Using synthetic semiochemicals to train canines to detect bark beetle–infested trees

The dog detection allows timely removal by sanitation logging of first beetle-attacked trees before offspring emergence, preventing local beetle increases. Detection dogs rapidly learned responding to synthetic bark beetle pheromone components, with known chemical titres, allowing search training during winter in laboratory and field. Dogs trained on synthetics detected naturally attacked trees in summer at a distance of > 100 m. An early detection of first beetle-attacked trees would allow timely sanitation felling before offspring emergence, curbing local beetle increase. We tested if detection dogs, trained off-season on synthetic pheromone components from Ips typographus, could locate naturally bark beetle–infested spruce trees. Indoor training allowed dogs to discriminate between the infestation odours (target) and natural odours (non-target) from the forest. Odour stimuli were shown by chemical analysis to be bioactive at extremely low-levels released (< 10−4 ng/15 min) in the laboratory. Detection dogs, trained to recognise four different synthetic pheromone compounds in the wintertime, were able to detect naturally infested spruce trees unknown to humans the following summer. The dog-handler pairs were able to detect an infested spruce tree from the first hours of beetle attack until several weeks after first attack, long before discolouration of the crown. Trained sniffer dogs detected infested spruce trees out to ≥ 100 m, as measured by GPS-collar tracks. Dog-handler pairs appear to be more efficient than humans alone in timely detecting bark beetle infestations due to the canine’s ability to cover a greater area and detect by olfaction infestations from a far longer distance than can humans.


54
Detection dogs are used to locate many objects including humans, explosives, and illicit drugs 55 (see BROWNE et al. 2006 and references therein;LORENZO et al. 2003). Trained canines have 56 also been used to detect invasive organisms (GOODWIN et al. 2010;HOYER-TOMICZEK et al. 57 2016) as well as endangered species (reviewed by BEEBE et al. 2016). Canines have also been 58 trained to detect small or cryptic insects such as termites (BROOKS et al. 2003), palm weevils 59 (NAKASH et al. 2000), and bed bugs (PFIESTER et al. 2008; VAIDYANATHAN AND FELDLAUFER 60 2013) and endangered Coleoptera (MOSCONI et al. 2017). The key benefits of using trained 61 detection dogs are their keen sense of smell (HEPPER AND WELLS 2015), and their ability to 62 cover large areas in a shorter time, when compared to humans (MOSCONI et al. 2017). In most 63 cases, biological material is used for the training (JOHNEN et al. 2013). 64 The European spruce bark beetle − Ips typographus (L.) is one of the most destructive forest 65 pests in Europe (GRÉGOIRE AND EVANS 2004). For forest protection, the rapid detection of 66 bark beetle infestations is required to successfully implement a management strategy that 67 relies upon removing recently infested trees within 2 −3 weeks of attack (SVENSSON 2007). 68 However, human detection generally requires close inspection (≤ 1m) of trees, and is 69 therefore time-consuming, costly, and not always practical. Therefore, detection generally 70 occurs 2−3 months after an infestation in N Europe, when tree crown colour fades and bark 71 falls off. By this time, most bark beetles have left the infested tree and may attack other, non-72 infested stands. Since a rapidly changing, but specific series of beetle pheromone components 73 and other semiochemicals are present for several weeks after an initial attack, the use of 74 detection dogs may prove a better alternative than human inspection. Upon attacking a tree, 75 male bark beetles secret an aggregation pheromone, consisting of a blend of 2-methyl-3-76 buten-2-ol and cis-verbenol (BIRGERSSON et al. 1984). A few days later, an inhibitory signal 77 (consisting mainly of ipsdienol) is emitted when bark beetle females have begun laying eggs 78 not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 12, 2018. ;https://doi.org/10.1101/262386 doi: bioRxiv preprint 2018-03-11 5 (25) (BIRGERSSON et al. 1984SCHLYTER et al. 1987). After the first week, an additional chemical 79 cue, indicating that the infested tree is fully utilized and competition is high, is evident. This 80 semiochemical, verbenone, is an oxygenation product by the beetle and by the interaction of 81 fungi and bacteria with damaged tree phloem (LEUFVÉN AND BIRGERSSON 1987;SCHLYTER et 82 al. 1989). 83 In this proof of concept study, we report the laboratory training of two detection dogs on a 84 series of synthetic semiochemicals associated with bark beetle infestations, and the ability of 85 these trained dogs to later detect and locate bark beetle infested trees in the field. Since the 86 semiochemical profile of attacked trees changes rapidly in both the quality and quantity of 87 semiochemicals released, we chose to use several synthetic chemical compounds as 88 representative stimuli in our canine training. We were also interested in determining if dogs 89 trained on synthetic pheromones in the winter months could later locate infested trees in the 90 summer months. Finally, we wanted to determine if a trained dog can detect natural 91 infestations from distances (10 to 100 times) further away than a human. 92

93
Canines 94 Two dogs, owned by SnifferDogs Sweden (Hjortsberga, Sweden), were used in this study. Dog 95 A was a nine-year-old female German shepherd that was previously trained as a search and 96 rescue dog for humans. Dog B was a one-year-old female Belgian shepherd (Malinois) that had 97 only basic obedience training, and had no previous formal detection expertise. 98
The copyright holder for this preprint (which was this version posted March 12, 2018. ; https://doi.org/10.1101/262386 doi: bioRxiv preprint 2018-03-11 6 (25) bark beetle pheromone and a product of the host tree was obtained from Fluka (Sigma-103 Aldrich, Stockholm, Sweden). Other chemicals used in the study were obtained from our 104 chemical stocks (see ANDERSSON et al. 2012). 105 Each pheromone component was stored separately in separate jars of glass, to avoid cross 106 contamination of odours. In each jar of glass a cotton pad (ICA Basic Bomullsrondeller, 107 Netherlands) was placed in the bottom and a small amount of each semiochemical were 108 dropped on to the cotton pad (10 µl methylbutenol, ≈10 mg cis-verbenol, 1µl ipsdienol, or 10 109 µl verbenone). The glass jars were then filled with cotton pads, and so molecules in gas phase training, we replaced the cotton pads each day. For release rates by GC-MS and dog training 119 response study we prepared 5 steel tins of each synthetic pheromone at the same time. These 120 tins were stored in room temperature (circa + 20 °C). Release rates were determined using 121 odour collections similar to Zhang et al. (2000). An inverted glass funnel (5 cm dia.) was 122 placed above the steel tin and air was drawn through a column packed with Porapak ® Q 123 (25 mg mesh 60-80; in a Teflon tube 3 mm i.d.) at 100ml/min at 15 min intervals. Compounds 124 were eluted from the column with 400 µl pentane (Sigma-Aldrich, Steinheim, Germany) into 125 a 400 µl insert placed in a 2 ml screw top vial (Agilent Technologies, Böblingen, Germany), 126 and 1 mg heptyl acetate was added as internal standard. Aeration extracts were analysed by 127 gas chromatography-mass spectrometry (GC-MS; Agilent 6890-5975, Agilent Technologies, 128 not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Laboratory tests
133 Initially, dog A was introduced to the bark beetle pheromones using the synthetic odour from 134 a commercial dispenser, ETOpheron ® (Pheronova AG, Switzerland), which is used in bark 135 beetle monitoring traps. Because of the dispensers construction of fabric with a plastic shell it 136 was not 100% sure that the dog learned the scent of the pheromone components as the target 137 odour or if it learned any other odour of the dispenser materials. It is easy to inadvertently 138 train a dog to detect an unexpected or impure source when attempting to train to a pure 139 compound. To be sure that the dog learned the right odours we subsequently trained the dog 140 on pure synthetic semiochemicals applied to cotton pads. Non-target odours, that could 141 disturb search, were also used in the training and consisted of items found in a forest setting 142 such as vegetation odours from spruce needles, cones, resin, bark, moss, and animal odours 143 (i.e. scent from feathers, fur, hoofs and faeces). All non-target (disturbance) odours were 144 collected in the forest, or donated by local hunters (fur and hoofs from moose, deer, and boar). 145 Both target and non-target odours were stored in jars of glass and transferred by aeration to 146 cotton pads to ensure that the background odour of cotton was present in both target and 147 disturbance odours. 148 The training platform used here (2D illustrations in Figure ESM_1 and video in ESM_4_V1), 149 was developed by Stig Meier Berg and Geir Kojedal, Spesialsøk, Selbu, Norway, based on an 150 idea from Hundcampus, Hällefors, Sweden (FISCHER-TENHAGEN et al. 2011). It is designed to 151 let the dog work independently, to minimize the cues from the handler and to be easily 152 manoeuvred by the handler creating a more effective learning situation with a high rate of 153 not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 12, 2018. ; https://doi.org/10.1101/262386 doi: bioRxiv preprint 2018- 03-11 8 (25) opportunities to reward the dog for desired behaviour. Disturbance odours were presented 154 together with one or several semiochemical stimuli in a movable tray with 7 positions ( Figure  155 in ESM_2). 156 For evaluation of the dog detection performance with decreasing amounts of odour molecules 157 over time nine trials were conducted to evaluate the dogs' identification performance with 158 each synthetic semiochemical. For this trials we used 4 of the prepared 5 steel tins containing 159 cotton pads with synthetic semiochemical. Since the trials were conducted over several days 160 (1 hour through 84 hours after the cotton pads being placed in the tins and stored in room 161 temperature) we used a new tin every day. This was done to make sure that the tins weren't 162 contaminated with any other scents such as odour from the dogs. Every trial session lasted for 163 approximately one minute (50-70 seconds). 164 To compare different stimuli linear layouts, mixing target and non-target scent, on the movable 165 tray, the two dogs were tested in three trials with each stimuli layout (ESM_2). 166 Outdoor tests 167 To train the dogs to pinpoint the target odour source outdoors, pieces of the cotton pads 168 containing synthetic pheromone odours as those used for platform training were hidden in 169 cracks of the bark of several species of trees. The cotton pieces were placed in the height of 170 the nose of the dogs and the dogs were shown where to sniff for the target (video 171 ESM_4_V2). When the dog found the cotton piece holding the target odour, it was 172 immediately rewarded by the sound of the clicker and a piece of food delivered between its 173 nose and the odour source. Several pieces of cotton with either target or non-target odours, 174 were put in cracks of the bark in a small area (30x30 cm) to ensure the dogs did not use visual 175 cues for close-range target location. When the dogs were able to consistently (~ 100%) locate 176 the pads, the dogs were gradually sent from longer distances to locate the tree with the cotton 177 not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 12, 2018. ; https://doi.org/10.1101/262386 doi: bioRxiv preprint 2018-03-11 9 (25) pad, allowing the dog to detect decreasing amounts of target molecules and to follow the 178 odour to its source. 179 Dogs were trained in the winter under a variety of weather conditions (e.g. rain, snow, sun). 180 Training trials using synthetic odour were conducted on average once a week during 2009 and 181 2010. The temperature ranged from 2 to 28 °C. The handler determined the search strategy to 182 best cover the assigned area based on wind conditions and terrain. These protocols were 183 employed to simulate future practical field survey conditions. 184 A proof-of-concept test, evaluating the detection by dogs of spruces that were known to be 185 recently attacked by bark beetles, was conducted at the Nature Reserve of Notteryd (near 186 Växjö, Småland, Sweden). The area consisted of wind-felled trees and standing healthy 187 spruces. In the spring of 2009, 95% of all spruces in the reserve were already killed by bark 188 beetles. The remaining spruces that were still alive stood together in clusters of 10-15 trees. 189 We felt these circumstances made this particular Nature Reserve an optimal area to first try 190 the dogs on natural attacks. Another series of tests were conducted at a production-forest in 191 Nottebäck, also near Växjö, Småland, Sweden, with the permission of the owner of the forest. Occasionally, the dogs reacted with an increased interest when a new non-target disturbance 225 odour was presented. When this happened, the handler stood silent and just waited until the 226 not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 12, 2018. ;https://doi.org/10.1101https://doi.org/10. /262386 doi: bioRxiv preprint 2018 dog stopped investigating the new non-target odour and, if the dog did not continue to search 227 by itself, gave the dog a new command to start sampling the other tins again. After a few 228 encounters with the new non-target odour the dogs' interest decreased since they learned that 229 there would not be any reward for that particular odour. Even though the dogs appeared to 230 alert on new, disturbance odours (mostly edible items like cookies and chips or scents from 231 other animals) the handler did not record such behaviour as an alert. When alerting on a target 232 odour, both dogs stopped sampling, and waited for their reward, in contrast to increased 233 sampling a tin in order to investigate a disturbance odour. 234 235 Chemical quantification by odour collection and GC-MS was routine with a limit of 236 quantification (LOQ) of < 0.1 ng/min. However, two days later, we found that most stimuli 237 titres, still well biologically active, decreased to below the LOQ. Using estimates based upon 238 linear regression, chemical data indicates that by the third day some compounds were very 239 close to zero (Table 1). 240

Chemical stimuli strength
The dogs responded to estimated doses of 10  ng/15min releases or less. The four different 241 semiochemicals were learned equally well, and responses to sub-picogram release rates of 242 stimuli aged up to 3.5 days remained stable (Table in ESM_3). 243

244
The responses of both dogs to target odours are summarized per target scent in Table 2. The  245 dogs achieved a mean of 99% correct indications; 1% of the incorrect indications were either 246 false positive (alerting to a non-target odour; dog A) or false negatives in the beginning of a 247 trial session (dog B) ( Table 2). None of the dogs sampled all tins in every repetition. In each 248 repetition four tins were presented, but the trainer could never know where the dog would 249 start searching or in which direction it would continue its search. The only dispenser tin 250 not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 12, 2018. ; https://doi.org/10.1101/262386 doi: bioRxiv preprint 2018-03-11 12 (25) always sampled was the tin holding the target scent. This explains the high success rate of 251 99% for correct positives for the target scent (at which tin the search will stop), but the much 252 lower rate, 55% for the correct negatives with direct sampling of empty tins before finding the 253 target scent. 254 To increase the dogs sampling of all presented tins, we tried the dogs in different kinds of 255 stimuli layouts with zero to three different target odours presented in the same trial, Figure A  256 in ESM_2. The only clear effect was for the layout with no target scent, where response 257 decreased with time, Figure B in ESM_2 258 Interestingly, over the >3 days of testing combined with chemical sampling, the correct 259 responses remained consistently high irrespective of substance ( Table 2). The positive 260 responses showed no decline with estimated chemical stimuli levels (Fig 1A), indicating that 261 stimuli levels were above animal detection limit for the period. The correct negative responses 262 (no alert to disturbance odours) declined with the estimated stimuli strength since the dogs 263 learned that these odours weren't going to be rewarded (Fig 1B). 264

Outdoor tests
265 During off-season training, dogs were introduced to cotton pads initially placed at nose height 266 in the cracks of the bark in different kind of trees (Video ESM_4_V2). Dog A, previously 267 trained as a search and rescue dog, just needed to come into contact with one of the new learned 268 target odours to expect a reward hence follow it to the source and pinpoint it to its handler. Dog 269 A also alerted the found target source by barking. Probably because this was the trained alert 270 when locating a hidden human as a search and rescue dog. Dog B, however, which had no 271 previous search training, had to learn how to follow the odour plume to the source. Dog B was 272 not trained to perform any other alert than pinpointing the source of the target odour. This dog 273 did not know any other way to receive its reward but putting its nose on the target source. The 274 target source became a button to push to get its reward. 275 not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was this version posted March 12, 2018. ;https://doi.org/10.1101https://doi.org/10. /262386 doi: bioRxiv preprint 2018 Bark beetle activity usually began at the end of April, when the temperature increased to over 276 20 °C, allowing us to test the dogs' ability to detect natural pheromone from attacking spruce 277 bark beetles. Dog A successfully found the first spruce that was under attack, on the first day. 278 The spruce in question showed no signs of the attack at first sight, but further inspection at 279 close range revealed that the first bark beetles were drilling their way in to the spruce bark and 280 the sound of their drilling could also be heard. This finding was crucial in demonstrating that 281 it is possible to train a dog on a synthetic odour and subsequently showing that it will alert to 282 the natural odour under field conditions. All training and detection beetle in the Nature 283 Reserve was terminated at the end of May when so many spruces were under bark beetle 284 attack that the smell from the attacked trees became obvious even for the human nose. 285 Both dogs were also successful in locating sparser attacks in production forest stands, where 286 attacks were neither known to the dog handler nor the forest manager. In the first area 287 searched, dog A detected and alerted to a single, wind-felled spruce that had been infested by 288 bark beetles. In the second area searched, the same dog found seven infested standing spruces. 289 Five of them stood together in a cluster among old attacks. Two were located in a felling 290 edge. 291 In the third area, the dog detected five infested spruces, both standing and wind-felled. In this 292 area all the spruces were located near a felling edge by a clear-felled area where felled trap-293 trees were placed. The dog started its search with detecting and alerting on the synthetic 294 pheromones from the trap-trees. When sent to continue its search the dog detected, 295 recognized, followed, and alerted on the natural pheromones emitted from the bark beetles in 296 standing trees (as shown in video ESM_4_V4). 297 The handler observed by GPS a majority of successfully located sources of natural pheromone 298 to be detected within 50 m, but both dogs located sources in a behavioural sequence over a 299 range of 50 − 100 m (Fig 2A). No differences in detection distance by GPS could be seen 300 not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.  (25) among areas with attacks (10) known or unknown (10) to the handler. Later analysis of the 301 GPS-tracks showed several occasions where the more experienced dog A changed direction 302 and was able to detect the pheromones from bark beetle attacked trees at a distance of over 303 100 m from the source and follow it to the source (Fig 2B). In the 20 areas visited, the dogs 304 found in total 193 trees infested by bark beetles, in 77 different groups of attacked trees. 305

306
Training canines to detect bark beetle-infested trees poses some important limitations, 307 including the relatively short season available for using trees at various stages of attack, as 308 well the risk of inducing a full-blown tree attack by placing pheromone for training purpose 309 on a host tree during the actual beetle flight period. While it is probably possible to train a 310 detection dog to locate spruces that have been attacked by bark beetles by just letting the dog 311 sniff an attacked spruce and reward the dog, such a "natural" method will not teach a dog to 312 recognize the different kinds of semiochemicals the bark beetle releases over the course of an 313 attack. Therefore, we chose to train the dogs to recognize a series of synthetic pheromone 314 compounds and using an indoor training platform. In this study, we demonstrate that canines 315 trained on synthetic bark beetle pheromone compounds at low (sub-picogram) levels, indoors, 316 can later recognize naturally-produced pheromone over long distances, outdoors. 317 Additionally, by using synthetic sources of the bark beetle pheromone in the laboratory, it is 318 possible to train dogs off-season long before the bark beetles start their flight period in the 319 field, and the dog handler has control over which odours the dog learns, one at a time and at 320 very low concentrations. The indoor training of canines also has the benefit in that other 321 environmental distractions are minimized, thereby allowing the dogs to concentrate on and 322 learn the target odours. 323 In the field, detection dogs that work over large areas ("off-leash") can often be seen lifting 324 their nose up in the air and then make a sudden change in direction of travel. This likely 325 not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.