-Glo™: LH2 detection system with the newly evolved M7:**III-a** system comprising our most stable components. We performed three parallel stability studies to compare both systems side by side at 37 ⁰C for 150 days. Within each study, three replicate 1 mL formulations were tested that comprised 1 mM substrate and 0.5 mg/mL enzyme in Detection Reagent Buffer, which allows simultaneous capacity for cell lysis and bioluminescence detection. Throughout the studies, aliquots were taken at indicated time-points and assayed for luminescence by addition of 1 mM ATP. The thermostability parameter, T₈₀, was determined by plotting time versus luminescence. Control formulations containing only substrate or enzyme alone were also included, with the held-out component spiked in at the time of activity measurement at each time-point. Comparison of real-time stability profiles between the complete single-liquid reagents and stand-alone controls revealed different critical factors for the performance of the Ultra-Glo™:LH₂ and M7:**III-a** systems. For Ultra-Glo™:LH2, both Ultra-Glo™ and LH₂ lost \~20% of initial activity based on brightness within 30 days of incubation at 37 ⁰C. Since 80% of 1 mM LH2 still saturates Ultra-Glo™ luciferase light production, the fact that the complete reagent formulation lost \~40% of initial activity within the same period confirmed that formation of dehydroluciferin has significant impact on complete reagent stability at ambient temperatures. For the M7:**III-a** system, it was noted that **III-a** alone maintained exceptionally high thermostability with very little change to its luminescence profile, which is consistent with HPLC-based analysis of accelerated stability studies. Although there was similar initial reagent stability for the complete M7:**III-a** system (T₈₀ = \~15 d) and the existing Ultra-Glo™:LH₂ system (T₈₀ = \~20 d), there was an \~25% increase in stability for the new system at timepoints beyond 60 days. This improvement in long-term stability for the M7:**III-a** system is likely due to the presence of **III-a**, as its superior stability relative to LH2 in substrate-alone controls became more prominent at these later timepoints in the study. In addition, since the stability trend for the M7:**III-a** was nearly identical to M7 alone, it is clear that the stability of enzyme component is the limiting factor now that the decomposition of the substrate has been eliminated. Together, the results of using an ATP detection assay as a model confirmed that the high thermostability of the luciferin analog, **III-a,** translates into improved real-time performance in a homogenous liquid detection format, providing essentially no loss in performance over 5 months of storage at 37°C/98.6°F. # Conclusion We have improved thermostability of the standard firefly luciferase substrate, LH2 through chemical modification to create 5,5-dialkylluciferins that reduce or eliminate oxidative decomposition to dehydroluciferin, a potent inhibitor of firefly luciferases and a major limitation for firefly-based ATP detection systems. Despite their initial poor utilization as substrates by Ultra-Glo™ luciferase, 5,5-dialkylluciferins **III-a** and **III-b** were found to display favorable interactions with the enzyme using a luciferin displacement assay. Subsequent direct evolution of Ultra-Glo™ with these analogs yielded a set of luciferase mutants that showed efficient utilization of them as substrates. Notably, the combination of mutant M6 and **III-b** produced a bioluminescence signal comparable in strength to that of Ultra-Glo™ luciferase and LH2. The importance of discovering a bioluminescent substrate resistant to spontaneous oxidative decomposition here was demonstrated by real-time stability assessment of a homogenous solution containing **III-a** and the luciferase mutant, M7. Throughout the course of a 5-month, real-time stability study we observed no loss in stability or decomposition of **III-a** in solution at 37°C. In a complete formulation, the pairing was \~25% more stable compared to the existing Ultra-Glo™:LH2 system across the length of the study and appeared to only be limited by the eventual loss of stability of the M7 enzyme component. This suggests that further engineering of stability into the luciferase variant to match that of **III-a** would be the next step in developing a homogenous, liquid ATP detection reagent with even higher stability. Our results show that as both individual and combined components, 5,5-dialkylluciferins and thermostable Ultra-Glo™ luciferases form the foundation of a high performance, ATP detection platform for users in industrial or other settings where reagent stability and simple workflow are of critical importance. More broadly, this novel combination also expands the available modifications of luciferins, providing new tools for potential development of the firefly system toward additional color modulations and pro-luciferin chemistries. Such substrates, with less potential to be oxidized into dehydroluciferin, could potentially improve a wide range of luciferin-based bioluminescence assays. For example, improvements to the luminogenic substrates used in oxidation assays for cytochrome P450 family enzymes could further enable homogenous and high- throughput formats by increasing substrate stability during assay setup and screening. Similarly, bioluminescence assays for reactive oxygen species (ROS) could benefit from more stable, less-oxidatively sensitive 5,5-dialkylluciferins for measuring ROS activity in mitochondria during live-cell based measurements, including those performed in multiplex format with other cell-health detection assays. Together, the improved stability and robustness of both enzyme and substrate components will continue to enable researchers to take advantage of the exceptional sensitivity and flexibility of bioluminescence assays in the future. # Supporting information We thank the Analytical Services Group at Promega Biosciences, LLC, for the characterization of the synthesized compounds and Kris Zimmerman for molecular biology support. [^1]: I have read the journal's policy and the authors of this manuscript have the following competing interests: Authors are paid employees of Promega Corporation. Promega Corporation manufactures and sells Ultra-Glo™ luciferase, luciferin, and ATP detection reagents. Ultra-Glo™ luciferase mutations are disclosed in the published patent application US20200071682A1 “Luciferase Enzymes For Use With Thermostable Luciferins In Bioluminescent Assays”, and the luciferin analogs are disclosed in granted patent US 10,400,264 and the published patent application US 20190338340A1, “5,5-disubstituted Luciferins And Their Use In Luciferase-based Assays” owned by Promega Corporation. The authors confirm that these competing interests do not alter their adherence to all the PLOS ONE policies on sharing data and materials.

# Introduction Since the creation of the International League Against Rheumatism (ILAR), China has cooperated in epidemiological studies of common rheumatic diseases. Research in the field has proceeded gradually in different areas in China, with a focus on identifying the general epidemiology of common rheumatic diseases in China. To determine whether changes in the national economy and people’s living environments and life styles during the past three decades have affected the prevalence of common rheumatic diseases, another epidemiological survey supported by the Asia Pacific League of Associations for Rheumatology (APLAR) was undertaken in 2012 in the Shantou area. This paper reports the results of that survey. # Materials and Methods ## Study population Two randomly selected population samples in the Shantou region were surveyed. One population had lived in buildings of 7 to 9 floors without elevators for more than 10 years, and the other had resided in buildings with elevators for more than 10 years. All subjects were 16 years or older. All participants gave their written informed consent after receiving detailed explanations of the study and its potential consequences prior to enrollment. Written informed consent of participants under 18 was given by their parents or caretakers on their behalf. This study was performed with the approval of the Human Ethical Committee of Shantou University Medical College. ## Working team and training The working team was headed by QYZ and included 6 primary health care workers who were familiar with the local area, 20 senior medical students and resident physicians, 3 rheumatologists, and 2 radiologists. The working team received standard training prior to the survey, including information regarding the survey’s contents, survey procedure, house-to-house survey methods, inquiry techniques, physical examination, daily summary, and data input. A pilot survey was carried out in order to verify the unified criteria and quality of the study. To ensure cooperation among the selected sample population before the survey began, the objective and significance of the study were widely propagandized, and the content and method of the survey were also made known in written form to all participating families. ## Research methods The protocol of the World Health Organization-International League Against Rheumatism (WHO-ILAR) Community Oriented Program for Control of Rheumatic Diseases (COPCORD) was implemented. The WHO-ILAR COPCORD Core Questionnaire was used by the primary health care workers and the senior medical students for the house-to-house visits during phases I and II; positive respondents identified in phases I and II were interviewed and examined by rheumatologists and senior resident physicians in phase III; relevant laboratory tests and radiographs for subjects suspected of having rheumatic diseases were carried out in phase IV; finally, the rheumatic diseases were characterized and classified, and the data were summarized in the last phase. If a participant was absent or could not be contacted during the survey, another visit was made. All radiographs were independently read by 2 radiologists who were unaware of the clinical data. Finally, 3 rheumatologists made diagnoses according to the examination and test results. ## X-ray examinations \(1\) At minimum, frontal radiographs of both hands were taken for the subjects suspected of having rheumatoid arthritis (RA). (2) Frontal radiographs of the pelvis and anteroposterior and lateral radiographs of the lumbar spine were obtained for those suspected of having ankylosing spondylitis (AS). (3) When knee osteoarthritis (KOA) was suspected, anteroposterior and lateral radiographs of the knee joints were taken (4) Finally, in participants who were thought to have gout, anteroposterior and oblique radiographs of the affected foot were obtained to confirm the diagnosis. ## Laboratory examinations \(1\) Rheumatoid factor and anti-keratin antibody were tested in subjects suspected of having RA. (2) Those participants in whom ankylosing spondylitis (AS) was suspected underwent HLA-B27 testing. (3) Uric acid levels were tested for those patients thought to suffer from gout. ## Diagnostic criteria RA was diagnosed according to the American College of Rheumatology (ACR) 2010 criteria. AS was confirmed according to the modified New York Criteria. The ACR 1986 Criteria were used to diagnose KOA. Gout was confirmed according to the ACR 1977 Criteria. Fibromyalgia syndrome was diagnosed using the ACR 1990 Criteria. ## Statistical analysis The prevalence was standardized according to the fifth national census population. The Poisson distribution was used to calculate the 95% confidence intervals (95% CI). Rate comparisons were assessed using the chi-square test. A comparison of the mean values between 2 samples was performed by an independent- samples *t*-test. Logistic regression was used in the risk factors analysis. All statistical analyses were two-tailed and involved the use of the Statistical Package for the Social Sciences (SPSS) version 21.0 (IBM, Armonk, NY, USA). A probability value of P \< 0.05 was considered statistically significant. # Results A total of 5504 residents lived in the 2 surveyed sites, including 3103 residents in buildings without elevators and 2401 residents in buildings with elevators. The final number of residents actually surveyed was 4056, of which 2337 resided in buildings without elevators and 1719 resided in buildings with elevators. The overall response rate for the questionnaire was 73.7%, and the response rates for those living with and without elevators were 71.6% and 75.3%, respectively. The average age of the 4056 residents surveyed was 48.4 years; there were 1948 men with an average age of 47.2 years and 2108 women with an average age of 49.4 years. The ratio of men to women was 1:1.08. The age and sex of the residents surveyed are shown in, and their occupations are displayed in. Among the 4056 residents surveyed, there were 831, 560, 282, 241, 170, 73, and 111 participants with rheumatic pain, knee pain, neck pain, lumbar spine pain, shoulder pain, elbow pain, and foot pain, respectively, who accounted for 20.5%, 13.8%, 7.0%, 5.9%, 4.2%, 1.8%, and 2.7% of the study population, respectively. The associated prevalence values for each of these types of rheumatic pain were 15.7% (95% CI: 14.7–16.9%), 10.2% (95% CI: 9.5–11.3%), 5.6% (95% CI: 4.9–6.3%), 4.5% (95% CI: 3.9–5.1%), 3.1% (95% CI: 2.6–3.6%), 1.4% (95% CI: 1.0–1.8%) and 1.8% (95% CI: 1.4–2.2%), respectively, after adjusting for age and sex. Among the surveyed population, 421 subjects had KOA, 56 had gout, 19 had RA, 12 had AS, and 5 had fibromyalgia (FM); the associated crude prevalence for each of these conditions was 10.38%, 1.38%, 0.47%, 0.30%, and 0.12%, respectively, and after adjustment, the standardized prevalence was 7.10% (95% CI: 6.31–7.89%), 1.08% (95% CI: 1.02–1.74%), 0.35% (95% CI: 0.17–0.53%), 0.31% (95% CI: 0.14–0.48%), and 0.07% (95% CI: 0–0.15%). ## The sex, age, housing elevator status, and occupational distribution of subjects with rheumatic pain and knee pain ### Sex As shown in, for subjects living in buildings without elevators, the standardized prevalence of rheumatic pain in women was 23.8% (95% CI: 21.4–26.2%), which was notably higher than the 13.4% (95% CI: 11.4–15.4%) in men (P \< 0.001). The standardized prevalence of knee pain in women was 15.4% (95% CI: 11.44–15.68%), which was also notably higher than the 7.4% (95% CI: 5.52–8.71%) in men (P \< 0.001); with the exception of the foot, the pain prevalence in other body parts was also higher in women than in men (all P \< 0.001). For subjects living in buildings with elevators, the pain prevalence in all parts of the body except the foot was lower than that in subjects living without elevators, regardless of gender (all P \< 0.001), and was also significantly lower in men than in women except in the shoulder and elbow (all P \< 0.05). ### Age As shown in, the prevalence of both rheumatic pain and knee pain displayed a notably elevated trend after age 35 among the subjects, regardless of whether they were living in buildings with or without elevators and regardless of their gender, but the rising trend was sharper in women than in men. ### Housing elevator status As shown in, the standardized prevalence of rheumatic pain and knee pain was 18.5% (95% CI: 17.4–20.6%) and 11.3% (95% CI: 10.24–12.84%), respectively, in subjects living in houses without elevators and was significantly higher than the 11.9% (95% CI: 9.9–12.9%) and 8.9% (95% CI: 7.55–10.25%) (both P \< 0.001), respectively, for subjects living in houses with elevators. The pain prevalence in all parts of the body except the foot was higher in subjects living in houses without elevators than it was in subjects living in houses with elevators (all P \< 0.01). ### Occupation As shown in, the prevalence of rheumatic pain was greatest in blue-collar workers (29.37%), followed by teachers (23.38%), white-collar workers (18.47%), individual industrialists and businessmen (12.77%), and finally, students (3.78%). The differences among the groups were statistically significant (χ² = 135.566, P \< 0.001). The prevalence of knee pain by occupation had a similar trend, ranging from blue-collar workers (19.95%) to teachers (15.38%), white-collar workers (12.29%), individual industrialists and businessmen (9.04%), and students (2.41%). Again, the differences among the groups were statistically significant (χ² = 88.213, P \< 0.001). For participants living in houses without elevators, the prevalence of rheumatic pain was highest in blue-collar workers, followed by teachers, white-collar workers, individual industrialists and businessmen, and finally, students; the differences among the groups were statistically significant (χ² = 90.687, P \< 0.001). For subjects who resided in houses with elevators, the prevalence of rheumatic pain was highest in blue-collar workers, followed by white-collar workers, teachers, individual industrialists and businessmen, and finally, students, and the differences among these groups were statistically significant (χ*2* = 31.207, P \< 0.001). The prevalence of knee pain in subjects who lived in houses without elevators was highest in blue-collar workers, followed by teachers, individual industrialists and businessmen, white-collar workers, and finally, students, and the differences among the groups were statistically significant (χ*2* = 46.412, P \< 0.001). In participants who had regular access to elevators at home, the prevalence of knee pain was highest in blue-collar workers, followed by teachers, white-collar workers, individual industrialists and businessmen, and finally, students, and the differences among the groups were statistically significant (χ*2* = 58.735, P \< 0.001). ## The sex, age, housing elevator status, and occupational distribution structure of subjects with common rheumatic diseases ### Knee osteoarthritis The prevalence of KOA was more than 2 times greater in women than in men. It increased with age; it was less than 2% before age 35, increased notably after age 45, reached 24.3% at age 65–74, and was 37.5% at age 85 or older. As shown in, the total standardized prevalence of KOA was higher in subjects living without elevators than in subjects with elevators (7.64% vs. 6.26%, P = 0.162) and was significantly higher in subjects aged 16–64 who resided in buildings without elevators than it was for those with access to elevators (5.89% vs. 3.95%, P = 0.004). As shown in, the prevalence of KOA was highest in blue-collar workers, followed by teachers, white-collar workers, individual industrialists and businessmen, and finally, students. Differences among the groups were statistically significant (χ² = 105.689, P \< 0.001). ### Gout The prevalence of gout was 6 times higher in men than in women (1.82% vs. 0.29%). The prevalence of this condition also increased with age and reached a peak at ages 65–74; except for one female patient who was younger than 35, no men under 30 or women under 45 had gout. The prevalence of gout was highest in individual industrialists and businessmen, followed by teachers, blue-collar workers, and white-collar workers; there were no students with gout. Differences among the groups were statistically significant (χ² = 12.856, P = 0.009). ### Rheumatoid arthritis and ankylosing spondylitis The prevalence of RA was 0.35% (95% CI: 0.17–0.53%), and was more than 2 times higher in women than in men (0.48% vs. 0.23%, 95% CI: 0.18–0.78% and 0.02–0.44%). The prevalence of AS was 0.31% (95% CI: 0.14–0.48%), and it was 5 times higher in men than in women (0.51% vs. 0.10%, 95% CI: 0.19–0.83% and 0–0.23%). These 2 rheumatic diseases showed no evident relationship with housing elevator status or occupation. ### Fibromyalgia All 5 subjects with fibromyalgia were female. The prevalence was 5/4056. The ages of those with fibromyalgia ranged from 45–64. Among the 5 subjects, 2 were blue-collar workers, 2 were white-collar workers, and 1 was a teacher. All the above subjects reported being under psychological stress. ## Changes in the prevalence of common rheumatic diseases in Shantou in the past 3 decades, and comparison of these changes with those in other major cities and countries ### Rheumatic pain Compared with the rural survey results of 1987, in the current survey, there was a lower prevalence of lumbar pain (4.5% vs. 13.0%, P \< 0.001), a similar prevalence of shoulder, elbow, and foot pain, and a higher prevalence for all other body parts (15.7% vs. 11.6%, P \< 0.001), knee pain (10.2% vs. 2.6%, P \< 0.001), and neck pain (5.6% vs. 2.0%, p\<0.001). Compared with the town survey results of 1992, except for a higher prevalence in all parts of the body and a lower prevalence both in lumbar pain and neck pain, the prevalence results in all other parts of the 2 surveys were similar. Compared with the urban survey results of 1995 and the rural survey results of 1999, the prevalence of rheumatic pain in all parts of the body was lower (15.7% vs. 18.1% and 19.8%; P = 0.018 and P \< 0.001, respectively) in particular for subjects living in buildings with elevators. The prevalence of rheumatic pain in this survey was also lower than the rate previously reported in Taiwan, Shanghai in both 1992 and 1998, Taiyuan, and Beijing (15.7.0% vs. 23.0%, 24.3%, 21.2%, 25.4%, and 40.3%, respectively) (all P \< 0.001). ### Knee osteoarthritis There was an increased prevalence of KOA in Shantou compared with that in 1992 and 1995 (7.10% vs. 1.3% and 3.2%, respectively) (both P \< 0.001), and the increase was more evident in subjects who lived in buildings without elevators than in those who had access to elevators (7.64% vs. 6.26%). The prevalence of KOA in this study was similar to those reported in Taiyuan (7.57%) and Shanghai (7.2%), but it was still lower than that in Beijing (9.6%, P = 0.064). ### Gout The prevalence of gout in Shantou was higher in the current study than it was in those reported in 1992, 1995, and 1999 (1.08% vs. 0.17%, 0.15%, and 0.26%, respectively; all P \< 0.01). The prevalence was also higher than that of rural, suburban, and urban Taiwan in 1994 (1.08% vs. 0.16%, 0.67%, and 0.67%; P \< 0.001, P = 0.067, and P = 0.067, respectively) \[14), and higher than that reported in Shanghai in 1992, 1997, 1998, and 2002 and in Beijing in 2011 (1.08% vs. 0.2%, 0.34%, 0.22%, 0.28%, and 0.09%, respectively; (all P \< 0.01). However, the prevalence of gout in the current survey was lower than that of the NAHSIT (the two Nutrition and Health Surveys in Taiwan) survey results in 1993–1996 and 2005–2008 (1.08% vs. 3.4%, 5.2%, respectively; both P \< 0.001) and was similar to that of the cadres examination in Beijing in 2005 (1.0%). ### Fibromyalgia The prevalence of fibromyalgia in the current study was higher than that of the former native survey results in 2004 and the Taiyuan survey result in 2007 (5/4056 vs. 2/2350 and 1/3915), but it was still lower than that reported in other countries, including the United States (US), Japan, India, Bangladesh, and Mexico. ### Rheumatoid arthritis and ankylosing spondylitis The prevalence rates of RA and AS showed no evident changes over time, and they were similar to those in other civil survey results. The prevalence of RA was similar to that in India, but was lower than that in Mexico and Caucasians, while the prevalence of AS was similar to that found in Western countries, but lower than that in India. ## Factors associated with rheumatic diseases Logistic regression was adopted to analyze the effects of sex, age, occupation, and housing elevator status on rheumatic diseases. The result showed that age, sex, stair-climbing, and occupation were all risk factors for knee pain and knee osteoarthritis, and that sex and age were risk factors for gout. # Discussion COPCORD was introduced by the WHO and ILAR last century to help developing countries control rheumatism. It proposed 3 stages: the first is an epidemiological, community-based survey of rheumatic diseases; the second includes treatment and health education for rheumatic diseases; and the third involves validation of the risk factors for rheumatism, both environmental and genetic, in order to prevent or decrease rheumatism. By participating in the ILAR-China epidemiological survey of Chinese rheumatism from the early 1980s to the late 1990s, the Rheumatology Department of Shantou University Medical College carried out successive surveys among 5 cohorts, including a total of 23,867 residents over 16 years of age, and found that there was a rising trend in the prevalence of rheumatic pain and an increase in the prevalence of gout in the Shantou area. The change in the prevalence of these common rheumatic diseases was considered to be related with a change in people’s socioeconomic status and living habits, and stair-climbing was considered to be a risk factor for knee pain and knee osteoarthritis. All questionnaires used in the aforementioned surveys were identical to the WHO- ILAR COPCORD Core Questionnaire; the only difference was that the subjects were examined soon after the questionnaires, whereas in the WHO-ILAR COPCORD Core survey, medical examinations were carried out within 3 days after the participants responded to the questionnaires. All questionnaires used in the above surveys were translated from Chinese to English and then from English back to Chinese, and they were validated by clinical verification and small-scale investigations. Moreover, there was a stable working team for the surveys in which members including the research director, rheumatologists, radiologists, and even the supervisors from the Chinese Rheumatism Association, APLAR, and ILAR were unchanged. The current survey was consistent with the previously mentioned COPCRD surveys. All these consistent surveys were conducted in Shantou area, where the overwhelming majority of the population has always remained Han people due to its culture and situation. Moreover, the working team and the methodology remained unchanged, greatly decreasing the potential for errors in the surveys and therefore ensuring comparable results over the past 3 decades. With land prices soaring due to urbanization and the expansion of city scales in the past 3 decades, buildings without elevators have risen higher and higher, while buildings with elevators sprang up around the year 2000 in Shantou and have become as common as buildings without elevators. In contrast, almost all subjects in the previous COPCORD surveys conducted in Shantou from 1987 to 1999 lived in buildings without elevators, so the current survey was conducted in two randomly selected population samples with and without elevators separately to further ensure comparable results. ## Rheumatic pain There have been 5 rheumatic epidemiological surveys on rheumatic pain since the 1980s in Shantou, and the prevalence of rheumatic pain has showed an upward trend from 1987–1999 but a decrease from 1999 until our survey was carried out. These results might be related to people’s strong work ethic and the fierce competition in the workforce since the 1980s, as well as the corresponding increase in the social economy and increasingly higher living standards, lower labor intensity, and the development of health services since the turn of the century. ## Knee pain and knee osteoarthritis The survey showed that the prevalence of knee pain and KOA had increased despite the overall declining prevalence of rheumatic pain, and it suggested that knee pain and KOA were not only related with stair-climbing but also with occupations. Our 1995 survey showed that the prevalence of knee pain, lumbar pain, and degenerative changes in residents living on the 4^th to 5^th floors of buildings without elevators was twice that of residents living in single-story houses, but we could not verify the relationship between stair- climbing and knee pain or KOA in a later study in Taiyuan. The results of the current survey showed that both the prevalence of knee pain and KOA were higher in residents living in buildings without elevators than in buildings with elevators, suggesting that the higher prevalence of knee pain and KOA might be related to stair-climbing. However, the different effects of stair-climbing on different age groups will require further clarification. Regarding the 2006 Taiyuan report, the relationships between stair-climbing and the prevalence of knee pain or KOA were drawn from comparisons of the prevalence rates in residents living on different floors in buildings without elevators, which had some limitations: one limitation was the lack of a comparison with residents living in buildings with elevators or in single-story houses, and the other was that there were too many confounding factors, such as the residents’ distribution on different floors being related to their social status for buildings owned by an employer, or that older and weaker people or those with knee trouble tended to live on the lower floors. With urbanization and the expansion of cities over the past 3 decades, land prices have soared, resulting in little to no construction of single-story houses except for a few luxury villas in both urban and more rural settings; in fact, even blocks of buildings under 4–5 stories without elevators are scarce. Therefore, it is difficult to carry out a cohort study between single-story houses and buildings without elevators. The comparison in this study was between residents living in buildings with and those without elevators, which was a rational cohort to study the association between stair-climbing and knee pain or KOA. The buildings in the 2 selected survey sites were completed at almost the same time 10 years ago; the houses were bought individually, and the residents moved into their residences at almost the same time. Therefore, there were few possible confounding factors in the survey. The survey results showed no apparent differences in the prevalence of KOA in residents younger than 40 years of age, regardless of whether their residence had elevators. The difference in KOA prevalence appeared after 45 years of age, which was in accordance with the development of KOA and previous epidemiological survey results, suggesting the reliability of the survey results. The reason for the increase in the prevalence of knee pain and KOA despite the decreasing overall prevalence of rheumatic pain is worthy of attention. The increased prevalence of knee pain and KOA may have been affected by the general prolongation of the life span due to economic development during the past three decades, which has resulted in a larger population of older people, who typically report a higher prevalence of knee pain and KOA. The survey results suggested that knee pain and KOA were not only associated with stair-climbing but also with a person’s occupation. When adjusted for age and sex, both the prevalence of knee pain and KOA were higher in residents living in buildings without elevators than in buildings with elevators. However, people who already had knee pain or KOA likely had a tendency to move to houses with elevators. At the same time, being sedentary may have falsely “improved” the pain, so it was thought that the prevalence rates of knee pain and KOA would be higher in subjects who chose to live in buildings with elevators than in those who moved into residences without elevators 10 years ago. However, the survey results 10 years later were contrary, which further verifies the effect of stair-climbing on knee pain and KOA, but no data regarding the prevalence of knee pain or KOA were obtained 10 years ago. Other factors, such as socioeconomic status, body mass index, waist circumference, comorbidities, education level, and smoking history, might also have contributed to the development of knee pain or KOA. Therefore, further study will be required to determine the role of these factors and explore their interactions. ## Gout The prevalence of gout has shown a rising trend in Shantou since the 1990s, and it has also risen in both mainland China and Taiwan. Our survey results showed that the prevalence of gout in Shantou was currently significantly higher than that at the end of the last century, which is undoubtedly related to the rapid development of national economics, greatly improved living standards, and changed lifestyles in the past 3 decades in mainland China; however, the prevalence of gout is still lower than that reported by Western countries in the 1990s. So far, survey results on gout in various areas of China or among various cohorts in the same area have differed. Therefore, it has been suggested that both genetics and lifestyle may influence this disease. ## Fibromyalgia With the exception of the surveys conducted in Taiyuan and Shantou by our research center, no epidemiological data on FM have been obtained in China. It is noteworthy that the prevalence of FM in this survey was higher compared to the survey results from 2004 and 2007 (5/4056 vs. 2/2350 and 1/3915). The 5 women with FM in this survey were approximately 50 years old, and they all reported experiencing mental or psychological pressure from their children’s employment or other domestic problems, which suggests the importance of considering people’s mental and psychological pressure during the rapid development of social economics, particularly in those vulnerable to such pressures. Although the prevalence of FM was higher in our study compared with that in the previous surveys, it was still lower than that reported in western countries, which may be related to the different paces of life, working pressures, or genetic factors in western countries. Further studies are needed to investigate this relationship. In conclusion, in the past 3 decades, the prevalence of RA and AS remained stable, whereas the prevalence of KOA, gout, and FM increased notably; these changes occurred alongside major changes in the social economic status and overall lifestyle in China. Age, sex, stair-climbing, and occupation were all risk factors for knee pain and knee osteoarthritis, and sex and age were also risk factors for gout. More risk factors should be investigated in future studies. # Supporting Information We would like to thank the Health and Family Planning Commission of Shantou City and the third and fourth Shantou Municipal Hospital for their assistance. We also appreciate the contribution of the medical students from the English-based medical program class of 2009 at Shantou University Medical College. This work was supported by APLAR COPCORD Grant 2012 and was also partly funded by a grant from Shantou University Medical College (NO.LDo30601). [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SYZ YG QYZ LPL. Performed the experiments: SYZ YG YPZ SBC JYC CQL JHP ZDH JQZ HJL GHH DMW HYL LPL QYZ. Analyzed the data: SYZ YG SBC QYZ ZDH JQZ HJL GHH DMW HYL. Contributed reagents/materials/analysis tools: JYC CQL HYL JHP LPL. Wrote the paper: SYZ YG QYZ LPL. [^3]: ‡ These authors are co-first authors.

# Introduction As world population’ life expectancy is increasing, promoting healthy behaviors and high health-related quality of life at old age, has become a major concern. Physical inactivity is a risk factor for several noncommunicable diseases and a leading cause of death for global mortality. Men have higher incidence and death rates of ischemic heart disease, diabetes mellitus during midlife, and of most cancers (not related to reproduction) than women. On the other hand, the health benefits of physical activity (PA) and exercise for older adults are well established. Those include decreasing the prevalence of common chronic diseases, namely, those previously mentioned, and cognitive decline, increasing physical function, mental health and, consequently, improving quality of life. Nonetheless, 58% of the European middle-aged and older men (over 55 years old) do not exercise or play any sports, mainly due to lack of time or motivation. Although the benefits of exercising outweigh the risks associated with being physically inactive, there is still a concern with the risk of injury. This especially in older, inactive and exercise/sport inexperienced populations. Consequently, finding alternative exercise programs that are effective, safe, and motivating enough to ensure long-term adherence to exercise for this population is essential. Exercise programs using recreational team sports, an adaptation of the official versions played as different small-sided games (SSGs), have been adopted as a motivating strategy to decrease physical inactivity and promote broad-spectrum health, physical fitness and well-being improvements in different populations. Recreational team sports have shown to have a major beneficial impact on cardiorespiratory fitness, which has been associated with the time spent with high heart rates (HR) during recreational team handball (TH) matches. Despite the high physical and physiological demands imposed by SSGs, the participants have reported moderate ratings of perceived exertion (RPE). TH is played by around 30 million of players worldwide being a particularly popular sport in Europe, namely in Portugal. If we add to this, the number of fans and supporters, the social capital emerging from this sport practice can be considered of great interest. Recreational TH has shown to be a high-intensity intermittent exercise mode, effective in improving physical fitness and cardiometabolic health (e.g. maximal oxygen uptake (VO_2max), blood pressure, aerobic performance, and blood lipid profile) in adult/middle-aged male former TH players, premenopausal overweight women, postmenopausal women and young adult men with no experience with the sport. In addition, it has proved to induce positive musculoskeletal adaptations (e.g. muscle mass, bone mineral content and density, and on bone metabolism) in young adult men and women, and also in postmenopausal women. SSGs are often used in recreational team sports exercise interventions as training tools. They are characterized by adapted rules compared to the official ones, such as number of players, size and shape of the court, allowed body contact, coach encouragement, among others, that influence internal and external load markers. Recreational TH as exercise mode has been implemented using different game formats, ranging from formal (7v7) to 3v3 formats. Notwithstanding the reported health benefits, no conclusive information exists on what is the most effective recreational TH game format to induce the reported adaptations. This issue is of great practical interest in the daily practice, as different number of participants may attend the training sessions, and also to guide future exercise interventions. In competitive soccer, the number of players per playing surface has been reported to impact the game demands. On other hand, in recreational soccer, high HRs were observed for different age, sex and social background groups, by playing SSGs, independently of number of players. Additionally, similar physical and physiological demands were reported for 21-year-old college students during recreational TH SSGs (4v4, 5v5 and 6v6). Nevertheless, the demands of this exercise mode have not yet been described for older populations, and the specific demands of other game formats frequently used in recreational TH-based exercise interventions (i.e., 5v5, 6v6 and 7v7) are still to be ascertained. In order to induce cardiovascular health improvements, average exercise intensity should range between 60–85%HR_max, and for optimal improvements it appears to be important to spend a significant amount of time of the training session with HRs above 85%HR_max. In fact, in a 12-week recreational TH intervention, post-intervention changes in VO_2max were largely correlated with the time spent with HRs \>90%HR_max. Recreational team sports intensities have shown to be in the range of these intensities in different populations. Nonetheless, this has not yet been described for middle-aged and older men. To optimize training load during recreational TH interventions, it is also important to describe the intensity of the different time periods within a proposed SSG. This to ascertain if a high intensity is maintained throughout the entire training session. Despite the interest of the maintenance of an effective exercise intensity during the matches, recreational TH internal and external load differences between match periods have only been addressed in one study with adult/middle-aged men with previous experience in TH. In that study similar cardiovascular load during the entire match duration (60 min) was reported. However, a decrease was observed in the second half in the frequency and distance covered in some of the locomotor categories, specific game actions and blood lactate (BL) values. Unfortunately, no study is currently available on the effects of game format duration on exercise intensity in over 60-year-old inactive men with no previous experience with recreational TH. Long-term adherence to the exercise programs is a major concern when planning exercise interventions. Recreational team sports have been considered as a social, fun and intrinsically motivational exercise mode, which are important characteristics for long-term adherence to exercise, namely, in the elderly male population. Therefore, it is of relevance to evaluate the self-reported fun levels (which reflect enjoyment) during recreational TH played as different game formats, as it may well be a positive affective response and an intrinsically motivation factor for the participation and adherence of an individual to an exercise program). Given the above reported premises, the aim of this study was to describe the acute physiological response, activity profile and fun levels of 5v5, 6v6 and 7v7 recreational TH game formats in over 60-year-old men when played over the official court (40x20 m). We hypothesized that 5v5 elicits higher cardiovascular (internal load) and activity profile (external load) demands due to the larger playing area, and, consequently, lower player density, and higher fun levels, as a result of higher involvement of the participants in the match. # Materials and methods ## Participants The recruitment process was done through advertisement in social media (Facebook), flyers/posters, and face-to-face meetings in local senior institutions. Seventeen male participants (67.4±3.3 (±SD) years; stature 168.2±5.5 cm; body mass 79.0±11.8 kg; fat mass 29.0±6.2%; body mass index 27.8±3.2 kg⋅m⁻²; peak oxygen uptake (VO_2peak) 27.9±4.1 mL⋅min^-1⋅kg⁻¹; systolic blood pressure 132±20 mmHg; diastolic blood pressure 78±8 mmHg; resting HR 69±12 b⋅min⁻¹; Yo-Yo intermittent endurance level 1 test (YYIE1) 480±256 m) with no previous experience with this sport, agreed to participate in this study. Inclusion criteria were: male participants, aged over 60 years, inactive (i.e., not complying with the PA guidelines for the last 6 months). Exclusion criteria were: participants with medical contraindications to perform moderate-to- vigorous PA or incapacity to run or grip a ball. All the participants were informed about the study purposes, risks and benefits and signed a written informed consent according to the Declaration of Helsinki. Ethical approval was provided by the local Institutional Review Board (CEFADE 19 2019). ## Experimental design All the participants were familiarized with the procedures involved in this study in the week preceding the data collection. Evaluations started with assessment of anthropometric variables, body composition, blood pressure, resting HR and VO_2max, in this order. Afterwards, on 2 separate days, the participants were tested for individual locomotor categories speed thresholds by performing each locomotor category at their individual speed, twice, over a 20 m distance, with 90-s recovery in-between, and for aerobic performance (YYIE1). Finally, internal and external load markers were monitored for each participant during 9 testing sessions. These sessions consisted of a standardized warm-up followed by recreational TH matches, 3 of each game format (5v5, 6v6 and 7v7), performed in a random order. These game formats were selected as they are typically used in recreational TH interventions that have shown to result in health improvements. There were 48 hours between each testing session and the participants were asked to refrain from intense PA in the 48h before the testing sessions. The court size was 40x20 m, resulting in \~80, 67 and 57 m² per player for 5v5, 6v6 and 7v7 game formats, respectively, to test the effect of player density. Players’ internal load was evaluated as exercise HR, BL concentration and differential RPE. Fun levels were also recorded at the end of all testing sessions. TH high-demanding game actions (i.e., jumps, throws, changes of direction, one-on-one situations and stops) and distances covered in selected locomotion categories were considered to profile participants’ external load. With the aim to account for inter-individual variability in external load, time- motion analysis was performed according to participants’ individual speed categories. All matches were performed during morning sessions and the participants wore t-shirts and shorts. The participants were hydrated at the beginning of the testing session and were allowed to drink water *ad libitum* to ensure the maintenance of proper hydration throughout the testing sessions. Each testing session comprised a 15-min standardized warm-up, consisting of running, coordination, flexibility, balance, and strength exercises, and three 15-min periods of recreational TH matches played either as 5v5, 6v6 and 7v7, interspersed by 2-min breaks. The warm-up started with back-and-forth progressive intensity runs in the TH court combined with articular movements for the upper and lower body during approximately 5 min. Then, the other 10 min aimed at flexibility and balance exercises for the upper and lower body and at strength exercises for the main muscles, namely, squats, frontal and side lunges, push-ups, and frontal and side planks. The mean HR during the warm-up of the testing sessions was 69%HR_max. After the warm-up, the second part of the testing sessions consisted of three 15-min periods of recreational TH matches. At every 3-min during the matches, the participants were instructed to change their positions assuring even rotation between the participants in the outfield and goalkeeper positions. There were no players’ substitutions during the matches and the participants were instructed to follow the basic TH rules. However, the balls used were smaller (47 cm circumference, GOALCHA, Fredericia, Denmark) and made of softer material than the official TH balls, and no body contact was allowed. This, to avoid injuries, since the participants had no experience with this sport. Only data from participants that performed all the three 15-min periods were analyzed. All testing sessions were instructed by a professional TH coach and physical education teacher and monitored by the research team. All the data collection and analysis were performed by the research team that comprised an experienced group of Sport Science, Physical Exercise and Health and Physical Education Teaching Master and PhD graduates, that had at least 5 years of experience with the testing procedures and analysis. ## Experimental procedures ### Anthropometric and health outcomes procedures Body mass (0.01 kg) and fat mass (%) were measured in a bioimpedance digital scale (Tanita Inner Scan BC 532, Tokyo, Japan) and stature (0.1 cm) was determined using a portable stadiometer (Seca 213, Hamburg, Germany), according to standardized protocols. Body mass index was calculated (kg·m⁻²). Blood pressure and resting HR measurements were assessed with an automatic upper arm blood pressure monitor (multiparameter patient monitor, Omron Z207, Kyoto, Japan). The participants were required to sit and rest for at least 5 min prior to the first blood pressure measurement. Two measurements were taken after 5 and 10 min of rest from the right arm, with the participants seated, in a relaxed position with their feet resting flat on the ground. The mean of the two measurements was considered for blood pressure analysis. If the two measurements differed by 2 mmHg or more, a third measure was taken. The lowest resting HR value was considered for analysis. To access VO_2max, the participants performed an incremental treadmill test until voluntary exhaustion (H/P/Cosmos, Quasar, Germany). For this purpose, the participants walked on the treadmill for at least 3 min for each stage. The participant’s HR was taken every min, and if the participant’s HR was not at steady state by the 3^rd min, the test continued at that same stage for another min. The first stage was considered the warm-up stage and was performed at 2.7 km⋅h^-1 and 10% inclination, the second stage at a 4 km⋅h^-1 and 12% inclination, the third stage at a 5.4 km⋅h^-1 and 14% inclination, the fourth stage at 6.7 km⋅h^-1 and 16% inclination and the fifth stage at 8 km⋅h^-1 and 18% inclination. The test was performed until voluntary exhaustion. The participants completed at least all the three first stages and the fifth stage was the highest reached. VO_2max and respiratory exchange ratio (RER) were determined by pulmonary gas exchange measurements (Oxycon Pro Metabolic Cart, Jaeger, careFusion, Germany) with the participants wearing a HR monitor (Polar Wearlink, Kempele, Finland). VO_2peak was considered as the highest 15-s mean value. The test ended at the participants’ voluntary exhaustion and the results were considered as VO_2peak if two of the following criteria were met: failure of VO₂ to increase with increased exercise intensity; RER ≥1.1; maximal HR (HR_max) ≥85% of age-predicted HR_max. The age-predicted HR_max was determined by the formula 208-(age x 0.7). Aerobic performance was evaluated by the YYIE1. The YYIE1 test was performed on the same indoor TH wooden floor court as the matches, after a 10-min warm-up consisting of running at different speeds and changes of direction. The test consists of 2x20 m shuttle runs with increasing speeds interspersed by 5 s of active recovery, with the participants walking around a cone placed 2.5 m behind the starting/finishing line. At set intervals, the running speed increases, starting at 8.0 km⋅h^-1. The total distance (m) covered during the test was recorded as test result for each participant. ### Internal load outcomes procedures One hundred and fifty-three HR recordings from 17 participants during the three game formats (9 matches per participant; 3 for each game format) were analyzed. Exercise intensity was assessed using HR monitors (Firstbeat Technologies Ltd., version 4.5.0.2, Jyväskylä, Finland). Selected HR zones were ≤60, 61–70, 71–80, 81–90, 91–100% HR_max. In this study, the individual HR_max was determined as the highest value reached either during the VO_2max test, YYIE1 or matches, according to a multiple testing approach. Capillary blood samples (30 μl) were drawn from the right earlobe to determinate BL concentrations (306 records from 17 players), at baseline (resting conditions) and at the end of the first and third period of the matches. For this analysis, a portable electroenzymatic lactate device analyzer (Lactate Pro 2 LT-1730, Arkray, Amsterdam, The Netherlands) was used. RPE is a practical, reliable and valid tool to estimate internal load and adding differential RPE (i.e., respiratory and muscular), may increase the sensitivity of internal load measurements. Therefore, differential RPE and fun levels (using a visual analogic scale; 0–10 AU) were registered at the end of all game formats. Participants were familiarized with the use of the considered psychometric scales in training sessions performed before this study. ### External load outcomes procedures Video recordings (153 evaluations; 17 participants) (SONY-DCR-SX65E, digital video camera recorder, Weybridge, United Kingdom) were collected for activity profile characterization using time-motion analyses. Players’ displacements were divided into eight locomotor categories: 1) standing still, 2) walking, 3) jogging, 4) fast running, 5) sprinting, 6) sideways medium-intensity, 7) sideways high-intensity and 8) backwards movement. High-intensity movements were the result of the sum of fast running, sprinting and sideways high-intensity categories. Individual speed thresholds were determined in order to account for the individual nature of the exercise intensity in each locomotor category. For this purpose, each participant was instructed to perform each locomotor category at their individual speed, twice, over a 20 m distance, with 90 s of recovery in-between. Telemetric photoelectric cells (Brower Timing System, IRD-T175, Utah, USA) registered the individual speeds. The distance covered in each category was calculated by multiplying each participants’ individual speeds by the time spent in each locomotor category. This study participants mean speeds in each locomotor category were 0 km⋅h^-1 for standing, 6 km⋅h^-1 for walking, 9 km⋅h^-1 1 for jogging, 12 km⋅h^-1 for fast running, 17 km⋅h^-1 for sprinting, 8 km⋅h^-1 for sideways medium-intensity, 10 km⋅h^-1 for sideways high-intensity and 8 km⋅h^-1 for backwards movements. Frequency of the selected high-demanding match actions, i.e., jumps, throws, stops, changes of direction and one-on-one situations, and total number of actions were registered via video-analysis of the matches. Accelerometer data was collected using Catapult MinimaxX S4 (MinimaxX S4; Catapult Sports, Canberra, Australia) in indoor mode with global positioning system units (GPS) technology in inactive state. Data was downloaded and processed using Catapult sprint Version 5.1.1 (Catapult Innovations, Canberra, Australia). Units were located in a specific vest on players’ upper back. The validity and reliability of the accelerometers have been described elsewhere. Player load (PL) (an estimate of physical demand combining the instantaneous rate of change in acceleration in 3 planes) variables were evaluated at a 100 Hz sampling rate. In this study, PL was presented as percentage of time spent in PL zones 0–0.1, ˃0.1–0.3, ˃0.3–0.6, ˃0.6–1.0, ˃1.0–1.5, ˃1.5–2.0, ˃2.0 and total accumulated PL. The matches were held under neutral temperature (20–22°C) and humidity conditions (50–60%). ## Statistical analysis Data was tested for normal distribution using Shapiro-Wilk test. Results are presented as means ± standard deviations (SD). Differences between game formats’ internal and external load variables were assessed by repeated measures analysis of variance (ANOVA) with Bonferroni post hoc test for multiple comparisons tests. Power calculations were performed to detect an effect size of 0.25 in a one-way ANOVA of repeated measures (within subjects). Using 3 groups and 3 measurements, with correlation between measures of 0.75, alpha of 5%, and power of 80%, 15 participants were needed. Practical significance was assessed by calculating Cohen *d* and interpreted as trivial (\<0.2), small (0.2–0.5), medium (0.5–0.8) and large (\>0.8). IBM Statistical Package for the Social Sciences (SPSS), Statistics for Windows, (Version 25.0, Armonk, New York, USA: IBM Corp.) was used for all analyses. Statistical significance was set at *p*≤0.05. # Results ## Internal load and fun levels during each game format Players’ internal load and fun variables for each game format (5v5, 6v6 and 7v7) are presented in and. No significant differences were found between game formats’ cardiovascular demands, RPE and BL, except for peak BL, which was significantly higher in 5v5 (5.6±2.1 mmol·l^-1) than in 7v7 (4.7±1.7 mmol·l^-1; *p =* 0.014, 95% CI: -1.4, -0.3, large). Players’ BL values during 5v5, 6v6 and 7v7 game formats are presented in. In all game formats, mean BL values increased significantly from baseline to the first period (5v5: *p\<*0.001, 95% CI: 1.2–2.7, large; 6v6: *p≤*0.001, 95% CI: 1.0–2.6, large; 7v7: *p≤*0.001, 95% CI: 0.8–2.2, large) and decreased significantly from the first to the third period (5v5: *p≤*0.001, 95% CI: -1.5,-0.6, large; 6v6: *p≤*0.001, 95% CI: -1.7,-0.6, large; 7v7: *p≤*0.001, 95% CI: -1.3,-0.5, large). ## Activity profile during each game format Players’ locomotor profile during 5v5, 6v6 and 7v7 game formats is presented in. During 7v7 game format, the frequency of walking was significantly lower than 5v5 (*p≤*0.001, 95% CI: 1.3–3.6, large) and 6v6 (*p≤*0.001, 95% CI: 1.8–4.7, large). Additionally, 6v6 percentage of time spent jogging (*p≤*0.001, 95% CI: -5.1, -1.4, large), and 5v5 and 6v6 total distance covered (*p =* 0.004, 95% CI: -5.9, -1.7, large; *p =* 0.034, 95% CI: -5.7, -0.8, large; respectively;) were significantly higher than 7v7. During 5v5 and 6v6, frequency (*p\<*0.001, 95% CI: -9.3, -3.7, large; *p≤*0.001, 95% CI: -13.9, -5.4, large; respectively), percentage of time spent (*p =* 0.007, 95% CI: -1.7, -0.4, large; *p\<*0.001, 95% CI: -2.9, -1.3, large; respectively), and total distance covered (*p =* 0.011, 95% CI: -138.0, -31.8, large; *p\<*0.001, 95% CI: -298.5, -111.5, large; respectively) in fast running were significantly higher than during 7v7. During 7v7 game format, high-intensity movements’ frequency was significantly lower than in 5v5 (*p\<*0.001, 95% CI: -10.6, -4.3, large) and 6v6 (*p≤*0.001, 95% CI: -14.6, -5.5, large). During 5v5 and 7v7, percentage of time spent (*p =* 0.030, 95% CI: 0.3–1.7, large; *p\<*0.001, 95% CI: -3.0, -1.3, large; respectively), and total distance covered (*p =* 0.020, 95% CI: 36.2–189.5, large; *p\<*0.001, 95% CI: -249.0, -113.3, large; respectively) in high- intensity movements were significantly lower than during 6v6. Moreover, 5v5 percentage of time spent in high-intensity movements was significantly higher than 7v7 (*p =* 0.005, 95% CI: -1.8, -0.5, large). Players’ high-intensity actions frequency during the matches is presented in the. During 5v5 and 6v6, the number of throws (*p\<*0.001, 95% CI: -4.7, -2.1, large; *p =* 0.031, 95% CI: -2.3, -0.4, large; respectively), stops (*p =* 0.017, 95% CI: -4.0, -0.8, large; *p =* 0.002, 95% CI: -3.3, -1.1, large; respectively) and total actions (*p =* 0.003, 95% CI: -13.2, -4.0, large; *p =* 0.017, 95% CI: -9.1, -1.8, large; respectively) was significantly higher than during 7v7 game formats. During 5v5, 6v6 and 7v7 game formats, players’ percentage of time spent in 0.0–0.1, ˃0.1–0.3, ˃0.3–0.6, ˃0.6–1.0, ˃1.0–1.5, ˃1.5–2.0 and above 2.0 PL zones were 17–19%, 40–41%, 16–17%, 8%, 10–11%, 5–6% and 1%, respectively, and the total PL accumulated during the matches ranged between 288 to 310. The number of low, medium, high, and total accelerations during the game formats, ranged between 13–17, 7–9, 9–11 and 29–36 and the number of low, medium, high, and total decelerations ranged between 8–10, 4–5, 2–4 and 14–18, respectively. No significant differences were found between the game formats in PL zones and in low, medium, high, and total accelerations and decelerations variables. ## Differences between match periods During 5v5, absolute and relative mean HR increased from the first to the second periods (*p =* 0.003, 95% CI: 1.5–4.9, large; *p =* 0.003, 95% CI: 0.9–3.0, large; respectively), remaining unaltered in the third period. Absolute and relative mean HR, in 6v6 (*p =* 0.035, 95% CI: -6.2, -0.9, medium; *p =* 0.034, 95% CI: -3.6, -0.5, medium; respectively) and 7v7 (*p =* 0.010, 95% CI: -9.2, -2.2, large; *p =* 0.008, 95% CI: -5.4, -1.4, large; respectively), decreased as the match time progressed, showing significant differences between the first and third periods. During 7v7 game format time spent above 80%HR_max significantly decreased from the first to the second (*p =* 0.027, 95% CI: -22.3, -3.7, medium) and third (*p =* 0.005, 95% CI: -27.3, -7.9, large) period, and from the second to the third period (*p =* 0.027, 95% CI: -11.4–2.0, medium). Total number of high-intensity game actions decreased from the first to the second period for 6v6 (*p =* 0.016, 95% CI: -6.6, -1.4, large) and 7v7 (*p =* 0.020, 95% CI: -4.1, -0.8, large), and from the first to the third period for 6v6 (*p\<*0.001, 95% CI: -8.1, -3.8 large). Distance covered in fast running (*p\<*0.001, 95% CI: -4.7, -2.5, large) and sprinting (*p =* 0.032, 95% CI: -0.5, -0.1, large) decreased from the first to the third period for 6v6. # Discussion The aim of this study was to describe the acute physiological response, activity profile and fun levels of 5v5, 6v6 and 7v7 recreational TH game formats in over 60-year-old men. This to provide physical exercise and sport professionals, evidence on the internal and external load characteristics of the game formats analyzed, that will allow them to make informed decisions according to the defined purposes for the training sessions. The main findings of this study were that game format had no significant impact on match internal load, although a tendency was observed for higher demands in 5v5 and 6v6 than in 7v7. Significant differences were evident in the external load variables, with 5v5 and 6v6, showing a higher number of high-intensity movements and total high-intensity game actions when compared to 7v7. ## Internal load and fun levels during the game formats During recreational TH training sessions using SSGs, mean HR is typically reported to be within 76–85%HR_max. In the present study, mean HR values for the three game formats were lower (76–77%HR_max) than values observed in young adult men and women, premenopausal women and adult/middle-aged men (81–85%HR_max), but equal to or higher than those reported for postmenopausal women (76%HR_max) enrolled in a TH intervention study that resulted in cardiovascular improvements. In fact, these values are in the range of the vigorous exercise intensity threshold (60–85%HR_max) proposed to promote cardiovascular improvements. Peak HR values were lower than those reported in studies using recreational soccer SSGs with elderly males (84–86 vs 99%HR_max, respectively) and, consequently, the percentage of time spent above 90% HR_max was also lower (4 vs 48% of total match time). However, studies using recreational floorball and recreational TH with similar age groups showed results in line with our study. Additionally, it is worth noting that in the present study we assessed the participants’ HR_max making use of a multiple approach in order to have an as accurate as possible value. Having an accurate HR_max is of great practical interest as time spent above 90%HR_max was reported to be related to improvements in cardiorespiratory fitness in recreational TH. Given that, the above differences in exercise HR may be the result of unsuitable HR_max assessment. No significant differences were found in exercise HR between the three game formats. This is in accordance with a recent study comparing 4v4, 5v5 and 6v6 game formats using the same pitch size (40x20 m) for young (20.8±1.1 years) active college students with no competitive experience in TH. However, our study reported a significant decrease in 6v6 and 7v7 game formats’ mean and peak HR in the third comparing to the first match period, while mean HR in 5v5 significantly increased from the first to the second period and was then maintained during the last 15-min period of the matches. This decrease in intensity was also shown in the activity profile. The main relevance of these results is that 5v5 game format seems to be more efficient in maintaining the cardiovascular load throughout 45-min matches, perhaps due to the greater involvement in the game imposed by the lower number of players. Nonetheless, exercise intensity during all match duration in the three game formats was within the range (60–85%HR_max) proposed for cardiovascular improvements. The practical implication of this study HR results is that they were in line with the results from other studies that used recreational team sports, especially TH, that reported cardiovascular improvements after 12-16-week interventions. Thus, future studies should address the role of training volume and intensity on practitioners’ health and fitness. In this study, peak BL was significantly lower in 7v7 than in 5v5, meaning that in 5v5, the participants, may achieve higher anaerobic intensities. Mean and peak BL concentrations (3.6±1.4 and 4.7±1.7 mmol·l^-1) were in line with the values reported in former TH players, during 7v7 matches (3.6±1.3 and 4.2±1.2 mmol·l^-1). Significantly higher values were found in all game formats in the first period comparing to baseline conditions and in 5v5, in the third period comparing to baseline. This is in line with a study involving former TH players playing recreational matches. Additionally, a decrease in BL values was found from the first to the third period in all game formats, which again, is in line with what has been shown for recreational TH players with previous experience with the sport, evidencing an intensity decrease from the beginning to the end of the recreational TH matches. This decrease is in accordance with the mean and peak HR decrease shown from the first to the third match period, as well as with the decrease in the number of high-intensity game actions during the match in 6v6 and 7v7 game formats, although not observed for 5v5. In our study, no differences were found in RPE between game formats, with the intensity being perceived as strong-to-very-strong (6.1–6.7 AU). These values were lower than those reported for former TH players playing the recreational version of this sport. However, higher than observed for postmenopausal women (4.8, AU, playing 5v5 and 6v6 matches) and for male college students (3.9 AU, playing 6v6 matches) performing recreational TH. The absence of differences in RPE when playing 5v5, 6v6 or 7v7 may be explained by participants’ lack of experience with this sport and by a spontaneous adjustment to game format demands as shown by exercise HR consistency across the proposed game formats. It is important to highlight the very high fun levels reported by the participants (9 out of 10) while playing recreational TH, since enjoyment is a key factor to increase motivation and assure long-term adherence to an exercise program. Also, the perceived high rate of fun in this population may mask the high metabolic, musculoskeletal, and cardiorespiratory strain during training interventions. ## Activity profile during the game formats Significant differences were found between the game formats’ external load, namely high-intensity locomotor activity variables. Standing frequency was significantly higher in 5v5 than in 6v6 and 7v7, which may be related to more stops being needed to recover since there are less players involved in the match. The walking frequency was significantly higher in 7v7 than in 5v5 and 6v6 and the jogging frequency showed no significant differences between the game formats. These variables were the ones in which the participants spent more time during the matches (76–77% of total match time) in all game formats. Furthermore, the frequency of fast running movements was significantly higher in 6v6 than in 5v5 and 7v7. Frequency of sprints, resulted higher in 5v5 than in 7v7. Moreover, the frequency of high-intensity movements was significantly higher in 5v5 and 6v6 than in 7v7. These results show that 5v5 and 6v6 may induce higher load on muscles and bones than 7v7. Standing time was significantly higher in 5v5 than in 6v6 and 7v7. When comparing to former TH players playing the recreational version of the sport, our participants’ standing time was higher (8–10 vs 4%), which may be related to their lower physical fitness (27.9±4.1 vs 40.2±7.0 mL⋅min^-1⋅kg⁻¹). The 7v7 game format revealed to promote significantly more jogging and less high-intensity distance covered than the other game formats. Interestingly, match time spent with high-intensity movements (2–4%) was similar to that reported in elite (4%) and recreational TH players with previous experience in the sport (6%). Time spent at high-intensity seems to be a good indicator when the purpose is to achieve cardiovascular adaptations. However, caution should be taken when comparing these results with those studies, due to differences in speed thresholds. The 5v5 game format showed a significantly higher number of throws and stops than 6v6 and 7v7, suggesting a greater individual game involvement due to the reduced number of players. The practical implication of these results is that 5v5 and 6v6 may be better options than 7v7, when the aim is to induce musculoskeletal improvements in this population. Moreover, when playing 5v5 and 6v6, the participants performed higher number of specific TH game actions, which leads to a higher participation in the training session, and consequently, may result in higher motivation. We had hypothesized that 5v5 game format would elicit higher cardiovascular (internal load) and activity profile (external load) demands due to the larger playing area, and, consequently, lower player density, and higher fun levels as a result of higher participant involvement in the match. However, changing the number of players (5v5, 6v6 and 7v7) in the same pitch size (40x20 m) did not result in significant differences in majority of these variables. The practical application of these results is that all the three game formats elicited high loading in over 60-year-old inactive men and therefore, all can be recommended as options for organizing recreational TH. In summary, this study results may guide physical exercise professionals and/or TH coaches, on the best practices when using recreational TH as exercise mode to promote physical fitness and health of older populations. Although in this study we analyzed the game formats typically used in recreational TH-based exercise interventions (i.e., 5v5, 6v6 and 7v7), a study limitation is the fact that we did not study other game formats’ demands, namely 4v4 or 3v3, as there could be differences in internal and external load in comparison to other populations. Additionally, other contextual variables should be addressed in the future, namely court dimensions and comparing indoor vs outdoor pitches, as this type of exercise program may be implemented in different environments. Future studies with more participants may be used to elucidate whether there are some minor advantages in relation to intensity and fun scores by using 5v5 and 6v6 in comparison to 7v7, in a 40x20 m TH court, for over 60-year-old inactive men. Future research should also test the training effects of the proposed different recreational TH formats on participants’ health and physical fitness. # Conclusions Recreational TH internal load demands are similar either played as small-sided (5v5, 6v6) or formal game formats (7v7), in the same pitch size (40x20 m), and are within the range to induce cardiovascular adaptations. This, across match time periods (i.e., 3x15-min). Higher frequency of high-intensity game actions was found in 5v5 and 6v6. Accordingly, these game formats may be better options when the purpose is to induce musculoskeletal improvements in this population. The higher number of total actions and throws found in 5v5 and 6v6 may also reveal to be of practical importance as a greater involvement may lead to a higher level of motivation and therefore, to higher fun levels and long-term adherence to the exercise program. Nonetheless, recreational TH practice is a highly motivational activity, whatever the chosen game format. From a practical point of view, this study results suggest that recreational TH can be a valid exercise option to promote health improvements in over 60-year- old men. A multiple game format approach may be used in recreational TH interventions to provide training variety and training sessions should last up to 60 min. Additionally, considering the very high fun levels reported during recreational TH matches and that lack of motivation to exercise is a major hurdle, future intervention studies using this exercise mode for this population are warranted. This study is part of the Handball4Health project, which is supported by the Portuguese Handball Federation, the European Handball Federation, Porto Sports Medicine Center (IPDJ, IP) and Gaia City Hall. We would like to thank all the participants for their committed participation. We would also like to express our gratitude to the members of the Handball4Health project, and the students from the University of Maia and the Faculty of Sport, University of Porto, who collaborated in the data collection and provided technical assistance. [^1]: The authors have declared that no competing interests exist.

# Introduction Quorum sensing (QS) is the process through which bacterial cells communicate enabling unicellular populations to coordinate their response to an external stimulus as a function of population density, for a review see. Gram negative bacteria such as *Pseudomonas aeruginosa* employ *N*-acylhomoserine lactone (AHL) QS signal molecules. *P. aeruginosa* is an opportunistic human pathogen responsible for causing infection in immune compromised individuals and is the leading cause of morbidity and mortality in cystic fibrosis patients. *P. aeruginosa* employs an AHL-dependent QS system employing two LuxR/I pairs (LasR/I and RhlR/I) where LasR and RhlR are transcriptional regulators which respond to the AHLs, *N*-(3-oxododecanoyl)homoserine lactone (3-oxo-C₁₂-HSL) and *N*-butanoylhomoserine lactone (C₄-HSL) respectively. These are produced via the LasI and RhlI AHL synthases respectively. Although LasI directs the synthesis primarily of 3-oxo-C₁₂-HSL, the analogues 3-oxo-C₁₀-HSL and 3-oxo-C₁₄-HSL are also produced, albeit at much lower levels. The *las* and *rhl* systems directly or indirectly regulate over 10% of the *P. aeruginosa* genome and are organized as a hierarchy in which LasR/3-oxo-C₁₂-HSL drives the expression of *lasI* (so constituting a positive feedback loop) as well as *rhlR* and *rhlI*. The *las/rhl* QS system plays a key role in controlling virulence factor production, biofilm maturation, swarming motility and the expression of antibiotic efflux pumps. *P. aeruginosa* AHLs have been detected *in vivo* during human infections. They are readily detectable in sputum from cystic fibrosis patients although determining their physiological QS concentration range is complicated as a consequence of their susceptibility to alkaline and/or enzymatic hydrolysis. Apart from modulating bacterial gene expression, AHLs such as 3-oxo-C₁₂-HSL (but not C₄-HSL) antagonize growth and virulence factor production in Gram positive bacteria such as *Staphylococcus aureus*. 3-oxo-C₁₂-HSL may also contribute directly to the outcome of host-pathogen interactions. 3-oxo-C₁₂-HSL influences smooth muscle contraction in blood vessels, exerts a marked bradycardia and modulates the junctional integrity and paracellular permeability of epithelial cells. It also modulates host inflammatory and immune responses, (reviewed in). For example, at concentrations below 10 µM, 3-oxo-C₁₂-HSL reduced lipopolysaccharide (LPS)-induced production of the pro-inflammatory cytokine IL-12 by monocytes whereas pro-inflammatory or pro-apoptotic effects were apparent at higher concentrations in both macrophages and neutrophils. The proliferation and function (cytokine production) of both mitogen-stimulated (e.g., and antigen- stimulated) T lymphocytes as well as antibody production by B lymphocytes, are inhibited by 3-oxo-C₁₂-HSL. Smith *et al.* reported that 3-oxo-C₁₂-HSL induced activation of the pro-inflammatory signaling components Cox-2 and NF-κB in transformed cell lines however, this does not occur in primary cells in the absence of LPS neither does 3-oxo-C₁₂-HSL act via known pathogen pattern recognition receptors. In the absence of LPS, 3-oxo-C₁₂-HSL does induce phosphorylation of mitogen-activated protein kinase (MAPK) p38, which could modulate cytokine production and also potentiate TNFα-induced poly(adenosine 5′-diphosphate- ribose) (PARP) cleavage, a biochemical marker of apoptosis. The direct target(s) of 3-oxo-C₁₂-HSL in mammalian cells have yet to be fully characterized. Since 3-oxo-C₁₂-HSL enters mammalian cells and retains intracellular activity, the most likely receptor(s) for 3-oxo-C₁₂-HSL has been suggested to be intracellular. In this context, Jahoor *et al.* obtained evidence to suggest that 3-oxo-C₁₂-HSL may bind to at least two isoforms (PPARγ and PPARβ/δ) of the peroxisome proliferator activated receptors (PPARs). These belong to the nuclear hormone receptor family which bind a range of endogenous and exogenous lipids and play roles in inflammation and lipid metabolism. 3-oxo-C₁₂-HSL may therefore modulate NF-κB signaling via the direct interaction with PPARs. Using an affinity matrix, Seabra et al. identified calprotectin as a target although this calcium binding protein is unlikely to be the primary receptor for 3-oxo-C₁₂-HSL. In order for QS signal molecules to interact with intracellular components they must first interact with the cell membranes. Structure-activity assays of 3-oxo-C₁₂-HSL have revealed that optimal immune modulatory activity in a mouse splenocyte proliferation assay in common with QS in *P. aeruginosa* requires a C₁₁ to C₁₃ acyl chain, an intact homoserine lactone ring and L-configuration at the chiral centre, suggesting lipophilicity is important for QS immunosuppressive activity. Based on these observations and given the broad biological activity of 3-oxo-C₁₂-HSL in both pathogens and eukaryotic cells (particularly its action on leukocytes), we have explored the interactions of long chain AHLs with simple membranes and T lymphocytes. The potential interactions of QS signal molecules with many types of membrane have so far largely been ignored, although a recent study by Lowery *et al.* indicated they may have effects on bacterial membrane permeability. This study presents evidence that 3-oxo-C₁₂ HSL and two close structural analogues, 3-oxo-C₁₀-HSL and 3-oxo-C₁₄-HSL are capable of insertion into the lipid bilayer in both artificial membrane and Jurkat T-lymphocyte cell systems. Although the AHL concentrations used are higher than those reported in some human host samples e.g. sputum from cystic fibrosis patients, the latter are likely to be an underestimate since local AHL concentrations up to 600 µM have been detected in culture supernatants of *P. aeruginosa* biofilms grown *in vitro*. In order to determine whether these compounds are capable of directly interacting with membranes they were studied using the fluorescent probe di-8-ANEPPS. This technique yields binding information such as affinity and overall binding capacity as well as whether addition of AHLs impacts on the membrane organization. The roles of QS acyl chain length (C₁₀ to C₁₄) and cholesterol in membrane interactions are also investigated and the potential roles of membrane microdomains (or rafts) in immune modulation are discussed. # Results ## AHL Immunosuppressive Activity is Dependent on membrane affinity (Lipophilicity) The concentrations of AHLs required to inhibit 50% proliferation (IC₅₀) of human peripheral blood mononuclear cells (PBMCs) was obtained from Chhabra et al 2003. These data are plotted against a calculated LogP value in, (calculated using the ACD/I-Lab Web service (ACD/LogP 12.0)) based on molecular structure. LogP is a measure of molecular lipophilicity and is frequently used in the pharmaceutical industry as an indicator of the likely bioavailability of a drug molecule. indicates that a strong negative correlation (Pearsons r = −0.895, *p* = 0.016) exists between membrane affinity and immunosuppressive activity. The observation that AHLs with increasing chain length (and therefore lipophilicity) impede cell proliferation to a greater extent provides the first fully quantitative evidence to suggest that the hydrophobic properties of AHLs play an important role in their ability to inhibit immune cell function. This will most likely manifest as an elevated ability to interact with biological membranes, influencing membrane electrical potentials which have been previously reported to effect indirectly membrane protein function. Alternatively AHLs with greater lipophilicity will interact more readily with any hydrophobic binding pockets that exist within receptor systems (i.e. either in a membrane or in a soluble protein). The aim of this investigation is to provide evidence which of these mechanisms AHLs elicit their well documented immunomodulatory effects. ## Interactions of AHLs with Phospholipid Membrane Vesicles; Effects on Membrane Dipole Potential Biological membranes, in addition to the well documented transmembrane potential also possess two lesser known electrical potentials termed the electrostatic membrane surface potential and the membrane dipole potential (for an extensive review of this trinity of membrane potentials see). The membrane dipole potential has a magnitude in the region of 300 mV and arises from the orientation of dipoles at and just under the membrane surface. On insertion of a molecule of interest into biological membranes, this can cause a perturbation in this potential which can be detected in both artificial and cellular membrane systems using the electrochromic probe di-8-ANEPPS. On this basis it is shown in that addition of 3-oxo-C₁₄-HSL, 3-oxo-C₁₂-HSL and 3-oxo-C₁₀-HSL to Phosphotidylcholine_100% and Phosphatidylcholine_70% Cholesterol_30% membranes vesicles led to a red shift in the excitation spectra of di-8-ANEPPS. A red shift (with a minimum of ∼440 nm and a maximum of ∼520 nm) is indicative of the ligands acting to decrease the membrane dipole potential upon insertion into the membrane. This finding is consistent with our previous work reporting that addition of reagents known to decrease the membrane dipole potential, give rise to di-8-ANEPPS difference spectra of similar profiles to those reported here. Changes in the membrane dipole potential can be tracked over time using the ratio of di-8-ANEPPS fluorescence at 460 nm and 520 nm excitation and a fixed emission (termed R(460/520)), which is sensitive solely to variations of the local electric field due to dipolar molecular properties. Upon titration of the AHLs into these artificial membranes a concentration dependent decrease in membrane dipole potential was observed as shown in. These data were plotted as the incremental change of di-8-ANEPPS fluorescence versus the concentration of AHL added and fitted to various ‘binding’ models (eq. 1 & 2) as shown in. Such observations are consistent with the interaction and insertion of AHL molecules with the membrane vesicles. All ligands were found to interact with both Phosphotidylcholine_100% and Phosphatidylcholine_70%Cholesterol_30% membranes via a simple hyperbolic (i.e. non-cooperative) binding mechanism, compare the dissociation constant (K_d) and saturation point (B_max) obtained in each case. Overall these figures depict that as acyl chain length increases it leads to a decrease of the observed the K_d, suggesting that longer chain AHLs have a higher membrane affinity than the shorter chain variants. This was found to be the case for both Phosphotidylcholine_100% and Phosphatidylcholine_70%Cholesterol_30% membranes. Titration of the AHLs into membranes containing 30% cholesterol did not result in a significant change in K_d, however the binding capacity of 3-oxo-C₁₂ HSL and 3-oxo-C₁₄ HSL (i.e. saturation in fluorescence units) of Phosphatidylcholine_70% Cholesterol_30% membranes was significantly greater than for Phosphatidylcholine_100% membranes (p\<0.05), suggesting that QS compounds may be accumulating in membrane microdomains present in Phosphatidylcholine_70%Cholesterol_30% membranes. Addition of 200 µM COOH-3-oxo-C₁₂ HSL to artificial membranes was found to cause only a nominal red shift in the di-8-anepps excitation spectrum. This shift was substantially less than that observed for 3-oxo-C₁₂ HSL. Titration of COOH-3-oxo-C₁₂ HSL into liposomes exhibited a dramatically lower effect on the dipole potential than 3-oxo-C₁₂ HSL and fit poorly to both hyperbolic and sigmoidal models. This observation was expected as COOH-3-oxo-C₁₂ HSL has a projected LogP substantially lower than 3-oxo-C₁₂ HSL. ## Interactions of AHLs with T-Lymphocytes; Effects on Membrane Dipole Potential In the previous section it was shown that addition of AHLs to simple phospholipid membranes resulted in a decrease in the membrane dipole potential, which is indicative of these compounds inserting into artificial membranes. As a result, it was of interest to study their interactions with T-Lymphocytes which we utilize as a model for the interaction of quorum molecules with their prospective eukaryotic host cell systems. indicates that addition of 3-oxo-C₁₄-HSL, 3-oxo-C₁₂-HSL and 3-oxo-C₁₀-HSL to this model cell system resulted in a red shift in the excitation spectra of di-8-ANEPPS (all normalized to controls containing the DMSO vector but not the QS) similar to those observed previously with the artificial membrane systems. This suggests that the AHLs used in this study were capable of insertion into the plasma membrane of Jurkat T-lymphocytes leading to changes in the dipole potential. Resazurin reduction-based cell viability assays were conducted as it is conceivable that treatment of the cells with AHLs may affect viability. It was found however, that no significant effects on cell viability take place at any of the concentrations utilized in our studies (data not shown). indicates that upon titration of di-8-ANEPPS labeled Jurkat T-lymphocytes with AHLs, a similar decrease in the R(460/520) parameter was observed as for titration with the artificial membrane systems. These data were plotted and fitted as before and are shown in. It was found that 3-oxo-C₁₄-HSL exhibited a strong interaction with the cell membrane with both a greater binding capacity (1.40±0.11 compared to 0.86±0.08, *p* = 0.02) and dissociation constant significantly less than 3-oxo-C₁₂-HSL (39 µM±6 µM compared to 153 µM±38 µM, *p* = 0.04) as determined by two tailed T test. The latter exhibited a more complex binding reaction that was poorly described by Eq. 1. Eq. 2 however was found to be able to describe the binding isotherm and this model indicates that cooperativity appears to be occurring. The cooperativity index (sometimes referred to as the Hill coefficient) for these studies was found to be 1.87±0.27. There are several possible interpretations of this finding, such as that two 3-oxo-C₁₂-HSL molecules come together on/in the membrane to promote their respective interaction. There are, however other explanations that accommodate this behavior of which the most likely is that the membrane is modified by the presence of the AHLs and this affects the subsequent binding of further molecules (see below). 3-oxo-C₁₀-HSL exhibited only a very slight interaction which fitted neither binding model to a satisfactory degree (R² = 0.57). indicates that pre-treatment of Jurkat T-lymphocytes with 160 µM 3-oxo-C₁₂ HSL caused a significant reduction in the B_max of P-glycoprotein (P-gp) for Saquinavir, in comparison to cells treated with equivalent volumes of 0.5% DMSO solvent (1.53±0.04 compared to 1.92±0.03, *p* = 0.002) in addition to a significant increase in K_d (39 µM±1 µM compared to 34 µM±1 uM, *p* = 0.007). A slight but insignificant decrease in binding cooperativity was also observed. Subsequent work has shown that pre- treatment of lymphocytes with saquinavir has no significant influence on the interaction of 3-oxo-C₁₂ HSL with Lymphocytes (Data not shown). These findings may therefore suggest that 3-oxo-C₁₂ HSL mediated lymphocyte membrane dipole potential modulation has indirectly acted to change the activity of P-glycoprotein. # Discussion ## The Interactions of AHLs with Artificial Membrane Systems The present study outlines the interactions between long chain AHLs and a number of membrane types including live cells from which it is possible to draw several important conclusions. First 3-oxo-C₁₄- HSL, 3-oxo-C₁₂-HSL and 3-oxo-C₁₀-HSL as shown in, were all observed to interact with PC_100% and PC_70%Cholesterol_30% artificial membrane systems. These model membranes are taken to represent the most abundant lipid types found in human cell types; their interaction profiles were found to correspond to simple hyperbolic (saturatable) binding models ( &). Based on the well characterized responses of the fluorescent probe we utilize in this study, we are able to conclude from our observations that AHLs are capable of direct insertion into biological membranes in the micromolar concentration range. ## The Interaction of AHLs with Cell Membranes One virtue of using the di-8-ANEPPS probe is that it may also be implemented with live cells as well as with our model membrane systems. Analogous studies with T cells, therefore, were found to indicate that all three AHLs are also capable of insertion into the cell membrane in the micromolar concentration range. Previous work examining the immunomodulatory activity of AHLs has suggested that insertion into T-lymphocyte membranes is unlikely to be responsible for immunosuppressive effects of 3-oxo-C₁₂ HSL at concentrations of less than 10 µM. This finding is corroborated by this study which suggests little membrane association of 3-oxo-C₁₂-HSL with T-Lymphocytes at concentrations less than 30 µM. Above this concentration, however, AHLs appear to have the ability to become inserted into membranes. These findings are also in agreement with suggestions that 3-oxo-C₁₂-HSL acts via multiple signaling pathways. The interaction of 3-oxo-C₁₂-HSL with T-Lymphocytes was found to be characterized most closely by a cooperative binding mechanism. This suggests that two molecules may come together on/in the membrane resulting in a fruitful binding/insertion complex. As this effect was not observed in the artificial membrane systems this observation implies that either 3-oxo-C₁₂ HSL acts to modify the more complex cell membrane which effects the interaction of additional levels of 3-oxo-C₁₂ HSL or that there exists an as yet unidentified 3-oxo-C₁₂ HSL receptor on the cell surface. The former possibility is less likely as the cell membranes possess both PC and cholesterol at about the levels we use in this study. The latter possibility however, is consistent with previous suggestions of the existence of a membrane receptor system for AHLs. To date the putative eukaryotic receptors identified (PPARs and calprotectin), are both intracellular however the existence of a membrane receptor has been suggested by Shiner *et al.* although its identity or nature has so far remained elusive. Our present work therefore constitutes support for the existence of a eukaryotic AHL membrane receptor which is active at concentrations up to the micromolar range. The concentrations of AHLs reported to influence membrane dipole potential in this study are slightly greater than those reported to cause significant inhibition of immune cell proliferation. One possible suggestion that reconciles this observation is that the immunosuppressive activity of AHLs is predominantly receptor mediated. The observation that the d-isomer of 3-oxo-C₁₂ HSL retains substantial immunomodulatory activity, however, implies that indirect membrane interactions also play an important role in the activity of these molecules. In addition, Kaufmann et al. reported more recently that many AHLs are capable of decomposing to form substantial quantities of tetramic acid breakdown products over the prolonged incubation times used (24 to 48 hours in many proliferation assays). The interactions of these tetramic acid breakdown products are as yet poorly understood and may offer an additional explanation for the greater than expected immunosuppressive activity of these molecules. ## The Effects of AHL chain length on the Interaction with Membranes The AHL acyl chain length was also studied with an increasing chain length observed to augment significantly the membrane binding affinity in both artificial and T-Lymphocyte membrane systems. This behavior was anticipated as an increasing AHL acyl chain length raises the molecular lipophilicity as indicated by the projected octanol/water partition coefficient. This parameter is known as Log P and often used by developmental pharmacologists to predict the lipophilicity of drug molecules which has an important bearing on their bioavailability. Classification of the LogP of the AHLs is of interest as it has been previously reported that increasing AHL acyl chain length from C₈ to C₁₄ increased its immunomodulatory activity in the micromolar range and this was found to negatively correlate with LogP. Overall these studies provides further evidence that the membrane interactions of AHLs may play an important role in inter-kingdom signaling including the immune modulatory effects of AHLs. indicates that modification of the acyl chain to introduce a negative charge (COOH-3-oxo-C₁₂ HSL, projected pKa = 4.78±0.1 calculated using ACD/pK_a algorithm) significantly reduces the ability of the compound to insert into artificial membranes and modulate the dipole potential. This result was predicted as the LogP of COOH-3-oxo-C₁₂ HSL is substantially lower than 3-oxo-C₁₂ HSL and provides evidence that changes in the di-8-anepps spectra observed occurred as a result dipole potential modulation, when AHLs inserted into membranes. ## The Interaction of AHLs with Membrane Microdomains A further sophistication to the systems that we investigate in the present study includes preparation of membranes that we have previously characterized and are known to exhibit microdomains. Such microdomains are rich in cholesterol and appear to be similar to cellular structures known as membrane rafts. The presence of cholesterol containing microdomains on titration of artificial membrane systems with 3-oxo-C₁₂-HSL and 3-oxo-C₁₄-HSL was observed to increase significantly the saturation point in comparison to membranes that were made up exclusively of phosphatidylcholine. This suggests that the AHLs used in this study may accumulate in cholesterol containing microdomains which leads to a decrease in the membrane dipole potential. The view that membrane microdomain localisation modulates the behavior of receptor signaling is held by research groups in addition to our own , and offers an interesting new possibility regarding the activity of receptor controlled signaling systems. There is a growing body of evidence suggesting that the dipole potential, which is higher in cholesterol rich microdomains than the surrounding disordered membrane, is capable of modulating membrane protein structure which may have implications for raft associated cell signaling. The decrease in membrane dipole potential observed in this study could therefore stimulate the indirect activation of signaling pathways by indirectly instigating a conformational change in transmembrane receptor structure. This principle could apply to both bacterial and eukaryotic organisms and also offers an explanation as to why a primary AHL site of action has so far remained elusive. ## Towards Defining Mechanisms for Inter-kingdom Signaling; Modulation of the Properties of a Membrane Protein Receptor System by AHL-dependent Changes of the Membrane Dipole Potential The final section of this study is directed towards defining a mechanism by which AHLs may elicit an inter-kingdom signaling process. In other words we seek to demonstrate how a prospective host cell (e.g. T cells) system may respond to the presence of AHLs at appropriate concentrations as an illustration of how such mechanisms may operate. In the present case we take advantage of the fact that the binding of Saquinavir, a HIV-1 protease inhibitor to the membrane protein P-gp is influenced by changes in the membrane dipole potential. The activity of this protein has been shown to have a dependence on the lipidic content of the membrane, particularly the sterol content and appears to be microdomain associated. Previously it has been shown that modification of the membrane dipole potential, through manipulation of sterol concentrations, can influence the activity of P-gp. In the present paper we have shown that pre- treatment of T-lymphocytes with micromolar concentrations of 3-oxo-C₁₂ HSL causes a significant reduction in the binding capacity of saquinavir and increase in dissociation constant. These findings are consistent with our previous work (see Asawakarn *et al.*) that indicated that the membrane dipole potential plays an important role in modulating ligand- membrane interactions. We demonstrated that disruption of cholesterol containing microdomains resulted in a decrease in the binding capacity of P-gp for saquinavir. As a similar trend was observed in the present work, this provides further evidence to support the hypothesis that on insertion into membranes, AHLs may act to disrupt cholesterol containing membrane microdomains. This could have important consequences on the activity of a plethora of raft-dependant membrane proteins by no means limited to P-gp, although it is interesting that P-gp activity is also associated with anti-microbial therapy. P-gp belongs to the ABC-transporter protein superfamily and is known to play a crucial role in the development of multiple-drug resistance (MDR). The ability of AHLs to modulate PgP activity would be advantageous to the bacteria and increase the immune systems susceptibility to cytotoxic agents also released by the pathogen. In addition, as MDR poses a significant problem for the treatment of conditions including cancer, the search for potential inhibitors has been intensive. Our present paper provides evidence of a membrane based mechanism of action of AHLs on immune cells and suggests that through their membrane interactions can act to modulate P-gp activity. AHLs or their derivatives may therefore provide a novel drug template family for the prevention of MDR. The concentration of AHLs required to modulate immune cell function in this way suggest that only cells in close proximity to the bacterial biofilm would be subject to immune modulation through these processes. This observation is in agreement with predictions by Teplitski et al 2010 who suggests the existence of such an AHL gradient, up to micromolar concentrations under physiological conditions. This could be of evolutionary advantage to *P. aeruginosa* as under this system only immune cells which pose a direct threat to the bacterial population are affected, allowing more remote immune cells to retain their function, potentially eliminating pathogens which would otherwise take advantage of a compromised immune system, developing into secondary infections which would compete with *P. aeruginosa*. Finally Chhabra *et al.* have suggested that insertion of AHLs into T-lymphocyte membranes causes immunosupression through the inhibition of immunological synapse formation. Membrane microdomains have been shown to play an important role in the formation of the immunological synapse. The insertion of AHLs into biological membranes observed in this study and the associated decrease in dipole potential could be interpreted as evidence that this is the case. # Materials and Methods ## Reagents Egg phosphatidylcholine (PC) was supplied by Lipid Products (UK). Di-8-ANEPPS was supplied by Invitrogen, UK. Saquinavir was supplied by Roche (UK). AHLs were synthesized as described previously. Tissue culture reagents, cholesterol and all other reagents were supplied at the highest purity available by Sigma Aldrich (Poole, UK). ## Membrane preparation and labeling with DI-8-ANEPPS This technique is outlined more comprehensively in. Briefly, PC_100% and PC_70%Cholesterol_30% (molar ratios) were dissolved in chloroform before drying under a stream of oxygen-free nitrogen gas by rotary evaporation until a thin film was formed. The lipid film was rehydrated with 280 mM sucrose, 10 mM Tris, pH 7.4 The resulting multilamellar solution was freeze- thawed 5 times in liquid nitrogen and finally extruded 10 times through polycarbonate filters with pores 200 nm in diameter (Nucleopore Corp., Pleasanton, USA) using an extruder (Lipex Biomembranes Inc., Vancouver, Canada) according to the extrusion procedure. This resulted in a monodisperse, unilamellar suspension of phospholipid vesicles. These were labeled exclusively in the outer bilayer leaflet with Di-8-ANNEPS. Here the phospholipid vesicles were incubated for at least 1.5 hours at 37°C in the dark in the presence of Di-8-ANEPPS dissolved in ethanol. ## Cell viability An alamarBlue (Invitrogen; Carlsbad, CA) resazurin reduction ass*ay* was used to measure of the number of viable and proliferating cells and was conducted according to the manufacturer's instructions. After incubation with AHLs compounds, 10% alamarBlue was added to each well. Cells were incubated for 4 h in a 5% CO₂ incubator at 37°C. A negative control containing culture medium and alamarBlue reagent without cells was included. Fluorescence measurements were made by excitation at 530–560 nm and measuring emission at 590 using a LS-55 plate reader (Perkin Elmer, MA, USA). ## Labeling of Jurkat T-Lymphocytes with Di-8-ANEPPS The cells were cultured in RPMI 1640 medium supplemented with 10% foetal bovine serum, L-glutamine (0.02M), penicillin (100units/ml) and streptomycin (100 mg/ml) and maintained at 37°C and 5% CO₂. Cells were counted using a trypan blue exclusion assay before harvesting by centrifugation at 300 g for 5 minutes. Cells were then labeled with Di-8-ANEPPS according to the methods outlined by Asawakarn *et al*. as follows: 0.5 µM Di-8-ANEPPS was added to a suspension containing 1×10⁶ cells per ml⁻¹ and incubated for 1.5 h at 37°C. ## Fluorescence measurements Fluorescence time courses were undertaken by adding desired amounts of an experimental reagent under study to suspensions of cells or phospholipid vesicles (400 µM lipid or 40,000 cells ml⁻¹) on a Fluromax-4 model spectrofluorimeter (HORIBA Jobin Yvon Thermo Electron, UK). Di-8-ANEPPS spectra were obtained by exciting the samples at 460 nm and 520 nm and measuring the emission ratio at 590 nm. The contribution of the dilution effect to the fluorescence signal was corrected by using equivalent liposome or cell suspensions and adding equivalent amounts of DMSO solvent (final concentration did not exceed 0.6%). Any changes in fluorescence signal were then subtracted from those obtained with the quorum compounds. These data were fitted to Equations 1 & 2 in order to determine the best description of the molecular process. The model which best fit these data was determined using an Extra sum- of-squares F-Test. Where *y* is the observed signal, B_max is the 100% ligand binding capacity (fluorescent units), K_d is the affinity of the QS compound for the membrane in concentration units and n represents the hill coefficient, i.e. as an index of cooperativity. Unless otherwise stated all experiments are reported from at least 3 replicates ie. n = 3. The authors would like to thank Siri Ram Chhabra and Alex Truman for AHL synthesis. We also thank the anonymous referees for their review whose comments have been included. [^1]: Conceived and designed the experiments: PO. Performed the experiments: BMD RJ. Analyzed the data: BMD PO. Contributed reagents/materials/analysis tools: PW PO. Wrote the paper: BMD PW PO. [^2]: The authors have declared that no competing interests exist.

# Introduction GKS-3 is involved in many cellular signaling pathways such as the insulin/PI3K or the Wnt pathways and participates in a high number of functions such as metabolism, cell proliferation, cell fate, survival and apoptosis. Besides, it also plays a key role in certain neuronal specific functions like long term potentiation (LTP) and depression (LTD) of synaptic activity. Dysregulation of GSK-3 has been postulated to participate in the etiology of neuropsychiatric or neurodegenerative diseases: bipolar disorder, schizophrenia, Alzheimer’s disease (AD), or Huntington’s disease, and of non-CNS diseases: type 2 diabetes or cancer. Consequently, GSK-3 inhibitors have been postulated as a promising therapeutic tool. Lithium inhibits GSK-3, and this has been postulated to contribute to its therapeutic efficacy, but also to its neurological toxicity. Together, this lithium’s adverse effects and those of GSK-3 genetic inhibition, warn about possible limitations of GSK-3 inhibitor based therapies. Understanding of the mechanism of this toxicity and of how to counteract it may be a key step for successful therapeutic use of GSK-3 inhibitors. GSK-3 is implicated in apoptosis but its modulatory effect can be different depending on the specific apoptotic pathway involved: intrinsic (type I) that involves release of cytochrome c and disintegration of mitochondria and extrinsic (type II) apoptosis that occurs upon the activation of death receptors, specifically the TNF receptor family including Fas and TRAIL. Consequently, lithium and other GSK-3 inhibitors are protective towards many apoptotic stimuli that affect mitochondrial integrity but increase apoptosis triggered by TNF, or Fas. Conceivably, this may have significant implications on the therapeutic potential of GSK-3 inhibitors. Mice with conditional expression of a dominant negative form of GSK-3 (Tet/DN- GSK-3 mice) are a useful tool to explore the neurological consequences of chronically decreasing GSK-3 activity in the brain. Tet/DN-GSK-3 mice display increased rate of neuronal apoptosis and impaired motor coordination that might relate to the frequent neurological motor side effects, such as hand tremor, experienced by lithium-treated patients. Interestingly, wild type mice chronically treated with lithium also show increased rate of neuronal apoptosis and a deficit in motor coordination that have been reported to occur through a mechanism involving Fas, as they are absent in Fas-deficient mice (*lpr* mice). Since lithium inhibits other enzymes like inositol-monophosphatase or histone- deacetylase, it cannot be ascertained that this toxicity mechanism is due to GSK-3 inhibition. Here we test if apoptosis and related behavioral consequences due to decreased GSK-3 activity are Fas dependent. # Materials and Methods ## Animals Tet/DN-GSK-3 mice in a C57/BL6J background were generated as described previously. Fas-deficient *lpr* mice (C57/BL/6J background) were obtained from Jackson laboratories (B6.MRL-Faslpr/J, stock number: 000482). All mice were housed at the Centro de Biología Molecular “Severo Ochoa” animal facility. Mice were housed four per cage with food and water available *ad libitum* and maintained in a temperature-controlled environment on a 12/12 h light-dark cycle with light onset at 07∶00 h. For behavioral analysis, mice were tested at the age of 2.5–3 months and they were sacrificed upon completion of the battery of tests. For histological studies only male mice were used. For behavioral studies males and females were used indistinctly. Statistical analysis as genotype x treatment interaction was evaluated by two way-ANOVA to rule out any effect of sex. All behavioral studies were performed during light phase. ## Ethics Statement Animal housing and maintenance protocols followed the guidelines of Council of Europe Convention ETS123, recently revised as indicated in the Directive 86/609/EEC. Animal experiments were performed under protocols (P15/P16/P18/P22) approved by the Centro de Biología Molecular Severo Ochoa Institutional Animal Care and Utilization Committee (Comité de Ética de Exerimentación Animal del CBM, CEEA-CBM), Madrid, Spain. ## PCR Mice were genotyped by using the following primers: For detection of the CamKII-tTA (tTA) transgene: primer tTA-C (5′-ACTAAGTCATCGGATGGAGC-3′) and primer tTA-F (5′-CGAAATCGTCTAGCGCGTCGG-3′) that amplify a 592 bp fragment. For detection of the DN-GSK-3 (tetO-R4) transgene: primer GSK-3A (5′- CATGGTCAGGTCATGGATGAGC-3′) and primer GSK-3B (5′-TAATCAGCCACTGATCCACCCAG-3′) that amplify a 642 bp fragment. Amplification protocol used for both combinations was: 5 min at 94°C followed by 30 cycles of 53°C for 1 min, 72°C for 1,5 min, and 94°C for 1 min and finally 72°C for 5 min. For discriminating the *lpr* mutation we used the protocol recommended by Jackson Laboratories with the three following primers: wild type (5′-CAAATCTAGGCATTAACAGTG-3′), mutant (5′-TAGAAAGGTGCACGGGTGTG) and common (5′-GTAAATAATTGTGCTTCGTCAG-3′) that yield a 179 bp band from the wild type allele and a 217 bp band from the mutant allele. ## Western Blot Analysis Mice were sacrificed using CO₂ and brains immediately removed and dissected on an ice-cold plate. Whole extracts were prepared by homogenizing the brain areas from right hemisphere in ice-cold extraction buffer consisting of 20 mM HEPES pH 7.4, 100 mM NaCl, 20 mM NaF, 1% Triton X-100, 1 mM sodium orthovanadate, 1 µM okadaic acid, 5 mM sodium pyrophosphate, 30 mM β-glycerophosphate, 5 mM EDTA, and protease inhibitors (2 mM PMSF, 10 µg/ml aprotinin, 10 µg/ml leupeptin and 10 µg/ml pepstatin). Samples were homogenized and centrifuged at 15000 g for 20 min at 4°C. The resulting supernatant was collected, and protein content determined by Bradford assay. Fifteen micrograms of total protein were electrophoresed on 10% SDS-polyacrylamide gel and transferred to a nitrocellulose membrane (Schleicher and Schuell). The experiments were performed using the following primary antibodies: anti-β-gal (Promega, 1∶2000) and anti-β-actin (Sigma, 1∶5000). The membranes were incubated with antibody overnight at 4°C in 5% non-fat dried milk. Secondary antibodies used were polyclonal rabbit anti-mouse immunoglobulins/HRP (DAKO Cytomation) (1∶2000) and ECL detection reagents (Perkin Elmer) were used for immunodetection. ## Immunofluorescence and Immunohistochemistry Left hemispheres were processed for histology placed in 4% paraformaldehyde in Sorensen's phosphate buffer (PFA) overnight, washed in PBS and then immersed in 30% sucrose in PBS for 72 hr. Once cryoprotected, the samples were included in OCT compound (Sakura Finetek Europe) frozen and stored at −80°C until use. 30 µm sagittal sections were cut on a CM 1950 Ag Protect freezing microtome (Leica) and collected and stored free floating in glycol containing buffer (30% glycerol, 30% ethylene glycol in 0,02 M phosphate buffer) at −20°C. For immunofluorescence, 30 µm sagittal brain sections were pretreated with 0,1% Triton X-100 for 15 min, 1 M Glycine for 30 min, and blocking solution (1% BSA and 0,1% Triton X-100) for 1 hour. Sections were then incubated overnight at 4°C with primary antibodies in blocking solution at the following concentrations: cleaved-caspase-3 (Cell Signaling Technology, MA) (1∶50), β-gal (Promega) 1∶2000 and NeuN (Chemicon) (1∶100). The following day, sections were washed in PBS. Then sections were incubated with donkey anti-rabbit Alexa 555 (Invitrogen) (1∶500), donkey anti-mouse Alexa 488 (Invitrogen) (1∶1000) and goat anti-mouse Alexa 488 (Invitrogen) (1∶1000) secondary antibodies for 1 hour. Finally, nuclei were counterstained with DAPI (1∶5000, Calbiochem). Sections were mounted on glass slides, coverslipped with Fluorsave (Calbiochem) and maintained at 4°C. Colocalization of markers was identified by taking successive Alexa 555 and Alexa 488 fluorescent images using a Laser Confocal LSM710 camera (Zeiss) coupled to an inverted microscope AxioObserver (Zeiss). For immunohistochemistry, brain sections were pretreated for 30 min in 1%H₂O₂/PBS followed by 1 h with 1% BSA, 5% FBS and 0.2% Triton X-100 and incubated overnight at 4°C with β-gal (ICN Biomed.-Cappel, 1∶2000) or cleaved-caspase-3 (Cell Signaling, 1∶50) primary antibodies. Finally, brain sections were incubated in avidin-biotin complex using the Elite Vectastain kit (Vector Laboratories). Chromogen reactions were performed with diaminobenzidine (SIGMA*FAST*™ DAB, Sigma) for 10 min. Sections were mounted on glass slides and coverslipped with Fluorosave (Calbiochem). For quantification of cleaved-caspase-3 immunostainings, four male mice per genotype were processed as follows: one every four serial 30 µm-sagittal sections were selected for staining between the 3.00 mm and 0.72 mm planes of the Paxinos and Franklin mouse brain atlas. Round-shape (5–10 µm in diameter) positive cells were quantified with the “Analyze particles” tool of Image J. Non-round shape cells with characteristic apoptotic bodies were detected manually. All analyses were performed blind and results were presented as the estimate of total positive cells in the whole brain structure. ## TUNEL Assay Sections were processed according to the In Situ Cell Death Detection Kit protocol (POD, Roche). Quantification was performed on 4 male mice per genotype by staining one every four serial 30 µm-sagittal sections spanning from the 3.00 mm to the 0.72 mm planes of the Paxinos and Franklin mouse brain atlas. Round- shape (5–10 µm in diameter) positive cells were quantified with the “Analyze particles” tool of Image J. All analyses were performed blind and results were presented as the estimate of total TUNEL-positive cells in the whole brain structure. ## Behavioral Testing ### Treadmill gait analysis Gait analysis was performed using the DigiGait™ system (Mouse Specifics Inc., Boston, MA). Briefly, digital images of paw placement were recorded at 80 Hz through a clear treadmill from beneath the animal. Mice were tested without pre- training in one session at a treadmill speed of 24 cm/s. Paws were marked with red colorant for better contrast. Plotting the area of each digital paw print imaged sequentially in time provides a dynamic gait signal, representing the temporal record of paw placement relative to the treadmill belt. Swing duration was measured as the time duration of the swing phase, when no paw is in contact with the belt. Paw angle variability is the variability of paw angle, considered as the angle that the paw makes with the long axis of the direction of motion. Stride length variation represents the standard deviation of the stride length for the set of strides recorded (reflecting the dispersion about the average value). Step angle variation was measured as CV and was calculated using the equation: *100 x standard deviation/mean value* (variability normalized to the mean). Step angle factors both stance width and stride length. Stance asymmetry is the ratio of left hind limb stance to right hind limb. The number of animals used for this test was: wt, n = 14 (5 males/9 females); Tet/DN-GSK-3, n = 12 (4 males/8 females); lpr/+, n = 22 (13 males/9 females); lpr/+;Tet/DN-GSK-3, n = 23 (12 males/11 females); lpr/lpr, n = 15 (11 males/4 females) and lpr/lpr;Tet/DN- GSK-3, n = 9 (5 males/4 females). ### Vertical pole test Testing was performed as previously described with minor modifications. The mouse was placed head-upward on the top of a vertical rough-surfaced pole (diameter 1 cm; height 50 cm) and the time taken to descend to the floor was recorded with a maximum duration of 60 s. The number of animals used for this test was: wt, n = 16 (6 males/10 females); Tet/DN-GSK-3, n = 18 (10 males/8 females); lpr/+, n = 28 (16 males/12 females); lpr/+;Tet/DN-GSK-3, n = 23 (12 males/11 females); lpr/lpr, n = 16 (13 males/3 females) and lpr/lpr;Tet/DN- GSK-3, n = 10 (5 males/5 females). ### Rotarod Test was performed with accelerating rotarod apparatus (Ugo Basile, Comerio, Italy). Mice were pre-trained during two days at a constant speed, 4 rpm the first day or 8 rpm the second day. Then, rotarod was set to accelerate from 4 to 40 rpm over 5 min and mice were tested four times. During accelerating trials, the latency to fall from the rod was measured. The number of animals used for this test was: wt, n = 16 (6 males/10 females); Tet/DN-GSK-3,n = 19 (10 males/9 females); lpr/+, n = 28 (15 males/13 females); lpr/+;Tet/DN-GSK-3, n = 25 (13 males/12 females); lpr/lpr, n = 16 (12 males/4 females) and lpr/lpr;Tet/DN- GSK-3, n = 10 (5 males/5 females). ### Open Field Locomotor activity was measured in clear plexiglas boxes measuring 43.2 cm×43.2 cm, outfitted with photo-beam detectors for monitoring horizontal and vertical activity. Activity levels were recorded with a MED Associates’ Activity Monitor (MED Associates, St. Albans, VT). Locomotor activity data were collected via a PC and was analyzed with the MED Associates’ Activity Monitor Data Analysis software. Mice were placed in a corner of the open-field apparatus and left to move freely. Variables recorded included: resting time (s), ambulatory time (s), vertical/rearing time (s), jump time (s), stereotypic time (s) and average velocity (cm/s). Data were individually recorded for each animal during 15 min. The number of animals used for this test was: wt, n = 6 (2 males/4 females); Tet/DN-GSK-3,n = 6 (1 male/5 females); lpr/+, n = 11 (7 males/4 females); lpr/+;Tet/DN-GSK-3, n = 14 (8 males/6 females); lpr/lpr, n = 12 (5 males/7 females) and lpr/lpr;Tet/DN-GSK-3, n = 6 (3 males/3 females). In order to rule out any sex difference statistical analysis was performed evaluating genotype × sex interaction for each parameter by a two way-ANOVA. The following results were obtained: swing (p = 0.001, F = 4.142), stride length variability (p = 0.054, F = 2.221), paw angle variability (p = 0.114, F = 1.805), paw area variability (p = 0.159, F = 1.612), vertical pole (p = 0.823, F = 0.436), rotarod (p = 0.662, F = 0.651), resting time (p = 0.772, F = 0.504), ambulatory time (p = 0.626, F = 0.700), vertical/rearing time (p = 0.999, F = 0.045), jump time (p = 0.878, F = 0.353), stereotypic time (p = 0.904, F = 0.311) and average velocity (p = 0.450, F = 0.935). As no sex differences were found among genotypes, males and females were used indistinctly. Behavioral tests were performed during three consecutive days as follows: first day, treadmill gait analysis and first training day of rotarod; second day, vertical pole and second training day of rotarod; and third day, four accelerating trials in the rotarod. The first group of animals tested performed an additional open field one day before starting behavioral tests. ### Statistical analysis Statistical analysis was performed with SPSS 19.0. Data are presented as mean values ± S.E.M. The normality of the data was analyzed by Shapiro-Wilk test. Statistical analysis of data with a normal distribution was performed following a one way-ANOVA test followed by a DMS or a Bonferroni *post-hoc* test. Statistical significance of non-parametric data was determined by Kruskal-Wallis test when analyzing all experimental groups, followed by a Mann-Whitney U-test for analysis of paired genotypes and Bonferroni correction was applied. Genotype frequencies were analyzed by means of chi-square test. A critical value for significance of p\<0.05 was used throughout the study. # Results ## Tet/DN-GSK-3 Mice in lpr Background are Viable and Grow Normally To explore whether the Fas dependent neuronal death observed in lithium-treated mice could be mimicked by inhibiting GSK-3 via a dominant-negative genetic approach, we decided to generate Tet/DN-GSK-3 mice in Fas-deficient background for subsequent behavioral and brain apoptosis analysis. Mice expressing DN-GSK-3 in forebrain neurons in a tetracycline repressible manner (Tet/DN-GSK-3 mice) were generated as previously described by combining mice expressing tTA under CamKIIα promoter (tTA mice) with mice carrying myc-K85R-GSK-3 and β-Gal sequences fused to a bidirectional tetO promoter (tetO-R4 mice). Fas-deficient mice (*lpr* mice) are naturally deficient in Fas receptor. Double transgenic Tet/DN-GSK-3 mice (that carry both the tTA-CamKIIa transgene, and the TetO-R4 transgene) were then combined with *lpr/lpr* mice as shown in. Resulting lpr/+;Tet/DN-GSK-3 (F1) were then bred with *lpr*/+ (F1) mice to obtain the six experimental genotypes out of the twelve possible ones. For identification of genotypes, three different PCR reactions were performed to detect the CamKII-tTA construct, the GSK-3 construct and the *lpr* mutation. To verify expression of the transgene, western blot and immunohistochemistry of the β-gal reporter were also performed. Analysis of the frequencies by means of a chi-square test revealed that all genotypes adjusted to the expected Mendelian distribution and they displayed no overt phenotype. They also displayed body weight similar to that of their control wild type (wt) littermates except for *lpr*/*lpr* mice that were bigger than wild type as previously reported. One-way ANOVA test (p = 0.026) followed by Bonferroni *post-hoc* test was applied to determine the level of significance and when males and females were tested separately, a similar effect was found for lpr/lpr mice. However, this difference was not evident in *lpr*/*lpr* mice harbouring any of the Tet/DN-GSK-3 transgenes. For further biochemical, histological and behavioral analysis only wt, Tet/DN-GSK-3, *lpr*/+, *lpr*/+;Tet/DN-GSK-3, *lpr*/*lpr* and *lpr*/*lpr*;Tet/DN-GSK-3 mice were used and they were analyzed at the age of 2.5–3 months. ## Absence of Apoptosis in Cortex and Striatum of Fas Deficient-Tet/DN-GSK-3 For the analysis of apoptosis, immunofluorescence and immunohistochemistry against cleaved (activated) caspase-3 and TUNEL staining were performed in the six experimental groups of mice at the age of 2.5–3 months. TUNEL is a common method for detecting DNA fragmentation that results from apoptotic signaling cascades and caspase 3 is one the major caspases activated during the execution phase of apoptosis. First, to verify that the cells undergoing apoptosis express the transgene, we performed double immunofluoresce for caspase-3 and the reporter β-gal. As described in our previous report, double immunofluorescence with caspase-3 and NeuN (neuronal marker) antibodies revealed an increase in the rate of apoptosis in Tet/DN-GSK-3 mice respect to wild type mice with the majority of apoptotic cells being neurons. For a more precise quantification and comparison among the experimental groups with varying *lpr* genotypes, immunohistochemistry of cleaved caspase-3 was performed. Counting of positive caspase-3 cells confirmed the previously reported increased rate of apoptosis in cortex (Cx) and striatum (St) of Tet/DN-GSK-3 mice respect to their corresponding control wt littermates (wt vs. Tet/DN-GSK-3: 334.24±33.61 vs. 578.81±111.68 positive cells in Cx, p = 0.023; 546.52±66.75 vs. 728.70±21.47 positive cells in St, p = 0.01). As expected, no effect in the number of caspase-3 positive cells was detected in brain regions that do not express the DN-GSK-3 transgene (TetO-R4 transgene) such as the thalamus or the cerebellum. Interestingly, levels of cortical and striatal apoptosis in Fas-deficient- Tet/DN-GSK-3 mice, either *lpr*/+;Tet/DN-GSK-3 or *lpr*/*lpr*;Tet/DN-GSK-3, were similar to control littermates (Tet/DN-GSK-3 vs. *lpr*/+;Tet/DN-GSK-3: p = 0.023 in Cx, p = 0.001 in St; Tet/DN-GSK-3 vs. *lpr/lpr*;Tet/DN-GSK-3: p = 0.062 in Cx, p\<0.000 in St). Therefore indicating that apoptosis resulting after sustained inhibition of GSK-3 in striatum and cortex is Fas dependent. Equivalent results were obtained when apoptosis was analyzed by TUNEL method (wt vs. Tet/DN-GSK-3: p = 0.007 in Cx, p = 0.01 in St; Tet/DN-GSK-3 vs. *lpr*/+;Tet/DN-GSK-3: p = 0.004 in Cx, p = 0.005 in St; Tet/DN-GSK-3 vs. *lpr/lpr*;Tet/DN-GSK-3: p = 0.007 in Cx, p = 0.005 in St). (statistical analysis of cleaved caspase-3 immunohistochemistry and TUNEL was performed following a one way-ANOVA test followed by a LSD *post-hoc* test). Furthermore, since the absence of DN-GSK-3 mediated apoptosis is not only seen in homozygous *lpr*/*lpr* background but also in the heterozygous *lpr*/+ background, just a partial attenuation of Fas-dependent signaling is sufficient to preclude the effect of DN-GSK-3 on apoptosis induction. ## Absence of Motor Deficits in Tet/DN-GSK-3 Mice in *lpr* Background Increased apoptosis in Tet/DN-GSK-3 is more prominent in striatum and cortex which are part of the basal ganglia circuit, involved in motor control. Accordingly, Tet/DN-GSK-3 mice have been reported to show deficits in motor tasks like the vertical pole and the footprint tests. Since we have found no evidence of increased neuronal death in striatum and cortex of Tet/DN-GSK-3 mice with Fas deficiency (*lpr*/+ or *lpr*/*lpr*;Tet/DN-GSK-3), we wondered whether motor deficits observed in Tet/DN-GSK-3 mice would also be absent in Fas- deficient Tet/DN-GSK-3 mice. To that end, the six experimental groups of mice were subjected to several behavioral tests that detect potential deficiencies in motor tasks. First we performed DigiGait test, which measures footprint pattern and other parameters of walking regularity. In good agreement with our previous report, Tet/DN-GSK-3 mice subjected to analysis of DigiGait parameters showed multiple gait abnormalities such as increased swing duration (ANOVA, p = 0.006, F = 3.374;), increased variability in stride length (ANOVA, p = 0.043, F = 2.342;), increased paw angle variability (ANOVA, p = 0.000, F = 10.306;) and increased paw area variability (Kruskal-Wallis, p = 0.000;) as compared to wt mice. Interestingly, *lpr*/+;Tet/DN-GSK-3 and *lpr/lpr*;Tet/DN-GSK-3 showed no abnormalities compared with their respective controls. Statistical analysis of swing duration, stride length variability and paw angle variability was performed by applying one way-ANOVA test followed by a LSD *post-hoc* test while analysis of paw area variability was performed using Kruskal-Wallis test for no parametric distribution of data. Then paired genotypes analysis was performed following a Mann-Whitney U-test and Bonferroni correction was applied. In summary, these results indicate that the increased variability in walking regularity found in Tet/DN-GSK-3 mice is reduced in Fas-deficient backgrounds. We then performed tests of striatal-dependent motor coordination like the vertical pole and the rotarod tests. In good agreement with the above shown results, we found significantly increased time to descend the vertical pole for Tet/DN-GSK-3 mice with respect to their control wt littermates (wt vs. Tet/DN- GSK-3: 17±2.2s vs. 10±0.8s, ANOVA: p = 0.036, F = 2.485). Interestingly, the difference was smaller and not significant between *lpr*/+;Tet/DN-GSK-3 (12±1.1s) or *lpr*/*lpr*;Tet/DN-GSK-3 (12±1.6s) and their respective controls. Statistical analysis was performed applying a one way-ANOVA test followed by a LSD *post-hoc* test. The accelerating rotarod is another test that detects striatal dependent deficits in motor coordination. This test measures the time spent by the mouse on a rotating cylinder (rod) as the speed of rotation is accelerating from 4 to 40 r.p.m. Results of rotarod test were represented as average of third and fourth trials. Despite a tendency to decreased time on rod of Tet/DN-GSK3 mice (wt vs. Tet/DN-GSK-3: 225±8.26 vs. 199±15.1, p = 0.101 in Kruskal-Wallis analysis followed by Mann-Whitney U-test.), no significant difference was observed for any group. Finally, to verify that these animals do not show impairment in general locomotive behavior, they were subjected to open field test and several parameters as inactivity time (p = 0.289, F = 1.277;), horizontal activity (p = 0.116, F = 1.876;), vertical activity (p = 0.0.064, F = 2.511;), jump time (p = 0.788;), stereotypic movements (p = 0.0.271, F = 1.32;) or average velocity (p = 0.098, F = 1.982;), were measured and no differences among genotypes were found. Statistical analysis was performed applying a one way-ANOVA test except for jump time for that a Kruskal-Wallis test was applied. # Discussion By combining conditional transgenic mice with neuronal expression of a dominant negative form of GSK-3 (Tet/DN-GSK-3 mice) and *lpr* mice, we have generated mice with sustained inhibition of GSK-3 in a Fas-deficient background for analysis of apoptosis in brain and for behavioral characterization of their ability to carry out motor tasks. Unlike Tet/DN-GSK-3 mice that show increased apoptosis in cortex and striatum, *lpr*/+;Tet/DN-GSK-3 and *lpr/lpr*; Tet/DN- GSK-3 mice showed levels of neuronal apoptosis similar to those found in wild type mice. In addition, analysis of motor coordination in tests of walking regularity, in the vertical pole test and in the rotarod test revealed that the motor deficits previously reported in Tet/DN-GSK-3 mice were no longer evident when genetic inhibition of GSK-3 occurs in a Fas-deficient background. Lithium prescription to bipolar disorder patients has decreased in the last years due to its common side effects. Understanding the molecular mechanism by which lithium exerts its neurological toxicity may lead to strategies to overcome its side-effects. In this regard, inhibition of GSK-3 by lithium has been postulated to contribute not only to its therapeutic efficacy, but also to its neurological toxicity. Since increased GSK-3 activity is believed to contribute to the etiology of other pathologies such as AD, lithium has been postulated as a possible therapy for AD. However, despite potential benefits of lithium for amnestic minimal cognitive impairment in a recent trial, clinical trials for AD have been hampered by high rates of discontinuation due to lithium’s adverse effects that is even more prominent in the elderly. Therefore, toxicity of lithium therapy is a problem that extends beyond mood disorders. Regarding the molecular basis of lithium toxicity, we have recently reported a mechanism for the neurological toxicity of chronic lithium in mice that is dependent on Fas signaling. As mentioned, lithium also inhibits other enzymatic activities such as inositol monophosphatase and histone deacetylase. Although this study does not provide direct evidence that decreasing Fas would decrease toxicity associated with lithium treatment, the results reported here demonstrate that neuronal apoptosis and motor deficits caused by sustained GSK-3 inhibition are Fas dependent. It is therefore conceivable that Fas signaling modulating drugs could be used in the future to improve clinical use of lithium not only for bipolar disorder but also for AD. Besides, more selective GSK-3 inhibitors are currently under development, and, similar to lithium, they are known to attenuate neuronal loss in AD animal models with increased GSK3 activity (such β-Amyloid infusion) but also to induce neuronal apoptosis when administered alone to wild type animals without any GSK-3 increasing stimulus. Therefore, modulation of Fas signaling might contribute to the successful application of selective GSK-3 inhibitors to clinics in general. It is worth noting that factors that attenuate Fas signaling have been suggested *per se* as a potential avenue for therapeutic intervention for AD since increased levels of Fas protein have been reported in the brain and cerebrospinal fluid of AD patients, and the *Fas* gene is located in the 10q24.1 region showing linkage to late onset AD, with polymorphisms in *Fas* having shown association with AD progression. What would increase the chances of success of a combination of GSK-3 inhibitors and Fas signaling-blockers for AD. Apart from the mentioned Fas-dependent mechanism for lithium- or genetic GSK-3 inhibition-induced apoptosis there is another reported mechanism by which decreased GSK-3 activity could contribute to exacerbate apoptosis through the extrinsic pathway. More precisely, GSK-3 has been reported to act at an intracellular juxtamembrane location in association with DDX3, the cellular inhibitor of apoptosis protein-1 (cIAP-1) and any of the death receptors to form an antiapoptotic complex. When GSK-3 activity is blocked, stimulation of death receptor potentiates caspase-3 activation. Therefore and in regard to the other above mentioned mechanisms requiring Fas activation, both mechanisms converge in the activation of a death receptor. As in our present study we have blocked apoptosis at the level of Fas (by using Fas-deficient mice), we cannot conclude if stimulation of the receptor or disassembly of the antiapoptotic complex, is predominant. Further experiments specifically aiming blockade of Fas signaling upstream or at the inhibitory complex would be needed to elucidate the contribution of both proposed mechanisms to apoptosis. One interesting feature of the results reported here is that heterozygous *lpr* modification of Fas is sufficient to correct both the apoptotic and the motor phenotypes of Tet/DN-GSK-3 mice. Therefore, provided the case that pharmacological inhibitors of Fas are developed in the future to facilitate lithium- of more selective GSK-3 inhibitor-based therapies, it is possible that just a partial attenuation of Fas-dependent signaling might sufficient to preclude the effect of GSK-3 inhibitors on apoptosis induction. However, we should also keep in mind the potential drawbacks of prospective Fas inhibiting therapies as Fas signaling is required for many aspects of brain physiology such as neural progenitor survival. An open question remains regarding the results reported here as whether the apoptosis observed in Tet/DN-GSK-3 mice and that disappears in the absence of fully functional Fas receptors is required for the motor phenotype that also gets normalized in a Fas deficient background. In fact, despite higher than in wild type mice, apoptotic rate in Tet/DN-GSK-3 mice is still low and does not produce a visible atrophy of affected structures. Alternatively, the effect in motor coordination could be caused not by the loss of a limited number of neurons in the circuit but by an alteration in the physiology of the global brain structure or circuit. In fact, our group has previously demonstrated that in a mouse model of striatal degeneration (HD94 mice) alterations in motor coordination occur in the absence of striatal cell loss, presumably by dysfunction of striatal neurons. It is therefore possible that Fas signaling may also affect physiology of healthy neuronal circuits. Thus an excessive Fas-FasL signaling could result in apoptosis of a limited number of neurons and in a more widespread dysregulation of neuronal physiology thus contributing to the behavioral consequences induced by GSK-3 inhibition. In summary, here have genetically proven that neuronal apoptosis and motor deficits caused by sustained GSK-3 inhibition are Fas dependent. This might have important implications to enable combined therapies that could facilitate the clinical use of lithium and or more selective GSK-3 inhibitors not only for bipolar disorder but also for other diseases in which excess GSK-3 activity is believed to contribute to pathogenesis. # Supporting Information The authors thank Dr. F. Hernández and Dr. J. Avila for helpful suggestions on many aspects of this research and for critical reading of the manuscript. We are also grateful to Javier Palacín, Desireé Ruiz, Alicia Tomico and Miriam Lucas for technical assistance. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: RGS JJL. Performed the experiments: RGS. Analyzed the data: RGS JJL. Contributed reagents/materials/analysis tools: RGS JJL. Wrote the paper: RGS JJL.

# Introduction Olanzapine (Olz) is an effective atypical antipsychotic drug used for treating severe psychiatric disorders, including schizophrenia and bipolar disorder. However, Olz administration is associated with excessive weight gain and severe metabolic side effects such as type II diabetes mellitus, hyperglycemia, dyslipidemia, and insulin resistance,. A novel antipsychotic drug that retains the clinical efficacy of Olz but causes less treatment-emergent weight gain would be an invaluable breakthrough in schizophrenia therapy. For the last few years, researchers have focused on developing novel antipsychotics with fewer metabolic side effects. Nonetheless, the crucial need for developing an ideal antipsychotic agent for schizophrenia still continues. It is suggested that the simultaneous blockade of the dopamine D₂ and serotonin 5HT_2A receptors with Olz through the various dopaminergic pathways is involved in the molecular mechanisms of Olz's therapeutic efficacy, resulting in the distinct clinical properties of this drug. Changes in hormonal peptide levels correlated with food intake (such as insulin and leptin) have been suggested to play a part in the weight gain induced by atypical antipsychotic drug treatments. Olz in particular significantly influences the regulation of plasmatic insulin and leptin, although the mechanism of action remains elusive. A large number of studies have also reported a relevant role for the affinities of atypical antipsychotics for the serotonergic 5HT_2A, 5HT₆, and 5HT₇ receptors; adrenergic α_1A receptor; and particularly histamatergic H₁ and serotonergic 5HT_2C receptors in their obesogenic effects. The blockade of the H₁ receptors has been repeatedly described as the most likely mechanism for atypical antipsychotic drug-induced weight gain. The inhibition of the H₁ receptors is directly involved in the activation of hypothalamic AMPK (5′ adenosine monophosphate-activated protein kinase) signalling, which stimulates food intake and positive energy balance and reverses the anorexigenic effects of leptin. A strong link between H₁ receptor affinity with antipsychotic agents and weight gain susceptibility has been reported. Clozapine and Olz, which have a higher affinity for the H₁ receptors (*K*_i = 1.2 nM and *K*_i = 2.0 nM, respectively), showed a greater propensity to induce weight gain. However, antipsychotic drugs with a lower H₁ receptor antagonist affinity, such as loxapine and amoxapine, caused neither weight gain nor weight loss in patients treated with these medications. Thus, the development of a novel antipsychotic agent with a similar 5HT_2A/D₂ receptor binding affinity ratio to that of Olz, and with a lower affinity for the H₁ receptors, may significantly advance schizophrenia therapy. We have previously examined two new analogues of Olz; OlzEt (2-ethyl-4-(4′-methylpiperazin-1′-yl)-10*H*benzo\[*b*\]thieno\[2,3-*e*\], diazepine), and OlzHomo (2-ethyl-4-(4′-methyl-1′,4′-diazepan-1′-yl)-10*H*-benzo\[*b*\]thieno\[2,3-*e*\] \[1,4\]diazepine) (newly synthesised by our research group), presenting an ethyl substituent at position 2 of the thiophene ring of Olz compounds. We have demonstrated in our previous published study that both of these analogues showed an *in vitro* lower affinity for the H₁ receptors while maintaining similar affinities for the D₂ and 5HT_2A receptors when compared to Olz. Since the blockade of the H₁ receptors has been repeatedly described as the most likely mechanism for atypical antipsychotic drug-induced weight gain, the present study further explored the therapeutic potential *in vivo* of these two analogues of Olz in an animal model of schizophrenia while assessing their metabolic side effects. We found that OlzEt and OlzHomo compounds display a significant reduction in metabolic side effects (weight gain and adiposity) compared to Olz. Thus OlzEt and OlzHomo may present as potential new drugs for schizophrenia therapy. # Materials and Methods ## Ethics statement All experimental procedures were approved by the Animal Ethics Committee, University of Wollongong, and conducted in accordance with the *Australian Code of Practice for the Care and Use of Animals for Scientific Purposes* (2004). ### Study (1) Animals and drug treatment regimes In the first series of experiments (Study 1), female Sprague Dawley rats (7 weeks old) were used to investigate weight gain and adiposity effects of a chronic treatment with Olz, OlzEt, and OlzHomo. Animals were obtained from the Animal Resources Centre (Perth, WA, Australia) and housed individually at 22°C, on a 12 h light-dark cycle with *ad libitum* access to water and standard laboratory chow diet (3.9 kcal/g, 74% carbohydrate, 16% protein, and10% fat). Animals were then randomly assigned to one of the following treatment groups: 3 or 6 mg/kg/day of Olz (Bosche Scientific, NJ, USA), OlzEt (Lichem, Hebei Boyuan Co., China), OlzHomo (Lichem, Hebei Boyuan Co., China), or vehicle (n = 8), three times daily at eight-hour intervals. Following 1 week habituation in their new environment, the animals underwent training to self-administer a sweet cookie dough pellet for 1 week. Cookie dough (62% carbohydrate, 22% protein, 10% vitamins, 6% fiber, and minerals) administration was performed as previously reported for 5 weeks. Over the course of this experiment, animals were weighed twice per week. Food and water consumption were also monitored every 48 hours for each animal and results were corrected for spillage. ### Post-mortem hormone, lipid and tissue analysis At the end of Study 1, female rats were fasted for 10 h prior to sacrifice by carbon dioxide asphyxiation. Upon sedation, blood was removed and collected in Lavender Vacutainer tubes containing EDTA (ethylenediaminetetraacetic acid; 5-HT, 5-hydroxytryptamine (serotonin) for hormonal testing. Samples were immediately centrifuged (1000 g for 10 min at 4°C), after which plasma was aliquoted and stored at −20°C until use. Fasting plasma insulin, leptin, and adiponectin levels were measured using commercially available Milliplex kits (Millipore Corp., USA) and Luminex 100. Plasma samples were processed by Southern IML Pathology for levels of glucose, cholesterol, triglycerides, high- density lipoprotein (HDL), and low-density lipoprotein (LDL) levels. White fat pads and sub-scapula brown fat pads were dissected from each animal and individually weighed. Brains were immediately removed, dissected into hypothalamus and prefrontal cortex, snap frozen in liquid nitrogen and then stored at -80°C until use. ### Study (2) Animals and drug treatment regimes In the next experiment (Study 2), the effects of Olz, OlzEt, and OlzHomo subchronic administration on PCP-induced behaviours were tested in male Sprague Dawley rats (180–200 g). Animals were housed in pairs in the same conditions described above. Following a 1 week habituation period, rats were treated orally with a sweet cookie dough pellet containing 3 mg/kg/day of Olz, OlzEt, OlzHomo, or vehicle (n = 12), three times daily at eight-hour intervals for 2 weeks. Animals were injected subcutaneously with either saline or PCP (10 mg/kg, synthesized in the School of Chemistry, University of Wollongong, Wollongong, New South Wales, Australia) 30 min following the final drug/cookie administration. Open-field behavioural testing was performed 15 min after this injection. ### Behavioural analysis and post mortem measurement The open field test was used to determine the behavioural effects of the pre- treatment of Olz, OlzEt, and OlzHomo on PCP-treated animals. To minimise stress during the experiment, animals underwent a 10 min habituation period for the open-field test one day prior to the experiment. As previously described, the locomotor activity was recorded for each tested animal in a black open square box (60 cm×60 cm×40 cm). Behavioural parameters including the total distance travelled (cm), mean velocity (cm/s), central and peripheral duration (s), and frequency of rearing were measured for 30 min and then analysed *via* Ethovision video-tracking software (Nodulus Information Technology, Wageningen, The Netherlands). Animals were euthanized 120 min following the open-field test as described above; brains were rapidly removed from the skull and dissected into prefrontal cortex and striatum, snap frozen in liquid nitrogen and stored at −80°C until required for the receptor binding assays. ### Radioligand binding assays The striatum, prefrontal cortex, and hypothalamus were used in radioligand binding assays to measure the receptor binding density of D₂, 5HT_2A, and H₁ receptors respectively. The assays were performed according to previously described procedures. In brief, the striatum, prefrontal cortex, and hypothalamus were homogenized separately and then centrifuged (27,000 g for 15 min at 4°C). The resultant membrane was incubated in the presence of 2 nM \[³H\]-Spiperone (specific activity, 15 Ci/mmol, 1mCi/ml; Perkin Elmer, Australia), with or without 2 µM (+) butaclamol (Sigma, Australia), 10 nM \[³H\]-Ketanserine (specific activity, 67 Ci/mmol, 1mCi/ml; Perkin Elmer, Australia) in the absence or presence of 10 µM methysergid (Sigma, Australia), or Pyrilamine (specific activity, 37 Ci/mmol, 1mCi/ml; Perkin Elmer, Australia) with or without 2 µM doxepin (Sigma, Australia), for D₂, 5HT_2A, and H₁ receptor binding assays, respectively. Radioactivity was measured by a beta liquid scintillation analyser (Perkin Elmer, Tri-Crab 2800 TR). ### Data analysis Data were statistically analysed using SPSS (version 17.0 SPSS, Chicago, IL, USA). Total weight gain, total food intake, energy efficiency, insulin, leptin, adiponectin, glucose, cholesterol, triglycerides, HDL, LDL, fat mass, and binding density were analysed by one-way analyses of variance (ANOVA) for each dose of Olz, OlzEt, and OlzHomo. Repeated ANOVA measures (COMPOUNDS×DAYS as repeated measures) were employed for cumulative weight gain, and food and water intake. Open-field parameters were also analysed by one-way ANOVA. Student's t-tests were used to determine the significance of differences between the saline and PCP-treated rats. Multiple comparisons were performed using Tukey or Games–Howell post hoc tests. Where Kolmogorov–Smirnov tests showed data to be distributed non-parametrically, Kruskall–Wallis tests were applied followed by Mann–Whittney U post hoc analysis. Correlations were identified using Pearson's correlation tests or Spearman's correlation tests for non-parametric data. Linear regression was performed in groups with significant correlations. Significance was set at P\<0.05. # Results ## Study (1) Weight gain and metabolic side effects of Olz, OlzEt, and OlzHomo ### Body weight gain, food and water intake Body weight gain was found to be significantly increased following both 3 mg/kg (F_3,27 = 7.11, P = 0.001) and 6 mg/kg (F_3,27 = 23.27, P\<0.001) Olz administration compared to the control group. However, the effects of both OlzEt and OlzHomo on body weight gain were not significantly altered compared to the control groups. OlzEt and OlzHomo showed a significant reduction in weight gain compared to the Olz group with 3 mg/kg (−26%, P\<0.05 for both compounds) and 6 mg/kg doses (−32% and −48%, P\<0.001, respectively). A repeated ANOVA measure (treatment×days) revealed a significant effect of time on the progressive enhancement of body weight for both doses of Olz (3 mg/kg and 6 mg/kg) (F_3.37,90.90 = 246.06, P\<0.001, and F_3.32,92.89 = 289.38, P\<0.001, respectively), and the interaction between these two factors (F_10.10,90.90 = 2.34, P\<0.05, and F_9.95,92.89 = 5.53, P\<0.001, respectively). As illustrated in, administration of Olz (at both tested doses) gradually increased weight gain from day 4 to the end of the treatment period. In contrast, the effect of OlzEt and OlzHomo on cumulative body weight was not significant compared to controls. Higher doses of Olz treatment (6 mg/kg) induced a significant increase in food intake (F_3,27 = 10.73, P = 0.001) that began after 6 days of treatment. In contrast, OlzEt and OlzHomo administration did not affect food intake for either of the tested doses. Post hoc analysis showed that total food intake after 5 weeks of treatment with OlzEt and OlzHomo (6 mg/kg) was significantly lower than Olz administration (14% and P = 0.001, 29% and P = 0.006, respectively)). A significant positive correlation between total body weight gain and total food intake was found after 5 weeks of treatment (r = 0.48, P\<0.001). ### Fat deposition Visceral fat deposition (intra-abdominal, including retroperitoneal) was significantly increased in 6 mg/kg Olz treated rats compared to the control group (*F*_3,28 = 8.98, *P* = 0.002). However, OlzEt and OlzHomo treated animals (3 mg/kg and 6 mg/kg, *P*\>0.05) did not show any significant difference in relation to visceral fat deposition compared to controls. Significant positive correlations were found between body weight gain and food intake with fat mass (*r* = 0.42, *P* = 0.002 and *r* = 0.59, *P*\<0.001, respectively). There were no effects of treatments on sub-scapula brown fat mass compared to the control group. ### Plasma hormone and glucose levels Olz treatment at a dose of 6 mg/kg caused a significant reduction in fasting plasma insulin levels in the tested animals compared to the control rats (*F*_3,23 = 10.46, *P*\<0.01). In contrast, OlzEt and OlzHomo administration did not affect the fasting plasma insulin levels in the tested rats compared to the controls. Additionally, both compounds showed higher insulin levels than those measured in Olz treated animals with 6 mg/kg doses (48%, *P* = 0.001, and 54%, *P* = 0.013, respectively). The insulin levels were not significantly altered in the treated groups (Olz, OlzEt, and OlzHomo) for the 3 mg/kg dose compared to the control animals. Post hoc analysis did not show any significant difference in the leptin levels of animals treated with Olz at both 6 mg/kg (*F*_3,22 = 6.92, *P* = 0.95) and 3 mg/kg (*F*_3,21 = 5.33, *P = *1.00) doses compared to the control groups. However, plasma leptin levels were significantly decreased following both 3 mg/kg and 6 mg/kg OlzHomo (*P = *0.01 and *P*\<0.01, respectively) compared to controls. Fasting plasma adiponectin levels were found to be significantly higher after chronic administration of Olz (at the 3 mg/kg and 6 mg/kg doses, *F*_3,23 = 22.45, *P*\<0.01, and *F*_3,24 = 18.28, *P*\<0.001, respectively) but were found to be lower following OlzHomo treatment (at the 3 mg/kg dose, *P* = 0.039) compared to controls. Both leptin and adiponectin levels for the animals treated with OlzEt were not significantly altered at both tested doses compared to controls). A significant negative correlation was found between plasma insulin levels and food intake (*r* = −0.46, *P* = 0.002), as well as a significant positive correlation between leptin levels and fat deposition (*r* = 0.34, *P* = 0.026). Plasma glucose levels were found to be not significantly altered following treatment with Olz, OlzEt or OlzHomo at either the 3 mg/kg dose or the 6 mg/kg dose compared to the control rats. ### Plasma lipid levels Levels of plasma cholesterol, triglycerides and LDL were found to be unaltered in the treatment groups (Olz, OlzEt, OlzHomo) at both the 3 mg/kg and the 6 mg/kg dose compared to controls. HDL levels were found to be significantly decreased in both the OlzEt 6 mg/kg treated animals and in the OlzHomo 3 mg/kg treated animals compared to controls (P = 0.027 and P = 0.007 respectively). However there was no significant difference in HDL levels in any of the other treatment groups at either dose. ### Aterations in H₁ receptor density Olz (3 mg/kg and 6 mg/kg) significantly reduced the H₁ receptor density in the hypothalamus of the female animals treated for 5 weeks (F_3,12 = 4.01, P\<0.05, and F_3,12 = 23.06, P\<0.001, respectively) compared to controls, while the H₁ receptor density remained unchanged in animals treated with either OlzEt or OlzHomo for both tested doses. There was a significant negative correlation between total body weight gain and visceral fat mass with H₁ receptor density in the hypothalamus (r = −0.59, P = 0.001, and r = −0.61, P = 0.001, respectively). ## Study (2) Behavioural and neurochemical study: A comparison between Olz, OlzEt, and OlzHomo treatments ### Behavioural testing Behavioural results are illustrated in. In the saline groups, no significant difference was found following 2 weeks of treatment with Olz, OlzEt, or OlzHomo in all the parameters measured in the open-field test. However, in animals acutely administered with PCP, total distance moved, mean velocity, and centre and periphery durations differed significantly according to the tested treatments (Olz, OlzEt, and OlzHomo). No sedative behaviour was observed in PCP- treated rats. Total distance moved was significantly reduced in Olz and OlzEt (F_3,31 = 4.97, P = 0.023, and F_3,31 = 8.19, P = 0.008) treated rats compared to controls. However, the effect of OlzHomo treatment on reducing locomotor activity was not significant compared to the controls (P = 0.42). Mean velocity was also reduced in animals treated with Olz (F_3,44 = 8.22 and P\<0.05). One-way ANOVA revealed significant changes in the centre and periphery durations following 2 weeks of treatment with OlzEt (F_3,33 = 4.37, P\<0.01, and F_3,41 = 3.21, P = 0.017, respectively). Moreover, t-test results showed a significant effect of acute PCP administration on the distance travelled in animals treated with Olz, OlzEt, OlzHomo or control compared to their equivalent saline groups. ### Alteration in D₂ and 5HT_2A receptor densities In the saline groups, 2 weeks of treatment with Olz and OlzHomo induced a significant reduction in D₂ receptor density in the striatum of the male rats compared to controls (F_3,20 = 34.83, P\<0.001, and F_3,20 = 31.25, P\<0.001, respectively). However, in the PCP-treated groups, a significant reduction was found following Olz and OlzEt treatments (F_3,20 = 40.75, P = 0.005) compared to controls. Similarly, 5HT_2A receptor density in the prefrontal cortex of the animals treated with Olz and OlzEt was significantly decreased in both saline (F_3,20 = 125.03, P\<0.001, and F_3,20 = 125.03P = 0.002, respectively) and PCP (F_3,20 = 92.13, P\<0.001) groups. # Discussion ## Pharmacological and behavioural evidence of effective analogues of Olz in an animal model: possible applications in the clinic N-methyl-D-aspartate (NMDA) receptor antagonist animal models of schizophrenia such as PCP have been widely used to test new drugs that have been developed for future schizophrenia therapies. These models offer reasonable validity with respect to the clinical symptoms of schizophrenia, and to some degree predict the efficacy of drugs in patients. In our study, we used the PCP rat model to determine whether sub-chronic pre-treatment with Olz, OlzEt, or OlzHomo could attenuate the characteristic PCP-induced behaviours in adult male rats. We chose to focus this study specifically in male rats since the effects of PCP injections have been shown to be more pronounced in male compared to female rats in several behavioural tests. Coinciding with previous reports, our results showed that PCP-treated rats showed a remarkable increase in spontaneous locomotor activity (i.e. total distance travelled or travel velocity) and anxiety/exploratory related parameters (duration/frequency in centre or periphery) of the open-field test compared to the saline group. It has been suggested that the behavioural effects of PCP treatment are due to the multiple mechanisms of action that may include altered dopamine, serotonin and noradrenaline transmission. For instance, a disruption to the firing pattern of dopaminergic neurons, which increases dopamine release in the frontal cortex and activates the mesolimbic dopaminergic neurons, may play a part in the PCP- induced psychotic behaviour which is similar to that seen in schizophrenia patients. Altered activity of glutamatergic neurons in the cortex, leading to elevated glutamate release and a reduced inhibitory feedback onto the principal neurons, is also involved in this mechanism. Regarding its clinical relevance, Olz has been shown to attenuate the hyperlocomotion and anxiety induced by PCP administration in animals. In our study, the observed PCP-induced behaviours were largely blocked in the Olz and OlzEt treatments in rats. Pre-treatment with Olz and OlzEt significantly inhibited the hyperlocomotion induced by PCP in male rats. Interestingly, OlzEt was more effective than the Olz treatment in suppressing the anxiety-like behaviours of PCP, such as longer time spent in the outer field and lower entries into the centre field, suggesting the potential antipsychotic capacity of OlzEt. In contrast, OlzHomo did not suppress the PCP-induced behaviours measured in the open-field test. It is suggested that the mixed antagonistic activity of Olz at multiple receptors, including dopamine D₁–D₄; serotonin 5HT_2A and 5HT_2C; muscarinic M₁; and adrenergic α₁, and α₂ receptors, may underlie its ability to block PCP-induced behaviours. For instance, an increased level of serotonin at synapses containing 5HT_2A receptors plays a part in PCP-induced hyperlocomotion. This effect may be prevented by 5HT_2A antagonism by Olz at the level of motor pathways in the spinal cord or in the brain. In our previous study, OlzEt showed a similar affinity as Olz for blocking the D₂ and 5HT_2A receptors in the striatum and prefrontal cortex respectively, which may partly explain the way in which OlzEt inhibits the PCP-induced behaviours *in vivo*. On the other hand, OlzHomo demonstrated a lower affinity for blocking these two receptors, which may contribute to its lack of efficacy for alleviating the PCP-induced hyperactivity. In fact, since the ambulation in OlzHomo/PCP-treated rats was comparable to the control/PCP group, we postulate that the potential for therapeutic effectiveness of an OlzHomo regime may be achieved at a higher dose than that at which Olz and OlzEt are administered. To further validate our hypothesis regarding the effect of PCP on the levels of D₂ and 5HT_2A receptors, we measured the neurochemical changes in the brain following the PCP challenge. Consistent with previous reports, our findings showed that the D₂ and 5HT_2A receptor densities in the striatum and prefrontal cortex respectively, in adult male PCP-treated rats, did not differ from saline-treated controls. However, subchronic treatment with Olz and OlzEt induced a long-lasting down-regulation in the binding capacities of D₂ and 5HT_2A receptors in both saline and PCP-treated animals. As previously suggested, the down- regulation of these two receptors may partly contribute to the blockade of PCP- induced behavioural changes, including hyperlocomotion,. However, this hypothesis should be taken with a degree of caution, since altered dopamine and serotonin receptor densities are not the only mechanisms underlying the behavioural changes induced by PCP administration. The extent to which such changes are involved in the therapeutic effects of Olz and OlzEt remains to be investigated. However, based on the pharmacological and behavioural results reported in our study, OlzEt warrants further examination for the treatment of schizophrenia. ## Prevention of weight gain and adiposity following OlzEt and OlzHomo regimes In contrast with study 2, female rats were used in study 1 since the metabolic side effects following antipsychotic treatment (such as Olz) reported in the literature have been more pronounced in females compared to males. The switch of gender for this study should not influence the results since the most appropriate animal model was chosen to be able to validate the different hypothesis in the two respective studies. Similar to previous animal studies and clinical reports, our study showed that chronic treatment with Olz (at both 3 and 6 mg/kg doses) induced an increase in body weight gain, with higher doses causing a greater effect. In contrast OlzEt, and particularly OlzHomo, prevented weight gain over the treatment period. Our study also confirmed previous findings in relation to hyperphagia and enhanced energy efficiency induced by Olz treatment. Notably, food intake was increased in animals treated with 6 mg/kg of Olz, indicating that energy consumption contributed significantly to body weight gain in this group of animals. On the other hand, food intake was not significantly increased in the 3 mg/kg Olz-treated rats, despite a significant increase in the body weight of these animals. These results suggest that a decrease in energy expenditure may be associated with the Olz-induced weight gain with the 3 mg/kg dose. Coinciding with previous studies, our findings report a positive correlation between body weight gain and visceral fat deposition despite Olz, OlzEt and OlzHomo inducing low visceral fat deposition. Both peripheral and central factors may be involved in Olz-induced weight gain and adiposity, nevertheless the exact mechanisms by which this drug causes metabolic adverse side effects still remains elusive. The fat deposition may be associated with the increase in body weight or be due to the direct effects of Olz on adipose tissue. Studies reporting Olz-mediated peripheral adipogenesis in the 3T3-L1 cell model showed an over-expression of *fatty acid synthase* and *adiponectin* genes. In accordance with this *in vitro* adipogenesis result and with previous animal and clinical studies, our findings support the effect of Olz administration on increasing plasma adiponectin levels. However, our data may appear counterintuitive given that other studies reported a reduction of adiponectin in obesity. The correlation between plasma leptin levels and visceral adiposity found in this study suggested that leptin may be a useful indicator of fat mass deposition induced by Olz. Interestingly, we found plasma leptin levels in animals treated with OlzHomo were significantly reduced, which may confirm the preventative effect of OlzHomo on visceral adiposity. A similar outcome was found in the leptin levels of female rats treated with ziprasidone, which has a low effect on weight gain and fat deposition. Numerous studies have shown that Olz-induced weight gain is associated with elevated leptin levels in schizophrenia patients. Since we have observed a link between enhanced adiposity and leptin levels, the lack of a significant effect of Olz treatment on leptin levels in our study may appear surprising. However, our treatment duration was subchronic (5 weeks), which can explain the discrepancy of our data with chronic studies performed with schizophrenia patients. Our study also showed a marked reduction in insulin secretion in the 6 mg/kg Olz treatment group with no significant change in the 3 mg/kg Olz, OlzEt, and OlzHomo-treated animals as illustrated in. These results support some recent findings that short term treatment with Olz decreased insulin levels in rats and schizophrenia patients. On the other hand, an extensive amount of literature has reported increased insulin levels following chronic treatment with Olz, particularly in patients who gained a significant amount of weight. However, as a result of increasing weight gain, chronic administration of Olz can lead to compensatory hyperinsulinemia and insulin resistance, which are commonly observed in clinical cases. The direct antagonistic effect of Olz delivered at a high dose (6 mg/kg in rats represents around double the dose for treatment in humans) on the muscarinic M₃ receptors in the pancreatic β-cells which regulate insulin secretion may contribute to the reduction in plasma insulin concentrations observed in our study. With regards to the levels of plasma glucose, our results did not show any significant difference between the different treatment groups (Olz, OlzEt and OlzHomo) compared to the controls, which is in accordance with results from recent clinical studies. Since blood glucose levels depend not only on insulin secretion, but also on tissue insulin utilisation, our results suggest that there is no insulin resistance at the early stage of olanzapine treatment in our study. Although weight gain was significant in rats treated with Olz *vs.* vehicle rats, these animals seemed to remain insulin sensitive. This is supported by our results showing no significant difference in the lipid profiles performed in the Olz treated rats compared to the vehicle rats. The affinities of atypical antipsychotics for the 5HT_2A, 5HT_2C, 5HT₆, α_1A and H₁ receptors and their obesogenic effects has been repeatedly reported in various studies. Particular emphasis was placed on the ability of antipsychotics to block the H₁ receptors. The blockade of the H₁ receptors is directly involved in the activation of the hypothalamic AMPK signalling pathway, which stimulates food intake and positive energy balance and reverses the anorexigenic effect of leptin. This study supports the potential role of H₁ receptor affinity in antipsychotic-induced weight gain and fat deposition. As shown in our previous report, OlzEt and OlzHomo have a lower affinity for binding to the H₁ receptors (*K*_i = 1.95, and *K*_i = 13.63, respectively) compared to that of Olz (*K*_i = 0.13). Therefore, the pronouced antagonism of OlzEt and OlzHomo at the H₁ receptors may be responsible for their significantly attenuated propensity to induce weight gain and metabolic dysfunction, which are associated with Olz treatment. Furthermore, corresponding with previous reports, the present study demonstrated a significant negative correlation between hypothalamic H₁ receptor density and weight gain and accumulative fat mass in rats. H₁ receptor density in the hypothalamus has been markedly reduced following chronic treatment with Olz (at both 3 mg/kg and 6 mg/kg doses) but not with OlzEt and OlzHomo, supporting the importance of the H₁ receptor in Olz-induced obesogenic side effects. In agreement with our observation, the down-regulation of H₁ receptor expression has been previously reported in the hypothalamic nuclei of rats treated with Olz. In addition, the nonsignificant alterations of the H₁ receptor density following OlzEt and OlzHomo treatments may explain the lack of orexigenic effects of these two compounds in the treated rats. Thus, the involvement of the H₁ receptors in Olz-induced obesity and fat deposition might be closely related. In general, while our study confirmed the effect of Olz administration on metabolic alterations in rats, we showed that OlzEt and OlzHomo administrations did not induce either enhancing effects on body weight and food intake or detrimental consequences on fat deposition and metabolism. Our findings appear to have reasonable predictive validity for different aspects of Olz-induced weight gain/adiposity and metabolic abnormalities which mimic the clinical situation. In conclusion, our findings confirmed the obesogenic effect of Olz administration, coupled with the down-regulation of the H₁ receptors in the hypothalamus. Our results showed that both OlzEt and OlzHomo appear to be promising new candidate compounds which did not result in weight gain, visceral fat deposition and metabolic dysfunction. However, based on the pharmacological proprieties alone, OlzEt presented with a similar profile to Olz during behavioural assessment in the open-field test with regards to blocking PCP- induced hyperactivities. These findings suggest that the long lasting down- regulation of D₂ and 5-HT_2A receptors induced by sub- chronic Olz and OlzEt treatment may play a part in blocking PCP-induced behaviours. The fact that OlzHomo had a reduced capacity to inhibit PCP-induced behaviours could also be explained by its lower affinity for the D₂ and 5HT_2A receptors in the brain compared to that of Olz and OlzEt. Therefore, if the OlzHomo regime was to be delivered at a higher dose than that of Olz and OlzEt, then the therapeutic effectiveness of OlzHomo may be increased. Given the limitations associated with animal models, we suggest that the present results be taken with caution. Only further behavioural studies and clinical trials will reveal the predictive validity of this preclinical model for the therapeutic efficacy and metabolic side effects of these two compounds. # Supporting Information We would like to thank Mr Marc Bouillon (School of Chemistry, University of Wollongong, Wollongong, New South Wales, Australia) for synthesizing the PCP for this study and Dr Mei Han (School of Medicine, University of Wollongong, Wollongong, New South Wales, Australia) for her contribution to our behavioural experiments. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: FFE SJ. Performed the experiments: SJ FFE. Analyzed the data: SJ JLA FFE. Contributed reagents/materials/analysis tools: SJ FFE. Wrote the paper: SJ FFE. Provided advices on the final versions of the manuscript: XH. Provided support in english writing and comprehension: JLA.

# Introduction Tuberculosis (TB) remains a major global public health problem. TB is an infectious disease that causes the highest number of deaths globally due to a single infectious agent. Indonesia ranks as the second-highest TB-burdened country in the world and carries the third-highest gap between the estimated number of incident cases and the notifications of new cases. Approximately 30% of active TB cases are currently undetected by Indonesia’s health services. The World Health Organization (WHO) recommends screening for early detection of tuberculosis cases to reduce transmission and improve patient outcomes. The screening is prioritized for people who have close contacts with TB (who live in the same household or frequent contact with the sputum smear-positive TB patients), and afterward for populations at risk of TB (people with HIV, diabetes, residents of the area with high TB transmission, *i*.*e*., slum areas). Screening with clinical symptoms has a sensitivity of 70% while screening with chest radiography has a sensitivity of 87% (3). However, chest radiography exposes patients to radiation and is not practical to use in remote areas. Due to TB’s significant clinical and economic impact, it is of great importance to develop screening tools that are accurate, point-of-care, easy to use, and produced at a low cost, thus can be widely used in lower-middle-income countries such as Indonesia. An electronic nose (eNose) was investigated as a diagnostic tool for TB by examining the patients’ exhaled breath. An eNose is an easy-to-use tool based on an array of sensors that can learn and diagnose a disease from the pattern of volatile organic compounds (VOCs) that are produced by *Mycobacterium tuberculosis* or by the host metabolism due to TB, which are different from standard conditions. The sensor array in the electronic nose comprises non- specific sensors exposed to a variety of odor-sensitive biological materials, such as breath, urine, or feces. An odor stimulus produces a specific fingerprint from the sensor array. Patterns or fingerprints from known odors are used to build a database and train a pattern recognition system; thus, unknown odors can subsequently be classified. Generally, eNose instruments consist of hardware components to collect and transport odors to the sensor array and electronic circuitry to digitize and store the sensor’s responses for signal processing. An eNose is practical to carry, easy to use, and without radiation exposure, making it suitable as a screening tool. To our knowledge, there has been no published study protocol concerning eNose as a screening tool for TB. To date, trials that investigate eNose as a screening tool for TB are planned or ongoing, which include our trial and another one in Paraguay (NCT04407325). We aim to investigate the potential of an eNose-TB as a screening tool in Indonesia compared to the screening with clinical symptoms and chest radiology, which are currently used as a standard. We further aim to analyze the time and cost of a screening algorithm with eNose-TB to obtain additional detection of one TB case. We formulated the following hypotheses: 1) Breath test has high sensitivity (\>90%) as a screening test for TB, and 2) the time and cost of a screening algorithm with eNose-TB to obtain additional detection of one TB case are shorter and lower, respectively, in comparison with the standard of screening by clinical symptoms and chest radiology that is currently used. # Materials and methods ## Design plan In a previous pilot machine learning study (unpublished), we trained the eNose- TB to recognize the breath pattern of TB patients and healthy controls. The TB patients were recruited consecutively from the Respira Lung Hospital, a hospital formed by five previous public lung clinics in Yogyakarta. Meanwhile, the healthy controls were recruited from TB patients’ neighborhood to represent the same socioeconomic conditions, but these participants did not have close contact with the TB patients. The inclusion criteria of the TB patients were: 1) diagnosed as TB, had at least 1+ on the WHO scale in minimum 1 out of 2 samples examined for microscopy, and positive result of Xpert MTB/Rif and culture (identification of *M*. *tuberculosis*) from the samples, 2) had never received TB treatment, 3) agreed to participate in the study by signing the informed consent, 4) did not smoke for at least one year before the study. The inclusion criteria for the healthy subjects were: 1) agreed to participate in the study by signing the informed consent, 2) did not smoke for at least one year before the study, and 3) had no sign nor symptoms of TB. The exclusion criteria were 1) unable to breathe normally for two minutes due to respiratory illness, 2) incomplete supporting examination data, and 3) invalid measurements of breath tests. As part of the routine examination, all participants were characterized by clinical symptoms (*e*.*g*., persistent cough, unintentional ≥5% weight loss, and night sweats), CXR, smear microscopic, Xpert MTB/Rif examination. For study purposes, we added culture, HIV test, and eNose-TB breath test. The participants breathed normally through a disposable mouthpiece of pipe connected to the electronic nose. The inlet was covered with HEPA-filter to protect the *electronic-nose* contaminated with bacteria and virus. Afterward, the electronic-nose was connected to a laptop, and the breath data was analyzed. We included 27 TB patients and 24 healthy controls. Bruins *et al*. and Zetola *et al*. showed that this was an acceptable sample size to recognize the breathing pattern by the eNose-TB machine. The eNose-TB raw data were manipulated by the baseline using a different method. The data were cleaned by windowing the data during the sampling phase with five windows, and a feature extraction method was performed in each window, namely the maximum value, the median value, and the standard deviation value. Afterward, a leave-one-out cross-validation (LOOCV) was used in training procedures as a validation method. LOOCV was used in the hyper-parameters tuning process of Support Vector Machine (SVM) and Recursive Feature Elimination (RFE) for a feature selection method. Logistic regression was used as an estimator in the RFE method. Seven breath samples were invalid, thus were excluded from the analysis. shows the Receiver Operating Characteristic (ROC) curve of the best model in sensitivity and specificity of breath test in the training phase; sensitivity was 95% (95% CI = 77–100%), and the specificity was 82% (95% CI = 60–95%). During the breath sample collection, there were no adverse events (e.g., breathless, infection, or bleeding) associated with the study intervention. After the previous pilot machine learning study was finished, we aim to continue the study with a two-phase planned project, namely the validation and screening phases. The purpose of this current study is to validate the eNose-TB and investigate it as a screening tool for TB. ### I. Validation phase of the electronic nose This phase uses a cross-sectional design. It is conducted in Surakarta General Hospital, Central Java, and primary health centers in the municipality of Yogyakarta and Kulon Progo district. The inclusion criteria are: 1) agree to participate in the study, 2) able to produce samples for Xpert MTB/Rif examination, and 3) able to produce exhaled air samples. The exclusion criteria are: 1) invalid measurements of breath tests, 2) incomplete CXR data, 3) missing specimens, and 4) inability to breathe normally for 2 minutes due to respiratory illness. All study participants are characterized by clinical symptoms (*e*.*g*., persistent cough, unintentional ≥5% weight loss, night sweats), CXR, smear microscopic, Xpert MTB/Rif examination, HIV testing, and breath test with the eNose-TB. Participants’ sputum is taken and analyzed with Xpert MTB/Rif examinations to confirm the diagnosis of TB. An acid-fast-bacilli Ziehl-Neelsen microscopy is done as a part of the routine examination in the hospital. If a participant cannot produce sputum, he/she will be asked to collect a stool sample, as this method has been shown as reliable in diagnosing TB. The study uses triple-blind masking, in which the research subjects, breath sample takers, and laboratory sample examiners do not know the results of each sampling that has been done. The final data processor is also blinded to the results of Xpert or smear microscopy. The breath sampling data are saved in graphic form, of which interpretation will be done later by the data processor at the final stage. ### II. Screening phase of the electronic nose After the validation phase, we will conduct a screening phase with a cross- sectional design. The study will be conducted in the municipality of Yogyakarta, a region with the highest TB prevalence in Yogyakarta Special Province, and in the Kulon Progo district of Yogyakarta, which contains many remote areas, where eNose-TB will have its most useful application. The municipality of Yogyakarta has 18 primary health centers and 21 hospitals. The estimated TB incidence in 2019 is 1,400 cases. The Kulon Progo district of Yogyakarta has 21 primary health centers and nine hospitals, and the estimated TB incidence in 2019 is 1,033 cases. The study population is adults and children in these two districts. In this phase, we will use an eNose-TB that has been validated; thus, it has a more robust trained pattern recognition technique, and as a result, it can give real-time measurements. As the active TB case finding activity, namely “Zero TB city”, is launched in Yogyakarta, the eNose-TB will be paired with the symptom screening and chest radiology (CXR) in this activity. The mobile clinic team consisting of doctors, radiology officers, laboratory personnel, and nurses will travel to the area with a high risk of TB, *i*.*e*., the waiting rooms of an outpatient clinic in primary health centers and hospitals, slums, boarding houses, boarding schools, and prisons. The inclusion criteria are: 1) agree to participate in the study, and 2) currently not in TB treatment, while the exclusion criteria are: 1) invalid measurements of breath tests, 2) incomplete CXR data, 3) missing specimens, and 4) unable to breathe normally for 2 minutes due to respiratory illness. Participants will be interviewed for symptoms of TB, which consist of the main symptoms (cough that lasts more than two weeks, night sweats, unintentional ≥5% weight loss) and other symptoms (coughing up blood, fever \>1 month, enlarged lymph nodes, shortness of breath and chest pain), and undergo CXR examination. They will then undergo a breath test. Patients with positive radiological results or positive breath tests or coughing for more than two weeks or coughing up blood or showing extrapulmonary TB symptoms will be asked to collect their sputum or stool samples for Xpert MTB/Rif examinations to confirm the diagnosis of TB. We will collect participants’ demographic and clinical data. We will analyze the screening algorithm’s time and cost with eNose-TB to obtain additional detection of 1 TB case. The screening phase also uses triple-blind masking, in which the research subjects, breath sample takers, and laboratory sample examiners do not know the results of each sampling. ## Tools and materials Figs and show the eNose-TB system, which contains a sampling system (air collecting bag, HEPA-filter, valve, and micro-pump), a sensor array system, and a data acquisition system (Arduino and a personal computer). The micro-pump has a flow rate of about 1 milliliter per minute. This micro-pump functions to suck the breath sample into the gas sensor chamber and to push it out after the sensing process is complete. Meanwhile, HEPA filters are used to filter virus particles or other foreign objects other than VOCs from breath samples. The manufacturer of the eNose-TB device is the Department of Physics, Faculty of Mathematics and Natural Sciences, Universitas Gadjah Mada, Indonesia. The software was also developed and validated by the same department. We use the Metal-Oxide-Semiconductor (MOS)-based gas sensors. All sensors used in the eNose-TB device are commercial sensors available in the market with a high level of replication. There are 16 sensors used, with the characteristics of global selectivity. Based on the datasheet and our tests’ results, each sensor responds to at least 5 different gases, thus the detection range is at least 80 different VOC groups. With that coverage, the sensor array will form a relatively consistent pattern for each group of breath. is a typical voltage response of 16 sensors in eNose-TB device to time. The sensor response begins with the baseline at the same level for 10 seconds, when the surrounding air is introduced into the gas sensor room. After that, the electronic valve will control that the breath sample is inserted into the gas sensor chamber for 120 s. The sensing process occurs after the 10th seconds to the 130th seconds, so the total sampling time is 120 seconds. When the sensing process is complete, the sensor room is cleaned with the ambient air until it returns to the original baseline level. To distinguish the breath response pattern between the negative and positive TB subjects, each sensor’s response during the sensing process is divided into 5 windows with an interval of 24 seconds each. Furthermore, we extract the features from each window in the form of the median and standard deviation values with pre-processing in the form of standard and scaler. Each sensor has 10 characteristic values, which are then used as input into the machine learning model as a classification model, namely the support vector machine (SVM). The three main parameters of SVM for each set is a cost value of 10, with the kernel in the form of a linear function and using an optimizer in the form of a recursive feature elimination (RFE). Because the number of data in the pilot machine learning phase (training phase) was limited, we used a system performance validation method in the form of a leave-one-out cross-validation (LOO-CV). The effects of temperature and humidity in the sensor room are measured in parallel (with a calibrated SHT31 sensor) during the sensing process. shows the change in temperature and relative humidity during the pilot machine learning study (training phase). Changes in temperature and relative humidity during measurement were only about 0.5 C and 15%, respectively, which means that there is no significant effect on sensor response when referring to sensor characteristics in general. ### eNose-TB software system The eNose-TB software system consists of two programming parts: microcontroller programming (Arduino MEGA 2560) and data logger software programming (Windows). Microcontroller programming is conducted with the Arduino IDE software, while the data logger programming uses Microsoft Visual Studio 2019 Community (open source) software, which uses the C# programming language. ## Sampling plan ### Collection of the exhaled breath samples The study participants are asked to take two initial deep breaths and exhale, using their protective masks. Next, subjects are asked to take a deep breath and exhale powerfully (a forced expiratory volume) to the air collecting bag from the third and subsequent breaths (maximum until the fifth breath) until the collecting bag is full. The collecting bag is sealed and connected to the eNose- TB machine via a collecting hose and HEPA-filter protecting the eNose-TB from microbes. A designated officer supervises the breath collection. The data are read and stored in the eNose-TB machine. Considering the risks of TB and COVID-19 transmission, we will place the eNose-TB device in a particular isolation room, and the officer who operates the eNose-TB uses level 3 personal protective equipment (N95 mask, gloves, goggles, gown, boots, or closed shoes). We also use disposable equipment (HEPA-filter, connector, and air collecting bag), apply disinfectant procedure after each breath sampling, and ensure that the physical distancing measurement is applied during the sampling collection. We plan to finish the data collection in December 2021, with the possibility of extending our study to another year and the addition of study sites if COVID-19 jeopardizes our plans. ### Variables The outcome variable in the first phase (validation phase) will be sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), while in the second phase (screening phase) will be positive agreement, negative agreement, and time and cost analysis of screening with the eNose-TB. We will also collect the following data variable: age, body mass index, sex, occupation, smoking habits, alcohol consumption, comorbidities (diabetes, asthma, Chronic Obstructive Pulmonary Disease/COPD, HIV/AIDS, flu, bronchitis, bronchiectasis, lung fibrosis, lung abscess, empyema, polycystic lung disease), co-medication (inhalation drugs and antibiotics), and food and beverages consumption before the breath test. Data of age, body mass index, sex, comorbidities, co-medication, and occupation will be asked directly to patients or their legal guardians. Smoking habits and alcohol consumption will be measured by asking participants, “When was the last time you smoked” and “When was the last time you consumed alcohol” (on a value of 1 and 2, 1 being ‘never or ≥8 hours ago’, 2 being ‘\<8 hours ago’). Food and beverage consumption before the breath test will be measured by asking participants, “When did you last take food or beverages” (on a value of 1 and 2, 1 being ‘not having prior intake or having prior intake ≥1 hour ago’, 2 being ‘having prior intake \<1 hour ago’), and “What kind of food and drink did you consume” (will be noted as description data, and later categorized as poultry meat, coffee, tea, milk, and none of them). All questions will be asked in the local language (Bahasa Indonesia). The interview guide can be accessed in the. Two researchers will double-enter all data into a database and ensure no missing data or typing errors. ### Sample size The validation phase is conducted on minimum 395 presumptive TB patients who are recruited consecutively. The sample size is calculated based on the following forma: $\text{n} = \frac{\text{Z}_{\propto /2}{}^{2}\text{xSNx}\left( {1 - \text{SN}} \right)}{d^{2}}/P$ with Z_∝/2 is the Z-score for the confidence level of 95% (1.96), SN is the pre-determined value of sensitivity (90%), *d* is the maximum marginal error (5%), and *P* is the disease prevalence in the study site (35%). Based on the proportion of TB cases in children among all cases in Indonesia, which is approximately 10%, we will recruit 40 children among all participants. The lower limit in age for child-participants is 4 years old, as they are considered able to follow the instructions for breath collection. In case the participants cannot produce sputum for the Xpert MTB/Rif examination, they will be asked to collect their stool samples, as this method has been shown as reliable in diagnosing TB. Based on the same formula as above, with Z_∝/2 value of 1.96, SN value of 90%, d value of 5%, and *P* value of 10%, the minimum number of study participants needed in the screening phase is 1383 patients. We will recruit the participants consecutively. Based on the proportion of TB cases in children, the number of children recruited is 140 among all participants. ## Analysis plan ### Microbiological analysis The samples that are taken will be analyzed with acid-fast-bacilli Ziehl-Neelsen microscopy and Xpert MTB/Rif, which are conducted according to the WHO and manufacturer guidelines, respectively, in the study sites’ laboratories. When the facilities are unavailable, the samples are referred to the TB-Microbiology laboratory, Faculty of Medicine, Public Health, and Nursing, Universitas Gadjah Mada. ### Data interpretation of eNose-TB Multivariate (chemometric) data analysis uses the open-source programming language R version 3.5.1 and Python version 3.7. The feature extraction method is conducted by taking the average value of each sensor’s sampling data. Evaluation of all datasets using SVM model resulted in the confusion matrix. The data is divided into two classes, *i*.*e*., the class predicting the classifier’s results and the actual class. In measuring performance using a confusion matrix, four conditions represent the results of the classification process, namely true positive (TP), true negative (TN), false positive (FP), and false-negative (FN). A ROC curve is used to perform analysis of 2 label classes. The caret library is also used in the ROC analysis procedure. ### Statistical analysis In the first phase (validation phase), we will calculate the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the breath test using the Xpert MTB/Rif as the reference standard. In each variable (such as age, body mass index), one stratum’s ROC-curve indicating the breath test’s sensitivity and specificity is compared with another stratum’s ROC-curve. An association between the breath test’s variable and sensitivity- specificity is indicated by a significant difference of an AUC between strata (*p* \< 0.05). In the second phase (screening phase), the performance of screening with a breath test will be compared with screening with clinical symptoms or CXR examination by calculating positive and negative agreements between the breath test and clinical symptoms or CXR examination. The time and cost of a screening algorithm with eNose-TB to obtain additional detection of one TB case will be calculated as the mean of time needed and mean of cost spent from the beginning of screening with the eNose-TB until the detection of the case. Statistical analysis is performed using STATA/SE 15 (License: Universitas Gadjah Mada). ## Ethical considerations The research will be conducted following the principles of the Helsinki Declaration of 2013 and Good Clinical Practice. The research has obtained the approval of the institutional review board at the Faculty of Medicine, Public Health, and Nursing, Universitas Gadjah Mada, Yogyakarta, Indonesia (KE/FK/1127/EC/2020) and has been registered in clinicaltrials.gov (NCT04567498). All subjects will sign an informed consent before participating in this study. In the validation phase, the attending physician at the study sites will provide information to patients regarding this study and offer patients a chance to meet one of the research teams or staff at the hospital trained to provide research information. The research team or staff provides information about this research verbally, based on what is written in the informed consent form. In the screening phase, nurses in the mobile clinic team will provide information to prospective participants about this study. All participants need to sign the informed consent form to participate in this research, and they are given 24 hours to decide to participate in this study. Subjects can leave the study at any time for any reason if they wish to do so without any consequences. The investigator can decide to withdraw a subject from the study for urgent medical reasons. If they withdraw, they will be diagnosed and treated with the standard treatment according to local guidelines. The breath sampling is conducted for a maximum of three breaths consecutively to avoid a hyperventilation-induced response when someone deep inhales and exhales 6–10 breaths consecutively. If a hyperventilation-induced response happens, the health officer will ask the participants to have rest by sitting or lying down for a while. If the subjects suffer from any adverse event, the investigator will treat them until they recover. Participants will not be paid to part in this research, but if there are costs incurred by participants to participate in this study or due to any adverse events, these costs will be compensated. The investigator will submit a summary of the study progress to the accredited institutional review board once a year and submit a protocol amendment immediately when the investigator makes one. Information will be provided on the date of inclusion of the first subject, numbers of subjects included and numbers of subjects that have completed the study, serious adverse events/serious adverse reactions, other problems, and amendments. The research team will store all data in a safe, locked cabinet, and only the research team has access to this data, or legal authorities and audit officers. The legal authorities and audit officers who are independent of the sponsor have access to these data and its analysis and can decide to terminate the study. The audit is conducted once a year by visiting the study site and examining the study data. Physicians who are not involved in the study (independent physicians) are also provided access to answer the prospective participants’ questions. The data will be made available upon study completion in keeping with the PLOS Data policy. The investigators will communicate the study results via publication and report to the sponsor and study sites. ## Differences from the study protocol Indonesia was written as the country with the third-largest burden of TB in the study protocol, as the 2020 edition of the WHO global TB report was not yet issued at the time of study protocol submission to the institutional review board. Meanwhile, when we submitted this manuscript, we used the latest data from the 2020 edition of the WHO global TB report. In this manuscript, we included the latest figure of the schematic circuit of the eNose-TB system, while the study protocol still used the previous figure of the eNose-TB system. In this manuscript, we wrote the number of child participants that will be recruited and added information regarding collecting and managing reported adverse events and other unintended effects of study interventions, plan of an audit of the study conduct, and the dissemination policy. # Discussion This is the first study protocol investigating the potency of an eNose-TB as a screening tool for TB. Based on this screening study’s findings, the extent of the potential of eNose-TB as a screening tool for TB, and the time and cost analysis of a screening algorithm with an eNose-TB would be known. The findings will enable the stakeholders and health professionals to implement effective screening strategies. The study is conducted during the time of the COVID-19 pandemic; thus, additional precautions are taken, such as using the appropriate protective personal equipment in collecting the samples, placing the eNose-TB device in a particular isolation room, using disposable equipment (HEPA-filter, connector, and air collecting bag), applying disinfectant procedure after each breath sampling, and ensuring that the physical distancing measurement is applied during the sampling collection. # Supporting information We would like to thank the staff at Surakarta General Hospital and the primary health centers for their cooperation for the upcoming study. 10.1371/journal.pone.0249689.r001 Decision Letter 0 Quinn Frederick Academic Editor 2021 Frederick Quinn This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 22 Jan 2021 PONE-D-20-33305 eNose-TB: a trial study protocol of electronic nose for tuberculosis screening in Indonesia PLOS ONE Dear Dr. Mahendradhata, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript. If you will need significantly more time to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Frederick Quinn Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: [http://journals.plos.org/plosone/s/supporting-information](about:blank). 3\. Please include additional information regarding the interview guide used in the study (lines 259-267) and ensure that you have provided sufficient details that others could replicate the analyses. For instance, if you developed the guide as part of this study and it is not under a copyright more restrictive than CC-BY, please include a copy, in both the original language and English, as Supporting Information. 4.We note that Figure(s) \[7, 8 and 9\] in your submission contain copyrighted images. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For more information, see our copyright guidelines: <http://journals.plos.org/plosone/s/licenses-and- copyright>. We require you to either (1) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (2) remove the figures from your submission: a.         You may seek permission from the original copyright holder of Figure(s) \[7, 8 and 9\] to publish the content specifically under the CC BY 4.0 license. We recommend that you contact the original copyright holder with the Content Permission Form (<http://journals.plos.org/plosone/s/file?id=7c09/content- permission-form.pdf>) and the following text: “I request permission for the open-access journal PLOS ONE to publish XXX under the Creative Commons Attribution License (CCAL) CC BY 4.0 (<http://creativecommons.org/licenses/by/4.0/>). Please be aware that this license allows unrestricted use and distribution, even commercially, by third parties. Please reply and provide explicit written permission to publish XXX under a CC BY license and complete the attached form.” Please upload the completed Content Permission Form or other proof of granted permissions as an "Other" file with your submission.  In the figure caption of the copyrighted figure, please include the following text: “Reprinted from \[ref\] under a CC BY license, with permission from \[name of publisher\], original copyright \[original copyright year\].”   b.    If you are unable to obtain permission from the original copyright holder to publish these figures under the CC BY 4.0 license or if the copyright holder’s requirements are incompatible with the CC BY 4.0 license, please either i) remove the figure or ii) supply a replacement figure that complies with the CC BY 4.0 license. Please check copyright information on all replacement figures and update the figure caption with source information. If applicable, please specify in the figure caption text when a figure is similar but not identical to the original image and is therefore for illustrative purposes only. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Does the manuscript provide a valid rationale for the proposed study, with clearly identified and justified research questions? The research question outlined is expected to address a valid academic problem or topic and contribute to the base of knowledge in the field. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Is the protocol technically sound and planned in a manner that will lead to a meaningful outcome and allow testing the stated hypotheses? The manuscript should describe the methods in sufficient detail to prevent undisclosed flexibility in the experimental procedure or analysis pipeline, including sufficient outcome-neutral conditions (e.g. necessary controls, absence of floor or ceiling effects) to test the proposed hypotheses and a statistical power analysis where applicable. As there may be aspects of the methodology and analysis which can only be refined once the work is undertaken, authors should outline potential assumptions and explicitly describe what aspects of the proposed analyses, if any, are exploratory. Reviewer \#1: Yes Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 3\. Is the methodology feasible and described in sufficient detail to allow the work to be replicable? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors described where all data underlying the findings will be made available when the study is complete? The [PLOS Data policy](https://journals.plos.org/plosone/s/materials-and- software-sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception, at the time of publication. The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? *PLOS ONE* does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above and, if applicable, provide comments about issues authors must address before this protocol can be accepted for publication. You may also include additional comments for the author, including concerns about research or publication ethics. You may also provide optional suggestions and comments to authors that they might find helpful in planning their study. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This is an excellent plan addressing an important question, and making use of a plan in progress to thrive for a zero-TB city of Yogyakarta. The plan is carefully designed, with a pre-specified sample size to develop and validate the test; and next, to sue is for screening for TB. 1\) If I understand it correctly, the measurement of breath samples collected will be processed by e-nose on the spot; if that is correct, how will the safety precautions to prevent the transmission of TB and Covid-19 be secured? will all sites have a dedicated room for e-nose measurements? 2\) the protocol aims at enrolling 40 children; will there be a loer limit in age, e.g., 10 years and over, to make sure that the child-participants have adult-phenotype pulmonary TB and that these young participants can follow the instructions for breath collection, and produce sputum for GeneXpert? 3\) who is the sponsor of the study? 4\) who is the manufacturer of the e-nose device, and the software? 5\) you plan to complete data collection by December, 2012; what if Covid-19 jeopardizes your plans? textual suggestions: page 4, line 82: due to infections from TB: change into 'due to TB' page 8, line 172: unable to: change into inability to..  page 13, line 287: Afterward,  .. change into: Next,  ..  line 292: As the anticipation of Covid transmission... considering the risks of TB and Covid-19 transmission..  page 14, line 303: smoking habits; my impression was that you plan to exclude smokers; please explain. line 310: 'when was last time you smoked' (I guess you will be asking questions in local languages or Bahasa Indonesia; when was the last time..(add 'the') line 313: 'when did you last take food or beaverages' page 21, line 459; is it a vagal reflex, or perhaps more likely, a hyperventilation-induced response, caused by hypocarbia and respiratory alkalosis? I think that this is what should be anticipated; Trendelenburg positioning is not indicated then, and just a bit of rest, sitting or lying down a while will settle the case. This is much easier than Trendelenburg positioning which will also be more uncomfortable for participants of the study. Reviewer \#2: The objectives are good. TB is a particular problem, especially amongst childrene and it i very difficult to separate from other symptoms to give the right interventions for control strategies. This manuscript is very confusing. It is all written in the present tense. It should be in the past tense as the study, I assume has already been done. It is very difficult to understand the initial study and then the validation study. Were these done in sequence or was the same set of data used for both?? Most of the Figures are totally irrelevant to the objectives. Where are the PCS, where are the clustered dendograms or similar to demonstrate that it is possible to distinguis using their enose the TB vs no TB cohorts and the difference in VOC profiles?? Most of the Figures are about the software system and how the screen pages in the instrument look. These are irrelevant How good is the replication of the sensors used in this instrument? How many sensors are used?? What is the range of detection of different VOC groups. Are they complimentary or what?? None of this information is present. Were sputum samples taken and analysed to confirm TB or no TB patients?? Most of the relevant data around these questions are not included. What about the sensitivity of metal oxide sensors to relative humidity. How was this overcome??? The key scientific aspects are ignored while the process of the software in the instrument are focused on. Unfortunately, this does not help in understanding the Results. This manuscript needs to be completely rewritten in a format which is understandable and approaches the work systematically. I assume the work was all done and needs to be worded in the past tense. The work is interesting and would be useful but the present version is really not up to scratch and needs very major rewriting and presentation. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0249689.r002 Author response to Decision Letter 0 8 Mar 2021 We have uploaded a file providing point-by-point response to the comments we've received 10.1371/journal.pone.0249689.r003 Decision Letter 1 Quinn Frederick Academic Editor 2021 Frederick Quinn This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 23 Mar 2021 eNose-TB: a trial study protocol of electronic nose for tuberculosis screening in Indonesia PONE-D-20-33305R1 Dear Dr. Mahendradhata, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Frederick Quinn Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Does the manuscript provide a valid rationale for the proposed study, with clearly identified and justified research questions? The research question outlined is expected to address a valid academic problem or topic and contribute to the base of knowledge in the field. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 2\. Is the protocol technically sound and planned in a manner that will lead to a meaningful outcome and allow testing the stated hypotheses? The manuscript should describe the methods in sufficient detail to prevent undisclosed flexibility in the experimental procedure or analysis pipeline, including sufficient outcome-neutral conditions (e.g. necessary controls, absence of floor or ceiling effects) to test the proposed hypotheses and a statistical power analysis where applicable. As there may be aspects of the methodology and analysis which can only be refined once the work is undertaken, authors should outline potential assumptions and explicitly describe what aspects of the proposed analyses, if any, are exploratory. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Is the methodology feasible and described in sufficient detail to allow the work to be replicable? Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors described where all data underlying the findings will be made available when the study is complete? The [PLOS Data policy](https://journals.plos.org/plosone/s/materials-and- software-sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception, at the time of publication. The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? *PLOS ONE* does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above and, if applicable, provide comments about issues authors must address before this protocol can be accepted for publication. You may also include additional comments for the author, including concerns about research or publication ethics. You may also provide optional suggestions and comments to authors that they might find helpful in planning their study. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: All my questions and queries have been appropriately addressed in the revised version; the current revision provides a clear plan for the planned research \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No 10.1371/journal.pone.0249689.r004 Acceptance letter Quinn Frederick Academic Editor 2021 Frederick Quinn This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 6 Apr 2021 PONE-D-20-33305R1 eNose-TB: a trial study protocol of electronic nose for tuberculosis screening in Indonesia Dear Dr. Mahendradhata: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Frederick Quinn Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Plasma lipids and lipoproteins concentrations are important risk factors for atherosclerosis and related vascular diseases. Twin and family studies suggest that about 50% of the variation in plasma lipid and lipoprotein levels is genetically determined. A search for the genetic contributions to variation in plasma lipid and lipoprotein levels has been ongoing for several decades. Since 2007, genome-wide association (GWA) studies have obtained great success and have implicated common variants in numerous loci and genes as being the genetic influences underlying lipid and lipoprotein levels,. Kathiresan's group recently performed a comprehensive meta-analysis of the GWA studies and identified 95 loci significantly associated with blood lipids. These associations were primarily found in European ancestry. Following studies successfully replicated most of these loci in East Asian population. However, some of the loci showed no significant association with lipid and lipoprotein levels in Asian population, including the *MAFB*, *NCAN/CLIP2/PBX4* and *MVK/MMAB* loci. It is therefore important to confirm whether known loci have consistent effects across ethnic groups. The purpose of this study was to replicate the previously reported genetic loci in the Chinese population. We firstly evaluated association between lipid levels and 15 loci selected from the three recent GWAS reports, in Cohort1, comprising 2, 533 Chinese individuals, and selected 10 out of the 15 loci on the basis of the strength of statistical evidence. We then tested association of the 10 loci with lipid traits in Cohort2, comprising 2,105 individuals, to confirm the findings in Cohort1. Finally, we combined the two cohorts results together, since both the two cohort were from Shanghai, with similar genetic background. Clinically, the most important plasma lipids and lipoproteins are triglycerides (TG), total cholesterol (TC), high density lipoprotein (HDL) cholesterol and low density lipoprotein (LDL) cholesterol. Several studies have suggested that the lipid ratio (TC/HDL-C) has greater independent predictive value for coronary heart disease (CHD) and cardiovascular events than either total cholesterol or LDL cholesterol levels. We therefore focused on five lipid traits: TG, TC, HDL cholesterol, LDL cholesterol and the TC/HDL ratio in this study. # Methods ## Ethics Statement The ethics committee of the Shanghai Institute for Biological Sciences approved this study. Written consents were given by the patients. ## Participants Participants in the present study comprised two groups, Cohort1 and Cohort2. Cohort1 was primarily designed for a case-control study of type 2 diabetes (including 1,360 non-type 2 diabetes controls and 1,173 type 2 diabetes patients). Cohort2 was a community-based prospective epidemiologic cohort of 2,105 subjects. Individuals known to be on lipid-lowering therapy were excluded. Both cohorts were recruited from Shanghai, China. The characteristics of participants are summarized in. For all individuals, height, weight, hip and waist circumference and blood pressure were measured by trained medical professionals using a standardized protocol. Body mass index (BMI) was calculated as weight (kg)/\[height (m)\]². Blood samples were collected after an overnight fast. Total cholesterol, HDL cholesterol, LDL cholesterol, triglycerides, and fasting plasma glucose (FPG) were measured enzymatically according to standard methods on the Roche modular P800 autoanalyzer (Roche, Mannheim, Germany) with the appropriate reagents (Roche Diagnostics CmbH, Mannheim, Germany). ## Selection of candidate variants We selected 15 out of 55 single nucleotide polymorphisms (SNPs) that achieved genome-wide statistical significance in three recently published GWA studies. Three criteria were adopted to choose SNPs: 1. Only one lead SNP was selected for each locus. For instance, three GWAS reported seven SNPs in the *FADS1/2/3* cluster, including rs174570, rs174537, rs102275, rs174556, rs1535, rs174546, and rs174547. Considering all were in a region with extremely high degree of linkage disequilibrium (D' = 1.0, r² = 1.0, The International HapMap Project), we selected one variant, rs174546, to represent this region. 2. We selected only those SNPs with a minor allele frequency higher than 5% in Chinese population to ensure that this study had enough statistical power. 3. Loci had been studied in Chinese population were excluded. ## Genotyping High-molecular-weight genomic DNA was prepared from venous blood using the QuickGene 610 L Automatic DNA/RNA Extraction System (Fijifilm, Tokyo, Japan). All genotyping experiments were done using TaqMan technology on an ABI7900 system (Applied Biosystems, Foster City, California). The standard 5 µl polymerase chain reaction (PCR) reactions were carried out using TaqMan Universal PCR Master Mix reagent kits under the guidelines provided. Genotype data were obtained in 97.5% of the DNA samples. Replicate quality control samples (5% samples) were included and genotyped with 100% concordance. ## Statistical analyses SHEsis was used to perform the Hardy-Weinberg Equilibrium (HWE) test. We assumed an additive model of inheritance, and conducted multiple linear regressions to assess the effect of the number of the specified allele of each SNP on five traits—concentrations of TC, TG, LDL cholesterol, HDL cholesterol and the TC/HDL ratio. Age, gender, body mass index (BMI), and type 2 diabetes status were included in the multiple linear regression models as covariant. Plasma TG and TC/HDL-C were logarithmically transformed before linear regression due to skewed distributions. We did not exclude the type 2 diabetes subjects, in line with previous genome-wide studies. Considering the potential correlation between diabetes and lipid levels, we included the diabetes status in the association model as a covariant. Multiple testing corrections were performed in stage 2 and combined analysis, *P*\<0.001 was considered as significant, given there were 10 SNPs and 5 traits analyzed. # Results ## Stage 1 replication 15 SNPs in 15 loci were genotyped in Cohort 1 comprising 2533 individuals at stage 1. Genotype distribution of each SNP did not deviate from Hardy-Weinberg equilibrium at the 5% level. Results of the multiple linear regression analysis adjusted for age, gender, BMI, and type 2 diabetes status are shown in. Of the 15 SNPs tested, 8 SNPs including rs10889353 in *DOCK7*, rs1501908 in *TIMD4-HAVCR1*, rs2954029 in *TRIB1*, rs1883025 in *ABCA1*, rs964184 in *APO(A1/C3/A4/A5)*, rs2338104 in *MMAB-MVK*, rs2650000 in *HNF1A*, and rs157580 in *TOMM40-APOE*, showed significant association (P\<0.05) with at least one lipid level trait. However, 7 other loci including rs10903129 located in *TMEM57,* rs12670798 in *DNAH11*, rs4936883 in *LIPG*, rs2304130 in *NCAN*, rs7120118 in *NR1H3*, rs6102059 in *MAFB*, and rs174546 in *FADS1/2*, which had previously been reported to be associated with plasma lipid levels in European ancestry, did not show evidence for association with any of the five lipid traits in the our Chinese sample. Given that the *FADS1/2/3* cluster was reported to be associated with lipid concentrations in all of the three GWAS papers, we included this locus in the stage 2 replication. Rs6102059 in *MAFB* showed a relatively low *P* value with TC (*P*\<0.09), therefore was also included. ## Stage 2 replication The 10 selected SNPs were subsequently genotyped in Cohort2, and 4 of them showed significant association with lipid traits after multiple testing correction (*P*\<0.001), including variants in *TIMD4-HAVCR1, TRIB1, ABCA1* and *APO(A1/C3/A4/A5)*. ## Combined analysis Considering the two cohorts were from the same place, we combined the two cohorts together to enhance the statistic power. We found three variants, including rs10889353 in *DOCK7* (Combined *P*∼6.5×10⁻⁴), rs2954029 in *TRIB1* (Combined *P*∼5.8×10⁻⁶) and rs1883025 in *ABCA1* (Combined *P*∼4.0×10⁻⁴), associated with total cholesterol concentration. Three variants, including rs10889353 in *DOCK7* (Combined *P*∼5.9×10⁻⁴), rs2954029 in *TRIB1* (Combined *P*∼2.3×10⁻⁷) and rs964184 in *APO(A1/C3/A4/A5)* (Combined *P*∼2.8×10⁻²⁸), showed significant association with triglyceride concentrations. Two variants including rs1883025 in *ABCA1* (Combined *P*∼2.0×10⁻⁵) and rs964184 in *APO(A1/C3/A4/A5)* (Combined *P*∼3.0×10⁻¹¹), showed association with HDL cholesterol concentrations. Three variants including rs2954029 in *TRIB1* (Combined *P*∼7.1×10⁻⁴), rs964184 in *APO(A1/C3/A4/A5)* (Combined *P*∼4.6×10⁻⁶) and rs157580 in *TOMM40-APOE* (Combined *P*∼2.0×10⁻⁸) showed significant association with LDL cholesterol concentrations. Three variants including rs1501908 in *TIMD4-HAVCR1* (Combined P∼1.9×10⁻⁵), rs2954029 in *TRIB1* (Combined *P*∼6.0×10⁻⁶) and rs964184 in *APO(A1/C3/A4/A5)* (Combined *P*∼2.5×10⁻⁹) showed association with the ratio of total cholesterol to HDL cholesterol. We found marginal association between rs174546 in the *FADS1/2/3* and triglycerides (Combined *P*\<0.01). We found no significant association between rs2338104 in *MMAN-MVK*, rs2650000 in *HNF1A* or rs6102059 in *MAFB* and plasma lipid levels either in Cohort2 or in the combined Cohorts. # Discussion In this study, we investigated whether the results of three independent genome- wide European association studies on plasma lipid and lipoprotein levels were replicatable in the Chinese population. Of the 15 loci selected from the European GWAS reports, 7 loci were successfully replicated. The most significant association was found between rs964184 in the *APO(A1/C3/A4/A5)* cluster and triglycerides (Combined *P*∼2.8×10⁻²⁸). This variant was also found to be associated with HDL cholesterol, LDL cholesterol, and TC/HDL. Differences in the G allele frequency (0.22 vs. 0.14) and effects on TG (0.14 vs. 0.30) were found between Chinese and Europeans, suggesting a higher risk allele frequency and weaker effect in Chinese population. The *APO(A1/C3/A4/A5)* cluster encodes important regulators of fasting lipids, and there is considerable evidence suggesting that variants in this region are associated with altered lipid metabolism. Fine mapping in this region may help us to find the functional variant. Another similar case is rs157580 in *TOMM4-APOE,* which encodes Apolipoprotein E, a main apoprotein of the chylomicron, essential for the normal catabolism of triglyceride-rich lipoprotein constituents. We found the A allele of rs157580 was associated with decreased LDL, TC and increased TG in Chinese, which was different from European population (A allele was associated with increased LDL, TC and TG). Two other SNPs rs4420638 and rs439401 in this region were reported to be associated with blood lipid profile in both Europeans and Chinese, and rs439401 (not included in this study) also showed different effect direction on LDL and TC between the two ethnic groups. Given the different allele frequencies (A allele of rs157580, 0.44 vs. 0.67) and different linkage equilibrium patterns of this region between Chinese and Europeans, these discordant results across ethnic groups could be explained by different linkage patterns between the causal variants and the tag SNPs that were studied. Nevertheless, these results confirmed the involvement of variants of this gene cluster in the lipid metabolism. We found variant in *ABCA1* associated with TC, TG and HDL. The association with TG was newly identified and retained significant after adjusting by other lipid traits (data not shown), suggesting variant in *ABCA1* gene played a wider effect on blood lipid profile in the Chinese population. Recently, Acuna-Alonzo et al reported a functional *ABCA1* gene variant exclusive to Native American and descent populations is associated with low HDL cholesterol levels and shows evidence of positive selection. These findings suggest the importance of *ABCA1* genetic variants in lipid metabolism. We also found variant in *DOCK7* associated with TC and TG, variant in *TIMD4-HAVCR1* associated with TC, TG, LDL and TC/HDL, variant in *TRIB1* associated with TC, TG, LDL, and TC/HDL, with similar effects and same effect directions on blood lipid traits as previous studies in Europeans. Our study confirmed that these loci are implicated in lipid metabolism in the Chinese as well as the European populations. The *FADS1/2/3* cluster locates on 11q12 encoding fatty acid desaturases, which convert polyunsaturated fatty acids into cell signaling metabolites and are functionally involved in lipid metabolism. Previous European studies found variants in the *FADS1/2/3* cluster to be associated with plasma concentrations of TG, TC, HDL cholesterol and LDL cholesterol. We found rs174546 in this locus marginally associated with TG (*P*∼0.01) in Chinese population, with same effect direction to that in Europeans. This result is consistent with another East Asian study, which reported a SNP in FADS1/2 is associated with TG in Japanese and associated with LDL in Mongolian. The linkage disequilibrium patterns of this region in the Chinese population somewhat differ with Europeans. For instance, the LD value between rs174546 and rs174570 is much higher in Chinese (D' = 1.0, r² = 1.0, HapMap, CHB) than in Europeans (D' = 1.0, r² = 0.32, HapMap, CEU). The different linkage disequilibrium pattern may therefore explain different association profile across the two ethnic groups. Rs2338104 in *MMAB-MVK* was reported to be associated with HDL cholesterol in studies by Willer and Kathiresan, but in our study this variant showed no association with HDL cholesterol, which is consistent with a large scale Japanese study. Similarly, variants in *NR1H3*, *LIPG*, *DNAH11*, *HNF1A* and *MAFB* didn't show significant association with blood lipid traits in our study. It is not necessarily the case that these loci do not influence lipid phenotypes in Chinese. One possible reason is that because of the modest effect sizes of the individual genetic variants on lipid traits our sample size is not enough to detect the association. Another possible reason is the different linkage disequilibrium pattern in Europeans and East Asian population. It is possible that these genes may influence lipid levels through other polymorphisms in East Asian populations. Fine mapping these regions by deep sequencing or additional screening of dense arrays would be needed to reveal association between these genes and lipid levels in the Chinese population. In conclusion, we successfully replicated association between 7 loci and plasma lipid concentrations in the Chinese population. Our study confirmed the implication of *APO(A1/C3/A4/A5), TOMM40-APOE, ABCA1, DOCK7, TIMD4-HAVCR1, TRIB1* and *FADS1/2* in plasma lipid and lipoprotein concentrations in Chinese population. # Supporting Information We warmly thank all the participants for their cooperation in this study and gratefully acknowledge all the physicians and nurses who assisted in studying the patients. [^1]: Conceived and designed the experiments: LH YL. Performed the experiments: ZZ LT ZC. Analyzed the data: ZZ LT. Contributed reagents/materials/analysis tools: ZZ LT ZC. Wrote the paper: ZZ LT. Revised the manuscript: D. Zhou MK D. Zhang CL. [^2]: The authors have declared that no competing interests exist.

# Introduction Sarcopenia was originally defined as age-related loss of muscle mass. Recently, the European Working Group on Sarcopenia in Older People (EWGSOP) has updated the operational definition of sarcopenia as a progressive and generalized skeletal muscle disorder that is associated with adverse outcomes including physical disability and mortality. In its 2019 revised definition, EWGSOP2 uses low muscle strength as the primary parameter of sarcopenia and the diagnosis is confirmed by the presence of low muscle quantity or quality. In view of the method of diagnosing sarcopenia being still complex and considered difficult to introduce into routine practice, the EWGSOP2 advises the use of the SARC-F questionnaire as a means of finding individuals with probable sarcopenia so as to carry out its assessment and provide treatment in clinical practice. The SARC-F is a symptom score based on 5 self-reported questions concerning strength, ambulation, rising up from a chair, climbing up a set of stairs, and falls. In longitudinal studies, it has been demonstrated to predict the adverse consequences associated with sarcopenia, such as physical disability, hospitalization, and mortality. Despite SARC-F being easy to conduct, inexpensive, and validated in different populations, its sensitivity is relatively low, as confirmed in a recent meta-analysis. To overcome this limitation, some authors have combined use of the SARC-F with other features in order to optimize the diagnostic properties of this screening tool (SARC-Calf combining calf circumference and SARC-F+EBM adding age and body mass). In the same way, Ishii test was devised so as to estimate the probability of sarcopenia by using a score based on three variables—age, grip strength, and calf circumference. Systemic sclerosis (SSc) is a rare multisystem autoimmune disease characterized by widespread vasculopathy and progressive fibrosis of the skin and other internal organs such as lungs, gastrointestinal tract, and kidneys. As a systemic inflammatory condition, prominently affecting patients' physical function and nutrition, SSc may be considered a major risk factor for sarcopenia. According to different definitions, sarcopenia has been diagnosed in nearly 20% of patients with SSc, which is similar to other rheumatic diseases, such as psoriatic arthritis (20%), rheumatoid arthritis (20.8%), and ankylosing spondylitis (22.7%). Considering that patients with SSc are particularly prone to develop severe clinical complications associated with comorbid sarcopenia, such as physical function decline and death, and that there are several case-finding instruments available not yet validated for SSc, we aimed to compare the sensitivity of SARC-F, SARC-CalF, SARC-F+EBM, and Ishii test as screening tools for sarcopenia in patients with SSc. In addition, we aimed to estimate the other standard measures of diagnostic accuracy and the area under the receiver operating characteristic (ROC) curves as the measurements to describe the accuracy of each screening test. # Methods ## Patients and study design A total of 142 consecutive patients with SSc were evaluated between March and December 2019, in a cross-sectional study carried out on a convenience sample of patients diagnosed with SSc followed up at a public university hospital. For an expected prevalence of sarcopenia of 20% in a sample of 94 patients with SSc, we could estimate a power greater than 80% to find a sensitivity ranging from 50% to 85%. Additionally, a *post hoc* calculation retrieved a power of 99% for sensitivity and 54% for specificity. A study flowchart is presented in. All patients were Brazilian and the vast majority inhabitants of the urban area of Porto Alegre, RS. A standardized and comprehensive research questionnaire was applied to each participant by the same researcher (VH). Disease duration was defined as time from the first non-Raynaud's symptom. Disease subtype was classified as follows: diffuse cutaneous SSc (involving trunk and acral skin), limited cutaneous SSc (restricted to extremities and/or face), or sine scleroderma. The severity of skin disease was evaluated by using the modified Rodnan skin score. Patients also completed the SARC-F questionnaire and data about calf circumference, body mass index, and handgrip strength were collected. Inclusion criterion was the fulfillment of either one of the two mostly used classification criteria for SSc: the ACR/EULAR 2013 classification criteria for SSc and the LeRoy/Medsger 2001 classification criteria for early SSc. Out of the 94 participants, 2 were classified as early SSc patients according to LeRoy/Medsger criteria and 92 were classified as SSc patients according to ACR/EULAR 2013 criteria. Exclusion criteria were: (1) the presence of any overlapping systemic autoimmune disease, (2) severe renal disease, defined as a glomerular filtration rate less than 30ml/min/1.73m², (3) any liver disease, defined as an elevation of aspartate aminotransferase or alanine aminotransferase above three times the upper limit of normal, (4) any chronic infection (e.g., hepatitis C virus, hepatitis B virus, human immunodeficiency virus), (5) severe chronic obstructive pulmonary disease, defined as forced expiratory volume in one second less than 50% of the predicted value, (6) any concomitant malignancy, and (7) any inflammatory myopathy, defined as previous history of myopathy and/or an elevation of creatine phosphokinase CPK or aldolase above 1.5 times the upper limit of normal. This study was conducted according to the principles expressed in the Declaration of Helsinki, all patients signed written informed consent and this research protocol was approved by the institutional Research Ethics Committee of the Hospital de Clínicas de Porto Alegre/Brazil (CAAE 06473019.0.0000.5327). ## Measurements Body mass was measured with a calibrated digital scale, with participants standing barefoot and wearing light clothes. Body height was measured with a standard fixed stadiometer. Body mass index (BMI) was calculated as weight (kg) per height (m²). Maximal calf circumference was measured as the widest circumference of the right calf with the legs relaxed and feet 20 cm apart from each other with an inextensible tape measure, according to the methods previously described. An anthropometric scale with a resolution of 100 g (Filizola S.A. Pesagem e Automação, São Paulo, Brazil), a 1 mm precision stadiometer, and a 1 mm precision measuring tape were used for these measurements. Handgrip strength was measured using a handheld dynamometer (Jamar Hydraulic Hand Dynamometer, Preston, USA) according to the methods proposed by Roberts *et al*. Patients had to squeeze the device as hard as they could three times in each hand in an alternating manner, and the maximum strength was defined as the highest of the 6 values. Cut-off points to define low strength were \<27 kg for men and \<16 kg for women according to the EWGSOP2. The Short Physical Performance Battery (SPPB) was applied to evaluate physical performance. It consists of three separate tests: balance, 4 m gait speed and chair stand test. In the balance test, the patient holds his balance for 10 seconds in three standing positions with eyes open: feet side by side, feet in semi-tandem stance, and feet in tandem stance. Only one attempt was permitted for each stance. In the gait speed test, patients walk a 4-m marked course at their usual walking pace, with the examiner timing their walk with a stopwatch. Two attempts were allowed on this test, with the fastest recorded time being used for the overall score. The chair stand test examines the ability to rise from a sitting to a standing position from an armless chair, with the arms folded across the chest. In the final part of the SPPB, a series of five consecutive chair stands, which should be performed as quickly as possible. The examiner times the patient’s performance with a stopwatch, counting aloud the number of stands completed. A score between 0 and 4 was assigned for each component, reaching a maximum of 12 points. According to the EWGSOP2, SPPB ≤8 defines low physical performance. Body composition was measured by whole-body dual-energy X-ray absorptiometry (DXA) (Lunar Prodigy Primo, GE Medical Systems, UK). Patients were wearing only underwear and were asked to remove all metal accessories and jewelry before measurements, which were taken in the morning. After that, patients were aligned in the center of the densitometer table with the feet positioned together and with the hands positioned with palms flat against the densitometer table (for larger subjects who do not fit within the constraints of the scanning field, hands were placed laterally against the hips). Only one patient had knee prosthesis. Following the EWGOSP2 consensus recommendations to use lean soft tissue assessed by DXA to infer muscle mass quantity, the appendicular skeletal muscle mass index (ASMI) was calculated as appendicular skeletal muscle mass (the sum of the muscle mass in both arms and legs) divided by height squared. Considering the cut-off points recommended by the EWGSOP2, men with an ASMI below 7.0 kg/m² and women below 5.5 kg/m² were defined as presenting low muscle quantity. ## Assessment of sarcopenia (EWGSOP2) Sarcopenia was defined according to the 2019 revised EWGSOP2 criteria. This definition uses low muscle strength (determined by handgrip strength) as the primary parameter of sarcopenia and the diagnosis is confirmed by the presence of low muscle mass (determined by DXA). In the presence of low muscle strength, low muscle quantity and low physical performance (determined by SPPB), sarcopenia is considered severe. ## Sarcopenia screening tools We used the SARC-F, SARC-CalF, SARC-F+EBM, and Ishii screening test to estimate the presence of sarcopenia. The standard SARC-F is composed of 5 items questioning the strength, assistance in walking, rise from a chair, climb stairs, and self reported number of falls in the last year (each one scored between 0 and 2). The score ranges from 0–10 and, in the original study, a score equal to or greater than 4 was predictive of sarcopenia and poor outcomes. The original SARC-F questionnaire was already translated to Portuguese and validated as a sarcopenia screening tool in Brazil with the optimal cut-off point equal to or greater than 6. In the current study, we applied this validated version; however, due to the lack of cut-offs standardization, we performed separate analyses and chose the value with a better performance in our specific sample which was equal to or greater than 4. The SARC-CalF is composed of 6 items: the standard SARC-F (5 items: strength, walking ability, rising from a chair, stair climbing, and self reported number of falls in the last year) and a sixth additional item (maximal calf circumference). Calf circumference is measured through scoring: zero representing the absence of low muscle mass (\>34 cm for men and \>33 cm for women) and 10 for presence (≤34 cm for men and ≤33 cm for women). The score ranges from 0–20. For the SARC-CalF, a total score of ≥11 indicates positive screening for sarcopenia. The SARC-F+EBM is a score that combines SARC-F with data about age and BMI. For age, patients with \< 75 years of age scored zero point, whereas ≥ 75 years of age scored 10 points. For BMI, patients not being underweight (\>21 kg/m²) scored zero point, whereas underweight (≤21 kg/m²) patients scored 10 points. The score ranges from 0–30. For the SARC-F+EBM, a total score of ≥12 indicates positive screening for sarcopenia. The Ishii screening test calculates the probability of sarcopenia based on three selected variables: age, grip strength and calf circumference. The formula to calculate the score is as follows: score in men = 0.62 (age– 64)– 3.09 (grip strength– 50)– 4.64 (calf circumference– 42), score in women = 0.80 (age– 64)– 5.09 (grip strength– 34)– 3.28 (calf circumference– 42). Alternatively, this score could be easily obtained from the values of the three variables combined on a simple score chart in each sex. For the Ishii test, a total score of ≥105 in men and ≥120 in women is suggestive of sarcopenia. ## Statistical analysis Statistical analyses were performed by using the Statistical Package for the Social Sciences version 23.0 (SPSS Statistics; IBM, Armonk, NY) and MedCalc Statistical Software version 16.8.4 (MedCalc Software, Ostend, Belgium). Variables with a normal distribution were presented as mean and standard deviation (SD), and non-normal quantitative variables were presented as the median and interquartile range (IQR). Sensitivity, specificity, positive likelihood ratio (+LR), negative likelihood ratio (-LR), positive predictive value (PPV), negative predictive value (NPV), and diagnostic *Odds Ratio* (DOR) were computed by using the EWGSOP2 criteria as the gold standard for diagnosis of sarcopenia. The diagnostic accuracy of the SARC-F, SARC-CalF, SAR-F+EBM, and Ishii screening tests were calculated so as to identify sarcopenia. The overall accuracy of screening tests was evaluated by ROC curves. The area under the ROC curve (AUC) and 95% confidence interval (CI) were calculated for all tests and Youden's J statistics was used to compare the performance of SARC-F with different cut-off values. An AUC greater than 0.9 has high accuracy, whereas 0.7 and 0.9 indicate moderate accuracy, 0.5 and 0.7 low accuracy, and 0.5 a chance result. To compare sensitivity, specificity, +LR, -LR, PPV, NPV, and AUC of the screening tests, we used one-way ANOVA with Tukey HSD ("Honestly Significant Difference") *post-hoc* test to indicate which groups were significantly different from others. There were no missing values of any variable in the entire analytic sample. All statistical tests were 2-sided. A *p* value of less than 0.05 was considered statistically significant. # Results Out of 142 patients evaluated initially, 37 were excluded for not meeting inclusion criteria. Of these 105 patients, 11 patients refused to participate, remaining a total of 94 patients diagnosed with SSc (7 men and 87 women). The mean age mean age of the total sample was 60.5±10.3 years (range 33–79 years of age). shows the clinical characteristics of patients with SSc stratified by sex. Sarcopenia was identified in 15 patients with SSc (15.9%) by the EWGSOP2 criteria and severe sarcopenia in 5 patients (5.3%). Average (SD) scores for screening tools were: SARC-F 2.56 (1.84), SARC-CalF 5.12 (4.96), SARC-F+EBM 5.01 (5.12), Ishii test 96.31 (37.75). Concerning the ability to evaluate sarcopenia, the ROC curves of the four screening tests against the EWGSOP2 definition of sarcopenia are shown in. presents the results of sensitivity/specificity analysis and AUC of these tests in the whole study population by using EWGSOP2 diagnostic criteria as the reference standard. The raw data of each tool vs. the gold standard diagnostic results were provided in 2x2 tables as. Area under the ROC curve of SARC-F screening for sarcopenia was 0.588 (95% CI 0.420–0.756, *p* = 0.283). The SARC-F results of sensitivity, specificity, +LR, -LR, and DOR with the EWGSOP2 criteria as the gold standard were 40.0% \[95% CI, 19.8–64.2\], 81.0% (95% CI, 71.0–88.1), 2.11 (95% CI, 0.98–4.55), 0.74 (95% CI, 0.48–1.13) and 2.84 (95% CI, 0.88–9.22), respectively. The optimal cut-off point of SARC-F in our sample was ≥4 (Youden index: 0.21), the same cut-off point recommended in the literature. Only 6 (40%) out of the 15 participants with sarcopenia were identified by the SARC-F questionnaire in our population. However, the SARC-F properly identified 4 out of 5 patients who had severe sarcopenia. As summarized in, the magnitude of the sensitivity could vary widely: from 40% for the SARC-F alone to 86.7% for the Ishii screening test. SARC-CalF showed better sensitivity (53.3%, 95% CI 30.1–75.2) and better specificity (84.8%, 95% CI 75.3–91.1) compared with SARC-F. The same occurred with the SARC-F + EBM, that presented better sensitivity (60.0%, 95% CI 35.7–80.2) and also a slightly better specificity (86.1%, 95% CI 76.8–92.0) than SARC-F alone and SARC-CalF. The best sensitivity (86.7%, 95% CI 62.1–96.3) and the best NPV (96.7%, 95% CI 88.8–99.1) were obtained with the screening test of Ishii *et al*, at the expense of a relatively small loss of specificity (73.4%, 95% CI 62.7–81.9). In contrast, the most specific tool was the SARC-F+EBM (86.1%, 95% CI 76.8–92.0), which also presented de highest +LR (4.31, 95% CI 2.17–8.56) and PPV (45%, 95% CI 29.2–61.9). Comparing the aforementioned ROC curves, SARC-F performed worse than the SARC- CalF, SARC-F+EBM and Ishii test as a sarcopenia screening tool (AUCs 0.588 vs. 0.718, 0.832, and 0.862, respectively). Additionally, when all tests were evaluated together, there were no differences among screening tests for specificity (*p* = 0.156), PPV (*p* = 0.473), NPV (*p* = 0.077), +LR (*p* = 0.639) and -LR (*p* = 0.098), whereas sensitivity (*p* = 0.020) and AUC (*p* = 0.026) were statistically different. *Post-hoc* direct comparisons between tests revealed differences only between SARC-F and Ishii test for sensitivity (*p* = 0.013) and AUC (*p* = 0.031). # Discussion An ideal screening test has to combine a reasonably high sensitivity to find cases in the tested population with a relative high specificity to reduce the number of false positives, avoiding unnecessary and expensive investigations. Aligned with results of previous reports, our study demonstrated that SARC-F presents a poor sensitivity but a high specificity. Also, in a recent meta- analysis including 7 studies (12,800 subjects), the pooled results of SARC-F sensitivity, specificity, and DOR with the EWGSOP first criteria as the gold standard were 21% (95% CI, 13–31), 90% (95% CI, 83–94), and 2.47 (95% CI, 1.64–3.74), respectively(10). In our sample, 15.9% of patients with SSc had sarcopenia, according to the EWGSOP2 criteria (gold standard), and SARC-F sensitivity, specificity and DOR were 40% (95% CI, 19.8–64.2), 81% (95% CI, 71.0–88.1) and 2.84 (95% CI, 0.88–9.22), respectively. Even though our findings indicate a relatively greater sensitivity, it still lacks clinical utility as 60% of patients with sarcopenia will test negative on SARC-F. The highest sensitivity was from Ishii test, according to which 13.3% of patients with sarcopenia will test negative. A screening tool for sarcopenia presenting high sensitivity is important for prompt identification of patients at risk in clinical practice, allowing to start at the earliest diagnostic confirmation and preventive strategies. On the other hand, the diagnostic accuracy of a screening tool also could be assessed using the AUC value. According to this approach, the observed performance of SARC-F as a screening tool for sarcopenia (AUC 0.588) is, therefore, considered insufficient, suggesting that SARC-F questionnaire is not an adequate tool for sarcopenia screening in patients with SSc. In our study, SARC-CalF, SARC-F+EBM, and Ishii test proved to be superior to SARC-F alone for sarcopenia screening, all of them presenting AUC greater than 0.7. In practical terms, PPV indicates the probability of having sarcopenia when the test is positive and the NPV, the probability of not having sarcopenia when the test is negative. Generally, previous studies have indicated higher NPV than PPV for sarcopenia screening tools. In our study, SARC-F presented the lowest PPV and NPV, and SARC-F+EBM the highest PPV and Ishii test the highest NPV. In 2016, the (Brazilian) Portuguese-translated version of the SARC-F questionnaire was validated in a population-based study. These authors also proposed to improve its efficacy by associating SARC-F to calf circumference, as an estimate of muscle mass. The SARC-CalF significantly improved SARC-F’s screening performance (AUC 0.736 vs. 0.592, *p* = 0.027), with a substantial increase in sensitivity (SARC-F 33% vs. SARC-CalF 66%) without compromising the remaining parameters. In a recent meta-analysis, including 5 studies (1,127 participants), the pooled results of sensitivity, specificity, and AUC with the EWGSOP first criteria as the gold standard were 58% (95% CI 46–70), 87% (95% CI 84–90), and 0.860 (95% CI 0.83‒0.89), respectively. In our study SARC-CalF also presented a significantly higher sensitivity, specificity and AUC compared to SARC-F alone in patients with SSc (53%, 84% and 0.718, respectively). Adopting a different approach, Kurita *et al* proposed to add “EBM” (“elderly” and “body mass” index information) to SARC-F in order to improve its diagnostic accuracy in patients with musculoskeletal disease. Using the EWGSOP2 criteria as the reference standard, SARC-F+EBM presented higher sensitivity (84.2% vs. 47.4%) and AUC (0.876 vs. 0.558) than SARC-F alone. Thus, the authors suggested that SARC-F+EBM may be a better approach to finding cases of sarcopenia in patients with musculoskeletal disease. In our study, SARC-F+EBM also presented significantly higher sensitivity and AUC than SARC-F in patients with SSc (60% and 0.832, respectively) and also the best specificity, +LR and PPV among the other tests evaluated (86%, 4.31, and 45%, respectively). In the original validation study of SARC-F+EBM, patients were selected after referral for spinal surgery or knee or hip replacement therapy and osteoarthritis was the most common diagnosis. Even though SSc patients may present associated osteoarthritis, in the present study severe functional limitation due to osteoarthritis was not frequent. Considering the specific clinical features of SSc that may contribute to sarcopenia, such as skin thickening and interstitial lung disease, we understand that the best performance of SARC-F+EBM in our study is not predominantly due to similarities with the original study's population. Aware of these SARC-F’s limitations, EWGSOP2 consensus mentions that clinicians may prefer a more formal case-finding tool to be used in populations where sarcopenia is likely, suggesting the Ishii screening test as an option in this setting. Applying this method, the probability of sarcopenia could be easily obtained from a score chart in each sex, combining three variables—age, grip strength, and calf circumference. When the sum of sensitivity and specificity was maximized, sensitivity, specificity, and AUC for sarcopenia were 85%, 88%, and 0.939 for men, and 75%, 92%, and 0.909 for women, respectively. In our sample, the Ishii test also presented the best sensitivity (87%), NPV (96.7%) and–LR (0.182), at the expense of a small decrease in specificity (73%). An important aspect to be considered is the choice of the cut-off values for sarcopenia definition. According to the EWGSOP2 consensus, reference values were provided to increase harmonization of sarcopenia studies. In a previous regional study, Barbosa-Silva *et al*. used a different cut-off for ASMI, since the value recommended by EWGSOP2 consensus was not able to identify low muscle mass within their sample. In contrast to the study by Barbosa-Silva *et al*, the present study, using EWGSOP2 consensus reference values, identified a prevalence of sarcopenia similar to those reported in previous studies of SSc patients. Therefore, instead of using the adapted cut-off values, we chose to report our findings using the reference values recommended by EWGSOP2. To the best of our knowledge, the present study was the first attempt to evaluate the diagnostic accuracy of the SARC-F questionnaire in a sample of patients diagnosed with SSc. As previously described, our results confirmed the low sensitivity of SARC-F and the better diagnostic accuracy of other tests compared to SARC-F, but in a different population with a high reported prevalence of sarcopenia. Therefore, we understand that the SARC-F+EBM combines the best set of diagnostic properties with the easiest application into clinical practice since it does not depend on the handgrip strength as Ishii test (dynamometers are widely available in research centers, but hardly ever present in doctors’ offices). In the context of personalized medicine, the proper choice of a screening strategy using easily applicable tools could provide relevant diagnostic information about sarcopenia in patients with SSc. Our study should be interpreted within its limitations. The sample size may not be large enough to detect some differences in accuracy of the screening tests in some subgroups of patients, especially among men (only 7 patients) and non- Caucasian (only 17 patients). Also, a limited sample size could be the reason why there was no difference for most diagnostic measures among the tests, as only sensitivity and AUC were different between SARC-F and Ishii test. In addition, due to our cross-sectional design it was not possible to address the direct impact of a positive screening test in disability, hospitalizations and mortality, as previously shown in other studies. Moreover, our comprehensive exclusion criteria could potentially cause selection bias and limit our findings' external validity. Finally, considering the clinical features of our sample that may interfere on sarcopenia measures, such as skin thickening, joint disease, interstitial lung disease and pulmonary hypertension, we acknowledge the limitations of using previously validated tools on a different population and encourage the development of specific tests for SSc patients. # Conclusion In view of sensitivity, PPV, +LR, -LR, DOR and AUC, SARC-CalF, SARC-F+EBM, and Ishii test performed better than SARC-F alone as screening tools for sarcopenia in patients with SSc diagnosed by EWGSOP2 criteria. Only specificity and NPV were greater in SARC-F. Considering diagnostic accuracy and feasibility aspects, SARC-F+EBM seems to be the most suitable screening tool to be adopted in routine care of patients with SSc. These findings need validation in larger samples and different settings, preferably in a longitudinal design to assess the prognostic properties of each screening test. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction In the central nervous system (CNS), myelin arises from oligodendrocytes (OLGs), which proceed through a regulated pathway that assembles the components of the myelin membrane. Myelination commences with differentiation of the bipolar early oligodendrocyte progenitor cell (OPC), and culminates with copious synthesis of classic myelin basic protein (MBP) isoforms and proteolipid protein (PLP), when extensive processes form and extend around an axon. Compact myelin is formed by flattening of the multiple, spirally-wrapped lamellae with the extrusion of cytoplasm, a process modeled in various ways, *e.g.*, the “liquid croissant” and “corkscrew” models. The amount of white matter in the brain increases with evolutionary complexity. Myelin continues to be formed until the early twenties in humans, and remodeling continues throughout adulthood in the healthy CNS. Multiple sclerosis (MS) is a disease that is characterized by inflammatory demyelination of axons, for which the molecular mechanism has remained unknown over 150 years since its first major clinical documentation. An “inside-out” model suggests that multiple sclerosis results from a cytodegenerative process aimed at the oligodendrocyte-myelin complex –: a process of gradual physical demyelination can then lead to an autoimmune response and a cycle of further degeneration, characteristic of the most common relapsing-remitting manifestation of multiple sclerosis. For many reasons, it is essential to attain an understanding of myelin formation and architecture at the molecular level in order to comprehend the causes and pathogenesis of this debilitating disease, as well as fundamental aspects of brain development and modeling. One of the most studied candidate auto-antigens in multiple sclerosis is the classic 18.5-kDa isoform of MBP, which is essential to the stability of central nervous system myelin where it plays numerous roles both in myelin development and homeostasis, acting both to adhere membrane leaflets to each other, and as a hub in protein-protein and protein-membrane interaction networks. The latter include cytoskeletal proteins such as actin and tubulin, as well as signaling proteins such as calcium-activated calmodulin and SH3-domain containing proteins. The murine 18.5-kDa MBP isoform has 168 amino acids, and is intrinsically disordered, like all members of this protein family. There are, however, three segments of the protein that become α-helical in the presence of lipids or membrane-mimetic solvents such as trifluoroethanol (TFE),. These three segments that undergo this specific disorder-to-order transition are denoted by us here as the α₁-segment (murine 18.5-kDa residues T33-D46), α₂-segment (P82-I90), and α₃-segment (Y142-L154), respectively. In addition to being membrane-anchoring motifs, these segments can also “moonlight” as protein-protein interaction sites. *In vivo*, 18.5-kDa MBP undergoes extensive post-translational modifications, which play a role in its ability to interact with a wide variety of partners. One of the most important post-translational modifications in MBP is phosphorylation by mitogen-activated and other protein kinases (reviewed in). Phosphorylation of MBP is altered during development and ageing, and the overall level is decreased in multiple sclerosis,. Phosphorylated MBP is associated with less compact myelin, and has been shown to be developmentally partitioned into detergent-resistant microdomains. The phosphorylation of MBP also protects it from proteases such as trypsin, which suggests that this modification alters the protein’s conformation. *In vitro* MAP-kinase phosphorylation of 18.5-kDa MBP occurs sequentially at T92 and T95 (murine 18.5-kDa isoform numbering), and modulates the ability of MBP to polymerise actin and tubulin, and to bundle microfilaments and microtubules. Additionally, pseudo-phosphorylations (Thr to Glu substitutions) at these sites have been shown to affect the protein’s intracellular trafficking and its interactions with the non-receptor tyrosine kinase Fyn in transfected cells, and to inhibit the binding of MBP to the SH3-domain of Fyn *in vitro*. The T92 and T95 MAP-kinase sites in 18.5-kDa MBP are located within a central, proline-rich, highly-conserved region encoded by classic exons III and IV. This region can form a poly-proline type II (PPII) structure *in vitro*, and we have demonstrated that it represents an important protein interaction motif for SH3-domains such as those of Fyn ,. In all mammalian species, this proline-rich region in MBP is immediately adjacent to the α₂-segment. The α₂-segment represents a primary immunodominant epitope in multiple sclerosis; it associates with the phospholipid membrane as an amphipathic α-helix, presenting the PPII structure formed by the proline-rich region to the cytoplasm for potential protein-protein interaction. Previous molecular dynamics (MD) simulations by us demonstrated that phosphorylation at the MAP-kinase sites within the proline-rich region influences the interaction of this amphipathic α-helix with the membrane which could alter the orientation of the PPII structure in the cytoplasm. Given that the MAP-kinase sites located within the proline-rich region of MBP are also only a few residues removed from the α₂-segment, it is feasible that MAP-kinase phosphorylation could directly influence the conformation, dynamics, and stability of both the PPII and the α-helical structural elements. We have proposed that the central region of MBP encompassing the amphipathic α₂-helical segment, and the adjacent PPII structure, constitute an important molecular switch in which membrane-MBP interactions, as well as MBP-protein interactions, are controlled by phosphorylation/dephosphorylation at the two MAP-kinase sites. In this study, we seek to gain insights into the possible mechanisms of this molecular switch by probing the effects of phosphorylation at the MAP-kinase sites on the stability of the α₂-helix, as well as on the conformation and dynamics of the PPII-structure. We have utilized four different 36-residue peptides (murine 18.5-kDa MBP residues S72-S107, denoted the α₂-peptide) that contain the α₂-helical segment (P82-I90) as well as the proline-rich segment (T92-S99), and that have varying phosphorylation status at T92 and T95. We have determined and compared the free-energy change of disorder-to-α-helical transition in the unmodified and phosphorylated peptides using equilibrium TFE- titration curves monitored by circular dichroism (CD) spectroscopy. We also present new solution NMR spectroscopic data characterizing the structure of the unmodified α₂-peptide in dilute aqueous conditions, building on our previous solution NMR spectroscopic studies in which the structure of the unmodified α₂-peptide in phospholipid and lysophospholipid environments had been evaluated. The spectroscopic experiments described here were complemented by MD studies on the identical unmodified and phosphorylated α₂-peptides in water, as well as associated with dimyristoylphosphatidylcholine (DMPC) bilayers, that extend our previous MD experiments on shorter 24-residue peptides (murine 18.5-kDa residues E80-G103). Overall, our results indicate that the MBP α₂-peptide in aqueous environment does not readily form an α-helical structure, nor does it form PPII structure in the proline-rich region, in contrast to the α₂-peptide in a lipid environment where both conformations are observed. Additionally, we find that phosphorylation at the MAP-kinase sites inhibits the formation of α-helical structure, and also affects the global conformation of the peptides through altered intramolecular electrostatic interactions. The results further our understanding of a critical region in 18.5-kDa MBP, and enhance our appreciation of possible consequences of aberrant amino acid substitutions or post-translational modification that may be involved in myelin destabilization during development,. # Materials and Methods ## Expression, Purification, and Construction of α₂-peptide Variants Unlabeled as well as fully ¹³C-¹⁵N-labelled recombinant murine MBP 36-residue α₂-peptide (S72–S107, murine 18.5-kDa sequence numbering – see) was expressed as a SUMO-fusion, and purified by chromatography as described previously. This construct represented the (i) unmodified peptide variant. The synthetic phosphorylated variants of the α₂-peptide were (ii) PhT92 (phosphorylation at residue T92), (iii) PhT95 (phosphorylation at T95), and (iv) PhT92–PhT95 (double phosphorylation at residues T92 and T95), and were purchased from Biomatik (Cambridge, ON). All other materials were as previously described. ## Solution NMR Spectroscopy of the Recombinant α₂-peptide in Aqueous Buffer Unmodified α₂-peptide samples for solution NMR spectroscopy were prepared by dissolving ∼2 mg of uniformly ¹³C-¹⁵N-labelled α₂-peptide in a solution containing 100 mM NaCl, 20 mM HEPES-NaOH (pH 7.5), and 10% D₂O. The final sample volume was ∼600 μL, and the sample was maintained at 295K for all measurements. ## Sequence-specific Resonance Assignments and Structure Calculations High-resolution ¹H, ¹³C, and ¹⁵N NMR spectra, using several complementary pulse sequences, were recorded on a Bruker Avance spectrometer operating at a Larmor proton frequency of 600.13 MHz. In order to obtain residue-specific assignments, a two-dimensional ¹H-¹⁵N-HSQC (heteronuclear single-quantum coherence) experiment was conducted in addition to several triple-resonance NMR experiments as follows: (a) HNCO/HN(CA)CO to assign the amide proton (¹H_N\[*i*\]), the amide nitrogen (¹⁵N\[*i*\]), and the carboxyl carbon atoms of the current and preceding amino acid (¹³Ć\[*i*\], ¹³Ć\[*i-1*\]); (b) CBCA(CO)NH/HNCACB to assign the C_α and C_β atoms of the current and preceding amino acids; (c) HCCH-TOCSY (total correlation spectroscopy) to assign the remaining ¹H and ¹³C side-chain atoms; (d) HACAN to assign the prolyl residues ; and (e) two-dimensional ¹H-¹³C-HSQC spectra without carbon decoupling to extract heteronuclear ¹J_CαHα coupling constants, which allows for differentiation of the PPII and random coil conformations. Water suppression was achieved using the double-pulsed field gradient spin echo technique (excitation sculpting) with the carrier frequency set to the water ¹H signal. Detailed experimental spectroscopic parameters are given in. The ¹H chemical shifts were referenced directly to the methyl signal of DSS (2,2-dimethylsilapentane-5-sulphonic acid) in an external sample tube, whereas the ¹³C and ¹⁵N chemical shifts were referenced to DSS indirectly. The spectra were processed using NMRPipe. All free induction decays were zero-filled, and apodized using a shifted, squared sinusoidal bell function prior to Fourier transformation and subsequent phase-correction. The ¹H-¹⁵N-HSQC spectrum was zero-filled up to 2048 and 4096 complex points along *F*₁ and *F*₂, respectively. The HN(CA)CO, HNCACB, and their complementary spectra were zero-filled up to 256, 256, and 2048 complex points along *F*₁, *F*₂, and *F*₃, respectively. The HACAN spectrum was zero-filled up to 256, 256, and 2048 complex points along *F*₁, *F*₂, and *F*₃, respectively. The ¹H-¹³C-HSQC spectrum was zero-filled up to 4096 complex points along each of *F*₁ and *F*₂. The acquisition parameters for all spectra are listed in. Spin systems were then assigned using Computer Assisted Resonance Assignment (CARA, version 1.8.4), modules contained in the CARA software package ([www.nmr.ch](http://www.nmr.ch)), and a collection of in-house scripts that have been previously described. All spin systems were created on the basis of the ¹³C, ¹⁵N, and ¹H ppm values of ¹³Ć\[*i*-1\], ¹⁵N\[*i*\], and ¹H_N\[*i*\] of each system in the HNCO spectrum. All other spin systems were picked and partially assigned automatically using another set of in-house scripts. Spin systems that were not identified by the scripts were assigned manually. Sequence-specific connectivity was obtained manually using iterative trials. Secondary structure estimation by chemical shift index (CSI) analysis was performed as described. The assignments (H_α, C_α, C_β, C′, N, H_N) were also input into the Vendruscolo Laboratory’s δ2D software, which uses an algorithm designed to calculate the probability that any amino acid in a primarily disordered protein has a particular secondary structure conformation, based on the chemical shift assignments. The chemical shift data have been deposited into the Biological Magnetic Resonance Bank (BMRB ID 19186). ## Forward and Reverse TFE-titration Curves of the α₂-peptide The four peptides used in this study were each titrated with trifluoroethanol (TFE) by diluting concentrated peptide stock (typically ∼130 µM peptide) in water into different solutions containing successively higher TFE concentrations. The final solution conditions in all cases were: 15 µM peptide, 20 mM HEPES-NaOH, pH 7.4, and TFE concentrations ranging from 0 M to ∼7.7 M (0 to 55%, v/v). All samples were incubated for ∼16 hours at 25°C prior to measurement. The TFE-titration experiments were monitored by CD using a Jasco J-815 spectropolarimeter (Japan Spectroscopic, Tokyo, Japan) at a fixed wavelength of 222 nm using a quartz cuvette with a 1-mm path length, thermostatted at 25°C using a Jasco PTC-424S/15 Peltier temperature controller (Japan Spectroscopic, Tokyo, Japan). At this wavelength and temperature, based on buffer measurements, the non-peptide components of the solution gave virtually no CD signal, and the measured ellipticity was due exclusively to the peptide. A total of 30 readings of each sample were taken over a 30-second period (*i.e.*, one per second), and averaged in order to reduce scatter in the curves. At least 3 independent TFE- titration curves of each peptide were measured, with each curve analyzed individually (see below). Reverse TFE-titration curves of the peptides were also measured by making up a concentrated peptide solution in high TFE (∼50% v/v), followed by dilution to lower TFE concentrations. The reverse-titration samples were incubated and measured in the same way as the forward-titration curves. ## Quantitative Analysis of TFE-titration Curves The TFE-titration curves of the four α₂-peptide variants were fit to a 2-state equilibrium model: disordered↔α-helical, with equilibrium constant *K^TFE*. For this transition, a linear dependence of Δ*G* on TFE concentration was assumed, allowing the equilibrium constant to be defined by:where *R* is the universal gas constant, *T* is the temperature in Kelvin, Δ*G^TFE* is the free energy of the transition at a given concentration of TFE, is the free energy of the transition in the absence of TFE (*i.e.*, in dilute buffer), *m* is a measure of the dependence of Δ*G^TFE* on TFE concentration, and \[*TFE*\]*_mid* is the concentration of TFE at which the disorder-to-helical transition is half-completed. The data were fit to the following equation as previously described for a 2-state equilibrium process, :where, Yobs is the observed CD signal, YD and YH are the y-intercepts of the pre-transition and post-transition baselines, respectively, and SD and SH are the slopes of the pre-transition baseline (between 0 M and ∼1 M TFE), and the post-transition baseline (above ∼4 M TFE), respectively. Fitting to) was done using Microcal Origin version 8 (Northampton, MA). In order to reduce the number of fitted parameters, and to determine the values of *m* and \[*TFE*\]*_mid* more precisely, *Y*_D, *S_D*, *Y_H*, and *S_H* were determined by linear regression analysis, and were then set as fixed parameters in non-linear curve fitting. All values are reported as an average of at least 3 independent experiments, and reported errors are standard deviations. ## Molecular Dynamics (MD) Simulations of the MBP α₂-peptides The GROMACS 4.5.5 software package with the Gromos96 ffG53a6 force-field was used to perform molecular dynamics (MD) simulations using the SHARCNET high performance computer cluster ([www.sharcnet.ca](http://www.sharcnet.ca)). Peptide models used in simulations were constructed from the lowest energy structures obtained from our solution NMR experiments in dodecylphosphocholine (DPC) micelles (PDB ID 2LUG). This structure was used essentially without modification except for the addition of PO₄^2- groups to the T92 and T95 residues as appropriate using the SYBYL-X 1.3 molecular modeling suite (SYBYL, Tripos Associates, St. Louis, MO). The following 4 models were considered: (i) unmodified, representing the original solution NMR structure; (ii) singly-phosphorylated at Thr92 only (PhT92); (iii) singly-phosphorylated at Thr95 only (PhT95); and (iv) doubly-phosphorylated at both Thr92 and Thr95 (PhT92-PhT95). All four peptides were simulated in H₂O, as well as in a DMPC lipid bilayer system, as described below. Both the N- and C-termini of the peptide were uncharged, and all histidyl side chains were unprotonated. All peptide bonds were in the *trans* conformation. All simulations in DMPC were done in duplicate, whereas the simulations in water were done in triplicate due to their increased dynamics and variability under these conditions. ### Molecular dynamics simulations in H₂O The four α₂-peptides with varied phosphorylation states were simulated at 37°C in a cubic virtual box with dimensions 14×14×14 nm. Each peptide was positioned in the center of the box, and the box was subsequently solvated with water molecules using the spc216 model. The final density of the system was 997.2 g/L. To obtain an overall net charge of zero, Na⁺ or Cl^- counter-ions were added as appropriate. Subsequently, the solvated and neutralized system was energy-minimized to a maximum overall force of \<1,000 kJ/mol/nm using the steepest descent minimization algorithm with the *rcoulomb* and *rvdv* cut-offs set at 1.2 nm. The equilibration steps were done at 1 atm and 310 K, using Berendsen isotropic pressure coupling and velocity rescaling with a stochastic term (*v-rescale*) thermostat for temperature coupling. For the production MD runs, the time step (*dt*) was set to 0.002 ps, and the same pre-equilibration settings were used with respect to temperature and pressure couplings. The equilibrated system was simulated for total of 160 ns. ### Molecular dynamics simulations in DMPC bilayers Each of the four α₂-peptide variants was also simulated at 37°C in the presence of a DMPC lipid bilayer, surrounded by H₂O with counter- ions on both sides; a system that we have used previously. The peptides were initially positioned horizontally on top of the bilayer according to previous experimental data, with the hydrophobic residues F86 and F87 pointing towards the membrane. In order to position the peptide relative to the DMPC bilayer correctly, the peptide was first put in a vacuum box of the same dimensions as the DMPC lipid system. The box containing the peptide was merged with the DMPC reference membrane, and potential steric clashes were eliminated automatically using *genbox* with the default van der Waals radii distance threshold of 0.105 nm. The final merged simulation box had a volume of 828 nm³ with a density of 955 g/L; an overall net charge of zero for the system was obtained by adding Na⁺ or Cl^- counter-ions, as appropriate. The peptide-DMPC simulation box was then energy-minimized using the steepest descent algorithm to a final maximal force of 1,000 kJ mol⁻¹ nm⁻¹, and this procedure was followed by two equilibration steps in which the peptide was restrained to avoid potential structural distortion. The first equilibration step was carried out for 100 ps at constant number of particles, volume, and temperature (NVT). The second equilibration step was carried out for 1,000 ps at constant number of particles, pressure, and temperature (NPT). These equilibration steps were important for allowing the lipids and water molecules to accommodate the structured peptide without significant unfolding. The equilibrated and energy-minimized system was simulated using a 0.002 ps time step for a total of 160 ns at 37°C and 1 atm. All equilibration and simulation steps were thermally- and pressure-coupled using the same parameters as for the H₂O simulations described in the previous section. ### Molecular dynamics simulations with varying temperature The thermal unfolding propensity of the four energy-minimized and equilibrated structures of the unmodified, PhT92-, PhT95-, and PhT92-PhT95-α₂-peptides associated with the DMPC lipid bilayer were measured qualitatively using temperature ramps (*i.e.*, simulated annealing). Due to disruption of the DMPC bilayer at high temperatures, only the peptide was subjected to the temperature-ramp protocol, while keeping the DMPC bilayer, counter-ions and H₂O molecules at a constant temperature of 37°C throughout these experiments. These temperature-ramp simulations were performed for a total of 25 ns and involved, sequentially: (i) a linear gradient from 37°C to 500°C during the first 10 ns (temperature ramp of 46.3°C/ns), followed by (ii) maintenance at 500°C for a total of 5 ns, and finally (iii) a linear decrease at a rate of 46.3°C/ns in temperature from 500°C back to 37°C, over a 10-ns time interval. The same temperature and pressure coupling settings that were used in the isothermal H₂O and DMPC simulations were also applied in these temperature-ramp experiments, with a time step of 0.002 ps. The suitability of this protocol for assessing differences in α-helix stability of the peptides was assessed by performing validation experiments in which all the residues within the α-helical segment (P82-I90) of the unmodified peptide were mutated to alanine, valine, and glycine respectively, followed by temperature- ramping using the exact same protocol as outlined above. The results of these validation experiments revealed that the poly-alanine helix denatured at a higher temperature, followed by the poly-valine and poly-glycine helices (see), as expected based on differences in α-helical propensity of these residues. ## Analysis of MD Experiments The 160-ns trajectories from the H₂O and DMPC bilayer simulations were analyzed frame by frame (each frame equivalent to 10 ps) using GROMACS utilities. Secondary structure evolution was evaluated with the dictionary of protein secondary structure (DSSP) algorithm. The tilt dynamics of the α-helix (residues P82-I90) within the α₂-peptide (residues S72-S107– see) with respect to the DMPC bilayer surface were evaluated. The tilt is defined as the angle Θ between the helix axis passing through its center of mass, and the axis parallel to the plane of the phospholipid head groups of the membrane. The dihedral angles of each residue in the peptide, as well as the membrane penetration depth of the α-helix, were determined using VMD software and custom scripts written in the TCL language. In the calculation of membrane-penetration depth, the surface plane of the leaflet was defined by selecting the phosphates of the DMPC lipids, and the geometrical center of the α-helical segment of the peptide was determined using the α-carbons of the amino acid residues. Using this information, the membrane-penetration depth was thus expressed as the distance between the center of mass of the α-helix and the surface of the membrane. The presence of PPII structure in the peptides over the 160-ns trajectory was determined by creating a classifier function based on defined φ and ψ angle thresholds. Peptides were classified to adopt a PPII structure if at least 2 adjacent residues satisfied the following criteria as previously defined : (a) −46° ≤φ ≥ −104°, (b) 116° ≤ ψ ≥174°, and (c) ω = 0. # Results ## Solution NMR Spectroscopy of Recombinant α₂-peptide Previously, we have evaluated the conformation of the 36-residue α₂-peptide (murine 18.5-kDa residues S72–S107) in the presence of DPC micelles, and found that the peptide adopted a well-defined amphipathic α-helix in the region P82-I90, whereas the remainder of the protein was largely disordered. Separate CD experiments conducted on this peptide in aqueous conditions (see also), indicated that although the central α-helical structure is much less well defined, there could be some residual secondary structure present under these conditions. This result was consistent with solution NMR data obtained on the entire 18.5-kDa protein. Here, we have performed several solution NMR experiments on the α₂-peptide under aqueous conditions in order to characterize better any residual secondary structure, as well as to define more precisely the conformation of the proline-rich region, which is the probable SH3-ligand. Consistent with the overall intrinsically-disordered nature of the protein, the nitrogen-HSQC spectra obtained under aqueous conditions showed a high degree of degeneracy, and so additional experiments were needed in order to make full resonance assignments. We collected a series of three-dimensional and two- dimensional spectra, and used the same assignment strategy as previously for the α₂-peptide in DPC micelles. These data allowed us to resolve the degeneracy observed in the HSQC spectrum, and to obtain nearly complete ¹H, ¹⁵N, and ¹³C backbone assignments for the peptide. The ¹H_α, ¹³C_α, and ¹³C_β resonances were completely assigned, with the exception of residues S72 and Q73, which were not identified in any spectra. These assignments were used to confirm sequential connectivity of the residues. A complete list of resonance assignments is given in. Several different algorithms were used to identify any propensities of the α₂-peptide to form secondary structural elements. Based on the C_α chemical shift deviations from random coil values, we found that the peptide does not have any obvious secondary structural elements, given that there are no two sequential residues that are ±0.7 ppm of the established random coil values. The deviations were all sequence-corrected. The secondary structure propensity (SSP) algorithm designed by the Forman-Kay group and collaborators was also used to determine secondary structure elements in the unmodified α₂-peptide. An SSP score of 1 indicates that the residue is in an α-helical conformation, whereas −1 suggests β-structure. To account for the intrinsically-disordered nature of the α₂-peptide, only the ¹H_α, ¹³C_α, and ¹³C_β chemical shifts were considered for the calculations based on the recommendations of the designers. This algorithm also determined that the α₂-peptide had a high propensity to remain in the random coil conformation under these conditions, as most scores remained close to zero. Dihedral angle determination by TALOS+ also found the torsion angles of the α₂-peptide to be dynamic, with high deviations from the predicted ψ and φ angles for most of the residues. One last algorithm was also used to determine secondary structure propensities. The δ2D algorithm designed by the Vendruscolo laboratory outputs the probability of each residue adopting α-helical, β-strand, PPII, or random coil conformation. The results suggested a high percentage of random coil throughout the peptide, with up to ∼25% probability of PPII conformation at its N- and C-termini. The probability of PPII conformation within the proline-rich region was surprisingly near nil, and there was an added likelihood of β-structure (up to 30% probability) present in this region. It should be noted that the percentages are different from those obtained in DPC, where the highest probability of PPII conformation (up to ∼25%) is located directly in the proline-rich region, with very little propensity for PPII conformation observed elsewhere. As a complement to these secondary structure analyses based on chemical shifts, we also used coupling constants, namely ¹J_CαHα to differentiate between random coil and PPII conformations. In this method, the PPII conformation can be differentiated by observing the amide nitrogen deviation from random coil values, as well as the deviation from random coil values of coupling constants. We calculated the ¹J_CαHα coupling constants for the region that spanned residues V91-S99. This segment encompasses the proline-rich region of the α₂-peptide, and was selected because it was the only segment that fulfilled the criteria of having sequential residues with a deviation in the ¹⁵N resonance \>1.1 ppm of the corresponding random coil value. Given that the calculated coupling constants were almost all within 1.1 Hz of the random coil values for the respective residues, the torsion angles associated with the amino acids in this region suggest a random coil conformation. ## Analysis of CD Spectroscopic TFE-titration Curves The propensity of the largely disordered α₂-peptide to adopt α-helical structure was evaluated by performing quantitative TFE-titration curve experiments, as has been done for other proteins and peptides. Fitting of the titration curve data to obtain the change in free energy (Δ*G*) of the disorder↔α-helical transition defines the thermodynamic feasibility of α-helical formation, with low values of Δ*G* suggesting a more stable α-helical conformation. The MAP-kinase phosphorylation sites in MBP are at residues T92 and T95, immediately adjacent to the C-terminal end of the α-helical segment, but it is unclear whether phosphorylation at these sites has an effect on the stability of the α-helix. We had not specifically addressed the stability question *per se* in our previous MD study. The CD-monitored TFE-titration curves of all four α₂-peptides (unmodified, PhT92, PhT95, and PhT92–PhT95) are sigmoidal, with a single pronounced transition, consistent with a cooperative 2-state folding process that would be expected for a peptide of this size. A 2-state process is also suggested by the presence of an isodichroic point in full CD spectra at low wavelengths (∼202.5 nm for unmodified and ∼1–2 nm lower for the variants). Reverse-titration curves, in which the α₂-peptide in the presence of high TFE concentration is diluted to lower TFE concentration, are coincident with the forward-titration curves, indicating a reversible transition that can be analyzed in thermodynamic terms. The TFE-titration data were fit well to a 2-state equilibrium model, (disordered↔α-helical) defining Δ*G^H2O* (free energy of transition in the absence of TFE, *i.e.*, in aqueous buffer) and \[*TFE*\]*_mid* (the concentration of TFE at which the transition is 50% completed). The \[*TFE*\]*_mid* of the unmodified peptide was found to be 2.31 M (∼17% v/v), whereas the Δ*G^H2O* is 14.9 kJ mol⁻¹. The positive value of Δ*G^H2O* suggests that α-helical formation in aqueous conditions (in the absence of TFE) is unfavorable, which is consistent with the observation of very little α-helical content under these conditions by CD and NMR spectroscopy. The changes in the \[*TFE*\]*_mid* and Δ*G^H2O* values resulting from phosphorylation can be represented as Δ\[*TFE*\]*_mid* (determined as \[*TFE*\]*_mid* of phosphorylated variant – \[*TFE*\]*_mid* of unmodified peptide), and ΔΔ*G^H2O* (calculated as Δ*G^H2O* of phosphorylated variant – Δ*G^H2O* of unmodified peptide). The three phosphorylated variants all have positive Δ\[*TFE*\]*_mid* and ΔΔ*G^H2O* values, ranging from ∼0.2–0.4 M and ∼1.4–2.5 kJ mol⁻¹, respectively. The data thus suggest that phosphorylation at the MAP-kinase sites in MBP generally tend to disfavor formation of the α-helical-rich conformation, especially when phosphorylation occurs at residue T92. ## Molecular Dynamics Simulation Experiments The one unmodified and three phosphorylated 36-residue α₂-peptide variants were simulated in water as well as in DMPC bilayers at 37°C for 160 ns. These experiments expand upon our previous MD experiments on shorter 24-residue peptides (murine 18.5-kDa residues E80–G103). In these 24-residue peptides, the residues E80–V91 were modeled as an α-helix, whereas the remaining residues T92-G103 were initially modeled as an extended PPII structure to provide a starting point, since experimental structural coordinates were not available at that time. In the MD experiments presented here, the starting structures were derived from NMR experiments that determined the structure of the α₂-peptide in the presence of DPC micelles, with little modification. Given that the DPC and DMPC lipidic systems have similar characteristics, this NMR structure elucidated in DPC is expected to be a good starting structure for MD simulations. The starting structures of the α₂-peptides used here have a slightly shorter α-helical segment (P82-I90) than previously modeled in the starting structures of the 24-residue peptides. Additionally, the PPII structure formed by residues in the proline-rich region in the α₂-peptide is shorter compared to the modeled 24-residue peptide (comprising ∼7 residues versus 12 residues). Moreover, the N-terminal region (S72–V83) and the C-terminal region (T92-S107) of the α₂-peptides undergo transient intramolecular interactions to form a “closed” conformation, which makes the starting structures for the simulations presented here markedly more compact than the largely extended conformation of our previous 24-residue peptide models. ### Molecular dynamics simulations in water In water, the MD experiments reveal that the α₂-peptides are highly dynamic and the α-helix is unstable, as expected, based on Δ*G^H2O* values and solution NMR data. As the simulations progress, significant portions of the α-helix becomes distorted, transitioning to π-helical, turn, or random coil conformations. All of the peptides, regardless of phosphorylation status, remain somewhat compact even as the α-helical segment unfolds, with P96, P97, and P98 generally being partially buried throughout the simulations. For the unmodified as well as phosphorylated peptides, α-helix unfolding tends to start at the C-terminus, and unfolding is most apparent in the PhT92- and PhT92-PhT95-α₂-peptides. The residues N-terminal (S72-N81) and C-terminal (T92-S107) to the α-helical region (P82-I90) are highly dynamic. The N-terminal residues largely adopt coil and bend conformations throughout the simulations, whereas the C-terminal residues rapidly and reversibly transition between coil, bend, β-bridge, β-sheet, and PPII conformations. The flexibility of the terminal regions also facilitates the formation of stabilizing electrostatic interactions between the negative phosphate group(s) and different basic residues within the phosphorylated peptides (see Discussion). During the course of the MD experiments, the evolution of PPII structure in the proline-rich region, C-terminal to the α-helix, was analyzed quantitatively by observing the number of frames in which this structure is present. In this analysis, PPII structure is considered to be present if at least 2 consecutive residues adopted dihedral angles consistent with the PPII conformation threshold (as described in the section). In the unmodified peptide, the average number of PPII conformations present is ∼8/ns with similar results for the phosphorylated variants. This observation indicates that phosphorylation may not have a significant effect on the formation of short PPII structures within the proline- rich region. ### Molecular dynamics simulations on DMPC bilayers The four α₂-peptides were also simulated for 160 ns in association with a DMPC bilayer. In these experiments, the starting peptide models used were the same as for the simulations in water. The peptides were initially placed on top of the bilayer, oriented with the hydrophobic residues F86 and F87 pointing towards the bilayer. At the beginning of the simulations, the phosphate groups in the phosphorylated peptides form electrostatic interactions with basic residues (See Discussion) and the central α-helical segment of all the peptides penetrates the bilayer with an average depth of ∼6–8 Å (measuring from the geometrical center of the α-helix to the phospholipid head groups located at the membrane surface). This membrane penetration greatly stabilizes the α-helix, allowing it to persist mostly intact throughout the entire simulation. In the unmodified peptide, as well as the single phosphorylated variants, any tendency for α-helical unfolding in DMPC appears to be confined to the last helical turn at the C-terminal end. In contrast, the α-helix of the PhT92–PhT95 doubly- phosphorylated peptide tends to undergo more significant unfolding at the N-terminal end of the helix, with a relatively slight perturbation at the C-terminal end. A notable effect of phosphorylation is that it can alter the tilt of the helix in the DMPC membrane. For the unmodified α₂-peptide as well as the PhT95-variant, the α-helical segments (P82-I90) throughout the simulations are tilted such that the C-terminal portion of the helix is more submerged in the DMPC bilayer, whereas the N-terminal half is closer to the membrane surface (Θ\<0, negative tilt). However, the helical tilt is different in the PhT92- and especially in the PhT92–PhT95-α₂-peptide simulations, with the N-terminal half of the peptide tending to be more submerged into the bilayer, whereas the C-terminal half is at or above the membrane surface (Θ\>0, positive tilt). The average helix penetration depth of ∼6 Å for the PhT92- and PhT92–PhT95-α₂-peptides also tend to be slightly lower compared to the unmodified and PhT95-α₂-peptides, which have penetration depths of ∼8 Å. The disordered regions that flank the α-helix remain mostly solvent-exposed, and are quite dynamic in all the α₂-peptides in DMPC. There appears to be more transient formation of various structures (including β-bridge, β-sheet, bends, and PPII conformation) in the region C-terminal to the α-helix compared to the N-terminal disordered region. Similarly to the simulations in water, the observed number of PPII structures in the proline-rich region observed in the DMPC simulations is ∼8/ns for the unmodified as well as phosphorylated variants. ### Molecular dynamics thermal-ramp simulations In order to obtain information on thermal unfolding propensity of the α-helix within the α₂-peptide in DMPC, thermal-ramp experiments were carried out in which the four α₂-peptide variants were simulated for a total of 25 ns, with the experiment divided into three distinct steps: 10 ns with linear heating from 37°C to 500°C, followed by simulation for 5 ns at a constant temperature of 500°C, and finally by linear 10 ns cooling from 500°C back to 37°C. It should be noted that in these experiments, only the peptides were heated and the temperature of the bilayer was kept constant at 37°C, because the structure of the bilayer was observed to be severely and irreversibly disrupted at higher temperatures. In these temperature-ramp experiments, observed thermal unfolding of the α-helix of the unmodified α₂-peptide starts at ∼250°C. Interestingly, unfolding is reversed when the peptide is cooled from 500°C to 37°C during the final 10 ns step. The PhT95-α₂-peptide is apparently very stable to the effects of heat with only limited observed unfolding, primarily involving the first 2 residues at the N-terminal side of the helix. The α-helical unfolding in the PhT95-variant, however, is somewhat less reversible compared to the unmodified peptide. The α-helix of the PhT92-α₂-peptide is observed to undergo more significant unfolding compared to either the unmodified α₂-peptide or the PhT95-α₂-peptide, with only the middle 4 residues of the α-helix retaining their starting conformation. Furthermore, observed α-helical unfolding in the PhT92-α₂-peptide is almost totally irreversible. In thermal-ramp experiments, the α-helix of the PhT92–PhT95-α₂-peptide is the only one that is observed to completely unfold, and this unfolding is totally irreversible. Overall, these data suggest that phosphorylation at position T92 in the α₂-peptides reduces the apparent stability of the α-helix in a membrane environment. # Discussion ## The Proline-rich Region of MBP is Disordered in Water The proline-rich region in mammalian MBP (murine sequence –T92-P93-R94-T95-P96-P97-P98-S99-) is highly conserved (encoded by classic exon IV) and contains a minimal XP-X-XP SH3-ligand domain. We have previously demonstrated that MBP binds to several SH3-domains *in vitro*, and probably *in cellula*, including those of Fyn kinase, cortactin, and zonula occludens 1 (ZO-1). Co-expression of MBP with a constitutively-active form of Fyn-kinase caused extensive membrane elaboration, and branching complexity in cultured N19-oligodendrocyte membrane processes. This phenotype was abolished by Pro-to- Gly substitutions at either proline residue within the XP-X-XP SH3-ligand consensus motif in MBP, thereby directly implicating this region in SH3-domain binding. Additionally, isothermal titration calorimetry experiments revealed that MBP variants with Pro-to-Gly substitutions within the proline-rich region, as well as pseudo-phosphorylation at the MAP-kinase sites (T92E and T95E), have reduced enthalpy upon binding to the SH3-domain of Fyn kinase. More recently, using NMR spectroscopy, we have reported chemical shift perturbations within the proline-rich region when MBP α₂-peptide was titrated with the SH3-domain of Fyn. Given the strong experimental evidence suggesting that the proline-rich region is indeed an SH3-ligand, it is important to investigate its conformation and specifically to determine if this region spontaneously adopts PPII structure usually recognized by SH3-domains. Previously, we have shown by CD spectroscopy that an 18-residue peptide of MBP (F86-G103, murine sequence), containing the proline-rich region adopts some PPII structure in both aqueous buffer and with DPC micelles, that appears to be slightly stabilized from thermal denaturation by phosphorylation at T95. However, given that there is an abundance of evidence that non-proline-rich regions within disordered conformations can adopt PPII structure, we cannot un-ambiguously assign the observed PPII structure in the CD spectra to the proline-rich region due to potential contribution from the residues outside of this region. Thus, the NMR experiments reported here are important for conclusively determining the conformation of this proline-rich region *per se*. The NMR experiments on the unmodified α₂-peptide in water suggest that PPII structure is largely disfavored within the proline-rich region of MBP, with some PPII propensity only being present in the disordered regions at the N- and C-termini of the α₂-peptide, according to the δ2D algorithm. This observation is in contrast to the α₂-peptide in DPC micelles, where the δ2D algorithm indicates that the proline-rich region has some propensity to form PPII structure. Additionally, the secondary structure assignment methods XTLSSTR, PROSS, and SEGNO, all characterize the proline-rich region in the α₂-peptide in DPC as adopting a PPII conformation. These data thus suggest that PPII formation within the proline-rich region is highly sensitive to a lipid environment or, more likely, may be enhanced by the presence of the fully-formed adjacent α-helix. The MD experiments allowed us to investigate the effects of phosphorylation on PPII formation in the proline-rich region. The MD results have to be interpreted with caution as the simulation time scale in these experiments is relatively short at 160 ns, due to limitations in computational resources presently available to us. Nevertheless, if we assume that the kinetics of unfolding/refolding of the PPII structure in the proline-rich region is similar for the unmodified and the phosphorylated variants, valuable insights can be gained in comparing the number of PPII structures observed during MD simulation experiments. The observation that phosphorylation does not significantly affect the formation of PPII structures of length 2-residues in the proline-rich region suggests that any effect of phosphorylation on the binding of SH3-domains may only involve direct or indirect electrostatic effects (*vide infra*). ## The Central α-helix Stability is Reduced upon Phosphorylation Disorder-to-order transitions are a common and often necessary feature of IDPs in order to interact with their binding partners, and these transitions play an important role in the multifunctionality of these proteins (*e.g.*). The disordered regions or fragments within proteins that adopt defined secondary structure upon interaction with their binding partners have been termed molecular recognition fragments or elements (MoRFs or MoREs). The central (P82-I90) segment that is contained within the α₂-peptides used in this study is an unequivocal α-helical molecular recognition fragment (α-MoRF). This region is disordered in water, but becomes α-helical upon interaction with phospholipids and other surfactants, or in the presence of membrane-mimetic solvents such as TFE (and references therein). Similar to the methodology used for other IDPs and peptides,, we have used equilibrium TFE-titration curves to evaluate the free energy change of the disorder-to-α-helical transition of the unmodified α₂-peptide as well as three phosphorylated variants of it. The Δ*G^H2O* of 14.9 kJ mole⁻¹ for the disorder-to-α-helical transition of the unmodified α₂-peptide indicates that, although α-helical formation is unfavourable in water, the free energy difference between the disordered and α-helical states is relatively modest, increasing the likelihood of a small population of the α-helical conformation in aqueous conditions. The approach that we have adopted here allows for a systematic comparison of the effects of phosphorylation, at the two threonyl residues that lie just at or outside the C-terminal end of the α-helical region, on the stability of the α-helix. Phosphorylation has been shown to alter the stability of α-helices in other proteins in a position dependent manner. Whereas phosphorylation close to the N-terminal end of an α-helix tends to be stabilizing, phosphorylation close to the C-terminal end of an α-helix can destabilize the helix due to electrostatic repulsion between the negative phosphate group and the α-helix dipole. Consistent with this general principle, we find that the singly- phosphorylated MBP α₂-peptide at position T92, as well as the doubly- phosphorylated α₂-peptide at positions T92 and T95, both have markedly higher values of Δ*G^H2O*, suggesting that the disorder-to-α- helical transition is less favoured. The singly-phosphorylated PhT95-α₂-peptide also tends to have a higher Δ*G^H2O* value compared to the unmodified α₂-peptide, although this difference is smaller than for the two other phosphorylated α₂-peptides. These results thus suggest that phosphorylation at the T92 position has a greater destabilizing effect on α-helical stability compared to phosphorylation at position T95. This conclusion is supported qualitatively by our thermal-ramp MD experiments, where the α-helix of the unmodified and PhT95-α₂-peptides have higher apparent thermal stability, when associated with the DMPC membrane, than either of the PhT92- and PhT92–PhT95-α₂-peptides. These results might be expected, given that T92 is closer to the α-helix (comprising residues P82-I90) than T95, being only two residues removed from the C-terminal end of the helix. Furthermore, our α₂-peptide simulations in DMPC at physiological temperature demonstrate that an electrostatic interaction occurs between the phosphate group on T92 and the residue K88 in the middle of the central α-helix (see and below). Formation of this stable electrostatic interaction could cause strain in the α-helix, which could also contribute to destabilization. Given that attaining an α-helical conformation is essential for MBP-membrane interaction, the results suggest that phosphorylation could modulate 18.5-kDa MBP’s interaction with membranes by altering the energy landscape. There have been other studies on intrinsically-disordered proteins or regions that suggest that C-terminal phosphorylation can inhibit the disorder-to-α- helical transition, and hence potentially compromise the effectiveness of α-MoRFs. For example, it has been reported that seryl phosphorylation at the C-terminal end of a region with α-helical propensity in the LD4 motif of paxillin inhibits the disorder-to-α-helical transition. Similarly, it has been reported that seryl phosphorylation at the C-terminal end of an α-MoRF in the protein 4E-BP1 can modulate the free energy folding landscape, by shifting the disorder-to-order equilibrium towards the disordered species, regulating the interaction of 4E-BP1 with its partner el4E. Hence, phosphorylation particularly at the C-terminal end of α-MoRFs may be an important and general molecular switching mechanism, regulating binding interactions by disfavouring the necessary disorder-to-order transition. ## Phosphorylation Alters the Orientation of the α-helix with Respect to the Membrane Our previous electron paramagnetic resonance (EPR) and NMR experiments of full- length 18.5-kDa murine MBP had suggested that the α-helix within the molecular switch region of MBP is titled within the membrane such that the C-terminal end of the helix is submerged within the membrane, while the N-terminus of the helix is close to the membrane surface. This tilt might be energetically favourable, in part, because it would enable the disordered, hydrophilic region N-terminal to the α-helix to be in the aqueous phase and not be buried within the lipid bilayer. Our previous MD studies on a 24-residue peptide of MBP (E80-G103, murine 18.5-kDa sequence), representing the molecular switch region and containing this α-helix within it, showed similar results to the EPR and NMR experiments. These simulations further suggested that phosphorylation reverses the tilt angle of the helix such that the N-terminus becomes more submerged in the membrane whereas the C-terminus is closer to the surface of the membrane. The MD experiments presented here on the unmodified longer 36-residue α₂-peptide (S72–S107), comfortably comprising the molecular switch region, also demonstrate that the α-helix is tilted within the DMPC membrane such that the C-terminal end of the helix is more submerged. However, the α-helix within the unmodified 36-residue α₂-peptide has a less negative tilt angle (Θ) compared to the α-helix within the 24-residue peptide. The slight difference in the α-helical tilt between the 24- and 36-residue peptides could be due to the shorter α-helix in the 36-residue peptide (of length 9-residues versus 12-residues at the start of the simulation), and the presence of the longer disordered regions in the 36-residue peptide that flank the helix. In the 36-residue α₂-peptide simulations, the central α-helix tilt angle Θ changes slightly upon single phosphorylation at T92, becoming less negative and approaching zero. In the doubly-phosphorylated α₂-peptide, the tilt angle Θ is actually positive, meaning that the N-terminus of the helix is more submerged in the bilayer compared to the C-terminus. An α-helical tilt angle near zero or positive allows the highly charged phosphate group at T92 to be further from the membrane surface and closer to water, enabling more efficient solvation, and easier formation of stabilizing electrostatic interactions with basic residues in the water-exposed disordered regions (see below). A more positive tilt angle Θ is less important in the case of PhT95 because the phosphate group is further away from the helix, and hence it is unsurprising that the tilt angle of the helix within this PhT95-α₂-peptide is very similar to that of the unmodified α₂-peptide. The differences in tilt angles mediated by phosphorylation could affect how the N- and C- terminal loop regions of this central α-helix are presented to the cytoplasm for binding to protein partners. ## Phosphorylation Modulates Electrostatic Interactions within the Molecular Switch Region of MBP The molecular switch region of MBP, represented here by the 36-residue unmodified α₂-peptide, has an overall net charge of +3 (assuming uncharged histidyl residues) with 2 acidic and 5 basic residues. Given that all but one of the basic residues is found outside the central α-helical segment, there is a greater potential for phosphate groups to be stabilized by electrostatic interactions, due to the flexibility of these largely disordered regions. One difference between unmodified and phosphorylated α₂-peptides, in MD experiments conducted in water as well as in DMPC, is the pattern of electrostatic interactions observed. These electrostatic interactions are driven exclusively by the need to stabilize the −2 charge of each phosphate group. Whereas there are no observable stable electrostatic interactions that persist throughout the simulations in water for the unmodified α₂-peptide, the phosphate groups in all three phosphorylated α₂-peptides each form two or three electrostatic interactions. These electrostatic interactions reduce the conformational freedom of the peptides and tend to form early in the simulation, usually persisting throughout the simulations. In the PhT92-variant, these electrostatic interactions primarily involve the residues R94 and K88, and occasionally R104, with the side-chains of the residues clustering simultaneously around the phosphate group. In the PhT95-α₂-peptide, electrostatic interactions occur between the phosphate group at T95, and two basic residues simultaneously. These interactions involve either K88 and R76, or R94 and R104. In the PhT92–PhT95 doubly-phosphorylated α₂-peptide, a total of four electrostatic interactions occur, involving the two phosphate groups and basic residues in the peptide, with the phosphorylated T92 forming electrostatic interactions simultaneously with K88 and R94, whereas phosphorylated T95 forms electrostatic interactions with either R76 or R104. The R94 residue can also be re-oriented to form an electrostatic interaction with phosphorylated T95, leaving the phosphorylated T92 stabilized by just one electrostatic interaction. In MD simulations conducted in DMPC, all the α₂-peptides were observed to be far less dynamic compared to water. In the unmodified α₂-peptide, an electrostatic interaction occurs between E80, in the N-terminal disordered region, and K88 in the middle of the α-helix that appears to be of only moderate stability, tending to be only transiently formed throughout the simulation. This electrostatic interaction partially restrains the N-terminal disordered region, and may play a role in ensuring a correct orientation for binding interactions. This electrostatic interaction is absent in the phosphorylated α₂-peptides, due in part to the preference of the K88 side-chain to interact with the phosphate groups. During the simulation of the PhT92 peptide in DMPC, we observed electrostatic interactions between the phosphate group at T92 and K88, as well as with R94. The third electrostatic interaction between the phosphate group at T92 and R104 that occurs in water is not present in the DMPC condition, due to the C-terminal region being less dynamic. In contrast to the PhT92 α₂-peptide, the electrostatic interactions of PhT95 in DMPC do not involve K88. Rather, the phosphate group in PhT95 is stabilized mainly by interaction with the guanidinium group of the neighboring R94 residue, and sometimes simultaneously with the R104 residue. The absence of an electrostatic interaction involving K88 in the PhT95-α₂-peptide may help to account for the increased α-helical stability in this peptide due to reduced strain (see above). In the simulation experiments of PhT92–PhT95 in DMPC, distinctly different patterns of electrostatic interactions were observed. One of the observed patterns involve the phosphate groups at T92 and T95 being stabilized by electrostatic interactions with K88 and R94, respectively. A second pattern involves phosphorylated T92 and T95 interacting with R94 and R104, respectively. These distinct patterns of electrostatic interactions may represent separate ensemble structures stabilized by double-phosphorylation in the molecular switch region of MBP. Overall, the differences that we see in electrostatic interactions as a result of differing phosphorylation status may be a mechanism for controlling the conformation of the disordered regions in the molecular switch of MBP. This may be important in properly orientating these disordered regions in the cytoplasm for more specific protein-protein interactions. # Conclusions In this study, we have investigated the structure, stability, and dynamics of a key, highly-conserved central segment within the 18.5-kDa isoform of MBP, containing a proline-rich region with two MAP-kinase phosphorylation sites adjacent to a region that can form an amphipathic, membrane-anchoring α-helix. The finding that the proline-rich region adopts PPII structure in the presence of DPC lysophospholipids when the adjacent α-helix is formed, but not under aqueous conditions when the helical region is disordered, suggests that the α-helical and PPII regions may be cooperatively linked. This phenomenon could be a mechanism for ensuring that SH3-domains, which typically recognize PPII structure, bind to MBP only when the protein is anchored to the membrane via its central α-helix, and not free in the cytoplasm (see also). Phosphorylation at the MAP-kinase sites has marked effects on the stability and conformation of the conserved central segment of MBP. Phosphorylation alters the energy landscape of the central segment by disfavoring α-helical formation, which could inhibit MBP-membrane association. Phosphorylation also appears to play a direct role in MBP-membrane interaction by altering both the tilt angle and the penetration depth of the helix in the membrane. Phosphorylation can also restrain the highly-dynamic regions that flank the α-helix through formation of local electrostatic interactions. This post-translational modification may be important in orientating these highly-dynamic regions for binding reactions. Additionally, the restriction in conformational freedom resulting from the formation of these electrostatic interactions may play a role in facilitating binding-induced folding mechanisms that are often critical in IDPs, by reducing the conformational entropy of the unbound state. Overall, the results presented here support the hypothesis that the central conserved region in MBP represents an important molecular switch during myelinogenesis and turnover, and illustrates possible mechanisms of control through phosphorylation/dephosphorylation at the MAP-kinase sites in MBP. # Supporting Information The molecular dynamics investigations were made possible by the facilities of the Shared Hierarchical Academic Research Computing Network (SHARCNET: [www.sharcnet.ca](http://www.sharcnet.ca)) and Compute/Calcul Canada. The authors are grateful for laboratory technical support to Mrs. Janine Voyer- Grant, and for software and hardware support to Ms. Valerie Robertson and Mr. Peter Scheffer (University of Guelph NMR Centre), and to Mr. Bill Teesdale (Physics, Guelph). We thank Dr. Shenlin Wang and Dr. Vladimir Bamm (Guelph) for many helpful discussions and comments on the manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: KAV KB MDA EP GH. Performed the experiments: KAV KB MDA. Analyzed the data: KAV KB MDA EP GH. Contributed reagents/materials/analysis tools: EP. Wrote the paper: KAV KB MDA EP GH. [^3]: Current address: Department of Electrical Engineering and Computer Science (Institut Montefiore), Université de Liège, Liège, Belgium

# Introduction Breast cancer (BC) represents the most common malignancy in females, accounting for estimated 266120 new cases in the United States, in 2018. According to the latest statistical report, breast cancer still accounts for the second most common cause of death in females (14% of all cancer deaths) after lung cancer. It was reported that approximately 6% of breast cancer patients have been diagnosed with synchronous distant metastasis, such as bone, liver, lung and brain. Although great advance has been made in the past several years, huge challenge still exists in patients with metastatic breast cancer (MBC), which is still associated with substantial morbidity and mortality. The median survival time of MBC patients is about 18–24 months, and the 5- and 10-year survival rates is as low as 27% and 13%, respectively. It was reported that about 30–50% of breast cancer patients are aged over 65 years, among which patients over the age of 80 constitute a large proportion. Approximately 25% of breast cancer patients over age of 65 or 10.6% of all BC patients is reported to be 80 years-old or older. Moreover, study also showed that the incidence of breast cancer among the elderly group with 80–84 years-old (400 per 100,000 people) are much higher than that with 50–54 years-old (200 per 100,000 people). These older patients (≥80 years-old) often receive inferior screening for breast cancer, and present as an challenge in their treatment. Due to comorbidities and frailty, the elderly patients are inevitably underrepresented in clinical trials. A previous key study revealed that only 9% of participants in clinical trial were aged over 75 years in comparison with 31% in the overall patients population. Although oncologic indication exists, the decline of physiologic function and poor survival expectation make both elderly patients and doctors reluctant to pursue standard treatment recommended by clinical guideline, especially for those with distant metastasis. Therefore, compared to the younger group, the disease may have distinctive biological characteristics in patients aged over 80 years old. So far, however, few previous reports have concentrated on analyzing the heterogeneity of metastatic pattern and prognosis among MBC patients older than 80 years. In current study, the distant metastasis pattern and prognosis of elderly patients (≥80 years-old) were analyzed and compared to other age groups in a large cohort of metastatic breast cancer patients by using the SEER database. # Methods ## Database and case selection We performed a retrospective cohort research by utilizing the custom Surveillance, Epidemiology and End Results (SEER) database \[Incidence- SEER 18 Regs Custom Data (with additional treatment fields), Nov 2018 Sub (1975–2016 varying)\]. The SEER program, a database established by the National Cancer Institute of the U.S., collected data of cancer patients that accounts for about approximately 28% of the U.S. population. The SEER\*Stat software (version 8.3.8, National Cancer Institute, Washington, USA) was utilized to access the data from SEER database. Patients with de novo metastatic (M1) breast cancer (Site recode International Classification of Diseases for Oncology-3 (ICD-O-3)/WHO 2008:Breast) diagnosed in 2010 through 2015 were identified from this database. Only patients diagnosed with invasive ductal carcinoma (IDC) or invasive lobular carcinoma (ICC) as their first malignancy were eligible. In addition, patients with unknown follow-up, unknown molecular subtype, or unknown metastatic involvement, including bone, liver, lung, brain, and distant lymph node (DL) were excluded. Finally, a total of 10479 eligible patients were included in this study. It is well known that the cancer biology may quite different between young breast cancer patients and older patients. Recently, there is no international agreement on the definition of young breast cancer. Most of the literatures defined breast cancer patients younger than 35 or 40 years old as young breast cancer patients. Hence, based on the diagnosed age, patients in our study were subsequently divided into four groups: \< 35 years old, 35–49 years old, 50–79 years old, and \> 79 years old. The flowchart for the patient’s selection was shown in. ## Covariates Multiple variables were included in this study, including demographic characteristics (race and marital status), disease characteristics (AJCC T and N stage, histologic grade and molecular subtype), and treatment modalities (surgery, chemotherapy, and radiotherapy). Specially, races, including American Indians, AK Natives, Asians and Pacific Islanders, were classified into other races. Marital status, including divorced, separated, widowed, and domestic partner were categorized into other status. The main outcome were overall survival (OS) and cancer specific survival (CSS). ## Statistical analysis Demographic and clinical characteristics between different cohorts were summarized by descriptive statistics and compared by using the Pearson’s Chi- square test or Fisher’s exact test. The Kaplan-Meier curves were plotted for the OS and CSS between different cohorts, and the Log Rank test for the comparison of difference among the curves. The multivariate logistic regression models were constructed to explore the association between the age and the sites of distant metastases. Subsequently, the univariate and multivariate Cox regression analyses were also conducted to identify the independent prognostic factors for the elder patients (\> 79 years) with metastatic breast cancer. Descriptive statistic, logistic regression analysis, and Cox proportional hazards analysis were performed by using the SPSS 24.0 (IBM Corp). The Kaplan- Meier curves and the forest plots were created by using the R software version 3.6.0. We defined a 2-sided P value of \<0.05 as statistically significant unless otherwise stated. ## Ethics statement The SEER database is an open database. Data released from the SEER database do not require informed patient consent, because cancer is a reportable disease in every state of the United States. The present study complied with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. # Results ## Population characteristics A total of 10479 metastatic breast cancer (MBC) patients diagnosed from 2010 to 2015 were finally extracted from the SEER database, among which 1036 (9.9%) patients were at the age of ≥ 80 years old, 7045 (67.2%) patients within 50–79 years old, 1993 (19.0%) patients within 35–49 years old, and 405 (3.9%) patients under 35 years old. In comparison with younger (\< 35 yrs) and middle-aged (35–49 or 50–79 yrs) patients, the older patients (≥ 80 yrs) had a higher rate of white race (84.6% vs. 62.5% vs. 68.8% vs. 74.2%, respectively, P \<0.001), and higher rate of histologic grade 1–2 (48.4% vs. 33.1% vs. 39.0% vs. 44.1%, respectively, P \<0.001). On the contrary, compared to younger and middle-aged patients, the older patients exhibited a lower rate of lymph node metastasis (28.1% vs. 14.6% vs. 16.8% vs. 20.8% for N0, respectively, P \<0.001), and received less treatment, such as surgery (77.5% vs. 55.1% vs. 58.8% vs. 66.8%, respectively, P \<0.001), radiotherapy (77.6% vs. 56.8% vs. 57.3% vs. 64.6%, respectively, P \<0.001), and chemotherapy (81.7% vs. 13.6% vs. 23.8% vs. 42.4%, respectively, P \<0.001). Regarding to the molecular subtype, specifically, the older MBC patients had a higher rate of type of HR+/Her2- (67.7% vs. 40.7% vs. 53.6% vs. 61.3%, respectively, P \<0.001), but lower rate of HR+/Her2+ (13.0% vs. 27.9% vs. 20.7% vs. 16.5%, respectively, P \<0.001) and HR-/Her2+ (5.9% vs. 17.8% vs. 10.7% vs. 9.0%, respectively, P \<0.001), when compared to other aged groups. The detailed information for clinicopathological features among different aged groups were showed in. ## Metastasis pattern In the whole cohort of MBC patients, the most common single metastases site was bone (34.9%), followed by lung only (7.6%), DL only (7.3%), liver only (6.0%), and brain only (0.7%), respectively. 24.8% of patients were diagnosed with two distant metastasis sites, while 14.0% of patients had three or more metastases sites. In addition, the metastatic patterns between different aged groups were also compared. As shown in and, the older group had the highest incidence of lung (13.5% vs. 0.2% vs. 10.5% vs. 4.9%, respectively, P \<0.001) metastasis only and the lowest incidence of liver (0.4% vs. 0.5% vs. 1.6% vs. 3.6%, respectively, P \<0.001) compared with the younger and middle-aged groups. Subsequently, we further performed a multivariate logistic analysis that adjusted for various confounding variables, including race, marital status, histologic type, histologic grade, AJCC T and N stage, molecular subtype, and therapies. The results indicated that the older group tended to have less bone metastasis (OR = 0.59; 95% CI \[0.50–0.70\]; P \<0.001), brain metastasis (OR = 0.56; 95% CI \[0.33–0.97\]; P = 0.014), liver metastasis (OR = 0.47; 95% CI \[0.36–0.62\]; P \<0.001), and multiple sites metastasis (OR = 0.71; 95% CI \[0.49–0.99\]; P = 0.041) than the younger group. On the contrary, relative to the younger group, the older group was more likely to occur lung metastasis (OR = 2.04; 95% CI \[1.54–2.71\]; P \<0.001) than the younger group. The distribution of molecular subtypes in patients with specific metastatic site was further investigated. The older patients with bone metastasis had the highest proportion (75.9%) of HR+/HER2- subtype compared with those who had other metastatic sites. The similar trends were also found both in the younger (48.7%) and the middle-aged (62.1% or 68.4%) group. Our result also indicated that the proportion of TNBC subtype increased in patients with visceral metastases, especially for brain metastasis. Additionally, it was noted that the proportion of TNBC subtype increased most in the older patients (32.4%) with brain metastasis, compared to the younger (17.4%) and the middle-aged (25.2% or 19.2%) patients. ## Survival Both of the OS and CSS between the four age groups were compared in this study. The Kaplan-Meier curves showed that the older group had the shortest OS and BCSS (P \<0.001), with the median survival time being 13 (mean, 17.5 months), 26 (mean, 29.9 months), 27 (mean, 30.3 months) and 22 (mean, 25.8 months) months in the older, younger and middle-aged group, respectively. After adjusted for various confounding variables, the multivariate Cox analysis also confirmed that age was an independent prognostic factor. Compared to the younger group, the older age was significantly associated with worse OS (HR = 2.11; 95% CI \[1.78–2.50\]; P \<0.001) and CSS (HR = 1.89; 95% CI \[1.58–2.26\]; P \<0.001). Moreover, expect for MBC patients with brain metastases, the older group showed poorer OS and CSS than the younger group in all subgroup stratified by different metastasis sites, including bone, lung, liver, DL, and multiple metastasis sites. We then performed univariate and multivariate Cox analysis to explore prognostic factors that associated with OS and CSS for older patients with metastasis breast cancer. The multivariate analysis, as shown in, showed that patients with white race (vs black; HR = 0.87; 95% CI \[0.82–0.94\]; P \<0.001), married (vs single; HR = 0.82; 95% CI \[0.76–0.88\]; P \<0.001), N1 (vs N0; HR = 0.81; 95% CI \[0.85–0.98\]; P = 0.010), molecular subtype HR+/Her2+ (vs HR+/Her2-; HR = 0.81; 95% CI \[0.75–0.88\]; P \<0.001) were significantly associated with increased CSS of older MBC patients. On the contrary, patients who had clinicopathological factors like histologic grade 3–4 (vs grade 1–2; HR = 1.39; 95% CI \[1.30–1.48\]; P \<0.001), T2 (vs T1; HR = 1.12; 95% CI \[1.01–1.24\]; P = 0.030), T3 (vs T1; HR = 1.22; 95% CI \[1.09–1.36\]; P = 0.001), T4 (vs T1; HR = 1.40; 95% CI \[1.27–1.56\]; P \<0.001), TNBC subtype (vs HR+/Her2-; HR = 2.91; 95% CI \[2.68–3.16\]; P \<0.001), bone metastasis (vs DL metastasis; HR = 1.15; 95% CI \[1.02–1.31\], liver metastasis (vs DL metastasis; HR = 1.52; 95% CI \[1.29–1.79\]; P \<0.001), lung metastasis (vs DL metastasis; HR = 1.27; 95% CI \[1.09–1.48\]; P = 0.002), brain metastasis (vs DL metastasis; HR = 2.46; 95% CI \[1.83–3.30\]; P \<0.001), multiple sites metastasis (vs DL metastasis; HR = 1.98; 95% CI \[1.76–2.24\]; P \<0.001) were dramatically associated with decreased CSS. Regarding to the treatment modalities, older MBC patients who received surgery (HR = 0.57; 95% CI \[0.53–0.61\]; P \<0.001) or chemotherapy (HR = 0.62; 95% CI \[0.58–0.66\]; P \<0.001) had dramatically favorable CSS in comparison with those received no treatment. A similar survival trend was also observed for OS. # Discussion In current study, we systematically analyzed the distant metastasis patterns of MBC patients aged over 80 years and compared it to younger or middle-aged counterparts. Our results indicated that the elderly patients had distinctive clinicopathological characteristics and metastasis patterns. Our data also demonstrated that age at diagnosis was an independent risk factor for distant organ metastasis. Additionally, various independent prognostic factors were also identified for MBC patients aged over 80 years. The proportion of elderly MBC patients (≥ 80 yrs) was 9.9% in current study, which echoed previously published studies implying that octogenarians represent approximately one in ten BC patients. Our data found that the elderly MBC patients had a higher proportion of white race than younger patients, which might be due to genetic heterogeneity or social and economic factors. We also found that the older patients had a higher proportion of HR+/Her2- subtype but a lower proportion of other molecular subtypes, including HR+/Her2+ and HR-/Her2+. These data were inconsistent with the results of previous study that showed a similar molecular subtype between octogenarian patients and patients with 60–70 years old, which partly due to its small number of cases. Recently, accumulating evidence have demonstrated that the expression level of HER-2 is positively related to the lymph node involvement and malignant behavior of breast cancer. These findings might partly explain the current data indicating that the older MBC patients had lower histological grade and less lymph node involvement than younger patients. Regarding the metastasis pattern in this study, the most and least frequent metastasis lesion were bone and brain, respectively. Our result was consistent with previous studies. Purushotham, et al. suggested a striking relationship between increasing age at diagnosis and decreased risk of distant metastasis to both bone and viscera in breast cancer patients. Consistent to this study, our data further demonstrated that the elderly patients were less likely to have distant bone, brain, liver, lymph node, and multiple sites metastasis than the younger counterparts. Aging-related deterioration of the immune system, also called ‘immuno-senescence’, might partially account for the phenomenon. Recent researches have demonstrated that co-opting the immune system is an important part of the metastatic process. Hence, we hypothesized that immuno-senescence accompanied by advancing age may act as protective factor for distant metastasis through depriving the key immune-cellular components associated with metastatic process. However, as an exception, we also found that the older group had a higher risk of lung metastasis than the younger group. One possible factor contributing to this phenomenon would be aging-related low-grade inflammation, which was reported to facilitate tumor progression. Moreover, Wculek, et al. also reported that an altered presence of neutrophils contribute to lung colonization of metastasis-initiating breast cancer cells. Additionally, the relationship between molecular subtype and metastasis were further studied in our study. Our data indicated a distinctive proportion of molecular subtype in different aged patients with different metastasis lesion. The proportion of HER2+ and TNBC subtype increased in patients with visceral metastases, which echoes previous study suggesting that ER-/PR- and/or HER2+ were more likely to metastasize to viscera compared to the reference category. Additionally, consistent to previous researches, our study also indicated that patients with brain metastases were more likely to be TNBC subtype, especially in the older group. This result implied that the genetic alteration in different aged group as well as different molecular subtype may associated with increased risk of metastasis to specific organs. Moreover, the ‘seed and soil’ hypothesis might also a possible contributor. The underlying molecular mechanisms still need to be further investigated. In our survival analysis, our Kaplan-Meier curve as well as the multivariable Cox model demonstrated that older age dramatically contributed to worse OS and CSS. Moreover, in almost all subgroup with different metastasis site, the old aged patients had the worst survival in comparison with the younger group. This phenomenon might be explained by multiple factors, such as decreased physiologic function, increased risk of non-cancer death, and reduced treatment willingness. Supporting this viewpoint are previous reports showing that elderly patients are often not only offered less surgical treatment but also received less postoperative adjuvant chemotherapy and radiotherapy. Consistently, our data further validated that elderly group were less likely to receive any treatment than younger group, although patients can still get survival benefit from surgery and adjuvant therapy even with distant metastases. This was probably due to concerns about increased comorbidities that induce decreased treatment tolerability. In addition, our study also identified various independent prognostic factors for MBC patients aged over 80 years. For example, we demonstrated that patients with white race or married status were significantly related with increased CSS, which could partly due to socioeconomic factors. We also found that patients with HR+/Her2+ subtype had a better prognosis than those with HR+/Her2-. We suspect that the recent great advance in anti-Her2 targeted therapy may account for it. Inevitably, several potential limitations may exist in this study. Firstly, this analysis is limited by selection bias due to intrinsic weaknesses of retrospective study. Secondly, metastatic data in SEER database is restricted to five organs (distant lymph node, bone, liver, lung and brain) and other distant metastasis sites are unclear. Thirdly, detailed information such as co- morbidities, endocrine therapy, and anti-Her2 targeted therapy are limited in the SEER database, which are thought to be closely related to patients’ survival. Hence, our results need to be further validated in future studies. # Conclusion In summary, our study summarized the metastasis pattern and survival outcome of MBC patients aged over 80 years from a large sample of the population. We found that MBC patients with different age presented with distinctive metastatic patterns, clinical characteristics, and prognostic outcomes. Our findings could provide a better understanding of clinical features in elderly MBC patients and may assist clinicians to make appropriate treatment. # Supporting information The authors acknowledged the efforts of the Surveillance, Epidemiology, and End Results (SEER) Program tumor registries in the creation of the SEER database. [^1]: The authors have declared that no competing interests exist.

# Introduction Despite the universal recognition of efficacious interventions and unprecedented resources to reduce MTCT (mother to child transmission), in 2009, only 26% of pregnant women living in low and middle income countries had been tested for HIV and among identified HIV-positive women, only an estimated 53% received an antiretroviral (ARV) regimen to prevent mother-to-child HIV transmission (PMTCT). As a result of the failure of health systems to effectively deliver PMTCT regimens, approximately 1,000 children under the age of 15 continue to be infected with HIV every day, the vast majority around the time of birth and through breastfeeding. While multiple studies have measured PMTCT efficacy, studies that describe determinants of PMTCT program effectiveness are rare. The few published reports that have assessed PMTCT program effectiveness have not addressed reasons for successes and failures. A multi-site PMTCT effectiveness study in four African countries called PMTCT Effectiveness in Africa: Research and Linkages to Care (the PEARL Study) was conducted between 2007–2009, and found that coverage of nevirapine (NVP) among HIV-infected women delivering in health facilities with PMTCT services varied dramatically and was only 55% overall. A better understanding of barriers to high PMTCT coverage is needed with the new 2010 World Health Organization (WHO) guidelines promoting more efficacious ARV combination regimens for women and extended infant prophylaxis during breastfeeding. Implementation of the new guidelines will require higher quality antenatal - and PMTCT care, more rigorous infant follow up, and ongoing maternal care after delivery. More complex PMTCT regimens will also require a greater understanding of what occurs at the health facility level and where investments need to be made. Using data from the PEARL Study, we examine how facility and service characteristics predict maternal-infant coverage with NVP at the time of delivery. # Methods The PEARL study was a multi-country evaluation of PMTCT effectiveness at the patient, facility, and community levels. We previously reported PMTCT coverage in a survey among women delivering in 43 health facilities in Cameroon, Cote d'Ivoire, South Africa, and Zambia between 2007 and 2009. PMTCT coverage was defined as the proportion of HIV-positive mother-HIV-exposed baby pairs in which both received single-dose NVP. Maternal dosing was confirmed by biochemical measurement in the cord blood and newborn dosing was confirmed by chart review. As part of this study, we completed facility surveys at delivery centers which also provided antenatal care. ## Data collection We used a modified version of “A Rapid Health Facility Assessment Tool: to Enhance Quality and Access at the Primary Health Care Level.” This tool was developed in 2006 by ICF Macro in collaboration with MEASURE Evaluation and a panel of experts from the United States Agency for International Development (USAID) and other cooperating agencies and modified to include detailed PMTCT information in parallel with other antenatal care information. The original tool is available online <http://www.mchipngo.net/controllers/link.cfc?method=tools_rhfa>. The questionnaire included general questions such as type of facility, estimated size of the catchment area, and location as well as four discrete modules on clinic operations. Each module contained 32–120 questions. Modules included direct observation of clinician patient encounters, exit interviews of patients, and questions of patients and providers. - Module 1 was a quality-of-care checklist completed by direct observation of an antenatal consultation. Up to six consecutive antenatal visits in each facility were observed. Only first ANC visit observations were used. - Module 2 was an exit interview with up to six consecutive pregnant women after the consultation to measure patient perceptions about care. Key measurements were satisfaction and understanding of each medicine or prescription given during the consultation. - Module 3 was a checklist of available infrastructure, equipment and supplies (including drugs). - Module 4 was an interview with the health center manager regarding services provided, numbers of personnel, and other staffing variables. The survey instrument was adapted and translated into French in Côte d'Ivoire and was pretested in the four countries prior to data collection. All questionnaires were entered into a Microsoft Access database and sent to PEARL's central data management and analysis unit. We examined each variable (n = 377) in the questionnaire individually to determine which ones were associated with PMTCT coverage. In addition, we created seven composite scores that summarized features of the clinic in several domains in a systematic manner. These scores were developed *a priori* by the study co-investigators based on logical groupings of characteristics, and included composite scores for antenatal care, PMTCT, supplies, staffing level, patient satisfaction, general infrastructure, and patient understanding of medications. Scores were adjusted by country to account for different standards of care (e.g. items relating to malaria were not considered for South Africa since South Africa does not include malaria prophylaxis in routine antenatal care). summarize the variables used and how the scores were constructed. In addition, time-motion variables were constructed from the average of up to six patient observations at each facility. Time was recorded at a) registration b) start of exam and c) visit finish (including receipt of medication at pharmacy), allowing the calculation for the median total time of the visit as well as the time spent post test counseling. Antenatal and PMTCT domains were assigned a score of (1) if appropriate care or treatment was given; (0.5) if appropriate care or treatment was given but not recorded in the chart; and (0) if appropriate care was not given. This was done because failure to record information, such as an HIV test, CD4 count, or blood pressure reading will lead to failure to act upon this critical information throughout the pregnancy. Infrastructure, supplies, and staff domains were assigned a score of (1) when available and (0) when not available. Scores were totaled for each domain, and divided by the number of items included. Non- applicable information and missing items were not considered in the scoring. Final domain scores were thus a proportion between 0 and 1, with a higher score reflecting more appropriate care. Patient satisfaction was scored on a scale of 1 to 5, with 5 being extremely satisfied and 1 being extremely unsatisfied. For patient comprehension, (1) point was given for each correct response for each criteria (dose, frequency, duration, and purpose), and (0) points were given for incorrect responses. ( summarizes the construction of health facility quality scores. ## Statistical analysis All PEARL study facilities with cord blood and facility survey data were included. Site characteristics were summarized using descriptive statistics. We computed means and standard deviations or medians and inter-quartile ranges for continuous variables and percentages for categorical variables. Using all facilities from all 4 countries and PMTCT coverage as a continuous outcome measure, separate regression models were fit for health facility characteristics of interest and the quality scores. Generalized estimating equations (GEE) with an exchangeable correlation structure were used to account for country program -related correlation between facilities in the same country. Statistical analyses were performed using SAS® version 9.1.3 (SAS Institute, Cary, NC, USA) and Stata® version 10.0 (StataCorp. 2007. Stata Statistical Software: Release 10. College Station, TX). ## Ethical review and role of the funding source Approval was provided by institutional or national review boards at the U.S. Centers for Disease Control and Prevention (CDC), the University of Alabama at Birmingham, and each of the participating countries a) Comite d'éthique des sciences et de la vie, Ministère de la Santé, Côte d'Ivoire, b) University of Zambia Research Ethics Committee, Lusaka Zambia, c) Cameroon Baptist Convention Health Board, Bamenda, Northwest Province, Cameroon, d) South African Medical Research Council Ethics Committee, Cape Town, South Africa. Verbal consent was obtained among the health care workers and the selected pregnant women who participated in the direct interview during this survey. The corresponding author had access to all data in the study and final responsibility for the decision to submit this manuscript. # Results ## Description of the health facilities Cord blood and facility survey data were collected from 32 facilities (8 facilities in Cameroon, 9 facilities in Cote d'Ivoire, 6 in South Africa, and 9 in Zambia). Most were health centers managed by their respective governments. General characteristics of facilities and their services are described in, An opt-out approach for HIV testing was used in 100% of facilities in Zambia, 63% in Cameroon, and none in Côte d'Ivoire and South Africa Antenatal care service provision among observed patients by country is described in. ## Relationship of individual facility and service characteristics to NVP coverage shows the relationships between facility characteristics, observed service provision, and NVP coverage. Higher coverage was associated with HIV test kits being found in ANC (as opposed to in the lab), HIV testing available in labor ward, availability of CD4 testing at the clinic, infant testing with PCR available most days of the month, and the presence of an antenatal register with PMTCT information. PMTCT coverage was not associated with type of facility, location of HIV testing during antenatal care, reported opt-in or opt-out testing strategy, same day HIV results, partner HIV testing, time of dispensation of NVP (at first antenatal visit, later in pregnancy or upon arrival at the facility in labor), or the presence of lay counselors. ## Relationship of a priori composite quality scores to NVP coverage shows the seven composite quality scores and two time variables stratified by country, and their relationship with NVP coverage. In univariate analysis using Pearson correlation (not adjusted for the clustering of facilities within countries), PMTCT coverage was correlated with higher PMTCT quality score (p = 0.003); infrastructure quality score (p = 0.017); patient satisfaction quality score (p = 0.031); and patient understanding of medications score (p = 0.006), and was marginally correlated with the antenatal quality score (p = 0.091). There was no correlation between PMTCT coverage and staff quality score (p = 0.168) or supply quality score (p = 0.812). PMTCT coverage was associated with longer total time at the clinic (p = 0.013), but not longer time spent for post-test counseling (p = 0.938). When we accounted for clustering using GEE, only antenatal quality score (p\<0.001) remained significantly positively correlated with PMTCT coverage, but PMTCT quality score and patient satisfaction score remained marginally significant. # Discussion This facility survey was conducted in sites at which we already knew the PMTCT coverage level from the previously-published cord blood study. The exercise thus presented a unique opportunity to both describe facility factors associated with PMTCT coverage and to begin to validate the survey tool as a predictor of program performance. To our knowledge, this is the first study to examine in a systematic way the relationship between antenatal clinic and service characteristics and an objective outcome such as PMTCT coverage. Many variables were associated with higher coverage, both individually and in two of the aggregate scores. Some, including availability of CD4 testing and infant PCR testing were intuitively satisfying and agree with our preconceived notions about how programs evolve to be more sophisticated and more successful. We did learn a few new things, however, including that variables pertaining to general antenatal care were even more predictive than PMTCT variables, and that the single factor distinguishing all of the worst-performing sites from the others was the lack of registers with PMTCT information. The predictive value of the non-PMTCT variables, and the lack of predictiveness of several PMTCT variables, reminds us that complex PMTCT services cannot be expected to function well where overall services are poor, even if all the elements of PMTCT are provided. Overall, we concluded that this tool has limited but not zero utility as a method of evaluating PMTCT sites. Certainly programs do need to know if the essential elements of PMTCT and antenatal care are in place, and common sense dictates that PMTCT requires the presence of drugs, trained staff, and properly stored test kits. However, presence of these elements does not necessarily ensure successful implementation, and answers obtained in this simple checklist- based survey and patient exit interview do not necessarily reflect the experiences of patients visiting the facilities. The overall low coverage of NVP in our study sites (55%, range 33% to 68%) indicates that something is amiss in these PMTCT sites. The aggregate PMTCT score did have some relationship with coverage overall, but not in all countries, and the PMTCT scores were relatively homogeneous compared to the very large observed differences in coverage. Provision of longitudinal medical care and the population's receptiveness to this medical care is a complex process dependent upon attitudes, beliefs, trust, relationships, and complex elements of clinic flow and individual interactions. Most of this cannot be adequately described using simple inventories such as this one, even with direct observation of some elements of care. Several variables that we expected to be associated with higher PMTCT coverage were not, and on close analysis of the questions and data we appreciated weaknesses in the survey tool that are important to help guide future program evaluation efforts. For example, we asked clinic managers if their site conducted “opt-out” HIV testing. Though several studies have shown that opt-out testing increases uptake. There was no association between opt-out testing and higher PMTCT coverage in our study. Asking clinic managers was probably not the right approach. In an operational sense, opt-out testing is defined not by the policy in place, but by the details of how testing is explained to clients and how they, their laboratory forms, and their blood move through the clinic during their visit. A checklist survey approach does not lend itself to an adequate description of how this occurs, and it is doubtful that we truly distinguished between opt-out and opt-in approaches. There are other examples as well. We found that availability of antiretroviral therapy (ART) at the clinic did not predict PMTCT coverage. Simply having ART at the same site does not tell us how well the ART and antenatal clinics work together to ensure that all eligible women receive ART during pregnancy. The details of *how* the services are integrated will determine successful provision of ART in pregnancy. Meaningful exploration of these issues requires a different type of evaluation. Our study has several recognized limitations. Since we decided to include only sites that provided both antenatal and delivery services, we included a relatively small number of facilities across four countries. The small sample size may have impacted the statistical power to detect associations between our composite scores as well as individual variables and PMTCT coverage. Second, with only six consecutive observations per facility selected, it is possible that the result for any one site might not be representative of that site's performance (sampling error). This could lead to exposure misclassification and undermine the ability to observe a true association. Our results broadly support the general principle of strengthening the health care system as an important strategy for improving PMTCT coverage. This analysis exercise provides an important reminder for programs that service provision is complex, and that provision of drugs, test kits, policies, and training does not ensure program success. It is also an important caution to program evaluators that even this carefully conceived and detailed site survey could not predict which sites did well in a very useful way. Once basic elements of PMTCT are in place, detailed clinic-level quality improvement and problem solving initiatives that focus on operational factors are probably just as essential. Further study surrounding quality improvement methods and their impact on coverage and outcomes is warranted. ## Members of the Pearl Study Team Cameroon: Pius Tih (Cameroon Baptist Health Convention Health Board \[CBCHB\]); Thomas Welty (CBHC); Allison Spensley (Elizabeth Glaser Pediatric AIDS Foundation \[EGPAF\]), Christophe Grundman (EGPAF); Catherine Wilfert (EGPAF and Duke University). Cote d'Ivoire: Didier Koumavi Ekouevi (Programme PAC-CI, Abidjan, Cote d'Ivoire and Université Victor Segalen, Bordeaux, France); Francois Dabis (Université Victor Segalen, Bordeaux, France); Serge Kanhon (Ministry of Health, Abidjan, Cote d'Ivoire). South Africa: David Coetzee (University of Cape Town, Cape Town, South Africa \[UCT\]); Kathryn Stinson (UCT); Peter Smith (UCT); Andrew Boulle (UCT); Felicity Gopolang (UCT). Zambia: Elizabeth M. Stringer; Jeffrey S A Stringer (Centre for Infectious Disease Research in Zambia \[CIDRZ\]); Benjamin H Chi (CIDRZ), Namwinga Chintu (CIDRZ); Mark Giganti (CIDRZ); Jessica Joseph (CIDRZ), Maximilian Bweupe (Zambian Ministry of Health); Nande Putta (CIDRZ); Alain DeGroot (CIDRZ), Humphrey Mulenga (CIDRZ); Wendy Z Mazimba (CIDRZ); Andrew O. Westfall (CIDRZ), Marc Bulterys (CDC-Zambia); Lawrence H Marum (CDC-Zambia); and Charles Cowan (Analytic Focus, LLC). US Centers for Disease Control and Prevention: Tracy Creek (National Center for HIV, STD, and TB Prevention, Global AIDS Program, Atlanta Georgia). World Health Organization: Nathan Shaffer (HIV Department, Geneva; formerly with CDC). # Supporting Information We would like to thank all the patients, doctors, nurses, community workers, and support staff who contributed to the completion of this study. In addition to the study team, we would like to acknowledge Jessica Joseph for her contribution to the analysis of this study, Mary Louise Newell and Latasha Treger for their contributions to this study. The findings and conclusions of this manuscript are solely the responsibility of the authors and do not necessarily represent the official views of the US Centers for Disease Control and Prevention. This study was presented in part at the 6^th IAS Conference on HIV pathogenesis, treatment and prevention, Rome, Italy, 17–20 July 2011 (Abstract \# MOPE493). [^1]: Analyzed the data: AOW. Wrote the paper: DKE ES TC AOW. Designed the study: DKE ES NC DC PT TC TW BHC CW NS JS FD. Collected the data: DKE PT KS NC. Interpreted the data: DKE ES NC BHC TC JS FD. Contributed to the writing of the manuscript: DC PT KS TW NC BHC CW NS JS FD. Agreed with the manuscript's results and conclusions: DKE ES DC PT KS TC AOW TW NC BHC CW NS JS FD. [^2]: The authors have declared that no competing interests exist.

# Introduction Dengue is by far the most devastating of all mosquito borne viral diseases, caused by dengue virus (DENV), a member of the family *Flaviviridae*. More than 3 billion humans live in dengue endemic regions of the world and currently more than 50 million infections occur annually with at least 500,000 individuals requiring hospitalization. Development of antiviral agents has so far focused on inhibiting viral enzymes, i.e. nonstructural protein 3 (NS3). An understanding of virus morphogenesis and protein-protein interactions could provide new targets for intervention. DENV is an enveloped virus with a positive sense single-stranded ∼11 kb RNA genome. Infection begins with attachment of virus particles to host receptors, several of which have been identified as putative receptors i.e. heparan sulphate, heat shock proteins, Hsp70 and Hsp 90, Glucose- regulated protein (GRP78/Bip), a 37-kDa/67-kDa high affinity laminin receptor, and Dendritic cell-specific intercellular adhesion molecule-3-Grabbing non- integrin (DC-SIGN). After binding to the receptor DENV undergoes either clathrin mediated endocytosis, or directly enters into the cytoplasm of cells by fusion at the plasma membrane (PM). Following internalization, the virus envelope undergoes fusion with the endosomal membrane due to low pH, and the viral nucleocapsid is released into the host cytoplasm. The viral RNA is then translated to yield structural and nonstructural viral proteins followed by transcription and replication of RNA, packaging and maturation in the perinuclear (PN) region and egress by exocytosis or budding at the PM. The entire process of morphogenesis is dependent on the movement of virus and viral components within the cell. The cytoskeleton is an integral component for intra- cellular trafficking, with all 3 components, the microfilaments, the microtubules (MT) and the intermediate filaments, involved. Dynein, is a microtubule associated motor protein and is responsible for the movement of cargo from cell periphery to cell centre, from the endoplasmic reticulum (ER) to the Golgi. A crucial role for dynein in the MT associated transport of viral proteins has been reported for hepatitis C, African swine fever, Hanta, polio and rabies viruses. Although the major events in morphogenesis are broadly understood for DENV, the detailed events underlying, entry and retrograde trafficking have not been entirely delineated. The present study uses confocal microscopy to visualize internalization, endocytosis and early trafficking of DENV-2 proteins and shows for the first time association of dynamin II with the envelope (E) protein during entry and relevance of dynein in the retrograde trafficking of envelope and core (C) proteins. # Materials and Methods ## Cells and Virus BHK-21 (Baby Hamster Kidney) (C-13, American Type Culture Collection) cells were used as the cell line was suitable for both infection as well as transfection studies. The cells were maintained in Minimal Essential Medium (MEM) with 10% fetal bovine serum (FBS) at 37°C and 5% CO₂, supplemented with 200 U/ml of penicillin/streptomycin and 1% glutamine. All cell culture reagents were procured from Gibco-BRL. DENV-2 (803347) was obtained from the Institute's Virus Repository and used in all the experiments. ## Antibodies and reagents For immunofluorescence assay primary antibodies included mouse monoclonal antibody (MAb) against DENV-2 E glycoprotein and C protein (a kind gift from Dr. Askov, Queensland University of Technology, Australia), goat antibody against dynamin II, mouse MAb against dynein intermediate chain (Santa Cruz Biotechnology Inc., CA, USA), mouse MAb against alpha tubulin (Sigma, USA). Secondary antibodies used were goat anti-mouse IgG conjugated to FITC or donkey anti-mouse IgG conjugated to Alexa 488, donkey anti-goat IgG conjugated to Alexa 594 (Molecular Probes, Eugene, OR, USA) and rabbit anti-mouse IgG conjugated to TRITC (Sigma). Lysotracker Red (Molecular Probes) was used to stain endosomes. ## Immunofluorescence Assay BHK-21 cells were seeded one day prior to infection at a minimum concentration of 1×10⁵ cells/ml per 55 mm petri dish (Tarsons) containing 9×22 mm cover slips. Cells were infected with DENV-2 at a multiplicity of infection (MOI) of 1 or 5. Adsorption was carried out for 1 h at 4°C to prevent virus entry and synchronize infection of cells. Cells were washed twice with cold medium to remove unbound virus and replenished with prewarmed medium. The cover slips were taken out of the culture at various time points post infection (p.i.) under sterile conditions, washed with 0.01 M phosphate buffer saline pH 7.2 (PBS) and fixed in ice-cold acetone for 10 min at −20°C for experiments with dynein. Cells labeled for dynamin II and endosomes were fixed with 3.7% paraformaldehyde for 30 min at room temperature (RT), followed by permeabilization with 0.1% Triton X-100 for 2 min and quenching with 0.01 M NH₄Cl. The fixed cells were washed with PBS and blocked with 1% bovine serum albumin (BSA) in PBS for 30 min. For dual staining, viral E or C proteins were labeled with specific mouse MAbs and the cellular components with the specific antibodies as mentioned in. The primary antibodies to viral protein and cellular components were added simultaneously to the cells and incubated for 1 h followed by washing with PBS. This was followed by incubation with appropriate secondary antibodies conjugated to fluorescent dyes, FITC, TRITC, Alexa 488 or Alexa 594 added simultaneously. DAPI (4′, 6′ diamino-2-phenylindole) was used to stain nucleus in all experiments. At the end of the staining process, cover slips were washed and mounted onto slides using mounting medium MOWIOL (Calbiochem, CA, USA). Control cells, which were not infected but submitted to the same procedures, were included in all experiments as mock infected cultures. ## Transfection with GFP-dynamitin plasmid The GFP-dynamitin expressing plasmid was a kind gift from Dr. Beate Sodeik, Hannover Medical School, Germany. BHK-21 cells, grown on coverslips in 6 well plates, were transfected with 2 ug GFP-dynamitin plasmid using Lipofectamine 2000 Plus (Invitrogen) according to manufacturers instructions. Briefly, 2 ug of plasmid was diluted in 250 ul of MEM mixed with 250 ul of MEM containing 10 ul of Lipofectamine and incubated for 30 min at RT. The DNA-liposome complex was then added to preformed monolayer of BHK-21 cells with 70% confluency in serum free medium and incubated at 37°C. Cells were then infected with DENV-2 (MOI of 1) at 6 h post transfection. The cells were fixed after 36 h p.i. and stained for DENV-2 E or C protein using mouse anti-E or anti-C MAb followed by anti- mouse IgG labeled with TRITC. For comparison mock-transfected cells were infected with DENV-2 and processed similarly. ## Image acquisition and image analysis The images were acquired using Laser Scanning Microscope 510 META (Carl Zeiss, Germany) with 40× and 63× oil objective lens corrected for oil immersion. For dual color analysis, green and red emissions were recorded sequentially through appropriate filters (505–530 band-pass filters for Alexa 488/FITC and 560 nm long-pass filters for Alexa 594/TRITC). Post acquisition image analysis was done using the 3D projection option of LSM software, surface rendering option of Huygens Essential (Scientific Volume Imaging, The Netherlands) and maximum intensity projection using Imaris (Bitplane Scientific software). Three- dimensional (3D) images, which cover the entire depth of the cell, were derived from deconvolved images. All voxels (volume pixels) in the image with a given colocalization level were joined, forming a 3D surface. In the surface rendering option of the software, the surface is split into different closed unconnected parts enabling independent analysis. ## Coimmunoprecipitation assay BHK-21 cells were infected with DENV-2 (MOI of 1). At 36 h p.i., the cells were washed with ice cold PBS (0.01 M), scraped and centrifuged at 2000 rpm. The cells were lysed in 5 pellet volumes of ice-cold lysis buffer (50 mM Tris- HCl,150 mM NaCl, 0.5% (v/v) NP40, 0.1 mM NaF, 10 mM DTT and protease inhibitor cocktail). The clarified cell lysate containing approximately 500 ug of protein was incubated with 10 ul of equilibrated Sepharose protein A beads for one hour at 4°C for preclearing and centrifuged to recover lysate. The precleared lysate was incubated overnight at 4°C with human anti-DENV IgG (purified from a serum sample that had plaque reduction neutralization titre of 12,500) for formation of immune complexes. The immune complexes were captured by Sepharose protein A beads by incubating for 3 h at 4°C, centrifuged and washed three times with lysis buffer. The beads were resuspended in 30 ul of sodium dodecyl sulfate (SDS) sample buffer (125 mM Tris, pH 6.8, 4% SDS, 20% glycerol), boiled for 5 min, centrifuged and the supernatant was loaded on 12.5% polyacrylamide gel under non reducing conditions. The proteins were transferred to nitrocellulose membrane for Western blot analysis. The blot was probed with anti-dynein antibody followed by goat anti-mouse IgG labeled with Horse radish peroxidase (HRP) (Sigma) and developed with diamino benzedene substrate. Mock infected BHK-21 cell lysate pulled down with Sepharose protein A beads served as negative control. Infected BHK-21 cells lysate was used as positive control. ## Real time PCR Viral RNA was quantitated in the infected cultures using a two step real time RT-PCR test reported previously. BHK-21 cells were infected at MOI of 1 and culture supernatant and cells were harvested at different time points starting from 0 h to 120 h p.i. # Results ## Kinetics of DENV-2 replication in BHK-21 cells Prior to undertaking experiments on entry and early intracellular trafficking, the kinetics of viral RNA production in BHK-21 cells was determined. BHK-21 cells were infected with DENV-2 at MOI of 1 (∼10⁹ RNA molecules) and the cultures were assessed for cell-associated and cell free viral RNA at different time points p.i. by real time RT-PCR. Despite washing off the unadsorbed virus, viral RNA (10⁵ molecules) was detected in cells as well as culture supernatant, from 0–6 h after infection. Similar levels of viral RNA at early time points after infection were shown in a previous report. It is possible that virus particles, which are not internalized, remain loosely bound, and are detectable for a long time in culture. The inefficiency of the process of infection for flaviviruses with a PFU: particle ratio of 1∶4000 has been reported before. The cell-associated viral RNA levels showed a rise from 12 h onwards till 72 h (10⁹ molecules) and then decreased to 10⁷ particles by 120 h. Viral RNA in the supernatant started increasing from 36 h and peaked at 84 h (10⁷ molecules) after which the level reached a plateau till 120 h. The cultures could not be tested after 120 h due to cell lysis. ## Association of dynamin II with DENV-2 E protein Dynamin II, a MT associated GTPase protein is known to regulate clathrin- mediated endocytosis. The association of dynamin II with DENV-2 E protein and the duration of the association were determined to understand the kinetics of internalization. BHK-21 cells were infected with DENV-2 at MOI of 5 to ensure infection of most cells simultaneously. Virus adsorption was carried out on ice to synchronize internalization. After adsorption virus was removed and the monolayer was washed with cold medium. Immediately after adding pre-warmed medium, cells were fixed and processed for staining after 0, 5, 10, 15, 30, 60 and 120 minutes. Cells were double stained for DENV-2 with mouse anti-E MAb and for dynamin II with goat anti-dynamin II antibody, followed by donkey anti-mouse IgG-Alexa 488 and donkey anti-goat IgG-Alexa 594. The E protein was presumed to represent the incoming virus, as it is the major surface protein on the virion. Colocalization of E protein and dynamin II was observed from 0 min to 30 min p.i. The colocalization foci were counted and analyzed in 20 fields at 0, 5, 15 and 30 min time points. The number of cells showing the presence of colocalization foci decreased from 50% at 0 min to 17% at 30 min. The size of the foci did not vary with time. The size of foci ranged from 0.75 um to 2 um, within which the spot of green fluorescence representing virus was 0.2–1 um (size of virus 50 nm) suggestive of each focus representing an aggregate of virus associating with dynamin II. In the area of colocalization the average fluorescence intensity for dyanmin II was 234 and did not change with time. The intensity of fluorescence for E protein decreased from 145 at 0 min to 84 at 15 min. The pattern of intensity reversed at points of no colocalization with average fluorescence intensity of 48 for dynamin II and 207 for E protein. The colocalization of E protein with dynamin II is depicted in the split image. The intensity of DENV E protein at a point of no colocalization is higher than that observed in the area of colocalization as shown by the intensity profiles. For a clearer depiction of the merging of virus and dynamin II, deconvolved images of the cell in were generated using Huygens surface rendering wherein free virus can be seen (green) around the focus of colocalization (yellow). To show that the complexes were present on the surface, z-stack images of the infected cells were acquired and analyzed by Maximum Intensity Projection (Imaris), which clearly shows the location of the aggregate on cell surface. Rotation of the image along its y-axis and 3D imaging using Zeiss software further confirms the surface location of the complex. The colocalization between E protein and dynamin II decreased by 30 min p.i.. At 60 min there was no association between E protein and dynamin II. shows the distribution of dynamin II in mock infected cells. The results illustrate the involvement of dynamin II in the process of virus internalization which peaked in the first 5 min of infection. ## Trafficking of virus within endosomes Lysotracker Red dye was used to follow the kinetics of DENV-2 association within endosomes. Lysotracker red is an acidophilic dye which selectively stains low pH containing compartments and would therefore detect DENV-2 in both early and late endosomes Cells were infected with DENV-2 at 5 MOI, similar to the experiment with dynamin II and stained for E protein and endosomes at 0 min, 10 min, 15 min, 30 min, 1 h, 2 h, 4 h, and 8 h post infection. Lysotracker red was added to cells 2 h prior to fixation and cells were labeled for E protein using mouse anti-E MAb followed by goat anti-mouse IgG FITC after fixing. Tracking the movement of endosomes showed that the endosomes positive for E protein, were visible close to cell periphery at 0 min p.i., were present throughout the cytosol within 15 min and collected in the PN region by 30 min. By 2 h only a few cells showed association and by 4 h there was no colocalization. The close association was further confirmed by the deconvolution analysis of the 30 min image. The number of E-positive endosomes observed was higher than the number of dynamin II-E protein colocalization foci per cell, reflecting the transient nature of dynamin II-E protein complexes. The results indicate that DENV-2 traffics within endosomes and is delivered to the PN region within 30 min. Infected cells were labeled for the presence of viral C protein at 1 h, 2 h and 4 h p.i. to detect capsid released within the cells. Absence of positive signal indicated that the concentration of released capsid was below detectable levels. The decrease in E protein-positive endosomes by 2 h indicates end of entry events. ## Association of DENV-2 E protein with microtubule (MT) The microtubule network is the major highway along which there is movement of endocytosed material. Colocalization of the E protein with alpha tubulin was therefore examined in BHK-21 cells infected at 1 MOI. Infected cells were fixed at different time points, 1 h, 2 h, 4 h, 8 h and 24 h p.i. The fixed cells were labeled first with mouse anti-E MAb followed by goat anti-mouse IgG FITC. The labeled cells were saturated with unlabeled goat anti-mouse IgG, then stained with mouse anti-alpha tubulin MAb followed by rabbit anti-mouse IgG TRITC. Close association between the two proteins was observed from 8 h and maximized at 24 h especially in the PN region, representative of the microtubule organization centre (MTOC). Mock infected cells are shown in. No colocalization was observed at early time points prior to 8 h p.i. Although confocal microscopy has its limitations, the observation of E protein associating with microtubules from 8 h onwards and not prior to that suggested that microtubules are intimately involved in trafficking of newly synthesized E protein and not the entering virions. ## Dynein dependent retrograde trafficking of DENV-2 E protein Dynein is a minus-end directed motor protein complex, which is implicated in the trafficking of viral proteins towards the PN region,. DENV morphogenesis is known to occur in the PN region in the endoplasmic reticulum (ER) and Golgi apparatus. To determine whether dynein was the motor protein involved in MT based trafficking of DENV-2 E protein, infected cells were fixed and doubled stained for dynein intermediate chain and viral E protein at different time points from 2 h to 72 h p.i. For E protein and dynein, both the primary antibodies were mouse derived therefore the labeling was done sequentially with an intervening step of blocking. After reacting the cells with mouse anti-E MAb followed by rabbit anti-mouse IgG TRITC, the cells were blocked with unlabeled goat anti-mouse IgG. Cells were then incubated with mouse anti-dynein MAb followed by donkey anti-mouse IgG Alexa 488. There was no E protein visible at 2 h (not shown). Panel 5A shows the pattern of E-dynein association at 4 h, 12 h, 24 h, 36 h, 48 h, 60 h and 72 h with the intensity profiles for both fluorochromes in the marked region of colocalization. Dynein in uninfected cells was distributed evenly in the entire cell. In the infected cell, intensity of dynein increased around the E protein. At 4 h, the E protein was present at very low concentration and colocalized with dynein. As time progressed, the concentration of E protein increased and by 12 h, a discrete bead like pattern of E-dynein complex, was seen which corresponded to the distribution of ER suggestive of trafficking to and from ER. At 24–48 h maximum concentration of the dynein-E complex was observed in the PN region. Association of viral protein in the Golgi apparatus has been reported before. At 60 h, colocalization was weak and by 72 h, dissociation between dynein and E protein was observed. Another facet of infection was revealed by the intensity profiles accompanying each split image. In infected cells the concentration of E protein and dynein were inversely proportional to each other as infection progressed indicating the requirement of dynein for early trafficking of E (intensity profiles in). Deconvolved images were used to generate 3D reconstruction of E-dynein at 4 h, 48 h and 72 h, which depicted the association followed by dissociation of dynein with E protein as infection progressed. The kinetics of association of dynein with E protein suggests that the newly synthesized E protein uses dynein for retrograde trafficking to the site of assembly and dissociates from it at 72 h when maximum virus is present in the cell supernatant. ## Co-immunoprecipitation of dynein with DENV-2 E protein Co-immunoprecipitation was used to prove that the colocalization between E protein and dynein observed by confocal microscopy represented formation of E-dynein complex. BHK-21 cells were infected with DENV-2 and lysed at 36 h p.i., a time point at which maximum colocalization was observed. IgG, purified from the serum of a dengue immune individual was used to immunoprecipitate the proteins from infected and uninfected cell lysates. The immune complexes resolved by SDS-polyacrylamide gel electrophoresis (PAGE) under non reducing condition followed by Western blot were identified with antibody directed against the intermediate chain of dynein. The reactivity of anti-dynein antibodies with cell-associated dynein is shown in Lane 1. The presence of dynein in the immune complex co-precipitated with human anti-dengue IgG coated sepharose A beads is visible in Lane 2. There was no signal in the immunoprecipitated uninfected BHK-21 lysate in Lane3. The results of the co- immunoprecipitation assay validated the colocalization of E protein-dynein visible in infected cells by confocal microscopy. ## In silico docking of E protein with dynein The light chain of dynein (LC8) bound to residues 128–138 of intermediate chain (IC74) is reported to contain the cargo binding site of the MT- dependent motor protein complex. Hex (Version 5.1), which is a molecular superimposition and docking program based on 8 Dimensional fast Fourier transform (FFT), was used to carry out docking between the E protein of DENV-2 and dynein using structures downloaded from the Protein Data Bank (PDB). DENV-2 E monomer (PDB:1TG8) was docked with dynein (PDB:2P2T). The model with the lowest energy was selected for further analysis using Accelrys Discovery Studio visualizer (version 2.5). Five residues on the dynein molecule were found to interact with five residues on the E protein. The five residues identified in the E protein were conserved in 87 other strains of DENV-2 (protein sequences downloaded from Genbank) when analyzed by Clustal W2 (MEGA 5.03). All five residues were located in domain I of the E protein but were discontinuous. Whether the residues are actually involved in binding will require further studies. ## Loss of Dynein motor activity by over expression of Dynamitin Dynein along with its cofactor dynactin, mediates MT-dependent retrograde trafficking. Over expression of p50/dynamitin, a subunit of dynactin acts as a dominant-negative inhibitor of dynein-dynactin interaction which blocks dynein mediated transport. The effect of dynamitin over expression on DENV-2 E protein trafficking was investigated. Cells were transfected with GFP-dynamitin and infected with DENV-2 at 6 h post transfection. The infected cells were stained for E protein at 36 h p.i., the time point at which maximum colocalization of E protein with dynein had been observed. In control nontransfected infected cultures, \>90% of the cells were infected and positive for E protein which accumulated in the PN region. In transfected infected cultures GFP-dynamitin positive cells showed low expression of E protein in the cytosol and not in the PN region as seen in adjacent dynamitin-negative cells. Therefore there was inhibition of expression of E protein and its trafficking to the PN region. The effect of over-expression of dynamitin on the transport of C protein was also investigated. DENV C protein is known to contain nuclear localization signals, and is transported to the nucleus. In the infected mock transfected cells C protein was observed in the cytosol and in the nucleus. In comparison in GFP- dynamitin transfected infected cells, the C protein was expressed but restricted to the cytosol. Cells showing high intensity of GFP-dynamitin (white arrowhead) had no detectable C protein. A similar pattern was also seen for the E protein (not shown). The results thus unequivocally proved the role of dynein in trafficking of DENV-2 E and C proteins. # Discussion The current study was undertaken to determine the interactions of cellular proteins with E protein during the internalization and trafficking of DENV-2 in mammalian fibroblast cells (BHK-21) using confocal microscopy. DENV has been shown to gain entry into different cell types using different modes of internalisation, i.e. clathrin/caveolae dependent endocytosis and fusion at PM. We studied the dynamin II dependency of virus entry. Dynamin II belongs to a family of GTPases associated with formation of nascent vesicles and promotion of vesicle fission and is closely associated with clathrin-mediated endocytosis. We were successful in showing a strong association between DENV-2 E protein and dynamin II using confocal microscopy. The E protein is the major surface glycoprotein of DENV, which is responsible for engaging the receptor. Therefore visualising the E protein in the early steps of virus entry was considered representative of tracking the virion. We used the E-dynamin II association to determine the kinetics of DENV-2 entry. Colocalization of DENV-2 E protein with dynamin II within seconds was in concordance with the function of dynamin II, which during clathrin mediated endocytosis aggregates around the vesicle and serves as pinchase-like mechanoenzyme to facilitate the formation of endocytic vesicles by severing nascent endocytic pits from the PM. The higher intensity of dynamin II at the foci of colocalization was perhaps a reflection of dynamin II aggregation at the site of endocytosis. The observation of lower intensity of E protein in the foci of colocalization compared to the points of no colocalization could be due to lower accessibility of the E protein to the antibody once the virus had entered the process of endocytosis. The presence of large amounts free virions on the cell surface was supported by the detection of substantial viral RNA at 0–6 h p.i. by real time RT-PCR. This kind of visualization of the endocytosis process is being shown for the first time. The maximum internalization occurs within the first 15 min of infection, however the E-dynamin II complex could be visualized at the cell surface till 30 min. This indicates that DENV-2 internalization is a process that goes on till 30 min p.i. The kinetics with confocal microscopy was in agreement with the internalization kinetics reported earlier by electron microscopy and estimation of internalized infectious virus by PFU assay, where maximum internalization of virus occurs within the first few minutes of infection. Labelling for endosomes and E protein revealed that DENV-2 localised in endosomes within 0 min p.i. and by 30 min the E protein positive endosomes were observed mostly in the PN region. Therefore internalization and the process of virus trafficking to the PN region via endosomes continue for 30 min. Association of dengue virus with early and late endosomes has been shown at 3 min and 17 min post infection respectively by live cell imaging. The absence of endosome free E protein in the cytoplasm during this period indicated that virus travels to the site of replication protected in the endocytic vesicles and uncoating occurs directly at the site of replication. Following endocytosis the low pH of the endosomes results in trimerization of the envelope protein, followed by fusion and uncoating of nucleocapsid. It has been reported for other flaviviruses that most of the steps in replication occur in the PN region. Intracellular shuttling of both proteins and vesicles requires cytoskeletal filaments and molecular motor proteins. Staining for MT and DENV-2 E protein showed close association between the two from 8 h onwards, with viral protein accumulating in the PN region. Cytoplasmic dynein transports cargo toward the microtubule minus end. Co staining with dynein revealed colocalization of DENV-2 E protein with dynein from 4 h p.i. Before this time point the E protein could not be seen either in association with dynein or independent. The last sighting of E protein was at 2 h within endosomes. Transcription of viral RNA has been reported to occur by 2 h. It follows therefore that uncoating and the first round of translation of genomic RNA should be occurring before 2 h. The fact that the E protein could not be observed in the cells till 4 h indicated that the concentration of the initially translated protein is too low to be detected by immunofluorescence assay. Therefore the E protein was seen colocalized with dynein only after the *de novo* synthesis of viral proteins had reached a substantial concentration. By 12 h most of the dynein was engaged by the viral E protein. The E-dynein complex aggregated to the PN region by 36 h, which is now known to be the region of high viral activity. Further proof of binding of E protein to dynein was provided by co-immunoprecipitation at 36 h post infection and also by protein-protein docking analysis. A putative site of interaction comprising of five residues was identified on the E protein. Three of the five residues are present in the recognition sequence on dynein-binding cargo proteins. Dynein mediated transport of the E protein on the microtubules is perhaps involved in the movement of E protein to the ER and from the ER to the Golgi apparatus, the major players in DENV morphogenesis. The loss of association by 60–72 h was co-incidental with maximum release of virus into the supernatant of infected cultures as proved by real time RT-PCR data. Loss of association with dynein indirectly suggests reduction in synthesis of E protein. Therefore, it is possible that when there is peak virus production, there is shutdown of viral protein synthesis. To prove that dynein mediated trafficking was essential for DENV-2 replication, dynein motor activity was disrupted by over expression of dynamitin and its effect was seen on the expression of viral E protein. Dynamitin, a subunit of dynein dynactin complex, is required for cargo transport. Over expression of dynamitin resulted in lower concentrations of viral E and C proteins and inhibited their translocation to target sites. Higher expression of dynamitin resulted in total inhibition of E and C protein expression, which was indicative of inhibition of virus replication. Thus dynein was crucial to the trafficking of newly synthesized DENV structural proteins, E and C. In conclusion the virions gain entry via dynamin II assisted endocytosis. They are translocated from cell periphery to PN region within endosomes. The newly translated DENV-2 protein binds to dynein and traffics on MT to PN region which is known to be the site of assembly. The association of dynein in the intracellular transport of DENV-2 proteins is being shown for the first time. We thank Dr. K. Guru Kumar for his help during Real time RT-PCR and Dr. A. C. Mishra, Director, NIV for his support. [^1]: Conceived and designed the experiments: NS SS DC. Performed the experiments: NS SS JK. Analyzed the data: NS SS DC. Contributed reagents/materials/analysis tools: PSS. Wrote the paper: NS DC. [^2]: The authors have declared that no competing interests exist.

# Introduction Integrins, the major family of cellular receptors for extracellular matrix proteins, comprise 18 α and 8 β subunits, which assemble into 24 known αβ heterodimers with different ligand binding specificities. Gene targeting studies in mice have revealed that integrins have essential functions in a wide array of developmental and homeostatic processes, ranging from embryo implantation and placenta formation early in development, to blood clotting and immunocyte function in adult animals. Within the integrin family, the laminin-binding integrins, α3β1, α6β1, α6β4, and α7β1, constitute a distinct subfamily. These integrins play essential roles in the morphogenesis and maintenance of skin, kidney, and lung epithelia (α3 and α6 integrins) and muscle (α7 integrin) by binding to laminin isoforms in the basement membranes underlying these tissues. In addition to ligand preference, the laminin-binding integrins share other biochemical similarities, including palmitoylation of the α3, α6, and β4 integrin subunit cytoplasmic tails, and physical interactions with tetraspanin proteins in the cell membrane. Tetraspanins are a family of 33 proteins in mammals that are characterized by 4 transmembrane domains, cytoplasmic amino and carboxyl termini, and one small and one large extracellular loop, which contains a characteristic cysteine motif. Tetraspanins interact with themselves (both homotypically and heterotypically) and with a subset of other integral membrane proteins, including integrins, to assemble multi-protein complexes within dynamic membrane domains termed tetraspanin-enriched membrane microdomains (TEMs). Localization of the laminin- binding integrins to TEMs may provide access to a distinct array of cytoplasmic signaling proteins, including PI 4-kinase, ERM proteins, and classical PKC isoforms,. The laminin-binding integrins have been extensively studied in the context of tumor cell biology because of their potent, context-dependent functions in regulating tumor formation, progression, invasion, and metastasis. Where studied, α7 integrin functions as a suppressor of tumor growth and metastasis in a variety of tumor types. In contrast, α6β4 integrin promotes metastatic progression in breast and skin carcinoma, and α6β1 integrin exerts pro-survival and pro-metastatic functions in prostate carcinoma. For α3β1 integrin the picture is complex. While α3β1 can promote breast cancer tumorigenesis in vivo, it is sometimes lost during prostate cancer progression, and forced α3β1 expression can suppress the growth of rhabdosarcoma in vivo and block skin carcinoma progression. Thus, in order to predict whether α3β1 will exert a pro- or anti-metastatic influence in specific cases, more insight into the molecular mechanisms of α3β1 function in tumor cells is required. Pro-metastatic functions for α3β1 integrin in breast cancer may involve multiple mechanisms, including promoting (i) Cox-2 expression, (ii) matrix metalloproteinase MMP-9 secretion, (iii) tumor cell crosstalk with endothelial cells, and (iv) Src, FAK, and Rac activation,. The activation of a Src/FAK-Rac signaling pathway may underlie α3β1's ability to promote unusually rapid migration on its laminin ligands, laminin-332 (LM-332) and laminin-511. Integrin α3β1-dependent motility on these ligands can be as much as ∼4–5 fold more rapid than motility on non-α3β1 ligands, such as fibronectin or collagen I. Many α3β1 functions in breast cancer cells may depend strongly on its association with tetraspanin proteins. Silencing of tetraspanin CD151, a major α3β1 integrin partner, has several effects on tumor cell behavior in the MDA- MB-231 breast carcinoma model, including (i) reduced migration toward LM-332 in transwell assays, (ii) reduced Matrigel invasion towards EGF, reduced EGF- induced spreading on laminin-111 (LM-111), and reduced subcutaneous or orthotopic growth upon implantation in nude mice, (iii) reduced subcutaneous tumor growth and reduced scattering in 3D Matrigel in response to endothelial cell conditioned medium, (iv) reduced TGFβ1-induced cell scattering in 3D Matrigel, and reduced lung colonization upon tail vein injection in nude mice, and (v) reduced adhesion on LM-332 and LM-111, impaired cable formation on 3D Matrigel, and impaired spreading on LM-111 in response to phorbol ester. While the findings described above support the view that CD151 makes critical contributions to the functions of laminin-binding integrins in breast cancer cells, most of these studies either focused primarily on α6 integrin, or utilized substrates such as Matrigel for which both α3 and α6 integrins can make functional contributions. Thus, it remains unclear whether the mechanisms by which tetraspanins regulate α3β1 integrin in breast carcinoma cells are identical to the mechanisms by which they regulate α6 integrins. Moreover, CD151 loss-of-function phenotypes are often interpreted in terms of CD151's ability to link its integrin partners to other TEM-resident proteins, but the loss-of- function phenotypes of other TEM resident proteins have generally not been carefully compared to those of CD151 in side-by-side studies. To begin to clarify which aspects of CD151's α3β1 integrin-regulatory functions may depend on which other TEM-resident proteins, we created breast carcinoma cells with profound RNAi-mediated silencing of CD9 and CD81, two closely related tetraspanins that engage in similar biochemical interactions – and which strongly depend on CD151 for association with α3β1. We then compared the phenotypes of our CD9/CD81-silenced cells directly to the phenotypes of CD151-silenced cells. Our data reveal that the CD9/CD81 complex and CD151 have overlapping but distinct functions in regulating α3β1-dependent behaviors in the MDA-MB-231 breast cancer model. Surprisingly, several CD9/CD81 functions, including promoting α3β1's association with PKCα and promoting α3β1-dependent tumor cell migration on LM-332, may not require CD151-dependent complex formation with α3β1 in this system. # Materials and Methods ## Antibodies, Extracellular Matrix Proteins, and Pharmacological Reagents Antibodies used in this were anti-CD9, ALB6, (Meridian Life Science); anti-CD81, M38, anti-CD151, 5C11 ; anti-α3 integrin, A3-X8 and A3-IIF5 ; anti-α6 integrin, GoH3 (eBioscience); anti-CD55, BRIC-216 (Millipore); rabbit monoclonal anti-PKCα (Abcam ab32376); rabbit polyclonal anti-PKCα (Santa Cruz sc208); anti-β-actin, AC-15 (Sigma); and polyclonal anti-α3 integrin cytoplasmic tail antibody, A3-CYT. Secondary reagents used were Alexa 488-goat anti-mouse, and Alexa 680-goat anti-rabbit (both from Invitrogen), Cy2 goat anti-mouse (Jackson ImmunoResearch), and IRDye 800-goat anti-mouse (Rockland Immunochemicals, Inc). Rat tail collagen I and Matrigel were from BD Bioscience. LM-332 was purified from SCC25 squamous carcinoma cell-conditioned medium as previously described. Phorbol-12-myristate-13-acetate (PMA) was from Sigma, and classical PKC isoform inhibitor, Gö6976, was from Tocris Bioscience. ## Cell culture, RNAi and Retroviral Transduction MDA-MB-231 breast carcinoma (ATCC), A431 epithelial carcinoma (available from ATCC; obtained from the lab of Martin Hemler, Dana-Farber Cancer Institute) and GP2-293 retroviral packaging cells (Clontech) were cultured in high glucose DME medium supplemented with 10% fetal bovine serum, 2 mM L-glutamine, 100 U/ml Penicillin, and 100 ug/ml Streptomycin (Invitrogen). For RNAi, double-stranded oligonucleotides encoding short hairpin RNAs (shRNAs) targeting the human CD9, or CD81, or CD151 mRNAs were annealed and then cloned into the pSIREN-RetroQ retroviral vector (BD Biosciences) as described. The shRNA targeting sequences used were CD9: 5′-AGAGCAUCUUCGAGCAAGAAA-3′; CD81: 5′-GGUCAUCCUG-UUUGCCUGUGA-3′, and CD151 sh3: 5′-AGUACCUGCUGUUUACCUACA-3′. To facilitate simultaneous knockdown of CD9 and CD81, the puromycin selectable marker of pSIREN-RetroQ was replaced with a hygromycin resistance gene using recombinant PCR, and the CD9 targeting construct was cloned into this modified pSIREN-RetroQ-hygro vector. Retroviral particles produced in GP2-293 cells were used to transduce MD-MBA-231 or A431 cells and stable transductants were selected with puromycin, hygromycin, or both. Stably transduced cells were sorted by flow cytometry for loss of CD9, CD81, or CD151, and maintained as polyclonal populations. For CD9 or CD81 re-expression in the CD9/CD81 doubly-silenced cells, recombinant PCR was used to generate CD9 and CD81 cDNAs containing two silent mutations within the RNAi target sequences. These constructs, named CD9RX and CD81RX, and were cloned into the pLXIZ retroviral expression vector for transduction into CD9/CD81-silenced cells, followed by zeomycin selection. Re-expression of CD9 or CD81 in the CD9/CD81-silenced cells was confirmed by flow cytometry. CD151 re- expression in the CD151-silenced cells was accomplished in the same manner, using our previously published CD151RX construct. ## Cell Spreading Assay To assess short term cell spreading, 5×10⁵ cells were plated in a T25 flask on day 1. On day 2, cells were starved overnight in serum-free medium (SFM), which was DMEM with 5 mg/ml cell culture grade BSA (Sigma cat. \#A1470) and 25 mM HEPES pH 7.2. On day 3, cells were treated with trypsin-EDTA for 3 minutes and harvested in SFM supplemented with 0.1 mg/ml soybean trypsin inhibitor and 20 µg/ml DNAse I (Worthington Biochemical, Lakewood, NJ). Cells were collected by centrifugation, resuspended in SFM, and plated at 2.5×10⁴ cells/cm² in 35 mm dishes or glass coverslips that had been coated with 1 µg/ml LM-332 or 20 µg/ml collagen I. After 30 min, cells were fixed, and photographed on a Leica DMIRE2 inverted microscope using a 20× phase objective. NIH Image J software was used to measure the spread area of at least 25 cells per cell type. Specific spreading was calculated by subtracting the mean area of unspread cells (fixed immediately after plating) from the spread cell areas measured at the end of the assay, as in. In some experiments, α3 integrin function-blocking antibody, A3-IIF5, or α6 integrin function blocking antibody, GoH3, was added to cells at 10 µg/ml for 10 min prior to plating. ## Time-lapse Cell Motility Assay Cells were serum-starved and harvested as described for the cell spreading assays above and then plated in 35 mm glass bottom dishes (MatTek Corp.) that had been coated with 1 µg/ml LM-332 or 20 µg/ml collagen I. Cells were allowed to attach and spread and then imaged using a stage incubator and video- microscope system described previously. Images were collected every 2 minutes for 3 hours, unless otherwise specified. Image J software was used to record XY coordinates of at least 20 cell centroids per movie in 4 minute intervals. Only cells that remained in view for at least 1 h were included in the analysis. Custom Java software was used to calculate mean velocity, net distance traveled, and persistence (net distance traveled/total distance traveled). In some assays, function blocking antibodies or the PKC inhibitor Gö6976 were added after 2 h, and cell motility was monitored for an additional 2–3 h. Quantification of rear cell membrane projections (tails) was performed by measuring (i) the total number of projections that lasted at least 4 min in at least 25 cells per cell type, and (ii) the total duration for each of the observed tails. One way ANOVA with Tukey post test was used to assess statistical significance of differences among cell types. ## Immunoprecipitation and Immunoblotting Cells were lysed in 20 mM HEPES pH 7.2, 150 mM NaCl, 1 mM MgCl₂ with 1% detergent, protease inhibitors (2 mM PMSF, 10 µg/ml aprotinin, 5 µg/ml leupeptin, and 5 µg/ml E-64) (Roche Diagnostics) and HALT phosphatase inhibitors (ThermoFisher Pierce). Detergents used were (i) Brij 99 (Acros Organics), (ii) a 1∶1 mixture of Brij 96V (Fluka) and Brij 99, or (iii) Brij 58 (Acros Organics). Lysate protein concentrations were normalized on the basis of a Red 660 protein assay (G Biosciences). Primary antibodies were added at 5 µg/ml, and immune complexes were collected with protein G sepharose (ThermoFisher Pierce) and analyzed by SDS-PAGE and immunoblotting. Blots were blocked with AquaBlock (East Coast Bio), and developed with primary antibody, followed by IRDye 800-goat anti-mouse or Alexa 680-goat anti-rabbit secondary antibodies, and scanning with an Odyssey infrared imaging system (LI-COR Biosciences). For PKCα association studies, cells were stimulated with 100 nM PMA for 30 min prior to lysis and co- immunoprecipitation experiments, as previously described. Immunoprecipitates were analyzed by SDS-PAGE under reducing conditions. To estimate the fraction of total cellular PKCα associated with α3β1 integrin, PKCα band intensities were quantified with Image Studio Lite software (LI-COR Biosciences), and the amount of PKCα co-precipitating with α3 integrin was divided by the amount of total PKCα detected in lysates, correcting for the volume of lysate input into the immunoprecipitation and the volume of lysate loaded on the gel. ## Adhesion Assay Cells were serum-starved, harvested, and resuspended in SFM, as described above for the cell spreading assay. Next, 1×10⁴ cells per well were plated in 96 well plates coated with 1 µg/ml LM-332, 20 µg/ml collagen I, or 10 mg/ml heat-inactivated BSA (negative control). After 45 min non-adherent cells were removed by rinsing, and adherent cells were fixed, stained with crystal violet, and quantified, as previously described. ## Cell Growth in 3D Matrigel Cells were resuspended in growth factor reduced Matrigel and plated at 5000 cells/well in a 24 well plate over a pre-polymerized cushion of Matrigel. After polymerizing for 30 min at 37°C, the wells were overlaid with standard growth medium. Colony growth was quantified by measuring the area of at least 25 colonies for each cell type at 1, 4, and 5 weeks post-plating, using Image J software. ANOVA with post-hoc t tests was used to determine the statistical significance of observed differences. # Results ## Efficient silencing of tetraspanins CD9 and CD81 in MDA-MB-231 breast cancer cells Tetraspanins CD9 and CD81 strongly associate with one another and with a similar spectrum of tetraspanin partners, including the immunoglobin superfamily proteins EWI-2 and EWI-F/CD9P-1. After preliminary experiments, in which silencing CD9 or CD81 alone in MDA-MB-231 breast cancer cells produced only modest phenotypic changes, we established cells with stable silencing of both tetraspanins (CD9/81si cells). Flow cytometry revealed that cell surface CD9 expression was reduced ∼90% and cell surface CD81 expression by ∼80% in the CD9/81si cells, as compared to parental MDA-MB-231 cells. For specificity controls, we selectively restored CD9 (CD9RX cells) or CD81 (CD81RX cells) using expression constructs designed to evade CD9 or CD81 RNAi. All cell lines were maintained as stable, polyclonal populations. ## Delayed spreading of CD9/CD81-deficient cells on LM-332 To begin to explore how the CD9/CD81 complex regulates α3β1 integrin function, we first determined the contributions of the two major MDA-MB-231 laminin receptors, α3 and α6 integrins, to cell attachment and spreading on LM-332. Functional blockade of α3 integrin, but not α6 integrin, virtually abolished cell attachment and spreading on LM-332. For cells that had already attached and begun migrating on LM-332, addition of an anti-α3 integrin function blocking antibody caused rapid cell rounding and a cessation of motility, while an anti-α6 integrin antibody had no effect within 2 h after addition. These data established that MDA-MB-231 cell responses to LM-332 are strongly α3β1 integrin- dependent. Spreading of CD9/CD81si cells on LM-332 was significantly delayed compared to the parental cells, as visualized 30 min after plating. Restoring either CD9 or CD81 expression in the CD9/81si cells restored spreading to near normal levels. Quantification of cell spreading area (the total spread area minus the mean area of unspread cells) confirmed an ∼60% reduction for the CD9/81si cells compared to parental cells, which was reversed in the CD9RX and CD81RX cells. The ability of either CD9 or CD81 re-expression to reverse the spreading defect of the CD9/CD81si cells was expected since silencing either tetraspanin individually was not sufficient to uncover a spreading defect. Thus, the presence of either CD9 or CD81 is required for efficient α3β1 integrin-dependent spreading at early time points. At 60 min after plating, all cell types were equally well spread (not shown), thus the loss of CD9 and CD81 delayed spreading, but did not ultimately prevent it. ## Impaired directed migration for CD9/CD81-deficient cells on LM-332 We used time-lapse video-microscopy to study post-attachment cell behavior on LM-332. Compared to the parental MDA-MB-231 cells, CD9/CD81si cells showed a significant reduction in the net distance traveled during the observation period.. Contributing to this reduction in net migration distance was both a modest reduction in the absolute migration velocity of CD9/CD81si cells, and a somewhat larger reduction in the directional persistence of migration. Persistence, a measure of the extent to which cells continue to migrate in the same direction over time, is defined here as the ratio of net distance traveled to total distance traveled. Re-expression of CD9 or CD81 in the CD9/CD81si cells partially or completely restored cell migration parameters to wild type levels. The altered migration pattern of the CD9/CD81si cells is readily appreciated when their cell tracks are plotted. Compared to parental and rescue cells, the CD9/CD81si cells migrated shorter distances before turning and migrating in another direction, resulting in more highly convoluted tracks. Together, the cell motility data in revealed that α3β1 integrin-dependent directed cell migration is defective in cells depleted of CD9 and CD81. ## Front-rear morphology is altered in CD9/CD81-silenced cells The defective directed migration of the CD9/CD81si cells may be related to altered front-rear morphology in these cells. Parental MDA-MB-231 cells migrating on LM-332 frequently developed long retraction tails at the rear of the cell. Both the frequency and duration of retraction tails was dramatically reduced in the CD9/CD81si cells. Re-expressing CD81, and to a lesser extent, CD9, restored retraction tail frequency and duration to more normal values. In addition to altered retraction tail morphology, CD9/CD81si cells also displayed altered localization of cortactin. Cortactin is normally recruited to the leading edge of lamellipodia, where it promotes the formation of new adhesions and lamellipodial persistence. In parental cells, cortactin typically co- localized with F-actin in a thin zone at the very leading edge of lamellipodia (top panels). In contrast, in the CD9/CD81si cells, although thick zones of F-actin were often present at the lamellipodial leading edge, cortactin staining was often absent from this region (bottom panels), suggesting that leading edge identity, actin dynamics, or adhesion formation may be altered in CD9/CD81si cells. The altered relationship between cortactin and F-actin localization in the CD9/CD81si cells can be appreciated from pixel intensity scans intersecting the lamellipodia of each cell type (graphs). Quantification revealed that ∼84% of parental cells stained positive for leading edge cortactin, while only ∼24% of CD9/CD81si cells were positive for leading edge cortactin (p\<0.0001, Fisher's exact test;). Thus, CD9/CD81si cells displayed morphological and/or molecular differences at both the leading and trailing edges of migrating cells. ## The CD9/CD81 complex and tetraspanin CD151 regulate overlapping but distinct aspects of α3β1 function in MDA-MB-231 cells Several previous studies have established that tetraspanin CD151 engages in stable, direct interactions with alpha subunits of laminin-binding integrins, – and promotes laminin-binding integrin association with other components of TEMs, including CD9 and CD81. To compare the function of CD151 to that of the CD9/CD81 complex in regulating α3β1 integrin, we established CD151-silenced MDA-MB-231 cells (CD151si cells). Flow cytometry revealed near total silencing of CD151 in these cells, with minimal impact on the expression levels of CD9, CD81, or α3 integrin. Co-immunoprecipitation experiments showed that a substantial fraction of the total α3 integrin (as measured by direct α3 immunoprecipitation) was recovered in either CD9 or CD151 immunoprecipitations (compare lanes 1&3 to lane 4). Much less α3 was detected co-precipitating with CD81 (lane 2), but we suspect that this may be due to epitope shielding of CD81 in mild detergent lysates (EGW & CSS unpublished observation). Silencing CD9 and CD81 virtually abolished recovery of α3 integrin in a CD9 immunoprecipitate, as expected (lane 5), but had little effect on the association of α3 with CD151 (lane 7). In contrast, in the CD151si cells, the CD9-α3 association was disrupted (lane 9) in addition to the anticipated loss of α3 recovery in a CD151 immunoprecipitate (lane 12). Collectively, these data show that CD151 is required to promote α3-CD9 association, but CD9 is not required to promote α3-CD151 association. This result is in accord with our prior study showing that CD151 may act as a direct physical linker of laminin-binding integrins to other TEM-resident proteins. To compare the functional impact of CD151 depletion to that of CD9/CD81 depletion, we performed additional adhesion, spreading, and motility experiments on LM-332. CD151si cells displayed a dramatic loss of adhesion in short term assays on LM-332, while CD9/CD81si cells adhered equally as well as the parental MDA-MB-231 cells. All cell types adhered well to the α2β1 integrin ligand, collagen I, with CD151si cells perhaps even showing enhanced adhesion compared to parental cells. In cell spreading assays on LM-332, CD9/CD81si cells again displayed delayed spreading, as previously observed in. However the CD151si cells showed an even more profound spreading defect on LM-332 than the CD9/CD81si cells. By ∼1 h post-plating, all 3 cell types had spread equally well (not shown), indicating that the spreading defect on LM-332 is transient, as previously observed. In contrast to the results on LM-332, all three cell types displayed similar rapid spreading on collagen I. The more severe spreading and adhesion defects in CD151si cells on LM-332 might have been anticipated, given that in CD151si cells, not only is CD151 virtually absent, but α3 association with the CD9/CD81 complex is also largely abrogated. However, analysis of post-spreading cell behaviors yielded unexpected results. In motility assays on LM-332 that were initiated after spreading was completed, the CD9/CD81si cells displayed reduced migration velocity, persistence, and net distance traveled. Surprisingly, the CD151si cells, once attached and spread, displayed wild type migration velocity, persistence, and net displacement. On collagen I, CD151si cells migrated significantly faster, and displayed significantly greater net distance traveled than either parental or CD9/CD81si cells. The directional persistence of CD9/81si cells appeared modestly reduced on collagen I, but the difference was not statistically significant. These results indicate that, although CD151 is critical for α3β1 integrin-dependent adhesion and spreading on LM-332, it appears less important for post-attachment migration in this system. Remarkably, the CD9/CD81 complex continues to be important for α3β1-dependent directed cell migration after cell spreading. This CD9/CD81 function might not involve a direct integrin association, since CD151si cells displayed wild type motility on LM-332, despite the fact that α3β1-CD9/CD81 association was disrupted in these cells (See). Consistent with the possible involvement of altered front-rear cell morphology in the CD9/CD81si cell migration phenotype, CD151si cells displayed a wild type number of tail retractions during migration on LM-332. To further explore potential functional differences between CD151 and the CD9/CD81 complex, we next examined cell growth in 3D Matrigel, a behavior to which both α3 and α6 integrins are expected to contribute. Over a 35 d assay, there was little apparent difference in colony size between parental and CD151si MDA-MB-231 cells, consistent with a previous report. In contrast, CD9/CD81si colony size was significantly reduced. Quantification revealed that CD151si and parental cell colonies were virtually identical through 28 d of growth, with CD151si colony size perhaps leveling off somewhat more by day 35. In contrast, CD9/CD81si colonies were significantly smaller than parental colonies at all time points examined. These data provide another example of a cell behavior that is regulated differently by the CD9/CD81 complex than by CD151. ## α3β1 integrin-PKCα association is disrupted by depletion of CD9/CD81, but not by depletion of CD151 in MDA-MB-231 cells Tetraspanin proteins may regulate laminin-binding integrin function in part by promoting integrin association with activated classical protein kinase C (PKC) isoforms,. To explore potential PKC contributions to the phenotypes of our CD9/CD81si and CD151si cells, we first examined α3β1 integrin-PKCα association in these cells upon PMA stimulation. Preliminary experiments revealed that PMA- stimulation significantly enhances α3β1-PKCα association above basal levels, as previously reported. In Brij 99 detergent lysates, in which PKCα-α3β1 integrin association is preserved, PKCα co-precipitated with CD9, CD151, and α3 integrin, but not with the CD55 negative control in parental MDA-MB-231 cells (lanes 1–4). Co-precipitation of PKCα with CD9, CD151, and α3 integrin was dramatically reduced in the CD9/CD81si cell lysate (lanes 5–7). Surprisingly, in a CD151si cell lysate, co-precipitation of PKCα with both CD9 and α3 integrin was maintained at wild type levels (, lanes 9 & 11), although it was lost from the CD151 immunoprecipitate, as expected (lane 10). Immunoblotting PKCα and β-actin in cell lysates confirmed that PKCα expression is unchanged in tetraspanin- silenced cells, and that similar amounts of total protein were input into each set of immunoprecipitations. Estimation of the fraction of total cellular PKCα that associates with α3β1 integrin in each cell type revealed an ∼75–85% reduction in α3β1-associated PKCα in the CD9/CD81si cells compared to parental or CD151si cells. In a recent study, the contribution of CD151 to a PKCα-α6β4 integrin association was assayed in Brij 58, a milder detergent than Brij 99. To determine whether there might be a larger pool of PKCα that could associate with α3β1 integrin in a CD151-dependent manner in a milder detergent, we repeated our analysis using Brij 58 lysates. However, we obtained identical results in Brij 58 as in Brij 99. To further confirm that CD151 expression is not essential for PKCα-α3β1 integrin association in MDA-MB-231 cells, we restored CD151 expression to the CD151si cells using an RNAi-resistant CD151 cDNA. Flow cytometry confirmed restoration of CD151 expression in these cells (CD151RX cells; Table I). Restoration of CD151 re-established α3β1 association with tetraspanin CD9, and restored adhesion on LM-332. Despite modulating association with CD9 down and then back up, neither silencing nor restoring CD151 expression reduced the amount of PKCα that co-precipitated with α3β1 integrin compared to that seen in wild type parental cells. Collectively, these data indicate that CD151 is not required to promote PKCα-α3 integrin in MDA-MB-231 cells, but that the CD9/CD81 complex indeed plays an important role. The loss of PKCα-α3β1 association in the CD9/CD81si cells is not due to a gross alteration in the ability of PKCα to translocate to the plasma membrane, since we observed similar membrane localization of PKCα upon immunostaining either parental or CD9/CD81si cells. Thus, while the mechanism by which the CD9/CD81 complex regulates PKCα-α3β1 association remains to be determined, our data indicate that it is likely to be more specific than a global disruption of PKCα's ability to associate with the plasma membrane. Since both α3β1-dependent directed migration and α3β1-PKCα association were impaired in the CD9/CD81si cells, but not the CD151si cells, we next tested for a possible functional role of PKCα in α3β1-driven cell motility. Time-lapse video-microscopy revealed that upon addition of the classical PKC isoform inhibitor, Gö6976, parental MDA-MB-231 cells migrating on LM-332 slowed significantly and displayed a modest reduction in persistence, resulting in a substantial reduction in net distance traveled. Thus, an inhibitor of PKCα recapitulated key motility phenotypes observed in CD9/CD81si cells. ## The CD9/CD81 complex regulates α3β1 integrin-dependent motility in an alternate tumor cell system To begin to investigate the generality of the role of the CD9/CD81 complex in regulating α3β1 function, we also created CD9/CD81-silenced A431 carcinoma cells. Flow cytometry confirmed that CD9 was 95% silenced and CD81 was 89% silenced in these cells. Compared to wild type parental cells, the CD9/CD81si A431 cells displayed normal adhesion on LM-332 (data not shown), as we had observed in our MDA-MB-231 cells. However, in two separate trials, the CD9/CD81si A431 cells displayed significantly reduced migration velocity and net distance traveled. There was also a trend towards reduced directional persistence that reached statistical significance in one trial, but not the other. Overall, these data indicate that the ability of the CD9/CD81 complex to promote α3β1 integrin-dependent motility is not restricted to the MDA-MB-231 breast cancer model, although, unlike MDA-MB-231 cells, A431 cells also exhibit an ongoing requirement for CD151 for rapid migration on LM-332. # Discussion ## Impact of CD9/CD81 depletion on α3β1 integrin function in tumor cells The first major finding of our study is that the CD9/CD81 complex is in fact required for certain aspects of α3β1 integrin function in tumor cells. Numerous studies have documented that depleting or genetically ablating tetraspanin CD151 significantly impairs the functions of the laminin-binding α3 and α6 integrins,,. However, far less had been known about the extent to which other tetraspanins, such as CD9 and CD81, are required for α3 or α6 integrin function. MDA-MB-231 breast carcinoma cells depleted of CD9 were reported to display enhanced proliferation in 3D Matrigel (over a 5 day period), and reduced spreading (but enhanced chemotactic migration) on fibronectin. CD9-silenced MDA- MB-231 cells also displayed reduced migration towards soluble, native collagen IV, a behavior that depended on discoidin domain receptor 1. In a recent study, an siRNA screen identified CD9 as a potent modulator of integrin function in transformed cells. RNAi silencing of CD9 reduced Matrigel invasion of multiple tumor cell types, including PC-3 and 22Rv1 prostate carcinoma, and MDA-MB-231 breast carcinoma cells. Lung invasion by CD9-silenced MDA-MB-231 cells was also curtailed compared to control cells at 48 hours after tail vein injection. While all of these studies pointed towards possible integrin-dependent functions for the CD9/CD81 complex in breast carcinoma cells, the specific integrins involved in each study were not defined, and, in particular, the specific contribution of α3β1 integrin was not assessed. In addition to the MDA-MB-231 model, reported effects of silencing or genetically ablating CD9 or CD81 in other cell types include (i) reduced dendritic cell chemotactic migration on fibronectin, (ii) reduced small cell lung carcinoma cell adhesion on fibronectin, with increased apoptosis, (iii) reduced α2β1 integrin-dependent suppression of focal adhesion formation and cell proliferation for endothelial cells on LM-111, (iv) reduced adhesion and migration of bone marrow-derived macrophage on Matrigel or fibronectin, (v) impaired αvβ5 integrin-dependent photoreceptor outer segment binding and engulfment by retinal pigmented epithelial cells, (vi) reduced expression of multiple β1 integrins and reduced cell spreading on Matrigel, LM-111, fibronectin, and collagen I in ovarian carcinoma cells, (vii) enhanced migration of bladder carcinoma cells on an unspecified substrate, and (viii) enhanced migration of primary melanocytes towards soluble fibronectin. Again, all of these studies pointed towards the CD9/CD81 complex as a regulator of integrin- dependent cell behaviors, but in many of them, the specific integrins involved were not well-defined, and none of the studies focused on α3β1 integrin, a major tetraspanin partner. Our new data now establish that depletion of the CD9/CD81 complex can have a significant impact on α3β1 integrin function in tumor cells. The loss of CD9 and CD81 delayed initial cell spreading and impaired directed cell motility on LM-332, two cell behaviors that we show here to strongly depend on α3β1 integrin but not on α6 integrins in MDA-MB-231 cells. Depletion of CD9 and CD81 had minimal impact on α2β1 integrin-dependent cell behaviors on collagen I, and re- expression of CD9 or CD81 reversed the phenotypes of CD9/CD81-silenced cells on LM-332. Thus, the impaired α3β1 functions in CD9/CD81si cells are due specifically to the loss of CD9 and CD81, and are unlikely due to off-target effects of RNAi or to a general reduction in cell motility. Similar results were obtained in A431 carcinoma cells, indicating that the ability of CD9/CD81 to regulate α3β1 integrin is not restricted to the MDA-MB-231 model. ## Overlapping but distinct effects of CD9/CD81 depletion versus CD151 depletion on α3β1 function in tumor cells A surprising finding in our study was the degree of divergence between the phenotypes of the CD9/CD81-silenced MDA-MB-231 cells and the CD151-silenced cells. CD151 function has often been rationalized in terms of a model in which CD151 physically links laminin-binding integrins to other tetraspanins and TEM proteins. A simple prediction of this model might be that the impact on α3β1 function of depleting CD151 is likely to be more severe than the impact of depleting other subsets of TEM proteins, since the loss of CD151 should largely sever α3β1's physical association with all other TEM constituents. Indeed, in CD151-silenced MDA-MB-231 cells, α3β1 association with the CD9/CD81 complex is severely reduced. Moreover, in short term adhesion and spreading assays on LM-332, CD151-silenced cells were severely impaired, while the CD9/CD81-silenced cells showed a more modest reduction in spreading that did not translate to a measurable deficit in a simple adhesion assay. However, once cells were fully spread, assays of cell motility on LM-332 revealed an ongoing role for CD9/CD81, whereas CD151 appeared dispensable. One potential explanation for this unexpected result is that during attachment and spreading, a smaller number of α3β1 receptors are initially involved in binding to the laminin substrate than the number of α3β1 receptors that are involved once the cells are fully attached and spread. Perhaps in the sub- optimal conditions that initially exist as cells first contact the laminin substrate, a physical association of α3β1 with fully intact TEMs is required for maximum function. Subsequently, when cells are fully spread and many more α3β1 receptors are engaged, association with fully intact TEMs may become less critical for α3β1 function in the MDA-MB-231 cell model. (A similar argument regarding receptor density has been made to explain the nominal effect of silencing CD151 on MCF-10A cell adhesion to LM-332). Nevertheless, even in fully spread cells, CD9/CD81 could continue to exert functions – perhaps independent of direct association with α3β1 – that are required for optimal α3β1-dependent motility. Our data indicate that such functions could include (i) helping to establish and maintain front-rear cell morphology and (ii) indirectly promoting PKCα association with α3β1. These two functions might be related, since cells expressing an α3 integrin mutant that lacks a cytoplasmic, PKC- controlled phosphorylation site display a reduction in tail retractions, reminiscent of what we observed for our CD9/CD81-silenced cells. In one hypothetical example of an indirect mechanism, CD9/CD81 could function by sequestering an unidentified factor X, which would otherwise inhibit the association of PKCα with α3β1 integrin. CD9/CD81 themselves can associate with α3β1 via CD151 in wild type cells, but CD9/CD81 continue to sequester the inhibitory factor X even when the CD9/CD81-α3β1 association is disrupted by the loss of CD151. However, when CD9/CD81 are depleted, factor X is released to block PKCα-α3β1 association. Many other indirect mechanisms are possible, including completely indirect mechanisms where CD9/CD81 could influence signaling pathways that somehow regulate PKCα-α3β1 association independently of any CD9/CD81 association with other TEM proteins. Another setting in which CD9/CD81 and CD151 functions diverged was in long term growth in 3D Matrigel. Compared to parental or CD151-silenced cells, CD9/CD81-silenced MDA-MB-231 cells displayed impaired growth over a 5 week assay. The minimal contribution of CD151 to growth in 3D Matrigel had been reported before for the MDA-MB-231 model. Both α3 and α6 integrins can make contributions to MDA-MB-231 cell behavior in Matrigel, so whether CD9/CD81 contributions in this setting reflect regulation of α3 or α6 function (or both) remains to be determined. ## Potential involvement of PKCα in the pro-migratory functions of the CD9/CD81 complex A third major finding uncovered in our study was the potential involvement of PKCα in CD9/CD81 pro-migratory functions. Congruent with the seemingly minimal contribution of CD151 to α3-dependent cell migration on LM-332, we also found that CD151 seems not to be required for α3β1 integrin-PKCα association in the MDA-MB-231 system. This result was also surprising, given that the ability of PKCβII to associate with α3β1 had previously been mapped to the α3 integrin *extracellular* domain. This would seem to strongly implicate CD151 in linking α3β1 to classical PKC isoforms, since CD151 and α3 integrin engage in a direct, extracellular interaction. One potential explanation for the apparent discrepancy between our results and the previously published study of Zhang et al is that a residual CD9/CD81 association with α3β1 integrin in our CD151-silenced cells might be sufficient to mediate all of the α3β1-PKCα association. In this scenario, the bulk of CD9/CD81 would associate with α3β1 via linkage through CD151, but a much smaller pool would associate directly. Silencing CD151 abolishes most of the CD9/CD81 association with α3β1, but the small, directly interacting pool remains and is capable of mediating all the PKCα-α3β1 association. The hypothetical mechanism depicted in may seem difficult to reconcile with our findings that CD9-α3β1 association appears virtually abolished in CD151-silenced cells. However, we and others have noted that α3β1 can retain low, residual association with other TEM-resident proteins upon near total loss of CD151 by RNAi, or even when CD151 is completely absent due to genetic deletion, (see also, lane 9 and the middle lane of in this study). Thus, although we currently favor an indirect mechanism, we cannot completely exclude the possibility that a residual CD9/CD81-α3β1 association in the CD151-silenced cells could directly mediate the PKCα-α3β1 association. We find it unlikely, however, that incomplete silencing of CD151 is responsible for preserving the wild type levels of PKCα-α3β1 association in our CD151-silenced cells. Flow cytometry revealed that CD151 expression was reduced by ∼99% in our CD151-silenced cells (Table I) with no impact on PKCα-α3β1 association. Yet silencing just 80–90% of CD9/CD81 was sufficient to reduce PKCα-α3β1 association by ∼75%. Thus, in the MDA-MB-231 system, the PKCα-α3β1 association appears to be much more sensitive to the total level of CD9/CD81 than to the amount of CD9/CD81 linked by CD151 to α3β1. Another possibility is that the rules governing α3β1-PKCα association in adherent MDA-MB-231 cells differ from those that govern α3β1-PKCβII association in non-adherent K562 cells, which were used in the Zhang et al study. For example, in the MDA-MB-231 system, the loss of CD151 may allow other as yet unidentified tetraspanins to associate directly with α3β1 integrin and recruit PKCα. Further experiments are required to clarify mechanism of α3β1-PKCα association in breast carcinoma cells. The mechanism of α3β1-PKCα association may also differ from that of α6β4 integrin-PKCα association. In the MMTV-ErbB2 mouse breast cancer model, phosphorylation of the β4 integrin cytoplasmic tail at the PKC-dependent S1424 site was reduced in the primary mammary tumors from CD151 knockout mice versus control mice. In addition, CD151-silenced MCF-10A breast cancer cells expressing ErbB2 also showed reduced β4 integrin phosphorylation at S1424 in response to EGF stimulation. Similarly, phorbol ester or EGF-induced PKCα recruitment to α6β4, and PKC-dependent phosphorylation of the β4 cytoplasmic tail, were also significantly reduced in CD151-silenced A431 cells. However, PKCα-α6β4 association and β4 tail phosphorylation under basal conditions were unaffected by the loss of CD151, and CD151 was not absolutely required for integrin-PKCα association under any condition tested. Thus, while CD151 can clearly promote PKCα association with laminin-binding integrins in certain contexts, it appears not to be required as a critical linker in all cases. Consistent with CD9/CD81-dependent PKCα association playing an important role in α3β1 function, we showed that motility on LM-332 was significantly impaired upon treatment with the classical PKC isoform inhibitor Gö6976. Potential pro-migratory consequences of PKCα association with α3β1 include regulation of α3β1 cytoskeletal association via PKC-controlled phosphorylation of the α3 integrin cytoplasmic tail, and, by analogy to α6 integrin, regulation of α3β1 diffusion mode. ## Conclusion PKCα is emerging as a potential promoter of tumor progression and metastasis. One recent study showed that a novel, PKCα inhibitory peptide can strongly inhibit spontaneous metastasis in the 4T1 murine mammary carcinoma cell model. In addition, PKCα expression level correlates with estrogen receptor/progesterone receptor-negativity, higher tumor grade, increased Ki-67 positivity, and poor prognosis in human breast cancer, and the metastasis inhibitor KiSS1 may function in part by downregulating PKCα. A recent study has also revealed a critical role for α3β1 in MDA-MB-231 breast carcinoma tumor formation in vivo. Together with these studies, our new data raise the distinct possibility that the pro-metastatic, pro-tumor activity of PKCα in breast cancer may derive in part from PKCα's ability to promote α3β1 integrin-dependent tumor cell motility. If PKCα does promote breast cancer metastasis through α3β1 integrin, our new data strongly suggest that the CD9/CD81 complex plays a key role. Since CD151 also promotes metastatic colonization in the MDA-MB-231 model (but seems not to be required for PKCα association with α3β1 in this model), simultaneously inhibiting both CD9/CD81 and CD151 functions might have an even more profound inhibitory effect on metastasis. Future studies should aim to elucidate the mechanism by which PKCα regulates α3β1 function in tumor cells, and the interplay between CD9/CD81, CD151, and other TEM-resident proteins such as the major CD9/CD81 partners, IgSF proteins EWI-2 and EWI-F/CD9P-1, in regulating metastatic cell behaviors. # Supporting Information We would like to thank C. Clayton Hazelett for helping to establish the CD9/CD81-silenced A431 cell line, and Mary E. Herndon for helpful discussions and a critical reading of the manuscript. Emily Amaro and Mousa Aboissa provided assistance with quantification of cell morphology. Cell sorting and flow cytometry were performed at the Carver College of Medicine/Holden Comprehensive Cancer Center Flow Cytometry Core. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: EAGW CSS. Performed the experiments: EAGW. Analyzed the data: EAGW CSS. Wrote the paper: EAGW CSS.

# Introduction Nutrient pollution results in man-made eutrophication, which is amongst the most pernicious forms of global change affecting aquatic ecosystems around the world. Human causes of eutrophication are the inefficient use of fertilizers, aquaculture and urban outflows and atmospheric nitrogen deposition from combustion. The ecological effects of eutrophication are well-known, including toxic algal blooms and high mortality of animals due to dissolved oxygen depletion at night. Water authorities attempt to mitigate eutrophication by establishing safe nutrient concentrations (e.g. OECD, 1982; Directive 91/676/ECC). However, the direct toxicity of nutrients to wildlife under chronic exposure is still poorly studied. Considering agricultural intensification continues unabated and water purification is costly, there is the pressing need to get better insight into the health effects that environmentally relevant nutrient concentrations have on wildlife. Nitrate (NO₃^-) is a widely distributed nutrient that naturally occurs at a low environmental concentration. However, it can reach up to 2000 mg NO₃^-/l in aquaculture tanks and 345 mg NO₃^-/l in surface waters in nitrate vulnerable zones. From 2012 to 2015 the surface area vulnerable to nitrate pollution increased from 1951898 km² to 2175861 km² just in Europe, representing 61% of the total agricultural area. Alongside surface waters, nitrate pollution degrades groundwater, with reported concentrations of more than 395 mg NO₃^-/l, which exceeds the legal thresholds for Europe (50 mg NO₃^-/l; Directive 91/676/ECC) and U.S. (44 mg NO₃^-/l; USEPA SWDA). Ground and surface waters are linked, buffering groundwater against shortages of surface water during drought. Moreover, climate change may intensify the effects of nitrate pollution on temperate rivers if shifts in precipitation regimes increase agricultural run- off. Nitrate toxicity has long attracted the attention of public health agencies after nitrate-induced oxidation of respiratory pigment (methemoglobinemia) was recorded in U.S. babies. Studies have since reported diseases other than respiratory issues in humans and in laboratory and domestic animals after drinking nitrate-polluted water, including mortality, oxidative stress, hypertension, birth defects, diabetes, impaired thyroid function, spontaneous abortions or cancer. For water-breathing animals, nitrate was generally considered of little concern, possibly because nitrate has low branchial permeability compared to the highly toxic ammonia and nitrite. This view changed after experimental evidence showed methemoglobinemia and alterations in hormone levels, behaviour, growth or in vulnerability to diseases in aquatic taxa under chronic nitrate exposure. However, these studies used eggs, juveniles or adults of one sex of different species, all of which are factors that may affect the toxic response. Moreover, toxic responses can be delayed, so that the combined use of short- (e.g. feeding assays) and long-term biomarkers (e.g. growth) will provide a more holistic view of nitrate toxicity to wildlife than the often used single-type biomarker approach. The eastern mosquitofish (*Gambusia holbrooki*) is one of the world’s worst piscine invaders, which has been introduced in many temperate regions due to a misguided strategy for mosquito control. Although extensively used in ecotoxicology, nitrate toxicity to mosquitofish has not been examined in detail. Reduced sperm counts and increased testicular weight in male mosquitofish were associated with concentrations of up to 22 mg NO₃^-/l in U.S. streams. However, there is no experimental evidence for other nitrate- induced alterations in mosquitofish. The effects of nitrate on other fish species are negative, almost neutral and even protective against a disease. Nevertheless, these studies were mostly conducted on captive-reared species, which may have a higher tolerance to nitrate than wild fish because nitrate accumulates in aquaculture tanks and nitrate pre-exposure increases tolerance. This rationale may apply to wild taxa if pre-exposure to the many pollutants occurring in natural waters induces co-tolerance. The present experimental study monitored over 8 weeks the effects of three ecologically relevant nitrate concentrations on wild males, females and juveniles of mosquitofish using endpoints associated with their ecological impact. If males are the sicker sex and juveniles are more vulnerable than adults to pollution, then we expected female mosquitofish to be the most tolerant to nitrate pollution. If the effects of nitrate pollution are subtle, then we expected nitrate effects to be more apparent in short- than in long-term biomarkers. Finally, responses in nitrate treatments should be comparable to those in controls if mosquitofish can cope with nitrate toxicity. Given the ecological relevance of the biomarkers used, our work will explore whether the ecological impact of the mosquitofish can be modulated by changing water- nutrient concentrations. # Materials and methods ## Fish origin and general fish maintenance The male and female mosquitofish used in this study were captured with dip nets in November 2012 in channels draining an agricultural area in the Llobregat river, Barcelona, Spain (41°16’52”N, 2°02’04”E). Fish were brought to the University of Barcelona in opaque plastic tanks provided with air-pumps and were acclimatised for one week to the laboratory conditions in two mixed sex stock 500 L tanks provided with an external filter, artificial plants and flowerpots for refugee. Fish were maintained in acclimation and experimental conditions as follows. A malaquite green/formaline bath was applied at a prophylactic dose upon arrival. Water was then fully renewed by using dechlorinated tap water as we did to maintain the experimental environmental conditions (see section 2.3). Water properties in the laboratory tap were: pH = 7.7, mg/l, ammonia \<0.5 mg/l, nitrite \<0.03 mg/l, nitrate = 7.4 mg/l, sulphates = 81.2 mg/l, chloride = 130 mg/l, bicarbonate = 221 mg/l and conductivity = 784 μS/cm. Pregnant females (*N* = 15) with overt signs of giving birth soon were introduced in batches of three in 100 L tanks provided with nets to collect recently newborn juveniles (N = 165). Both adults and newborn were kept under 22±1°C and 12 h light:12 h dark cycle and fed daily with crushed commercial Sera Vipan flakes and weekly with frozen bloodworms for adults and live *Artemia* nauplii for newborns. Fish were fed once daily until satiety and uneaten food and faeces were removed daily with a dipnet. Each tank had a biological filter to prevent metabolic waste built-up (NH₄⁺ and NO₂^-) and ensure water oxygenation. ## Ethics statement The experimental procedure was authorised by the Natural Environment and Biodiversity Division at the Catalan Department of Agriculture and Fisheries (Num. DAAM 8290). Fish capture and maintenance were approved by the Committee for an Ethical use of Experimental Animals at the University of Barcelona (Num. 87/15). All fish were humanely euthanized on the termination of the experiment in compliance with Spanish legislation for the management of invasive species (Real Decreto 1628/2011). ## Experimental nitrate concentrations and exposure conditions Sodium nitrate (NaNO₃, CAS Number: 7631-99-4) was used to make two nitrate solutions (50 and 250 mg NO₃^-/l, equivalent to 11.5 and 57 mg NO₃^—N/l, respectively) using dechlorinated tap water, which was also used in the control treatment (\<10 mg NO₃^-/l). The lowest nitrate concentration is the safety nitrate threshold for European waters (Directive 91/676/ECC) and the highest level is within the range reported in aquaculture and in rivers draining nitrate vulnerable zones in Europe and tropical countries. Experimental nitrate concentrations represented a 0, 5- and 25-fold increase, respectively, for mosquitofish in relation to the nitrate concentration at the collection site (9.9 ± 3.0 mg NO₃^-/l, based on quarterly water analyses over one year). Male and female mosquitofish were visually size-matched per sex (total length, female: TL = 37.6 ± 0.4 and male: 25.9 ± 0.2 mm) and exposed for 8 weeks to the experimental nitrate solutions in 20 L aquaria (N = 5 tanks per treatment and sex) with six males or females in each replicate. For juveniles, the same experimental setting was used but these were exposed in batches of 11 siblings per tank. The exposure started by increasing nitrate concentrations in each aquarium drop by drop (\~3 h) via a 5 mm ø tube connected to a tank with clean water or one of the two nitrate solutions. This drop-by-drop system was used to refill each tank with fresh water from each experimental condition after 50% of water was changed every three days. Water samples randomly analysed from the different treatments using the colorimetric kit VISOCOLOR indicated that the water quality conditions remained constant through the experiment at 24 h of the next water change. ## Overview of the biomarkers measured to appraise nitrate toxicity The effects of chronic nitrate exposure on males, females and juveniles of mosquitofish were examined using 13 variables. All of them are indicators of fish health, but calorimetry and stable isotope signatures inform the quality of fish for piscivores (see section 2.4.2). Alterations in the quantity of food eaten or in the reaction time to a stimulus show how nitrate may alter mosquitofish performance in ecosystems. Indicators related to fish growth, body condition based on mass-length relationships, energetic content and histopathology were recorded on the termination of the experiment (8 weeks). However, the feeding behaviour of mosquitofish was monitored at 0, 4, 6 and 8 weeks. All fish were euthanized at 8 weeks using an overdose of the anaesthetic MS-222. ### Growth and body condition based on mass-length relationships Male and female mosquitofish were anesthetised with MS-222 (0.02%), measured (TL, mm) and weighted (0.001g) at Time 0 and at 8 weeks. Newborns other than those used in the experiment were used to estimate the size of experiment juveniles at Time 0 (TL = 8.9 ± 1.2 mm) to avoid compromising the health of the tiny tested individuals due to handling. Fish size measures were used to calculate the specific fish growth rate (G) using the equation G = (ln L_t−ln L₀) / t_n, where ln L_t is the natural logarithm of fish length at 8 weeks, ln L₀ is that of fish length at Time 0 and t_n is the duration of the experiment (8 weeks). We ranked fish in each tank by body length at Time 0 and t_n to identify fish individuals and be able to calculate G because our tagging equipment (e.g. elastomer) is not suitable for such small fish. Moreover, we calculated the Scaled Mass Index (SMI) as an index of body condition: SMI = W_i (L_o/L_i)^bSMA, where W_i and L_i are the weight and length of each fish individual, respectively, L₀ is the arithmetic mean length of all the tested mosquitofish and b_SMA is the slope of a standardised major axis regression of the mass-length relationship. The SMI is regarded as a correlate of energy and fitness measures. ### Energetic reserves Changes in fish δ¹³C and δ¹⁵N stable isotope signatures and calorimetry were used as two complementary measures of energetic reserves. Stable isotopes are widely used in studies of trophic ecology because the isotopic composition of predator tissue is naturally altered by the type and amount of food assimilated. Moreover, tissue isotopic differences are due to changes in metabolism, including increased lipid storage, because lipids are about 6–7% depleted in ¹³C relative to protein. We used the δ¹³C and δ¹⁵N ratio as proxy for lipid content in white muscle because this is the tissue widely used in fisheries (e.g.). However, lipid content varies amongst fish tissues, so that we used the caloric content of the whole fish as an additional measure of energy content. For stable isotope analyses, we freeze-dried fish muscle samples from below the dorsal fin of 3 adult fish from each tank and we ground them to fine power. Two sub-samples of 0.30 mg each were placed into tin buckets and crimped for combustion to determine δ¹³C and δ¹⁵N using a Flash EA1112 and TC/EA coupled to a stable isotope mass spectrometer Delta C through a Conflo III interface (ThermoFinnigan). Analytical accuracy was controlled using replicate assays of certified standards indicating an analytical error of ±0.1‰ and ±0.3‰ for δ¹³C and δ¹⁵N, respectively. Isotope ratios are expressed conventionally as δ values in ppt (‰) according to the following equation: *δX* = ((*R*_sample/*R*_standard)– 1) ·1000, where *X* (‰) is ¹³C, ¹⁵N, and *R* are the corresponding ratios ¹³C/¹²C ¹⁵N/¹⁴N, related to the standard values: *R* standard for ¹³C is Pee Dee Belemnite, for ¹⁵N is atmospheric nitrogen. For calorimetry, we used three adults per tank and we pooled all juveniles from each tank to reach the detection limits of the IKA Calorimeter c7000 (Germany). Samples were oven-dried at 60ºC for 48 h, weighted and the caloric content was expressed as joules per gram (J/g). ### Histopathology A random sample of three fish from each tank was processed for histology. The head and viscera of adult fish were fixed individually in 10% buffered formalin, dehydrated in ethanol, cleared in xylene and embedded in paraffin wax. Juveniles were processed as a whole due to small size. Sagittal sections of 5 μm thick were cut in all fish at the same position and stained with conventional haematoxylin-eosin. Observations were made under an Olympus BH light microscope at 400x magnification. We focused on liver and gills because nitrate uptake is through gills and liver is the target of many toxicants. Gill alterations and the number of mucous cells in the gills were recorded and expressed as a ratio out of the 50 secondary lamellae and spaces examined. Liver alterations and the number of melanomacrophague centres were expressed as a ratio out of the number of microscope fields examined. We took photographs with a Jenoptik ProgRes C3 camera and used the ImageJ software to quantify the area (μm²) of perivisceral fat in 3 sections of each juvenile fish as an additional measure of energy storage. Outcomes were expressed as average per juvenile. ### Feeding behaviour Temporal changes in feeding behaviour were quantified in all experimental fish, which were not fed 24 h before the assay to ensure all fish were hungry. The assay was conducted in an aquarium with a plastic sheet in one side, where the tested fish were left for 3 minutes to acclimatise before the trial. The trial began when the sheet was gently removed and ten size-matched unfrozen *Artemia* adults in 200 ml of water were gently released into the experimental area on the opposed side of the tested fish. Live *Artemia* nauplii were used for juveniles. The assay was repeated individually for all fish and the same observer, sat in front of the aquarium, recorded: i) latency defined as the time spent to capture the first item, ii) voracity defined as the time needed to capture the first items (four in adults and ten in juveniles), and iii) satiety defined as the total number of items eaten without stopping \> 2 minutes. If a fish ate all 10 prey items, then we added *Artemia* individuals on the water surface until satiety. ## Statistical analyses All analyses were conducted using the R software and the functions outlined below. Spearman rank correlation coefficients were used to examine associations amongst all biomarker responses. The effects of chronic nitrate exposure on mortality, fish growth, SMI, caloric content, the C/N ratio, histopathological and behavioural measures were assessed using generalized linear mixed models (GLMMs). Data on males, females and juveniles were analysed separately because there were significant differences in all response variables. To avoid pseudoreplication, aquarium ID was included as random intercept in all models to account for the fact that fish were exposed to nitrate in batches (6 adults or 11 juveniles). For juveniles, aquarium ID was nested within mother ID as random intercept to account for systematic differences amongst clutches. Nitrate was included as fixed effect in all models. The interaction between nitrate and time was included in the behaviour model to test whether the effects of nitrate on fish varied with exposure time. The distribution of all response variables was visually inspected and the error distribution in GLMMs was chosen accordingly (e.g. Gaussian for fish growth, Poisson for the number of prey eaten). Model assumptions were checked by inspecting diagnostic plots of residuals. The function *Anova* within the package *car* was used to assess significance at P ≤ 0.05. # Results Mosquitofish were evenly distributed by size and sex amongst treatments and fish did not differ in size amongst aquaria at Time 0, either for males (F = 0.54; P = 0.90) or females (F = 1.22; P = 0.28). However, females (Mean ± S.E. = 37.6 ± 0.42 mm) were significantly bigger than males (25.9 ± 0.18 mm; t = 25.8; P \< 0.001). There was no mortality in males due to nitrate and only minor mortality was recorded for juveniles (3.6%) and females (3.3%). Nonetheless, mortality did not differ significantly between treatments (Nitrate: χ² = 1.59, P = 0.66) or sexes (Sex: χ² = 0.43, P = 0.93). ## Growth, body condition and energetic reserves An eight-week nitrate exposure did not alter significantly the growth rate or body condition, as defined by the SMI, of males, females and juveniles. The caloric content of juveniles at 50 mg NO₃^-/l and in the controls at \<10 mg NO₃^-/l was markedly higher than at 250 mg NO₃^-/l. However, males significantly increased in caloric content at 50 mg NO₃^-/l compared to the other two concentrations. Females showed no significant differences amongst treatments. Nitrate also did not affect the δ¹³C and δ¹⁵N measures in the white muscle of all tested fish. ## Histopathology A detailed examination of all slides did not reveal overt clinical signs of disease, but some tissular changes were observed in liver and gills (and Figs). Occasional telangiectasia and slight epithelial lifting were observed in secondary lamellae. However, we did not observe other tissue alterations such as hyperplasia, hypertrophy or increased number of mucous cells. All liver samples had regularly aligned cords of hepatocytes. However, slight changes in the staining intensity of the cytoplasm were observed in liver tissue, probably related to glycogen deposits and occasional lipid droplets. The presence of macrophage aggregates was restricted to adult fish, with females having a larger number than males, but without significant changes due to nitrate. The amount of perivisceral adipose tissue in juveniles at the highest nitrate concentration was lower than in juveniles in the other treatments. ## Feeding behaviour Males, females and juveniles showed differences in latency, voracity and satiety, but without significant changes due to nitrate apart from males from the sixth week onwards. Males at 50 and 250 mg NO₃^-/l exhibited lower satiety values and lower voracity (i.e. \> time to capture prey) than those in the controls at \<10 mg NO₃^-/l. Overall, females and juveniles had a higher voracity and satiety than males, and juveniles had greater latency times than adults. Females and juveniles also tended to increase voracity and satiety and to reduce latency times throughout the experiment compared to males. ## Pair-wise correlations amongst biomarkers Spearman rank correlation coefficients were generally low amongst all biomarkers measured in juveniles, males and females of mosquitofish. There was a strong positive correlation between calorimetry and growth (ρ = 0.73, P = 0.002) and between calorimetry and the amount of perivisceral fat cells in juveniles (ρ = 0.72, P = 0.003). However, a marked negative correlation was found between satiety and latency time of males (ρ = -0.78, P \<0.001). A strong negative association was observed in females for d¹⁵N and growth (ρ = -0.74, P = 0.002), even though the relationship was positive between C/N and growth (ρ = 0.77, P \<0.001). # Discussion This is the first comprehensive study examining the chronic effects of nitrate on a widely introduced fish species, as exemplified by the eastern mosquitofish (*Gambusia holbrooki*). Moreover, this is one of the few ecotoxicological studies using short- and long-term biomarkers (e.g. growth, histopathology, feeding assays) in females, males and juveniles of the same species. Overall, we did not find overt clinical signs of disease, which supports the prevailing idea that many invasive species, including mosquitofish, have wide tolerance to changes in water quality. However, the fact that nitrate altered food intake or energetic reserves in males and juveniles suggests that concentrations \>50 mg NO₃^-/l cannot be considered completely safe. Many studies have shown that males are more likely to acquire diseases than females, including fish. We did not find gross pathological alterations in any fish, but the more marked effects of nitrate on males provide some support for males being the sicker sex. The weaker response of the tested fish to nitrate is unlikely to be much attributed to pre-acclimation to nitrate at the collection site (9.9 ± 3.0 mg NO₃^-/l), as reported for amphibians. However, this outcome does not exclude the possibility of nitrate tolerance being increased due to pre-exposure to other ions, including those of water hardness, which is high at the collection site and in the laboratory dechlorinated tap water due to a calcareous geology. The effects of metals and chlorine compounds on fish in the animal facility probably were negligible because tap water is filtered through active charcoal and the aquarium product Sera Aquatan is used to further guarantee water is free of metals and chlorine. The fact that juveniles had a greater tolerance to nitrate than males was unexpected because young fish are generally more sensitive to chronic pollution than adults. However, 96-h LC50 tests revealed that susceptibility to nitrate increases with body size in the Siberian sturgeon *Acipenser baeri*. Although we cannot reveal the mechanisms for the mild effects of nitrate on juveniles because nitrate metabolites in tissues were not measured (e.g. nitric oxide), differences in space and behaviour between adults and juveniles might partially explain outcomes. Adults were kept at lower densities (6 fish per tank) than juveniles (11 fish per tank) and not surprisingly, we observed more agonistic interactions among adults due to confinement. Mosquitofish populations are often confined in small water bodies and female- biased, including in the collection site of the studied fish (authors *pers*. *observ*.). The biased sex-ratio has been attributed to the high life-span of females compared to males. We built on this knowledge by showing that females may dominate in number because they are more tolerant than males to polluted waters, where mosquitofish often occur. Female mosquitofish had higher feeding rates than males regardless of the nitrate treatment, which is consistent with previous data in clean water. Given that higher nitrogen excretion rates have been reported in females, it is possible that tolerance to environmental nitrate can be predicted from nitrogen excretion rates in fish. Nonetheless, sensitivity to nitrate probably depends on many factors in wild fish, including temperature, predation, and the fact that a parasite with more severe effects on males than females is more sensitive to nitrate than the fish host. In contrast to mainstream literature in which external factors other than pollutants are often not included in toxicological assays, our study accounted for intraspecific interactions; that is, several individuals were exposed to nitrate in the same tank instead of fish being exposed individually. Agonistic interactions probably are amongst the most important factors to explain *G*. *holbrooki* performance alongside sex because females are more aggressive towards conspecifics than males.This might explain why more females died during the experiment than males, although mortality did not differ significantly amongst treatments. Feeding traits were more affected by nitrate than other biomarkers measured in male mosquitofish, which supports that food intake is amongst the most sensitive biomarkers in ecotoxicology. Although we cannot identify the mechanisms, it might be a response-mediated by stress hormones (e.g. cortisol) because high levels of these hormones often reduce appetite in fish and other animals. However, cortisol levels remained stable in females of Siberian sturgeon (*Acipenser baeri*) after 30-day exposure to 250 mg NO₃^-/l, as opposed to the reproductive hormones testosterone and estradiol. Moreover, there is correlative evidence for reduced sperm count in male mosquitofish at \< 22 mg NO₃^-/l. These studies illustrate that nitrate is an endocrine disruptor through *in-vivo* conversion to nitric oxide, which is involved in many metabolic pathways, suggesting that it is possible that the effects of nitrate on fish probably would have been stronger than observed if we had used biochemical biomarkers. Nevertheless, many biochemical alterations often do not have far reaching impacts on individuals, reason for which they are considered of less ecological relevance than behavioural assays, including the feeding traits we measured. Even though fish differed in food intake amongst nitrate treatments, we did not observe overt signs of disease, including reduced fish growth. Reduced food ingestion in males may be attributed to fatigue because nitrate forms methaemoglobin, which transports oxygen worse than haemoglobin. However, fish can cope with moderate methaemoglobinemia, especially in hard water, such as ours in the laboratory, which may have mitigated nitrate adverse effects. Growth was expected to decrease in mosquitofish because iodine uptake, which is needed for thyroid functions and animal development, is altered by nitrate, but concentrations up to 11 mg NO₃^-/l did not impair the thyroid function in perch (*Perca fluviatilis*) and Crucian carp (*Carassius carassius*). The neutral effect of nitrate we saw on mosquitofish growth agrees with Freitag et al., who found that concentrations up to 450 mg NO₃^-/l had no effect on the thyroid hormone levels in Atlantic salmon (*Salmo salar*). Our outcome is also consistent with studies in other freshwater taxa showing that nitrate effects on growth and survival occur at \> 500 mg NO₃^-/l. However, the neutral effect of nitrate on mosquitofish does not exclude the possibility that fish exposed to nitrate may reduce their ability to cope with other pollutants if nitrate alters the internal ionic composition of fish at an osmoregulation cost and probably impairs important enzymatic complexes, such as those involved in detoxification. Histopathological analyses revealed no relevant tissue alterations because slight epithelial lifting and other alterations we saw are not pathological but tissue processing artifacts. Changes in caloric content only matched with histological data for juveniles at 250 mg NO₃^-/l, which reduced energy reserves as also exemplified by peripheral fat content, but no major changes in δ¹³C and δ¹⁵N measures of muscle occurred in fish from any treatment combination. These findings suggest that no single tissue can be a good proxy of overall fish energy reserves because they vary greatly amongst tissues. Moreover, our findings confirmed that no biomarker, including mass-length relationship indices such as the SMI, can be assumed to accurately reflect ‘true condition’ without analysing body composition. The lack of response of the SMI may be attributed to the fact that nitrate did not markedly change fish weight, possibly because although fish reduced food intake, fish were fed daily until satiety. However, the biochemical composition of fish tissues might have changed due to nitrate because pollutants often alter tissue stoichiometry with potential far reaching impacts for fish predators. In this regard, juvenile mosquitofish altered energy content in tissues, but food intake or growth were not affected, which suggests that growth is prioritised over lipid storage, probably to reduce size-dependent predation mortality. In our study, energy costs are likely to be mostly attributed to intraspecific interactions and osmoregulation due to nitrate. Osmoregulation cost probably was caused mainly by the anion nitrate (NO₃^-, see) and, to a minor degree, by the cation sodium (Na⁺) of the salt (NaNO₃). The sodium concentration in the highest nitrate concentration was below 1 parts per thousand (ppt) and there is no experimental evidence for major changes in mosquitofish metabolism at 20 ppt or in mosquitofish plasma osmotic concentration at 10 ppt. Surprisingly, we found that the caloric content of males was higher at 50 mg NO₃^-/l than at 250 mg NO₃^-/l and in the controls at \<10 mg NO₃^-/l. Reduced caloric content in males at 250 mg NO₃^-/l compared to 50 mg NO₃^-/l can be due to osmoregulation cost increasing nitrate concentration. However, this rationale does not explain why males at 250 mg NO₃^-/l had a similar caloric content to those in the controls at \<10 mg NO₃^-/l, although the latter had the highest feeding rates in the study. Less energy stored implies that males at \<10 mg NO₃^-/l had an additional cost than those at 50 mg NO₃^-/l, which may be the courtship display. Control males were reproductive active and we observed copulation attempts with other males, a behaviour that often occurs in male poeciliids in the absence of females. Courtship display has an energetic cost, which probably was reduced at 50 mg NO₃^-/l because nitrate, even at lower concentrations, reduces testosterone and this hormone promotes the sexual characteristics of males. ## Conclusions Our study shows that females of the invasive fish *G*. *holbrooki* are more tolerant to nitrate pollution than males and juveniles, but that there are all weak effects combining short- and long-term biomarkers. Therefore, the ecological impact of this invasive fish seems not to be much affected by nitrate pollution, especially if populations are female-biased. However, our study cannot inform the indirect effects that nitrate may have on *G*. *holbrooki* through the alteration of aquatic food-webs, including a possible reduction in prey numbers accompanied by impaired food intake in males. There is the pressing need for an–omics screening (e.g. transcriptomics) to identify simultaneously all the metabolic pathways that are altered in fish exposed to nitrate in order to improve the mechanistic understanding of the effects of this widely distributed subsidy and pollutant in aquatic ecosystems. # Supporting information We thank the staff at the animal house facilities, especially Jordi Guinea. We also thank lab technician Cèlia Arcarons for histology sample processing, as well as Mercè Durfort and Jordi Correas for histology methods guidance. Bomb calorimetry analysis was done with the guidance of Xavier Remesar and Rosa M. Marimon. We also thank Pili Rubio for stable isotopes processing, Helena Satorras and Mònica Utjés for image editing and Mario Monroy for support in fish collection and lab procedures. We would also like to thank the two anonymous reviewers for their suggestions on this text. [^1]: The authors have declared that no competing interests exist.

# Introduction The recipe of the Harry Potter saga’s success might reside in part in the very unique way its author has installed a familiar kind of social network in a fantasy world. In order for the reader to be seduced by the story of any novel, the social network narrated in the book must not be too distinct from the ones typically found in real life. Interestingly enough, a novel respectful of the social reality of its time also constitutes an interesting temporal compression (many years of the life of many characters held in but one, or a few, books) that allows clever text mining and natural language algorithms to more easily catch the main features of the social network depicted in the novel: its topology (i.e. its degree distribution: is Harry Potter social network scale- free?), its clustering degree (are the friends of Harry friends themselves?) and the way it is being constructed in time (does the social network grow in a random way or does it follow some form of preferential attachment?). Not only is it possible at the end of a book (or the series) to extract and analyse the complete topology of the characters relationship, but it is as well possible to follow the way new characters enter this social network and connect to the existing nodes. Accordingly, a first investigation described in this paper is the study of the various social networks (both their static features and their dynamic nature) found in several popular series and one classic: The seven “Harry Potter” books by J.K. Rowling, the (presently) five “A Song of Ice and Fire” books by George R. R. Martin, the three “His Dark Materials” books by Philip Pullman, the (presently) three “Lunar Chronicles” books by Marissa Meyer, the six “The Mortal Instruments” books by Cassandra Clare, the three “Liveship Traders” and four “Rain Wild Chronicles” books by Robin Hobb, the fifteen “The Wheel of Time” books by Robert Jordan, and “Les Misérables” by Victor Hugo. Our approach starts with an automated construction of the social networks, based on the processing of dialogs in the text. The characters intervening in each conversations are identified, and a network is formed between them based on these interactions. For instance, if in a given part of the text Harry speaks to Hermione, a new link will connect Harry to Hermione in the evolving social network if there were none. Similarly, the first time Jean Valjean addresses Cosette, both the new node “Cosette” and its new link to “Valjean” will be added. The temporal evolution of the network is assumed to follow the succession of dialogs in the books, meaning that new characters appear in a network as soon as they are involved in a conversation. However, we do not believe such an assumption has a major impact on the presented results. The various steps leading to the creation of this network, as well as the finished social network, provide with a series of key features that form what appears to be a signature of each story, characterizing important elements relating to it, such as the scope, the number of protagonists, and even the author’s style and reading level (for instance, whether or not an author assumes that the reader can keep track of the speakers in a conversation, or the relative proportion of narration within a conversation). This series of attributes leads to a way to characterize each book, and even draw parallels between several of them. The rest of the article first summarizes the main assumptions and key decisions taken to automatically construct and follow in time the social networks out of the several books. Then, a more complete and sufficiently detailed technological description of the algorithmic steps will be presented in order for any interested reader to easily perform a very similar analysis of their favourite books or series. Finally, the results will be presented and discussed for the forty-seven best-selling books in terms of typical network measures (size, clustering and degree distribution) and the way these networks evolve in time (preferential attachment). # The main assumptions and algorithmic ideas First of all, the software technologies presented in this paper should be as automated and user-friendly as possible for anyone interested to easily extract the final social network and its time evolution out of their favourite book. All the software tools and the way to use them will be made available in the bibliography of the paper. The full extraction starts with a text version of the complete book or the complete series. We have arbitrarily restricted our analysis to forty-seven books, but nothing prevents to easily move ahead and enlarge our statistic samples to any size. The key assumption to construct social networks lies in the exploitation of dialogs and successions of dialogs to automate the extraction in time of the nodes and the links. It is far from surprising to base a social network analysis upon communication among the nodes, since the most common interactive mode between people remains verbal communication. Dialogs are also generally easy to detect and to separate; however, indirect speech will be ignored as a result, as it is syntactically impossible to distinguish from regular narration. An important algorithmic step of the full protocol is to first isolate one single dialog and insert it in a larger piece of text (including some surrounding narration) referred to as its context. The following key step is to identify the speaker in this context. Then this speaker is connected to the characters he speaks to, and possibly to some others not directly present but mentioned during this dialog. The connectivity pattern to associate with this dialog will be described next but could be: the speakers with all the others, or a fully connected sub-graph (all characters with all). Incorporated in the algorithm is the possibility to weight or colour the links (as can be seen in the network) as a result of a sentimental analysis performed on the basis of the words found in the context. In this paper, this additional description will be of no use as our focus is primarily on topology, while for a further characterization of the author’s style or a study extended to weighted graphs, this additional data could be very profitable. One last assumption is to follow the network construction in time as a succession of the contexts just described above. This is a simplification: the events in a novel are not always told in chronological order. A classical figure of style is the use of flash-backs such as, for instance, at the heart of the narration in “les Misérables”; similarly, temporal inconsistencies may occur with shifting points of view, as are employed in “A Song of Ice and Fire”. Nonetheless, our interest lies mainly in at what point in the story each new character is introduced, and who they are originally attached to when introduced, to verify that a preferential attachment occurs; with this objective and restrictive frame of analysis (the social network’s evolution is only considered over the course of the book’s story, ignoring all backstory and previously-established continuity), this first-order approximation is expected to hold. # Implementation The principle of social network extraction from literature may seem simple in theory but is a very complex task in practice that involves many disciplines such as Natural Language Processing, Named-entity recognition, Co-references resolution, Aliases association, etc. Different steps are required to tackle the task properly, they are reviewed in detail in the following sections but an introduction of the important notions is first required. An author tells a story by switching between descriptions of the events occurring during the story (i.e. the narration) and descriptions of the conversations happening between the different ***characters*** involved. Both are important as they provide information about the characters intervening in the action, but the identification and analysis of the conversations provide a more precise description of the way social links are built through the storytelling. Each ***conversation*** depicted by the author consists in a succession of ***dialogs*** usually indicated in the text by double quotes (See Figs and for examples). As in real life, when someone speaks (the ***speaker***), he addresses himself to one or more persons; ***the audience*** (See). One last notion must be introduced, the ***context*** in which a conversation takes place. As explained before, the author switches between conversation and narration all trough the storytelling and even during a conversation to provide the reader with more information about the context in which it takes place (i.e. the location, who is involved, who they are… See for a practical example). The analysis of the context is necessary to differentiate between the characters participating in a conversation from those who are just mentioned in it (e.g. “I saw Ronald this morning”: Ronald is not actually participating to the conversation but is mentioned nonetheless), and to identify the speakers even though their actual last name, first name or alias is not being used. The speaker is usually identifiable (subject identification inside a dialog by means of grammatical analysis. See section) as well as the characters to whom he talks by analysing the entire conversion with all the dialogs it contains. The challenge resides in the identification of all the names that refer to each character, and building a network dynamically using the information extracted from the conversations. To summarize, a novel is composed of many conversations that describe the social interactions between all the characters involved in the action at the specific point in time. Building a network by extracting those information is a complex succession of operations necessary to identify those characters and the way they relate to each other (i.e. who talks to whom). **Lexicon**: ***Character:*** Person that participates to the story. ***Speaker:*** Character that speaks to other characters involved. ***Audience:*** Characters that listen to a speaker. ***Dialog:*** Line of text occurring in a novel denoted by double quotes to indicate that someone is speaking. ***Conversation:*** Succession of dialogs representing a conversation between two or more characters. ***Context:*** Before, during as well as after a conversation, the author may tell us more about the context in which a conversation takes place. This is what we call the context. To reach this goal an algorithm composed of four consecutive steps was developed: Pre-processing, Extraction of dialogs and conversations, Characters identification and Network building. They are composed of multiple sub operations as depicted in that are detailed in the following sections. ## Pre-processing The text to analyse has to respect a specific format to be processable by the algorithm, formatting it properly is the goal of this step. The pre-processing involves manual intervention on the raw text for some aspects (i.e. Removal of headers, table of content,…) and the application of automated procedures to extract the necessary information from the novel. **Manual pre-processing** The manual step consists in the acquisition of a proper raw text file. The actual text to analyse may be found in different file formats (i.e. HTML, DOC, EPUB, PS, PDF,…) that need to be converted to TXT. (The software Calibre was used for this purpose). The books may also be formatted in different ways (i.e. Use of double quotes or simple quotes to indicate dialogs, use of multiple spacing or specific formatting to indicate the chapters,…) that are not appropriate for an automated processing to take place. At the end of this manual step, whatever the way chosen to apply those modifications, the text must answer only two requirements: *Use double quotes to denote dialogs (single quotes for apostrophes)*. *Indicate chapter or scene breaks using empty lines with all other empty lines removed*. **Automated pre-processing** The next step is fully automated. It takes the cleaned raw TXT file as input to compute various information detailed here: Split the text in sentences using the CLiPS Pattern tree parser with two modifications applied to the content. First, some punctuation marks are transformed into others that are recognizable by the tree parser function. Then, the multiple lines forming one dialog are grouped in a same entity (by forcing an even number of quotation marks in a single entity). A sentiment score is computed for each line of text using the built-in opinion mining tool of CLiPS. (Other libraries like NLTK with SentiWordNet or CoreNLP were considered, but CLiPS was retained as final solution for this implementation as it suited best the needs of this algorithm). Also, if a scene break or chapter break is found, it is indicated by a boolean value. Produce a parse tree using the CLiPS Tree parser for each sentence and do a division into chunks (i.e: Chunking consists in the division of a text in parts of word that are syntactically correlated. For examples, see in the section ‘CLiPS Tree Parsing’ and the section ‘Characters identification and alias resolution’). This will also allow the determination of the grammatical roles of the words composing the sentences. (i.e. Subject, verb, complement,…). Save the metadata computed on files for later use. To sum up, this part of the algorithm is concerned with the cleaning of the text as well as the preparation of numerous metadata used for the next steps of the program. Those metadata do include, for each line of text, the sentiment score, the parse tree, the identification of scene or chapter break and the index. (i.e. Each line of text is identified by its index, the number of the line inside the entire novel) ## Extraction of dialogs and conversations A novel tells us a story in which characters interact at different moments in time. While reading a book, we are able to differentiate the contexts that take place easily, as well as the different dialogs and members of the audience. This is much harder for a program to tackle without a perfect knowledge of the language used. This step tackles the task of solving this aspect of the problematic, it identifies the dialogs and conversations in each dialog and also extracts the context of a conversation. The algorithm proceeds as follows: **Evaluating dialog spacing:** This consists in a simple counting operation of the number of sentences present between two lines of dialog. To allow a proper division of the story into conversations using an automated process, the values of dialog spacing are analysed during the next step to compute a threshold indicating when one conversation must start and end. Drawing the distribution of those values yields histograms presented in Figs, and. A few general conclusions can be derived from these results. First, the overall shape of the distribution (zeroes excluded since they represent chapter or scene breaks) remains the same and shows a rapidly decreasing profile. Second, each author appears to have a different typical profile as it was observed during the analysis of other series like “Game of Thrones” (\[–\]). This dialog spacing is a distinctive mark of the author’s style. Indeed some authors are more prone to write long sequences of narration, whereas others tend to do the opposite, showing respectively higher or lower value of dialog spacing. Third, the author is more important to the profile than the length of the novel is, even though we reach higher values of spacing in longer novels regardless of the author. **Conversation size thresholding:** Using the values of dialog spacing, a threshold is computed to evaluate the maximum distance between two lines of dialog belonging to the same conversation. Unsurprisingly, the frequency distribution heavily favours small numbers since lines of dialog tend to be clustered in conversations. Such observation indicates that it is possible to extract most conversations automatically by relying on a threshold value of dialog spacing. This threshold corresponds to the usual number of spacing (i.e. Sentences of story telling) found between two dialogs. After the analysis of a few books, an empirical formula was derived. Namely, the threshold is the highest possible value of spacing, such that its frequency is higher than both 10 and double the frequency of the spacing one unit above, and such that all lower values of spacing have higher frequencies. However, if there is a higher value of spacing with a frequency above 100, this value is used as a threshold instead. This is summarized in the formula that follows: $$\begin{array}{ccl} {threshold} & = & {\max\left\{ s \middle| s \in spacing\mspace{600mu} \land \mspace{600mu} \right.} \\ & & {\left\lbrack \mspace{600mu}\left( \mspace{600mu} frequency(s)\mspace{600mu} \geq \mspace{600mu} 10\mspace{600mu} \right)\mspace{600mu} \land \right.} \\ & & {\left( \mspace{600mu} frequency(s)\mspace{600mu} \geq \mspace{600mu}\left( 2\mspace{600mu}*\mspace{600mu} frequency(s + 1) \right)\mspace{600mu} \right)\mspace{600mu} \land} \\ & & {\left( \forall\mspace{600mu} t\mspace{600mu} \in \mspace{600mu} spacing\mspace{600mu}:frequency(t)\mspace{600mu} > \mspace{600mu} frequency(s) \right)\left. \mspace{600mu} \right\rbrack} \\ & & { \vee frequency(s)\left. \mspace{600mu} > \mspace{600mu} 100 \right\}} \\ \end{array}$$ For “Harry Potter and the Philosopher’s Stone”, the computed value is 8 (i.e. If more than eight lines do separate two dialogs, they do belong to two different conversations), as shown on. It is then used to group the dialogs and to single out the successive conversations appearing in the novel. **Context generation:** To improve the process of speaker identification coming later on during the process, the conversation is being extended to a context. (i.e. The named entities found inside the corresponding context are used to enrich the information already extracted from the dialogs forming a conversation) It is built by listing every sentence from the end of the previous conversation up to the start of the next one. (See) **Dialog metadata extraction:** For each dialog found inside the novel some metadata are computed and grouped to facilitate further processing. The result is a map of every dialog associated with the index of the dialog (i.e. the index of the corresponding line inside the novel, indicating the time at which the dialog occurs) as well as two fields, ‘from’ and ‘to’, containing the names identified and corresponding respectively to the speaker and the audience. Here they are explained with an example in. **Metadata**: **Index of the sentence:** Position of the line of dialog inside the entire text. **Identified speakers (‘from’):** Character that speaks at that moment. **List of the identified characters within the dialog (‘to’):** Characters mentioned in the dialog. This list contains all the NNPs (proper nouns) identified in it. **Index of the context to which that dialog belongs:** Corresponding to the number of the context (i.e. enriched conversation) in which the dialog falls after the conversation thresholding. At the end of this step, dialogs are grouped into conversations, each of which has an associated context, and each dialog is enriched with metadata that serves as the basis to build a social network. ## Characters identification and alias resolution Another problem occurring while reading a novel is the speaker and characters identification. “Who talks to whom?” illustrates the problem perfectly. Once again, this task is easy for a human but may be more difficult for a program depending on the writing style of the author (Namely, whether the book is intended for a young or an adult readership). Identifying the characters involved in the action and more specifically distinguishing the speaker from the audience is another difficult problem to solve. This part of the algorithm runs in two steps and relies on the metadata previously computed (i.e. the parse tree and chunks already provide the information about the NNPs that are looked for) (See). **First**, the program takes each dialog of the novel and checks its grammatical structure computed while pre-processing the novel. During the metadata extraction in the previous step, the program already checked for the presence of dialog tags (bits of narration interspersed within the dialog to indicate a speaker, e.g. “he said”) and extracted a speaker if the subject of that dialog tag was a proper noun (NNP tag), or a succession of proper nouns and stores it as the “from” field of the dialog occurrence. Otherwise, the speaker is not yet clearly identified, and that field is blank. (See for an example and section for more information). **Second**, the problem of aliases is handled. Almost every characters possesses one or more aliases used through the story telling by the author at different moments (i.e. name, diminutive, title, full name,…) This step handles this problem by building an alias network which provides an association of the different names with their aliases and replaces their occurrences by the same name; the key of the character. Doing so, instead of risking to create a network full of characters different in appearance but actually identical, the algorithm is able to reduce the size of the resulting network by properly identifying the characters. Two main flaws arise from the use of the alias table however: the inability to identify characters from other proper nouns, and the handling of NNP tags that are applied to several characters, such as titles and family names. “ The part- of-speech tags are based on the Brown Corpus reduced in size to fit the orientation of the Penn Treebank used. It eliminates redundancy by taking into account lexical as well as syntactic information. Most NLP tools use those. ” The former problem derives from the use of the Pen Treebank) tag set as the primary identifier of character names. Indeed, a variety of proper nouns are tagged as NNPs (Example: Gryffindor) with no regard to what they refer to. Because of this, we are unable to differentiate capitalized neologisms or named locations from character names. The second error comes from titles and family names when several characters share them. It is impossible to fully solve those aliases, but a workaround (the alias table) was implemented to prevent those names from being referred to as mush as possible (e.g. “Professor McGonagall” would be included in the entry “McGonagall” but not professor, but if the author refers to her used only her title then the node “Professor” would still appear). It should also be noted that the algorithm is incapable of differentiating between two characters who would have the same name. This could be corrected by employing a disambiguation algorithm; however, that only results to minor errors: characters in a fictional narrative, especially in writing, are rarely named the same way, as that could cause confusion even for a normal reader. This is consistent with other research in stylistic analysis over works of fiction, such as Argamon et al.. It should be noted that a disambiguation algorithm may help the second error mentioned above; this may thus be the subject of future research. **Speaker identification**. During the metadata extraction step, speakers are identified locally for each dialog (i.e. within the dialog itself), using dialog tags as announced. This means that the resulting speaker identification rate is highly dependent on the author’s style; namely, whether or not they frequently remind the reader of who is speaking, and whether or not they use the character’s name or pronouns to do so. The resulting rate of speaker identification (SIR), averaged over all analysed books, is close to 50%, but with a high variance (standard deviation of about 10%). A breakdown by author, however, shows the expected result: the average rate of identification is very different depending on the writer. (See) This leads to an acceptable rate of speakers identification. To improve the results obtained, some assumptions are made. For example, in a conversation where only two characters were identified as speakers, the algorithm infers that the characters speak in turn, and fits that pattern to the dialogs where no speaker has been identified yet as best as possible. The same happens when more than two characters are involved in a conversation: the algorithm takes into account the speakers identified for the neighboring dialogs to propose the best possible inferences. Doing so, the average rate of speaker identification reaches over 98%. One question arises though: what about the accuracy of this identification? At this stage of development the accuracy cannot be measured for a simple reason: the correct speakers should be identified manually to assess the precision of the classifier. As an example, in *Harry Potter and the Philosopher’s Stone*, the correct speaker was identified for 56.9% of individual occurrences. However, the bulk of the analysis does not rely on individual occurrences, but on networks built from entire conversations. Within each conversation, all the identified speakers are interconnected; thus, as long as a speaker was identified within a conversation, it matters little whether or not it was assigned the correct specific lines of dialog, as it will result in the same network. This is measured by comparing the list of speakers appearing in each context of the extracted network to the same list in each context of a manually- generated correct graph, and counting how many are present in both lists. For the example of *Philosopher’s Stone*, the proportion of speakers that were correctly found goes up to 82.4% when performing this bulk analysis. To get a better sense of what this represents, a differential between the extracted graph and the correct graph is shown in. 73.3% of the edges from the extracted are present in the correct graphs. The discrepancy can be explained by some of the assumptions made, especially when assuming a strict order in which characters speak within a conversation. This means that any error propagates quickly within a conversation, for instance by “flipping” the speakers, whereas the characters identified for that conversation might remain correct. To synthesize, this step is the most important. It tries to identify all the characters as well as their aliases and allows the identification of a speaker for a majority of the dialogs encountered. However, the interest of the speaker identification may seem useless at first glance. Indeed, building a social network should lead in a first approximation to an undirected network in which all the characters that do present social links are linked together. The speaker identification allows the algorithm to go further by giving a direction to each link, providing for a directed, dynamic network with an edge weight corresponding, for example, to the sentiment transmitted by the speaker. This aspect goes however out of the bounds of this paper and will be the object of later analysis. Here, the capabilities of the algorithm are just demonstrated. ## Network building The final step of the process consists in the creation of the network. It is the simplest part in principle, but the potential outputs are numerous. Do we want a simple social network (undirected, unweighted, atemporal), do we want the network to be extracted from only one chapter, one context, one book or even an entire series? Do we want it to be dynamic? With sentiment scores? Do we want the scores obtained to be adapted over time or not? Many other networks can also be extracted such as the alias networks, dialog networks… However, for this paper, only social networks are used. Two types of social networks are extracted during the processing, each giving a different characteristic: the “Context Networks” and the “Incremental Networks”. The **“Context Networks”** are built using only the dialogs corresponding to a single conversation, for which the network is built. The resulting graph is undirected, since the interactions are considered to be reciprocal between two characters. All characters appearing in the context are interconnected. This is not mandatory: we could identify the speaker and the audience for each dialog and build a directed graph based on that information. Those networks can be used to analyse a precise sequence of storytelling, but are not the ones on which the following results are based; they will be the subject of further research. An example is given in for the context number 64 in the first book of Harry Potter. The **“Incremental Networks”**, on the other hand, are built iteratively from the context networks by merging all the context network until some point in the story; in other words, by taking all context networks from the first conversation to a given conversation, and merging them together. An example is given in with 498 contexts considered on the left side ( left), roughly corresponding to a specific period of the novel in which only the family members of the main character, Harry, are interacting, as well as a few professors; and the global network on the right side ( right), built using the 1670 dialogs available. Every network is exported to the CSV format and drawn as a PNG image, while the final incremental graph, representing the social network at the end of the story, is written to the additional formats GEXF and JSON. The possibilities of outputs are obviously numerous and will undergo additional research in the future. shows an example of such a final graph from the Harry Potter series. In summary, the algorithmic steps just described are able to generate many different types of networks all answering to specific questions. In this paper, we have decided to restrict our analysis on one main aspect: the topology and the time evolution of the characters social network (preferential attachment or not) as well as how they do compare with their real counterparts. # Results As outcome of the execution of the algorithm many results can be obtained. For the purpose of this article, mainly one example resulting from the social network extraction on “Harry Potter” is examined more deeply, even though the other books and series were analyzed. ## Degree distribution The degree distributions for each book (See for the distributions of the book from the Harry Potter series) show similar profile and appear similar to the typical distribution of a scale-free network. It should be noted that, while there is a discrepancy with an ideal power law, it is fairly small (as indicated by the *R*² of the power function fitted to each distribuion) and mostly due to the small size of the networks, meaning that low frequencies for the highest degrees cannot effectively be reached. (See) The exponent of the power approximation falls in the \[−1, −2\] range, which is outside of the \[−2, −3\] predicted by theory or shown by large, scale-free networks, such as the Web graph. However, social networks do show exponents falling in that range (\[–\]). That the networks extracted from novels fall into the lower end of that range can be attributed to two factors that contribute to increasing the probabilities of higher degrees: Conversations tend to involve several people at once, either due to several people participating in a conversation or because another character is mentioned during the exchange. The specificities of fiction, in particular the so-called “Law of Conservation of Detail”, which dictates that any element mentioned in a work of fiction should be relevant (and thus be brought back up again). These factors, added to the networks’ small size, contribute to increasing the real frequencies for higher degrees. This is true, even when considering a series as a single entity, as shown on. ## Clustering coefficient The clustering coefficients of the incremental networks produced were also computed (See Figs). It shows a clear common trend for all the books: the clustering coefficient starts at a value which is heavily dependent on the first few contexts occurring in the book to quickly stabilize at a value close to its final one. The values then converges slowly. To assess the significance of the clustering coefficient in a rigorous way, the best point of comparison to use is a similar random graph, whose clustering coefficient is given by *C*_*rand* = \< *k* \> /*N*. The same values are computed for the extracted social networks, i.e. their **average clustering coefficient**. The coefficient of a random graph is the proportion of two neighbouring nodes that are connected, equivalent to the probability that two random nodes are connected. The comparison is presented in the table from. The order of magnitude of the coefficient is greater for all the books considered, the same type of results that characterizes scale-free networks, as shown for instance in: one of the common characteristic of social networks is a high ratio of their clustering coefficient to that of a random graph with similar properties (average degree and size). Going further in the analysis, the clustering coefficient can be explained with more detail by studying the type of narration of each book. For instance, a book that focuses on a single character or group of characters will present a higher clustering than stories focused on several characters. Similarly, extending the scope of the story across time or space will have a tendency to reduce the resulting social network’s clustering. For instance, compare two very different series: *Harry Potter* and *A Song of Ice and Fire*. The former, focused on a single protagonist, features a much higher clustering coefficient than the latter, in which the story follows several protagonist, often very separated in-universe. Another example is within a single series itself, *The Wheel of Time*: the clustering coefficient changes greatly from books where the protagonists are few and/or gathered for a single event (“New Spring”, “The Eye of the World”, “Winter’s Heart”) to when the characters are very separated (“The Dragon Reborn”, “The Gathering Storm”). (See ) ## Preferential attachment Preferential attachment has also been measured and evaluated. The plots in Figs and represents the probability that a new node be attached to a node of a given degree, with those degrees being taken a posteriori from the global graph. Those profiles are tested by comparing them to a fitted linear function. From those graphs, we can observe that books with a single protagonist have a much higher tendency towards preferential attachment. Indeed, we can verify that in other series like “His Dark Materials”, the *R*² values goes down tremendously due to the introduction of a new protagonist in every new book, where the value of *R*² remains similar in the series of “Harry Potter” as well as in the other books. ## Clustering of the books During the extraction of each book’s social network, a number of signature features were identified: the conversation size threshold, the speaker identification rate, the degree distribution’s exponent, and the average clustering coefficient in the final network. Two methods of clustering were implemented to test these four features: hierarchical clustering, using the Ward variance minimization algorithm, on, and a K-means classification, shown in. The dendrogram shows a good clustering of certain series: “Harry Potter”, “The Rain Wild Chronicles”, “The Lunar Chronicles” and “The Mortal Instruments”, for instance, appear to be clustered. Other series are more scattered, which can be attributed to a fluctuation of style: for instance, “The Wheel of Time” is separated in three sub-clusters, gathering similar books within the series together. The K-means classification yields better results: although some series are still split across several clusters, they are much less numerous than before. With the clustering computed in, any pair of book is either in the same cluster and part of the same series, or part of different series and in different clusters, with a probability of 83.35%. These are preliminary results for a proof of concept, and can be improved further. We cannot yet suggest a systematic way of pruning the dendrograms to obtain optimal clusters using this method. The number of clusters on the K-means analysis was assumed to be the number of distinct authors (10, counting that some books from “The Wheel of Time” were written by a different author). The K-means clustering was tested with other numbers of clusters (from 2 to 15 clusters), and that value did provide an optimum in the separation of the books. This is a weakness of the K-means algorithm, since it assumes that the number of clusters is known in advance # Conclusion and future work This paper attempted to replicate the type of network analysis that sociologists perform in their daily studies. Is it important for literature to be respectful of social realities, or are the authors of successful novels imaginative enough in imposing in the same social topology of their history the same fantasy as they create in the fictional worlds in which their characters act? We proposed a novel algorithm capable of generating multiple types of networks (directed, undirected, weighted, weighted using a sentiment analysis, dynamic) built from novels. Doing so, social networks, alias network, conversation networks now can be generated and analysed. A first social network analysis was also proposed and indicates similarities in the way the social network is presented in novels compared to real life social networks with explanation concerning the differences that might be observed (observation of a power-law degree distribution, high clustering coefficient and an a posteriori preferential attachment verification). Finally, some features were identified that might be relevant in the identification of the author’s style like the dialog spacing and alias network topology. Moreover, this network analysis provided with a number of characteristics related to the book’s story and the author’s style: The dialog spacing frequency (and the resulting threshold value), indicating the relative presence of narration within conversations The speaker identification rate, specifically after the metadata extraction step, indicating how much the author reminds their readers of the speakers’ identity The degree distribution in the final network The average clustering coefficient in the final network, indicating the scope (in time, space and number of protagonists) as well as its relation to the clustering coefficient of a similar random network and its temporal evolution Once further refined, these features will be combined with other method of stylistic analysis, such as the frequency of function words, and the stylistic genome (“stylome”) introduced by van Halteren et al.. As the algorithm is still under development, there are many different ways to improve it and other further analyses are yet to be made considering the vast number of networks that can be generated. A few tasks seem to be most interesting to tackle quickly though. Those concern the formal verification of the character’s identification process as well as the alias table accuracy, solving the coreferences to avoid relying too much on inferences based on the ordering of the dialogs, as well as a sentiment analysis of the dynamic network produced. With those objectives as a priority, we hope to refine the algorithm further and apply it to other novels as well as other media and try to verify that those novels are built following the same schema of social network. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MCW TN. Performed the experiments: MCW TN. Analyzed the data: MCW TN. Contributed reagents/materials/analysis tools: MCW TN. Wrote the paper: MCW TN HB. [^3]: Current address: ULB, CoDE-IRIDIA, 50, Av. F. Roosevelt, CP 194/6, B-1050 Brussels, Belgium

# Introduction Since its first association with antibiotic-associated disease (AAD) in 1977, *Clostridium difficile* has been recognised as the most commonly identified cause of nosocomial diarrhoea world-wide. *C. difficile* infection (CDI) typically occurs following antibiotic therapy, in which disruption of the resident gut flora leaves the intestine susceptible to *C. difficile* outgrowth. CDI can result in a range of clinical sequelae; asymptomatic carriage to severe diarrhoea, pseudomembranous colitis and death. In the U.K. cases of CDI peaked in 2007 at 57,255 cases but, more recently CDI rates have declined to 21,682 in 2011, presumably due to the modification of antibiotic use and the implementation of improved containment barrier protocols \]. In the US, *C. difficile* is associated with over 14,000 deaths annually \[[www.cdc.gov/HAI/organisms/cdiff](http://www.cdc.gov/HAI/organisms/cdiff) updated 2010\]. The total identifiable cost of CDI was estimated to be £4000 per case in England in 1996. On this basis such infections conservatively cost the U.K. £87 million in 2011. Disease has largely been associated with the production of two large toxins, toxin A (TcdA) and toxin B (TcdB), which are encoded by *tcdA* and *tcdB* (respectively) along with three other genes on the pathogenicity locus (PaLoc). Variations within this locus are recognised by a typing scheme, which recognises at present 31 different toxinotypes. These include the toxinotype III group to which Ribotype 027 isolates have been associated with global CDI increase. These strains typified by the NAP01/Ribotype 027 isolates (A+B+CDT+) were responsible for 41.3% of all CDI cases between 2007–2008 in the U.K.. Ribotype 027 isolates have been associated with increased disease severity (such as toxic megacolon) and relapse rates. Another toxinotype that has also been associated with a global increase of CDI outbreaks, especially in Asia, are the A−B+ toxinotypes (types VIII, X, XVI, XVII, XXX & XXXI). In 2008 the proportion of A−B+ isolates recovered from Korean CDI cases was 25.7% compared to 4.2% of isolates recovered in 1995. After initially being thought as non-pathogenic it is now known that A−B+ toxinotypes can cause a wide spectrum of disease including pseudomembranous colitis and mortality. Toxin A−B+ isolates have typically been typed as Ribotype 017, however, in recent years some A−B+ strains isolated in China & Australia appear as distinctly separate ribotypes. Along with toxigenic strains, naturally occurring non-toxigenic (A−B−) *C. difficile* are associated with asymptomatic carriage in both adults and infants. In these strains the PaLoc region encoding the toxins is replaced by a short sequence of 115 bp. Non-toxigenic carriage rates depend on age with several reports highlighting infants (≤2 years-old) and the elderly as sources of community *C. difficile*, with carriage rates as high as 26.5% and 2.7% , respectively. Sporulation, which is the transformation of vegetative cells to metabolically dormant endospores, is an important process for *C. difficile* transmission. As *C. difficile* spores are shed in the faeces of CDI patients, any surface or device contaminated with faeces can act as a reservoir for infection. *C. difficile* spores are highly resistant to several hospital cleaning agents, desiccation, pH extremes and high temperatures. Deakin *et al.*, elegantly showed that sporulation is essential for a persistent and relapsing disease and that transmission can occur through environmental contamination, direct contact and airborne transmission. *C. difficile* strains that exhibit increased disease severity and relapse are typically associated with increased sporulation rates compared to non-epidemic strains and historical strains. Within this study, we investigated clinical strains that naturally express different combinations of the toxins in the golden Syrian hamster model of disease. This model mirrors many clinical aspects of human CDI, as hamsters pre- treated with clindamycin results in haemorrhagic caecitis, which manifests as ‘wet tail’ and eventually death. In this manuscript we have used this model to characterise the disease outcomes of three clinical *C. difficile* strains; BI-7 – toxinotype III (A+B+CDT+), isolated from an epidemic outbreak in Portland, U.S. 2003, M68 - toxinotype VIII (A−B+CDT−), isolated during a *C. difficile* outbreak in Dublin, Ireland 2004 & CD1342 –PaLoc negative (A−B−CDT−), isolated from an asymptomatic paediatric patient in Oxford, U.K. 2009. # Materials and Methods ## Bacterial strains and spore preparation *C. difficile* BI-7 was donated by Dr Trevor Lawley (Wellcome Sanger Institute, Cambridge, U.K.); M68 was obtained from Dr Richard Stabler (London School of Hygiene and Tropical Medicine, London, U.K.) and CD1342 was obtained from Dr Kate Dingle (Oxford University, Oxford, U.K.). BI-7 is Ribotype 027, toxinotype III (A+B+CDT+), M68 is Ribotype 017, toxinotype VIII (A−B+CDT−) and CD1342 is Ribotype 005, toxinotype PaLoc negative (A−B−CDT−). Strains were grown on CCEY agar supplemented with cefoxitin-cycloserine, egg emulsion (Lab M, Lancaster, U.K.) and erythromycin (100 mg/L; M68 & CD1342) or clindamycin (20 mg/L; BI-7) at 37°C under anaerobic conditions. Animal inoculation spores were made according to Buckley *et al.* and were enumerated, to calculate the inoculating dose, by 10-fold serial dilutions plated onto supplemented CCEY agar plates as above. ## Minimum inhibitory concentration (MIC) The MIC of *C. difficile* BI-7, M68 and CD1342 to erythromycin and clindamycin was determined by the broth doubling dilution method as described by Andrews. Briefly, rows of pre-conditioned BHI broth (90 µl) were supplemented with a concentration range of 1024–0.5 µg/ml of either antibiotic. Wells were inoculated with ∼5000 spores/100 µl (as determined from spore preparations above) and incubated at 37°C for 48 h anaerobically. A no antibiotic positive control row was setup alongside an uninoculated sterility control row. After incubation, plates were visually inspected and compared to the controls; the MIC end point was determined as the lowest concentration of antibiotic at which there is no visible growth. Results are given as the median MIC from at least three assays. ## Ethics statement All procedures were strictly conducted according to the requirements of the Animals (Scientific Procedures) Act 1986 approved by the Home Office, U.K. (project licence 60/4218). ## Animal experiments All hamster procedures, including telemetry chip insertion, clindamycin dosing and *C. difficile* challenge, were done following to Buckley *et al.*. For survival assays, animals were monitored for any signs of morbidity, including ‘wet tail’. Animals were culled if core body temperature fell below 35°C (previously shown to be a relevant and effective endpoint). *C. difficile* faecal shedding was monitored in those animals that survived challenge. When an animal succumbed to infection, to establish level of colonisation, the caecum and colon were removed aseptically at *post-mortem*. To enumerate the total bacterial load (spores and vegetative cells), each section was opened longitudinally, and the contents were removed by gentle washing in 10 ml PBS (luminal associated bacteria). The tissues washed in 10 ml PBS and homogenized in 5 ml of PBS for 1 min (tissue associated bacteria), and viable counts were determined for the homogenates. Serial 10-fold dilutions were plated on supplemented CCEY agar plates. To determine the numbers of spores present in the samples, the samples were heated for 15 min at 56°C, and the numbers of spores present were determined by the viable count method as described above. To monitor the bacterial recoveries as the infection progressed, at least 5 animals were each culled at 1-, 3- & 11-days post challenge (d.p.c), where the caecum and colon were excised and processed as above. Results are shown as mean number of recovered bacteria from 2 biological replicates, where a total of at least 5 animals were included per time point. Colonies were confirmed using multilocus variable-number tandem-repeat analysis (MVLA). ## Detection of *in vivo* toxin levels Production of *C. difficile* toxins were detected *in vitro* using filtered caecum content from animals taken during *post mortem* as described by Buckley *et al.*. Briefly, monolayers of Vero cells (kidney epithelial cells) were washed with preheated sterile PBS before addition of serial diluted filtered gut content in supplemented EMEM and incubated for 18 h at 37°C (5% CO₂). Cells were washed with PBS, fixed in 1% formalin for 10 min then washed again. Fixed adherent cells were stained with Geimsa for 30 min then washed before addition of 0.1% SDS and left for 1 h. Optical density was taken using an EL_x808 Ultra microplate reader (Bio-Tek instruments) at 600 nm and compared to non-infected hamster caecal and colon gut contents as a negative control. If the toxin dilution was able to cause cell toxicity (cell rounding) this leads to the loss of cell adherence resulting in a reduced staining and hence optical density of the wells. Results are expressed as LOG reciprocal titre. ## Histology Caecum samples were prepared for simple histology as described by Goulding *et al*. Caecal pathology was scored in a blinded fashion, grading neutrophil margination (0, no neutrophil accumulation; 1, local acute neutrophil accumulation; 2, extensive submucosal neutrophil accumulation; 3, transmural neutrophilic infiltrate), haemorrhagic congestion (0, normal tissue; 1, engorged mucosal capillaries; 2, submucosal congestion with unclotted blood; 3, transmural congestion with unclotted blood), hyperplasia (0, no epithelial hyperplasia; 1, twofold increase in thickness; 2, threefold increase in thickness; 3, fourfold or greater increase in thickness), and percent of epithelial barrier involvement (0, no damage; 1, less than 10% of mucosal barrier involved; 2, less than 50% of mucosal barrier involved; 3, more than 50% mucosal barrier involved). Results are expressed as mean pathology score per strain for each criterion. ## Statistical analysis All statistical analyses were performed using the GraphPad Instat 3.10 (GraphPad Instat Software). A Mann-Whitney analysis of variance analysis (ANOVA) was used to determine significant difference in bacterial recoveries between all time points examined. *P* values ≤0.05 were considered significant. # Results ## Antimicrobial susceptibility *C. difficile* M68 and CD1342 were highly resistant to both erythromycin and clindamycin, MIC\>1024 µg/ml & \>256 µg/ml, respectively. Strain BI-7 showed resistance to clindamycin (MIC = 64 µg/ml) but was highly susceptible to erythromycin (MIC = 0.125 µg/ml). Using this data we determined that a one-day clindamycin hamster treatment model could be used without any diminutive effects on the initial inoculum for each strain. ## Telemetry monitoring and survival of infected hamsters Challenge of clindamycin pre-treated hamsters with 10,000 spores of *C. difficile* BI-7, M68 or CD1342 resulted in a 100% colonisation rate. Hamsters challenged with the toxin negative (A−B−) strain CD1342 resulted in a 100% survival rate with no signs of morbidity. In contrast, when challenged with M68 (A−B+) 55% (6/11) of hamsters survived, however all animals showed classical symptoms of *C. difficile* infection, i.e. ‘wet tail’. Those animals that succumbed to infection with M68 had a mean time to cull of 82.55 h±16.9 h SEM, however a wide range between 43.7 h and 127.0 h was observed (n = 5). An extended period of high body temperature was associated with those hamsters that survived M68 infection, e.g. a peak body temperature of 39.3°C was observed before the onset of symptoms in the surviving hamster profile shown in. In comparison hamsters that succumbed to M68 challenge showed slightly elevated body temperatures before the rapid temperature drop associated with CDI in hamsters. Animals challenged with BI-7 resulted in a 100% fatal infection with a mean time to cull of 26.3 h±0.75 h SEM. When animals succumbed to infection a rapid temperature drop was observed similar to other ribotype 027 isolates. ## Faecal shedding of CD1342 & M68 Shedding profiles of either CD1342 or M68 were obtained from the faeces of animals that survived *C. difficile* challenge. High numbers of CD1342 were recovered from the faeces 2-d.p.c. (c. 1.8×10⁶ CFU/g⁻¹ faeces) followed by a slow decrease with small numbers of spores recovered at 11-d.p.c. (c. 2.2×10³ CFU/g⁻¹ faeces). Animals challenged with M68 also peaked 2-d.p.c. although recoveries were at least 2 LOG lower initially (c. 7.2×10⁴ CFU/g⁻¹ faeces) compared to CD1342. ## Colonisation kinetics non-toxic of CD1342 To assess the colonisation kinetics of *C. difficile* CD1342 after inoculation, hamsters were culled at 1-, 3- & 11-d.p.c. to quantify bacterial levels in the caecum (CAE) and colon (COL). One-day after challenge with 10⁴ spores/animal, total *C. difficile* caecum levels reached approximately 5.7×10⁷ CFU/organ with higher bacterial levels in the lumen (-LA) compared to those bacteria more intimately associated with the tissue (-TA) (5.2×10⁷ & 5.0×10⁶ CFU/organ, CAE-LA & CAE-TA, respectively). Levels recovered from the colon were similar compared to the caecum. At 1- & 3-d.p.c. the percentage of spores present was high representing 54 (CAE-LA), 39 (CAE-TA), 83 (COL-LA) & 80% (COL-TA) of total bacteria isolated, respectively in both tissues, whilst at the experimental end-point (11-d.p.c.) the total number of bacteria isolated generally decreased by ∼1 LOG CFU/organ across all organ sites. ## Colonisation kinetics of M68 To measure the colonisation kinetics of *C. difficile* M68 in the caecum and colon, hamsters were culled at 1-, 3- & 11-d.p.c. and additionally if the animals succumbed to infection. At 1-d.p.c., caecum levels reached approximately 7.3×10⁷ CFU/organ whilst levels in the colon were slightly less, 8.8×10⁶ CFU/organ. Again bacterial recoveries were higher in the lumen than those more intimately associated with the tissues. By 3-d.p.c. the pattern of colonisation remained similar to that seen on day 1, although the levels of spores recovered from most tissues increased. The recoverable bacterial levels of those animals that succumbed to infection (45%; ∼83 h post challenge) were high across all tissues sites sampled, although showing no significant difference to those animals culled at a similar time point (3-d.p.c.), which showed no clinical symptoms. At experimental end-point (11-d.p.c.), surviving hamsters had high *C. difficile* levels in all tissue sites however there was a c.a. ∼1 LOG reduction especially in the organisms more intimately associated with the tissues. ## Colonisation kinetics of BI-7 Hamsters challenged with BI-7 resulted in a rapid fatal infection after ∼26 h. Total bacteria recovered at this time were significantly lower by at least 1 LOG than either CD1342 & M68 at the same time point (1-d.p.c.) with caecal recoveries highest compared to colon (6.9×10⁶, 3.1×10⁵, 5.6×10⁵ & 2.5×10⁴ CFU/organ; CAE-LA, CAE-TA, COL-LA & COL- TA, respectively; *p*≤0.0061 for all tissues vs both CD1342 & M68). Levels of spores recovered were significantly lower compared to those seen with CD1342 at 1-d.p.c. (*p* = 0.0003 all tissues). ## *In vivo* toxin levels *In vivo* toxin activity was measured semi-quantitatively *in vitro* using a Vero cell toxicity assay by serially diluting filtered caecal luminal contents, and calculating the maximum fold-dilution at which cell toxicity was still detected (cell rounding). At each time point tested, little cell rounding activity was detected in the caecal filtrate from animals challenged with CD1342. Challenge with M68 resulted in a wide spread of caecal toxin measurements with peak toxin detected at 3-d.p.c.. Hamsters that succumbed to infection with M68, ∼3-d.p.c., were associated with increased caecal toxin levels however these data were not significantly different compared to those hamsters which were culled at day 3 (*p* = 0.0568). The difficulty with this data is it is not possible to differentiate between those animals culled at 3-d.p.c. that may have subsequently succumbed to infection compared to those that would have survived. In comparison to animals that succumbed to challenge with M68, animals infected with BI-7 and culled at ∼26 h showed significantly higher levels of toxin activity (∼LOG 3.8 vs 5.4; *p* = 0.0215). ## Histological changes Considering the lack of clinical symptoms, histological analysis of the caeca from hamsters challenged with CD1342 and sacrificed 1-d.p.c. showed surprising changes; whilst the epithelial layer seemed intact unclotted red blood cells were associated within the villus structure. Accompanying this haemorrhagic congestion was an increase in circulating submucosal neutrophil cells. Similar caecal pathology was also observed in all hamsters infected with CD1342 and culled at 3-d.p.c. By 11-d.p.c. hamsters challenged with CD1342 showed no caecal pathology, tissue was similar to uninfected hamsters. Similarly, animals challenged with M68 and culled 24 h later showed a modest increase in circulating neutrophils and increased numbers of red blood cells within the capillaries. Whilst animals culled at 3-d.p.c. showed typical characteristic pathology exemplified by high epithelial cell loss, transmural neutrophil infiltrate and high levels of unclotted red blood cells associated with the villus structure and the lumen. Intestinal pathology was most pronounced at 3-d.p.c. but intestinal epithelial cell loss and inflammatory cell infiltration persisted until experimental end-point (11-d.p.c.). In addition, these animals showed tissue hyperplasia in the terminal colon that persisted to day 11 (data not shown). Hamsters that succumbed to infection with M68, showed more caecal pathology in comparison to those lacking symptoms and culled at 3-d.p.c. Typically more unclotted red blood cells was associated with the tissue and severe epithelial cell loss was apparent. At clinical end-point, caecal tissue from hamsters challenged with BI-7 showed pathological changes typically associated with *C. difficile* infection. Caecal tissue showed high levels of transmural neutrophil infiltrate and unclotted red blood cells with wide-ranging epithelial cell loss and extensive damage to the villus structure. # Discussion Here we present a detailed virulence study of three human isolates of *C. difficile*, BI-7, M68 and CD1342, in the hamster model of infection. Antimicrobial susceptibility assays showed both M68 and CD1342 had high-level resistance to clindamycin (\>256 µg/ml), whilst BI-7 had an intermediate level of resistance (64 µg/ml). Clindamycin is an important clinical antibiotic that has been implicated with induction of CDI with many strains showing high resistance including all A−B+ strains isolated from South Korean CDI cases. Previously, we described the infection kinetics of a clindamycin-sensitive Ribotype 027 U.K. isolate, R20291 (clindamycin MIC = 8 µg/ml), where 100% of hamsters succumbed to infection with a mean time of 46.7 h. The much faster infection kinetics displayed by the closely related Ribotype 027, BI-7, probably reflects the more efficient germination and subsequent survival of this strain within the clindamycin treated environment of the hamster gut. Due to this high- level clindamycin resistance hamsters were challenged with *C. difficile* one- day prior to clindamycin infection without any detrimental effects on inocula. Challenge of hamsters with *C. difficile* strain CD1342 resulted in a 100% colonisation rate, with animals remaining colonised until the end of the study 14-d.p.c. This is similar to other studies with non-toxigenic strains where colonisation was observed until at least 31-d.p.c.. Although PCR detection for the presence of the PaLoc and for the CDT-encoding genes in CD1342 were negative (data not shown), this strain still caused mild caecal pathology characterised by local acute epithelial cell loss, haemorrhagic congestion and neutrophil cell influx. This suggests that, at least in this strain, *C. difficile* CD1342 possesses an as yet uncharacterised virulence factor that is able to cause cell loss and damage, activating the immune system in the process. In the absence of the dominating effects of the toxins, i.e. using naturally occurring non-toxic strains, the hamster model of infection is ideally suited to elucidating potential transmission and/or colonisation factors for *C. difficile* infection. Sequencing of these types of non-toxic strains may give us more insight to such factors. Bacterial germination and outgrowth of CD1342 was rapid within the animal as even assuming 100% germination from the 10⁴ spores used to challenge the animals bacteria in the caecum and colon had multiplied by at least ∼3 LOG CFU within the first 24 h. High levels of spores were also observed, demonstrating the rapid *in vivo* germination, replication, sporulation and shedding of this strain in this short timeperiod. The rapid growth and high rates of shedding, thus enhanced transmission potential, coupled with potential acquisition of virulence factors could result in new clades of *C. difficile* with enhanced virulence similar to that observed with ribotype 027 strains, resulting in new global epidemics. The intestinal environment is a ‘hotbed’ for genetic exchange mediated by bacteriophages and these exchanges have resulted in the acquisition of virulence factors like antibiotic resistance determinants and pathogenicity islands in several bacterial species, such as the LEE locus in enterohemorrhagic *E. coli*. This type of genetic transfer has potential implications for the introduction of toxin encoding determinants, such as the PaLoc, to be transferred to previously non-toxic strains. This has potential implications on the potential use of non-toxic strains as probiotics to toxic *C. difficile*, When challenged with the toxin A negative *C. difficile* strain, M68, 100% of hamsters were colonised and shed this strain, with 45% of hamsters succumbing to disease. As all hamsters showed classic symptoms of CDI (wet tail) before potential recovery, the use of the telemetry system proved invaluable as we were able to discriminate those animals that had transitory disease from those that rapidly succumbed. These data show that, similar to a clinical setting, strain M68 (only producing a functional TcdB) is not only able to cause disease but can cause lethality in an *in vivo* model, which suggests that TcdA is not essential for disease initiation. However, with only a 55% survival rate in hamsters challenged with this strain, there may be a role for TcdA in fulminant CDI with this strain. Whilst it is not possible to directly attribute the role of individual genes in this type of study, the use of isogenic mutants to clarify the role of specific toxins has also been subject to controversy. In particular there has been confusion over the role of TcdA in CDI with isogenic mutants of *C. difficile* strain 630 where toxin genes have individually been disrupted generating conflicting data. Lyras *et al.* first reported a minimal role for TcdA whilst Kuehne *et al.* found an essential role for TcdA in CDI in the hamster. These differences may reflect the technologies used in mutant generation, differences in SNP profiles between the strains from different labs and even the method used to determine endpoint of experiments using the animal model. In contrast, our data is similar to that reported by, where hamsters colonised with an A−B+ clinical isolate (CF2) had a 50% survival rate. This suggests the role for TcdA in pathogenesis may vary dependent on strain and experimental conditions. Differences may reflect the acceptable endpoints criteria in different countries. However, within this experiment in which the endpoint is more refined a mixed picture is observed. Why some animals succumb to infection with M68 is unclear at present. Those animals that succumbed to infection showed no significant differences in total bacterial organ recoveries compared to animals culled at a similar time (3-d.p.c.). The small increase in vegetative cells observed could produce more TcdB, as shown by the toxin assay, causing more epithelial damage compared to those animals that survive. Interestingly those animals that survived challenge with M68, typically displayed an elevated core temperature above the normal range. Macrophages exposed to either TcdA or TcdB have been shown to release interleukin 1β (IL-1β), a key cytokine that, along with IL-6, can cause an increase in body temperature in rodents. Patients with severe *C. difficile* colitis often display clinically elevated IL-1β in their stool samples. Such an increased cytokine profile could be responsible for the febrile response seen in surviving animals. This possibility questions whether the surviving hamsters were better able to mount an appropriate immune response to toxin exposure, maybe due to genetic differences as the hamsters used in this study are from an out-bred colony. Since the initial Canadian outbreak, Ribotype 027 isolates have spread globally, causing major outbreaks in almost every continent. These strains have been associated with increased disease severity & reoccurrence rates, leading to this group of strains to be classed as ‘hypervirulent’. When challenged with BI-7 100% of hamsters rapidly succumbed to disease, ∼26 h post challenge. In a similar study, Razaq *et al.* observed increased mortality rates with epidemic Ribotype 027 isolates, especially with a clindamycin resistant isolate. Histological analysis of hamsters challenged with BI-7 showed severe epithelial cell loss, transmural neutrophil infiltrate and extensive damage to the submucosal structure. Such damage is, at least in part, due to the high toxin titres seen from the tissue samples, which considering the modest level of recovered bacteria is surprising. This high toxin titre could be due to more toxin being produced or the produced toxin is a more efficient enzyme due to changes in the protein sequence. Recently Lanis *et al.* showed that TcdB from strain R20291 (Ribotype 027) was more toxic due to conformational changes that occurred at a higher pH when compared to TcdB from strain 630. Another explanation for this rapid mortality could be the presence of the binary toxin, CDT. Through the induction of microtubule-based protrusions, CDT may enhance the adherence of *C. difficile* to host cells. In conclusion, using the hamster model of infection we characterised the infectivity profiles of three *C. difficile* isolates varying in toxin carriage. The toxin negative strain colonised hamsters and was shed with high efficiency but caused tissue damage, however no clinical symptoms were observed during the infection process. Whereas the toxin A negative strain caused mortality in 55% of hamsters, potentially associated with the increased toxin detected. Those animals that survived challenge displayed a febrile response highlighting potential host genetic differences involved in survival of CDI. Challenge with *C. difficile* BI-7 resulted in a rapid fatal infection in 100% of the animals, causing extensive tissue damage and high toxin titres observed. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AB GRD. Performed the experiments: AB JS LM DC JI. Analyzed the data: AB GRD. Contributed reagents/materials/analysis tools: GRD AB. Wrote the paper: AB GRD.

# Introduction The *SOX2* gene encodes a member of the SRY-related HMG-box (SOX) family of transcription factors involved in the regulation of embryonic development and in the determination of cell fate. In particular, it is well known that SOX2 plays important roles in maintenance of embryonic stem (ES) cell self-renewal and pluripotency. Among adult tissues, SOX2 is expressed in the brain, retina, tongue, lung, esophagus and stomach, and plays crucial roles in the differentiation and morphogenesis of these organs. We previously reported that SOX2 mRNA and protein were expressed in normal gastric mucosae, but frequently down-regulated in human gastric cancer tissues and cell lines, some of which are due to aberrant DNA methylation. We further revealed that SOX2 plays important roles in growth inhibition through cell cycle arrest and apoptosis, indicating that SOX2 may have tumor-suppressive functions in gastric cancer cells. However, the downstream target genes of SOX2 involved in gastric carcinogenesis remain largely unknown. MicroRNAs (miRNAs) are small, approximately 22-nucleotide, noncoding RNAs that regulate the expression of hundreds of genes by targeting their mRNAs posttranscriptionally. miRNAs bind to the partially complementary target sites in 3′-untranslated regions (3′-UTRs) of mRNAs, inducing direct mRNA degradation or translational inhibition. To date, it has been reported that the miRNA expression profiles differ between in normal tissues and derived tumors, including gastric cancer, and many miRNAs can act as tumor suppressors or oncogenes. Recently, it was reported that miRNA-134 and miRNA-145 repress SOX2 expression by targeting its coding region in mouse ES cells and the 3′-UTR in human ES cells, respectively. However, there have been no reports on miRNA(s) that can regulate SOX2 expression in human gastric cancer. In the initial step of this study, we performed immunohistochemical analysis of the SOX2 protein in human gastric cancer tissues, in which the DNA methylation statuses of *SOX2* had already been examined, and found that a certain number of SOX2 expression-negative cases did not show DNA hypermethylation, leading us to the idea that there is another mechanism underlying SOX2 down-regulation. Accordingly, in this study, we aimed to find miRNAs that target SOX2 expression in human gastric cancers. We found that miRNA-126 (miR-126) repressed SOX2 expression by targeting its 3′-UTR, and then performed functional analyses of miR-126 in gastric cancer cells. To further clarify the importance of miR-126-mediated SOX2 down-regulation in gastric carcinogenesis, we attempted to identify downstream target genes of SOX2 in gastric cancer cells. # Results ## The *SOX2* 3′-UTR is a predicted target of miRNA-126 and -522 In order to find novel miRNAs that regulate SOX2 expression in gastric cancer, we performed computational analysis using a miRNA target database, MicroCosm Targets (formerly miRBase Targets), and tried to identify miRNAs that target the *SOX2* 3′-UTR according to the following criteria. Considering the position, number, and sequence conservation of miRNA target sites among species, we selected two miRNAs, miR-126 and miR-522, as potential miRNAs targeting the *SOX2* 3′-UTR. miR-126 has two predictive target sites, which are both near the stop codon of the *SOX2* open reading frame (ORF) in the 3′-UTR, whereas the predicted target site of miR-522 is highly conserved in seven species and also located near the *SOX2* ORF stop codon in the 3′-UTR. ## miR-126 inhibits SOX2 expression in multiple cell lines To validate the results of computational analysis, we examined whether or not miR-126 and miR-522 can repress the expression level of the endogenous SOX2 protein in SOX2-expression-positive gastric cancer cell lines. As shown in the upper panel of, transfection of the miR-126 mimic molecule (Pre-miR-126) as well as SOX2 siRNA markedly reduced the endogenous SOX2 protein level compared with a non-specific negative control oligonucleotide (NC) in HSC43 cells, but the miR-522 mimic molecule (Pre-miR-522) did not. To generally evaluate the possibility that miR-126 inhibits SOX2 expression, we transfected Pre-miR-126 and Anti-miR-126 inhibitor (Anti-miR-126) into multiple gastric cancer cell lines. We used the following cell lines for transient transfection experiments: MKN45 (SOX2 positive; miR-126 intermediate) and TGBC11TKB (SOX2 positive; miR-126 negative) for Pre-miR-126 transfection; and HSC43 (SOX2 positive; miR-126 positive) and NUGC3 (SOX2 very low; miR-126 positive) for Anti-miR-126 transfection. Remarkable reductions of the SOX2 protein level were observed in Pre-miR-126-transfected MKN45 and TGBC11TKB cell lines. Conversely, Anti-miR-126 transfection up-regulated the SOX2 protein levels in both the HSC43 and NUGC3 cell lines, indicating that not only exogenous Pre-miR-126 but also endogenous miR-126 can regulate SOX2 protein levels in gastric cancer cells. We also performed quantitative real-time RT-PCR analysis of *SOX2* mRNA expression, and found that exogenous miR-126 modestly but significantly suppressed the *SOX2* mRNA level in HSC43 cells. Because the SOX2 protein is known to be abundantly expressed in ES cells, we examined whether or not miR-126 inhibits SOX2 protein expression in a mouse ES cell line (SOX2 positive; miR-126 negative). Interestingly, exogenous miR-126 transfection dose-dependently decreased the SOX2 protein level in the mouse ES cells, suggesting that miR-126 represses SOX2 expression in various species and cell lineages. ## miR-126 directly targets the *SOX2* 3′-UTR through two predicted binding sites To determine whether or not the predicted target sites for the miRNAs in the 3′-UTR of *SOX2* mRNA are responsible for the SOX2 down-regulation, we performed luciferase reporter assays with a vector containing the *SOX2* 3′-UTR downstream of the luciferase reporter gene. As shown in, significant repression of luciferase activities were observed in HEK293T cells co-transfected with the pGL4-*SOX2* 3′-UTR vector and Pre-miR-126 or siRNA that targets the *SOX2* 3′-UTR. On the other hand, Pre-miR-522 had no significant effect on the luciferase activity of the pGL4-*SOX2* 3′-UTR vector compared with NC. These results, combined with those of Western blot analysis, indicate that miR-126 is a more potential candidate miRNA that represses SOX2 expression in gastric cancer than miR-522. To examine the direct interaction of miR-126 with the potential target sites in the *SOX2* 3′-UTR, we carried out luciferase reporter assays with the deletion mutant vector as to the putative miR-126 target sites. As shown in, miR-126 has two predicted binding sites, A and B, in the 3′-UTR of *SOX2* mRNA. We therefore performed luciferase assays with the wild type pGL4-*SOX2* 3′-UTR vector (Wt), the vectors with each predicted miR-126 target site deleted, A (Del-A), B (Del-B), or both sites, AB (Del-AB). Intriguingly, each single deletion mutant vector exhibited a low inhibitory effect on luciferase activity compared with the Wt vector after Pre-miR-126 co-transfection. Moreover, the double deletion mutant vector, Del-AB, showed complete reversal of the inhibitory effect of the Pre-miR-126 co-transfection, indicating that miR-126 directly inhibits SOX2 expression by targeting the two binding sites in the 3′-UTR of *SOX2* mRNA independently. ## The inverse correlation between miR-126 and SOX2 expression in some cultured and primary gastric cancer cells To assess the relationship between miR-126 and SOX2 expression in gastric cancers, we initially examined SOX2 mRNA and protein levels by RT-PCR and Western blot analysis, respectively, in 10 gastric cancer cell lines without DNA methylation of *SOX2*. Five of the 10 cell lines showed low or undetectable levels of the SOX2 protein, whereas three of them exhibited a low *SOX2* mRNA level. We subsequently examined miR-126 expression by TaqMan real-time PCR analysis in these 10 gastric cancer cell lines. Four (NUGC3, GCIY, NUGC4 and HSC58) of the 10 cell lines, whose SOX2 mRNA and protein levels were low, exhibited relatively high expression of miR-126, whereas the other cell lines, except for HSC43 cells, exhibited relatively low expression of miR-126 and a high *SOX2* mRNA level. These data indicate that the miR-126 expression level is mostly opposite to the SOX2 mRNA and protein levels in gastric cancer cell lines. To further compare the expression pattern of miR-126 with that of SOX2 in primary gastric cancers, we initially examined the expression levels of SOX2 protein in 15 primary gastric cancer tissue samples without DNA methylation of *SOX2* by immunohistochemistry. We found that almost all non-cancerous mucosae showed SOX2-positive signal only in the cell nuclei within the neck of the gastric glands, whereas nine of the 15 cases exhibited low or undetectable levels of the SOX2 protein compared with the paired non-cancerous mucosae. Next, total RNA was isolated from these 15 gastric cancers and paired non- cancerous tissues, and the miR-126 expression levels were determined by TaqMan real-time PCR analysis. Four of the 15 cases exhibited significantly high levels of miR-126 expression, whereas three of them did low miR-126 levels in comparison with the adjacent non-cancerous mucosae. Among the miR-126-up- regulated cases, three (FG6, FG21 and FG24) exhibited lower levels of SOX2 protein than paired non-cancerous mucosae, suggesting that high levels of miR-126 expression contribute to low levels of SOX2 protein at least in some primary gastric cancers. There was no significant correlation between the miR-126 expression and sex, age, depth of tumor invasion or histological type (data not shown). ## miR-126 enhanced anchorage-dependent and -independent growth of gastric cancer cells We next evaluated the effect of miR-126 on tumor cell growth. We initially examined the proliferation rates of SOX2-expression positive gastric cancer cell lines, MKN45 and HSC43, after transient transfection of Pre-miR-126. As shown in, the Pre-miR-126-transfected-MKN45 and HSC43 cells exhibited significant growth advantages compared with the control NC-transfected-cells. Moreover, the SOX2 siRNA-transfected-MKN45 cells, but not HSC43 cells, significantly increased proliferation compared with the control cells, suggesting that miR-126-mediated growth stimulation may occur in a SOX2-dependent manner, at least in MKN45 cells. To determine the role of miR-126 in gastric tumorigenesis, we next carried out soft agar colony formation assays of gastric cancer cell lines after Pre-miR-126 transfection. As shown in, the Pre-miR-126- and SOX2 siRNA-transfected-MKN45 cells formed larger colonies than the NC-transfected cells in soft agar at 9 days after transfection. We then performed soft agar colony formation assays using a CytoSelect™ 96-Well In Vitro Tumor Sensitivity Assay Kit, which can be used to count the colony-forming cells by means of a colorimetric method, such as the WST-8 assay, making the assays quick and accurate. Exogenous miR-126 over-expression as well as SOX2 siRNA transfection significantly enhanced the anchorage-independent colony formation of MKN45 and HSC43 cells compared with the control cells at 9 to 10 days after transfection, respectively. These results suggest that miR-126 may promote the tumorigenicity of gastric cancer cells through suppression of SOX2 expression. ## miR-126 controls novel SOX2 target genes in gastric cancer cells To further understand the potential effects of miR-126-mediated SOX2 down- regulation on the gene expression change in gastric cancer cells, we first attempted to identify candidate downstream target genes of SOX2. We transiently expressed exogenous SOX2 in NUGC3 cells by using an adenovirus system, and changes in expression were determined by cDNA microarray analysis (GEO accession No. GSE23589). Among 41,174 probes, 366 known genes were up-regulated (\>2.0-fold) and 369 known genes were down-regulated (\<0.5-fold) by SOX2-over- expression in NUGC3 cells compared with in control GFP-over-expressing cells. Representative microarray results are summarized in, and we found the significant up-regulation of exogenous SOX2 (\>20.20-fold), supporting the validity of this experiment. Intriguingly, there were many cancer-related genes that could be novel downstream targets of SOX2 (for example, *LTF*, *PPP2R1B*, *TGFBR2*, *SERPINE1*, *MMP9*, *HMGA1*, *SOX9* and *PLAC1*), and squamous cell differentiation markers *KRT6E* and *KRT6C*, whose amino acid sequences are highly conserved among the KRT6 family members and virtually identical to one of the known SOX2 downstream genes, *KRT6A*. We validated the microarray results by RT-PCR analysis in NUGC3 cells after SOX2 over-expression, and representative results are shown in. Most of these genes also showed changes in their expression after SOX2 over-expression at least in one more gastric cancer cell line among the two to three cell lines we investigated (data not shown). To determine the target genes of SOX2 controlled by miR-126 in gastric cancer cells, we next performed SOX2 knockdown experiments and further screened for candidate target genes. SOX2 knockdown by Pre-miR-126 and siRNA was confirmed by Western blot analysis in SOX2-expression-positive gastric cancer cell lines MKN45 and HSC43, and the subsequent expression changes of the putative SOX2 downstream target genes were preliminarily analyzed by RT-PCR in these cell lines, and then by quantitative real time RT-PCR in HSC43 cells. Among over 20 cancer-related genes we investigated, only two showed changes in expression after SOX2 knockdown (data not shown). First, differentiation marker *KRT6A* expression, which was up-regulated by SOX2 over-expression, was significantly down-regulated by Pre-miR-126 as well as SOX2 siRNA transfection in HSC43 cells. Second, placenta- and tumor-specific *PLAC1* expression, which was down- regulated by SOX2 over-expression, was significantly up-regulated by SOX2 knockdown with Pre-miR-126 and siRNA in HSC43 cells. These results indicate that miR-126 can control *KRT6A* and *PLAC1* expression by down-regulating SOX2 expression in gastric cancer cells, and these genes might be downstream target genes of SOX2 contributing to gastric carcinogenesis. # Discussion We previously reported that SOX2 expression was frequently down-regulated in human gastric carcinoma tissues (about half of the total cases), some of which was due to aberrant DNA methylation (about 16% of the total cases). Therefore, the mechanisms underlying loss of SOX2 expression have not yet been defined in more than half of the cases. In this study, we demonstrated that miR-126 decreased the SOX2 mRNA and protein expression levels in gastric cancer cell lines. In addition, we found that miR-126 expression was inversely correlated with SOX2 expression in certain cultured and primary gastric cancer cells without DNA methylation of *SOX2*, indicating that aberrant miR-126 expression may be a novel mechanism underlying SOX2 down-regulation in gastric cancer. Furthermore, Pre-miR-126 over-expression promoted anchorage-dependent and -independent growth of gastric cancer cells *in vitro*, and increased oncogenic *PLAC1* expression in a gastric cancer cell line. These findings suggest that miR-126 may be an oncogenic miRNA that controls SOX2 expression in gastric cancer cells. Besides gastric cancer, Pre-miR-126 over-expression reduced the SOX2 protein level in mouse ES cells, suggesting that miR-126 may generally control SOX2 expression, at least in two species (human and mouse) and various cell lineages, including ES cells. MiR-126 is known as an endothelium-specific miRNA, and has been reported to promote angiogenesis by targeting *SPRED1* and *PIK3R2*, which normally inhibit VEGF signaling. Moreover, it has been reported that miR-126 inhibits apoptosis of acute myeloid leukemia (AML) cells and enhanced the colony-forming ability of mouse bone marrow progenitor cells through targeting Polo-like kinase 2 (*PLK2*), a tumor suppressor. On the contrary, miR-126 has also been reported to be a tumor suppressive miRNA, inhibiting tumor cell growth through targeting p85beta in colon cancer cell lines, and targeting IRS-1 in HEK293 and MCF-7 cells, respectively. Although it is controversial as to whether miR-126 is a tumor suppressive or oncogenic miRNA, at least in the present study, we demonstrated that miR-126 acts as an oncogene by targeting SOX2 in gastric cancer cells. These functional differences in oncogenesis might be considered to be a “lineage-dependency model for cancer”, that is, developmentally important genes also have crucial roles during tumor progression in lineage-specific manners. However, further studies are needed to clarify the biological roles of miR-126 in gastric carcinogenesis and other tissues. The initial computational analysis indicated that both miR-126 and miR-522 are candidate miRNAs that target the *SOX2* 3′-UTR. However, miRNA over-expression experiments showed that Pre-miR-126 but not Pre-miR-522 reduced SOX2 protein levels and *SOX2* 3′-UTR luciferase activity. Moreover, loss-of-function experiments and reporter assays involving deletion mutants of *SOX2* 3′-UTR luciferase vectors revealed that miR-126 directly targets the 3′-UTR of *SOX2*. This difference between miR-126 and miR-522 as to SOX2 regulation is likely to be due to the following reasons. First, the 5′ region, which is called the miRNA “seed” sequence (nucleotides 2–7), of miR-126 completely matches the 3′-UTR of *SOX2*, whereas miR-522 does not. It is well known that perfect “seed” pairing is required for both target site recognition and repression of the target transcript. Second, miR-126 has two binding sites in the 3′-UTR of *SOX2* mRNA, but miR-522 has only one. It has been reported that when a miRNA has multiple binding sites in the 3′-UTR of its target gene, the binding sites could be simultaneously targeted by the miRNA. These findings combined with our present data suggest that the presence of multiple complementary target sites and perfect matches between these ones and miRNA “seed” region are good indicators for finding functional miRNAs. In this study, miR-126 expression was found to be relatively high in SOX2-expression-negative gastric cancer cell lines, and was aberrantly up- regulated in some primary gastric cancer cases compared with the paired non- cancerous mucosae. However, the mechanism underlying this aberrant miR-126 expression in gastric cancer remains to be elucidated. It was previously reported that the over-expression of miR-126 in a kind of AML, core-binding factor (CBF)-AML, is associated with partial demethylation of the CpG island but not with amplification or mutation of the genomic locus. In fact, we observed that some gastric cancer cell lines exhibited restored miR-126 expression after treatment with a demethylating agent, 5-aza-2′-deoxycytidine (data not shown). These findings indicate that miR-126 expression may be epigenetically regulated in gastric cancer cells. We performed cDNA microarray analysis to identify the downstream target genes of SOX2 in gastric cancer cells, and found that many tumor-associated genes exhibited significant changes in expression after SOX2-overexpression (e.g., *LTF*, *PPP2R1B*, *TGFBR2*, *SERPINE1*, *MMP9*, *HMGA1* and *SOX9*). Furthermore, most of them also exhibited changes in their expression after SOX2-overexpression, at least in two gastric cancer cell lines. These results indicate that SOX2 might regulate the expression of these tumor-associated genes, thereby contributing to gastric carcinogenesis. However, we could not observe any significant changes in expression of these genes after SOX2 knockdown by Pre-miR-126 or siRNA, at least in the cell lines we tested. There are some possible reasons for this discrepancy. First, it has been reported that a different expression level of SOX2 switches the regulation of target gene expression from up- to down-regulation, or vice versa. In this study, the expression levels of SOX2 were quite different among the cell lines that were used for the over-expression and knockdown experiments. Second, it is well known that stem cell transcription factors, such as SOX2, OCT3/4 and Nanog, cooperatively interact with their target genes' promoters and control their gene expression, being so-called “transcriptional cofactors”. These expression differences and/or transcriptional cofactors might also be critical for control of expression of the downstream target genes of SOX2 in gastric cancer cells, and further studies are necessary to clarify the roles of SOX2 in the regulation of its target genes. Expression of *KRT6A* and *PLAC1* was significantly changed by both SOX2 over- expression and knockdown, suggesting that SOX2 is the critical regulatory factor for these two genes in gastric cancer cells. *KRT6A* is a member of the cytokeratin gene family, and recently it was reported that ectopic SOX2 over- expression up-regulated the *KRT6A* mRNA level in a lung adenocarcinoma cell line. In the present study, we demonstrated the possibility that *KRT6A* expression is also positively regulated by SOX2 in gastric cells, but the role of *KRT6A* in gastric carcinogenesis remains unclear. Thus, further investigations are necessary to elucidate the roles of *KRT6A* in gastric carcinogenesis. On the other hand, *PLAC1*, a recently described X-linked gene exhibiting expression restricted to the placenta, is also expressed in a wide variety of human cancers, including gastric cancer. Koslowski *et al.* reported that siRNA- mediated knockdown of PLAC1 decreased cell motility, migration and invasion, and induced G1-S cell cycle arrest with nearly complete abrogation of proliferation in breast cancer cell lines. In this study, we demonstrated that SOX2 negatively regulates *PLAC1* expression in gastric cancer cell lines, and propose a novel hypothesis that miR-126 inhibits SOX2 expression and consequent changes in the expression of some SOX2 target genes, such as *PLAC1*, thereby contributing to gastric carcinogenesis. In conclusion, for the first time, we demonstrated that miR-126 is a novel oncogenic miRNA, which targets SOX2, and that downstream pro-oncogenic target genes of SOX2, such as *PLAC1*, may contribute to gastric carcinogenesis. These findings have important implications for not only explaining the loss of SOX2 expression in gastric cancers, but also for understanding the transcriptional regulatory mechanisms of SOX2 in other various cell lineages, such as ES cells. Taken together, our findings may lead to new diagnostic and therapeutic approaches for gastric cancer, and provide new insights into the transcriptional regulation of SOX2. # Materials and Methods ## Ethics Statement Written informed consent was obtained from all subjects, and the study was approved by the Ethics Committee of Tokyo Medical and Dental University. ## Cell lines and tissue samples We used 10 human gastric cancer cell lines (HSC43, MKN45, TGBC11TKB, NUGC3, KATOIII, AGS, HSC44PE, GCIY, NUGC4 and HSC58) and one human embryonic kidney cell line (HEK293T) in this study, as described previously. All the cell lines were cultured in appropriate medium. Mouse ES cell line BL6 was obtained from Dr. Hirobumi Teraoka (Tokyo Medical and Dental University Medical Research Institute, Japan), and was cultured as described previously. A total of 16 primary gastric carcinoma tissue samples and corresponding non-cancerous gastric mucosae were obtained, as described previously. ## miRNA mimic and inhibitor transfection Gastric cancer cells were transfected with Precursor Molecules mimicking miR-126 (Pre-miR-126), miR-522 (Pre-miR-522) (Applied Biosystems, Foster City, CA), SOX2 siRNA (sense, 5′-GGAAUGGACCUUGUAUAGAUC-3′; and anti-sense, 5′-UCUAUACAAGGUCCAUUCCCC-3′, Sigma-Aldrich, St. Louis, MO), anti-miR inhibitor miR-126 (Anti-miR-126) (Dharmacon, Lafayette, CO), or scrambled sequence miRNA (Pre-miR-NC) (Pre-miR Negative Control \#1, Applied Biosystems) to give a final concentration of 10 to 100 nmol/L (nM) by using MicroPorator MP-100 (Digital BioTechnology, Seoul, Korea), according to the manufacturer's instructions. At 24–72 h after transfection, cells were harvested for Western blot or RT-PCR analyses. ## Western blot Western blot analyses were performed as described previously. The primary antibodies used were rabbit anti-SOX2 (1∶1000: Cell Signaling Technology, Danvers, MA) and mouse anti-α-tubulin (1∶200; Santa Cruz Biotechnology, CA). The secondary antibodies used were alkaline phosphatase-conjugated anti-mouse IgG or anti-rabbit IgG (1∶2000; Bio-Rad Laboratories, Hercules, CA). Blots were developed with Immun-Star™ AP Substrate (Bio-Rad Laboratories). We used α-tubulin as an internal protein loading control, and the band intensities were defined as described in the footnote of when 100 µg of protein was loaded per lane. ## RT-PCR and quantitative real-time RT-PCR Total RNA was extracted by using Trizol reagent (Invitrogen, Carlsbad, CA) and treated with DNA-free™ (Applied Biosystems). RT-PCR and quantitative real-time RT-PCR were performed as described previously. The primer sequences used for all genes are shown in. For semi-quantitative RT-PCR, *GAPDH* expression was used as an internal loading control, and the band intensities were defined as described in the footnote of under the conditions of 35 PCR cycles. ## Dual luciferase reporter assay The 3′-UTR oligonucleotide of *SOX2*, a 1050 bp fragment containing the last 36 bps of the *SOX2* coding region and the putative target sites of miR-126 and miR-522, was amplified by PCR with the following primers: sense, 5′-GCGCTCTAGAGCCATTAACGGCACACTGCC-3′; and anti-sense, 5′-GGCCTCTAGATACATGGATTCTCGGCAGAC-3′. Luciferase constructs were obtained by ligating the wild type 3′-UTR oligonucleotide of *SOX2* (Wt) or nucleotides with the miR-126 target sites deleted (Del-A, -B or -AB) into the *Xba*I site of the pGL4.13 (*luc2*/SV40) *firefly* luciferase reporter vector (Promega, Madison, WI). HEK293T cells were co-transfected using HiPerFect (QIAGEN, Hilden, Germany) with 10 ng of the pGL4.13 vector containing or not containing the 3′-UTR sequence (for normalization of the non-specific effects on pGL4.13-3′-UTR vector of miRNAs), 4 ng of the pGL4.74 (*hRluc*/TK) *renilla* luciferase control vector (for normalization of the transfection efficiency), and 30 nM Pre-miR-126, Pre- miR-522, SOX2 siRNA, or Pre-miR-NC. Luciferase activity was measured 24 h after transfection using a Dual-Luciferase Reporter Assay System (Promega). Relative luciferase activity was calculated by normalizing the *firefly* luminescence as to the *renilla* luminescence. ## Immunohistochemistory Paraffin-embedded tissue samples were sectioned, deparaffinized, and then pretreated by autoclaving in 10 mM citric acid buffer for 15 min to retrieve antigenicity. After the peroxidase activity had been blocked with 3% H₂O₂-methanol for 15 min, the sections were incubated with 10% normal goat serum in PBS to block nonspecific protein binding, followed by incubation with primary antibody against SOX2 (1∶300; Millipore) at 4°C overnight. Then, the sections were incubated with horseradish peroxidase-labeled goat anti-mouse-rabbit antibody (Dako, Carpinteria, CA) for 30 min at room temperature, and the signal was amplified and visualized with diaminobenzidine- chromogen, followed by counterstaining with hematoxylin. Expression was considered to be “positive” when 10% or more cancer cells were stained. ## Quantitative real-time RT-PCR of miRNA Total RNA was extracted by using Trizol reagent (Invitrogen) and then treated with DNA-free™ (Applied Biosystems) for cell lines. On the other hand, paraffin- embedded tissue samples were sectioned into 10 µm-thick, deparaffinized under RNase-free condition, and then total RNA was extracted by using RecoverAll™ Total Nucleic Acid Isolation Kit (Applied Biosystems, Foster City, CA) according to manufacturer's instructions. Quantitative real-time RT-PCR of miRNA was carried out using a TaqMan Reverse Transcription Kit (Applied Biosystems), TaqMan MicroRNA Assays (Applied Biosystems), and a LightCycler TaqMan Master (Roche Diagnostics, Mannheim, Germany), according to the manufacturers' instructions. The expression levels of miRNA were calculated by the delta-delta Ct method using RNU6B as an internal control. ## Cell proliferation and soft agar colony formation assays We transfected Pre-miR-126, SOX2 siRNA and Pre-miR-NC into HSC43 and MKN45 cell lines to give a final concentration of 50 nM by using MicroPorator MP-100. After 48 hours, the transfected cells were trypsinized, counted and replated in quadruplicate on 96-well plates (5×10² cells for HSC43, 2.5×10² cells for MKN45 per well). Cell proliferation was evaluated on days 1, 3, 5 and 7 after replating by determining the number of cells with a Cell Counting Kit-8 (Dojindo, Kumamoto, Japan), according to the manufacturer's instructions. For soft agar colony formation assays, we used a CytoSelect™ 96-Well In Vitro Tumor Sensitivity Assay Kit (Cell BioLabs, Inc., San Diego, CA), according to the manufacturer's instructions. Briefly, the transfected cells, as described above, were trypsinized, counted and plated in quadruplicate on 96-well plates with Agar Matrix Layer (2×10³ cells for HSC43, 1×10³ cells for MKN45 per well). After incubating the cells for 7 to 8 days at 37°C and 5% CO₂, the soft agar in each well was solubilized, and viable cells, that is, colony-forming cells, were measured with Cell Counting Kit-8 (Dojindo). ## Microarray analysis Adenovirus (Ad)-SOX2 and control Ad-GFP vectors were generated as described previously, and used to infect NUGC3 cells at the optimum MOI (infectious units/cell) of 20. At 72 h after infection, total RNA was extracted by using Trizol reagent (Invitrogen) and then treated with DNA-*free*™ (Applied Biosystems). cDNA microarray analysis was conducted by DNA Chip Research Inc. (Kanagawa, Japan) with Whole Human Genome oligo DNA arrays (Agilent Technologies, Santa Clara, CA). The microarray data is Minimum Information About a Microarray Experiment (MIAME) compliant and has been deposited in a MIAME compliant database, Gene Expression Omnibus (GEO). The GEO accession number is GSE23589. # Supporting Information We thank Dr. Hirobumi Teraoka for providing the mouse ES cell line and Dr. Hiroshi Fukamachi for the valuable discussion. [^1]: Conceived and designed the experiments: TO YA YY. Performed the experiments: TO KG. Analyzed the data: TO. Contributed reagents/materials/analysis tools: TO YH SS. Wrote the paper: TO YA YY. Participated in the manipulation of miRNAs: YH KG. Participated in the luciferase assay: SS KG. [^2]: The authors have declared that no competing interests exist.

# Introduction Healthy loading of the tibiofemoral joint of the knee during activities of daily living including gait involves significant tibial anterior shear and tibial internal rotation torque. The anterior cruciate ligament (ACL) is the primary restraint to anterior shear and a major secondary restraint to internal tibial rotation. Therefore, ACL deficiency through sports trauma results in anterior tibial translation and tibial internal rotational instability of the knee. Knee movement is a function of external forces and of muscle forces. In ACL- deficiency, knee joint stability is provided through the action of concavity compression of the tibiofemoral articulation on the medial side, where the compressive forces push together the concave surface of the joint. However, this stability mechanism is not present at the lateral knee compartment as the lateral tibial plateau is convex, resulting in an unstable and more mobile compartment. As a result, during normal knee joint loading with a tibial rotational torque, the rotational axis of the knee moves medially, creating an excessive translation of the lateral compartment. This excessive movement then causes secondary conditions including damage to the other passive restraints to these motions, such as cartilage, menisci, and the collateral ligaments. ACL deficiency is implicated with an increase in the rate of osteoarthritis and limits athletes in their activity. There is a subset of ACL deficient patients who are able to return to pre-injury activity without surgical intervention; these are termed copers. Coping is achieved through avoiding muscular contraction that produces an anterior shear force through, for example, avoiding full contraction of the quadriceps especially during the early stance phase and when the knee is at full extension. An alternative coping mechanism counteracts quadriceps contraction through co- contraction of the hamstrings and through the adaptation of muscle firing. The other set is that of non-copers who undergo ACL reconstruction surgery, where there may be a residual internal rotation instability. Prior work has shown that activating muscles crossing the knee with functional electrical stimulation (FES) is able to reduce anterior tibial translation (ATT), a surrogate measure of the anterior shear force. It has also been shown that FES, assisted with a knee brace, can be used to learn a muscle contraction pattern that then, once learned, persists despite halting the use of FES. Thus, the underpinning hypothesis of this work is that FES can restore normal ATT at the lateral compartment of the knee by entraining the contraction of specific knee muscles. The main muscles involved in the movement of the knee are the quadriceps, gastrocnemius and hamstrings. Of these, the hamstrings afford the most potential to reduce anterior tibial shear force and thus restore ATT to normal as they are anatomically located to apply a posterior pull to the tibia. Biceps femoris long head (BFLH) is the best candidate for selective activation in order to resist the peaks of anterior shear force and internal rotation torques during the stance phase of gait. It has been shown in a modelling study that activation of BF is able to decrease the anterior tibial shear force when knee flexion is less than 40°. Additionally, because BFLH attaches to the fibular head on the lateral aspect of the knee, it is expected that it will also be able to resist the large internal rotation torque and hence the large pathological motion of the lateral compartment in ACL deficiency. Thus, it is hypothesized that activation of BFLH is able to restore knee stability in non-copers to allow them to become copers. This preliminary study addresses two main questions: first, can muscle activation reduce the internal rotation torque and the anterior tibial shear force? Secondly, what is the optimum muscle activation to achieve this? Also, the effects of the muscle activation on other loading across the knee is examined. The study is a combination of computational modelling and an *in vivo* experimental study in healthy control subjects. # Materials and methods ## Physical experiments This study was approved by the Imperial College London Research Ethics Committee and written informed consent was obtained from all participants. Twelve healthy subjects (5 male, 7 female; height 1.67 ± 0.08 m; mass 66.74 ± 16.80 kg; age 26.08 ± 2.29 years) were recruited and underwent normal gait and functional electrical stimulation (FES) gait in sequence. During normal gait, subjects walked across the walkway in a self-selective comfortable walking speed, taking several steps prior to landing the right foot entirely on the force plate, and continuing for several steps. FES gait immediately followed the normal gait. The skin of the right BFLH region was prepared with 70% isopropyl alcohol skin wipes and two FES electrodes (Odstock 2 Channel Stimulator, Odstock Medical Ltd., UK) were placed over the region: one placed in the middle of the line between the ischial tuberosity and the lateral epicondyle of the tibia, and the other placed two hand widths distal to the first. The frequency of the stimulator was set to the manufacturer recommended level of 40 Hz and simulation current was set to a minimum value of 40 mA. The subject was asked to stand on their left leg in a neutral position, while the stimulator was activated. In all cases this resulted in visible flexion of the right knee, confirming the excitation of BFLH. The intensity was then adjusted to the maximum level that the subject was able to comfortably withstand. The FES stimulation current was set up to start with one second of ramp up, followed by four seconds of maximum current and then one second of ramp down. The stimulator was manually started by the subject and timed so that the stimulation current was at its maximum value from when the right foot stepped on the force plate, through heel strike, until toe off. Several practice trials were made to allow the subject to become accustomed to the required timing after which motion capture commenced. Ground reaction forces (GRF) were recorded at 1000 Hz from a force plate (Kistler Type 9286BA, Kistler Instrument AG, Winterthur, Switzerland). A ten-camera motion analysis system (Vicon Motion Systems Ltd, Oxford, UK) recorded the motion of the right lower limb at 200 Hz; eighteen retro-reflective markers were attached to the foot, thigh and pelvis with an additional two clusters of three markers attached to the shank and thigh. The subjects walked for six trials for normal gait and six for FES gait, of which a random selection of three trials each were used for data analysis. ## Lower limb musculoskeletal model Open source musculoskeletal modelling software, Freebody V2.1, was used. This has been validated for knee joint forces and muscle activity using direct measures from instrumented implants and electromyography. The segment-based lower limb model consists of the foot, shank, thigh, pelvis and patella (the patella segment is assumed to be massless, and its position and orientation is determined based on the knee flexion angles and the geometry of the patellofemoral joint). The model inputs are the kinematics data from the retro reflective markers and the kinetic data from the force plate. The model calculates the intersegmental forces and torque at the proximal end of each segment. Each subject’s anatomical geometry was created by linear scaling of an MRI-based anatomical dataset. The dataset consists of 163 muscle elements representing 38 lower limb muscles. The muscle attachment sites, joint centres of rotation, and tibiofemoral contact points were manually digitized from the MR imaging of a male subject (1.83 m, 96 kg, 44 years). The model quantifies the muscular and joint reaction forces experienced by the lower limb during the recorded movement through minimisation of a cost function: $$min{\sum_{i = 1}^{163}\left( \frac{f_{i}}{f_{max_{i}}} \right)}^{3}$$ where *f*_*i* is the muscle force of muscle element *i* (*i* = 1,…,163) and *f*_*maxi* is the maximal muscle force of muscle element *i*, which is determined by multiplying published physiological cross-sectional areas of muscle element *i* by an assumed maximum muscle stress of 31.39 N/cm², constrained by the equations of motion of the whole lower limb: $$\begin{bmatrix} \mathbf{S}_{\mathbf{i}} \\ \mathbf{M}_{\mathbf{i}} \\ \end{bmatrix} = \begin{bmatrix} {m_{i}E_{3 \times 3}} & 0_{3 \times 3} \\ {m_{i}\mathbf{c}_{\mathbf{i}}} & \mathbf{I}_{\mathbf{i}} \\ \end{bmatrix}\ \begin{bmatrix} {\mathbf{a}_{\mathbf{i}} - \mathbf{g}} \\ \overset{¨}{\mathbf{\theta}_{\mathbf{i}}} \\ \end{bmatrix} + \begin{bmatrix} 0_{3 \times 1} \\ {{\overset{˙}{\mathbf{\theta}}}_{\mathbf{i}} \times I_{i}{\overset{˙}{\mathbf{\theta}}}_{\mathbf{i}}} \\ \end{bmatrix}\ \begin{bmatrix} E_{3 \times 3} & 0_{3 \times 3} \\ \mathbf{d}_{\mathbf{i}} & E_{3 \times 3} \\ \end{bmatrix}\ \begin{bmatrix} \mathbf{S}_{\mathbf{i} - 1} \\ \mathbf{M}_{\mathbf{i} - 1} \\ \end{bmatrix}$$ where *i* is the segment number or joint number (numbering from distal to proximal), ***S_i*** the proximal intersegmental forces, ***S***_***i*−1** the distal inter- segmental forces, ***M_i*** the proximal intersegmental torques (notional joint torques), ***M***_***i*−1** the distal intersegmental torques (notional joint torques), ***I_i*** the inertia tensor, ${\overset{¨}{\mathbf{\theta}}}_{\mathbf{i}}$ the angular acceleration about COM, ***a_i*** the linear acceleration of COM, *m*_*i* the segment mass, *E*_3×3 the identity matrix, ***c_i*** the vector from the proximal joint to the segment COM and ***d_i*** is the vector from the proximal to the distal joint. In order to quantify the effect of higher muscle activation of BFLH produced by the FES at the knee, a revised optimisation method is proposed: $$min{\sum_{i = 1}^{162}\left( \frac{f_{i}}{f_{max_{i}}} \right)}^{3}$$ where *f*_*i* is the muscle force of muscle element *i* (*i* = 1,…,162) and $f_{max_{i}}$ is the maximal muscle force of muscle element *i*. In the revised optimisation method, the muscle force of BFLH is set as a constant value during the stance phase to replicate the physical stimulation of the muscle by FES. This value is set at a muscle activation, *c* times the maximum force of BFLH. As the attachment sites of BFLH are on the shank and thigh segments, the equations of motion of the shank and thigh segments were modified by the inclusion of an additional term to give: $$\begin{bmatrix} \mathbf{S}_{\mathbf{i}} \\ \mathbf{M}_{\mathbf{i}} \\ \end{bmatrix} = \begin{bmatrix} {m_{i}E_{3 \times 3}} & 0_{3 \times 3} \\ {m_{i}\mathbf{c}_{\mathbf{i}}} & \mathbf{I}_{\mathbf{i}} \\ \end{bmatrix}\ \begin{bmatrix} {\mathbf{a}_{\mathbf{i}} - \mathbf{g}} \\ \overset{¨}{\mathbf{\theta}_{\mathbf{i}}} \\ \end{bmatrix} + \begin{bmatrix} 0_{3 \times 1} \\ {\overset{˙}{\mathbf{\theta}_{\mathbf{i}}} \times I_{i}\overset{˙}{\mathbf{\theta}_{\mathbf{i}}}} \\ \end{bmatrix} + \begin{bmatrix} E_{3 \times 3} & 0_{3 \times 3} \\ \mathbf{d}_{\mathbf{i}} & E_{3 \times 3} \\ \end{bmatrix}\ \begin{bmatrix} \mathbf{S}_{\mathbf{i} - 1} \\ \mathbf{M}_{\mathbf{i} - 1} \\ \end{bmatrix} - \begin{bmatrix} {\left( {c \times f_{BFLH_{max}}} \right) \cdot \mathbf{n}_{\mathbf{B}\mathbf{F}\mathbf{L}\mathbf{H}}} \\ {\left( {c \times f_{BFLH_{max}}} \right) \cdot (\mathbf{r}_{\mathbf{B}\mathbf{F}\mathbf{L}\mathbf{H}} \times \mathbf{n}_{\mathbf{B}\mathbf{F}\mathbf{L}\mathbf{H}})} \\ \end{bmatrix}$$ where *c* is a constant, $\mathbf{f}_{{\mathbf{B}\mathbf{F}\mathbf{L}\mathbf{H}}_{\mat hbf{m}\mathbf{a}\mathbf{x}}}$ the maximum force of BFLH, ***n***_***BFLH*** the line of action of BFLH and ***r***_***BFLH*** the moment arm of BFLH. In this study, *c* was increased in increments of 0.05 until the peak anterior tibial shear was reduced to zero, where *c* is a value between 0 and 1, to make sure that the BFLH force does not exceed its maximum activation value. The reduction in BFLH activation theoretically causes a reduction in tibial internal torque, which was calculated as the product of the reduction of BFLH muscle force and its moment arm at the time frame at which peak anterior tibial shear occurred. ## Data analysis The stance phase was expressed in a 0–100% duration with a step interval of 1% using cubic spline data interpolation. Walking speed, knee joint torque, anterior shear force, knee contact force and patella tendon force were measures of interest and expressed as the mean value of three trials. Knee joint torque and knee contact force were presented in the tibial coordinate frame. Patella tendon force was calculated from the force balance across the patellofemoral joint. To test the hypothesis that the peak of the tibial internal rotation torque and the anterior shear force were reduced by applying the FES over the BFLH, the differences between normal gait and FES gait were compared using a one-tail paired-samples t-test with an α level of 0.05. All data processing and analysis was conducted in MATLAB (The Mathworks Inc., Natick, MA). # Results Walking speed during FES gait (0.25m/s) was significantly reduced by 7% compared to normal gait (0.27m/s; *p* = 0.036). The peak value of the tibial internal rotation torque across all subjects was 0.0012± 0.0010 Nm/BW during normal gait. It was reduced by 63% to 0.0005 ±0.0004 Nm/BW (*p* = 0.032) when BFLH was stimulated by FES. The first peak of adduction torque and of flexion torque were not significantly different during FES gait compared to normal gait (*p* = 0.3457 and *p* = 0.2623, respectively;). In the standard optimisation method, the peaks of anterior shear force and internal rotation torque both occurred at late loading response (between 12.6% and 18.8% of the stance phase). Mean peak anterior shear was 0.2892 ± 0.0766 BW. The muscle activation of BFLH at peak anterior shear was 0.0152 ± 0.0214. Increasing BFLH activation incrementally resulted in an incremental reduction in the anterior shear force. The activation of BFLH (expressed as a *c* value) required to reduce the peak anterior shear force to zero in the revised optimisation ranged from 0.15 to 0.40 with a mean *c* value of 0.208 ± 0.084. Applying the mean value of 0.208 to all subjects, reduced the peak anterior shear force to below zero in 11/12 subjects and was 0.0778 BW for the other subject. At the time frame at which peak anterior shear force occurred, the reduction in tibial internal torque was calculated as the product of the reduction of BFLH muscle force and its moment arm at that time frame and normalised by body weight. This level of muscle activation at 0.208 caused a 188% reduction of the internal rotational torque of 0.0226 ± 0.0167 Nm/BW (*p* = 0.0002). The knee compressive force with 0.208 BFLH muscle activation is shown in. The first peak of lateral knee compressive force was increased by 276% (*p*\<0.0001) during FES gait when compared to normal gait, resulting in an increase in overall knee compressive force of 144% (*p* = 0.0003). There was no significant difference for the first peak of medial knee compressive force (*p* = 0.2373). The patella tendon force with 0.208 BFLH muscle activation is shown in. The first peak of patella tendon force was increased significantly (*p* = 0.0000) by 61% from 1.5700 ± 0.4635 BW in normal gait to 2.9100 ± 0.7922 BW during the FES gait. # Discussion This study tested, using a combined modelling and experimental approach, the hypothesis that selective activation of the BFLH, one of the hamstrings, can theoretically and practically reduce the anterior tibial shear and knee internal rotation torque at the knee. The hypothesis was derived due to the anatomy of the muscle, which attaches on the fibular head that articulates with the lateral tibia and so has the potential to resist a large internal rotation torque and hence the pathological motion of the lateral compartment that occurs in ACL deficiency. We found that the anterior shear force and the knee internal rotation torque were reduced when BFLH was stimulated with FES. FES gait was slower than healthy gait and therefore it is possible that the reduction in internal rotation torque may be due to this small 7% reduction in speed in addition to that due to the FES assisted muscle activation. However, it is expected that this effect due to speed is small compared to the large 63% reduction in internal rotation torque. Furthermore, the BFLH stimulation does not affect the knee adduction torque and flexion torque (p\>0.01). The modelling approach used two optimisation methods to solve the muscle indeterminacy problem; both of these optimisation methods show that when the ACL is loaded during weight acceptance in FES gait the peak tibial internal rotation torque was reduced. The reduction of the tibial internal rotation torque indirectly affects the value of the anterior shear force. Theoretically, as BF inserts on the fibula its activation in a flexed knee is able to pull the tibia posteriorly. In this study, the peak anterior shear force was significantly reduced when FES was applied during weight acceptance, before full knee extension. This work is consistent with the model simulation by Shelburne et al and the experimental study by Chen et al showing that by increasing the muscle activation of the hamstrings, ATT was reduced by 0.2 cm with the knee in 20° to 50° of flexion. Also, in healthy gait body weight is transferred onto the forward limb in the weight acceptance phase. In contrast, for FES gait, the posterior pull of the extra activation of the BFLH by the FES resulting in slower than normal gait, as found in our study. Here, the modelling cost function was modified from its standard form by assigning a weighting, *c*, to simulate BFLH stimulation. The value of *c* for each subject that reduced anterior shear force to zero was found and the mean value of *c* across all was 0.208. This mean value was then used, resulting in only one subject having a very small positive anterior shear force, demonstrating that the use of a mean value to simulate external activation using FES is appropriate. This particular subject required a *c* value greater than 0.208 to decrease anterior shear force. This may be due to the subject’s characteristics: this was the tallest and heaviest subject. This work also follows the literature in which a similar *c* value of 0.25 was used to simulate the electrically stimulated muscle activation of gluteus medius to reduce the medial knee joint reaction force. In the literature, hamstrings activation without FES has shown that 56% of the maximal hamstring muscle force could reduce the ATT to a normal level during the stance phase of gait. That study modelled motions in the sagittal plane only and so cannot be compared for tibial internal rotation. Focusing on ATT only would suggest that the hamstrings on the medial side could also reduce anterior shear force and this has been shown in other modelling studies. However, as these do not assess tibial internal rotation torque, their results cannot be compared here. It should be noted that in this study over activation of the hamstrings resulted in a higher knee contact force due to the co-contraction of the quadriceps muscles to overcome the flexion torque due to the hamstrings activation. This has been addressed in a previous study by Catalfamo et al. who proposed that a 50% biceps femoris stimulation is more appropriate than a 100% stimulation to reduce ATT due to the pathological increase in knee joint forces and we have provided further evidence for this proposal. This study has some limitations. Firstly, the test cohort comprised only healthy control subjects; future work should focus on conducting experiments on ACL deficient subjects to test the applicability of this method in a clinical cohort. We would expect the results in such a cohort to be amplified as an ACL deficient subject would have reduced ability of the passive stabilisers to resist the ATT and internal rotation torque, thus emphasising the effect of the musculature. Secondly, future work should investigate not only the effect of activation of BFLH in ACL deficient subjects, but also include the effect of stimulating other muscles. A confounding factor in ACL deficient subjects is that they already demonstrate altered muscle activation patterns that might result in a different pattern of internal rotation torque and anterior shear force. Thus, such studies might also include an investigation of compensatory muscle activations due to selective activation of BFLH, perhaps through the use of electromyography. Third, the use of static optimisation to determine the muscle forces needs to be further validated, as it may not reflect in vivo muscle force generation. Fourth, the timing of FES can be improved by placing the switch under the subject’s heel, to enable the FES to be set to to high stimulation during heel strike of the injured leg and set to low stimulation during heel strike of the contralateral leg. However, in this study the four seconds of stimulation was enough to make sure the high stimulation occurred during stance phase and was synchronized with the revised optimisation. Finally, the application of a constant muscle activation for the whole of stance phase as achieved here is neither desirable, nor practical. Technology to allow selective activation at the peak of anterior tibial shear should be developed for appropriate clinical use. # Conclusion This study is the first to have shown that selective activation of the BFLH can reduce the anterior tibial shear and tibial internal rotation torque at the knee in healthy subjects. It opens the way for new rehabilitation therapies for ACL deficient subjects using FES. This research was performed within the Medical Engineering Solutions in Osteoarthritis Centre of Excellence (reference number: 088844/Z/09/Z), which is funded by the Wellcome Trust and the EPSRC. Nur Liyana Azmi is supported by the Malaysian Ministry of Higher Education. [^1]: The authors have declared that no competing interests exist.

# Introduction The work-up of patients with suspected heparin-induced thrombocytopenia is difficult due to major practical problems and methodological restrictions. Frequently, suspicion is expressed at times when haematological consultancy services and sophisticated laboratory tests are not immediately available: during night-shifts and weekends. However, a decision whether or not heparin shall be stopped and treatment with alternative anticoagulants started must be taken rapidly to prevent major complications. Clinical scoring systems such as the 4Ts score are available to guide decision-making at bedside, but rating is difficult in inexperienced hands and their positive predictive value is too low for diagnosing HIT. Functional assays such as the serotonin release assay (SRA), the heparin-induced platelet activation assay (HIPA) or the heparin-induced platelet aggregation test (PAT) are regarded as gold-standard for the diagnosis of HIT. However, these tests are time-consuming and available in very few laboratories only. Thus, physicians mostly rely on immunoassays to establish the presence of anti-platelet factor 4 (PF4)/heparin antibodies. The diagnostic accuracy of enzyme-linked immunoassays (ELISA), which are used most frequently, is well evaluated. However, ELISA results are available once daily at most because determination is time-consuming and requires a specialized laboratory. Several rapid immunoassays were developed to overcome this limitation: particle gel immunoassay (PaGIA), lateral flow immunoassay, latex immunoturbidimetric assay, and chemiluminescent immunoassay. PaGIA and lateral flow immunoassay can be implemented in routine laboratories, because they do not require specialized analysers or expert knowledge. Chemiluminescent immunoassay and latex agglutination assay can be easily automated. However, evaluation studies comparing the diagnostic accuracy, reproducibility and costs of these assays in clinical practice are very limited, particularly with regard to the Swiss health care system. To address these issues, we conducted a comprehensive evaluation study comparing accuracy, reproducibility, and costs of all commonly used immunoassays in clinical practice. # Methods ## Study design and population To estimate the utility of immunoassays for diagnosis of HIT in clinical practice, we selected a cohort of patients evaluated for suspected HIT at our institution (Inselspital University Hospital, University of Bern, Switzerland). To avoid any selection bias, we screened the database for a period with 190 consecutive patients with residual plasma samples. The following clinical data were extracted retrospectively: age, sex, setting (surgery, intensive care unit \[ICU\], or internal medicine), as well as results of the 4Ts score. We recorded quantitative results of ELISA, semi-quantitative results of PaGIA, and PAT if conducted at the time of assessment. We accepted missing test determinations, because we focused our investigation on real-life clinical practice. Residual citrated plasma samples were used to additionally perform a recently developed rapid immunoassay, the IgG-specific chemiluminescent immunoassay (AcuStar HIT- IgG). The study received approval by the local Ethics review board (Kantonale Ethikkommission Bern; 21.12.2014). ## Work-up of patients with suspected HIT in clinical practice A protocol for management of patients with suspected HIT was implemented in 2004 and periodically renewed. Haematology consultancy team was informed by attending physicians or laboratory technicians (HIT test requested). Haematologist contacted attending physicians to clarify clinical probability, predominantly with the help of the 4Ts score. PaGIA was conducted in most cases as a preliminary rapid test within few hours, followed by a polyspecific ELISA in the course of the following week. PAT was conducted as a confirmatory test every three months at the discretion of the haematology consultant. ## Handling of samples and determination of immunoassays Blood samples were obtained using citrated plastic syringes (Monovette®, Sarstedt, Nümbrecht, Germany; 1:10 trisodium citrate 0.106 mol/L). Platelet poor plasma was generated by double centrifugation at 1500 g x 10 min. Aliquots were snap-frozen and stored at -70°C. PaGIA (DiaMed SA, Cressier sur Morat, Switzerland) was conducted by pipetting 10 μL of patient plasma to the upper reaction chamber, followed by 50 μL of particle suspension. Results were recorded after 5 minutes of incubation at room temperature and 10 minutes’ centrifugation with the DiaMed ID-centrifuge. Semi-quantitative results were obtained by titration studies as previously described. Polyspecific ELISA was determined as reported elsewhere using a microtiter plate reader, measured at 405 nm (GTI-PF4 Enhanced, Genetic Testing Institute, Waukesha, WI, USA). AcuStar HIT-IgG (Instrumentation Laboratory, Bedford, MA, USA) was conducted in batches on a BIO-FLASH® analyser (Inova Diagnostics, San Diego, California, USA). Samples were thawed rapidly at 37°C. Assays were calibrated using AcuStar HIT- IgG calibrator 1+2, and controls were tested before every test run. Results of 4Ts score, ELISA and PAT were not available to the technician conducting AcuStar HIT assay. ## Definition of HIT HIT was defined as a positive functional assay, the PAT test. Following recent guidelines and recommendations, PAT was not conducted in a number of patients with a negative ELISA and/or a negative PaGIA in combination with a low risk/ intermediate risk 4Ts score. These cases were classified as HIT negative. Inconclusive results were excluded as recommended by several authors (n = 8). We conducted two sensitivity analyses considering different definitions of HIT to address possible issues regarding PAT as the reference gold standard: (1) a broad definition of HIT was applied, considering patients with an ELISA result above OD 2.5 or an PaGIA titre of at least 32 in addition to PAT positive samples, and (2) considering PAT as definition of HIT only, excluding all patients without PAT results. We determined a two-point PAT as previously described using a commercially available light transmission aggregometer APACT 4004 (LABiTec, Ahrensburg, Germany). Patient plasma was tested using platelet rich plasma from 4 selected donors. PAT was classified as positive if aggregation was more than 50% in at least 2 out of 4 donors with 0.5 U/mL heparin and suppressed with 100 U/mL heparin. Neither PaGIA nor AcuStar HIT-IgG results were available to the technician performing PAT and no other result were available to the technician conducting PaGIA. ## Costs We calculated the analytical cost for determining the individual tests taking the following factors into account: reagents, controls, calibrators, consumables, and work expended. We assumed that 5 tests per week are determined, that tests are conducted on different days in case of PaGIA, AcuStar HIT-IgG, and that ELISA is run once weekly analysing all 5 samples in one run. In case of PaGIA, we applied 2.3 steps of titration per analysis (average number in our institution). Costs are based the on Swiss health care system. Labor costs were estimated according to the average wage of a qualified technician in Switzerland (CHF 42,- per hour). ## Statistical analysis Descriptive statistics was used to characterise the study population. Accuracy was determined by calculating the sensitivity and specificity as well as likelihood ratios for the individual tests according to the presence of HIT. Calculations were repeated using two additional definitions of HIT as sensitivity analysis. Specificities between different thresholds and assays were compared using the z-test. In addition, post-test probabilities of different test results according to pre-test probability were reported. Calculation of diagnostic accuracy was done at three different thresholds; all of them are already established. The low threshold corresponds to the recommendation of the manufacturer. The intermediate and high thresholds were identified in previous diagnostic accuracy studies with larger patient populations using receiver operating characteristics analysis. Thus, an intermediate threshold of PaGIA does not directly correspond to an intermediate threshold of ELISA. Reproducibility was determined by calculating coefficients of variation (CV) for within-run repetitions (10 times) as well as day-to-day (5 times). A CV of 0.1 was set as requirement for imprecision, according to general principles of ELISA, clinical considerations and previous publications. We calculated a sample size of 190 to demonstrate differences in specificity. We anticipated a frequency of true-negative values of 85%, considered a difference in specificity of 5% to be clinically relevant and a fixed a 95% CI. Analyses were performed using the Stata 13.1 statistics software package. (StataCorp. 2013. Stata Statistical Software: Release 13. College Station, TX: StataCorp LP), Figs were created using Prism 6 (GraphPad Software, Inc., La Jolla California USA). # Results ## Patients The flow of participants is shown in. Residual plasma samples were available in 187 out of 190 patients. Eight patients were excluded from analysis because of uncertain diagnosis of HIT. Finally, 179 patients were included for analysis. Among them, AcuStar HIT-IgG was conducted in 179 patients (100%), PaGIA in 171 patients (95.5%), and ELISA in 144 patients (80.4%). Median age was 70.0 years (inter-quartile range \[IQR\] 61.3, 76.9), 42.5% were female. Detailed patient characteristics are shown in. The clinical setting was surgery in 38.6% of the cases, internal medicine in 32.4% of the patients, and intensive care unit in 29.1% of the patients. The clinical risk according to 4Ts score was estimated to be low in 24.6%, intermediate in 15.6%, and high in 1.7%. The 4Ts score was missing in 104 patients (58.1%) because samples were received from external hospitals. ## Antibody test results Thirteen patients were classified as HIT-positive because of a positive PAT test. One hundred and forty-one patients were classified as HIT-negative because of a negative PAT, and 21 patients because of a negative ELISA and/or PaGIA result and a low/intermediate 4Ts score. Thus, the prevalence of HIT in the present study population was 7%. ELISA was conducted in 144 patients (80.0%), with a major difference between HIT-positive patients (median OD 2.90) and HIT- negative patients (median OD 0.17; ;). Results were missing in 41 patients (22.9%), mostly because of a negative PaGIA and a low risk 4Ts score. Data on PaGIA results were available for 171 patients (95.5%), the distribution among HIT-positive and HIT-negative patients is shown in and. AcuStar HIT-IgG were conducted in all 179 samples resulting in very low levels in HIT-negative patients and high levels in HIT-positive samples. ## Comparative diagnostic accuracy Quantitative immunoassay results were considerably higher in patients with HIT compared to patients without HIT in all of the tests. The distribution of results in relation to different thresholds of the immunoassays is shown in. In the current population, sensitivity was 100% for all assays at low and intermediate thresholds. However, the 95% confidence intervals are wide. At high threshold, sensitivity was 84.6% for the AcuStar HIT-IgG and 58.3% for PaGIA only. Specificity was considerably below 90% at low thresholds for all assays except for AcuStar HIT-IgG (92.8%; 95% CI 87.7, 96.2). Specificity increased significantly at intermediate thresholds (PaGIA: 95.6%; 95%CI 91.1, 98.2; ELISA: 96.8; 92.4, 99.2; AcuStar HIT-IgG: 97.6; 94.0, 99.3). Specificity of AcuStar HIT-IgG was significantly higher than PaGIA and ELISA at low threshold. Essentially identical results were observed in first sensitivity analysis, considering PAT as reference gold standard only (Table A and Table B). Sensitivity analysis 2, considering a broad definition of HIT revealed lower numbers for sensitivity and a comparable specificity for all of the assays. However, sensitivity was above 90% at low thresholds as well. ## Reproducibility Within-run imprecision was within predefined limits for all assays (CV\<0.1). A CV of 0.04 was determined for ELISA, and 0.03 for AcuStar HIT-IgG. Antibody titre was identical in 9 out of 10 PaGIA measurements (one replicate with a titre of 4 instead of 2). Day-to-day variation was within the predefined limits as well (CV≤0.1). For ELISA, CV was 0.07, and for AcuStar IgG 0.03. Four out of 5 PaGIA titres were identical (a titre of 4 instead of 2 was seen in one sample) confirming previous results. ## Costs We calculated total costs per test for the individual assays as follows: CHF 51.02 for ELISA, CHF 117.70 for AcuStar HIT-IgG, and 83.13 for PaGIA. In case of ELISA, reagents summed up to CHF 26.10 per sample (including controls and calibrators), consumables to CHF 1.90, and average wage of CHF 24.5 (175 minutes of work expended). For PaGIA, reagents costs CHF 65.19, consumables 0.44, and wage CHF 17.50 (25 minutes). Costs of AcuStar HIT-IgG were CHF 106.20 for reagents, CHF 1.00 for consumables, and CHF 10.50 for wages (15 minutes). # Discussion As observed in clinical practice, all immunoassays demonstrated adequate diagnostic accuracy measures at intermediate thresholds; sensitivity was above 95% and specificity above 90%. In general, specificity was higher with AcuStar HIT-IgG and lower at low thresholds for all assays. A good reproducibility was shown for all tests. Relevant differences exist with regard to analytical costs in the Swiss health care system. Even though this study was conducted in clinical practice, our results are in- line with previous publications. First polyspecific ELISA were developed in the nineties, and a number of studies evaluated their diagnostic accuracy. A recent meta-analysis estimated the pooled sensitivity to be 96.7% at low threshold and 98.4% at intermediate threshold (specificity 86.8% and 94.9% respectively). In the same meta-analysis, diagnostic accuracy of PaGIA was calculated to be 96.5% (low threshold), and 98.9% (intermediate threshold). Specificity of PaGIA was 93.7% and 95.9% respectively. Specificity of PaGIA was considerably lower in our investigation, probably due to a limited precision (restricted number of patients) or some inconsistencies of test determination in clinical practice. Fewer studies focused on chemiluminescent immunoassay, pooled data are available from the same meta-analysis (sensitivity at low/ intermediate thresholds for AcuStar HIT-IgG 98.9%/ 78.6%; specificity of AcuStar HIT-IgG 94.9%/ 98.7%). In our data, sensitivity was higher at intermediate thresholds, most likely because of a limited number of HIT-positive patients. Most previous data on reproducibility originate from a North American proficiency testing program. CV was 0.29 for a polyspecific ELISA (GTI). Our data, showing much lower variation, suggest relevant variation in laboratory procedures as a source for this variability. A good reproducibility was shown in one study for PaGIA and in another for chemiluminescent immunoassay as well. Costs of “rapid” tests, PaGIA and chemiluminescent immunoassays were higher compared to the ELISA assay. However, rapid immunoassays can be conducted in a 24 hours service, reducing the time span between suspicion of HIT and antibody testing from several days to few hours. A recent published budget model estimated that this might save a relevant amount of anticoagulant treatment costs, by far outweighing the higher analytical cost. The strengths of our study are that we included consecutive patients evaluated for HIT in clinical practice. Thus, our results are applicable to real-life practice. In addition, we evaluated a number of important criteria that determine the utility in clinical practice in a joint assessment: accuracy, reproducibility, as well as costs. Furthermore, we directly compared commonly used immunoassays. Even more, we applied the costs of the Swiss health care system, providing recommendations for Swiss physicians and laboratory managers in particular. Our study has several important limitations. First, the number of observations and the number of true HIT cases is limited. This is reflected by wide confidence intervals and a sensitivity of 100%, which is higher than previously estimated. However, a high sensitivity of all assays is already known and our primary outcome was specificity. In addition, our results obtained in clinical practice confirm previous studies conducted in more rigorous settings and this is the most important message of the present investigation. Second, a relevant number of measurements are missing and some patients had to be excluded due to missing plasma samples (n = 3) or PAT results (n = 8). This might lead to selection of patients and affect the internal validity of the study. However, we focused on a high applicability to clinical practice, which is an important issue in diagnostic accuracy studies as well. In addition, exclusion of patients with inconclusive reference standard results is recommended by other authors. Third, PAT as reference standard is criticized because of a limited sensitivity. We conducted a sensitivity analysis to address this issue by applying a broad definition of HIT, considering all patients as HIT positive with an ELISA OD above 2.0 and/ or an PaGIA titre of at least 8 (Table C and D of). Sensitivity was more limited for all assays but still similar to previous publications with a comparable pattern between assays and thresholds. Finally, our analysis of cost was based on the Swiss health care system limiting the applicability to different countries. Moreover, we did not take the analyzer for determination of chemiluminescent immunoassay into account, assuming that this test will only be considered if a respective analyzer by Instrumentation Laboratory, is already implemented for other purposes. Another limitation is that we conducted chemiluminescent immunoassays retrospectively on stored samples. Thus, our investigation might not reflect inconsistencies of test determination in real- life practice. Our data suggest that rapid immunoassays have accuracy and reproducibility properties that are at least as good as an ELISA assay. Furthermore, they have the potential to relevantly improve patient care due to a shorter turn-around time. Not only that unnecessary switching to alternative anticoagulants will be avoided, but the risk of thromboembolic complications while waiting for laboratory results will be prevented. Even more, they offer the potential of relevantly saving treatment-related health care costs. The easy handling of PaGIA without the need for specialized knowledge or analyzers facilitates the implementation of a 24-hours-service even in small to middle-sized medical laboratories. In contrast, chemiluminescent immunoassays are fully automatable. We suggest applying intermediate thresholds at least for polyspecific assays in order to improve specificity. However, we do not recommend to implement an intermediate threshold for AcuStar HIT-IgG because sensitivity was limited in other studies. Our study was restricted on the presence of HIT as the only clinical outcome. Future prospective multicenter studies might focus on clinical outcomes such as thromboembolic, bleeding complications, and deaths as well. In addition, evaluation studies are needed in order to investigate the diagnostic accuracy and clinical outcomes of complete diagnostic algorithms comprising clinical scoring systems immunoassays as well. In conclusion, rapid immunoassays for diagnosis of HIT have favorable characteristics in terms of diagnostic accuracy and reproducibility in clinical practice. The implementation of 24-hour service using these tests might not only improve care in patients with suspected HIT but has the potential to save treatment-related health care costs. # Supporting information We thank all the residents in haematology who contributed to this work by gathering clinical information and calculating the 4Ts score. [^1]: MN received research funding from Instrumentation Laboratory. MN has received research grants, lecture fees or consultancy fees from Bayer, Roche diagnostics, Instrumentation Laboratory, Boehringer Ingelheim, and CSL Behring. MN currently serves as an Academic Editor for PLOS ONE. This does not alter our adherence to PLOS ONE policies on sharing data and materials. [^2]: **Conceptualization:** AB MN LA. **Data curation:** AB YA. **Formal analysis:** AB MN. **Funding acquisition:** MN. **Investigation:** YA. **Methodology:** MN. **Project administration:** MN. **Resources:** MN LA. **Software:** MN. **Supervision:** MN LA. **Validation:** AB YA MH LA MN. **Visualization:** MN. **Writing – original draft:** AB MN. **Writing – review & editing:** AB YA MH LA MN.

# Introduction The production of banana (*Musa* spp.), one of the most important fruit crops in the world, is seriously threatened by cold stress and pests such as *Fusarium oxysporum* var. *cubense*. Development of new banana cultivars with resistance to diseases or cold stress is one of the best ways to overcome this problem. However, it is extremely difficult to create such cultivars through conventional breeding because most commercial cultivars are sterile triploids bearing parthenocarpic fruits. Plant regeneration *via* somatic embryogenesis is the base of banana germplasm improvement using biotechnological techniques. Unfortunately, some important banana cultivars are recalcitrant in regard to the embryogenic response. Solution of this problem represents a major challenge for future studies aiming at improvement of the banana germplasm. Cell wall plays a very important role in the plant development. The chemical components of the cell walls are modulated during plant growth and development. Several previous studies have reported about developmental changes in cell wall components such as arabinogalactan proteins and pectins in some plant species such as maize (*Zea mays* L.), chicory (*Cichorium*), coconut (*Cocos nucifera* L.), barley (*Hordeum vulgare*) and olive (*Olea europaea* L.). However, to our knowledge, no study was devoted to cell wall pectins during somatic embryogenesis of banana. The walls of plant cells are primarily composed of cellulose, hemicellulose (e.g., xyloglucans, xylans, and mannans), pectins, and a small amount of structural proteins. Pectins, one major class of chemical components, make up to 35% of the primary cell walls in dicotyledonous plants and non-graminaceous (non-grass) monocots. Homogalacturonan (HG) is the most abundant pectin polysaccharide, making up to 65% of total pectin. The structural domains of pectin are built on more or less methyl- and acetyl-esterified galacturonan. One main characteristic of pectin is the extent of methyl-esterification on the carboxyl group of polygalacturonic acid. The degree of HG methyl-esterification has been reported as the key determinant of plant and organ development involving processes such as cell division, expansion, and adhesion. Furthermore, a minimum stretch of nine unmethylated galacturonic acid (GalA) residues can form Ca²⁺ linkages, which may promote the formation of so-called “egg-box” model structure. Hence, the methyl-esterification status of HG can have dramatic consequences on cell wall texture and mechanical properties, thereby contributing to cell shape and growth. Somatic embryogenesis is characterized by well-defined embryogenic stages, which are generally similar to those in zygotic embryogenesis. This process requires strict spatial and temporal control over cell division and elongation. In some plants, digestion of cell wall pectins by pectinase can result in complete or partial disappearance of the extracellular matrix (ECM) at the surface of embryogenic cells (ECs) and/or proembryos, thus leading to their collapse. These observations point to the importance of pectins for somatic embryogenesis and ECM structural integrity. Currently, immunohistochemical techniques using well characterized antibodies have been applied to better define plant cell wall components and to localize cell wall polymers *in situ* within complex tissues and organs. These techniques enable monitoring of structural changes, organization and partial changes of function in the plant cell wall. Indeed, the application of immunohistochemical technique by using monoclonal antibodies JIM5 and JIM7 led to the identification of pectic epitopes in the extracellular matrix surface network (ECMSN) of calli in chicory and kiwifruit (*Actinidia deliciosa*). The same epitopes were also detected during organogenesis from callus of wheat (*Triticum aestivum* L.), and also during microspore embryogenesis of *Citrus*, cork oak (*Quercus suber* L.), olive and *Capsicum annuum* L.,. With the help of the 2F4 antibody, Liners et al. monitored the distribution of pectic polysaccharides in cell walls of carrot (*Daucus carota* L.) suspension cells and sugar beet (*Beta vulgaris* L.) calli. Another immunohistochemical study described the changes of JIM5 and JIM7 epitopes during somatic embryogenesis of coconut. Ruthenium red is a cationic stain with six positive charges, which forms electrostatic bonds to the acidic groups of sugars, for example carboxyl groups and sulfate groups. Usually, it is used to study mucilage secreted by plant seeds. Mucilage is composed of complex acidic or neutral polysaccharide polymers of high molecular weight, mostly methyl-esterified HG, rhamnogalacturonan I, and low amount of galactans and arabinans. Ruthenium red was also successfully used to study amount of galacturonic acid (GalA) in carrot callus. In this study, ruthenium red along with an arsenal of pectin antibodies, including six recently developed antibodies specifically recognizing HG epitopes with varying degrees of methyl-esterification, were used to map the developmental distribution and regulation of these epitopes during banana somatic embryogenesis. Moreover, the two above mentioned independent localization techniques were complemented by biochemical and immunodot analyzes. # Results ## Histological examination of somatic embryogenesis in banana Previously, PAS has been used to detect glycogen in animal tissues and carbohydrates, proteoglycans and glycoproteins in plant tissues. It is based on the reaction of periodic acid which selectively oxidizes the glucose residues and creates aldehydes reacting with the Schiff reagent (product of fuchsine or pararosaniline and sodium bisulfite) and producing purple-magenta color. In the present study, PAS staining was used to study the histology of NECs of cultivar ‘Baxijiao’ used as a control as well as in ECs and somatic embryos of cultivar ‘Yueyoukang 1’. This histological analysis revealed different cellular organization in NECs versus ECs as well as all basic developmental embryogenic stages including ECs, pre-globular, globular, pear- shaped and cotyledonary somatic embryos of cultivar ‘Yueyoukang 1’. In more detail, very few small starch granules were found in the cytoplasm of NECs, especially around their nuclei. In comparison to NECs, ECs were smaller but they contained more and relatively bigger starch granules localized around their nuclei. Small ECs were surrounded by some bigger cells, which were strongly stained by PAS and contained the biggest starch granules. Therefore, these cells were termed “starch-rich cells” in this study. Starch content decreased in pre-globular and globular embryos while many starch granules re-appeared in the cells localized in lateral and basal parts of late-stage pear-shaped and cotyledonary embryos. The formation of vascular tissue was initiated in cotyledonary embryos. PAS staining revealed mucilage-like areas and layers, especially at the surface of NEC and EC clumps as well as at the outer surface of pre-globular, globular and pear-shaped embryos and parts of cotyledonary embryos containing starch-rich cells. ## Histochemical staining of surface-localized pectins with ruthenium red A surface treatment of the whole mounted fresh material (not fixed and not embedded and sectioned) with ruthenium red was used to label extracellularly secreted pectins in the mucilage covering both non-embryogenic and embryogenic cell clumps as well as somatic embryos. Overall, the intensity of pink/red stain on NECs of cv. ‘Baxijiao’ was much weaker than on ECs and somatic embryos of cv. ‘Yueyoukang 1’. In more detail, only few surface localized NECs and their clumps were stained by ruthenium red. In contrast, many more ECs and cell clumps were stained with ruthenium red and the pink/red color was much stronger in these ECs as compared to NECs. Further, outer surface of somatic embryos at pre- globular, globular and pear-shaped stages was the most strongly stained by ruthenium red. The pink/red color of ruthenium red became weaker at the surface of late stage cotyledonary embryos. Moreover, staining appeared as a net-like structure at the surface of these cotyledonary embryos. ## Immunodot analysis of pectins in nonembryogenic cells and during somatic embryogenesis of banana A representative set of six commercially available monoclonal antibodies (CCRC-M34, CCRC-M38, JIM5, JIM7, LM18 and LM20) has been reported to bind to HG epitopes showing different degrees of esterification. The fully de-esterified and low methyl-esterified HG were recognized by CCRC-M38, JIM5, LM18 and partially to highly methylesterified HG were recognized by JIM7, LM20 and CCRC-M34. These antibodies were tested and selected to study developmental distribution and methyl-esterification of these pectic epitopes in the cell walls of NECs, ECs and somatic embryos of banana by immunodot technique. Furthermore, the 2F4 antibody binding to de-esterified/calcium cross-linked HG and by LM5 antibody mainly binding to (1→4)-β-D-galactan of rhamnogalacturonan I were also used (RG-I). Additionally, negligible signals were found by using the LM7 (recognizing non-blockwise deesterified HG) and the LM19 antibodies (binding to methyl-esterified HG) in all developmental stages (data not shown). This immunodot analysis revealed some semi-quantitative differences in the abundance of diverse pectic epitopes in NECs showing higher abundance of LM5 and JIM5 epitopes, followed by lower abundance of the LM18 epitope. The abundance of other HG epitopes such as 2F4, CCRC-M34, CCRC-M38, JIM7 and LM20 was relatively low in NECs. The abundance of pectic epitopes in ECs was similar to that of NECs with exception of the 2F4 and LM5 epitopes. The 2F4 epitope was almost undetectable and the LM5 epitope was less abundant in ECs in comparison to NECs. These differences might be related either to non-embryogenic versus embryogenic character of cell cultures or to the different cultivars used in this study (cultivar ‘Baxijiao’ for NECs versus cultivar ‘Yueyoukang 1’ for ECs). In pre- globular/globular embryos, a relative abundance of highly methyl-esterified HG recognized by LM20, JIM7 and CCRC-M34 antibodies was obviously higher in comparison to ECs. In late stage cotyledonary embryos, pattern of pectin epitopes resembled previous developmental stage of pre-globular/globular embryos except for a decrease in the abundance of the CCRC-M34 epitope. These data suggested that several pectin epitopes were changed during progression of somatic embryogenesis in cv. ‘Yueyoukang 1’. ## Distribution and methyl-esterification of pectin epitopes in NECs The immunofluorescence labeling was used to study the presence of diverse pectic epitopes in cell walls. Possible steric hinderance was considered for pectic epitopes during *in situ* immunolabeling due to the possible molecular crowding and/or masking/blocking in intact cell walls. Therefore, both qualitative and quantitative statements about relative abundance of pectic epitopes (as visualized by immunofluorescence labeling) were avoided. Nevertheless, in situ immunolabeling revealed that low and/or fully de-esterified HG epitopes such as LM18, CCRC-M38 and JIM5 as well as in the LM5 epitope representing (1→4)-β-D- galactan were present in NECs of cultivar ‘Baxijiao’. In contrast, the 2F4 epitope of HG (de-esterified/calcium ion cross-linked HG) was detected only in few cell walls. When spatial distribution was taken into account, CCRC-M38 and LM5 epitopes were distributed all over the section in the cell walls of both cortical and inner cell types while the LM18 and JIM5 epitopes were located in the surface mucilage areas and layers covering clumps of NECs. On the other hand, the highly methyl-esterified pectic epitopes CCRC-M34, JIM7 and LM20 were poorly detected in the intact cell walls of NECs. Next, we tested distribution patterns of these pectic epitopes after chemical de-esterification with NaOH. After such base treatment, immunolabelings of the JIM5-, JIM7- and LM20 epitopes fully disappeared while they were slightly stronger for the 2F4 and the LM18 epitopes. Further, immunolabeling of the LM5 epitope slightly decreased but no obvious changes were detected for the CCRC-M38 epitope after the base treatment . ## Distribution and methyl-esterification of pectin epitopes in ECs We compared immunolabeling patterns of pectin epitopes between ECs of cultivar ‘Yueyoukang 1’ and NECs of cultivar ‘Baxijiao’. Immunolabeling of the 2F4 epitope was weak in ECs, as it was in NECs. The JIM18 and LM5 epitopes were immunolabeled all over the section, in the cell walls of both surface and inner cell types of ECs while the same epitopes appeared to be immunolabeled predominantly at the surface of NEC clumps ( B, D). Generally, immunolabelings of epitopes of highly methyl-esterified HG were relatively stronger in ECs in comparison to NECs. The labeling of the LM5 epitope was relatively stronger in the starch-rich cells but it was weaker in the typical ECs. The response of pectic epitopes in ECs to the base treatment was generally very similar to that in NECs. After de-esterification with NaOH, immunolabeling of the 2F4 and the LM18 epitopes appeared to be relatively stronger. On the other hand, immunolabelings of JIM5, CCRC-M34 and LM20 epitopes became depleted, while they were weaker for JIM7 and LM5 epitopes. ## Distribution and methyl-esterification of pectin epitopes in pre-globular and globular embryos In pre-globular and globular embryos, the 2F4 epitope was detected by immunolabeling in the ECM layer surrounding these embryos. The JIM5 epitope was also strongly immunolabeled in this ECM layer while it also appeared in the subepidermal cells of such embryos. The CCRC-M38 epitope was immunofluorescently detected and equally distributed all over the pre-globular embryo. A comparable result was obtained also for the LM20 epitope. The LM18, JIM7 and LM5 epitopes were detected by immunofluorescence labeling in epidermis and subepidermis while they were hardly or not detectable in inner cells of globular embryos. Finally, the CCRC-M34 epitope was nearly undetectable by immunofluorescence labeling. As in previous cases, base treatments caused disappearance of the JIM5 and LM20 immunolabelings. Further, immunolabelings of the JIM7 and LM5 epitopes were largely reduced. On the other hand, immunolabelings of the 2F4 and LM18 epitopes appeared to be stronger after base treatment. Finally, no obvious difference in the immunolabeling intensity/distribution was detected for the CCRC-M38 epitope after NaOH treatment ( cf. 6C). ## Distribution and methyl-esterification of pectin epitopes in late-stage embryos Both 2F4 and LM18 epitopes were immunolocalized besides surface-localized ECM also to the cortical cells of the late-stage embryos ( A, B). The CCRC-M38 epitope was localized to outer cortical cells and to the parenchyma cells in the middle part of embryo and the JIM5 epitope was localized especially in the cell- cell junctions of cortical cells. The CCRC-M34, JIM7 and LM20 epitopes were found in the epidermal cells and some inner parenchyma cells of late-stage embryos ( E–G). The immunolocalization of LM5 epitope was restricted to the cell-cell junctions of cortical cells in these embryos. Base de-esterification of HG caused severe depletion of the immunolabeling in the case of JIM5, CCRC-M34 and LM2 epitopes. Moreover, also labelings of JIM7 and LM5 epitopes seemed to be reduced after the NaOH treatment (Fig. 7NP). The immunolabelings of 2F4, LM18 and CCRC-M38 epitopes were not obviously affected by base treatment. Altogether, these immunolabeling results suggested a tight developmental regulation of several pectic epitopes and their differential sensitivity to the NaOH de-esterification treatment. For better overview, a summary of immunofluorescence labeling results in control and NaOH-treated samples is presented in the. ## The changes in the GalA content and pectin DM during somatic embryogenesis of banana Comparison of D-galacturonic acid (GalA) content in different developmental stages during banana somatic embryogenesis revealed that it was the highest in the late-stage embryos. The GalA content was significantly lower in NECs, ECs and pre-globular/globular proembryos, however, the differences among them were not significant. On the contrary, DM was the lowest in the late-stage embryos and the highest in NECs. The moderate levels of DM were detected in ECs and pre- globular/globular embryos. # Discussion Pectins are major plant cell wall polysaccharides representing up to 35% of the primary cell wall in dicotyledonous plants and non-graminaceous (non-grass) monocots while HG is the major pectin component. Biosynthesis and modifications of pectins, especially their methyl-esterification, have been proposed to be involved in plant development and cell adhesion. Moreover, biochemical properties of cell walls including content of galacturonan and uronic acid are variable according to taxonomy and distinctive cell walls occur in some groups of the monocotyledonous plant species. Banana taxonomically belongs to the genus *Musa*, family *Musaceae* and order *Zingiberales* of monocotyledonous flowering plants. It has been reported that banana pericarp tissue contained high levels of both uronic acid and glacturonan. Consistently with this report, we have found a relatively high content of the galacturonic acid in banana embryogenic cultures in this study. This indicates that some monocotyledonous plants contain pectic polysaccharides in the levels which are comparable to those in dicots. Here we report on the developmental histochemical (ruthenium red) and immunofluorescence (using antibodies against diverse HG epitopes and the (1→4)-β-D-galactan epitope of RG-I) localizations of pectins along with developmental study on pectin DM during somatic embryogenesis of banana. Generally, all immunofluorescence data on intact cell walls should be interpreted with caution because immunolabeling of some pectic epitopes in certain cell walls/cell types might be blocked due to the sterical hindrance of these epitopes in these walls during immunolabeling procedure. We tried to partially avoid this problem by using several antibodies recognizing diverse epitopes at the same molecule, namely at pectic HG. ## Developmental localization of pectins in banana embryogenic cultures Histochemical staining with ruthenium red revealed pink/red colored surface- localized pectins, especially at the surface of banana embryos at pre-globular, globular and pear-shaped developmental stages. This was in good correlation with immunolocalization patterns of surface-localized pectin epitopes (such as LM18, JIM5, JIM7 and LM20) at the same embryogenic developmental stages. Roughly, it was also consistent with immunodot labeling of these pectin epitopes during embryo development. Further, immunodot analysis correlated to immunolabeling with pectin antibodies suggested that banana embryogenic cultures were relatively rich in the JIM5 and LM5 epitopes, but poor in the 2F4 and CCRC-M34 epitopes. Among pectic epitopes, especially the highly methyl-esterified HG ones, recognized by LM20 and JIM7 antibodies, appeared to be developmentally regulated. Thus, both immunodot and immunofluorescence labelings of these two epitopes were weaker in embryogenic cells but stronger in pre-globular/globular and late embryos during embryo development. The immunolabeling intensities of other pectic epitopes on immunodots and immunofluorescence-labeled tissue sections also varied during embryo development, though the differences between diverse developmental stages were not as obvious as for LM20 and JIM7 epitopes. This study revealed also some differences in pectins between nonembryogenic and embryogenic cultures of banana. For example, the LM5 epitope seemed to be more abundant in NECs as in ECs when immunodot and immunofluorescence labelings were compared to each other. This might be related to the fact that this epitope was mainly immunolocalized to the small groups of starch-rich cells in EC aggregates whereas it was detected in all cells in NEC aggregates. Moreover, clusters of NECs were covered by extracellular layer containing both LM18 and JIM5 epitopes. This might suggest extracellular secretion and local accumulation of these pectic epitopes in this ECM surface layer. On the other hand, the same epitopes were localized solely to the cell walls of individual ECs, suggesting that they were not secreted towards ECM surrounding ECs. Some of these results are in good agreement with previously published reports. For instance, larger intercellular spaces and the cell wall junctions contain JIM5 epitope in kiwifruit endosperm- derived callus. The same epitope was found to be widely distributed over the middle lamella between adjacent meristematic cells of coconut calli while the JIM7 epitope was not so abundant and it was differentially distributed in cell walls. Iwai et al. reported that non-methyl-esterified pectins were abundant in carrot NECs while methyl-esterified ones were mostly detected in ECs of carrot. During development of globular somatic embryos of banana, the LM18 and JIM5 epitopes were detected in the outer surface of their epidermal cells. This was also the case for the LM18 but not for the JIM5 epitopes in the late-stage embryos. Moreover, globular and late-stage embryos possessed at their outer surfaces the 2F4 epitope, especially after NaOH treatment. Previously, it was reported for monocot maize that highly methyl-esterified pectins recognized by JIM7 antibody were localized to the ECM covering embryogenic cell clumps of embryogenic callus. Moreover, the JIM7 epitope was abundant in the outer wall of surface cells and the continuous layer over the cells in meristematic tissue of wheat representing another monocot species. In banana, methyl-esterified JIM7 and JIM20 epitopes of HG were detected in epidermal/subepidermal cells of pre-globular and globular embryos and later also in the inner parenchyma cells of late-stage embryos. ## Is the 2F4 epitope of HG involved in cell-cell adhesion of banana ECs? Formerly, acidic (low methyl-esterified) pectins were thought to play a crucial role in cell-cell adhesion. A minimum stretch of nine de-esterified GalA residues can form Ca²⁺ linkages, thus promoting the formation of so- called “egg-box” model structure, which may eventually strengthen the cell wall. The corresponding 2F4 epitope, however, likely does not play a major role in cell-cell adhesion in banana embryogenic cultures because it was weakly detected on both immunodots and tissue sections of tightly adhered ECs while it was more present in less-adhered NECs. In fact, this is in good agreement with published results that “egg-box” structures with characteristic calcium bridges were more abundant in carrot NECs as in ECs. Additionally, the 2F4 epitope was not present in shoot apical meristem of *Sinapis alba* which is composed of tightly adhered cells. Moreover, pectins in monocot wheat are suggested to play only a minor role in cell-cell adhesion. Thus, it seems to be plausible that mechanism of intercellular attachment and cementing of cells is determined also by fine tuning of methyl-esterification degree of acidic pectins, as well as by highly methyl-esterified pectins in diverse plant species. ## Changes in degree of pectin methyl-esterification during somatic embryogenesis of banana Modifications of HG such as methyl-esterification are important for cell fate determination and plant development. Consequently, different levels of pectin methyl-esterification, as revealed by immunolabeling with JIM5 and JIM7 antibodies, have been reported for several plant species during somatic embryogenesis. In chicory, the JIM5 epitope was localized to the outer part of protodermal embryo cells and to the intercellular spaces while the JIM7 epitope was much less abundant during somatic embryogenesis. Immunogold labeling with the JIM5 antibody revealed a high amount of low methyl-esterified pectins in the outer cell walls of *Citrus* proembryos. Similarly, pectins recognized by the JIM5 antibody were rich in the peripheral wall below the exine and in the dividing inner cell walls of olive microspore-derived proembryos as well as in the cells of cork oak proembryos. In most of these studies immunolabeling of pectins with JIM7 antibody was very weak. Our immunodot and immunolocalization data with JIM5 and JIM7 antibodies on banana embryogenic cells and somatic embryos are consistent with above reports. In addition, we have used a much broader set of monoclonal antibodies binding to pectin HG epitopes with different degree of methyl-esterification from fully de-esterified (CCRC-M38) to highly methyl-esterified (LM20) in this study. This mapping of diverse pectic epitopes on immunodots and tissue sections suggests that fully de-esterified and low methyl-esterified ones such as CCRC-M38, LM18 and JIM5 were abundant in banana ECs while highly methyl-esterified ones such as CCRC-M34, JIM7 and LM20 antibodies were hardly detected. After ECs developed into embryos, the highly methyl-esterified pectins became more abundant. ## Conclusions and future prospects This study provides new information about the developmental localization of cell wall pectins and about methyl-esterification patterns of pectin HG epitopes during banana somatic embryogenesis. The main conclusions are: 1\. De-esterified and low methyl-esterified HG epitopes were detected in the surface localized ECM surrounding banana ECs and embryos. 2\. Partially and highly methyl-esterified HG epitopes were more abundant in the cell walls of pre-globular and globular embryos. 3\. De-esterification of pectins with base treatment caused imunolabeling depletion of highly methyl-esterified HG epitopes but increased immunolabeling of low methyl-esterified ones. 4\. The data from the present study support the hypothesis that the mechanism of intercellular attachment and cementing of cells is likely determined not only by calcium pectate gels but also by certain acidic de-esterified pectins. These results could be potentially valuable for development of future strategies aiming to improve the regeneration capacity of ECs in banana. This can be done, for example, by manipulation of pectin modifying enzymes such as pectin esterases and pectate lyases, as it was recently proposed for cyclamen somatic embryogenesis. In the future, genetic manipulation of pectin modifying enzymes combined with mapping of pectin epitopes in transgenic plants might prove to be useful for biotechnological applications using somatic embryogenesis in banana. # Materials and Methods ## Plant material Embryogenic cells (ECs) of banana cultivar Yueyoukang 1 (*Musa* spp. AAA) as well as nonembryogenic cells (NECs) of cultivar Baxijiao (*Musa* spp. AAA) were cultured in ZZl medium, which is half-strength MS medium supplemented with 1.1 mg/L 2, 4-dichlorophenoxyacetic acid (2, 4-D), 0.23 mg/L zeatin and 10 mg/L ascorbic acid. The pH of medium was adjusted to 6.0 before autoclaving. The cultures were incubated at 28±2°C under light on a reciprocal shaker at 90 rpm, and sub-cultured at 7 d interval. Seven days after the last subculture, ECs were inoculated on RD1 embryo-regeneration medium. The cultures were incubated in the dark at 25±1°C and 29±1°C to promote development of somatic embryos. ## Histological staining of tissue sections with periodic acid-Schiff (PAS) NECs, ECs (both at seven days after the last subculture), pre-globular/globular embryos (3-week-old regenerated material from the cultures incubated at 25±1°C) and late embryos (5-week-old regenerated material from the cultures incubated at 29±1°C) were collected and fixed with 3.7% (v/v) formaldehyde in stabilizing buffer MTSB \[50 mM piperazine--N,N'-bis(2-ethanesulfonic acid (PIPES), 5 mM MgSO₄.7H₂O, 5 mM ethylene glycol-bis(2-aminoethylether)- *N,N,N',N'*-tetraacetic acid (EGTA), pH 6.9\] at room temperature for 1 h. After washing in MTSB and phosphate-buffered saline (PBS) (pH 6.9), samples were dehydrated in graded ethanol series diluted in PBS and infiltrated with Steedman's wax according to previous studies. Thin sections (8 µm) were mounted on microscopic slides coated with 0.2% polyethylenimine, de-waxed in absolute ethanol, re-hydrated in ethanol/PBS series and washed in PBS. Histological sections of NECs, ECs and somatic embryos were stained with PAS according to manufacturer instructions (Sigma). ## Staining of fresh material with ruthenium red To test surface-localized pectins, suggesting an extracellular transport and accumulation of these pectins in mucilage-like ECMSN around NECs, ECs and somatic embryos at different developmental stages, the whole mount samples (without any fixation, embedding and sectioning) were collected at various developmental stages (see above) and incubated in 0.01% (w/v) ruthenium red by shaking (120 rpm) at 30°C for 2 h. Histochemical staining of surface pectins appeared as pink/red color and it was evaluated and documented under Leica binocular microscope (Leica, Germany). ## Antibodies and chemicals All antibodies used in this study and their corresponding epitopes/antigens are presented in. These antibodies were obtained from PlantProbes (UK), except for CCRM antibodies which were obtained from Complex Carbohydrate Research Center (Athens, USA). All chemicals were of analytical reagent grade and were obtained from Sigma (St. Louis, USA) unless indicated otherwise. ## Immunofluorescence labeling Immunolabeling was carried out exactly as described by Xu et al., except for 2F4 antibody. For this antibody, the buffer containing 1 mM CaCl2 in 50 mM PIPES (pH 7.4) was used for sample fixation. Additionally, T/Ca/S buffer (20 mM Tris-HCl, 0.5 mM CaCl2, 150 mM NaCl, pH 8.2) was used instead of PBS before labeling with the secondary antibody during immunolabeling procedure. The secondary antibody for 2F4, CCRC-M34 and CCRC-M38 primary antibodies was anti-mouse IgG-FITC (F9006, Sigma), and for all JIM and LM antibodies it was anti-rat IgG-FITC (F6258, Sigma), respectively. Sections incubated only with secondary antibodies were used as negative controls. A minimum three slides were used for each antibody at each developmental stage. Fluorescence was examined and documented with an Olympus BH-2-FRCA microscope. ## Preparation of alcohol-insoluble residue (AIR) AIR was prepared according to the method described by Louvet et al.. In brief, samples were washed three times with 70% ethanol (v/v) at 70°C for 30 min after homogenization. The supernatant was removed after centrifugation. The pellet was crushed in liquid nitrogen and freeze-dried. ## Immunodot assay Pectin was extracted from AIR with 0.5% (w/v) ammonium oxalate buffer at 100°C and the concentration was adjusted to 1 mg/ml. Samples were spotted (as 5 µl drops) onto nitrocellulose membrane by a micropipette. The membrane with dots was air dried at room temperature for 1 h. Assays with JIM5, JIM7, LM5, LM18, JIM20, CCRC-M34 and CCRC-M38 antibodies were carried out as described by Willats et al.. T/Ca/S buffer was used to replace PBS for the assay with 2F4 antibody. After the final wash, the membrane was developed in 3, 3′-diaminobenzidine tetrahydrochloride (DAB) kit from TCI (Shanghai) Development Co., Ltd. ## Measurement of pectin methyl-esterification degree (DM) with colorimetry The DM was calculated as moles of methanol per mol of GalA. The methanol assay was adopted from the methods described by Klavons and Bennett and Louvet et al.. In detail, 5 mg of AIR was saponified in 2 ml of 0.25 M KOH at room temperature for 1 hr, then neutralized with phosphoric acid (to pH 7.5). After centrifugation at 10.000 g for 10 min, aliquots of the supernatant (1 ml) were loaded into a 15 ml tube. Alcohol oxidase (1 ml, 1 U/ml, diluted in distilled water, Sigma) was added to each tube. After gently mixing, the tube was incubated at room temperature for 20 min. Thereafter, 2 ml of a mixture containing 0.1% 2,4-pentanedione in 1 M ammonium acetate and 0.14% acetic acid was added. Following 15 min of incubation at 60°C, samples were cooled on ice and absorbance was measured at 420 nm (HAITACHI, U-2900). Methanol in potassium phosphate buffer (pH 7.5) (0–20 µg/ml range) was used as a standard. GalA content of the sample was determined colorimetrically by the metahydroxydiphenyl assay adopted from Blumenkrantz and Asboe-Hansen and from Louvet et al.. In detail, 5 mg of AIR was hydrolyzed in 125 µl of 13 M sulfuric acid at room temperature for 30 min. A second hydrolysis was performed. The sample was diluted 5 times with distilled water and saponified at a final concentration of 1 M NaOH at 100°C for 2 h. The supernatant was adjusted to pH 8 with NaOH followed by neutralization with HCl. After centrifugation at 10.000 g for 10 min, aliquots of the supernatant (0.5 ml) were loaded into a 15 ml tube. Pre-cooled samples (on ice-water bath) were supplemented with 1.5 ml of pre- cooled 0.025 M sodium tetraborate buffer in concentrated sulfuric acid and incubated at 100°C for 5 min. After cooling in a water-ice bath, 25 µl of 0.15% MHDP in 0.5% NaOH was added to the samples. Following 10 min incubation at room temperature, absorbance was recorded at 520 nm. GalA (0–100 mg/l range) was used as a standard. Three replicates were made for each treatment and each experiment was repeated twice. GalA was calculated as micrograms of GalA per gram of AIR. ## Chemical de-esterification The extensive degradation of pectic polymers was performed by base catalyzed de- esterification. Samples were prepared as described above for immunofluorescence labeling. After dewaxing, sample sections were treated with 0.05 M NaOH at 4°C for 30 min followed by modified immunolabeling protocol (washing with buffer two times for 10 min, blocking with glycine). [^1]: Conceived and designed the experiments: JS CX. Performed the experiments: CX LZ XP. Analyzed the data: JS CX. Contributed reagents/materials/analysis tools: JS CX. Wrote the paper: JS CX. [^2]: The authors have declared that no competing interests exist.

# Introduction To address the human and financial impact of obesity and related chronic diseases like type 2 diabetes, countries across the world are advocating the promotion of healthy lifestyles and positive lifestyle change. This is demonstrated through the publication of national guidelines, such as those which advocate for healthier diets and recommend 30 minutes of moderate to vigorous physical activity on five or more days a week. Such guidelines are based upon evidence demonstrating the association between particular health outcomes and lifestyles such as tobacco smoking, alcohol consumption, physical activity, and dietary habits. Some commentators have suggested, however, that the translation of this research into health policy has tended to result in ‘siloed’ strategies that attempt to modify one unhealthy lifestyle at a time. Such an approach may be inefficient, since an increasing number of studies in the UK, Belgium, Finland, the Netherlands, the US, New Zealand and China, report that unhealthy lifestyles tend to co-occur non-randomly among the same individuals. Therefore, interventions which tackle multiple unhealthy lifestyles simultaneously may be more appropriate, as has been argued in the case of diabetes prevention. With some notable exceptions, however, previous work which has sought to identify the social determinants of multiple unhealthy lifestyles has focused upon individual characteristics (especially socioeconomic factors) and hitherto paid little attention to the role of neighborhood characteristics (such as affluence or geographical remoteness). This is an important gap to address, since the places in which people live have long been used as targets for experiments and policy interventions. There is increasing widespread belief among policy makers and academics that neighborhoods can influence health outcomes independent of characteristics operating at the individual-level, but less is known on the extent that neighborhoods also determine the co-occurrence of unhealthy lifestyles. Although individual-level factors (e.g. income) are important correlates of multiple unhealthy lifestyles, it is possible that neighborhood characteristics, such as socioeconomic circumstances, could amplify (in the case of deprived neighborhoods) or buffer (in affluent areas) the impact of those individual factors. Addressing this hypothesis was the aim of this study. # Methods ## Data Our analyses focused on the 45 and Up Study. Between 2006 and 2008, participants were randomly selected from the Medicare Australia database (the national provider of universal health insurance in Australia) and self-completed a survey on lifestyle, health status and socioeconomic circumstances. Response to the survey is estimated at 18%, though previous work has shown that the results relating to relative risks from the 45 and Up Study are similar to those from a representative population health survey. All participants were resident in New South Wales (NSW), the most populous state in Australia. The University of New South Wales Human Research Ethics Committee approved The 45 and Up Study. ## Outcome variable: the ‘unhealthy lifestyle index’ Previous work has tended to construct outcome variables by summing binary indicators of unhealthy lifestyles. We took a similar approach using eight measures of unhealthy lifestyles available within the 45 and Up Study. These variables were selected based upon published national guidelines for tobacco smoking cessation, alcohol consumption, moderate to vigorous physical activity and a range of dietary indicators. The data and refinement of these variables was as follows: ### 1) Tobacco smoking Current smoking status was derived from affirmative responses to the question *“Are you a regular smoker now?”* For participants reporting a history of smoking, those who had smoked within the past year were classified as current smokers (coded as 1), whereas those who had not smoked within the last 12 months were classified as non-smokers (coded as 0). ### 2) Alcohol consumption Participants were asked *“how many alcoholic drinks do you have each week?” and “on how many days each week do you usually drink alcohol?”* These variables were used to identify the approximate number of alcoholic drinks consumed each day. A binary variable was constructed to distinguish between participants consuming less than (coded as 0), or at least two alcoholic drinks a day (coded as 1). ### 3) Physical activity The Active Australia Survey was used to ascertain the number of minutes spent in moderate to vigorous physical activities (MVPA) each week. Previous work has demonstrated this survey to have a satisfactory level of test-retest reliability. In line with national guidelines, participants 30 minutes of MVPA on five or more days a week (coded as 0) were differentiated from those who did not achieve this level of MVPA (coded as 1). Participants who met the guideline of 2.5 hours of MVPA a week, but not spread over 5 or more days, were classified as not meeting the national guideline (coded as 1) which is explicit in recommending regular, rather than concentrated participation. ### 4) Fruit consumption Participants responding two or more (coded as 0) to *“about how many serves of fruit do you usually have each day?”* were distinguished from those consuming less (coded as 1). Fruit juice was measured separately in the survey and not considered appropriate for this study as it would be impossible to differentiate between nutrient rich fresh juice and that from concentrate which is often high in sugar content. ### 5) Vegetable consumption Responses to the question *“about how many serves of vegetables do you usually eat each day?”* were counts stratified by cooked and raw varieties. We summed both responses and differentiated participants eating at least five portions of vegetables per day (coded as 0) from those eating fewer than five (coded as 1). ### 6) Consumption of red meat and processed meat Participants indicated the number of meat products eaten each week by the type of meat. Guidelines on consumption also differ by the type of meat. We coded participants as 1 if they ate between 3 and 5 weekly portions of red meat (beef, lamb or pork), or zero weekly portions of processed meat (bacon, sausages, salami, burgers). Participants were coded as 0 if they ate fewer than 3 or more than 5 weekly portions of red meat or 1+ portion of processed meat. ### 7) Low-fat milk Participants were asked to indicate which type of milk they consumed most of the time. Those who drank reduced fat milk or skim milk (coded as 0) were differentiated from other participants who either drank no milk, whole milk, or another variety (coded as 1). ### 8) Fish The number of portions of fish (or other seafood) eaten weekly were indicated as a count. Participants eating three or more portions of fish (coded as 0) were differentiated from those eating fewer portions per week (coded as 1). Summing the responses of each of the aforementioned binary variables gave a score ranging from zero to eight co-occurring unhealthy lifestyles. This variable is referred to hereafter as the *‘unhealthy lifestyle index’.* ## Health status and other individual-level measures Previous studies have demonstrated association between the co-occurrence of unhealthy lifestyles and health status. To perform a similar validation, we utilized three indicators available in the 45 and Up Study. General health and life quality were both self-reported and scored from 1 to 5: excellent, very good, good, fair, poor. Binary variables were constructed by aggregating responses excellent through to fair, leaving participants reporting poor health or quality of life as separate categories. A third health variable pertained to mental health, as measured by the Kessler Psychological Distress Scale (K10). The K10 is a widely used instrument comprising 10 questions on whether a person felt tired for no reason, nervous, hopeless, restless, depressed, sad or worthless during the last four weeks. Scores for each of the 10 questions ranged from 1 (“none of the time”) to 5 (“all of the time”). When each of the scores are summed, participants with aggregate of 22 are identified as being at a high risk of psychological distress. Other individual-level characteristics which have been previously reported as being associated with multiple unhealthy lifestyle indices were also included as control variables. These included age, gender, annual income, education qualification, economic status (**employed, unemployed, retired, inactive due to long term illness or disability**), couple status and country of birth. ## Neighborhood-level measures To define neighborhoods, this study used Census Collection Districts (CCDs) which have a mean of 225 residents and were the smallest geographical scale at which 2006 Census data was disseminated. We focused on the level of affluence and geographic remoteness of neighborhood environments as widely used indicators were available. Local affluence was measured using the Socio-Economic Index for Areas (SEIFA) ‘Index of Relative Socio-Economic Advantage/Disadvantage’. This variable was initially in rank format, so it was re-expressed in percentiles; higher percentiles indicated more affluent neighborhoods. Geographical remoteness was measured using the ‘Accessibility/Remoteness Index of Australia’ (ARIA). ARIA is a score ranging from 0 to 15, with scores of 2.4 and over used to distinguish between urban and inner regions (\<2.4) and rural or remote areas (\> = 2.4). ## Sample A sample of 206,457 participants with complete data on unhealthy lifestyles (smoking status, alcohol consumption, physical activity and dietary measures) and health status (self-rated health, self-rated quality of life, and risk of psychological distress) were selected from 267,151 in the 45 and Up Study. We imputed the gender-specific mean to address missing data for continuous independent variables (a ‘missing’ category was used for categorical variables). The most substantive missing outcome was for the number of minutes spent in MVPA (n = 22,136, 8.3% of the sample). Persons missing any of the outcome variables were more likely to be older and less educated, not employed or in a couple, on lower incomes and living in more deprived neighborhoods. No substantive differences in missing outcome data were found by gender or country of birth. ## Statistical analysis The distribution of the unhealthy lifestyle index across the sample was assessed using percentages. For each of the 9 lifestyle clusters (0 to 8 inclusive), the percentage response of each individual lifestyle was calculated and graphed to examine levels of co-occurrence. To assess the extent of correlation between the unhealthy lifestyle index and health status, multilevel binary logistic regression was used to fit associations with self-rated health, quality of life and psychological distress as outcome variables. In each of these models, the unhealthy lifestyle index was initially fitted as continuous variable, but was then substituted for a categorical version to test for non-linear relationships. Those models controlled for gender, age, education, income, economic status, couple status, country of birth, local affluence and geographical remoteness. The coefficients and 95% confidence intervals were exponentiated to odds ratios. We then proceeded to investigate the distribution of the unhealthy lifestyle index across demographic, socioeconomic and neighborhood characteristics. The unhealthy lifestyle index was normally distributed, which afforded the application of multilevel linear regression to fit associations with each of the explanatory variables. A multilevel framework was used to disentangle associations between the unhealthy lifestyle index and factors operating at different levels of analysis; persons at level 1 nested within Census Collection Districts (neighborhoods) at level 2. The initial step in the model building strategy involved fitting a ‘null’ model (i.e. with no independent variables) to calculate the intra-class correlation coefficient (ICC). The ICC in the case of our model indicated that 1.4% of the amount of variation in the unhealthy lifestyle index could be attributed to neighborhoods. Following this, the next steps were to add in individual- and neighborhood-level characteristics sequentially, noting the magnitude and direction coefficients and to what extent they were statistically significant using 95% confidence intervals. Interaction terms were fitted to explore for gender differences by age, and cross-level interactions between individual- and neighborhood-level characteristics (local affluence and geographic remoteness). In particular, a focus was on the potential interaction between individual- and neighborhood- level socioeconomic circumstances. Statistically significant associations were identified by using the log-likelihood ratio test (*p*\<0.05). All data manipulation and analyses in this study were conducted using STATA V.12 (StataCorp, College Station, TX, USA) in 2013. # Results provides descriptive information on the unhealthy lifestyle index and its components. 1.5% of the sample scored zero on the unhealthy lifestyle index, whereas only 0.2% reported all eight unhealthy lifestyles. Nearly 50% of the sample reported 3 or 4 unhealthy lifestyles. The unhealthy lifestyle index followed a ‘normal’ distribution. Among people who reported up to four unhealthy lifestyles, the most common of these were not eating enough fish, followed by not meeting guidelines on vegetable consumption and moderate to vigorous physical activity. Smoking, drinking too much alcohol and a processed or red meat intensive diet only tended to be more prevalent among people who reported many other unhealthy lifestyles (i.e. scores over 4 on the unhealthy lifestyle index). We expected that unhealthier lifestyles would be associated with poorer health, a lower quality of life and a higher risk of psychological distress. confirmed these expectations. Compared to people who scored zero on the unhealthy lifestyle index (i.e. the ‘healthiest’ group), for example, those who scored 8 (the ‘unhealthiest’ group) were 7 times more likely to rate their health as poor (95%CI 3.6, 13.7), 5 times more likely to report poor quality of life (95%CI 2.6, 10.1), and had a 2.6 times greater risk of psychological distress (95%CI 1.8, 3.7). In general, men scored higher on the unhealthy lifestyle index than women irrespective of age. This was despite lower (i.e. healthier) mean scores among older men. A similar improvement in lifestyle was observed for women between age 45 and 74, but from 75 onwards, the mean score on the unhealthy lifestyle index increased. The combination of decreasing scores among men and a parabolic trend for women meant the gender gap in the unhealthy lifestyle index at age 45 diminished substantially into older age. reports that people who lived in more affluent areas tended to score lower on the unhealthy lifestyle index, regardless of income. However, people earning less than \$20,000 a year and living in the most affluent areas scored very similar on the unhealthy life index with those on higher incomes living in more deprived neighborhoods. In contrast, people with moderate incomes living in the most affluent areas scored significantly lower on the unhealthy lifestyle index than people living in deprived neighborhoods on incomes above \$70,000 per year. People living in more rural and remote neighborhoods scored slightly higher on the unhealthy lifestyle index than their peers in urban areas (Coefficient: 0.02, *p* = 0.045). # Discussion ## Main finding In line with previous studies, the co-occurrence of unhealthy lifestyles was associated with an increased risk of poor self-rated health, quality of life, and a high risk of psychological distress. Previous reports on the co-occurrence of unhealthy lifestyles have tended to focus on the characteristics of individuals to demonstrate social patterning. The *spatial* patterning, by contrast, has received substantively less attention. The main finding from our study was that the socioeconomic context in which people reside does have an influence on unhealthy lifestyle co-occurrence, over and above the impact of characteristics at the level of the individual. A higher income was more beneficial overall, but among people with the same level of income, the co- occurrence of unhealthy lifestyles was lower if they were resident in more affluent neighborhoods. Where a person lived appeared to matter most if their income was less than \$20,000 per annum; those on low incomes in the poorest neighborhoods having highest mean number of multiple unhealthy lifestyles overall. For policy makers, this suggests that people living in poorer neighborhoods are a high risk group for the co-occurrence of unhealthy lifestyles; even if they are simultaneously on relatively high incomes. In the same vein, our findings also imply that affluent neighborhood circumstances may support healthier lifestyles even among those on low incomes. As such, future research should look to evaluate the trade-offs of investing in multiple unhealthy lifestyle interventions at the population-level versus those which target specific geographical areas, such as deprived neighborhoods. ## Interpretation The associations between neighborhood socioeconomic circumstances and co- occurrence of unhealthy lifestyles reported in this paper using Australian data are broadly similar to previous work in Europe. This is the first to address the issue of unhealthy lifestyle co-occurrence in Australia and the consistency of the neighborhood effects across international boundaries is reassuring. We also tested a variable which described the rurality and remoteness of the neighborhood. The slightly higher risk among those in more rural and remote circumstances relative to their counterparts in urban areas was significant, though small in comparison to that attributed to local socioeconomic circumstances. For policy makers tasked with implementing multiple lifestyle interventions, the key message based on these results is that deprived neighborhoods could be the focus of the efforts regardless of whether they are in urban or rural areas. Previous work has reported gender and age differences in the co-occurrence of unhealthy lifestyles, but there has been no report of gender differences narrowing with age, or the suggestion of a parabolic distribution among women of 45 years and older. This is an intriguing result which could be interpreted in different ways. If we are to take these trends as reflective of lifecourse trajectories, then it is simultaneously good and bad news. It is positive that unhealthy lifestyles among men appear to decline with age, but also deeply concerning that unhealthy lifestyles co-occur more often among older women. The lifecourse trajectory explanation for this patterning of unhealthy lifestyles by gender and age can only be speculative, however, given the data are cross- sectional. It would require follow-up of this sample over time to confirm this hypothesis. In the absence of this data, it is difficult to discount other possible explanations. What appears to be a decline in unhealthy lifestyle co- occurrence among older men may be an artifact of higher mortality rates among those males who, prior to death, would have reported a fairly high co-occurrence of unhealthy lifestyles. To test this hypothesis, longitudinal data with linked mortality records would be required. Nevertheless, the survival hypothesis would not appear to have a strong *prima facie* case to explain the parabolic trend among women. A third possibility, therefore, is potential for cohort effects, in which these age and gender differences are the product of early life experiences for people growing up in different periods of time. This may result in systematic differences in the co-occurrence of unhealthy lifestyles according to (unmeasured) variables which are correlated with age, such as health literacy. Unfortunately, no data on health literacy was available in the 45 and Up Study to test this hypothesis. The patterning of unhealthy lifestyle co-occurrence by age and gender among middle-to-older age adults is, therefore, an important avenue for further exploration with longitudinal and linked data that is beyond the scope of this cross-sectional study. ## Strengths and limitations The emphasis on people aged 45 and older was inherent within our study design, which means that our results are unlikely to reflect the situations of those under 45 years old. Although this is a limitation, the focus on middle-to-older age is under-researched in the context of investigations into multiple unhealthy lifestyles and can therefore also be interpreted as a strength. Further strengths include the large sample size and also the number of lifestyle measures, which included the consumption of fish, milk, red and processed meat, which tend not to be included in other studies that focus only on smoking status, physical activity, the consumption of alcohol and intake of fruit and vegetables. Although the 45 and Up Study was not designed to be nationally representative, previous work has demonstrated that relative risk estimates are broadly comparable to a representative population survey. While ethnic differences in health and lifestyle are known, our study did not explicitly investigate to what extent unhealthy lifestyles clustered by ethnicity. By definition, the study also did not explore variations between Indigenous and non-Indigenous Australians and how these play out spatially; this marks another avenue for future exploration. Another area for further investigation is the question of what it is about deprived neighborhoods that increases the risk of unhealthy lifestyle clustering, such as a potential lack of access to green spaces, or other opportunity structures. Finally, a reliance on cross-sectional data prohibits causal inference, though longitudinal analyses will be possible to test the robustness of the associations in this paper when the follow-up wave of the 45 and Up Study becomes available. # Conclusion Previous work on the determinants of co-occurrence in unhealthy lifestyles has tended to focus mainly upon the characteristics of individuals. The results of this study suggest that the socioeconomic circumstances of where a person lives could have an impact on lifestyle co-occurrence which is independent of their individual characteristics. Where a person lives appears to have a more substantial influence if they are on a low income, yet even people on higher incomes tend to have unhealthier lifestyles if they also live in poor neighborhoods. The key message for policy makers is that unhealthy lifestyles co-occur more strongly among residents of deprived neighborhoods, regardless of their individual demographic and socioeconomic circumstances. Future research, therefore, should investigate the trade-offs of a population-level approach towards intervening on multiple unhealthy lifestyles versus one which targets resources towards specific geographical areas. We thank all of the men and women who participated in the 45 and Up Study. The 45 and Up Study is managed by the Sax Institute in collaboration with major partner Cancer Council New South Wales; and partners the Heart Foundation (NSW Division); NSW Ministry of Health; *beyondblue: the national depression and anxiety initiative*; Ageing, Disability and Home Care, NSW Family and Community Services; UnitingCare Ageing and the Australian Red Cross Blood Service. We acknowledge the use of the 2006 census and boundary data provided by the Australian Bureau of Statistics. To preserve the anonymity of participants in the 45 and Up Study, some parameters of the Census Collector District (CCD) level data cannot be reported. This location-indexing data from the 45 and up Study are of highly restricted access and are made available only through SURE (<https://www.sure.org.au/>). [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: XF. Performed the experiments: XF TAB. Analyzed the data: XF. Contributed reagents/materials/analysis tools: XF TAB. Wrote the paper: XF TAB.

# Introduction Global energy consumption has increased exponentially since the Industrial Revolution, with a concomitant increase in the network of electrical transmission line across the landscape. The development of this network has had effects on animals which are sometimes favorable (e.g. increasing the availability of places for bird nesting on the pylons) and sometimes not (e.g. increased risk of electrocution for some birds, mammals and reptiles and collisions by birds). Electrocution of birds has been reported to be an important mortality factor. There has been increasing awareness of this problem amongst land managers and the public at large, as well as increased research into the distribution of electrocution events and likely mitigation measures that might be adopted. Mortality caused by electrocution due to poorly-designed pylons and power lines can seriously affect avian species particularly when population size of the affected species is low or its distribution is limited. Under this situation, avian power line mortality becomes a likely cause of population decline, especially when combined with other causes of human-induced mortality such as shooting, poisoning, trapping, collision with human-made objects (e.g. wind turbines), habitat destruction or exposure to environmental contaminants. Potential impact of electrocution and collision with electricity infrastructure is particularly significant for Falconiformes, because their morphology and behaviour make them more prone to electrocution, but Ciconiformes, Strigiformes and Passeriformes may also be affected. Although the problem is clearly identified and some potential solutions have been put forward, land managers still need to convince the public, and especially politicians, about the necessity of adequate investment in conservation programs (including mitigation) aimed at reducing man-induced mortality risks. Electrocution risks are influenced by many factors that can be divided into two groups: landscape factors and individual factors. The former comprise vegetation structure and composition, landscape topography, prey density and perch availability. Individual factors include pole-top configuration, clearances among electrical components, raptor morphology and raptor behaviour. For example, inexperienced immature and subadult birds, as well as females (larger than males in species with reversed size dimorphism such as raptors) are more prone to electrocution than other birds, which could result in global population effects. Furthermore, electrocution has been shown to be the main cause of declines in one of Europe's rarest raptor, the Bonelli's eagle (*Aquila fasciata*), and the reduction of electrocution mortality, principally of subadult birds, is likely to be critical to the survival of this endangered species in the Iberian Peninsula. Resolving electrocution problems has been critical to the maintenance of other raptor species such as the endangered Spanish imperial eagle (*Aquila adalberti*), or the Eurasian eagle owl *Bubo bubo*, in Europe, or the critically endangered Californian condor (*Gymnogyps californianus*) in North and Central America. Currently, the causes of electrocution in birds have been identified and several measures with different successful results have been undertaken. Fortunately, electrocution risk to birds is not described by a random distribution, but a quasi-Poisson one in which few pylons account for most electrocutions and hence, mitigation is feasible. For example, identification of dangerous pylons and the application of appropriate insulation techniques has resulted in lower electrocution rates whereas other mitigation measures such as perch deterrents have not succeed. The network of above ground power lines has grown continuously in the last 50 years and this has been accompanied by an increase in other man-made structures, including roads, railroads and new power generation facilities (e.g. solar panel fields and wind farms). At least at the local level, once mortality sources have been identified (e.g. electrocution and collision with power lines), their impact can be lessened. However, it is not entirely clear that lessons learned and actions undertaken at the local level can be applied across a wider area to solve large scale conservation problems such as bird electrocution. Here we provide information and a practical example of how mitigation measures implemented on a regional level under the conservation program of the Spanish imperial eagle have resulted in a positive shift of demographic trends of one of the most endangered raptors in the world. # Materials and Methods ## Study species The Spanish imperial eagle (*Aquila adalberti*) is a long-lived resident tree- nesting raptor endemic of the Iberian Peninsula. With an estimated population of 250 pairs (National Working Group, unpublished data 2008), it has been considered one of the most endangered raptors in the world,. The Spanish imperial eagle population has increased in most of the Iberian populations since 2000. Its main threats are mortality caused by electrocution, poisoning, habitat fragmentation, shooting and the decline of its main prey, the European rabbit (*Oryctolagus cuniculus*) due to viral haemorrhagic disease. Electrocution on power lines has been reported to be the main known cause of death for the species, accounting for 60% of mortality cases with a strong sex-biased distribution towards female birds. This loss of individuals and resultant bias in the sex ratio in the small Doñana population, coupled with the K-species characteristics of the Spanish imperial eagle have been suggested as the cause of the rapid population decline there. ## Study area and mortality records The study area included the Andalucía region (area  = 87598 km², southwestern Spain), representing the 17,3% of the total surface of Spain, and the Doñana National and Natural Parks (total area  = 53709 km², 37°N 6°30′W). A 35 years temporal dataset (1974–2009) on mortality of Spanish imperial eagle was recorded. This included results of population censuses, and data on electrocution and non-electrocution of birds recovered in both the Doñana (local level) and Andalucía populations (regional level). Data from Doñana were obtained from the Doñana field diary (Doñana Biological Station archives) and from the Station Ringing Department. In addition, from 1986 to 1988, data were obtained from 32 young Spanish imperial eagles equipped with solar radio-transmitters (Type HSPB 1400 3xA, Wildlife Materials Inc.), fitted as backpacks to young when they were 50–60 days old. Also, a compilation of data on eagle mortality was obtained from the regional government (Consejería de Medio Ambiente, Junta de Andalucía) which kept records as part of the regional conservation program for this species. ## Mortality surveys Data on dead birds were gathered by means of specific field surveys during different time periods as follows. Every two months searches of the ground under 100 km of medium-tension power line in and around territories of Spanish imperial eagle were made during 1980–1982. From 1987 onwards the ground under an additional 500 km of power lines was checked every two months in juvenile dispersal areas as determined by radio-tracking. Furthermore, from 1990 to 1994 a total of 6288 selected power poles were surveyed annually throughout the entire distribution area of Spanish imperial eagle (including both natal and dispersal areas of immature birds), as well as adjacent areas normally beyond the species' range, to assess mortality in other species. Since 1995 until 2009, regular annual surveys of power lines in breeding and dispersal areas have been conducted. During the entire period, we also get additional data from radio tagged individuals (around 150 individuals in total). A detailed description of field methods is available in. Raptor electrocution mainly occurs at low-tension power lines (16–45 kV). However, in contrast to higher-tension power lines that pertain to a few large companies for which detailed data are available, accurate data about the extension of low-tension power lines were not available for the entire study area given the many owners of particular power lines and the lack of a compiled Geographical Information System (GIS) database. For this reason, data concerning the extension of medium-tension (200–110 kV) distribution power lines were recorded as an adequate surrogate of the low-tension wiring network extension in Spain. These data were obtained from the Spanish Association of Electricity Industry ([www.unesa.es](http://www.unesa.es)). ## Mitigation measures Mitigation measures included the identification of mortality hotspots, development of predictive cartography and prospective modeling. Both proactive actions (making safe dangerous pylons before mortalities occurred) and reactive measures (making safe after mortalities were recorded) were undertaken. Since pylon design and habitat had highly significant effects on raptor mortality, accounting for 82% of the variance in death rates, mitigation measures aimed to ameliorate the impact of power lines were implemented. These included the construction of new pylons with suspended insulators, avoiding the use of pylons with an exposed loop of wire above the insulator, and ensuring that new power lines were located away from both breeding areas and areas of temporary settlement of juveniles eagles. In the case of existing power lines measures were focused on correcting dangerous pylons by replacing exposed rigid insulators with suspended ones and installing protective systems on the pylons to prevent birds from coming into contact with wires. The selection of priorities in retrofit dangerous power poles was made using information on factors affecting mortality distribution. ## Statistical analysis Data were divided into two different periods: before and after the approval of the regulation of power line design in 1990 by the Andalusian regional government (Regulation 194/1990, June 19^th of the Junta de Andalucía). This regulation established mandatory rules aimed at minimizing or eliminating the negative impacts of power lines facilities on avian populations in and close to natural protected areas in the region. Although the Regulation was approved in 1990, the first mitigation measures were applied in 1992, so the two periods that were considered in the analysis were 1974–1992 and 1993–2009. We compared the number of birds found electrocuted per nesting pair and per km of medium-tension power lines at both regional and local levels, between the two periods using a Mann-Whitney non-parametric test. Linear regression curves including each variable versus time were fitted for the two different periods. The slopes of the two curves (before and after) were calculated and the change of the value assessed. Homogeneity of slopes tests were used to examine the interaction between the period and each of the continuous variables in influencing responses. The significance threshold was set at alpha  = 0.05. Statistical calculations were performed within STATISTICA 7.0. # Results Since 1974, a total of 158 Spanish imperial eagles have been recorded dead in Andalucía, 101 of them (63.92%) inside Doñana National Park. Electrocution was the most frequent cause of death accounting for 39.87% of the total mortality events. Since 1974 when the first death by electrocution was recorded, 37 Spanish imperial eagles have been found electrocuted in Doñana (36.63% of the mortality cases) and 26 cases in Andalucía (41.94% of the records). The population of Spanish imperial eagle has increased in Andalusia from 22 pairs recorded in early 70s to 60 pairs recorded in 2009 (average annual percentage of population change  = +3.46%), with this population currently representing 24% of the world population of the species. However, the Doñana's population, which is separated from other breeding populations (the nearest nesting conspecifics breed 300 km away;), has not experienced a parallel increase during the same time period, starting from 13 pairs in the 70s, reaching up to 16 pairs in the 80's and followed by a period of decrease in the 90's and early 2000 (average annual percentage of population change  = −0.40%). In relation to the two time periods considered, before and after the approval of the Andalusian regulation of power line design, the average annual percentage of population change has increased from +2.24% to +4.74% for the Andalucía population and decreased from +0.92% to −1.80% in the Doñana population. In the latter case, after 1992, a dramatic increase in annual adult mortality due to an increase in illegal use of poison in hunting areas surrounding Doñana National Park was recorded. Currently, the Doñana population is increasing slightly; nine pairs bred in 2009. In relation to mortality causes, the number of birds recovered dead by any cause has decreased considerably in Doñana (−72.40% of change in the cases recorded before and after the mandatory regulation was approved in 1990) and Andalucía (−45.17% reduction). Comparing the same time periods, both Doñana (−96.90%) and Andalucía (−61.95%) experienced declines in the number of dead eagles being recorded as victims of electrocution. Significant changes in the slope of regression curves between the two periods are reported in. Although mortality between these two periods decreased there was a continuous increase in the amount of overhead electrical wiring at regional and national levels. Interestingly, data show a significant change in the number of birds found electrocuted per nesting pair and per km of medium-tension power lines both in Andalucía (Mann-Whitney, U = 58.50, Z = 2.75, P = 0.006) and in Doñana (Mann- Whitney, U = 43.50, Z = 3.56, P\<0.001). Since 1992 until 2009 a total of 6560 dangerous pylons were made safe along 1446 km of power lines in Andalucía. The total budget for these measures amounted €2,624,000 (an average of €400.00 per corrected pole). This figure represents slightly over half of the total investment in conservation of the species for this period which adds up a total of €4,481,665.12. # Discussion Large scale conservation problems and consequent appropriate management actions operate at different temporal scales. Whereas human-induced factors affecting mortality may operate in relatively short time frames (e.g. bird electrocution, wind farm collisions, shooting, poisoning, etc.), management actions require longer temporal scales in order to determine how effective really are. This leads to conflicts between politicians, managers of the public administrations and scientists, whose relevant timeframes differ from one another. In the situation we describe, long-term data series of both population and mortality for Spanish imperial eagle are essential in determining the effectiveness of mitigation measures and their likely effect on population persistence. Therefore, we used a 35-years data series of the Spanish imperial eagles' population from Doñana and Andalucía to report a well-studied example of how a large scale conservation problem has been identified and how, after applying evidence-based mitigation measures, there has been a positive shift in the demographic trend of one of the most endangered raptors of the world. Our results show a shift in the main causes of mortality between the two periods, before and after the approval of mandatory regulation against bird electrocution in the Andalucía region. Whereas during 1973–1992 the main problem was electrocution, after 1992 the main mortality source shifted to the illegal use of poison in the Doñana surrounding area. In late 80's and early 90's, several studies highlighted the risk of electrocution to population persistence of the Spanish imperial eagle, , and mitigation measures were instigated accordingly. As a consequence, electrocution rates have changed from accounting for nearly 60% of total mortality events, to 39.87% (this study). In a general review of raptor electrocution studies, Lehman et al. stated that “without information on bird numbers, a post-mitigation decrease in mortality could be attributed to pole modifications, when in fact it is the result of a population decline”. This is not the case of Spanish imperial eagles for which the reduction in electrocution fatalities has been accompanied by a general increase in the population, and demonstrates that at least some large scale conservation problems can be resolved. However, the question remains as to the total cost of fixing the problem, and whether that cost is affordable. In seeking affordability of mitigation methods it is important to try to maximize benefit while minimizing cost. One of our main recommendations in solving an electrocution problem is to concentrate on the most dangerous pole designs. In our study, there was an association between pole design and habitat type. For example, pin-type poles in the natural habitat accounted for 26.3% of the mortality of Spanish imperial eagles , but only 6.7% of poles in natural habitat are of this type. So, mitigation measures aimed at correcting pin-type poles not only achieved the aim of reducing mortality, but were cost effective. Our results show how, after mitigation, there has been a strong decrease in bird electrocutions, both in the Doñana and the Andalusian populations, with a 97% and 62% of reduction respectively. This is clearly good news from the conservation point of view and well exceed calculations of Janss and Ferrer, who forecasted a 79.7% reduction in mortality as a result of correcting 37.2% poles in Doñana. The importance of human-induced mortality should also be considered in the context of population dynamics. Demographic variables are not independent of one another when endangered small populations are analyzed, as the case of Spanish imperial eagles. In long-lived spatially structured populations, pre-adult and adult mortality play a key role in influencing population persistence. Therefore conservation measures should be focused on mitigating detrimental effects on these parameters. In the case of Spanish imperial eagles, adult survival is higher than pre-adult and floater survival and hence small electrocution rate of adult eagles leads to more severe consequences on population dynamics than the death of non-breeding individuals. During the period 1995–2001 several projects carried out by the Autonomous Communities (including Andalucía) and the Spanish Ministry of the Environment were implemented to protect and conserve the Spanish imperial eagle, and some of these included measures to avoid bird electrocution. In addition, there has been an increasing pressure to legally require that efforts to mitigate avian electrocution are made. As a consequence, regional and national laws were approved with regard to avian electrocution (Decree 178/2006, October 10^th, of the Junta de Andalucía), “that establishes mandatory rules for the protection of avifauna against high-tension power lines”, and the Spanish Royal-decree (263/2008, February 22^nd), “which establishes mitigation measures in high-tension power lines to protect birds”. Mitigation measures included the insulation of cross arm braces, which was the most effective and practical tool to reduce electrocution of raptors at metal lattice power poles, taking into account that insulating materials should be periodically checked and replaced. Other effective measures included the fixation of spirals and the crossed bands for conductor-marking and the installation of static wire-marking to avoid collisions. In contrast, other actions such the employment of raptor models as deterrents were not effective. ## Conclusions Conservation and the preservation of biodiversity require financial investment from both the public and private sector budget. In our case, nearly €2.6 million have been spent on mitigation of bird electrocution during 1992–2009, which equals an investment of €154,352.94 per year. The Spanish imperial eagle population in Andalucía has increased from 31 to 60 pairs in the same period. Taking into account the high budgets assigned to the construction of new power lines and alternative power sources (e.g. wind farms, solar panel arrays), our results demonstrate that solving bird electrocution is an affordable problem if political interest is shown and financial investment is made. Furthermore, other avian species could have benefited from these measures. Future research should be focused on the development of new high-efficiency, low-cost devices that reduce electrocution risk of distribution power lines. These mechanisms should be easy to install. They should allow birds to perch and nest while protecting the lines to ensure that power supply is not disrupted due to bird electrocution events or short-circuiting by nest material. Finally, conservation actions should include adequate spatial planning, avoiding the placement of power lines in areas of special conservation interest. The combination of an adequate spatial planning with a sustainable development of human infrastructures will contribute positively to the conservation of the Spanish imperial eagle and may underpin population growth and range expansion, with positive side effects on other endangered species. We are grateful to all the people who have helped in the field work. Special thanks are given to the Consejería the Medio Ambiente of the Junta de Andalucía. An anonymous referee made valuable comments on an early draft of the paper. This paper complies with the current laws of Spain. P. López-López wrote this paper during a pre-doctoral stay at Doñana Biological Station (CSIC) in Sevilla, Spain. [^1]: Conceived and designed the experiments: PLL MF AM. Performed the experiments: PLL MF. Analyzed the data: PLL MF. Contributed reagents/materials/analysis tools: MF EC AM. Wrote the paper: PLL. Edited manuscript: PLL MF EC MM. [^2]: The participation of one the coauthors as a member of the company Natural Resources Ltd. does not alter the authors' adherence to all the PLoS ONE policies on sharing data and materials. All data used for this study is freely available upon request to the authors.

# Introduction <u>H</u>orizontal <u>G</u>ene <u>T</u>ransfer (HGT), the movement of genetic material between individuals without reproduction, is a major evolutionary force within microbial communities and impacts genome dynamics across all life. Although HGT events often provide direct fitness benefits to recipient cells, such as antibiotic resistance, integration of foreign DNA is an inefficient process. As a result, newly acquired regions often interfere with physiological, genetic, and regulatory pathways to cause changes independent of phenotypes under immediate or direct selection pressures. Numerous studies have demonstrated the existence of such costs by documenting changes to fitness, growth rate, or other phenotypes after the transfer of relatively small genomic regions. However, few studies have examined costs associated with megaplasmid transfer. A variety of non-mutually exclusive mechanisms potentially contribute to costs of HGT. For instance, recently acquired genes are typically expressed at inefficient levels leading to limitations in resources such as ribonucleotides, amino acids, or ATP. Additional genes can occupy molecular machines required for basic cellular functions, such as polymerases and ribosomes, and sequester these limiting enzymatic resources from more critical activities. Foreign proteins may not fold correctly in their new cellular contexts, which could lead to disruption or triggering of stress responses. Recently acquired regions may disrupt flux through cellular systems, leading to the buildup of toxic intermediates. While such costs have been directly observed in laboratory experiments, retrospective studies across genomes add an additional layer of complexity as there exists an inverse correlation between gene retention after HGT and number of protein-protein interactions affected. In most cases the precise molecular mechanisms underlying observed costs of HGT have not been identified, however, both the magnitude and molecular basis for costs could be greatly affected by both the size and gene content of the acquired region. Costs of HGT have typically been studied by focusing on phenotypic changes after HGT of relatively small plasmids and lysogenic phage, even though large-scale transfers (\>60,000 bp) occur at appreciable rates throughout bacteria,. We have developed an experimental system to investigate the costs of large-scale HGT by using a ∼1Mb megaplasmid which is self-transmissible throughout *Pseudomonas* species. Transfer of this megaplasmid occurs in both liquid and solid media and requires a type IV secretion system. This HGT event introduces approximately 700 ORFs into recipient cells, including many “housekeeping” genes as well as almost a full complement of tRNA loci as is characteristic of chromids. Importantly, it does not appear as though full pathways are present for megaplasmid encoded housekeeping gene pathways, so function very likely requires direct interaction with chromosomal networks. In a parallel manuscript, we demonstrate that acquisition of this megaplasmid lowers fitness of *Pseudomonas stutzeri* by ∼20% and here we report on multiple additional phenotypes affected by large-scale HGT. Specifically, we find that megaplasmid acquisition leads to sensitivity to quinolone antibiotics, DNA intercalating agents, temperature, and killing by other bacterial species. We further find that HGT changes bacterial behavior in that biofilm formation is decreased and motility is increased. This widespread pleiotropy could signal that multiple phenotypic costs occur throughout transfer events. Moreover, that such pleiotropic relationships between phenotypes are mediated at a systems level by single HGT events creates a unique situation where phenotypic evolution occurs as a by-product of evolutionary amelioration after transfer rather than direct selection on phenotypes themselves. In sum, we document the significant potential for secondary effects of HGT to alter phenotypic evolution and adaptive trajectories across microbial populations. This system further underscores the indirect power of costs of HGT to rapidly generate phenotypic diversity across closely related bacteria. # Results ## Megaplasmid Acquisition Decreases Thermal Tolerance Our initial observations suggested that, although we were successfully able to select for conjugation of pMPPla107 from *P. syringae* pv. *lachrymans* 107 to *P. stutzeri* DBL332 when plated out on selective antibiotics at 27^oc, conjugations failed when selected at 37^oc (data not shown). We have since demonstrated, as one can see in, megaplasmid acquisition sensitizes *P. stutzeri* to growth at 37°C and greater. Although both *P. stutzeri* DBL332 and DBL390 grow well at either temperature, independently created strains containing pMPPla107 (DBL365 and DAB412) appear stressed at 27°C and fail to grow at 37^oc. To further quantify this effect, we measured how changes in temperature alter competitive fitness for two related megaplasmid containing strains (DBL365 and DBL453). As one can see in, presence of pMPPla107 decreases competitive fitness by 22% and 12% for DBL365 and DBL453 respectively at 35^oc compared to 27^oc. Analyzed within a full factorial ANOVA framework, this effect of temperature is significant (F_1,3 = 45.33, p = 0.0067). Furthermore, analyzed as contrasts within the ANOVA framework, each strain's fitness is significantly lower at 35°C compared to 27°C (p\<0.05). Since DBL453 is derived from DBL365 by recombining out the tetracycline marker, these results are not due to the marker itself. Therefore, two separate assays confirm that large-scale HGT of a megaplasmid decreases thermal tolerance in *P. stutzeri*. ## Megaplasmid Acquisition Decreases Antibiotic Resistance We used Biolog assays to identify phenotypic changes associated with acquisition of megaplasmid pMPPla107 by *P. stutzeri* DBL187. Overall, aside from an overall consistent negative effect of pMPPla107 on growth of *P. stutzeri,* acquisition of the megaplasmid only significantly changed a handful of phenotypes across replicates. One striking result is that megaplasmid acquisition decreases resistance of *P. stutzeri* to a variety of quinolone antibiotics as well as DNA intercalating agents such as 7-hydroxycoumarin. We followed up these results by performing replicated growth curves in both nalidixic acid and 7-hydroxycoumarin over a series of different concentrations after independently conjugating the megaplasmid into another *P. stutzeri* strain background (DBL332). As shown in, megaplasmid presence lowers the minimal inhibitory concentration of *P. stutzeri* to both 7-hydroxycoumarin and nalidixic acid, thus replicating the results from the Biolog assay. To further quantify this difference in antibiotic resistance, we measured competitive fitness between DBL332 and DBL365 in 0 and 4 µg/mL nalidixic acid. As one can see in, presence of pMPPla107 decreases competitive fitness by 31% and 28.5% for DBL365 and DBL453 respectively in the presence of 4 µg/mL nalidixic acid. Analyzed within a full factorial ANOVA framework, the effect of antibiotic is significant (F_1,2 = 20.7564, p = 0.045). Furthermore, analyzed as contrasts within the ANOVA framework, each strain's fitness is significantly lower at 4 µg/mL nalidixic acid compared to standard SW-LB (p\<0.05). Since DBL453 is derived from DBL365 by recombining out the tetracycline marker, these results are not due to the marker itself. Therefore, two separate assays confirm that large-scale HGT of a megaplasmid decreases resistance to quinolone antibiotics in *P. stutzeri*. ## Megaplasmid Acquisition Decreases Biofilm Formation We witnessed that, after extended periods of growth in liquid media, *P. stutzeri* DBL332 forms a mass at the bottom of pipette tip within the culture after approximately 2-4 days. This mass remains regardless of how long cultures are left to incubate (longest tested period was 10 days, data not shown). However, strain DBL365 does not form such a mass. To quantify this effect, we grew replicate cultures of strains DBL332 and DBL365 in SWLB media containing a single pipette tip. As one can see in, even though population sizes of bacterial population suspended in liquid media are roughly equivalent (6.6×10⁹ and 4.8×10⁹ CFU/mL for DBL332 and DBL365 respectively), population sizes of bacteria attached to the pipette tip are much greater in the strain that lacks the megaplasmid (1.4×10⁸ compared to 7.6×10⁶). Therefore, whereas roughly 2% of *P. stutzeri* forms a mass on the pipette tip in the absence of pMPPla107, only 0.15% of the population forms this mass in a strain containing the megaplasmid. Similar results were observed for this assay with DBL453 as well as an additional *P. stutzeri* strain (DBL408) which acquired the megaplasmid independently of DBL365. We also note that we attempted to perform traditional biofilm assays to compare strains that contain or lack the megaplasmid, but slower growth of the megaplasmid strains made comparisons difficult to interpret (data not shown). ## Megaplasmid Acquisition Increases Motility We tested whether megaplasmid acquisition altered motility or chemotaxis in *P. stutzeri* using standard assays. Briefly, 1 µL of liquid culture was plated into semisolid agar and strains were placed into an incubator. After two days, size of the bacterial halo surrounding the inoculation point was quantified. As shown in, strains containing the megaplasmid consistently had larger halos than strains that lack the megaplasmid. Quantification shows that this increase in halo size is approximately 27% (F_1,4 = 59.294, p = 0.0015). Similar results were observed for this assay with DBL453 as well as an additional *P. stutzeri* strain (DBL408) which acquired the megaplasmid independently of DBL365. ## Megaplasmid Acquisition Increases Sensitivity to Supernatants from Other Bacterial Species *P. aeruginosa* strains are known to produce a variety of antimicrobial products during growth within laboratory culture, including pyocins and quinolones. Since interactions between bacterial species occur frequently within the environment, and are essential for HGT of the megaplasmid to occur, we tested for whether megaplasmid acquisition altered sensitivity of *P. stutzeri* to *P. aeruginosa* supernatant. We found that supernatant from *P. aeruginosa* interfered with growth of *P. stutzeri* cultures, but only when strains contained the megaplasmid. Similar results were observed for this assay with an additional independently created *P. stutzeri* strain (DBL408). Therefore, megaplasmid acquisition alters interactions between bacterial species by sensitizing strains to growth inhibition by other bacterial species during normal growth. # Discussion For horizontally transferred regions to be maintained within a population, they must either provide a large enough benefit or be transmitted at high enough rates across individuals to avoid loss due to selection or genetic drift. That such benefits may be the primary target of strong selective pressures within a given environment, as with antibiotic resistance, doesn't preclude the existence of neutral secondary phenotypic changes or HGT-associated costs which are deleterious in other environments. Although such costs of HGT appear to be widespread, there have been few efforts to investigate phenotypic effects of megaplasmid transfer. Furthermore, even when costs are observed, measurements are often limited to single phenotypes even though multiple cellular systems could be affected,. Here we explore a system where HGT increases bacterial genome size by ∼20%, using a megaplasmid which is self-transmissible throughout pseudomonads. We report that megaplasmid acquisition alters numerous phenotypes within *P. stutzeri* in unprecedented ways, which highlights the potential for large-scale transfers to shape evolutionary dynamics within natural populations. This system provides a unique foundation to explore how evolution affects pleiotropic interactions between phenotypes altered as a result of large-scale HGT events, but also a powerful model to dissect individual interactions underlying these costs at a molecular level. In a separate manuscript, we have demonstrated that megaplasmid acquisition by *P. stutzeri* impacts competitive fitness and bacterial growth under standard laboratory conditions. Here we show that megaplasmid acquisition is also accompanied by secondary changes to a variety of phenotypes affecting cellular physiology, environmental survival, and interactions with other species. For instance, megaplasmid acquisition increases sensitivity to quinolone antibiotics as well as stresses such as heat. To the best of our knowledge this is the first report of horizontal gene transfer directly lowering antibiotic resistance, and is especially interesting given that environmental stress can increase quinolone resistance. That these responses are specifically affected by the test environments, as opposed to correlated effects on slower growth across a range of conditions, is highlighted by lack of sensitivity to a variety of other tested conditions. We also demonstrate that megaplasmid acquisition increases sensitivity to a substance present within the supernatant of *P. aeruginosa* cultures. As with quinolone sensitivity, to the best of our knowledge this is the first report of a plasmid mediating sensitivity to supernatant from bacterial cultures. Although we currently do not know the molecule(s) responsible for this effect, multiple bacteriocins and other antimicrobial targets known to be found within the supernatant will be the target of future studies. We have further shown that acquisition of the megaplasmid increases bacterial motility (or lowers the chemotaxis threshold) within soft agar and decreases biofilm formation in liquid culture. It is important to note that all of these phenotypic changes are effectively neutral under the defined laboratory conditions, which we use to select for successful conjugation, because this megaplasmid is engineered to provide tetracycline resistance. Therefore, under selective conditions in the lab, cells that can't acquire the megaplasmid will die due to antibiotic selection. While selective pressures in nature are likely more complex, these secondary changes could have dramatic effects on ecological strategies or niches between closely related bacterial populations. This megaplasmid is representative of other large-scale gene transfers in terms of coding capacity, genetic content, and divergence from the recipient genome. In spite of data demonstrating a negative bias for retention of highly connected genes after HGT events, megaplasmids often contain and can transfer numerous housekeeping genes. The megaplasmid within our system itself contains 38 tRNA loci, polymerase subunits, DNA recombination and repair systems, a putative ribosomal protein, as well as other proteins that could be involved in housekeeping functions. At the moment we don't know whether HGT associated costs are dependent upon any of these genes interfering with chromosomal pathways. However, results obtained herein could represent general outcomes after HGT events or may only be emergent properties of HGT by specific types of larger vectors like chromids. One major question arising from these results concerns the independence of phenotypic shifts after HGT. Do all observed changes result from a single protein-protein interaction, numerous individual detrimental interactions, or the disruption of complex and interwoven regulatory networks? Furthermore, is the breadth of altered phenotypes a general property of HGT events as a whole or does the number of changes increase with size or gene content of transferred region? One candidate pathway does stand out as a potential mediator of these phenotypes *a priori*. Acquisition of the megaplasmid brings with it hundreds of new genes, the protein products of many of which are membrane localized. Since cell size is limited, the incorporation of additional membrane bound proteins likely disrupts molecular signatures and dynamics of the membrane and could easily trigger or disrupt the envelope stress response. The envelope stress response is conserved throughout bacteria, and responds to a variety of membrane stresses through the action of proteases, anti-sigma factors, as well as a host of other regulators. Since membrane integrity is critical for bacterial survival, the envelope stress response often sits at the top of regulatory cascades that control numerous phenotypes. That previous results suggest quinolone sensitivity and sensitivity to *P. aeruginosa* supernatant are correlated with cell membrane integrity provides support for this model –. Furthermore, chaperone function links both heat and envelope stress responses, since one of the main determinants for both pathways is improperly folded proteins. Lastly, motility and biofilm formation are directly regulated by the envelope stress response in *Pseudomonas*, as both the flagellum and pili are critical membrane bound structures,. The addition of so extra DNA through HGT could also affect gene regulation at a global level through modification of chromosomal conformation. Proteins like HN-S play important roles in limiting detrimental effects of HGT by silencing transferred regions, but are also critical for packaging the chromosome. HN-S like proteins encoded by IncHI plasmids potentially mediate multiple phenotypic effects in after HGT in *Salmonella* including: increasing competitive fitness at low temperatures, increasing survival at high temperatures, and a reduction of motility. MvaT is a *Pseudomonas* analog of HN-S, and an MvaT homolog (Pmr) found within the *Pseudomonas* pCAR1 plasmid mediates multiple phenotypic changes including increased resistance to chloramphenicol. Plasmids often encode nucleiod associated proteins (NAPs) such as HN-S, so that regulatory alterations of chromosomally encoded pathways may be a fairly general feature of plasmid driven HGT. It is striking that the phenotypes associated with acquisition of pMPPla107 by *P. stutzeri* are nearly the exact opposite of changes reported after HGT within for other systems. The pMPPla107 megaplasmid does contain at least three loci that resemble nucleoid associated proteins (including two divergent copies of MvaT/Pmr and a locus similar to IHF), but it is currently unknown whether interactions between these loci and chromosomal counterparts are responsible for observed costs. It is also possible that similar interactions between chromosome and plasmid occur across IncHI, pCAR1, and pMPPla107, but that *P. stutzeri* DBL332 regulates downstream pathways opposite of *Salmonella* and other pseudomonads. Along these lines, quinolone antibiotics are known to disrupt gyrase function and alter the level of DNA supercoiling within the cell. Sensitivity to quinolones could therefore arise after HGT because chromosomal maintenance requires a threshold amount of functional gyrase. Presence of the megaplasmid could increase this threshold or lower the amount of available gyrase, either of which would lower the concentration of quinolone required for antibiotic effects. Changes in DNA supercoiling are known to affect cellular responses to various stresses including heat shock. However, one should note that a complex feedback loop exists between DNA supercoiling and the envelope stress response, and both may be important contributors to costs of HGT demonstrated within this manuscript. Alternatively, a variety of other independent physiological and regulatory pathways could be responsible for these changes including nutrient limitation triggering the stringent response or the disruption of multiple quorum sensing pathways altering regulation across the genome. Outside of any direct fitness benefits from HGT, that multiple phenotypic changes are linked through single HGT events could skew evolutionary dynamics in unpredictable ways. For example, it is well known that adaptive trajectories and “evolvability” can be influenced by the order that beneficial or compensatory mutations fix. Secondary effects of HGT could bias future evolutionary paths within populations by altering magnitude or direction of epistatic interactions between adaptive mutations. Furthermore, since recently acquired regions often function sub-optimally, the total number of potential beneficial mutations within an environment could be increased due to compensatory changes for the HGT event. This influx of beneficial mutations correlated to HGT could impact both the order and magnitude of fixed adaptive mutations through clonal interference. Along these lines, significant differences in evolutionary potential could arise based on, for instance, whether antibiotic resistance is introduced through HGT or *de novo* mutation. Such a balance could be further impacted in positive or negative ways by environment specific feedbacks on costs of HGT. We also note that pMPPla107 conjugates across strains at high rates within the laboratory environment, on both solid media and in shaking liquid cultures, with transfer dependent on a type IV secretion system. Therefore, the costs observed after HGT of this megaplasmid may not need to be compensated in order for pMPPla107 to be maintained within a population. These observations suggest an interesting line of research exploring interactions between costs of HGT, transmission rates, and megaplasmid persistence within populations through time. Looking forward, amelioration of costs of HGT could lead to dramatic phenotypic differences between strains descended from a recent common ancestor due strictly to differences in paths of evolution and pleiotropic relationships from HGT. These pleiotropic relationships could be reinforced or disrupted depending on which suites of mutations fix over time. For instance, amelioration of megaplasmid associated costs could produce one cluster of strains that is phenotypically indistinguishable from the non-megaplasmid ancestor while another cluster compensates only for costs at 27^oc and is unable to grow at 37°C. Furthermore, as a result of costs of the megaplasmid to *P. stutzeri*, increased resistance to quinolone antibiotics such as nalidixic acid or ciprofloxacin could evolve solely due to compensation for the megaplasmid in the absence of direct selection by quinolones. Since HGT introduces foreign genes and pathways into novel genomic contexts, each transfer event brings with it great potential to disrupt existing genetic and physiological networks within the recipient cell. Although numerous results have analyzed and dissected costs after the transfer of relatively small plasmids and phage, we have developed a model system with which to explore phenotypic costs associated with large-scale HGT (∼20%). The phenotypic shifts we see are dramatic and include changes of bacterial tolerance to numerous stresses as well as alterations of behavior. Individually, a subset of these changes have been observed in other systems as a result of HGT or as a pleiotropic result of *de novo* adaptive mutations. However, taken as a whole, these results highlight the power of epistasis between the recipient genome and recently acquired regions to completely shift genetic and phenotypic expectations between closely related organisms. Furthermore, our results demonstrate the amazing breadth of phenotypes potentially affected by one HGT event and emphasize how singular evolutionary events can re-wire, reshape, and influence even the most well-studied genetic pathways. # Materials and Methods ## Bacterial strains and plasmids The ∼976 kb plasmid pMPPla107 was originally described in Pseudomonas syringae strain MAFF301305 (also known as pv. *lachrymans* 107) by Baltrus et al. 2011. A draft assembly sequence for pMPPla107 can be found at Genbank accession <u>CM000959.1</u>. All strains and plasmids used in the study are listed in. The focal *P. stutzeri* strain in this paper, 23a24, was isolated from soil, and was chosen for its high competence for natural transformation. This strain has been selected and phenotypically modified in a variety of ways to carry out the necessary experiments, with these modifications listed in. ## Culture conditions All experiments were carried out at 27°C unless otherwise stated. Salt water LB (SWLB) was used as base media for liquid cultures and agar plates. All liquid cultures were incubated on a rotary shaker (200 rpm). Antibiotics were used at the following concentrations: 20 µg/mL tetracycline, 50 µg/mL kanamicin, 50 µg/mL rifampicin, 10 µg/mL streptomycin. Xgal was used at a concentration of 40 µg/mL. For competitive fitness experiments in the presence of nalidixic acid, a concentration of 4 µg/mL was used. All dilutions and cell suspensions took place in sterile 10 mM MgCl₂. ## Competitive Fitness Assays To quantify the fitness cost of pMPPla107 in multiple environments, we set up independent competitive fitness assays, for paired *P. stutzeri* strains (i.e., one that lacked (DBL332) & one that contained (DBL365) a megaplasmid tagged with tetracycline resistance) using a pMP-free control strain (DBL390) that was phenotypically marked with gentamycin resistance and *lacZ*. Each week, we revived the strains to LB agar plates and then SWLB liquid (containing tetracycline at 10 µg/mL for pMP strains) by inoculating overnight cultures with cells from the corresponding agar plate. Subsequently, because the pMP impairs growth (e.g., extends the length of lag phase), we normalized the growth phase among the strains by conditioning each strain in tet-free SWLB for 48 hours under competition conditions by inoculating 2 mL cultures with 5 µL of the revival culture and incubating at 27°C. Furthermore, because the cost of pMPPla107 is severe at elevated temperatures and in the presence of nalidixic acid, we skewed the initial ratio of the pMP strain relative to DBL390 (∼5∶1 test:control ratio), whereas the pMP-free strain was set up ∼1∶1 with DBL390. Briefly, using optical density (OD600) to normalize cell density, we made a competition master mix (MM) for each strain large enough to make eight replicate 2 mL competitions per environment (e.g., 60 mL for 8 replicate competitions in three environments). However, because Nal 4 µg/mL was one of the environments tested, we initially set up the master mix with 2X the cells per unit volume (e.g., cells for 60 mL in 30 mL) and independently diluted the master mix 50∶50 with Nal 8 (i.e., 10 mL MM & 10 mL Nal 8 SWLB). The remaining 20 mL of the MM was diluted with 20 mL Nal 0 and used for the 8-replicate 27C/Nal0 & 8-replicate 35C/Nal0 competitions. For each replicate competition, relative fitness was calculated as the ratio of the growth rate (i.e., number of doublings) of the strain tested (e.g., DBL332, DBL365) relative to that of DBL390 during direct competition. The number of doublings for each strain in each competition was calculated by comparing the final density (i.e., \[CFU\]) and ratio of each independent competition to the average initial estimate calculated from eight replicate serial dilutions of each MM. ## Thermal Tolerance Assays Strains DBL332 and DBL365 were each grown up overnight in four replicate liquid cultures at 27°C in SWLB. A dilution series was then created from each tube and plated on two different SWLB agar plates. One plate was incubated at 27°C while the other was incubated at 37°C. This assay was carried out with two independently created megaplasmid strains (DBL365 and DBL412). After 4 days, both plates were photographed. Competitive fitness assays were performed as described above, with the following adjustments. The first set of replicate cultures was incubated for two days at 27°C, while the second set of replicates was incubated for two days at 35°C. 35°C was chosen as the temperature for competitions because DBL365 and DBL453 can undergo at least one doubling in these assays at this temperature. Each competitive fitness assay experiment contained 8 replicate cultures, and a total of 4 independent competitive fitness experiments were carried out for a total of 32 measurements for each strain. Within the ANOVA, strain and temperature were fixed variables while experiment was treated as a random variable. ## Antibiotic Resistance Assays Strains DBL332 and DBL365 were each grown up overnight in liquid cultures at 27°C in SWLB. Each culture was then used to inoculate a 96-well plate containing varying concentrations of either nalidixic acid or 7-hydroxycoumarin. This 96-well plate was placed into a Biotek Synergy-H1 plate reader, and incubated at 27°C with shaking for 24 hours. Every hour we took measurements of OD₆₀₀ from the assay plate in order to create a curve representing antibiotic inhibition. We chose 24 hours as the measurement represented in because the starting OD₆₀₀ values for each strain are equal. This assay was repeated with two independently created megaplasmid strains (DBL365 and DBL408), although only results from DBL365 are shown. Competitive fitness assays were performed as described above, with the following adjustments. A second set of replicates was inoculated into SWLB containing 4 µg/mL nalidixic acid. After two days, all cultures were plated out to SWLB agar plates containing Xgal, and the ratio of test/control strains was counted. Each competitive fitness assay experiment contained 8 replicate cultures, and a total of 3 independent competitive fitness experiments were carried out for a total of 24 measurements for each strain. Within the ANOVA, strain and antibiotic were fixed variables while experiment was treated as a random variable. ## Biofilm Assay Strains DBL332 and DBL365 were each grown up overnight into 2 mL liquid cultures at 27°C in SWLB. After one day, a second culture was inoculated at a 1∶100 dilution in 2 mL from the initial overnight culture and a sterile micropipette tip (Rainin, 1–20 µL) was placed into the culture. This culture is incubated 4 days. At this point the pipette tip was extracted and thoroughly rinsed with 10 mM MgCl₂ into a 2 mL Eppendorf tube containing 4 glass beads to remove biofilm from the tip. The tube was then shaken in a FastPrep machine for 20 seconds at 4 m/s. The dislodged bacteria were enumerated by plating a dilution series on SWLB, incubating at 27°C, and counting colonies after 3 days. Each assay experiment contained 4 replicate cultures, and a total of 4 independent competitive fitness experiments were carried out for a total of 16 measurements for each strain. Within the ANOVA, strain and population location were fixed variables while experiment was treated as a random variable. ## Motility Assay Strains DBL332 and DBL365 were each grown up overnight in 2 mL liquid SWLB cultures at 27°C. Then, we normalized the cell density (OD₆₀₀ of ∼1.0) for washed cell suspensions of each growth culture by suspending the harvested pellets in the appropriate volume of 10 mM MgCl₂ (e.g., cells from I mL of OD₆₀₀ = 2.0 were suspended in 2 mL). Using a sterile rounded wooden applicator (i.e, broken cotton tipped applicators), we inoculated the center of 2.5 mL motility agar (i.e., 1/2 strength LB and 0.25% agar) plates by first dipping the applicator into the washed cell suspensions. These plates were then incubated at 27°C for 48 hours, at which point pictures of each plate were scanned and the size of the motility halo was analyzed using ImageJ. Each assay experiment contained 8 replicate cultures, and a total of 5 independent motility assay experiments were carried out for a total of 40 measurements for each strain. Within the ANOVA, strain was a fixed variable while experiment was treated as a random variable. ## Bacteriocin Assay *P. aeruginosa* strains with killing activity were grown overnight in LB liquid culture at 37°C with shaking. 2 mL were placed in a 2 mL Eppendorf tube and centrifuged at 10,000 rpm for 5 minutes. Supernatant was carefully filter sterilized into a fresh tube without disturbing the pellet. Target strains were grown overnight in LB at 27°C with shaking. A 1∶100 dilution in LB was made using the target strain overnight and grown for 4 hr. 300 µL were taken from these cultures and used to inoculate 3 mL of melted agar (0.4%). Inoculated agar tubes were mixed well and utilized as an overlay on desired medium plates. Let overlay solidify with plate lid on for about 10–20 min. Spot overlay with extracted killing supernatant and let dry. Incubate plates at 27°C for 1–2 days. ## Data All raw data analyzed within this manuscript can be found within. # Supporting Information We thank Artur Romanchuk, Corbin Jones, and Jeff Dangl for helpful discussions concerning some experiments, and Kevin Hockett for assistance with the supernatant assay. We further thank Laura Williams for helpful comments and critiques of the manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: DAB KD BAS. Performed the experiments: DAB KD BAS AFM SM CF RM. Analyzed the data: DAB KD BAS. Wrote the paper: DAB KD BAS.

# 1 Introduction The 2019 Coronavirus Disease (COVID-19) is a highly contagious disease caused by infection of the SARS-CoV-2 virus. The disease began to spread in China in mid December 2019, and as the volume of intercity travel escalated around the Lunar New Year period, the number of infected individuals began to soar in mid January 2020 With no travel restriction in place due to the low level of vigilance or unawareness of the disease during the early phase of the outbreak, the spreading of the disease had gone almost unobstructed. Travel restriction began to be implemented throughout China since January 24, 2020, which has proven to be effective in curbing the spread of the virus. However, international traffic has not ceased and infectious individuals (who may or may not show any symptom at the time of travel) have actually travelled to different countries with the virus they contracted. Recent studies have also showed that travel restriction did contribute to the control of the spreading of COVID-19 within China as well as in a global context. By May 17, 2020, China had confirmed a total of 82,954 cases of COVID-19 infection, with death toll reaching 4,634. While China has begun to see declining numbers of infected cases in most cities from late February, other countries started to report surging number of cases in some cities or regions. As of May 17, 2020, the cumulative number of cases of COVID-19 infection was 4,804,011 worldwide, with South Korea, Italy and Iran reporting surges of infected cases within two weeks in late February and early March. Moreover, the global mortality rate has increased from around 3% (March 8, 2020) to 6.84% (May 14, 2020). In our recent work (available on February 19, 2020), intercity travel data obtained from Baidu Migration has been collected and integrated into the traditional Susceptible-Exposed-Infected-Removed (SEIR) model to account for the effects of inflow and outflow traffic between 367 cities in China. Parameters of susceptible-to-exposed infection rates, exposed-to-infected infection rates, and recovery rates for 367 cities have been identified by fitting the augmented SEIR model with historical data available from the National Health Commission of China. The predicted spreading profiles of the 367 Chinese cities (which were available on February 19, 2020) have been highly consistent with the actual profiles, including the times of infection peaks and the percentages of infected individuals in the 367 cities. In this work, we build on the data and estimated parameters obtained for 367 cities in our prior work, and establish a library of profiles (sets of codes) of different spreading profiles. Suppose a new outbreak has occurred in a given population. The numbers of infected and recovered cases during the early phase of the spreading form an initial profile. This initial or incomplete profile is compared with the historical full profiles obtained previously. Then, by selecting the best fit historical profiles, we identify the candidates parameters for prediction of the future spreading profile in that given population (of a city or region). It should be emphasized that the set of historical profiles obtained previously covers various possible spreading dynamics, representing a variety of contact topologies and transmission mechanisms including the travel effects that have been integrated in the model used to capture the transmission dynamics of the 367 cities. Thus, a new outbreak in a given population would likely follow one or a combination of the profiles based on the augmented model proposed in our previous work, and hence can be reconstructed from the historical sets of profiles. In this work, we develop a procedure for implementing the selection of historical profiles, identification of best-fit parameters and construction of future profile. We have applied the procedure to predict the epidemic progression in the cities of South Korea, Italy and Iran. Results of this study showed that the first wave of epidemic progression in most cities in South Korea peaked between early March and early April, 2020, and in Italy between late April and mid May, 2020, while Iran would have its peak in late April, 2020. We have also investigated the number of infected individuals in each city or region. Our method provides the average number of individuals eventually infected, along with a predicted deviation range at 95% confidence level. For Korea, we predicted that Daegu and Gyeongsang North Road would have around 7,619 and 1,287 people eventually infected (i.e., 0.306% and 0.062% of the city’s population), respectively, whereas the number of infected individuals in other cities in South Korea would be fewer than 300, i.e., less than 0.01% of city population. For Italy, we predicted that Lombardy and Emilia-Romagna would eventually have about 90,000 and 30,000 infected cases (i.e., 0.802% and 0.604% of the region’s population), respectively, and the number of people eventually infected in other cities in Italy would be below 10,000. Moreover, Iran would have more than 9,000 and 6,000 confirmed cases in Tehran and Isfahan (i.e., 0.39% and 0.2% of the city’s population). In addition, the number of people infected in most other cities would be larger than 1,000 (\>0.1% of the city’s population). From the progression trends of the epidemic in these three countries, provided control measures continue to be in place, our model show that the number of people infected in most regions of these three countries will peak before the end of May 2020. Hence, the first wave of the spreading of the disease is expected to come under control before the end of May 2020. In the remainder of the paper, we first introduce the official daily infection data used in this study. The augmented SEIR model is briefly reviewed, mainly to introduce the parameters of the model used for prediction of spreading profiles. The key procedure for matching historical profiles and prediction of future spreading profiles will be explained. Results of application of the proposed method to prediction of the peaks and extents of outbreaks in South Korea, Italy and Iran will be given. Finally, we will provide a discussion of our estimation of the propagation and the reasonableness of our estimation in view of the measures taken by the authorities in controlling the spreading of this new disease. # 2 Data The World Health Organization currently sets the alert level of COVID-19 to the highest, and has made data related to the epidemic available to the public in a series of situation reports as well as other formats. Our data include the number of infected cases, the cumulative number of infected cases, the number of recovered cases, and death tolls, for individual cities and regions in South Korea and Italy, from February 19, 2020, to May 12, 2020, and in Iran from February 19, 2020, to March 22, 2020, Data organized in convenient formats are also available elsewhere. Samples of data for Daegu, Gyeongsang North Road (South Korea), Lombardy, Amelia Romagna, Tehran and Mazandaran are shown graphically in. It should be noted that the data obtained for South Korea, Italy and Iran correspond to initial stages of the epidemic progression as the number of infected cases are still climbing, as of March 6, 2020. At present, there are two types of tests for confirming COVID-19 infected cases. One type of tests aims to confirm the presence of the SARS-Cov-2 virus in the body of an individual, which is commonly done via detecting the viral RNA through a polymerase chain reaction (PCR). The other type establishes the presence of antibodies in an individual, i.e., whether the individual being tested has been infected in the past, regardless of him or her carrying the virus at the time of testing. In this work, the official number of infected cases corresponds only to individuals who have been tested positive for the presence of the SARS-Cov-2 virus. # Method ## The augmented SEIR model The travel-data augmented SEIR model describes the spreading dynamics in terms of a basic fourth-order dynamical system with consideration of intercity travel in China. Consider a city *j* of population *P*_*j*. The states of the model are the number of susceptible individuals *I*_*j*(*t*), the number of exposed individuals (infectious but without symptom) *E*_*j*(*t*), the number of infected individuals *I*_*j*(*t*), and the number of recovered or removed individuals *R*_*j*(*t*). The model takes the following form in discrete time: $$\begin{array}{r} {X_{j}(t + 1) = F_{\text{aSEIR}}\left( X_{j}(t),M_{j},\mu_{j} \right)} \\ \end{array}$$ where *X*_*j*(*t*) = \[*S*_*j*(*t* )*E*_*j*(*t*)*I*_*j*(*t*)*R*_*j*(*t*)\]<sup>*T*</s up> is the state vector on day *t*, *F*_aSEIR(.) is the travel-data augmented function, *M*_*j* is the set of inflow and outflow travel strengths for city *j*, and *μ*_*j* is the set of parameters for city *j*, i.e., $$\begin{array}{r} {\mu_{j} = \left\lbrack \alpha_{j},\beta_{j},\kappa_{j},\gamma_{j},\delta_{j},k_{l} \right\rbrack} \\ \end{array}$$ where *β*_*j* is the rate at which a susceptible individual is infected by an infected individual in city *j*, *α*_*j* is the rate at which a susceptible individual is infected by an exposed individuals in city *j*, *κ*_*j* is the rate at which an exposed individual becomes infected in city *j*, and *γ*_*j* is the recovery rate in city *j*, *k*_*I* is the possibility of an infected individual moving from one city to another, and *δ*_*j* is the eventual percentage of the population infected in city *j*. Moreover, the eventual infected population in city *j* is given by $N_{j}^{s} = \delta_{j}P_{j}$. To facilitate comparison and matching of profiles, we introduce the normalized states as $\frac{{\overline{I}}_{j}(t)\Delta}{= {I_{j}(t)}/{N_{j}^{s}(t)}}$, $\frac{{\overline{E}}_{j}(t)\Delta}{= {E_{j}(t)}/{N_{j}^{s}(t)}}$, $\frac{{\overline{S}}_{j}(t)\Delta}{= S_{j}(t)/N_{j}^{s}(t)}$, and $\frac{{\overline{R}}_{j}(t)\Delta}{= {R_{j}(t)}/{N_{j}^{s}(t)}}$. Thus, can be represented in *normalized form* as $$\begin{array}{r} {{\overline{X}}_{j}(t + 1) = \mathcal{F}_{\text{aSEIR}}\left( {\overline{X}}_{j}(t),M_{j},\mu_{j} \right)} \\ \end{array}$$ where $\left. 0 \leq \right|{\overline{X}}_{j}\left. (t) \middle| \leq 1 \right.$. Since the above model has taken into account the human migration effect as well as the necessary transmission mechanism, we may consider the basic set of parameters to represent the characteristics of the propagation profile of city *j*. The complete set of parameters have been identified for 367 cities in China, which will serve as a set of codes for various propagation profiles of COVID-19 so far obtained. For brevity, we do not repeat the results here. While two different cities may have different population size and percentage of eventual infected population, the rates of infection and recovery should be similar across a group of cities, i.e., *μ*_*i* ≈ *μ*_*j*. Thus, in normalized form, we have $$\begin{array}{r} {\parallel {\overline{X}}_{i}(t) - {\overline{X}}_{j}{(t) \parallel < \epsilon\;\;{\text{for}\mspace{600mu}\text{some}}\mspace{600mu}\epsilon > 0,}} \\ \end{array}$$ for cities *i* and *j* within a group of cities having similar parameter sets. This also means $$\begin{array}{r} {N_{j}^{s}X_{i}(t) \approx N_{i}^{s}X_{j}(t)\;\;\text{or}\;\;\delta_{j}P_{j}X_{i}(t) \approx {\delta_{i}P_{i}}X_{j}(t)} \\ \end{array}$$ for the group of cities having similar rates of infection and recovery. Thus, provided the historical archive has adequately covered the possible dynamical profiles, we are able to perform fast prediction for any city *o*, by fitting an incomplete set of data (corresponding to an early outbreak stage in city *o*) and using the model parameters already obtained previously, as detailed in the following subsection. ## Prediction method The proposed data-driven prediction algorithm is based on the set of historical data of the spreading profiles of COVID-19 in 367 cities in China, namely, 367 sets of normalized time series of the form: $$\begin{array}{r} \begin{aligned} {\overline{I}}_{i}^{(c)} & {= \left\{ {\overline{I}}_{i}(1),{\overline{I}}_{i}(2),\ldots{\overline{I}}_{i}\left( K_{i} \right) \right\}} \\ {\overline{R}}_{i}^{(c)} & {= \left\{ {\overline{R}}_{i}(1),{\overline{R}}_{i}(2),\ldots{\overline{R}}_{i}\left( K_{i} \right) \right\}} \\ \end{aligned} \\ \end{array}$$ where *i* = 1, 2, …, 367, and *K*_*i* is the length of the data recorded in city *i*. Superscript “(*c*)” denotes data of Chinese cities. Now, suppose an outbreak occurs in city *o*, and only *k*_*o* days of data have been obtained in normalized form as $$\begin{array}{r} \begin{aligned} {\overline{I}}^{(o)} & {= \left\{ {\overline{I}}_{o}(1),{\overline{I}}_{o}(2),\ldots{\overline{I}}_{o}\left( k_{o} \right) \right\}} \\ {\overline{R}}^{(o)} & {= \left\{ {\overline{R}}_{o}(1),{\overline{R}}_{o}(2),\ldots{\overline{R}}_{o}\left( k_{o} \right) \right\}} \\ \end{aligned} \\ \end{array}$$ where *k*_*o* \< *K*_*i*. Then, assuming the spreading profile of city *o* is related to that of city *i* in the historical archive, as permitted by virtue of the validity of , we formulate the following optimization problem to predict the epidemic progression in city *o*: $$\begin{array}{r} \begin{array}{cl} {P_{0}:} & {\min\limits_{i \in (1,N)}f_{i}} \\ {{s.t.}\mspace{720mu}} & {(i)\mspace{720mu} f_{i} = \sum\left\lbrack w_{I}\left( {\overline{I}}_{o}(j) - {\overline{I}}_{i}^{(c)}(j) \right)^{2} + w_{R}\left( {\overline{R}}_{o}(j) - {\overline{R}}_{i}^{(c)}(j) \right)^{2} \right\rbrack} \\ & {\left( \text{ii} \right)\mspace{720mu}\mu_{L} \leq \mu \leq \mu_{U}} \\ & {\left( \text{iii} \right)\mspace{720mu} w_{I},w_{R} > 0.} \\ \end{array} \\ \end{array}$$ where *N* is the number of Chinese cities in the historical archive, *w*_*I* and *w*_*J* are weighting coefficients, *μ*_*L* and *μ*_*U* are the lower and upper bounds of the searching space, respectively. By solving the the nonlinear optimization problem, we can find the most closely resembling growth curve from the historical profiles, e.g., city *i*. Then, we apply the the augmented SEIR model with the profile code given in the parameter set for city *i* to predict the future spreading trend of city *o*. Furthermore, we can choose the top *n* best candidates with the smallest error as the candidate set for prediction, giving an average predicted propagation profile and a deviation range based on *n* best-fit profile codes. # 4 Results We apply the aforedescribed procedure to the data obtained so far for cities or regions in South Korea, Italy and Iran, as listed in. For each city or region, we identify a group of 10 profiles of best fit from the historical archive, and retrieve the corresponding sets of profile codes for generating the propagation profiles in the coming days. Using these 10 profiles, we produce an average progression profile, which is also accompanied by a deviation range at 95% confidence level. For Iran, we have only collected data for cities and provinces up to March 22, 2020, as data after this data was no longer available for individual cities and provinces. Figs, and show the data and the predicted number of infected individuals, each with a deviation range of the predicted average trajectory, for South Korea, Italy and Iran, respectively, and Figs, and show the corresponding cumulative values. Statistics of infection peaks are shown in. Statistics of the percentage of population eventually infected and the number of individuals eventually infected are shown in. Our key findings are summarized as follows: 1. The number of active infected individuals in South Korea, Italy and Iran is expected to continue to increase until the end of May 2020. While South Korea saw peaks in most of the cities between March 8, 2020 and April 9, 2020, most cities in Italy saw peaks between April 15 and mid May, while most provinces in Iran would saw peaks before the end of May 2020, as shown in the distributions of the peak times given in. 2. For South Korea, our results show that Daegu (with a population of 2,487,823) and Seoul (with a population of 10,018,537) are the two hardest hit cities, with 7,619 ±2,096 and 1,287 ±197 people eventually infected, accounting for 0.306% ±0.084% and 0.062% ±0.009% of the city’s population. In other cities in South Korea, the number of infected people will be fewer than 300, i.e., below 0.01% of the population. 3. For Italy, our results show that Lombardy (with a population of 10,078,012) and Emilia-Romagna (with a population of 4,459,477) have the highest number of cases, with about 90,000 and 30,000 people eventually infected, accounting for 0.802% and 0.604% of the region’s population. In other Italian cities or regions, the number of infected people will be below 10,000. 4. For Iran, we expect Tehran (with a population of 13,267,637) and Isfahan Province (1,292,283) to be most severely affected, reaching more than 9,000 (0.067%) and 1,695 ±92 (0.13%) cases eventually, respectively. Other Iranian cities will see more than 1,000 eventual infected cases. Our prediction shows that about 0.5% of the country’s population have been infected until May 14, 2020. 5. Provided the authorities continue to impose strict control measures, our model shows that most regions (cities and provinces) of these three countries would see peaks of infection cases before the end of May 2020. Hence, the first wave of epidemic will come under control by June 2020 for these three countries. Our prediction on the South Korean cities has revealed a very rapid progression of the epidemic, with 5,000 infections emerged within 10 days and peaks to be expected in most cities or regions within about 2 weeks. The Korean authorities have managed to test an overwhelmingly large number of people (140,000 until March 5, 2020) within a short time, thus preventing a large number of infected and infectious individuals not being quarantined in time. This strategy has an obvious advantage of offering a clear picture of the extents and locations of the infected individuals in the country at the early phase of the epidemic progression. The epidemic progression is found to be more rapid than typical, reflecting on the effectiveness of the control measures being taken. Italy has the second highest death toll after China, reaching 197 on March 6, 2020. The fatality rate is about 4%, which is the highest in the world. With infection cases soaring to 3,916 (on March 6, 2020), Italy had implemented control measures to contain the spread of the virus by shutting down schools and suspending public events in regions where outbreaks were reported. The epidemic has been expected to progress in a typical pace (with the present set of parameters), unless more stringent measures are in place. The situation in Iran is also critical, with the number of infected cases escalated to over 4,000 in less than 2 weeks. Iran has reported death of two lawmakers as of March 7, 2020, and has been struggling to control the contagion, which has spread to 31 provinces. The progression profile is again typical, however, expecting to peak in around 3 weeks. Like Italy, most cities show typical spreading profiles, and the peaks and subsequent decline are not expected to advance sooner unless more stringent measures are implemented to control the contagion. Finally, depending on the effectiveness of treatment, recovery rates vary, and judging from the predicted trends shown in Figs, and, the epidemic progressions for the three countries are expected to subside by the end of May, with South Korea expected to recover sooner than the others. # 5 Conclusion The spreading of the 2019 New Coronavirus Disease (COVID-19) has evolved to a global contagion, which has spread to 87 countries within two months and more 190 countries until May 14, 2020. The numbers of confirmed infection cases in South Korea, Italy and Iran have surged in late February and early Match, and continued to progress in the last two months, reaching 10,926, 222,104 and 112,725, respectively, on May 14, 2020. The global fatality rate, however, has increased from below 3% on March 8, 2020 to more than 6% on May 14, 2020. In this study, we build on the results of our previous work that have established a library of parameters of an augmented SEIR model, corresponding to the historic spreading profiles of 367 cities in China. This library forms a set of profile codes that cover a variety of possible epidemic progression profiles. By comparing the early incomplete data of epidemic progression collected for a specific population with the historic profiles, we select a few candidate profiles from the historic archive using a nonlinear optimization procedure. The corresponding profile codes of the selected historic progression profiles can then be used to produce estimates of the future progression for that specific population. We apply this method to predict the spreading profiles for South Korea, Italy and Iran, and specifically, to provide a method for estimating the proportion of infected population in these countries. Results have shown that the three countries would see infection peaks in most cities or provinces before the end of May 2020, with South Korea’s cases reaching their peaks much earlier than the others. The percentage of population eventually infected will be less than 0.3%, 0.5% and 0.5% for South Korea, Italy and Iran, respectively. The epidemic is expected to come under control before June 2020 in these countries, and depending on the effectiveness of treatment, particular cities may see full recovery or zero infection sooner or later than others. It is worth noting that the epidemic progression in South Korean cities are found to be more rapid than typical, implying that the authorities might have taken effective measures to control the spread. The predicted progressions for Italy and Iran, on the other hand, are found to display profiles that are typical of those in the historical archive, and unless more stringent measures are taken, the peaks and subsequent decline of the infection numbers will unlikely come sooner or more rapidly than the predicted trajectories. Finally, we should stress that the proposed data-driven coding method is applicable to predicting epidemic progression in any given population and the accuracy of prediction will depend on the adequacy of the available data in allowing a reliable match to be identified from the historical archive. In this study, the profile predicted in early March turned out to be consistent with the actual data collected up to mid May for the three countries. However, due to limited coverage of the data collected, the data-driven model may not perform satisfactorily if it is applied to a new epidemic which has a significantly different set of spreading characteristics (i.e., model parameters being significantly different from those collected in the historical database) or in a population or country having a significantly different contact topology, social behavior, travel patterns, effectiveness of control as well as climate. In fact, deviation was observed in the Hong Kong data in late March when compared with the profile predicted earlier in February, and the deviation was found to be a result of an unexpected surge in inbound travelers which were mostly overseas students returning from the UK and USA. Thus, had superspreader events or other unexpected events occurred, the spreading profile predicted by the data-driven model might deviate from the actual pattern. Furthermore, the model does not consider the effects of different control strategies. Continuous effort will be made to enhance the database so as to widen the coverage of possible progression profiles as well as to incorporate the effects of different control measures on the epidemic progression profiles. # Supporting information 10.1371/journal.pone.0234763.r001 Decision Letter 0 Goletti Delia Academic Editor 2020 Delia Goletti This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 23 Apr 2020 PONE-D-20-06792 Prediction of COVID-19 Spreading Profiles in South Korea, Italy and Iran by Data-Driven Coding PLOS ONE Dear prof Chi K. TSE, the manuscript is very interesting. Please, revise according to the referess points: 1\. present more clear the results 2\. update the results with today data 3\. give a prospetive of your findings Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. We would appreciate receiving your revised manuscript by Jun 07 2020 11:59PM. When you are ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission- guidelines#loc-laboratory-protocols> Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. We look forward to receiving your revised manuscript. Kind regards, Delia Goletti, M.D., Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: N/A \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: In the submitted manuscript, the authors provide a model of prediction of SARS-CoV-2 spreading in three main Countries: South Korea, Italy, and Iran. Their model is based on a previous publication by the same authors and on data collected in 367 Chinese cities. The data and the conclusion are clearly presented and the manuscript is easy to read, but some points have to be clarified and better discussed, especially in the light of updated observed data of viral spreading in the regions under investigation. One point that can be better explained, to make it easier for the large audience of Plos One to understand the hypothesis of the authors, are the variables taken into account for the construction of the model. Indeed some social/political/climatic variables could differently affect viral spreading in different countries and/or cities. The effect of such factors should be at least discussed in the conclusions. Another important point to discuss is the difference between the predictions presented and the actual situation (see for example line 200-205). We do understand that the authors performed the analysis having the data till the 6th of March but they now have access to the observed cases in the different countries and regions and some of the observations importantly deviate from the model of spread here presented. The information on observed cases should be updated and discussed proposing criteria for the adjustment of the presented model. In paragraph line 212-215, why do the authors conclude that the epidemic will end before June 2020, on the basis of which data? Minor points: To speak about SARS-CoV-2 spreading would be more appropriated than COVID-19 spreading. To indicate that for infected individuals the authors mean the people tested positive would also be appropriated, especially when presenting the data as proportion of population infected. Indeed, only seroprevalence studies may actually estimate the proportion of population that underwent infection. The names of Italian administrative regions should be corrected in text and figures: Lombardy, Emilia Romagna, Apulia, Basilicata. Reviewer \#2: The manuscript describes “Prediction of COVID-19 Spreading Profiles in South Korea, Italy and Iran by Data-Driven Coding.” The title is interesting, however, the manuscript is not well organized and understandable because of the lack of a good explanation of the results. The population of each city, province, and region (with a reference) should be mentioned in the results. For example, lines 194-197, the population of Daegu and Seoul should be mentioned. If Seoul’s population is x million with x people eventually infected, it is x% of the city’s population. In addition, it seems that the mentioned places of Iran are Provinces (not cities) and the population of whole province should be considered, however, they should be spelled also correctly (e.g. Fig.6, f, h, i, k, l). Moreover, some parts of the manuscript do not have related references which should be considered. Finally, the limitations of the present study should be addressed. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0234763.r002 Author response to Decision Letter 0 19 May 2020 (Already uploaded with the revised manuscript) We would like to thank the Editor and all reviewers for their helpful comments and suggestions, which have been taken into consideration in the revision of this paper. Reviewer: 1 Comments to the Author In the submitted manuscript, the authors provide a model of prediction of SARS- CoV-2 spreading in three main Countries: South Korea, Italy, and Iran. Their model is based on a previous publication by the same authors and on data collected in 367 Chinese cities. The data and the conclusion are clearly presented and the manuscript is easy to read, but some points have to be clarified and better discussed, especially in the light of updated observed data of viral spreading in the regions under investigation. Authors’ Response: Thank you for the positive support and encouragement. One point that can be better explained, to make it easier for the large audience of Plos One to understand the hypothesis of the authors, are the variables taken into account for the construction of the model. Indeed some social/political/climatic variables could differently affect viral spreading in different countries and/or cities. The effect of such factors should be at least discussed in the conclusions. Authors’ Response: Thank you for the positive support and encouragement. The point raised is indeed very legitimate for data-driven model. We have included a brief discussion at the end of the paper (Conclusion) to highlight this issue, and specifically, performance of any data-driven model would depend on the breath of coverage of the historical data. Thus, we do admit that the model needs to be continuously updated to cover different set of spreading characteristics. At present, we the data is limited, and the data-driven model may not perform satisfactorily if it is applied to a new epidemic which (i.e., model parameters being significantly different from those collected in the historical database) or in a population or country having a significantly different contact topology, travel patterns, effectiveness of government's control as well as climate. Another important point to discuss is the difference between the predictions presented and the actual situation (see for example line 200-205). We do understand that the authors performed the analysis having the data till the 6th of March but they now have access to the observed cases in the different countries and regions and some of the observations importantly deviate from the model of spread here presented. The information on observed cases should be updated and discussed proposing criteria for the adjustment of the presented model. Authors’ Response: The manuscript was finished on March 8, 2020. Now, two months have passed. We have updated the results based on new datasets. All the figures are updated. It is found that the general profile pattern remains the same for the three countries under study, despite some adjustment on the actual number of infected cases predicted and the exact times of the peaks. We believe that the progression profile is basically being captured by the historical data of the 367 cities collected. However, deviation can still be expected due to outlier events such as superspreader events and irregular travel patterns that may cause deviation from the general patterns. For instance, we found significant deviation in the Hong Kong data compared with a similar prediction conducted earlier, and there was unexpected surge in the number of infected cases in mid March due to an unexpectedly large number of inbound travelers as a result of overseas students returning from UK and USA. Regarding the updates in this revised version, we should mention that after March 22, Iran no longer publishes detailed data for individual provinces. Thus, we cannot achieve detailed data for individual provinces for forecasting after March 22. The main updates have been included in the blue texts in the Abstract, Section 4 and Conclusion of the revised paper. In paragraph line 212-215, why do the authors conclude that the epidemic will end before June 2020, on the basis of which data? Authors’ Response: We provided further explanation in the revised paper as follows. Basically, from the progression trends of the epidemic these three countries, provided control measures continue to be in place, our model show that the number of confirmed cases of COVID-19 infection in most regions of these three countries will peak before the end of May 2020. Hence, the first wave of epidemic progression would come under control before the end of May 2020. This has been mentioned in Page 3 and Page 7 of the revised paper. Minor points: To speak about SARS-CoV-2 spreading would be more appropriated than COVID-19 spreading. Authors’ Response: According to WHO (<https://www.who.int/emergencies/diseases/novel-coronavirus-2019/technical- guidance/naming-the-coronavirus-disease->(covid-2019)-and-the-virus-that-causes- it), the disease is abbreviated as COVID-19 and the virus is SARS-CoV-2. We have found that the literature has more adopted “COVID-19 spreading”. In the revised paper, we have added SARS-CoV-2 in the Introduction but has continued to adopt only COVID-19 for simplicity. To indicate that for infected individuals the authors mean the people tested positive would also be appropriated, especially when presenting the data as proportion of population infected. Indeed, only seroprevalence studies may actually estimate the proportion of population that underwent infection. Authors’ Response: Thank you for pointing this out. In Section 2, as far as our study is concerned, we have clarified the kind of data we have, corresponding to the number of infected cases. See Section 2 on Page 3 of the revised paper. The names of Italian administrative regions should be corrected in text and figures: Lombardy, Emilia Romagna, Apulia, Basilicata. Authors’ Response: We have tried to unified all the names for the cities and regions of the three countries. Reviewer: 2 Comments to the Author The manuscript describes “Prediction of COVID-19 Spreading Profiles in South Korea, Italy and Iran by Data-Driven Coding.” The title is interesting, however, the manuscript is not well organized and understandable because of the lack of a good explanation of the results. Authors’ Response: We have updated the results and improved the descriptions in the revised paper, so that the idea of the proposed data-driven model is more clearly explained to the readers. The updates are highlighted in blue in the revised data. We have included more information, as suggested, such as information about populations of individual regions. The population of each city, province, and region (with a reference) should be mentioned in the results. For example, lines 194-197, the population of Daegu and Seoul should be mentioned. If Seoul’s population is x million with x people eventually infected, it is x% of the city’s population. Authors’ Response: See Table 1 of the revised paper. We have also modified the description in the text as per your suggestion. In addition, it seems that the mentioned places of Iran are Provinces (not cities) and the population of whole province should be considered, however, they should be spelled also correctly (e.g. Fig.6, f, h, i, k, l). Authors’ Response: Table 1 indeed are provinces’ populations. Moreover, some parts of the manuscript do not have related references which should be considered. Authors’ Response: The following references have been cited: • Jia JS, Lu X, Yuan Y, Xu G, Jia J, Christakis NA. Population ow drives spatio- temporal distribution of COVID-19 in China. Nature 2020, 29: 1-1. doi: <https://doi.org/10.1038/s41586-020-2284-y10.1126/science.aba9757>. • Du Z, Wang L, Cauchemez S, Xu X, Wang X, Cowling BJ and Meyers LA. Risk of transportation of 2019 novel coronavirus disease from Wuhan to other cities in China. Emerg Infect Dis 2020; 26(5) DOI: 10.3201/eid2601.200146. • Wu JT, Leung K, Leung GM. Nowcasting and forecasting the potential domestic and international spread of the 2019-nCoV outbreak originating in Wuhan, China: a modelling study. The Lancet 2020; 395(10225): 689-97. • Humanity tested. Nat Biomed Eng 2020; 4: 355-356. • Zhan C, Tse C, Lai Z, Chen X and Mo M. General model for COVID-19 spreading with consideration of intercity migration, insuffcient testing and active intervention: application to study of pandemic progression in Japan and USA, Preprint available at medR[xiv.org](http://xiv.org), 2020, doi: <https://doi.org/10.1101/2020.03.25.20043380>. Finally, the limitations of the present study should be addressed. Authors’ Response: Indeed, our data-driven model does have its weakness. due to limited coverage of the data collected, the data-driven model may not perform satisfactorily if it is applied to a new epidemic which has a significantly different set of spreading characteristics (i.e., model parameters being significantly different from those collected in the historical database) or in a population or country having a significantly different contact topology, social behavior, travel patterns, effectiveness of control as well as climate. In fact, deviation was observed in the Hong Kong data in late March when compared with the profile predicted earlier in February, and the deviation was found to be a result of an unexpected surge in inbound travelers which were mostly overseas students returning from the UK and USA. Thus, had superspreader events or other unexpected events occurred, the spreading profile predicted by the data-driven model might deviate from the actual pattern. We have provided some comments in the Conclusion. 10.1371/journal.pone.0234763.r003 Decision Letter 1 Goletti Delia Academic Editor 2020 Delia Goletti This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 3 Jun 2020 Prediction of COVID-19 Spreading Profiles in South Korea, Italy and Iran by Data-Driven Coding PONE-D-20-06792R1 Dear Dr. Chi K. TSE,  We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Delia Goletti, M.D., Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0234763.r004 Acceptance letter Goletti Delia Academic Editor 2020 Delia Goletti This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 25 Jun 2020 PONE-D-20-06792R1 Prediction of COVID-19 Spreading Profiles in South Korea, Italy and Iran by Data-Driven Coding Dear Dr. Tse: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Delia Goletti Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist. [^2]: Current address: Department of Electrical Engineering, City University of Hong Kong, Kowloon, Hong Kong, China

# Introduction Obesity is a major public health concern resulting from a mixture of environmental, genetic, neural and endocrine factors. The distal gastrointestinal tract harbors \>10¹⁴ microorganisms with significant differences in the taxonomy and concentration of the bacteria along the digestive track reflecting major variations in the environmental niche. There are complex links between the digestive microbiota and obesity, and a new area of research has emerged based on the links between intestinal microbiota, weight change, the relief of malnutrition, and the use of antibiotics and probiotics. The highest bacterial concentration, approximately 10^11–12 microorganisms per gram of content, resides in the colon and is mainly comprised of anaerobes. In contrast, much lower bacterial concentrations, approximately 10^3–4 microorganisms per mL of content, are present in the upper two- thirds of the small intestine. *Lactobacillus* sp., *Escherichia coli* and *Enterococci* have been found as the predominant species in the duodenum and jejunum. Probiotics and antibiotics can alter the intestinal flora, and the role of *Lactobacillus*, *Bifidobacteria* or *Enterococcus* is easier to understand as the duodenojejunal flora contains mostly these species. The small intestine is responsible for most nutrient digestion and absorption by humans. Proteins and lipids are almost completely absorbed before entering the large intestine, along with simple sugars, such as glucose, very few disaccharides (lactose and sucrose), and a portion of starch. In the colon, microorganisms ferment undigested starch (including resistant starch), unabsorbed sugars, plant cell wall polysaccharides and mucins into the short-chain fatty acids (SCFAs) butyrate, acetate and propionate, which provide approximately 10% of the calories humans absorb. Human microbiota projects are being initiated throughout the world, with the goal of correlating human physiological phenotypes with the structures and functions of their indigenous microbial communities. However, the nature of the changes in the intestinal microbiota associated with obesity is a subject of controversy, and major discrepancies between different studies have appeared. The development of experimental models to study the relationship between gut microbiota and obesity has mostly been based on the study of feces. Recently, we showed that fecal analysis may not be the optimal method to examine the link between obesity and gut flora and that more focus should be given to the microbiota of the small intestine because this is where the calories are absorbed. However, to date, few studies have tested the microbiota of the upper intestinal track, and to the best of our knowledge, this compartment of microbiota has been explored by metagenomic analysis only once in humans, namely on ileostomy effluent samples collected from individuals who have had an ileostomy for 20 years. We report here the first metagenomic analysis of duodenal samples from obese and normal volunteers to examine the microbial population and functions of upper intestinal microbiota. # Materials and Methods ## Human Subjects Duodenal samples from healthy volunteers were collected in the framework of a clinical study (mrtm02.01) on gastrointestinal lipolysis performed with a solid- liquid test meal. Duodenal samples from obese patients (BMI\>28) were collected under similar test meal conditions. These studies were not initially designed for studying gut microbiota at the time they were performed, and samples had been kept frozen in sterile conditions at -80°C since 2003. ## Ethics statement Experiments were performed at the CPCET (Centre de Pharmacologie Clinique et d'Etudes Thérapeutiques, La Timone Hospital, Marseille) after the clinical protocol was accepted by the institutional review board of the local ethics review committee (CCPPRB, Comité Consultatif de protection des Personnes dans la Recherche Biomédicale, Marseille). Written informed consent was obtained from all participating patients. ## Test meal The mixed solid/liquid meal used for the clinical studies contained 80 g string beans, 90 g beef, 70 g French fries, 10 g butter, 15 mL olive oil, 5 mL sunflower oil and water for a total volume of 700 mL. Before mixing, the string beans, the beef and the French fries were put into a mincer with 2 mm holes. ## Experimental setup for collecting duodenal samples After an overnight fasting period, the volunteers/patients were intubated with a single-lumen duodenal tube (outside diameter 5 mm) and a separated single-lumen gastric tube (outside diameter, 3 mm) as previously described. The distal end of the duodenal tube was placed at the ligament of Treitz for aspiration of duodenal contents (-10 cm H2O). The test meal containing a non-abasorbable marker (PEG4000) was introduced into the stomach through the gastric tube using a 50 mL syringe over a period of 5 minutes. The duodenal fluid was then collected continuously by aspiration and fractioned every 15 minutes. Duodenal samples (1 mL) were immediately mixed with 1 mL glycerol and 40 μL protease inhibitors and frozen in liquid nitrogen before storage at -80°C. The samples selected for the present study were all collected at an average time of 90 minutes after meal intake. ## 16S rDNA V6 Pyrosequencing Total DNA was extracted from the samples using a method modified from the Qiagen stool procedure \[QIAamp DNA Stool Mini Kit (Qiagen, Courtaboeuf, France)\]. Primers were designed to produce an amplicon length (576 bp) that was approximately equivalent to the average length of reads produced by the GS FLX Titanium (Roche Applied Science, Meylan, France) as previously described. The primer pairs commonly used for gut microbiota were assessed in silico for sensitivity to sequences from all phyla of bacteria in the complete Ribosomal Database Project (RDP) database. Based on this assessment, the bacterial primers 917F and 1391R were selected. The V6 region of 16S rRNA V6 was pyrosequenced with unidirectional sequencing from the forward primer with one-half of a GS FLX Titanium PicoTiterPlate Kit 70×75 per patient with the GS Titanium Sequencing Kit XLR70 after clonal amplification with the GS FLX Titanium LV emPCR Kit (Lib-L). ## Metagenomic deep sequencing using Illumina MiSeq Five samples of weight gain individuals and five samples of normal weight individuals were extracted using the protocol 1 and were amplify by illustra GenomiPhi V2 DN Amplification Kit(GE Healthcare Bio-Sciences Corp. Piscataway, NJ 08855–1327, USA) to get enough genomic DNA. The DNAg of samples were then pooled and sequenced on the MiSeq Technology (Illumina, Inc, San Diego CA 92121, USA) with paired end and barcode according to the Nextera XT library kit (Illumina). The genomic DNA was quantified by a Qubit assay with the high sensitivity kit (Life technologies, Carlsbad, CA, USA) and dilution was performed to require 1ng of each sample as input. The « tagmentation » step fragmented the genomic DNA. Then limited cycle PCR amplification completed the tags adapters and introduce dual-index barcodes. After purification on Ampure beads (Lifetechnolgies, Carlsbad, CA, USA), the libraries were then normalized on specific beads according to the Nextera XT protocol (Illumina). Normalized libraries are pooled into a single library for sequencing on the MiSeq. The pooled single strand library was loaded onto the reagent cartridge and then onto the instrument along with the flow cell. Automated cluster generation and paired-end sequencing with dual index reads was performed in a single 39-hour run in a 2x250-bp. A total information of 6.5 G bases was obtained from a 695K/mm2 density with 91.17% (14,763,000 clusters) of the clusters passing quality control (QC) filters. Within this pooled run, the average index representation was determined 3.80%. The average 478, 239 paired end reads were filtered according to read quality. ## 16S rRNA pyrosequencing analysis The 16S raw data for all samples was processed and compared using QIIME pipeline 1.7.0, which contains a suite of Python scripts for data analyses. The raw reads were demultiplexed with split_libraries.py and were trimmed using a minimum read length of 150 bp and an average quality score of 30. One mismatch was allowed along the primer sequences. The number of homopolymers authorized in a sequence was limited to 6. The high quality reads were classified to their corresponding operational taxonomic unit (OTU) at 97% similarity using the open-reference OTU picking strategy in QIIME pipeline. The representative OTU sequences were taxonomically classified using the RDP classifier algorithm implemented in QIIME and using the most recent Greengene database gg_12_10 (<http://greengenes.secondgenome.com/downloads/database/12_10>). The NAST multi- aligner implemented in QIIME performed the multiple sequence alignment that is used to build phylogenic trees. Before building the phylogenic tree, the chimera identification sequence was performed by USEARCH. We also applied a QIIME random subsampling normalization to the OTU table of samples for the sample with the fewest read numbers. Moreover, the OTU table of all samples was filtered, discarding all OTUs that did not contain a minimum of three reads. The QIIME results, including tree, mapping and OTU files, were import into Phyloseq, an R package, for manipulation, alpha indices determination and graph visualizations. The 16S pyrosequencing raw data has been submitted to the SRA archive under the accession number SRP059828. ## Metagenomic analysis The metagenomic paired-reads were assessed for quality (minimum average of 25) and assembled using the Panda package. The prodigal software performed the detection of open reading frames from metagenomic assembled reads (parameter—p = m) data. The KEGG assignment of the ORFs against the genepep KEGG database was provided by BLASTP, with an E-value of 10⁻⁴, a minimum bit score of 50 and sequence coverage \>70. The corresponding tables from geneID to KO numbers, from KO to EC Enzyme or to pathways numbers, as well as the KEGG mapper tool (<http://www.genome.jp/kegg/mapper.html>) were used to build the different metabolic pathways. The assignment to Cluster of Orthologous Group (COG) was performed using RPS-Blast against the COG position-specific scoring matrix (PSSM) from the NCBI Conserved Domain Database, with a minimum E-value of 10⁻⁴, a bit score \> 50 and sequence coverage \> 70. The blast results were parsed for COG numbers that provided a classification into the different COG categories. The metagenomic raw data has been submitted to the SRA archive under the accession number SRP059828. ## Statistical analysis The relative abundance differences between obese and normal populations were analyzed using the nonparametric (Kruskal-Wallis test) statistical method. For data comparisons, we used EpiInfo version 6.0 software (Centers for Disease Control and Prevention, Atlanta, GA, USA). A *p* value \< 0.05 was considered significant. The raw p-values were adjusted using the Benjamini-Hochberg correction. # Results ## Subject and sample Characteristics We tested five lean healthy volunteers (3 males) and five obese patients (3 males). Lean subjects had a BMI of 20.7±2.3 kg/m² (19–24.5) and a mean age of 29.6 (26–34), whereas obese subjects had a BMI of 36.8±8.4 kg/m² (28.3–47.2) and a mean age of 39 (26–54). Although the variance of BMI in the obese group was high, the difference in BMI between the two groups was significant (*p*\<0.01) and allowed a clear distinction between lean and obese subjects, in the absence of body composition data. At the time of collection of the duodenal samples used for the metagenomic study (90 min after meal ingestion), no significant difference (*p*\>0.05) in the rate of gastric emptying was observed between lean (68.5 ± 21.9%) and obese (68.4 ± 25.5%) subjects, who received the same test meal. In both cases, around two third of the meal had already been emptied from the stomach and through the duodenum together with gastric, pancreatic and biliary secretions. Changes in microbiota profile due to differences in gastric emptying were therefore unlikely. ## Duodenal gut microbiota The microbial structure and composition of the duodenal gut microbiota was characterized by a 16S rRNA pyrosequencing approach. The average 16S rRNA pyrosequencing read length was 374 bp. summarizes the number of pyrosequencing- trimmed reads obtained for each of the ten samples. The analysis of the high- quality trimmed reads, which included a random subsampling normalization, indicated that the gut microbiota of the obese and normal individuals was composed of 11 bacterial phyla. Although we observed an inter-individual variability in taxonomic composition, *Firmicutes* and *Actinobacteria* were the most predominant phyla of the bacterial composition of the duodenal microbiota within obese and control groups. These predominant phyla were followed by less abundant ones, including *Proteobacteria*, *Fusobacteria*, TM7, *Bacteroidetes* and *Tenericutes*. Overall, the phylum taxonomic profile was very similar between the obese and control groups, with small differences for *Firmicutes* (62% in the control group vs 67% in the obese group; *p* = 0.91) and *Proteobacteria* (9.5% in the control group vs 4% in the obese group; *p* = 0.25). Compared with distal gut microbiota, the microbiota of the duodenal site showed major differences exemplified by the almost complete absence of the *Bacteroidetes* phylum (only present at 0.2%) and by the substantial abundance of *Actinomyces* and *Streptococcus* OTUs. We merged OTUs at the genus and species taxonomic ranks to compare the abundance levels for the individuals within each sample category. This comparison between the obese and control categories did not show any significant differences for all tested species and genera except for the *Rubrobacter* genus (*p* = 0.019). With more observed species, the global richness of the obese group was higher than the control group as shown by rarefaction curves. However, this difference was not significant using a 3% OTU distance and the Kruskal-Wallis statistical test. In addition, the alpha diversity Simpson indices indicated a weak biodiversity, as only a few abundant OTUs dominated the microbiota composition that mainly comprised *Streptococcus* (30–32%), *Actinomyces* (12–17%), *Propionibacterium* (3–8%) and *Granulicatella* (2–4%) genera. The many remaining OTUs were in weak abundance (most of them \< 1%). We also investigated the distribution of aerobic and facultative anaerobic genera residing in the duodenal microbiota of the obese and control groups using the taxonomic classification provided by 16S amplicon analysis. The difference in aerobic genus counts between the obese and control groups was not significant, with 55 and 52 genera identified, respectively. Likewise, the difference in anaerobic genus counts was also not significant, with 55 and 52 genera identified for the obese and control groups, respectively. However, the relative abundance of aerobic and anaerobic genera indicated a significant difference between the obese and control groups (Chi-square test; *p* \< 0.001). Compared with the control group, the obese group presented a higher proportion of anaerobic genera and a lesser proportion of aerobic genera. ## Duodenal gut microbiome Using a metagenomic approach, we investigated the functional capabilities of the duodenal microbiome within obese and control samples. We performed COG and KEGG metabolism category classifications. The percentage of COG assigned to carbohydrate metabolism tended to be lower in the obese group than that in the control group (*p* = 0.15). In contrast, the proportion of COG assigned to lipid metabolism tended to be higher in the obese group than in the control group (*p* = 0.1). The KEGG analysis gave results equivalent to those of the COG classification, including those for lipid and carbohydrate metabolisms. We investigated the type and the relative abundance of the enzymes involved in lipid pathways, including fatty acid biosynthesis and degradation. In the fatty acid degradation pathway, we found that the Acyl-CoA dehydrogenase (EC 1.3.99-) targeting the early enzymatic reaction of fatty acid beta-oxidation was enriched (COG and KEGG analyses) in the obese group compared with the control group (*p* = 0.0018). The degradation of a fatty acid requires multiple repetitions of the fatty acid beta-oxidation process (Lynen helix) that leads to the removal each round of two carbon atoms from the acyl chain and to the release of one Acetyl- CoA molecule for the Krebs’ cycle. Moreover, other Acyl-CoA dehydrogenases (EC:1.3.3.6 and EC:1.3.8.3) targeting specific acyl chain lengths but catalyzing the same enzymatic reaction were only detected in the obese group. The fatty acid biosynthesis pathway highlighted the presence of many enoyl-\[Acyl carrier protein\] reductase enzymes, including FabK, FabI, FabL. These enzymes are components of the type II fatty acid synthase system (Fas) and catalyze the terminal reaction in the fatty acid elongation cycle. The diversity of the enoyl reductase enzymes results from different substrate specificities that can enhance the regulation and the distribution of the products synthetized in the pathway. In our data, FabK was enriched in the control (*p* = 0.027) compared to the obese group; FabL and FabI were only detected in the control group. In addition, the relative abundance of the fatty acid synthase (Fas) tended to be higher in the control group (*p* = 0.07). An inspection of the glycerophospholipid metabolism pathway revealed the presence of phospholipase A1 (EC:3.1.1.32) only in control group (3 of 5 individuals, *p* = 0.05). An analysis of the corresponding protein best BLAST hits indicated that the sequences were homologous to the outer membrane phospholipase A (OMPLA), which is widespread among gram-negative bacteria and is known as a virulence factor in *Campylobacter coli* and *Helicobacter pylori*. In addition, we found that the duodenal microbiota of the obese group showed a reduced abundance of genes encoding sucrose phosphorylase (EC:2.4.1.7) (*p* = 0.015) and 1,4-alpha-glucan branching enzyme (EC:2.4.1.18) (*p* = 0.046), suggesting an alteration of the sucrose/glycogen balance in the obese flora. # Discussion We used pyrosequencing of 16S rRNA amplicons and metagenomic analysis to compare the duodenal microbiota in obese individuals to normal weight individuals. Although our samples were collected under similar conditions of test meal and gastric emptying they were taken from a study that was not initially designed for studying gut microbiota, they were kept frozen in sterile conditions at -80°C, eliminating the possibility of contamination. Moreover, before analyses, we verified that all of our samples had a good DNA load. To date, most of the 16S rRNA sequencing- and metagenomic-based studies have analyzed the distal part of the gut using feces and reported differences in the relative abundances of bacterial communities in the gut microbiota of obese versus normal weight people. However, it is noteworthy that the current 16S rDNA studies of gut microbiota within obese populations were not able to detect bacterial concentrations that were below 10⁷ organisms per gram of feces because of sequencing capability limitations. As a result, these sequencing- based methods are generally not able to access the complete richness of a feces sample and are biased by the heterogeneity of the copy number of the 16S rRNA gene that is present in an individual bacterial genome, which can lead to an overestimation of bacterial proportions. Indeed, the characterization of the 10^11–12 microorganisms per gram of feces that was used in these studies remains superficial. In contrast with the distal human gut, the bacterial concentrations in the duodenum reach only 10^3–4 cells per mL of content. Because of the lower bacterial concentration present in the duodenum, we were able to characterize the full species richness of samples by deep sequencing, as shown in the rarefaction curves. Indeed, when the sequencing effort is adequate to collect the complete species richness of a sample, the curve tends to the asymptote. A limitation of our study was that we used 12-year-old frozen samples. Excessive degradation of DNA reduces the efficiency of shotgun sequencing and previous studies showed that storing conditions of stool samples modestly affect the structure of their microbial community. It was found that the structure of microbial community is not affected when fecal samples are freeze and stored immediately. Lauber *et al*. reported that the phylogenetic structure of the microbiota did not significantly differ between three and fourteen day old fecal samples stored at a range of temperatures. Moreover it was found that fecal samples kept at -80°C for up to six months also retain a microbiota that was similar in composition to a fresh sample from that individual. Based on these, we believe that the quality and the microbial community of our samples remained stable as all our samples were immediately frozen after collection at −80°C and were never thawed and refreeze. We found that the duodenal microbiota presents important differences compared to the microbiota of feces. Indeed, the predominant phyla of the duodenal microbiota were *Firmicutes* and *Actinobacteria*, whereas *Bacteroidetes* were almost completely absent. This may be related to a limited availability of mucin as carbon source for *Bacteroidetes*, which are characterized by a high number of genes coding mucin-degrading enzymes like glycosyl hydrolases, proteases/peptidases, sulfatases and sialidases/neuraminidases. Indeed, the mucin layer is much thinner in the small intestine than in the stomach and the colon. Approximately 60% of the genera OTUs belonged to *Streptococcus*, *Actinomyces*, *Propionibacterium* and *Granulicatella*. Similarly, in a recent study, the phylogenetic mapping of the small intestinal metagenome of three different ileostomy effluent samples from a single individual indicated that *Streptococcus* sp., *Escherichia coli*, and *Clostridium* sp. were most abundant in the small intestine. Similarly, in a culture-based study in infants, *Actinobacteria* and *Firmicutes* were found to be the dominant phyla in ileostomy samples, whereas *Bacteroidetes* were only detected following the reversal of the ileostomy in the final fecal sample. In contrast, studies of stool microbiota revealed that *Firmicutes* and *Bacteroidetes* were the predominant phyla. Moreover, we found that obese individuals had a significantly higher proportion of anaerobic bacteria in their duodenal microbiota. This difference was mostly associated with the presence of *Veillonella*, *Bulleidia* and *Oribacterium*. Previously, *Veillonella* was detected significantly more in the gut microbiota of children with type 1 diabetes. *Veillonella* are able to ferment glucose and lactate to propionate, acetate and succinate. The Acyl-CoA dehydrogenase (FAD) involved in the first enzymatic reaction of fatty acid beta-oxidation was enriched in the obese group. This high occurrence of Acyl-CoA dehydrogenase in obese subjects might be associated with a higher beta-oxidation capacity and energy mobilization. Conversely, a higher production of fatty acids would be favored in normal subjects with both a low occurrence of FAD and a high occurrence of OMPLA phospholipase A1. Indeed, OMPLA is particularly active in *E*. *coli* cells with a ‘fad’ mutation and a perturbed cell envelope structure. The *E*. *coli* fad strain does not perform beta- oxidation and is characterized by an appreciable turnover of phospholipids and sizeable amounts of fatty acids resulting from phospholipid hydrolysis. If one assumes that obese patients have an excessive uptake of food and particularly fat, their intestinal microbiota may have adapted to high levels of dietary fats and free fatty acids released upon gastrointestinal lipolysis. Free fatty acids could thus be used as a carbon and energy source for microbial growth. High fat loads are also associated with increased endotoxemia, suggesting that fat and its lipolysis products have a deleterious effect on gut microbiota, leading to LPS release. The harvest and degradation of fatty acids by bacteria might be viewed as an adaptive response to their antibacterial effects. Duodenal sensing mechanisms linked to the release of fatty acids and their levels might also be impacted by microbiota. Fatty acids released in the upper duodenum trigger the release of CCK which first stimulates pancreatic secretion and thus digestion. CCK has however a dual function and also acts as a satiety agent together with other gut hormones like GLP-1 and PYY. It has been suggested that obesogenic microbiota induced high-fat feeding may alters CCK action and lead to dysregulation of glucose homeostasis. The mechanism by which obesogenic microbiota may induce CCK resistance has not been explored yet in humans. One hypothesis could be that the microbiota of obese patients may lower fatty acid levels by degrading them more efficiently. This could impact the fatty acid- induced release of gut hormones involved in satiety mechanism and regulation of glucose homeostasis. In conclusion, the low bacterial concentration and particular taxonomic composition of the duodenojejunal microbiota makes the evaluation of its variation by stool sample analysis extremely difficult. To the best of our knowledge, this is the first time that human duodenal samples have been analyzed by metagenomic techniques, and we found that the duodenal microbiota of obese individuals shows an increased capacity to degrade fatty acids, whereas the flora of control individuals shows an increased capacity to store fatty acids. Because the concentration of living bacteria is much higher in fermented products used as probiotics than in the duodenal flora (10⁹ vs 10⁵ microbes per mL, respectively), the impact of probiotics is probably more important on the duodenal than on the distal gut microbiota. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: EA FC BH DR. Performed the experiments: EA FA DB CR RL. Analyzed the data: EA FA DB BH FC. Contributed reagents/materials/analysis tools: JCL CR CM. Wrote the paper: EA FA BH FC DR.

# Introduction Pulmonary embolism (PE) is a serious multiple-organ-involved disease commonly originating from a deep venous thrombosis (DVT), with a morbidity of approximately 69 per 100,000 people. Patients who have been treated for PE have an 8% mortality rate, whereas untreated PE patients have a mortality rate as high as 30%, according to a previous study performed around 20 years ago. PE can be broadly classified as either massive or submassive. Patients with submassive PE can be treated with anticoagulation medication alone as they are generally hemodynamically stable, whilst patients with massive PE usually present with hemodynamic instability and are treated with either pulmonary embolectomy or thrombolytic therapy. The major risk factors for the development of PE include intrinsic factors such as previous venous thromboembolism and age \>70 years, and acquired factors such as malignancy, cancer chemotherapy, paralysis, major or lower limb trauma, lower limb orthopedic surgery, general anesthesia for \>30 minutes, heparin-induced thrombocytopenia and antiphospholipid antibodies. Other minor risk factors for PE are an inherited hypercoagulable state, obesity, pregnancy or puerperium, estrogen therapy, prolonged immobility, nephrotic syndrome, etc.. Although these risk factors are important for PE control and prevention in clinical practice, about a quarter of patients with PE have no apparent provoking risk factor, half have a temporary provoking risk factor such as a history of PE and recent surgery, and a quarter have complications from various cancers. The diagnosis of PE is a relatively complex and rigorous process because ‘confirmed PE’ indicates the need for PE-specific treatment and ‘excluded PE’ justifies the validity of withholding such treatment. To establish a PE diagnosis, the following symptoms, signs, history and medical examinations must be considered: clinical presentation, such as dyspnea, chest pain, cough, hemoptysis, syncope, tachypnea, tachycardia, fever, etc.; assessment of clinical probability using prediction rules such as the Wells score and the revised Geneva score; elevated D-dimer (a degradation product of cross-linked fibrin) level in plasma; evidence from compression ultrasonography and computed tomographic venography; evidence from ventilation–perfusion scintigraphy; evidence from computed tomography; pulmonary angiography, etc. Recently, research in the epidemiology, predisposing factors, natural history and pathophysiology of PE have been advancing greatly. These studies have helped to improve the diagnosis, treatment, and prognosis significantly. The acute case fatality rate for PE now ranges from 7 to 11% according to a prospective cohort of studies and is continuing to decrease. Nevertheless, many issues regarding the natural history and pathophysiology of PE require further study. Any advances made regarding the above issues will facilitate selection of the best management strategies for a typical patient suffering from a given condition, taking into account the impact on outcome, as well as the risk/benefit ratio of particular diagnostic or therapeutic means. PE can be difficult to diagnose as the clinical signs and symptoms are non-specific. Thus, further research on the relative symptoms, signs, disease history and pathophysiological characteristics of PE is still important. Questions that still need addressing include: Do any differences exist in the distribution of lesions within the lungs of PE patients? How do the demographic characteristics, symptoms, signs, disease history and phenotypes of pathophysiology differ between male and female patients? In this report, 149 PE patients were recruited at our Emergency Department from January 2010 to December 2014 to evaluate the above questions. # Materials and Methods ## Patients This study was conducted in accordance with the World Medical Association Declaration of Helsinki. The Review Board of the Ethics Committee of Medical Research at the Center Hospital of Minhang District (170 Xinsong Road, Minhang District, Shanghai 201199, China) approved the study protocols (reference number: SHMHCH 2010–002). Written informed consent was obtained from all patients according to the guidelines of the Chinese National Ethics Regulation Committee; the procedure was explained to all patients and we emphasized that their data would be used in this study. All patients were informed of their rights to withdraw consent personally or via kin, caretakers, or guardians. We sequentially recruited 149 PE patients treated in our hospital Emergency Department from January 2010 to December 2014. The diagnoses complied with the “2014 ESC Guidelines on the diagnosis and management of acute pulmonary embolism”. Patients presenting with PE symptoms and signs were screened by clinical examination and assessment of clinical probability, in combination with other tests at the emergency hospital phase. For suspected PE with shock or hypotension, computed tomographic pulmonary angiography (CTPA) was first adopted to confirm PE. For CTPA-negative patients, echocardiography was then performed to provide evidence of any acute pulmonary hypertension or right ventricle dysfunction. For those suspected PE patients presenting without shock or hypotension, the clinical probability of PE was assessed using clinical judgment or a prediction rule. Patients with high clinical probability of PE had their diagnosis confirmed by CTPA. Plasma D-dimer measurements were performed on patients with a low or intermediate clinical probability of PE. CTPA was used to confirm the diagnosis of PE in patients with elevated D-dimer levels (\>250 ng/mL D-Dimer Units). For patients in whom both CTPA and echocardiography were negative, the differential diagnosis would then include consideration of acute valvular dysfunction, tamponade, acute coronary syndrome (ACS) and aortic dissection. Ancillary bedside imaging tests including transesophageal echocardiography and bedside compression venous ultrasonography as well as catheterization would be performed. To survey the mortality of the PE patients treated in our hospital, the death cases during the three-month follow-up period were recorded and the death rate was calculated. ## Data collection The questionnaire explored demographics (age and gender); symptoms (dyspnea, cough, palpitation, chest pain, fever, syncope and hemoptysis); and personal history (PE, tumor, heart disease, chronic pulmonary disease, alcohol consumption, cigarette smoking, obesity, hypertension, diabetes and cerebrovascular cardiovascular disease). Any data collected via the questionnaire was confirmed by in-hospital measurement when the relevant assessment method became available. Physical signs in each patient, including engorgement of the neck veins, edema of the lower extremities, respiratory rate, systolic pressure, diastolic pressure, heart rate and cardiac sounds, were examined and collated at the Emergency Department registry. To identify any complications in the cardiovascular system, a 12-lead surface electrocardiogram was performed in the Emergency Department. As well as the structure and dysfunction of the heart, the following were recorded: P-pulmonale, right bundle branch block (RBBB), V1–V4 T-wave inversions, a large S wave in lead I, a large Q wave in lead III, an inverted T-wave in lead III (S1Q3T3), atrial fibrillation, pulmonary arterial hypertension, tricuspid insufficiency, and left ventricular ejection fraction (LVEF). The chief technician reviewed the final electrocardiographic and echocardiographic results for each patient. Any uncertainties regarding the electrocardiographic and echocardiographic results were resolved by discussion between the technicians. Both technicians who reviewed these data were blinded to the overall clinical data and group division. In order to screen for pneumonia, pleural effusion, and heart shadow changes, an X-ray examination was performed immediately after the patient attended the Emergency Department. For all suspected PE patients, CTPA was performed to reveal the position and extent of any lung injury caused by the embolism. The position of the embolism was then categorized as follows: right pulmonary artery; left pulmonary artery; upper lobe, middle lobe and lower lobe of the right lung; and upper lobe and lower lobe of the left lung. Blood gas analysis revealed the following biochemical results: D-dimer, creatinine, lactate dehydrogenase, creatine kinase, brain natriuretic peptide, etc. These were closely monitored during and after the emergency rescue process. ## Statistical analysis Continuous variables were presented as means ± standard deviation (SD) and categorical data were presented as numbers (percentage). Differences between female and male groups were examined by using t-tests or χ² tests according to the characteristics of the data distribution. The significance level (α) was set at 0.05. All statistical analyses were performed using Stata/SE 12.0 for Windows (StataCorp LP). # Results ## Clinical characteristics of PE patients In total 149 patients with PE were recruited for this report, of whom, 73 were males and 76 were females. The male: female constituent ratio was close to 1.0 and is coincident with other reports. The average age of the 149 PE patients was 73.5±13.4 years, with most patients tending to be older. The overall fatality rate was 7.4%. Severe pneumonia, cardiogenic shock, hemodynamic compromise and/or respiratory failure were found to be the main causes of death. ## Comparison between male and female PE patients To identify differences between the male and female PE patients regarding age, symptoms, signs, disease history, pathophysiology and outcome, the patients were divided into male and female groups. As shown in, there was no significant difference between male and female PE patients in the following variables. Symptom: presence of cough, palpitation, chest pain, or fever; Sign: syncope, hemoptysis, engorgement of the neck veins and edema of lower extremities, respiratory rate, heart rate; History: history of heart disease, PE or DVT history, history of immobilization and/or surgery; blood gas analysis: PO2, PaCO2, pH; electrocardiogram: rates of P-pulmonale, RBBB, V1–V4 T-wave inversion and atrial fibrillation; imaging tests: presence of pneumonia, pleural effusion and increased heart shadow; biochemical tests: levels of blood creatinine, lactate dehydrogenase, creatine kinase and brain natriuretic peptide; outcome: rates of recovery and death. It is noteworthy that many of these variables, such as the presence of cough, chest pain, and a history of immobilization and/or surgery, displayed different tendencies between male and female patients that were not considered to be significant because of the relatively small sample size. Interestingly, whilst the number of patients with dyspnea, tumor and chronic pulmonary disease; systolic and diastolic pressures, was significantly higher in males than in females, the number of patients with V1–V4 T-wave inversion and elevated blood D-dimer levels was significantly lower in males than in females. ## PE lesions are more common in the right lung To evaluate differences in the locations of PE, the data from all patients with confirmed PE using CTPA were analyzed further. As shown in , the numbers of PE located in the right pulmonary artery, right upper lobe, right middle lobe, right lower lobe, left pulmonary artery, left upper lobe and left lower lobe were 27.2%, 48.5%, 25.0%, 51.5%, 18.4%, 25.0% and 25.0%, respectively. The total proportion of right lung emboli (i.e., any confirmed embolism regardless of whether it was sited in a blood vessel or lung lobe) was 89.7%, which was significantly higher than that of the left lung (42.6%). ## Comparison of the PE location in males and females The PE location in male patients with confirmed PE was compared with that of female patients. As shown in, the number of PEs located in the right pulmonary artery, left pulmonary artery, right lung, left lung, both lungs and right lung only was 32.9%, 21.9%, 94.5%, 38.4%, 34.2% and 60.3% in males, respectively; and 17.1%, 11.8%, 69.7%, 39.5%, 28.9% and 40.8%, respectively, in females. The numbers of PE lesions in the right pulmonary artery and right lung were significantly higher in male than in female patients. # Discussion Differences relating to sex and gender are observed in the pathophysiological processes of many diseases. In this study, we found differences according to sex in the symptoms, signs, disease history, hemodynamic heart consequences, biochemical indices, and the location of lung lesions in PE patients. Firstly, the frequency of right lung PE lesions was significantly higher than in the left lung generally. Secondly, our data suggested that the probability of PE lesions in the right pulmonary artery and/or right lung was significantly higher in male patients than in female patients. Thirdly, the occurrence of dyspnea, number of patients with tumor, number of patients with chronic pulmonary disease, and the average systolic and diastolic pressure were significantly higher in males than in females. Finally, in contrast with this, the rates of V1–V4 T-wave inversion and elevated D-dimer blood levels were significantly lower in males than in females. Differences according to sex and gender in the pathophysiological procedure of PE have also been reported. Males were found to have lower survival rates than females. Normotensive female PE patients were on average older and had a submassive PE stadium more frequently. Changes in the right ventricle function over time during the course of a 3-month follow-up might differ between male and female patients with acute pulmonary thromboemboli, and the recovery process could be slower in females. As far as we are aware, few previous reports have concerned the above four issues, and we have therefore addressed them in our study. Females hospitalized with PE were found to have significantly lower odds of 30-day mortality compared with males. Of the PE patient population in our study, the death rate of male and female patients was 8.2% and 6.6% respectively; however, this difference was not significant. The reason that PE lesions are more prevalent in the right lung is unknown. Anatomic characteristics and lung circulation might be fundamental factors for this phenomenon. Asthma, chronic obstructive pulmonary disease, chronic bronchitis, emphysema, pneumonia, lung cancer and acute respiratory distress syndrome are the most common diseases that target the lung, but it is difficult to find evidence that these diseases are also more prevalent in the right lung. Thus while our findings are useful in the management of PE, the importance of checking for this phenomenon in other lung diseases should also be emphasized. Right ventricular dilatation and dysfunction are common complications of PE because of increased right ventricular after-load. These complications may lead to right ventricular failure and subsequently death. Echocardiograms are essential in ruling out complications such as these, and also intracardiac thrombi. In this study, although an echocardiogram was not performed on all patients at the Emergency Department, the electrocardiogram and imaging tests did show pathologic changes in the heart. These pathologic changes included increased heart shadow, P-pulmonale, RBBB, V1–V4 T-wave inversion and atrial fibrillation. As these tests are not specific to PE-related heart dysfunction, and we had no background information regarding the heart function of these patients, whether the above observations are specific consequences of PE is unknown. The reasons for patient death in this study were advanced age and missed hospital treatments. PE is difficult to diagnose and may therefore be missed because of non-specific clinical presentation. However, early diagnosis is fundamental, since immediate treatment is highly effective. Depending on the clinical presentation, initial therapy is aimed primarily at either life-saving restoration of blood flow through occluded pulmonary arteries or the prevention of potentially fatal early recurrences. Thus, the results presented in this report will complement current strategies in PE control and treatment. All patients in this report had CTPA data; embolism lesions were found in as many as 90% of PE patients who underwent CTPA at our hospital, which is considerably higher than noted in other reports. In this study, less than 10% of patients displayed shock or hypotension at the time of attending the hospital, which might be a result of local public health policies. # Conclusions Gender differences regarding the symptoms, signs, disease history, lesion position and pathophysiology exist in patients with PE and should be considered in clinical practice. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: XD. Performed the experiments: YL LZ CL ML ND JS. Analyzed the data: XD. Contributed reagents/materials/analysis tools: YL LZ. Wrote the paper: XD.

# Introduction Successful methods for the treatment of tracheal stenosis are yet to be developed. Tracheal stenosis occurs in response to an injury of the airway mucosa, caused by prolonged endotracheal intubation, long-term tracheostomy, or an airway burn. A prolonged tracheostomy and repeated surgeries are needed in most cases, but they will significantly impair the patient’s ability to speak and swallow. Although various surgical techniques have been developed, such as balloon dilation, endoscopic laser resection, and end-to-end anastomosis, the patient’s outcome can frequently be complicated by recurrent stenosis, due to the formation of new scars and to granulation following surgery, as the surgery itself causes an injury to the airway mucosa. Novel approaches against post- surgical restenosis have long been needed. The modulation of wound healing in the airway mucosa represents a potential treatment option. The sendai virus (SeV) has a strong affinity to the airway epithelium. The wild-type form of SeV causes respiratory tract infections in rodents, but is not pathogenic in humans. Its therapeutic use is predicted to be safer than that of DNA viral vectors, because the RNA genome of SeV does not go through a DNA phase. Therefore, the SeV vector has been suggested as a potential gene-transfer vector for the treatment of airway diseases. A fusion gene-deleted, non- transmissible Sendai virus vector (SeV/ΔF) was prepared in a preclinical study. Its preventive effect in a model of tracheal obliteration following transplantation was reported without any systemic side effects. We recently presented a novel model of acute mucosal injury, in which the tracheal mucosa of rats was scraped with a nylon brush through the tracheostoma, and we reported the successful gene transfer into the injured mucosa of those rats, through SeV/ΔF delivery. In this model, the wound-healing response after the mechanical injury to the airway mucosa is seen as inflammation or fibrous proliferation in the regenerating tracheal mucosa. *c-myc* is a regulator gene responsible for the upregulation of many genes that promote cell proliferation. The elevated expression of c-Myc has been detected not only in a broad range of human cancers, but also in chronic wounds. The main cause of tracheal stenosis is the excessive cell proliferation that occurs within the limited tracheal space during wound healing. Therefore, the suppression of c-Myc expression represents a potential strategy for the treatment of tracheal stenosis. The far upstream element (FUSE)-binding protein (FBP)-interacting repressor (FIR) is a *c-myc* transcriptional repressor.\[–\] FIR strongly suppresses *c-myc* transcription through its inhibition of the TFIIH/P89/XPB helicase (P89) activity. FIR also induces apoptosis by suppressing *c-myc*. The SeV-encoding FIR (FIR-SeV/ΔF) was identified as a candidate for cancer gene therapy. We reasoned that FIR-SeV/ΔF might also suppress *c-myc* in the tracheal mucosa, and thus show therapeutic potential in the treatment of tracheal stenosis. Therefore, we hypothesized that gene therapy using FIR-SeV/ΔF could prevent tracheal stenosis in an animal model of induced mucosal injury, via the inhibition of c-Myc caused by the expression of transduced FIR. The aim of this study was to provide evidence to support this hypothesis, using the above-mentioned rat model. # Materials and Methods ## Animals All protocols for the handling and the experimental use of animals were approved by the Committee on the Ethics of Animal Experiments and by the Safety Board on Recombinant DNA experiments of the National Defense Medical College (Permit Number: 14069 and 2012–24). All surgery was performed under anesthesia, and all efforts were made to minimize suffering. Twenty-eight adult, female Sprague–Dawley rats (200–220 g) were used. ## Preparation of non-transmissible recombinant SeV vectors A fusion gene-deleted, non-transmissible SeV vector encoding FIR (FIR-SeV/ΔF; 8.2 × 10⁹ cell infectious units \[CIU\]/mL) was prepared as previously described. FIR-SeV/ΔF was manufactured and provided by the DNAVEC Corporation (Tsukuba, Japan), and stored at -80°C until used. ## Surgical procedure The animals were anesthetized with medetomidine (0.8 mg/kg; intraperitoneal (IP) injection) and with ketamine HCl (40 mg/kg; IP injection) during the surgery. The tracheostomy was performed at the tracheal rings 5 and 6. The laryngotracheal mucosa above the tracheostoma was scraped 10 times with a 1.8-mm nylon brush (Kobayashi Pharmaceutical, Osaka, Japan), through the tracheostoma as previously described. To assess the preventive effect of FIR-SeV/ΔF, 19 rats were assigned to an untreated control group (*n* = 10) or to a FIR-SeV/ΔF- treated group (*n* = 9), respectively, at random. To confirm the transgene expression of FIR in the tracheal tissue using RT-PCR, nine rats were assigned to a normal group (*n* = 3), to an untreated control group (*n* = 3) or to a FIR-SeV/ΔF-treated group (*n* = 3), respectively, at random. Ten microliters of FIR-SeV/ΔF (8.2 × 10⁹ CIU/mL) or of normal saline solution was administered through the tracheostoma to the injured airway mucosa, for three consecutive days after the tracheal injury. Animals were observed carefully following the surgery, to prevent an acute airway obstruction. The crusts that were attached to the tracheostoma were removed every 24 h until the animals were euthanized. Animals were euthanized with an intraperitoneal injection of a massive dose (150 mg/kg) of pentobarbital, and the tracheal tissue at the tracheal rings 1 and 2 was excised five days following the tracheal scraping. If the animals exhibited severe stridor, difficulties with ambulation, or a body weight loss that was greater than 20% due to tracheal stenosis, they were euthanized for humane reasons. The survival curves were drawn, and the survival rates were compared between the groups. ## Histological and quantitative analysis of lumen stenosis To assess the pathological changes occurring at the airway mucosa, and to assess the degree of stenosis in the tracheal lumen, the excised tracheal tissues were fixed in 10% formalin, embedded in paraffin, and then cut into 4-μm-thick axial sections. The images of the H&E-stained axial sections were captured at a low- power magnification. The image J software, version 1.44p (National Institutes of Health, Bethesda, MD), was used to measure the area of the lumen-tracing mucosal surface as the narrowed lumen, and that of the cartilage surface as the initial lumen. The percentage of stenosis was calculated using the following formula, as was previously described: (1 − area of the mucosal surface lumen/area of the tracheal cartilage lumen) × 100. ## Quantitative RT-PCR analysis of FIR RNA expression To confirm the transgene expression of *FIR* in the tracheal tissue using quantitative real time RT-PCR, the tracheal tissues excised from the normal, the untreated control, and the FIR-SeV/ΔF-treated animals (*n* = 3 for each group) were prepared, as mentioned above. The tissue samples were fragmented into small pieces (\<25 mg), and immersed into an RNAlater solution (Applied Biosystems, Tokyo, Japan). The RNA was isolated using an RNeasy Mini Kit (Qiagen, Valencia, CA). The primer sequences that were used are as follows: human FIR forward (exon 2) 5′-CCATAGCTCTCCAGGTCA-3′ and reverse (exon 6) 5′-CGTAGACGCGGCACATGA-3′; rat β-actin forward 5′-TGGAGAAGAGCTATGAGCTGCCTG-3′ and reverse 5′-GTGCCACCAGACAGCACTGTGTTG-3′. The mRNA levels of the relevant molecules were measured using quantitative real-time RT-PCR, with the One Step SYBR PrimeScript RT-PCR Kit (Takara Bio, Shiga, Japan) in the Thermal Cycler Dice Real Time System II (Takara Bio, Shiga, Japan). All reactions were performed in triplicates using a PCR assay, and under the same cycling conditions: The reverse transcription reaction was maintained at 42°C for 5 min, and was ended with an incubation step at 95°C for 10 s. The PCR step was carried out for 40 cycles at 95°C for 5 s; 60°C for 30 s; the final dissociation step was carried out at 95°C for 15 s, 60°C for 30 s, and 95°C for 15 s. The levels of accumulated fluorescence were analyzed using the second derivative maximum method, and the ΔΔCt method with the LightCycler data analysis software, TP 900 version 4.02 (Takara Bio, Shiga, Japan), following the melting-curve analysis, and then the expression levels of the target genes were normalized to the expression level of β-actin in each sample. The results are shown as the mean ± standard error of the mean (SEM) of the three independent measurements. ## Immunohistochemistry for confirmation of the suppression *c-Myc* expression Immunohistochemistry was used to confirm that the SeV-encoded FIR suppressed c-Myc expression. Following the deparaffinization and the hydration of the tracheal specimens from the untreated control and the FIR-treated animals, the slides were covered with 10 mM of sodium citrate buffer, pH 6.0, and were heated in an autoclave for antigen unmasking at 120°C for 5 min. The endogenous peroxidase activity was blocked using 3% H₂O₂ in methanol for 5 min. The streptavidin-biotin complex method (VECTASTAIN Elite ABC mouse IgG Kit PK-6102; Vector Laboratories, Inc., Burlingame, CA) was used according to the manufacturer’s instructions. The slides were incubated for 1 h with an anti c-Myc primary antibody (9E10, sc-40; Santa Cruz Biotechnology, Inc., Dallas, Texas) at room temperature in a moisture chamber. The slides were then incubated with a secondary biotinylated antibody solution and with the VECTASTAIN ABC Reagent at room temperature, each for 30 min, and then visualized with 3,3′-diaminobenzidine (DAB) and counterstained with hematoxylin. The c-Myc expression was observed under high-magnification microscopy, and the images were captured using a CCD camera (DP26; Olympus, Tokyo, Japan). To quantify the c-Myc-positive area in the nucleus, the regions of interest (ROIs) in a given display image at a 400-fold magnification were carefully defined over the mucosa of the trachea. Three ROIs were randomly cropped in each captured image. To quantify the c-Myc-positive area in the nucleus, the percentage of DAB-stained area in the ROIs was calculated using the following formula: (DAB-positive area/hematoxylin-positive area) × 100, using the Image J software, version 1.44p (National Institutes of Health). ## Statistical analysis The results are shown as the mean ± standard error of the mean (SEM). To evaluate the significance of the differences in the percentages of stenosis, in the average mRNA expression levels, and in the DAB-positive areas between the two groups, Mann-Whitney U-tests were performed. The analysis of survival was performed using the log-rank test with the JMP software, version 10.0.0 (SAS Institute Inc., USA). The significance was determined using a *P*-value \< 0.05. # Results ## FIR-SeV/ΔF prevents tracheal stenosis To assess the preventive effect of FIR-SeV/ΔF in a rat model of induced mucosal injury, we evaluated the pathological changes in the tracheal lumen and the degree of lumen stenosis. Rats that had their airway mucosa scraped were administered a normal saline solution (untreated controls) or FIR-SeV/ΔF through the tracheostoma. The pathological changes in the tracheal lumen were assessed five days after following. Representative hematoxylin–eosin (H&E)-stained axial sections of the trachea of untreated controls revealed hyperplasia of the airway epithelium, and a thickened submucosal layer with extensive fibrosis, angiogenesis, and collagen deposition causing lumen stenosis. By contrast, the FIR-SeV/ΔF-treated animals showed a decrease in the wound-healing response in the airway mucosa, when compared to the untreated controls. Hence **F**IR-SeV/ΔF decreased the degree of tracheal stenosis (28.9 ± 7.30% vs. 43.4 ± 12.4%, respectively; *P* \< 0.05), and thus prevented tracheal stenosis following a mucosal injury. ## FIR-SeV/ΔF improves the survival rate Animals were euthanized five days following tracheal scraping, or when the animals displayed severe health problems due to tracheal stenosis. In the untreated group, four animals were sacrificed for humane reasons before the end of the observation period. In the FIR-SeV/ΔF-treated group, no animal was sacrificed. The FIR-SeV/ΔF-treated animals had a significantly increased survival rate compared to the untreated controls (*P* \< 0.05). The Kaplan–Meier survival curves during the five-day follow-up period are shown in. It suggests that FIR-SeV/ΔF improves the survival rate by preventing tracheal stenosis. ## Transgene expression of FIR in tracheal tissues To evaluate the transgene expression of FIR, the mRNA expression of FIR in excised tracheal tissues was determined using quantitative real-time reverse transcription-polymerase chain reaction (qRT-PCR). The relative mRNA expression of FIR was significantly increased in the FIR-SeV/ΔF-treated animals, when compared to the normal or the untreated control animals (1.96 × 10^-2 ± 1.61 × 10^-2 vs. 1.01 × 10^-4 ± 0.64 × 10^-4 vs. 1.98 × 10^-4 ± 4.86 × 10^-4, respectively; *P* \< 0.001). The expression of FIR following the delivery of the FIR-SeV/ΔF transgene was confirmed in the tracheal tissues. ## c-Myc is suppressed by FIR-SeV/ΔF To evaluate how c-Myc was regulated in the mechanically injured airway mucosa by FIR-SeV/ΔF, immunohistochemical staining using an anti-c-Myc antibody was performed. Representative sections of the paraffin-embedded rat trachea are shown in. c-Myc was mainly overexpressed in the nucleus of the tracheal basal cells of the untreated animals, but was less expressed in the same cells of the FIR-SeV/ΔF-treated animals. The area of c-Myc-positive tracheal epithelial cells in the FIR-SeV/ΔF-treated group was significantly smaller than that of the untreated control group (7.09 ± 1.88% vs. 27.7 ± 2.02%, *P* \< 0.001). c-Myc was probably suppressed by FIR-SeV/ΔF in the regeneration of the airway epithelium of the injured tracheal mucosa. # Discussion In the present study, the FIR-SeV/ΔF-treated animals showed a significant decrease of tracheal stenosis and a significant improvement in the survival rate. An elevated expression level of the transgene FIR mRNA was observed in the FIR-SeV/ΔF-treated tracheal tissues. c-Myc was overexpressed in the tracheal basal cells from the area of untreated mucosal injury in the control animals, but was down-regulated in the same cells from the FIR-SeV/ΔF-treated animals. Our data clearly suggests that FIR-SeV/ΔF has therapeutic potential in the treatment of tracheal stenosis through its inhibitory effect on the proliferation of cells in the regenerating airway epithelium. Owing to the risks of pathological immune responses or of oncogenesis with the current viral vectors, the use of gene therapy is limited to clinical trials for the treatment of cancer or for that of hereditary diseases linked to genetic defects. No experimental study on gene therapy for the treatment of tracheal stenosis in a model of acute mucosal injury has been documented. To make gene therapy for benign airway diseases practical, it will be necessary to develop novel vector systems that efficiently deliver genes to the airway epithelium, but that pose a lower risk of dangerous immune responses or of mutagenesis than the current viral vectors do. SeV has unique features as a vector, such as a strong affinity for the airway mucosa, a high transduction capacity, an optimal persistence, and sufficient evidence of its therapeutic safety.\[–\] In addition, our previous study revealed the successful transgene expression without any unwanted local immune response, and with an efficient duration when administering SeV/ΔF by spraying to the injured laryngotracheal mucosa. The administration method of spraying is simple and easy; therefore, it shows potential for clinical use in the treatment of human laryngotracheal diseases. The safety advantage of SeV compared with other viral vectors is that the SeV genome is located exclusively in the cytoplasm of the infected cells, and does not go through a DNA phase; i.e., there is no danger of the undesired integration of foreign sequences into the host chromosomal DNA. Extensive preclinical studies have shown the therapeutic potential of this vector for use in the treatment of airway disease, ischemic brain disease, and malignant tumors. In addition, the potential of the SeV vector for vaccination via the airway mucosa by nasal drip administration has been reported. SeV is currently being used in Japan in clinical trials for the treatment of ischemia of the lower limbs, without any reported serious adverse events over a 6-month follow-up. Our SeV vector system has the potential to act as an efficient and safe gene delivery system for the treatment of airway diseases. To our knowledge, this represents the first reported demonstration of successful gene therapy using an SeV vector in a model of acute mucosal injury. The far upstream element (FUSE)-binding protein (FBP) increases the transcription of *c-myc* by increasing the ability of the transcription factor II human (TFIIH) protein to release the paused RNA polymerase enzyme. This is counteracted by the activity of the FBP-interacting repressor (FIR), which binds to both the FBP and the TFIIH, and down-regulates the helicase activity of TFIIH. Recent reports indicate the important role of c-Myc not only on tumor growth but also on the wound-healing response. Several preclinical studies have reported the effects of inhibiting c-Myc on the treatment of lung fibrosis, on the stenosis of coronary arteries, and on malignant tumors. c-Myc can be considered an important molecular target to prevent tracheal stenosis due to the unnecessary cell proliferation in the injured airway mucosa. Therefore, the *c-myc* suppressor FIR can be considered a potential candidate for tracheal stenosis therapy. It is necessary to focus on the time-course of the wound-healing process that occurs in the airway mucosa, when aiming to prevent LTS after a mucosal injury. If the wound-healing response is completed without any unwanted cell proliferation, then the lumen may remain free of stenosis. Immediately following an injury, platelets or thrombocytes aggregate at the injury site to form a fibrin clot during the coagulation phase (i.e., 0–2 days following an injury). Next, the immune cells are activated during the inflammation phase (i.e., 2–3 days following an injury). In the early proliferative phase (i.e., 3–5 days following an injury), the epithelialization usually precedes the activation of cell proliferation in the submucosal layer. The late proliferative phase (i.e., 5–14 days following an injury) is characterized by angiogenesis and fibroblast growth. The remodeling phase (i.e, several weeks following an injury) continues for a prolonged period with the deposition of collagen produced by fibroblasts. Once wound-healing response has advanced to the remodeling phase, surgery is the only treatment option available. However, recurrent stenosis following the surgery has been a long-standing problem. Novel approaches to inhibit cell proliferation in the early phase of wound healing have been needed. In this study, FIR-SeV/ΔF effectively prevented LTS through its suppression of *c-myc* transcription in the proliferative phase, because the airway affinity-based transgene expression of the SeV vector achieved a peak in several days, and then persisted for around two weeks, as we have previously reported. The tracheal epithelium comprises goblet cells, ciliated cells, and basal cells. The tracheal basal cells represent a multipotent progenitor cell type that proliferates actively to ensure the renewal of the injured tracheal epithelium. In the present study, c-Myc was overexpressed in the tracheal basal cells from the site of untreated mucosal injury in control animals, but was downregulated in the same cells from the FIR-SeV/ΔF-treated animals. The initial thickening of the airway mucosal cells is the process that is mainly responsible for recurrent airway stenosis. The modulation of the wound-healing response following an injury to the airway mucosa is a key component of the treatment of tracheal stenosis. Thus, gene therapy using the *c-myc*-targeting FIR-SeV/ΔF could be an attractive treatment option in the treatment of tracheal stenosis in the future. The possible concerns of this treatment are the long-term preventive effect and an unwanted delay in the wound-healing process. Once the wound-healing response is completed, it is presumed that the stenosis-preventive effect will be maintained. However, as the observation period in this research project is short, a long-term effect of the transient overexpression of FIR through the use of the recombinant SeV vector should be examined in future experiments. Another concern is that, if the *c-myc* suppression continues for too long, it may impair wound healing undesirably when FIR-SeV/ΔF is used in combination with surgical treatment. The transient expression through SeV/ΔF is considered to be suitable for the limited suppression of the acute phase of the wound-healing response. Although this research revealed the mechanism of the c-Myc-dependent cell proliferation in the regenerating epithelium, a histological examination of the FIR-treated animals showed that the thickening of the submucosal tissue was also inhibited. An alternative function of FIR, i.e. other than the control of c-Myc, has been suggested and has also been shown in recent research. FIR inhibits the cell-cycle via its disruption of P27Kip1 (P27) expression. P27 arrests cells at the G1 phase, and SAP155, a subunit of the essential splicing factor 3b subcomplex in the spliceosome, is required for proper pre-mRNA splicing by P27. Notably, FIR forms a complex with SAP155, and thus alters the expression of c-Myc and of P27, through the disruption of the well-established functions of FIR and SAP155A. Accordingly, it has been reported that c-Myc is regulated at the transcriptional and post-transcriptional level by the FIR-SAP155 complex. FIR has the potential to inhibit cell proliferation not only by suppressing *c-myc,* but also by blocking the cell cycle via P27. A further investigation of the alternative function of FIR is needed in future experiments. Considering its enhanced safety, its efficient transgene expression, and its low contact time requirement compared to other vectors that are currently used, our SeV vector can be administered safely and easily to the airway mucosa by inhalation. Therefore, our FIR-encoding SeV vector shows great potential for therapeutic use in a clinical setting, e.g., for prophylactic inhalation for the treatment of the acute phase of an airway burn, for prolonged intubation, and for surgical treatments. In summary, we have demonstrated that the airway-targeted expression of FIR could efficiently prevent tracheal stenosis, through its suppression of *c-myc* transcription. We believe that the *c-myc*-targeting, SeV-encoding FIR can be developed as a safe and effective therapeutic agent for the treatment of airway disease. We would like to thank Dr. Daisuke Kamide and Dr. Yoshihiro Miyagawa for their technical assistance with the tissue processing and with the quantitative polymerase chain reaction. [^1]: YU is an employee of DNAVEC Corporation. The other authors declare no competing interest. This does not alter the authors’ adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: DM KA AS. Performed the experiments: DM KA NT HS. Analyzed the data: DM KA. Contributed reagents/materials/analysis tools: YU. Wrote the paper: DM KA. Edited the manuscript: MT TY AS HS KM.