Canine Scent Detection of Human Cancers

PAGE 1

Canine Scent Detection of Human Ca ncers: A Biological, Cell Signaling, Behavioral, and Experimental Approach By Emily Moser A Thesis Submitted to the Divisions of Natural Sciences New College of Florida In partial fulfillment of the requirements for the degree Bachelor of Arts Under the sponsorship of Alfred Beulig Jr. Sarasota, Florida May, 2010

PAGE 2

Canine ii Canine Scent Detection of Human Ca ncers: A Biology, Cell Signaling, Behavior, and Experimental Approach Emily Moser New College of Florida, 2010 Abstract Interest in the hypothesis that dogs are able to detect cancer in humans began in 1989 when a letter sent to the journal, The Lancet, implied that a dog had detected melanoma on its owners skin. Since then, considerable anecdotal and experimental evidence has been found to support this hypothesis. In fact, studies on the topic ha ve lead researchers to believe that dogs can detect several diffe rent types of cancer ea rlier than machines built for the same purpose. Canines are able to detect whether people have cancer by smelling cancer biomarkers in their breath with a sensitivity and specificity sometimes as high as 99%. The purpose of this thesis is to review previous st udies on the topic and explore in greater detail how cancer biomar kers are produced and how dogs smell them. In addition two experiments are included. The first is a study I participated in on canine scent detection of ovarian cancer. The sec ond is a study I led which used the same techniques to train a dog to de tect the presence of peanuts, a potentially lethal allergen, in food. Alfred Beulig Jr. Division of Natural Sciences

PAGE 3

Canine iii Acknowledgments I would like to thank Dr. Michael McCulloch for inviting me to work on the ovarian cancer study. This research experience was an integral part of my thesis, and led to my first publication in a peer review journal. To the members of my Baccalaureate committ ee, Dr. Beulig, Dr. Clore, and Dr. Bauer thank you for your time and support. Finally, I am eternally gratef ul to my parents, Mark a nd Paula Moser, and boyfriend, Nathan Kruer, for their encouragement and patience.

PAGE 4

Canine iv Table of Contents 1. Introduction 1 2. How m ammals detect odors and pheromones. 2 2.1 General Olfaction Pathway 3 2.2 Pheromone sensing in mammals before 2000 5 2.3 Progress made in odorant signaling 6 2.4 Progress made in pheromone signaling 9 2.5 Formyl peptide receptors 12 2.6 Functions and implications of FPRs 14 2.7 Conclusion 15 3. Olfactory detection of cancer ba sed on MHC dependent odor components 16 3.1 The immune system in general 16 3.2 T cells 17 3.3 The major histocompatability complex 19 3.4 Theories on the distinct odor of the MHC 20 3.5 General Cancer biology 24 3.6 Specific mechanisms tumor cells employ to escape recognition 26 3.7 A possible application-early dete ction of cancer 28 3.8 Conclusion 28 4 Systematic Review of previous canine scent detection of cancer studies 29 4.1 Methods 29 4.1a Study identification 29 4.1b Study eligibility 29

PAGE 5

Canine v 4.1cData extraction 30 4.2 Results 30 4.3 Discussion 34 4.4 Conclusion 38 4.5 Table 1 39 4.6 Table 2 40 4.7 Table 3 41 4.8 Figure 4 42 5. Experiment 1: Canine scent detection of ovarian cancer 43 5.1 Sub-Hypothesis 44 5.2 Methods 44 5.2a Patient and control br eath donors 44 5.2b Equipment and breath sampling 45 5.2c Dogs 46 5.3 Experimental Setup 47 5.4 Training 48 5.5 Testing 49 5.6 Data Management and Analysis 50 5.7 Analysis of DMSP absorbent chemistry strips 51 6. Two Up and Coming Cancer Detection Methods 51 7. Experiment 2: The peanut project 53 7.1 Sub-hypothesis 53 7.2 Why peanut dogs are important 54

PAGE 6

Canine vi 7.3 Methods 55 7.3a Subject 55 7.3b Experimental setup 55 7.3c Personnel 55 7.3d Sample stations 56 7.3e Classification of dogs response 56 7.4 Pre-training 56 7.5 Training 57 7.6 Testing 58 7.7 Instrumental Conditioning (T raining) 59 7.8 Results 62 7.9 Discussion 65 7.10 Conclusion 68 8. Appendix 1 Pictures from canine ovarian detection experiment 70 9. Appendix 2 Sequential phases of dog training and test ing 71 10. Appendix 3 Canine scent detectio n of ovarian cancer: Data Recording Sheets 72 11. Appendix 4 PC/GC/FT-ICR MS 73 12. Works Cited 75

PAGE 7

Canine 1 1. Introduction Initial interest in the hypothesis that dogs might be able to detect cancer in humans began in 1989 when Williams and Pembroke sent a letter to The Lancet in which they described a case where a woman was encouraged to get a skin lesi on examined after her dog showed an inordinate amount of interest in the spot on her skin (Williams and Pembroke, 1989). After clinical examination, the lesion proved to be a malignant melanoma. Several similar cases have been reported since then. Pickel and Willis et al. bot h published papers in 2004 indicating that the hypothesis suggested by Williams and Pembroke mi ght be valid. To date, studies have been published on the detection of bladder (Willis et al., 2004), lung (McCulloch et al., 2006), breast (McCulloch et al., 2006), prostate (Gordon et al., 2008), ovari an (Horvath et al., 2008), and melanoma (Williams and Pembroke, 1989; Pickel et al., 2004) cancers using a dogs sense of smell. Early detection of many cancers, although essential for treatment effectiveness, can be both difficult to achieve and introduce new health risks. High-resolution scanning technologies such as CT, MRI and PET are highly accurate, but are costly and carry the risk of unnecessary biopsies for benign lesions (Jett, 2005). CT scan s also increase radiation exposure, which could increase risk of cancer over a period of several decades of us e in screening (Martin and Semelka, 2006). Lowresolution scanning with ultrasound, while avoiding radiation exposur e, is less apt to find small tumors. The combined use of ultrasound, mammog raphy, and breast MRI is effective in saving the lives of women without a prior history of breast can cer, but is still not wit hout risk of unnecessary and invasive biopsies in benign cases (Gotzsche and Ni elsen, 2006). Serum biomarker tests such as PSA used in prostate cancer screen ing can similarly find early stage disease, but also increase risk of unnecessary biopsies (Harris a nd Lohr, 2002). Other tumor marker tests such as CA-125, PSA,

PAGE 8

Canine 2 CEA, CA 19-9, and CA 15-3 have frequent false positives as they can be elevated in m any noncancerous inflammatory conditions; these too, if used in screening, would lead to many unnecessary biopsies. Thus, there is still a great need for new low cost, low-risk methods for primary cancer screening in the general population. If indeed dogs are able to detect cancer in humans by sniffing biological samples such as exhaled breath or urine, then new avenues for earlier diagnosis of cer tain cancers could be possible. Interest in the potenti al analysis of volatile organic compounds for diagnostic purposes is growing and such a talent would be of great va lue in the development of diagnostic tests. 2. How mammals detect odors and pheromones Odorant and pheromone sensing is essential fo r the survival of many species of animals. The olfactory system in mammals can detect and di stinguish a vast array of volatile chemicals with a large variety of structures. The sensitivity of some animals to odorants is astounding. Canines can smell in parts per trillion (Johnston, 1998)! This is like detect ing a few molecules diluted in enough water to fill an Olympic sized swimming pool In the past 9 years significant discoveries have been made in the field of olfaction. Since the year 2000, new receptors, like trace-amine associated receptors (Liberles and Buck, 2006) or TAARs, guanylyl cyclase D, or GC-D (Leinders-Zufall, 2007), formyl peptide receptors, or FPRs (Liberles et al ., 2009 and Riviere et al., 2009), and V2r83, vomeronasal 2 ol factory receptors found in sensory organs like the Gruenberg ganglion (Lin et al., 2004) and Fl eischer et al., 2007), and the Sept al organ of Masera (Kaluza et al., 2004) have been discovered in a variety of different species. In addition, critical information about the organization and pathways of the olfactory signals has be en found that forces scientists in the field to re-evaluate past work, and make new conclusions about mammaian sensory

PAGE 9

Canine 3 capabilities and the f uncti on of certain structures in the se nsory pathway. In this section, the general state of knowledge in the field of olfactio n prior to the year 2000 will be discussed with a focus on olfactory and pheromone signaling. Then ne w discoveries will be examined, with a focus on the discovery and significance of formyl peptid e receptors to pheromone sensing in mammals. 2.1 General Olfactory Pathway in Mammals The olfactory system is responsible for the detection of 1) pheromones, which are chemicals released by animals to affect conspeci fics and regulate populations and their social interactions (Wilson, 1963), and 2) odorants, or volatile chemicals derived from food or the environment (Beets, 1970). As of 2000, scientists hypot hesized that to detect the 2 different types of chemicals, mammals completely separated odor and pheromone detection into two different neural pathways in the brain (Buck, 2000). T hough, this hypothesis has been proved partially incorrect, there are advantages to reviewing the proposed organiza tion of the olfactory pathways, before addressing recent deve lopments in the field. Mammals have an olfactory epithelium (OE), a specialized epithelial tissue inside the nasal cavity, which contains olfactory receptors (ORs). Olfactory receptors are 7 transmembrane domain G-protein coupled receptors, in the GPCR superfamily that recognize a wide variety of structurally diverse odorants (Buc k and Axel, 1991). When odorants bind to the receptors a signal is sent via neurons to the main olfactory bulb (MOB) and from ther e to the olfactory cortex where the signal is further processed. Interestingly these neurons ar e bipolar; they both receive and transfer information. Each receptor is a member of one of about a thousand different types, coded by one gene per neuron type. The receptors have strong punctuate expression and each type is expressed in one of four spatial zones in the epithelium, interspers ed with other types of receptors.

PAGE 10

Canine 4 The spatial organization rem ains constant as the signal is passed from the neurons to the MOB where the zones remain delineated (Ressler et at ., 1993). The signal originates from the binding of an odorant to a receptor. This binding stimulates the G protein to activat e adenylyl cyclase III (AC-3) which in turn causes cAMP production. Th e cAMP opens cyclic nucleotide-gated cation channels which causes a calcium influx and membra ne depolarization (Fires tein, 1992). The signal is passed to glomeruli, (same t ype neurons pass to one of two gl omeruli). Different combinations of olfactory receptors detect different odorants ca using different combinations of glomeruli to be activated. A smell can be very different dependi ng on the pattern of glomeruli activated, even when many of the same receptors are signaling (Beets, 1970). From the glomeruli the signal is passed to a mitral/tufted cell and on to the olf actory bulb. (See Figure 1.) Mitral/tufted cells are referred to together as it is di fficult to differentiate between th e two in function and structure (OConnor and Jacob, 2008). Figure 1. The olfactor y system: from receptors to mitral cells. After the odorant binds to receptors in the cilia, the sign al is transported up the olfactory neurons, through the axons, to the glomeruli, and from there, to the mitral cells. (Picture taken directly from http://www.personal.kent.edu/~cearley/Che mWrld/smell/olf_bulb.gif .) From the olfactory bulb, the signal seems to be transmitted along the myelinated fibers of the lateral olfactory tract in highe r vertebrates (or the medial olf actory tract in lower vertebrates

PAGE 11

Canine 5 like fish) (Nieuwenhuys, 1963) and e nds in the amygdala, piriform co rtex, or entorhinal cortex. (This olfactory pathway is unique, as most sensory information travels first through the thalamus before being processed in the cerebral cortex (Kalat, 2007)). The amygdala is not a direct processing center; rather, studies suggest it is involved in feeding regulation (Slotnick, 1985, and Halasz, 1990). Similarly, the entorhinal cortex se ems to function as an olfactory memory structure as opposed to a primary processing center (Staubli et al., 1986). In contrast, the piriform cortex is responsible for identifying odors. Pyramidal cells in the rostral part of this cortex project to the ipsilateral anterior olfactory nucle us, tenia tecta and olfactory tubercle, while those in the medial and posterior piriform cortex project to the caudal piriform and entorhinal areas (Luskin and Price, 1983a). From those secondary olfactory centers, the signal is eventually sent to the hippocampus, where some behavioral responses are stimulated (Macrides et al., 1982) and to the neocortex, where the detection and discrimination of odors occur (Onoda et al., 1984). 2.2 Pheromone sensing in mammals before 2000 Pheromone sensing was, and still is, much less understood than olf actory. In fact, the pheromone ligand had not even been identified by 2000 (Buck, 2000). However, it was known that there were two types of pheromone receptors, V1Rs and V2Rs, which like the ORs in the olfactory epithelium, are from the GPCR superfamily. In mi ce, they are located in the vomeronasal organ, a tubular structure in the nasal se ptum that is connected to the nasal cavity by a small duct, discovered by Ludwig Jacobson in 1813. The receptors are expressed in tw o different longitudinal zones, with the upper zone expressing V1Rs and the lower zone expressing V2R receptors (Dulac and Axel, 1995). Similar to the olfactory epithelium, each neuron expresses one VR gene, and neurons expressing the same VR are scattered in one zone. From the neuron, a signal is passed to

PAGE 12

Canine 6 the acc essory olfactory bulb (AOB). The signal is transduced by the neurons and transmitted to glomeruli and then to mitral cells. Seemingly unlike in the main olfactory bulb, several glomeruli types send information on to the same mitral cell (Takami and Graziadei, 1991). However, the organization of these cells remained unclear and th e pathway the signal took after the mitral cells was unknown. 2.3 Progress made into odorant signaling Since the year 2000 there has been significant progress made in deducing the pathway the olfactory signal takes in the mammalian system. Individual odorants bind to a specific subset of receptors and individual receptors bind to a specif ic subset of odorants (reviewed in Su et al., 2009). These receptors vary in their tuning and respond to a different nu mber of odorants. In addition, the larger the concentrations of an odorant, the larger the number of receptors that respond. Odorants also affect the duration of the signal that is produced depending on the receptor it binds to; some odorants elicit a short response, while others elicit a long-l asting response (Hallem et al., 2004). The length of response can va ry depending on the rece ptor as well as the odorant. Much more on how the chemical signals ar e translated to electrical signals in mammals has been discovered. Before the hydrophobic odorant ca n bind to the receptor it must pass through an aqueous fluid. The odorant does this by bi nding to odorant binding pr oteins (OBPs) which solubilize the airborne odorant (Su et al., 2009). The OBPs tran sport the proteins to specific receptors. Then odorants bind to the receptor po cket surrounding the transmembrane domains 3, 5, and 6 through hydrophobic and van der Waals forces (Katada et al., 2005). Some amino acids in the carboxyl terminal domain and third intracellula r loop of the olfactory receptors seem to be involved in the coupling and activation of G olf (Katada et al., 2004).

PAGE 13

Canine 7 A guanine nucleotide ex change factor called Ric8B, enhances the accumulation of the G protein at the cell membrane which improves th e OR coupling (Von Dannecker et al., 2006). The activated G protein stimulates AC-3, which results in an increase in cyclic AMP production. The increased concentration of cAMP causes the cy clic nucleotide gated channel (CNG) to open allowing an influx of calcium. The calcium influx then results in the opening of calcium-gated chloride channels (Stephan et al., 2009). See Figure 2A, (Kato & Touhara, 2009). Odor perception depends on more than just the chemical stru cture and concentration of odorants, (Kato and Touhara, 2009). Odor adaptation, or the ability of th e olfactory system to adjust its sensitivity at different stimulus intensities (Kato and Touhara 2009), allows high sensit ivity retention during long lasting or repetitive odor stim ulation. Olfactory sensory neurons can be subjected to negative feedback regulation that causes CNG channels to close in the presence of too much calcium (Zufall and Leinders-Zufall, 2000). See Figure 2B (taken from Kato and Touhara, 2009). Elevated levels of calcium bind to calmodulin and inhib it the CNG channels and activate phosphodiesterase. CaMK2 inhibits AC3, and PKA and GRK3 desens itize the receptor (Kato and Touhara, 2009). Figure 2A

PAGE 14

Canine 8 Figure 2B Olfactory transduction pathway (Fig 2A) & Negative fe e dback (Fig 2B) taken directly from Kato and Touhara (2009). From the olfactory sensory neuron the signal is passed to glomeruli. The glomeruli form synapses with the dendrites of mitral and tufted (M/T) cells (Su et al., 2009). These second order neurons then transfer the signa l to innervate higher brain re gions. These regions integrate information from other sensory modalities, like information from past experiences and information concerning the mammals behavioral state. The M/T cells synapse with pyramidal neurons in the olfactory cortex. The pyramidal ne urons act as coincidence detect ors, and only fire their action potential when certain M/T cells are active t ogether (Poo and Isaacson, 2009). The pyramidal neurons form connections with other regions such as the orbi tofrontal cortex, thalamus, and hypothalamus, integrating the different sites (Su et al., 2009) and stimulating a response to the original odorant. New receptors have been discovered in the main olfactory epithelium. TAARs, trace-amine associated receptors, act as a sec ond class of olfactory receptor that is activated by volatile amines found in urine (Libreles and Buck, 2006). The TAARs were discovered similarly to formyl peptide receptors. (FPRs will be discussed later in this paper). Libreles and Buck (2006) used real time

PAGE 15

Canine 9 qPCR with prim ers matching GPCRs not previously implicated in odor, pheromone, or taste detection to amplify the unknown GPCRs, and then used RNA in situ hybridization to confirm the olfactory sensory neurons expressi on of Taar7d and Taar9. They used primers specific for each mouse Taar gene in qPCR reacti ons with cDNA from the olfact ory epithelium and other mouse tissues to make sure that the genes were only expre ssed in the OE. However, they failed to test all relevant tissues, as several important sensory organs (discussed below) had not been discovered yet. The TAAR receptors will be mentioned again in a section on pheromone signaling. Another type of receptor discovered, called the guanylyl cyclase D (GC-D), is also expressed by olfactory sensory neurons in the ma in olfactory epithelium. These OSNs project to the dorsal part of the olfactory bulb and respond to everything fr om peptide hormones and CO2 to pheromones (Leinders-Zufal et al., 2007). In additi on, some ORs are expressed in the Septal organ of Masera (SO), which comprises an island of se nsory tissue located on either side of the nasal septum (Kaluza et al., 2004). The SOs proposed f unction is an alerting role or mini-nose which responds to ligand from food and the environment (Su et al., 2009) These newly identified organs and receptors add to the intricate, yet complicated olfactory system. 2.4 Progress made into pheromone signaling Since 2000 there has been some progress in determining the pheromone signaling pathway. However, the nature of the pheromones regula ting behavior in mammals is largely unknown (Rodriguez and Boehm, 2008). In 2006, it was discove red by Levai et al., ( 2006) that there are 44 different OR genes expressed in the apical zone of the VNO epithelium. In addition, it has become apparent that, although in general pheromones ar e recognized in the VNO, and odorants in the MOE, there is some overlap, and some pheromone s can be processed in the MOE ( Lin et al.,

PAGE 16

Canine 10 2004), while som e odorants are recognized by rece ptors in the VNO (Sam et al., 2004). Because this is such a significant disc overy, a more detailed di scussion of the experiment by Sam et al., (2004) will be included. Sam et al., (2001) used calcium imaging of single murine VNO neur ons containing Fura-2 dye. They tested the mouse VNO neurons to see if they could detect odorants by creating 9 mixes of odorants from 82 sources. If a neuron responded to a mix, then it was tested on each of the odorants in the mix to identify what had activated it. One to one and a half percent of the neurons responded to a single mix. Some neurons were also stimulated by 18 single odorants labeled as camphoric, floral, musky, sweet, or woody. Interes tingly, some neurons responded to more than one odorant, and some odorants activated more than one neuron. However, the neurons remained for the most part very specific. For example the VNO neurons could differentiate between indole and skatole, two molecules that are only one methyl group different in structur e. It is important to note that appropriate concentratio ns of odorants were used, and th e intensity of neuron activation was similar to that seen in neurons by normal pheromone ligands. (In other words, the experimental design was well done, and the controls conducted were adequate.) Sam et al., (2001) suggest that these odorants may act similarly to pheromones, a nd stimulate certain instinctual behaviors. Vomeronasal signal transduction remains uncl ear, but some suggestions for chemosensory transduction in VSNs have been made. V1Rs are coupled with G i2, and stimulate G -mediated calcium signaling (Touhara and Vosshall, 2009 ). V2Rs are likely coupled with G o to stimulate the calcium cascade. The signal transduction involves diacylglycerol and a di acylglycerol-activated cation channel, along with a transient recepto r potential channel TRCP2 which is strongly expressed in both apical and ba sal compartments of the VNO (Limann et al., 1999) (reviewed in

PAGE 17

Canine 11 Rodriguez and Boehm 2008; and Touhara and Vossh all, 2009). The mechanism for activation of TRPC2 via a VR/phospholipase-C pathway remain s unknown and additional ion channels could likely be involved (Touhara and Vosshall, 2009). Signals transduced from V1R neurons and ORexpressing VSNs go to the anterior accessory olfactory bulb, while V2R neurons signal to the posterior AOB. The neurons form several conve rged glomeruli and from there the sensory neuronal axon terminals synapse with the sec ondary neurons (Del P unta et al., 2002). The secondary neurons project to the bed nucleus of the accessory olfactory tr act (BAOT), and to the bed nucleus of the stria terminalis (BST), and the medial and posteromedial cortical nuclei of the amygdala (Brennan and Kendrick, 2006). From there, information is relayed to the MPOA-AH and hormones are released to cont rol the behavioral responses triggered by the pheromones. Significant progress has been made in identify ing some ligands that act as pheromones and trigger stereotypic behavior in some mammals. There are several volatile molecules present in mouse urine, called mouse urinary proteins (M UPs) that are VSN a gonists, including 2sec -butyl4,5-dihydrothiazole, 3,4-dehydroexo -brevicomin, farnesene, n -pentylacetate, 6-hydroxy-6methyl-3-heptanone, isobutylamine, 2-hepta none, and 2,5-dimethylpyrazine (Novotny, 2003). Some MUPs can activate V2Rs in the basal laye r of the VNO in the absence of volatile urine, (Stowers et al., 2004). In addition, synthetic MUPs alone can cause aggressive behaviors in male mice (Touhara and Vossell, 2009). There are also major histocompatibility comple x (MHC) peptides that have been shown to selectively activate some vomeronasal neurons located in the basal VNO epithelium (LeindersZufall et al. 2004). The MHC peptides are able to induce a pregnancy block, called the Bruce effect, in which exposure to a foreign male bl ocks implantation of embryos sired by a familiar male, (Leinders-Zufall et al., 2004). Lienders-Zufall et al., was able to demonstrate aspects of the

PAGE 18

Canine 12 Bruce effect by using electrophysio logical studies, in which whol e-cell current recording from VNO neurons produced membrane depolarization a nd action potential when the MHC peptides were applied. Other possible ligands include extraorbital l acrimal gland specific peptides (ESPs), (also called exocrine gland-secreted peptides), that are secreted by extraorbital lacrimal glands (Kimoto et al. 2007). ESPs are male spec ific, and they activate V2R neurons. This is important because it suggests that V2R-expressing neurons can re spond to nonvolatile pheromones (Touhara and Vossell, 2009). One newly discovered pheromone receptor, called TAAR, responds to volatile amines found in urine (Fleischer et al., 2007) that act as alarm pheromones originating from stressed conspecifics. As mentioned previously, TAARs are also expressed in the main olfactory epithelium. Another organ that expresses TAARs was only recently discovered and is called the Gruenberg Ganglion (GG). The GG also contains sensory neurons that express ORs activated by volatile alarm pheromones required for the freezi ng behavior in mice (Brechbuhl et al., 2008). These new pheromones, receptors, and organs ad d to the intricate a nd specific pheromone signaling pathways found in mammalian species. One of the most interesting pheromone receptors discovered was published only a few months ago. This receptor is known as the formyl peptide receptor and has been suggested for acting as a disease sensing receptor in mice and ot her mammals. The rest of this paper will focus on the discovery, function, and implication of thes e receptors in mammals, particularly mice and canines. 2.5 Formyl Peptide Receptors

PAGE 19

Canine 13 Liberles et al., (2009) discovered five pep tide receptors in vomeronasal epithelium using high-throughput screen for GPCRs expressed in mouse vomeronasa l tissue. Around the same time, Riviere et al., (2009), identified the same recep tors through reverse transcriptase PCR using primers corresponding to putative heptahelical recepto r transcripts. The five transcripts identified Fpr-rs1, Fpr-rs3, Fpr-rs4, Fpr-rs6, and Fpr-rs7, all belong to the formyl peptide receptor gene family. Two FPR mouse genes not transcribed sole ly in the VNO tissue play a role in immunity and are expressed in immune cells and other cell types (Migeotte, 2006). Both groups of scientists, Riveiere et al., and Liberles et al., performed quantitative PCR a nd determined that the 5 genes were only expressed in vo meronasal tissue. Both gr oups also performed RNA in situ hybridization to show that there was strong punctuate expression of the transcripts in the neuroepithelium similar to patterns observed in V1R, and V2 R genes. Both groups also used double in situ hybridization to show that there was no co-expression of FPR genes pertaining to other families, and that expression was monogenic. Liberles et al., (2009) then relied on BLAST and other data bases to search for genes encoding FPR related proteins in other mamma ls. The authors concluded that VNO FPRs are probably only present in a few se lect mammals, possibly only in cer tain rodents, and the likelihood they are expressed in dogs, humans, cows, or horses is very small. Riviere et al., (2009) evaluated the expression of Fpr-rs3 in rat and gerbil VNO tissue and found punctate messenger RNA and protein expression similar to what was observed in mice. In their abstract Riviere et al., (2009) suggest that they believe there are formyl pep tide related genes in mu ltiple mammalian species VNOs, though they do not specify any others beside s mice, rats, and gerbils. Riviere et al.,(2009) went on to determine what ligands might bind to the 5 FPR receptors. They transiently expressed the receptors in HEK293 cells. Then, the li gands known to activate the two immune FPR

PAGE 20

Canine 14 recep tors, FPR1, and FPR-rs2, were tested by measur ing agonist-induced calcium transients in cell loaded with ratiometric calcium sensitive to a reporter dye. Ligands tested included: formyl peptide, fMLF, CRAMP, lipoxin A4, diluted urine, and an elevated extr acellular K1 solution. It was discovered that the five receptors were stimulated by CRAMP, uPAR, LXA4, and fMLF. Riviere et al., (2009) concluded th at though the functional role of the 5 receptors remains to be determined, their agonist characteristics suggest the receptors may allow for the detection of various contaminated compounds, and the identificati on of sick conspecifics as explained below. 2.6 Possible Functions and Implicati ons of Formyl Peptide Receptors Though mice have been observed to avoi d sick conspecifics it was unknown how (Kavaliers et al., 2004, 2005, and 2006). A possible expl anation has resulted from the discovery of the formyl peptide receptor genes. N-terminal formyl groups are located on peptides derived from bacteria, mitochondria and plant chloroplasts (Gig lione et al., 2000). The recognition of formylated peptides by VNO neurons could allow for the detec tion of edible plants, signal the decay of food, identify a bacterial infection in a conspecific, or even pick up on individuality cues from waste products that could signal the presence of pred ators in an area (Liberles et al., 2009). This discovery may open up a whole new field of research into the molecular basis for smelling disease (Munger, 2009). For example, there has been a lot of research recently into canine scent detection of human cancer. Trained dogs can smell a persons breath, urin e, or tissue sample, a nd indicate whether or not that person has bladder (Willis et al., 2004), lu ng (McCulloch et al., 2006), breast (McCulloch et al., 2006), prostate (Gordon et al., 2008), ovari an (Horvath et al., 2008), and/or melanoma (Williams and Pembroke, 1989; Pickel et al., 2004) cancers. No mechanism fo r this scent detection

PAGE 21

Canine 15 of cancer has been determ ined, but if canines were to have formyl peptide-like receptors, then such receptors could be an important factor for cancer and other disease detection. This hypothesis is worth mentioning, but according to Liberles et al. (2009) is dubious as dogs are unlikely to have FPR receptors in their vomeronasal organ. In fact, dogs do not even have the non-olfactory FPR1 and FPRL2 receptors (Haitina et al., 2009) though they do have an FPRlike 1 receptor that could take the place of th e two missing receptor type s (Haitina et al., 2009). However, as we have learned from previous experiments, often receptors are missed in an organism if scientists do not know where to look for them (Lib reles and Buck 2006). The rat OR gene repertoire is larger than the dog repertoire, but canines have a higher level of diversification and the rat genome is less polymorphic (Roquier and Giorgi, 2007). The dog olfactory genome has been sequenced many times (Olvender et al., 2004; Quignon et al., 2003 and 2005; Robin et al., 2009) but the function of many GPCR gene sequen ces have not been determined. Perhaps dogs express FPR-like receptors that ha ve not yet been characterized. 2.7 Conclusion The olfactory system is a complicated and in tricate system which allows animals to sense thousands of different olfactory and pheromonal signals. Recent discoveries have illuminated some of the processes for signal tran sduction; however, more work in almost every area of the pathway should be conducted to better understand the system and its effects and influences on animals. One receptor in particular, the formyl peptide recepto r might be instrumental in mammalian detection of disease and sickness in consp ecifics, in the detection of food sources and identification of the foods physical condition, (contamin ated or healthy), a nd in the sensing of nearby predators. Munger (2009) stated it best,

PAGE 22

Canine 16 As the biological roles of individual olfactory subsystem s are elucidated, we begin to truly understand how animals detect a nd dissect their complex chemosensory worlds. The nose is a busy and interesting su bject for researchers these days. 3. Olfactory detection of cancer ba sed on MHC dependent odor components Though there is no accepted theory for how dogs are able to detect cancer in humans by sniffing biological samples, Balsei ro and Correia (2006) have pres ented an interesting hypothesis. They suggest that dogs can smell volatile organi c compounds produced by tumors due to changes in the human leukocyte antigen (HLA) molecules in the maligna nt cells. The purpose of this section is to describe in more detail what the Balseiro and Correia (2006) hypothesis suggests and to delve deeper into the subjects upon which the hypothesis is contin gent. Some topics that will be addressed include; the mechanisms employed by the human immune system to detect and destroy infectious and foreign agents, the role the MH C plays in the immune systemspecifically to prevent malignant cell prolifer ation, theories on how the MHC may affect body odor, simple cancer biology, and mechanisms cancerous cell s employ to escape detection by the immune systemspecifically mechanisms for altering MHC mo lecules in the infected cells. If indeed the hypothesis by Balseiro and Correia is correct, then possibilities fo r earlier diagnosis and treatment of cancer would be possible. 3.1 The Immune System The immune system functions to protect the body from pathogens. Th ere are two types of mechanisms that the immune syst em employs to protect its cells. Innate immune responses are immediate and the first line of defense for cel ls (Karp 2008). One innate immune response cell important in combating cancerous tumour cells, along with serving other functions, is a type of

PAGE 23

Canine 17 nonspecific lym phocyte called a natural killer (NK) cell. These cells have been known to cause the death of certain t ypes of cancer cells in vitro and may provide the mechanism for destroying a malignant cell in vivo before it develops into a tumor (H anna and Burton, 1981). Along with other cells, NK cells help initiate the adap tive response (Bottino et al., 2005). Adaptive immune responses take a certain amount of time to gear up for an attack against a pathogen, or other abnormal cell. After this lag time is over how ever, the adaptive respons e is very strong and selective (Karp, 2008). Two types of adaptive immune responses are humoral responses and cellmediated responses. Humoral responses focus on pathogens found outside of individual cells (Karp, 2008), and are not as importa nt for the purposes of this paper. However, the cell mediated immune response is very important for tumour prev ention and will be the focus of the rest of this section. 3.2 T Cells The major players in cell medi ated responses are T lymphocyt es. T cells recognize and kill foreign or infected cells (Eggert et al., 1999). Abnormal cells are detected by receptors present on the surface of T lymphocytes. The T-cell receptors (TCRs) bind to fragments of antigens that are held on the surface of antigen-presenting cells (APCs) that are abnormal. Each T cell expresses only one type of T-cell receptor which in turn binds selectively to only one kind of antigen fragment (Karp, 2008). When an APC is detected by a T receptor, the immune system is alerted to the infection. After it is alerted, the T cell may reproduce, called proliferat ion, and these new cells will differentiate into mature killing cells with the same T-cell receptors as the original T cell (Eggert et al., 1999). Since the receptor remains the same, the new T cells will only be able attack and kill the cells that contain the antigen recognized by that firs t T cell. This is a necessary

PAGE 24

Canine 18 precau tion to prevent newly synthesized T cells fr om promoting death in normal cells. There are three subclasses of T cells; cytotoxic T ly mphocytes (CTLs), helper T lymphocytes (TH cells), and regulatory T lymphocytes (TReg cells) (Karp, 2008). T hough all of these subc lasses are important, the cytotoxic T lymphocytes are the T cells that will be referred to in the rest of this paper unless indicated otherwise. CTLs kill abnormal cells by forcing them to undergo apoptosis. These CTLs possess a surface protein called CD8+. When a T cell receptor encounters an antigen presenting cell, the TCR docks on the MHC that is presenti ng the abnormal peptide fragment (Chang et al., 2004). The interaction between the MHC and T CR can be strengthened by additional contact between CD8+ cells and MHC prot eins. CD8+ cells are instrument al in the docking process and when they are inhibited or depleted by certain antibodies, some vi ruses and pathogens can spread with ease (Jin and Bauer, 1999). As will be discussed later, when mutations occur that impair the function of the immune system, cells of the body can mutate and become cancerous. 3.3 The Major Histocompatibility Complex (MHC) The major histocompatibility complex (MHC) is most known for its ability to detect self from non-self (Eggert et al., 1999). It consists of approximately 20 different genes containing over 2000 different alleles (Karp, 2008). Due to its highly polymorphic nature it is unlikely that any two organisms would have the same sequence of MHC a lleles (Eggert et al., 1999), except for identical twins. As was mentioned above, the MHC is resp onsible for presenting antigens on the surface of the cell. MHC molecules are not extremely select ive, and they can bind to different kinds of peptides that share certain features. As there are many different MHC molecules on the surface of just one antigen presenting cell, a vast array of peptides can be displayed. However, cells cannot display all possible peptides, because there are just too many variations in peptide structure and

PAGE 25

Canine 19 sequence (Karp, 2008). As it is difficult for the immune system to combat a pathogen that it cannot detect, an organisms susceptibility to a pa thogen depends on how well its MHC molecules can present the pathogens peptide fragments on the su rface of a cell. For this reason, some people can be more vulnerable to a virus, or antigen, if they are unable to adequately display fragments of that antigen on their cell surfaces (K arp, 2008). The MHC alleles presen t in a cell are determined by sexual selection (Potts and Wake land, 1993). Research indicates that a person is attracted to an individual who has a large number of different MHC molecules from the person, because then that persons own offspring will likely have a greater variety of MHC molecules (Eggert et al., 1999), and due to that variety, would be less susceptible to a larger number of diseases. One interesting fact relating to this is that a person would be less susceptible to a disease that their ancestors had been exposed to, because most likely their ancestors would have passed on their MHC genes that were sufficient at fighting those diseases (Karp, 2008). 3.4 The major histocompatibility complex is gene rally divided into two classes (Wedekind and Penn, 2000). Major histocompatability complex class 1 (MHC1) molecules are expressed by almost every cell in the body, while MHC class 2 (MHCII) molecules are generally only expressed on professional APCs, (such as dendritic cel ls, B cells, and macr ophages) (Karp, 2008). Professional APCs generally i ngest extracellular molecules through phagocytosis and express fragments of those molecules on their surface. The T cells that read these antigens are helper T cells. However, MHC class 1 molecules present pep tides synthesized inside the cell. The cytotoxic T cells that were discussed earlier read thes e peptides (Wedekind and Penn, 2000). When T cell receptors recognize foreign peptides expressed by the MHC1, the immune system reacts to cause the death of cells expressing those foreign peptides. Mutations or viruses in the cell both could be

PAGE 26

Canine 20 responsible for the synthesis of unknown and forei gn proteins (Karp, 2008). W hen a cell is normal, the MHC1 still presents peptide fragments of sy nthesized proteins on the surface of the cell. But no T cells will be able to bind with the normal peptides. Some T cells produced have T cell receptors that show a high affinity for normal cell peptides. These T cells are destroyed in the thymus before they have the oppor tunity to initiate an immune response against any healthy cells (Karp, 2008). Without the MHC and T lymphocytes the immune system would not be as able to detect problems occurring in the cells of our body. 3.5 Theories on the distinct odor of the MHC It has been fairly well esta blished through several studies th at there is a detectable odor humans, mice, and other animals exude due to their major histocompatibility complex (Penn & Potts 1998). These MHC associated odors can a ffect mating preferences of organisms in interesting ways. For example studies on house mi ce indicate that the sexes prefer mates with different MHCs from their own (Singer, 1997). A test done on humans showed similar results. Wedekind et al. (1995) did a study in which male students (N=44) were asked to wear the same shirt to sleep for two consecutive nights. When female students (N = 49) where asked to smell these shirts and rate the attractiv eness of the odors, a significant number of women chose the shirts worn by men with different MHC genotypes from th eir own. An interesting aside is that in a related study by Wedekind and Furi (1997), it wa s found that women on the contraceptive pill did not prefer odors of men with different MHC genotypes from their own. However, even though it has been shown that some animals can detect an odor associated with anothers MHC, it is still undetermined how MHC genes influence individual odor.

PAGE 27

Canine 21 Figure 3. Proposed Mechanisms for how MHC influences individual odor. This figure illustrates the five different hypotheses pr oposed for how MHC genes influence an individuals odor. Copied dire ctly from: Penn and Potts, 1998. Five prominent hypotheses have been suggest ed that would explain the mechanism for odor production associated with the MHC. (See Figure 3) The first mechanism, mentioned by Singh et al. (1987) is known as the MHC mol ecule hypothesis. This hypothesis assumes that

PAGE 28

Canine 22 there are fragm ents of volatile MHC molecules that can be found in the urine of individuals. Evidence for this hypothesis also comes from Singh et al. (1987), in which the authors found that a majority of class 1 molecules degrade to small fragments wh en introduced to urine and that rats were only able to differentiate between MH C class 1 types in the urine of different rats when the MHC molecules were degraded, but not when they were intact. However, according to Wedekind and Penn (2000), MHC molecules are large and nonvolatile, proof against the MHC molecule hypothesis, since mice can discriminate between urine samples from a distance (Singer et al., 1996), suggesting they are smelling volatile molecu les. In addition, Singer et al. (1993) conducted an experiment showing that mice can differentiate between urine, even after the proteins in the urine were denatured. Singh et al. (1987) presented another, more reasonable hypothe sis in the same article. The authors suggested that a mixture of MHC class 1 molecules in asso ciation with the peptides they are presenting on the cell surface could be transported from the blood into the urine and would impart an individual specific odor to the urine. They came up with this hypothesis after completing a study in which they tested animals for their ab ility to distinguish between urine samples from donors of one strain of MHC type from donors of a different strain. Specifically, rats could discriminate between PVG urine, et al.from a PVG.R1 urine type containing a polymorphic class I Aav1 type molecule (Singh et al., 1987). PVG urine is urine produced by genetically inbred mice with a PVG congenic series in which all non-MHC ge nes were nearly identical (Pearse-Pratt et al., 1999) containing a polymorphic class I Ac type molecule This hypot hesis was elaborated on by Singer et al. in 1997, and is now commonl y known as the peptide hypothesis.

PAGE 29

Canine 23 In the peptide hypothesis MHC m olecules bind to peptide fragments, and their volatile metabolites, such as carboxylic acids, are t hought to provide the odor (Wedikind & Penn, 2000). Singer et al. (1997) suggests that variations in the relative con centrations of individual compounds in acidic mixtures cause the difference in odor du e to MHC class 2 types. Singh et al. (1987) goes on to conclude that the amino acid conjugates of the odorous acids cont ain peptide functional groups that could be expressed by MHC glycoproteins and could be derived from the normally bound peptides. Singer et al. (1997) identified methylbutyric aci ds, dimethyl sulfone, phenol, pcresol, 4-epthylphenol, benzoic aci d, and phenylacetic acid as possible volatile acids. However, in their study only phenylacetic acid wa s statistically significant in signaling cla ss 2 type by differing relative concentrations in various mice. The third hypothesis is k nown as the microflora hypothesis. This hypothesis was presented as just a musing in which the author, Howard (1977) suggests that odor may be associated with MHC 2 indirectly. It has been reported that bact erial flora on the human skin is individual (stated by Howard, 1977 without citation). Therefore, individual odor could be produced through metabolites from b acterial flora on the skin, or in intestinal and genitourinary tracts. As the presence of these particular orga nisms might be under the control of MHC 2 linked IR genes (Howard, 1977) this could be a possibl e explanation of how od or is linked to MHC genes. More simply, in this hypothesis it is sugges ted that allele-specific populations of microbial flora are shaped by MHC genes and that this ma y alter odor produced by humans and animals. Though this hypothesis is interesti ng, there is no consistent ev idence for it (Wedikind and Penn, 2000). The fourth hypothesis is known as the carrier hypothesis. It was presented by Pearse-Pratt et al., in 1992. The authors suggest that MHC mo lecules change their sh ape to bind volatiles,

PAGE 30

Canine 24 instead of peptides, as in the peptide hypothesis, and carry the volatiles to scent or urine glands (reviewed by W edkind and Penn, 2000). A problem w ith this hypothesis is that MHC molecules bind to hydrophilic peptides, and no mechanism is suggested for how the MHC molecules could change their conformation in order to bi nd hydrophobic aromatic-binding molecules. The last hypothesis, the peptide-microflo ra hypothesis was suggested by the authors Wedikind and Penn (2000). They pr edict that MHC molecules in fluence odor by binding unique subsets of peptides (as in the pe ptide-hypothesis), which are then carri ed to preputial, coagulating, axillary or other microbe-harboring glands, wh ere their metabolites are made volatile by commensal microflora, similar to the microflora hypothesis. So far, this hypothesis is the most accepted because it not only includes a mechanis m for how individual odor is shaped by the peptide binding properties of the major histoc ompatibility complex, but it also provides a mechanism for how the molecules are made volatile and transported to the urine, breath, and sweat of individuals. Despite the fact that there is no one proven mechanism for how the major histocompatibility affects odor, there is proof that it does (Wedekind et al., 1995). Does cancer in a persons body change the MHC which in turn affects the odor produced by that person? In order to speculate on this questio n with any accuracy some general information about cancer should be addressed. 3.5 General Cancer Biology Cancer cells, unlike normal cells grow uncontrollably and indefinitely without stimulatory growth signals and despite the presence of inhibi tors (Karp, 2008). They do not necessarily grow any faster than normal cells, bu t their growth is not dependent on cell signals and they do not

PAGE 31

Canine 25 undergo apoptosis or other m echanisms for cell d eath, as a normal mutated cell would. Cancer cells are particularly deadly because they can metastasize and spread through lymphatic or vascular pathways to create secondary lethal tumors. Due to the prevalence and severity of cancer, studies on the disease have been continuous and wide spread for decades (Karp, 2008). Treatments such as radiation are available, but unfort unately no cure has been found thus far. Cancer can be caused by several agents, a ll of which alter the genome. Some causes of cancer include ionizing radiation, DNA and RNA-containing viruse s, and carcinogenic chemicals (Huebner and Todaro, 1969). Some even hypothesize that diet can affect a persons risk of cancer, because depending on where a person lives, they are more likely to get certain types of cancer but are at reduced risk for other types (Sinha et al ., 2003). For example people in the US are more at risk for colon and breast cancer, but less at risk for gastric cancer (Karp, 2 008). However, while geographical correlations with can cer are relatively easy to show, f ood consumption is not the only factor that could influence cancer risk. For example, Floridians had a greater risk of skin cancer in 2005 than inhabitants of Alaska (http://www.cdc.gov/ cancer/skin/statistics/st ate.htm), but that does not mean the increased risk was due to a high co nsumption of oranges or other foods eaten in Florida. More likely the risk was due to Floridians grea ter exposure to ultraviolet radiation. With more knowledge on the causes of cancer, more prev entative measures could be taken to avoid at least some types of cancer. Cancer is found in one third of individuals in Western countries and it is one of the two leading causes of death (Karp, 2008). Even though cancer is very prevalent when compared to other diseases a human can get, the chances of ge tting cancer on a cellular level are small. Several mutations or losses of gene function have to occu r before a tumor cell becomes malignant (Algarra et al., 2004). These mutations occur in a single line of cells in a multi-step process as first the

PAGE 32

Canine 26 cell(s) m ay lose self regulating controls that would induce apoptosis, and then lose their susceptibility to other ou tside controls and signals. Unfortunate ly, cancer is monoclonal and it only takes one correctly mutated cell for cancer to occur (Karp, 2008). After a cell becomes cancerous it generally still continues to mutate and undergo epigenetic changes. Th ese continuous mutations make cancer difficult to treat because the cancer cells that arise can be resi stant to the drugs being used previously. The cancer immune surveillance th eory (Albarra et al., 2004, Bu rnet, 2004et al.) suggests that there are intracellula r mechanisms the body usually uses to combat malignant cells. There are several tumor-suppressing genes expr essed that code for proteins generally released when cells start showing unusual growth. Ex amples of such genes includ e tumor protein 53 (TP53), ADP ribosylation factor (ARF), and retinoblastoma protein (RB) (Kar p, 2008). Mutations in these genes are often found in cancerous cells. As discussed in the immune system section, the bodys own immune system is generally able to kill cells which produce abnormal proteins due to mutations (Botino et al., 2005). However, as these proteins are still being produced by the bodys own cells, sometimes the tumor associated peptides can still look sufficiently like a normal peptide when presented on the cell surface (Karp, 2008). For th is reason, the immune system often fails to recognize the proteins as abnormal. In addition, even if the T cell receptors are able to recognize a protein as inappropriate, the tumors can often de velop mechanisms to escape destruction (Campoli and Ferrone, 2008). 3.6 Specific mechanisms tumor cells empl oy on MHC molecules to escape recognition One of the mechanisms most widely used by tumor cells to escape recognition is to alter the MHC class 1 expression on the cell surface (A lgarra et al., 2004). In fact, human leukocyte

PAGE 33

Canine 27 antigen (HLA) class 1 antigen changes have been discovered in 16-50 % of solid tumors found in head and neck squam ous cell carcinoma, and in br east, lung, colon, cervi x, prostate, and melanoma cancers (Chang et a., 2004). A variety of a bnormal expressions of human leukocyte antigen (HLA) molecules are present in tumor cells. Some of the different abnormal expressions include HLA down-regulation and complete HLA loss. Down regulation can be caused by abnormalities in the expression and/or function of components of the MHC 1 (Seliger et al., 2000), bu t not by structural mutations (Chang et a, 2004). MHC defects have been identified in several types of malignant lesions and such mutations could lead to impairment in the processing of tu mor-associated antigens and in the presentation of tumor associated antigens to T cells (Seliger et al., 2000). Interest ingly, according to Seliger et al. (2000) expression can sometimes be partially re stored by supplying exogenous peptides and incubating the down-regulated cells at a low, no n-physiological temperature which increases the stability of the antigen processing machinery. Complete HLA loss can be caused by Beta-2 microglobulin mutations (Chang et al., 2004). Selective HLA class 1 loss can be caused by muta tions or loss of genes encoding the lost HLA class 1 heavy chains or by mutatio ns that inhibit tran slation and/or transc ription (Chang et al., 2004).When complete HLA loss occurs in cancerous cells, often mechanisms to escape from natural killer cells are also pr esent (Bottino et al., 2005), such as the loss of the activatory NK receptor MHC class I polypeptide-related sequence A (MICA) and MHC class I polypeptiderelated sequ ence B (MICB) alleles (Algarra et al., 2004). This is because NK cells monitor the expression of MHC class 1 molecules. If a potential tumor cell does not express specific surface receptors that inhibit NK cells, by expressing re ceptors such as killer immunoglobulin (Ig)-like inhibitor receptors (Bottino et al ., 2004), the NK cell attacks and kill s the cell. Specific types of

PAGE 34

Canine 28 MHC m olecules that are often downr egulated or completely absent in tumor cells include; HLA-A, HLA-B, and HLA-C molecules (Balseiro and Correi a, 2005). Such HLA expression alterations in cancer cells could be what produce the different profile of odors th at are detected by canines. 3.7 A Possible ApplicationEa rly detection of cancer Early detection of many cancers, although essential for treatment effectiveness, can be both difficult to achieve and introduce new health risks (See next chapterMoser and McCulloch, 2010). If indeed canines are smelling alterations in MHC expression in order to detect cancer in humans, then an alternate detection method is po ssible that would not only have great sensitivity and specificity, but would also be non-invasive. This detection method, suggested by Balseiro and Correia, is to use an electronic nose to detect soluble or volatile organic MHC altered molecules in the breath, blood, or urine of a patient. The electr onic nose is a gas sensor array that can be used to detect MHC dependent odor types (Montag et al ., 2000). It has been used to distinguish urine odor types of MHC congenic mice strains, MHC mu tant mice strains, MHC class 1 mutant mice, and HLA-A2 transgenic mice. Interestingly, this ga s array was developed to be used in odor type research on the urine of subjects for behavioral studies on MHC dependant mate preferences. It would be quite astonishing if something developed to assist in behavioral studies was eventually used to detect such a deadly disease as cance r! See page 52 for more on the electronic nose. 3.8 Conclusion The detection of cancer based on MHC depe ndent odor components hypothesis put forth by Balseiro and Correia (2006) is an interesting and somewhat new way of approaching possible methods for earlier detection of cancer. More research needs to be done to determine the exact

PAGE 35

Canine 29 relationship between MHC molecule s in cancer cells, and odor produc tion, but the possibility is present for a new non-invasive method for the earl y detection of various human cancers. (See Page 50 for more on new devices inve nted to detect cancer.) 4. Systematic Review of Previous Canin e Scent Detection of Cancer Studies This section is a systematic review of all the previous canine scen t detection of human cancer studies published. In this section I will review the methods a nd accuracy of previous studies and give suggestions for future ones. Through this review I hope to as certain whether dogs can detect human cancers with enough sensitivity and speci ficity to be useful for diagnostic purposes. 4.1 Methods 4.1a Study Identification We conducted a systematic search in order to identify all known published data on canine scent detection of cancers, following th e roadmap suggested by Pai et al., (2004). We sought to find all published papers in which re searchers trained dogs to detect human cancers using only their sense of smell, employing biological samples such as skin, breath, urine, and excised tumor samples. Searches in PubMed (1949 to January 2009; www.pubmed.gov ), and EMBASE (1980 to January 2009; http://embase.com/ ), were completed using broad search terms such as canine scent and its synonyms, cancer and its synonyms, and the related articles link at the PubMed citation record of each eligible st udy. This search was designed to find all published trials in which dogs were trained to detect cancer in humans. We also manually searched for references from within the bibliographies of all eligible studies. 4.1b Study Eligibility

PAGE 36

Canine 30 We screened titles and abstracts and retained those that were described as a study testing a canines ability to detect hum an cancer. As so fe w studies have been published, this was the only eligibility requirement applied. We acquired full -text copies of all the retained studies, and screened those articles to insure that they met the above criteria. All ar ticles were included or excluded before extraction of data began. 4.1c Data Extraction Two reviewers (E.M. and M.M.) independently ex tracted data on the tumor type, target odor source, sample size, blinding of trainers to cancer odor location, storage details of the test samples, whether a data audit was completed, whether an i ndependent observer was pr esent, outcome of the testing, and statistical tests used in analysis, the source, bree d, and number of dogs, the canine training method used, the duration and frequency of training, and the type of controls used. (see tables 1, 2, and 3). 4.2 Results We found 531 potentially relevant studies on the subject of canin e detection of human cancers, and of these, excluded 520 and reta ined 11 (see figure 4). Full texts of the retained studies were examined. Five of these studies were excluded because they did not test canines detecting cancer, and one was excluded because it was a systematic re view and did not include original data. A sixth unpublished manuscript was added to the 5 remaining papers to bring the to tal to 6 papers for review. Willis et al.s study was on canines ability to be trained to detect bladder cancer by smelling a patients urine (Willis et al., 2004). In the study, liqui d urine was stored for up to 5 months at -40

PAGE 37

Canine 31 C and then used while still wet, or dried overnig ht and stored at room temperature for as many as 4 weeks. The authors used two methods to prepare scent samples for the dogs. Wet urine was pipetted onto filter paper and placed into Petri dishes allowing more surface area for vaporization of the volatile molecules. The remaining urine was dried in a beaker overnight. There were 27 target and 54 control samples av ailable for training of the dogs. Six dogs of varying breeds were trained 1-2 hours a day, 5 days a week, for 7 months. The operant cond itioning clicker method (Pryor, 1999) was used for training. During testing, the sa mples, 6 controls and 1 target, were prepared in a separate building and placed in ra ndom order unknown to the trainer, creating a fully blinded test. At least one control was age matched to within 8 years of the target and extensive care was taken to ensure that all of the cont rols were produced by volunteers with urological problems similar to that of the target patient. All target and control samples used for testing were unknown to the dogs. (This was the case in all studies reviewed.) Out of 9 runs the dogs indicated the correct sample and ignored the controls 41 % of the time (95% confidence interval: 23%52% estimated by bootstrap methods), compared with 14% expected by chance. Though these results are statistically significant, they are still well below that needed for use in clinical practice. Pickel et al.s (2004) study was on canines ab ility to detect and i ndicate melanoma by smelling lesions on living subjec ts. As the melanoma lesions being sniffed by dogs were insitu and had not been treated or surgica lly removed from the patients, no form of storage for the target samples was necessary. Two dogs, a golden retrie ver and a Standard Schnauzer, both highly trained AKC champions, were trained for testin g. One dog was trained over several weeks in about 200 area trials, during which he was asked to first retrieve a tube containing a mixture of basal, squamous, and melanoma tissue samples. Later the dog was required to search for the target tube in a grassy area containing many empty tubes. The second dog only participated in 11 such area

PAGE 38

Canine 32 trials. Two other types of tests we re com pleted, a box test (6 trials completed) and a test with tissue samples planted on healthy volunteers (around 73 tria ls completed), before the 7 patient searches for insitu lesions were conducted. In the patient sear ches completed, 8 to 30 adhesive bandages were placed in several locations on the patients body, including a bandage ove r the target lesion. Though the handler did not know where the target sample was locate d, the patient did; therefore, the experiment could not be considered double blind. A successful run occurred when the dog indicated the bandage covering the target mela noma. The two dogs were successful between 75% and 85.7% of the time. McCulloch et al.s (2006) study was on canines ability to detect and indicate lung and breast cancer by smelling samples of a patients exha led breath. Breath samples were collected on polypropylene fibers and stored for a maximum of 60 days at room temperature. There were 55 target lung samples, 31 target breast samples, and 83 control samples availabl e for use in training. Five dogs, three Labrador Retrievers and two Po rtuguese Water Dogs who were pets of owners living close to the research lo cation, were trained for 3 weeks, 20 times a day, every day. They were trained to indicate the target breath sample (containing breath from patients with lung or breast cancer) among 4 control samples also present. When the dog indicated the correct sample by sitting in front of it, the trainer used a clicker to indicate the dog was correct, and then reinforced the behavior by rewarding the dog wi th snack treats, praise, and pla y. During testing, controls were age matched to the targets. Confounding factors taken into consideration included smoking habits, diabetes, dental infection and recently eaten meal. During the double blind testing the dogs indicated the target br east cancer breath with a specifici ty of 98% (95% CI: 0.90, 0.99) and a sensitivity of 88% (95% CI: 0.75, 1.00). The same dogs indicated the target lung cancer breath with a specificity of 99% (95% CI: 0.96, 1.00) and a sensitivity of 99% (95% CI: 0.99, 1.00).

PAGE 39

Canine 33 Horvath et al.s (2008) study was on canines ability to detect a nd indicate ovarian cancer by sm elling ovarian tumor samples. The samples were stored for an unknown length of time at -80C. Out of 51 individual cancerous samples, 31 were allotted for use during training and 20 for use during testing. Only one dog, a Rise nschnauzer, was trained. The trai ning lasted for 12 months and was conducted in two parts: learning the odor signatures (6 mont hs) and learning odor discrimination (6 months). In the first part of training the dog was requested to sniff rags on the floor. When the dog showed interest in the targ et rag with the ovarian cancer sample, it was snatched away. This method supposedly reinfo rces the dogs natural hunting behavior and strengthens the dogs motivation to se lect the target (Horvath et al ., 2008). In the second stage, to learn discrimination, target samples were plac ed in two of 10 wooden boxes. The other boxes contained control samples of abdominal fat, mu scle, small bowel tissue, and pieces of healthy postmenopausal ovary. Control samples were not ag e matched nor symptom matched to the targets during testing. Single and double blind testing was completed with the dog having a sensitivity (measure of true positives indicated) of 100% and a specificity (measure of true negatives indicated) of 97.5%. The dogs accuracy was remarkable given that some of the control tissues had been removed from areas adjacent to the tumor within the same patients. Gordon et al.s (2008) study was on canines ability to detect breast and prostate cancer by smelling urine samples held in test tubes. The urine wa s stored for a maximum of 5 months at -20C and then stored in freezers at the trainers home. There were 53 target and 134 individual control samples allotted for use during traini ng. The 10 dogs of varying breeds were owned by trainers participating in the st udy and were trained at different locations. Six of the dogs were trained to detect breast cancer and four were trai ned to detect prostate cancer. The clicker training method summarized above was used with food treats as rewards for alerting to the target sample.

PAGE 40

Canine 34 The controls used were urine sam ples from subj ects who were not known to have cancer. Each dog was trained to discriminate only one cancer type from among the 6 controls. Blinding was introduced late in the training. Du ring testing, at least one contro l was age matched within 10 years of the target. In addition, there was substantial overlap of medica tions and medical history in both controls and patients with cancer. Testing wa s disappointing and overa ll performance was not statistically significant with success rates of 18% for detection of prostate cancer, and 22% for breast cancer. McCulloch et al.s study in progr ess is on canines ability to de tect ovarian cancer by smelling exhaled breath condensate. The breath condensate is being stored at -40C for a projected duration of one to two years. Five dogs, all pets of owners living near the research location with no prior scent training, are being tr ained using operant training with a clicker and food rewards, training one time per week. Controls will be age ma tched and patient and control histories will be recorded for future analysis. The testing will be completed under double blind conditions in which the dog will differentiate the target sample from four controls. However, the outcome is not yet available. 4.3 Discussion So far, five different teams of scientists have attempted to train dogs to detect and indicate lung, breast, melanoma, prostate, bladder, and/or ovarian cance r by smelling breath, urine, and tissue samples. Out of the nine studies complete d, five have met with statistically significant results and four of these have met with possibly clinically significant results. However, each of these studies could be improved to produce even better results. Training methods, sample storage method, target and control types used, and reporting methods could all be enhanced.

PAGE 41

Canine 35 Several different methods of training were us ed by the different teams of scientists. Clicker training, an operant conditioning method involving using a clicker to indicate success to the dog with food as a reward, was the most popular met hod. This is a typical method that many training facilities have used with consistent success. Pickel et al., (2004) and Horvath et al., (2008) used different methods, but still with reasonable su ccess. Surprisingly, though Gordon et al., (2008) used clicker training and food rewards, the dogs were not able to learn to correctly discriminate and indicate the target cancer samples. Inconsistency in training is the most obvious explanation for the dogs failure in this instance. A general training method was agreed upon by all the trainers with the specifics left to the individual trainers. The assumpti on was that they would know their dogs best and therefore be able to fine-tune the training to f it their dogs. Not only did the dogs have different trainers using diffe rent methods, but they also were trained at different times, and for different lengths of time. These inconsistencies in training may have made it nearly impossible for the dogs to perform well. One possible explanation for th e poor results in Gordon et al .s (2008) study is the present lack of existing accurate ways to predict pr ostate cancer aggressiveness, the so-called "tiger/pussycat" dilemma. Researchers at the UK's Institute of Cancer Research are working to develop such a test, based on the E2F3 gene (Foster et al., 2004), that may someday distinguish between aggressive and passive forms of prostate cancer. It is possible th at underlying variations in disease aggressiveness within the pool of patients us ed in canine scent de tection of prostate cancer may have confounded results. The studies reviewed used di fferent sample types, and stor ed the samples for different lengths of time under different conditions. Although dogs noses have been proven to be extremely sensitive, storage conditions such as temperature, storage container, and storage duration are very

PAGE 42

Canine 36 influential. It would be beneficial to find an optimal storing pro cedure, in order to preserve the sam ples. In the study by Willis et al., (2004), when the urine samples used were dried, the dogs did significantly worse (22% success rate) compared to when the urine sample was relatively fresh and dripped onto filter paper (50% success rate). This decline in performance by the dogs could be due to loss of volatile molecules during the drying pr ocess. This finding indicates that storage and presentation of the smell, for example dried inst ead of wet urine in the case of Willis, affects a dogs ability to detect the target odor. Freezing and then defrosting urine may also reduce the quality of the sample. Dogs have not yet been tested on urine samples that are fresh. In the reviewed studies, the dogs perfor med better overall when presented with breath samples or actual tissue samples, than when presented with ur ine samples. Perhaps there are more cancer biomarkers present at the source of the cancer, (in the melanomas and other tumors), and in the exhaled breath, than in the patients urine. This result may be of interest to scientists creating diagnostic equipment, influencing the sample t ypes that should be focused on to detect cancer biomarkers. Non age or symptom matched controls may have led to confounds in several studies. For example, the lack of properly age matched cont rols could have affected Willis et al.s (2004) results. In their experiment, one of the two samples provided by the two oldest patients was always the target sample used in a tes ting run. If during training the dogs learned to indicate the sample from the oldest subject, instead of the cancerous sa mple, they could still have completed successful testing runs more than 50% of the time without ever detecting the cancerous sample. Without more information it is difficult to tell if Horvath and Gordon also had this problem in their experimental setup, but it is likely. In future studi es, the majority of the controls should be closer in age to the target samples. It may be difficult in studies of malignancies common in older people, such as

PAGE 43

Canine 37 bladder cancer, to find a sufficient number of age m atched controls who have had other malignancies definitely ruled out, particularly low grade prostate cancer. In addition, future research should investigate base line odor signature variat ions in control individuals, to provide a more reliable comparison baseline against cancer cases. Although Willis et al., (2004) had trouble age matching controls they did a superior job of symptom matching the controls. All age matched c ontrols had some form of urological disease, and usually at least 2 of the ot her controls were from people w ith urological problems. In the studies completed by McCulloch et al., (2006) and Horvath et al., (2008), specificity may have been overestimated, as only h ealthy controls were used. Several statistical tests were used in the studies we examined, with differing levels of success. One error became apparent in the study by Gordon et al., (2008). In this study, replicate runs were completed during testing, but those r uns were treated as independent and were not analyzed with general estimating equations or a sim ilar statistical test to account for clustering of standard errors within donors. Other errors in analysis made by several authors included not reporting whether an independent ob server was present and/or a data audit completed. Without this information it is difficult to evaluate the qual ity of a study. In addition, calculating sensitivity and specificity is important when presenting results because they not only show the success rate, but also the degree of variabi lity, both of which are vita l for diagnostic purposes. One limitation of our study was the small number of papers reviewed. Five studies are not enough for a complete and comprehensive analysis of the topic and more research should be done in order to confirm that the prev ious studies are valid. The studie s already done should be repeated by other scientists to reduce any experimental bi as present and to see if better results can be attained. One experiment in particular that shoul d be repeated is the e xperiment by Gordon et al.,

PAGE 44

Canine 38 (2008) as several training inconsistencies m a y have caused the bad results.. Another study, not published, was on dogs ability to detect prostate cancer by smelling a patients urine. This study did not produce significant results and should also be repeated in order to see what errors may have occurred. 4.4 Conclusion Canine scent detection appears to be a valid method for cance r detection and may prove the principle that cancer can be detected by analysis of biological material su ch as urine or exhaled breath. Exhaled breath seems to be a better noninvasive biological sample than urine for biomarker analysis. Work done with malignant tissue samples was very accurate, but has no advantages over standard pathologic al examination of tumor tissue in that it still requires surgical removal. Work done with dogs sniffing human sk in was also accurate, but may perhaps suffer from patient acceptability, as many people do not trus t canine scent detection to be as reliable as other, more mechanical, methods. Future work sh ould focus on what biolog ical sample would be optimal and/or sufficient for diagnostic review In addition, future experiments should be considered with other cancer t ypes which would also benefit from improved early detection using earlier and less inva sive techniques.

PAGE 45

Canine 39 4.5 Table 1. Com parison of Scent Detection Studies Author/ Year Willis 2004 Pickel 2004 McCulloch 2006 Horvath 2008 Gordon 2008 McCulloch (in process) Tumor type Bladder Melanoma Breast Lung Ovarian Breast Prostate Ovarian N during testing: control/target 54/ 9 89/ 7 83/31 83/55 80/20 54/9 66/11 Planned: 60/60 # of target/ control samples present during a test run 1/6 1/7-29 1/4 1/4 2/8 1/6 1/6 1/4 Target odor source Dried and liquid urine pipetted onto filter paper in Petri dish Melanoma lesions covered by adhesive bandages on living subjects Breath, sampled on polypropylene fibers Breath, sampled on polypropylene fibers Tumor and control tissue samples Urine in test tubes Urine in test tubes Exhaled breath condensate and fiber tube Max storage time of test samples at highest degree used 5 months -40C and/or at room temperature for 4 weeks No storage (in situ, prior to surgery) 60 days at room temperature 60 days at room temperature Unknown duration at -80C 5 months -20C Then in freezer for undefined time 5 months -20C Then in freezer for undefined time One year at -40C Type of test Single blind (Only one person present) Single blind (Living subject not blind) Single and Double blind Single and Double blind (Fu lly blind) Single and Double blind Blinding unknown but present Blinding unknown but present Double blind Data audit Video Unknown Video Video By DVD Unknown Unknown Video Independent observer None Present Present Present Present Unknown Unknown Present Outcome: sensitivity/ specificity (or success rate) (41% success rate) (75-85.7% success rate) 88/98% 99/99% 100/97.5 % 22% sensitivity (17% success rate) Not yet available Statistical tests used Bootstrap techniques; t-test; rank sum test None noted Fischer twosided exact test and GEE Fischer twosided exact test and GEE Binomial probability distributeon Simple binomial or multinomial probability calculations Chi square Simple binomial or multinomial probability calculations Chi square Fischer twosided exact test and GEE

PAGE 46

Canine 40 4.6 Table 2 Canine and Training Related Information from Scent Detection Studies Target samples for training were melanoma ti ssue samples planted on healthy volunteer. Author/ Year Source of Dogs Breed of dogs # of dogs Training method Duration & Frequency of training Total # of target/control samples available to the dogs during training Willis 2004 Unknown Varyied 6 Operant clicker training 7 months 1-2 hrs/day 5X/wk 27/54 Pickel 2004 Highly trained AKC champions Golden Retriever Standard Schnauzer 2 unknown Several weeks 100-200 area trials On average 73/730* McCulloch 2006 (Lung) Pets Labrador Retrievers and Portuguese Water dogs 5 Clicker training w/ food rewards 3 weeks 20X/day Daily 55/83 McCulloch 2006 (Breast) Pets Labrador Retrievers and Portuguese Water dogs 5 Clicker training w/ food rewards 3 weeks 20X/day Daily 31/83 Horvath 2008 Unknown Riesenschnauzer 1 Unknown 12 months 31/unknown Gordon 2008 (Breast) Pets of trainers Varied 6 Clicker training w/ food rewards 12-14 months 15-30/day 2-7X/wk 53/134 Gordon 2008 (Prostate) Pets Varied 4 Clicker training w/ food rewards 12-14 months 15-30/day 2-7X/wk 46/120 McCulloch (in process) Pets Varied 5 Clicker training w/ food rewards 4 months 1040X/day1X/wk (study in process)

PAGE 47

Canine 41 4.7 Table 3 Comparisons of control samples used in scent detection studies 1Control samples from patients with diabetes, chronic cystitis, be nign prostatic hyperplasia, and healthy menstruation cycles were used. Author/ year (Tumor type) Control odor source Age matched control use Symptom matched control use Other confounding factors controlled for Willis 2004 (Bladder) Dried and liquid urine pipetted onto filter paper in Petri dish At least one control +/8 yrs Controls were matched to targets with similar urological symptoms Glucose levels, leucocytes and protein cystitis, and blood in urine1 Pickel 2004 (Skin Melanoma) Normal skin covered by adhesive bandages on living subjects Not Applicable Not Applicable None McCulloch 2006 (Breast) Breath, sampled on polypropylene fibers Yes2 None Smoking habits, diabetes, dental infection, and recent meal McCulloch 2006 (Lung) Breath, sampled on polypropylene fibers Yes2 None Smoking habits, diabetes, dental infection, and recent meal Horvath 2008 (Ovarian) Muscle, small bowel, and postmenopausal ovary samples None None None Gordon 2008 (Breast) Urine in test tubes At least one control age matched within +/10 yrs Medications and medical history were considered3 Food and drink ingested within prior 24 hours, and use of deodorants and perfumes Gordon 2008 (Prostate) Urine in test tubes At least one control age matched within +/10 yrs Medications and medical history were considered3 Food and drink ingested within prior 24 hours McCulloch (study in process) (Ovarian) Exhaled breath condensate and fiber tube None None Several4 2Mean age of controls and target samp le were statistically comparable. 3There was substantial overlap in the medical conditions and medi cations used in both the controls and patients with cancer. (Gordon et al., 2008) 4 Alcohol use, smoking, physical activity, socio-economic status and education, county of residence, age at menopause if applicable, age at menarche, family history of breast and ovarian cancers, weight and height, CA-125, level of physical activit y, body mass index, and concurrent chronic obstructive pulmonary disease, periodontal disease, rheumatoid arthritis, asthma, rhinitis, diabetes, renal disease, and card iovascular disease were controlled for.

PAGE 48

Canine 42 4.8 Figure 4. Study Eligibility Flowchart Potentially relevant papers identified and screened for retrieval (n= 531) Abstracts and titles excluded during first screeni ng (n=520) Relevant papers retrieved for more detailed evaluation (n=11) Papers excluded (n=6) Reasons for exclusion: Did not include canines detecting cancer (n=5) Systematic review not original data (n=1) Potentially appropriate papers to be included in the review (n=5) Papers added (n=1) Reason for addition Manuscript in process that met inclusion criteria Papers included for review (n=6), by tumor type Breast (n=2) Blad der (n=1) Melanoma (n=1) Lung (n=1) Ovarian (n=2) Prostate (n=1) Note: One paper included both lung and breast cancer and another paper included both breast and prostate cancer.

PAGE 49

Canine 43 5. Experiment 1 Canine Scent Detection of Ovarian Cancer After deciding on the thesis topic cani ne scent detection of human cancers I realized the importance of observing an actual study to fully understand the experiment al procedures on the topic. As an undergraduate at Ne w College of Florida I would not be able to conduct an adequate study myself due to the difficulty in obtaining organic products (such as urine or breath) from patients with cancer. I decided to look for an ongoing study to participate in somewhere in the United States. I emailed the first authors of signifi cant papers on the subject, and asked (1) if they were conducting a current study on th e topic, (2) if I could obser ve the project over January 2009, or (3) if they knew of, and coul d provide contact information for, anyone else conducting a similar study. I received one positive response. It was fr om Michael McCulloch, Research Director of Pine Street Foundation, in San An selmo, California, who had publis hed a paper in 2006 on canine scent detection of lung and breas t cancer. After several emails (in which I provided a CV and references), and a phone interview, I was invited to help in the 12 year ovarian epithelial cancer study from December 30th, 2008 to January 25th, 2009. The researchers were training about seven dogs to detect and indicate, by sitting next to, samples of breath given by patients with ovarian cancer from control breath samples. The ova rian study was being conducted similarly to McCulloch et al.s (2006) experi ment on canine scent detection of lung and breast cancer, with a few improvements and changes in the procedure. My first day at the research center I met al.l of the dogs (and people), walked the canine subjects in between training sessions, and mainly just observed the training procedure. Once I was orie nted I learned how to use a random number table to rotate the samples used during testing in a ra ndom order, was responsible for putting together the experimental setup in the mo rning, recording subjects behavior during trials, videotaping the

PAGE 50

Canine 44 trials for data auditing purposes, entering data taken during trials into an excel sp readsheet, observing breath sampling, and working the clicker in single blind trials. In addition, throughout the training and testing I made cr itical suggestions on how to im prove the current procedures. One of my suggestions which was implemented to improve the experiment was to take the subject completely out of the room while the sa mples were being rotated, instead of having the handler cover the dogs eyes, or bl ock the canines view of the samp les in some other manner. In this way the dogs only had one job while in the training room; to find th e cancer breath sampleand did not have to stay calm, or be distracted while the samples were being moved. With the permission of Dr. McCulloch the methods of the canine detection of human ovarian cancer study are described below. Note th e study has not been finished, so some of the later testing and data analysis has not yet been completed, and the results have not yet been compiled. 5.1 Sub-Hypothesis Dogs can be trained to detect early and late stage ovarian epithelial cancer biomarkers in a patients breath with high specificity and sensitivity by way of reward-b ased clicker training. 5.2 Methods 5.2a Patients and Control Breath Donors Eligible patients were: women 21 years of age or older with (1) hi stologically confirmed ovarian epithelial cancer (target group), (2) diag nosed with polycystic ovarian syndrome or endometriosis (unhealthy control group), or (3) were h ealthy volunteers with no prior ovarian or

PAGE 51

Canine 45 breast cancers and with no 1st or 2nd degree relatives with ovarian or breast cancer or known BRAC1 or BRAC2 mutations (healthy controls). The target group included newly diagnosed wome n with any stage of the disease, who had not yet received treatment for the cancer, so th at no treatments could interfere with the breath samples taken. All patients were required to abstain from the inta ke of Cox-2 inhibitors, vitamin E, omega-3 fatty acids, antioxidants, bromelain, co enzyme Q10, curcumin, vitamin A, or alcohol. In addition, the patients lived near the study sa mpling centers, read either English, Spanish, or Chinese, were non-smokers, and were w illing to provide breath samples. Patients were recruited to three medical cente rs in California includ ing California Pacific Medical Center, Pine Street Foundation, and UCSF Hellen Diller Family Comprehensive Cancer Center, and to one facility in Maine-the Univers ity of Maine. The patients and controls had to complete a questionnaire on their age, alcohol use, smoking history, ag e at menarche, body mass index, co-morbidities, physical act ivity, socioeconomic status, educat ion, country of residence, age of menopause (if applicable), most recently eaten meal, and medical history in case such factors were later determined to affect exhaled chemicals in the breath. Each participant provided written informed consent. At the beginning of the st udy an expected enrollment of 120 was projected. There was no compensation provided to subjects fo r providing breath sample s, or to dog owners for volunteering their dogs for use in the study. 5.2b Equipment and Breath Sampling Exhaled breath condensate (EBC) was collected in RTube EBC collectors from Respiratory Research Inc. ( www.RespiratoryResearch.com ). Rtubes are cylindrical polypropylene tubes 8.75 in. long, 0.9 in. outer diameter, and 0.8 in. inner diameter. An AADCO Pure Air Generator

PAGE 52

Canine 46 ( http://www.aadcoinst.com/Pureair.htm ) was utilized to provide zero air which patients breathed in during testing to ensure su rrounding air did not compromise the study. Breath was collected for one min on a DMSP absorbent strip ( http://www.sspinc.com/prodspecs/ssp-m100.cfm ) placed in the tube which was later sent to a chemistry la b to be analyzed. Three cotton cartridges were attached to the tube, 5 breaths each, and then froz en for later canine trials. Then, a frozen sleeve was fitted around the tube and caused the breath to condense in the tube as the patient breathed into the device for 30 minutes. After 30 minutes co tton swabs, q-tips, and cotton cartridges were wiped around the inside surface of the respiratory tube to remove the condensate. All samples were stored at 40 0C for a projected duration of one to two years. Breath samples for each person were collected during one visit. 5.2c Dogs The canine subjects trained in the study while the author (EM) was assisting included 2 Black Labradors (5 yr male and 2 yr female), 1 Ye llow female Labrador (3 yr), 1 male Lab-Golden mix (3 yr), and 1 male Miniature Poodle (2 yr). The canine subjects were provided by the Research Director, local dog owners, and by Guide Dogs for th e Blind (San Rafael, Ca lifornia). The subjects had at least basic obedience trai ning as outlined by the American Kennel Club, and were selected based on their level of eagerness to sniff objects and respond to commands. Between training sessions the dogs were kept in kennel crates of an appropriate size for each dog and monitored by a volunteer to ensure they were comfortable. Water was freely available, and the dogs were walked between sessi ons. Veterinarians were available if an animal was injured or ill, but th e subjects remained healthy throughout the study.

PAGE 53

Canine 47 5.3 Experimental Setup Training R oom. Training and testing occurred in a 3.0m x 7.3m room with vinyl tiling and overhead fluorescent lighting (Appendix 1.1). The room was not climate controlled and ambient temperatures varied year round. To prevent lingeri ng odors in the room, the floor was cleaned with Murphys Oil Soap and water, and swept regularly. Personnel. Certified (by Guide Dogs for the Blind-Sa n Rafael) dog handlers led the dogs, one at a time, into the room and encouraged the dogs to sniff the stations with the phrase, Go to Work. Generally two observers sat behind a curtain at the far corner of th e room in order to record the dogs responses manually and with a video recorder. Breath sample stations. Each station consisted of a polypr opylene plastic st orage container measuring 15 in. long, 12 in. wide, and 10 in. tall filled half-way with concrete. The lids of the containers had a 4.5 in. diameter hole allowing clear half-pint pol ypropylene containers measuring 1.5 in tall to rest inside. The ha lf-pint containers held cotton balls and Q-tips (see picture in appendix 1.2) or cotton cartridge s (appendix 1.3) containing a breath sample. Dogs were allowed to smell the breath samples, but were prevented from touching them by 7 qu arter in. holes drilled into the lids of the sample containers. The samp le container lids had numbers scratched into the top corresponding with the assigned patient numbers In this way no one c ould tell the contents (target or control) of the samp ling containers visually, without checking the numbers against a separate list. There were a total of five stations located in a st raight line 1 yard apart across the training room.

PAGE 54

Canine 48 Breath sa mple locations. A trial consisted of a dog walking past five stations, sniffing the samples (Appendix 1.4). One stati on contained the target cancer patient breath sample, and the other four contained control brea th samples. A random number tabl e was used by the experimenter to rotate the location of the target sample in a random manner to ensure the canine subjects could not predict the location of the can cer breath sample. There were 10 trials in a session, and each dog completed between 1 and 4 sessions one day per week. Classification of dogs response. Correct responses were (1) indi cating by sitting next to the sample station containing the target cancer sa mple (hit) and (2) smelling but not indicating a control sample (false negative). Incorrect responses were (1) smelling but not indicating the target sample (miss), (2) indicating a control sample (fal se positive), or (3) hesitating or performing an incomplete response to target or cont rol samples (incomplete indication). 5.4 Training Training consisted of three main phases (Appendix 2). After 30 consecutively correct trials dogs were advanced to the next phase. Phase 1. During the first training phase both the experimenter and th e handler knew the location of the target sample. The target sample consisted of cotton swabs, q-tips, and/or later cotton cartridges with the breath of a patient with ovarian cancer and a tiny dog treat (to interest the dog in the contents of the sample container). The other four stations contained clean cotton swabs and q-tips that contained no breath. At this phase the dog was usually led into the training room and allowed as much time as needed to sniff the sa mples. Once the dogs reached the correct sample

PAGE 55

Canine 49 they were told to sit by the handler. As soon as the canine sat at the correct sam ple the handler activated the clicker, rewarded the dog w ith a treat, and praised the dog verbally. Phase 2. Phase 2 was the same as phase 1 with the ex ception of the treat being removed from the target breath samples and those samples bei ng replaced with new target breath samples not contaminated by treats. The control samples remain ed blanks, the dogs were still commanded to sit by the handler, and the handler continued to know the location of the target sample. Phase 3. During phase 3, the control samples remained blank, and the dog was given no indication of the target sample by the handler. There was no command to sit, and in fact, the handler was blind to the location of the target sample. For this reason, the experimenter behind the curtain in the corner of the room would push th e clicker to signal success to the dog. After the clicker sounded, the handler rewarded the dog with treats and praise befo re leading the subject from the room. 5.5 Testing Single-blind trials. The single blind trials were similar to phase 3 of training, except control breath samples from either healthy volunteers, or from unhealthy patients (diagnosed with either polycystic ovarian syndrome or endometriosis) were used in place of blank samples. This greatly increased the difficulty level for the dogs, as they now could only differentiate between the target and control samples by detecting cancer-associated smells in the target samples. Zero trials. In order to assure that the subjects w ould not guess if they could not identify a cancerous sample, zero trials were conducted in which all five sta tions contained control breath samples. In a successful trial, th e dog would enter the room, go to work, sniff all the samples,

PAGE 56

Canine 50 and then leave the room without sitting. As th e handler did not know when a zero trial was occurring, the subjects were rewarded with praise (no food) after they le ft the training room. Double-blind trials. The control and target samples used were from novel patients so the subjects could not remember the correct sample from previous trials In addition, the experimenter (observer behind the curtain) did not know if a sample was from a patient with ovarian cancer or from a control patient. As the status of th e samples was unknown, the canine subjects were not rewarded for indicating a correct sample, but were instead praised after l eaving the testing room (as in the zero trials). 5.6 Data Management and Analysis After the trials were completed, data take n by the observer was entered onto an excel spreadsheet for future data analysis. See Appendi x 3 for an example of the information recorded. Each trial was videotaped, and the entire data set of dog performance will be audited for accuracy by comparing the paper data to videotape records. Only data from the double-blind testing will be analyzed to determine specificity and sensitivity. Confidence intervals for these resu lts will be estimated using general estimating equations (GEE) random effects linear regressio n. Each dog had the opportunity to sniff and potentially indicate any of five breath samples per trial. Each st ation is considered a unit of analysis and the use of four c ontrols along with one target samp le in a trial will not affect sensitivity or specificity. The 4 controls and 1 target breath sample re mained the same during a session (10 trials). Only the location of the target sample change d. As the dogs were not rewarded during the double blind testing it is possible the subjects behavi or and success rate might change during a session.

PAGE 57

Canine 51 Thus sensitivity and sp ecificity will be calculated for two points, (1) for a ll the double-blind trials combined, and (2) for only the first trial of each session (for the first time the canine subject is presented with the novel breath samples). In addition, Fisher 2-sided exact tests will be used to determine differences between patients and co ntrols (as recorded on the questionnaire). 5.7 Analysis of DMSP Absorbent Chemistry Strips. The DMSP absorbent chemistry strips were frozen until they could be sent to Dr. Touradj Solouki at the University of Ma ine. Dr. Solouki and his team c onstructed a Preconcentrator/Gas Chromatography/Fourier Transf orm Ion Cyclotron Resonance Mass Spectrometer (PC/GC/FTICR MS) which is used to determine biomarkers present in the breath of patients with ovarian epithelial cancer. The machine has an incredibly high resolu tion and high accuracy rate for determining the mass and abundance of charge d molecules in substances. Because this spectrometer is instrumental in detecting cancer biomarkers in the breath, the machine and how it works will be described in appendix 4. 6. Two Up and Coming Can cer Detection Methods Gold Nanoparticle Sensor. After the biomarkers for ovarian can cer are identified the next step will be to create a sensor (for use in hospitals) th at can detect cancer biomarkers in the breath of patients. Surprisingly, such a device has already been invent ed by Peng et al., (2009). They developed a hand-held sensor consisting of an array of chemoresistors based on functional gold nanoparticles in combination with pattern recogniti on methods that can differentiate healthy breath volatile organic compounds (VOCs) from the VOCs in breath of lung cancer patients. The breath tested does not have to be preconcentrated or dehumidified, a process that involves the use of

PAGE 58

Canine 52 expensive m achinery. Peng et al., (2009) determ ined 33 lung cancer biomarkers using GC-MS in combination with solid-phase microextraction and selected four key biomarkers that the nanoparticles would react most to. Thus far, the handheld sensor seems to be 86% accurate with 90% reproducibility, when tested with simulated breath samples. This sensor has great potential, yet more research needs to be conducted before the device is judged satisfactory. For instance, control breath samples were given on ly by healthy volunteers, even though unhealthy patients should also be tested to ensure th e devices specificity. In addition, the sensor has primarily been te sted on only late 3 to 4 stage can cer patients, instead of on early stage cancer patients. This testing is imperative if the apparatus is going to be touted as a noninvasive early lung cancer detection device. In addition, clinical trials ne ed to be completed instead of relying on prepared simulated breath patterns in calculating specificity and sensitivity of the devise. However, this device and others like it, mark the beginning of a new era of cancer detection using br eath analysis. Electronic nose sensor. As mentioned in a previous chapter, an electronic nose is an artificial olfaction system that is an array of non-sele ctive solid-state sensors that responds to the concentration of a combination of chemicals contained in a sa mple. Electronic nose technologies have been tested several times to determine if one can be cons tructed to detect lung cancer (Machado et al.; 2005; Mazzone et al., 2007; Peng et al., 2009; and DAmico et al., 2009). DAmico et al., (2009) performed an interesting study in which they determined the electronic noses sensitivity and specificity when tested against control patients with lung diseases like Chronic Obstructive Pulmonary Disease (COPD) Bronchitis, Pleurisy, and Interstitial lung disease. This study was well done, and importantly all the vital characteris tics of the patients and

PAGE 59

Canine 53 controls were recorded and displayed in the pub lication including age, sm oking history, diagnosis, and treatments. However, the study would have been stronger if the breat h of more early stage lung cancer patients had been used more instead of late stage patients. The gas sensor array had a global 79% of patients (cont rol and target) correctly classified and a sensitivity in respect to lung cancer of 89.3%. Though a sensitiv ity of 89.3% is significant, canines have had a much higher success rate (sensitivity and specifici ty 99%) (McCulloch et al., 2006). 7. Experiment 2 The peanut project: a po ssible procedure for training peanut-sniffing dogs Canines can detect peanuts conc ealed in other foods and alert an owner to their presence. Evidence for this is mostly anecdotal, ( http://abclocal.go.com/wls/sto ry? section=resources&id=6766540 9:12 4/27/09), but also seems reasonable, as canines have an incredibly accu rate sense of smell (Johnston, 1998). The purpose of our study was to determine if one method of inst rumental conditioning, involving clicker training, could work to produce a highly trai ned canine that would be able to act as a peanut sniffing dog in a controlled laboratory setting. Th e training procedure was adapted from the above study in which dogs were trained to detect ovarian cancer in a human by smelling that persons breath. A previous experiment (McCulloch et al., 2006) utilizing the same training procedure was remarkably successful with the dogs detecting the target sa mple containing lung cancer with a sensitivity of 0.99 (95% confidence interval, 0.99, 1.00) and specificity of 0.99 (95% confidence interval, 0.96, 1.00). 7.1 Sub-Hypothesis

PAGE 60

Canine 54 Given that the training procedure used in the cancer study (McCulloch et al., 2006) was successful, and as both the scent detection of ca ncer study, and the peanut sniffing study both had sim ilar goalsto teach a dog the task of discriminating a smell (cancer or peanut scent) from another substance (breath or f ood, respectively)we hypothesized that the same procedure used to condition canines to detect cancer could be used successfully to conditio n a canine to detect peanuts. 7.2 Why Peanut dogs are important Humans can have allergies to several different food including but not limited to foods like milk, egg peanut tree nut (walnut, cashew, etc.), fish shellfish soy, and wheat, (Cummings et al., 2010; Kandil and Davis, 2009). Som e 3.3 million Am ericans suffer from peanut or tree nut allergies. Allergy to pea nuts is one of the most deadly examples of a food allergy. The prevalence of this allergy has doubled in children in a five-year period and only around 20% of children outgrow peanut allergies. Public education on food allergens, specifically on how to read food labels, and on additional preventative measures for reducing the accidental ingestion of a known allergen is needed for todays society. Howeve r, even a well educated and diligent person can accidentally eat a food they are severely allergic to. An additional method studied in this paper for reducing accidental i ngestion of an allergen comes in an unlikely form, a service dog. Dogs trained to detect and indicate the presence of peanuts is not a novel idea, as th ere is already a private center in Texas training dogs for this purpose. It is not surprising that dog s are able to detect peanuts, even in low quantities of food as their se nse of smell is very accurate. They can smell in parts per trillion (Walker et al., 2003), which is like being able to smell a few drops of fragrance added to enough water to fill an Olympic sized swimming pool. However, there is no peer reviewed literature on the topic of the use of can ines for scent detection of peanuts used as

PAGE 61

Canine 55 working dogs, and there is very little infor mation on the appropriate way to train a dog to detect and indicate the presence of peanuts. With a trained peanut sniffing dog by his/her side, a child with severe peanut allergies might feel safer, and be at less risk of exposure to an allergenic material. 7.3 Methods 7.3a Subject. One canine, a male golden retriever, 2 year s old was trained to detect peanuts. The golden retriever is a pet of a professor on cam pus and was housed with the professor throughout the experiment. The dog was chosen for this stud y because he lives near the research site, is friendly, playful, and comfortable around humans. 7.3b Experimental Setup. Training and testing was done in the kitchen of Bon house on the New College of Florida campus. The kitchen is a narrow, clean room, with linoleum floors. Training was conducted 2-3 times a week on Mondays, Wedne sdays and some Fridays from 4:30 to 6:00 p.m. The experiment started on Febr uary 23, 2009 and ended on April 15, 2009. 7.3c Personnel Two people were present in the room during training. One handler would lead the dog into the room and would encourage the d og to find the sample containing peanuts. The other person, an observer, or recorder, move d the target sample betw een trials and observed trials behind an opaque curtain so as not to cue the dog to the position of the target sample. That observer would also record Sam s responses on a data table.

PAGE 62

Canine 56 7.3d Sample stations There were 5 sam ple stations located on the floor of the experiment room in a straight line spaced two feet ap art. Each station consisted of a polypropylene plastic storage container (64 fl oz, Publix brand) with a circ le cut into the top in which a 4 oz mini round polypropylene food container (Glad Ware brand) was placed. The 4 oz food containers held the control and target samples. There were 5 holes cut into the lids of these containers to allow the dog to smell the samples without tasting or touc hing them. Four of the containers contained uncontaminated food items such as corn, water, soynut butter, etc. The target container held peanuts or byproducts of peanuts such as raw pe anuts, and peanut butter mixed with other foods such as granola bars, bread and jelly, sunflower s eeds, etc. The location of the target during each trial was pre-determined using a random number ta ble so that the dog was not able to predict the locations of the sample by learning an unintended pattern. 7.3e Classification of dogs response. A test trial started when Sam entered the room and sniffed at each of the five stations. After indicating the location of the target sample, he was led out of the room and the target sample was moved to a new station. A correct response was (1) the dog sitting or lying down directly in front of the sample station where the sample containing peanuts was located (hit) and (2) sniffing a nd then walking by the control samples without indicating them (correct). Incorrect responses were (1) indicating on a control sample (false alarm), or (2) sniffing but not indicating a target sample (miss). 7.4 Pre-training Before training for the experiment began some pre-training was conducted. Sam already was acceptable at sitting when reque sted, but a more precise techni que of asking for the behavior was conditioned. The procedure for requesting Sam to sit included; the handler looking directly at

PAGE 63

Canine 57 Sam putting her hand in front of her as if signa ling stop, and then commanding to sit. During this period, whenever Sam sat on request he was rewarded with praise. Sam was also conditioned to enter the traini ng room and sniff each food sample presented on the floor. He learned this beha vior within the first two traini ng sessions without much trouble. During these sessions the handler kept Sam on a leash and led him to each station. The handler would tap on each food container until Sam showed interest. The control food containers were empty, but the target container he ld peanuts and treats. This tast y combination often led Sam to pay more attention to the target food sample. Ever y time he sniffed the target the handler would act excited, praise Sam, and ask him to sit. Once he sa t, he was praised more and awarded a treat. Sam reproduced the correct behavior after the first two sessions. 7.5 Training The training method employed instrumental conditioning methods with dog treats and praise as reinforcement. Training progressed in 3 phases, as adapted from McCulloch et al.s (2006) work. There were 5-10 trial runs in a session and 1-2 sessions each day of training. Phase 2 was considered complete when Sam had correctly discriminated the target from four control samples for 10 consecutive trials. There was no set protocol for determining when the other two phases were completed. Between each session Sam was taken outside fo r a walk. Generally the walk lasted around 10 minutes. He would often play and run around with the experi menters. During inter-trial intervals Sam would sit in the hallway outside of the training room with the handler. He was often petted and praised during this ti me, though he would often take time to get a drink of water and lay

PAGE 64

Canine 58 down on the floor as well. Sam always had access to water during the inter-trial intervals, and before the start of training. During the first phase of training, the control containers were empty and the target food container held three non-salted peanuts. The loca tion of the target sample was known to both the trainer and the observer. At the st art of a trial Sam was led into the training room and was told to go to work by the trainer. To encourage the do g to seek out the target sample a dog treat was placed in the target container. Once Sam sniffed the correct sample the clicker sounded. At this point the handler would ask Sam to sit, and then praise and rein force Sams behavior with a food treat. The trial was considered completed afte r the clicker had sounded, and the dog praised and led from the room. The target sample was moved duri ng the inter-trial interval as specified above. In the second phase of training the target cont ainer only held peanuts as opposed to peanuts and dog treats. During the first part of this phase the controls remained empty, but after Sam was performing successfully, corn wa s added to the controls. In addition, the handler did not purposefully give Sam any signals in dicating the correct sample. If he sat next to a control sample the handler ignored the sit and re peated go to work. The e xperimenter and handler knew the location of the targ ets in this phase. During the third phase of training only the experimenter knew where the target sample was located. The target samples varied in the amount and form the peanuts were in. For example sometimes only peanut butter was present and othe r times only peanut du st. During later training soynut butter was added to the co rn in the control samples. 7.6 Testing

PAGE 65

Canine 59 During the double blind testing the handler did not know where the target sam ple was located, and the recorder hid completely behind a sheet in order to prevent any involuntary body language that could cue Sam to the location of the target. Once Sam i ndicated a sample, the handler told the recorder out loud where Sam had sat, and the recorder clicked, or not, as appropriate. Each control container had novel food inside, and each target sample included novel food mixed with peanuts. During the first three days of testing Sam was not given any warm up training before novel foods were introduced. However, in the last th ree days of training, befo re the start of testing Sam was given 5 trial runs to warm up. These warm up runs were conducted in the same way as phase 2 of training, except the am ount of peanuts in the target we re reduced successively each trial. 7.7 Instrumental Conditioning (Training) Clicker Training As in the study by McCulloch (2006), clicker training (Pryor, 1984) was employed during training. The sound that a clicker produces acts as secondary reinforcer that has the capacity to reinforce behavi or through its association with the primary reinforcer (Williams, 1994), which in this case was food and praise. As soon as a dog associates a click with the reinforcement the animal tends to increase the frequency of the behavior it was doing when the click occurred (as predicted by Thor ndikes law of effect, 1911). Wit hout the clicker or some other bridge it would have been difficult to indicate to Sam exactly what he was doing correctly when he produced the desired behavior. Generalization and Discrimination During the first stages of training, a dog might not know exactly which behavior it is being rewarded for. In that case, it is likely the animal will generalize

PAGE 66

Canine 60 during dis crimination tasks like this one. However, as the dog becomes more familiar with the task it is more likely to learn to discriminate betwee n, in our case, food samples with peanuts and food sample without. Shaping Sometimes during training a shaping technique is used to condition an animal to execute more precise behaviors. At first, the animal is reinforced for a behavior close to the desired behavior, and then only reinforced for the desi red behavior. For example, at first Sam was rewarded for just sniffing the target sample, but la ter he only received the reward after sitting next to the target. If Sam sniffed the target and hesi tated, but did not sit, he was ignored until he returned to the target and sat completely. Context Effects In order to condition Sam at a fast rate he was conditione d in the same room and at the same time each week. However, if Sam was really going to be used as a sniffer dog, this kind of training would not be appropriate. That is because Sam would have to produce the conditioned behavior of indicating a sample containing peanut s in any context, not just when he was in a certain room where distractions were limited, and variables were minimized. However, as I said, for our purposes it was alright to stay in the same context, and was actuall y beneficial. Since we had such an exact routine Sam learned what to e xpect. Whenever he was taken out on a walk and then led into the hallway, he knew training was a bout to begin. In other words our presence and the walk before training primed Sam to remind him of past training experiences so he would be ready to perform when asked. These pre-training behavior s primed past conditioni ng experiences into his short term memory so that he was able to remember his conditioning with ease. Positive Reinforcement Sam was reinforced for indicating the correct sample with praise and treats. Both of these are primary reinforcers. One problem with praise is that it can be highly

PAGE 67

Canine 61 variable. If the handler is tired, then the dog m ight not be praised as highly for performing well, as he would be if the handler was in a better mood. In order to keep at least one reinforcer constant, food was awarded as well. Positive Contrast Effect Occasionally when Sam seemed to have lower motivation than usual an increased quantity of food was given to him dur ing conditioning in order to not only reinforce his good behavior, but to provide him with a greater source of motivation. The increased reward tended to have the desired effect and Sam performe d faster, and with more motivation than before. This effect was observed qualitatively as Sam seem ed more involved in his task, and less likely to get distracted by other objects or smells in the room after being rewarded more food on a previous trial. Aversive stimulus The only time a aversive stimulus wa s employed, was when Sam sat at the wrong station during a test. In these circumstances he was ignored for about 10 seconds, and then was told, no. If he continued to sit he was encouraged to k eep looking and the handler would point behind Sam as if to tell him to walk towards the other samples. Omission training During one testing session Sam was not performing well. He did not seem motivated to enter the ro om to start training, and once he was in side he did not smell the samples, but just looked at the handler. During this session Sam was put on timeout. He was led out of the room, without a reward and left in a room without reinforcemen t for about 10 minutes. When he returned he only seemed mildly more interested in training, but other variables could have been at play during that day of training that will be discussed later.

PAGE 68

Canine 62 Behavior chain (Chaining) A behavior chain can be observed in Sams responses during trials. Sitting next to the target elicited a click which in turn predicted tr eats and praise for Sam. Smelling food samples was reinforced by finding the target, while entering the room was reinforced by the opportunity to smell the samples. After Sam wa s reinforced with food, leaving the room was reinforced because then he had the opportunity to re-enter the room and start over. (See below diagram). Diagram 1 Sam: Entered the experiment room saw the food sample s sniffed the samples detected the target sample sat at the target heard the click received the rewards left the room started over. Each response in the chain produced a discrimi native stimulus that reinforced the previous behavior, and set the occasion for the next one. Spontaneous recovery At the beginning of training the handl er tried to condition Sam to visit every station at least once during a trial. Even after Sam sat next to the correct sample early on in a trial, he would be led to the other stations. However, Sam was never rewarded for this behavior except by a small amount of praise. After the initial training period, the tr ainers did nothing to encourage Sam to continue smelling stations after finding th e target food. Later, after Sam was off leash, and even during testing, he still would occasionally visit la ter stations after being rewarded at an earlier station. His behavior was interesting because after sporadic a nd minimal reinforcement during initial training, and then no reinforcement at all (extinction training), Sam was showing spontaneous recovery. (This is only one explanation for the behavior and would have to be tested in an alternate experiment to be proven.) 7.8 Results

PAGE 69

Canine 63 The results of the study have been record ed in Table 1 and Figure 1. There was no inform ation available on Sams success during the first two tr aining sessions, because the observers/recorders were still adjusting to the data sheets and the method for recording Sams behavior. Phase 1 of training was shortened du e to concern of overshadowing, but by that time Sam did seem to have learned the association between sit ting at the target, and receiving a reward. During phase 2, which lasted for four sessions, Sam progressed steadily in his performance until he successfully completed all ten trials perfectly at which time he was graduated to phase three. Sam did not seem at all confused when the trials were single blind and he performed at 100% for three successive sessions. Corn was added to the control samples, but did not seem to stymie him in the least, as he continually ignored the corn and discriminated the peanut samples. As he was showing a high level of conditioning the trials were made more difficult with the addition of a more salient food, soynut butter, in the control sample. (This food was assumed to be more savory to Sam than corn, however no studies were done to prove this assumption.) Sam performed perfectly the first trial and sat at the ta rget sample, but then he made a few mistakes and sat at the wrong stations. He ended up completing 87.5% of his trials correctly. After a 12 day break in training Sam performed at 70%, but then after another 5 da ys he only performed at 30%. After this dramatic drop in performance, his beha vior improved after the training routine was reestablished. In trial 16, water was added to the target and contro l samples to see how Sam would react to the new variable. He missed the target on the first trial, but then performed well. During the first day of testing Sam performed we ll. He missed the first three trials, but then correctly indicated the target stimulus the rest of the day. The second day of testing Sam performed sporadically, though still above chance at a 60% success rate. The day after that he performed even worse only responding correctly 40% of the trials. On this day every incorrect trial

PAGE 70

Canine 64 occurred when Sam sat at saltine crackers mixed with soynutt butter. From then on, Sam performed correctly about 70-80% of the time. Th e last two sessions, sessions 26 and 27, are not shown in the table because Sam was sick and unable to perform Table 1. This is the compiled data for the experiment. The session number is shown in the left column. The percent of trials that Sa m completed correctly is shown in the middle column. The last column indicates the traini ng method used, and the phase of training Sam was in. During phase 1 Sam was on a leash. During phase 2 Sam was off-leash and the controls remained empty. During phase three, the handler did not know the location of the target samples, and the control samples vari ed from containing not hing, to holding soynut butter, corn, and/or water. During testing al l of the samples were novel foods. Warm ups occurred before testing sessions and consiste d of successively smaller amounts of peanuts in the target food container. For more detail s see the section titled training procedure. Key for Table 1 Symbol Meaning T Target sample C Control Sample P Peanuts PB Peanut butter C Corn Table. 1 Testing and Training Data and Procedure Session number Percent Correct Training method Phase 1 N/A T=P&T C= E 1 2 N/A T=P&T C= E 1 3 50% T=P&T C= E 1 4 80% T= P&PB C= E 2 5 90% T= P&PB C= E 2 6 80% T= P&PB C= E 2 7 100% T=P C=E 2 8 100% T=P C=E single blind 3 9 100% T= P C=C single blind 3 10 100% T= P C=C single blind 3 11 87.5% T = PB C= soynut butter 3 12 70% T = PB C= soynut butter break 3 13 33% T = P C= soynut butter break 3 14 100% T= PB C=Soynut butter 3 15 100% T= PB, then P C=Soynut butter 3 16 80% T= P+ water C=Soynut butter+water 3 17 70% Testing 18 100% Testing (same samples as 17) 19 60% Testing 20 40% Testing 21 80% Testing 22 100% Warm up 23 80% Testing 24 100% warm up

PAGE 71

Canine 65 Percentage of Correct Responses 0% 20% 40% 60% 80% 100% 120% 135791113151719212325 Session numberPercent correct Figure 1 illustrates the results of the training and testing recorded in the first two columns of table 1. The blue line at 20% indicates chance level of responding. 7.9 Discussion Emotional variables There were some variables that consiste ntly seemed to affect Sams behavior during training and testing. The first variable was Sams mood. Occasionally, Sam would seem much less motivated then usual. This was witnes sed the last day of trai ning when Sam was sick, but also during sessions 12, 13, 19, and 20. It was pr obably not a coinciden ce that these sessions corresponded to the days that Sam performed wors t. Sessions 12 and 13 were after long breaks in training. Sam had a hard time concentrating on th e task and was often distracted. His actions seemed similar to students at New College who ha ve trouble staying awak e and paying attention during Monday classes. Sessions 12 a nd 13 were also a little more diffi cult than earlier trials due to the control and target samples used. Trial 19 wa s on a Monday and again Sam seemed distracted and unmotivated. Sams lowest performance ever was session 20, and again it was on a Monday. There were probably several factors that led to his incorrect res ponses that day. Firs t, a visitor that had been staying at Sams house with him for some time had left that morning. Sams owner said

PAGE 72

Canine 66 that Sam seemed depressed throughout the entire da y due to that event. During training that day there were also severe winds, which could have stirred up the air and spre ad distracting scents around. With all of these variables it is unsurprising that Sam did not perform at his best. It might be useful and interesting if in future experiments some type of quantitative measure of the subjects interest/motivation/mood could been deve loped to see if there truly is a correlation between the emotional state of th e subject and his performance. Overshadowing Interestingly, during session 20, Sam indicat ed the same incorrect control sample every time. He sat next to saltine crackers mi xed with soynut butter 4 times, 3 times in a row. During the other two incorrect tria ls Sam did not indicate any sa mple. It is strange that Sam indicated soynut butter, because that was a food us ed in earlier training sessions. It was expected that Sam would have learned that soynut butter was incorrect befo re that session. One explanation is that instead of Sam learning that soynut butter was the wrong stimulus dur ing those earlier trials, he had learned that peanut butter was the correc t stimulus. In session 20, the target was a peanut granola bar. The peanuts were ha rder to smell, and there was no peanut butter present. The only other trial where soynut butter and peanuts were presented together Sam performed correctly only 33% of the time. This indicates that there might have been an overshadowing effect due to the saliency of the target stimuli. Peanut butter sme lls much stronger than plain peanuts, and Sam may have learned much more about the peanut butters association with the reinforcers in these trials, then about the peanuts associa tion. The other, just as plausibl e explanation, would be that the saltine cracker and soynut butter sample had become contaminated with peanuts. If so, then he was sitting at a correct target unbeknownst to the trainers.

PAGE 73

Canine 67 Overshadowing and Generalization During phase 1 a m ixture of p eanuts and dog treats were used as the target stimuli. As dog trea ts are much more salient and familiar to Sam, then peanuts, he may have learned to go to the sample with the do g treat instead of learni ng to go to the sample with peanuts. Though in phase 2, after the treats were removed from the samples, Sam did well with just the peanuts, later in testing the association with food tr eats and reinforcers could have been an issue. This possible pr oblem was especially evident in session 25. Sam falsely indicated beef stew as the target three times. These thr ee trials were the only incorrect trials that Sam completed in that session. Beef stew may have smelled similar to the beef dog treats or dog food that is sometimes given to dogs. If Sam had ever been exposed to that type of food earlier on, before training, then he may have associated the beef stimulus with the dog treats used as a target stimulus in phase 1 of training. Due to phase 1 of training, Sam may have been inadvertently conditioned to react to any treat he had been given in past experience, as the target. In other words, Sam may have learned to generalize target stimuli to include any kinds of stimuli that were similar to dog treats he had prior exposure to. The same generalization problem could have happened in session 17, when Sam falsely indicated that carrots were the correct stim uli 2 out of the 3 trials that he missed that day. Sam had been given carrots as treats before training st arted and he may have indicated them as the target instead of th e peanut containing targ et for that reason. Reinforcement inconsistencies There were two other unfortunate problems with the training and testing procedure used. As mentioned earlier, Sa m was led on walks before training where he was petted and played with. Sam always seemed to enj oy this time, and was always eager to go outside. However, these walks were filled with lots of reinforcement that were not contingent on any particular behavior. Reinforcemen t during training was essentially very similar, only with the

PAGE 74

Canine 68 added reward of a food t reat. If Sam was more highly motivated by praise, than by food, then he might not have been as inclined to work as hard for a reward that he co uld get during other times besides training. In additi on, during inter-trial intervals, Sam w ould often get a lot of attention and petting while he was waiting to be allowed into the experiment room. Therefore, Sam could have been less motivated to work for praise, as he could get it anytime he wa s not being tested or trained. The other inconsistency during tr aining occurred due to a fluke. The treats used to reward Sam ran out halfway through training. The same tr eats were unavailable, so instead new treats were used. Though smaller, the trea ts were thought to be more savory and of higher quality, so it was expected that Sam would like the new treats ju st as much as the original ones. Unfortunately, the reinforcer substitutability wa s not tested ahead of time. If this second ki nd of treat had been considered by Sam to be less than the original trea t, then a problem of sans-negative contrast effect might have resulted. In this case Sam might have become frustrated with the smaller reward and may have performed with less enthus iasm and motivation than he had before the change in treats. 7.10 Conclusion If this experiment was to be repeated se veral changes in the procedure should be considered. First, a complete lis t of all treats given to the can ine subject should be compiled, and none of the stimuli should be similar to those treats. Second, the subject should be taken on walks before and after sessions, but an effort should be made to not overly praise the dog. Third, the quantity and quality of treats should be kept constant throughout the experiment, with only occasional jackpotting to increase motivation. Four th, no treats should be used as a target, even in the first phase of training. Lastly, it might be useful and interesting if some type of quantitative

PAGE 75

Canine 69 m easure of the dogs mood and/or should be devel oped to see if there tr uly is a correlation between the emotional state of th e subject and his performance. Unfortunately, the training procedure we us ed was probably not appropriate for training a peanut sniffing dog. The method worked well for McCu lloch et al., (2006) beca use there were very few variables in the control and target samples. Each sample contained only breath, from which the dog was to detect cancer biomarkers. In the p eanut study, Sam was expect ed to detect peanuts when there were several new foods present, some more salient then others. In addition, the peanuts in the targets varied widely in quantity and quality, (e.g. whole, crus hed, sugar coated, butter, etc.). These foods would probably be considerable more distracting to a dog, then breath samples were, and the canine in question could have prior experi ences with several of the control foods that would affect behavior. We we re attempting to condition Sam to behave in a much more complicated manner then originally thought. With these factors in mind it makes much more sense why a peanut sniffing dog would take several mo nths to train and cost over $20,000 dollars to buy ( http://www.peanutdog.com/index.html 5/8/09 6:14 pm). However, even with all of the difficulties in th is experiment Sam still consistently sat at the target sample significantly above chance. His performance indicates that the instrumental conditioning method used was at le ast partially successful. With mo re training, it is quite possible Sam could perform almost perfectly in the future.

PAGE 76

Canine 70 8. APPENDIX 1 Pictures from Canine Ov arian Detection Experiment Note the concrete holding the storage contain ers in place. 1.1 Testing and Training Room 1.2 Cotton and Q-tips in sampling container 1.3 Cotton Cartridge 1.4 Trial Run

PAGE 77

Canine 719. APPENDIX 2 Sequential phases of dog training and test ing. (Compiled by Michael McCulloch) Phase Location of cancer sample among five stations Contents of station with target stimuli Contents of other four stations Sequence of events at the station with cancer sample Location of cancer sample known by: Criterion for advancement Training 1. Sniffing 2. Command (sit / down) 3. Indication by dog 4. After dog sits, then: Clicker Food reward I Randomly chosen Cancer patient breath sample and food Blank tubes Praise Experimenter and handler 1. Sniffing 2. Command (sit / down) 3. Indication by dog 4. After dog sits, then: Clicker Food reward II Randomly chosen Cancer patient breath sample Blank tubes Praise Experimenter and handler 1. Sniffing 2. If dog indicates correctly: Clicker Food reward III Randomly chosen Cancer patient breath sample Blank tubes Praise Experimenter only Testing 1. Sniffing 2. If dog indicates correctly: Clicker Food reward Singleblinded trials Randomly chosen Cancer patient breath sample Control breath sample Praise Experimenter only 1. Sniffing 2. If dog indicates on any stations, then: No clicker No food reward Zero trials Cancer sample not present Control breath sample Control breath sample No praise Experimenter only 1. Sniffing 2. Possible indication by dog No clicker No food reward Doubleblinded trials Randomly chosen Cancer patient breath sample or control breath sample Control breath sample No praise Experimenter only 30 correct in a row aRandom order among trials, not known to the handler. bSample status not known either to th e experimenter or to the handler. cOnly the location of the tested sample within the lineup of five stations was known to the experimenter, but the sample status (cancer or control) was unknown

PAGE 78

Canine 72 10. APPENDIX 3 Canine Scent Detection of Ovar ian Cancer: Data Recording Sheets (Table compiled by Michael McCulloch) Rec# Dog Kind Vid# TR I II III IV V Correct Notes 1 2 3 4 5 6 7 8 Ko mh rh G W T S H G W T S H G W T S H G W T S H G W T S H F L D Y A 1 2 3 4 5 6 7 8 Ko mh rh G W T S H G W T S H G W T S H G W T S H G W T S H F L D Y A 1 2 3 4 5 6 7 8 Ko mh rh G W T S H G W T S H G W T S H G W T S H G W T S H F L D Y A 1 2 3 4 5 6 7 8 Ko mh rh G W T S H G W T S H G W T S H G W T S H G W T S H F L D Y A 1 2 3 4 5 6 7 8 Ko mh rh G W T S H G W T S H G W T S H G W T S H G W T S H F L D Y A 1 2 3 4 5 6 7 8 Ko mh rh G W T S H G W T S H G W T S H G W T S H G W T S H F L D Y A 1 2 3 4 5 6 7 8 Ko mh rh G W T S H G W T S H G W T S H G W T S H G W T S H F L D Y A 1 2 3 4 5 6 7 8 Ko mh rh G W T S H G W T S H G W T S H G W T S H G W T S H F L D Y A 1 2 3 4 5 6 7 8 Ko mh rh G W T S H G W T S H G W T S H G W T S H G W T S H F L D Y A Date: ________________ [sample # in low er left] G = trainer instructs by gesturing W = waiting T = tugging on leash S = command to sit H = hesitation of dog = visited = fully correct indication = no indication = false indication = correct after false false after correct/ 1 2 3 4 5 6 7 8 Ko mh rh G W T S H G W T S H G W T S H G W T S H G W T S H F L D Y A Kind : 1 CAfood/blank/both 2 CA/blank/both 3 CA/blank/exp 4=CA/Ctr l/exp 5 = strange 6 = zero 7 = single 8 = double Indic: 0 = incorrect 1 = correct tria l 2 = correct zero Video counter: mark counter on tape. TR: Name of Trainer Notes: F = food L = leash D = distracted Y = happy to work A = data audited

PAGE 79

Canine 73 11. APPENDIX 4 PC/GC/FT-ICR MS The Fourier Transform Ion Cyclotron Resonance Mass Spectrometer (FT-ICR MS) was put together by Marshall and Comisarow in 1973 (Comisarow and Marshall 1974a and 1974b). Ion Cyclotron Resonance Mass Spectrometer. The ICR MS machine works by having (previously) ionized particles f unneled by an ion guide into an analyzer cell (also known as a penning trap). The analyzer ce ll is surrounded by a superconduc ting magnet and the bigger the magnet, the greater the final resolution will be. The charged particles are drawn toward the magnetic field (running through the center of the cell). Two excitation plates on either side of the cell have electrodes on an external circuit. Chirps, or oscillating radio frequency pulses are sent to the excitation plates via the electrodes. The chirps emit a range of frequencies, starting low (stim ulating larger io ns) but getting higher (to incite the smaller ones) with time. Si nce each ion has a particular cyclotron frequency that it travels at in the cyclot ron, the ions will be excited, and increase in radius, only when the chirp emitted is at the ions specific frequency. In a sample, identical ions pack together and move in sync around the cyclotron, in what is called an ion packet. The osci llating radio frequency pulses not only initiate the increase in ion circling radius, but also stimulate ion packets until they have reached their parking orbit, or largest radi us (about 1 cm. from the excitation plates). After the chirps cease, the packets spiral back to thei r original orbit and the ma chine captures the decay of the orbits over time. In addition to the excitation plates, th ere are detector plates which are in charge of detecting the charged particles tr aveling in the cyclot ron. As the ions near a detector plate, electrons of an equal magnitude to the nearby charged particles are displaced through an outside

PAGE 80

Canine 74 circuit and travel through a resistor. T hrough a fairly convoluted process, the resistor measures the voltage of the electrons as they move through the outside circuit. Th e voltage readings are amplified and digitized to produce the raw data. The raw data is a composite signal of all the cyclotron frequencies measured from all the ions measured in the sample layered on top of each other. Fourier Transform. With the help of the Fourier transf orm algorithm, a computer converts the raw data (in amplitude over time) into a spectrum of the ion frequencies. The spectrum of frequencies is then translated into a mass spectrum (mass over charge, or m/z). Preconcentrator/ Gas Chromatography Solouki et al., (2004) coupled a 3 stage Entech 7100 series preconcentrator (Entech, Simi Valley, CA) with a SRI GC system (Las Vegas, Nc) and a IonSpec 7-T FT-ICR mass spectrometer (IonSpec C op., Lake Forest, Ca) in order to simplify sample preparation and improve selectivity for analytical characterization. Human exhaled breath contains complex mixtures of molecules and these molecules can be sorted into like groups as they travel through hundreds of meters of fine tubing. The molecules segregate into groups due to specialized coatings on the inside of the tubes that affect the molecules in different ways. After the molecules are separated they are captured in a machine that suspends the molecules at negative 200 degrees until they can be transported to the ion cyclotron resonance mass spectrometer for analysis. For a more detailed review of the PC/GC/FT-ICR MS used see the instrumentation section of Solouki et al., 2004.

PAGE 81

Canine 75 12. Works Cited 1. Algarra, A. Garcia-Lora, T. Cabrera, F. Ruiz-Cabello, F. Garrido. (2004) The selection of tumor variants with altered e xpression of classical and nonclassical MHC class I molecules: implications for tumor immune escape, Cancer Immunol Immunother 53, 904. 2. Balseiro, S.C., H.R. Correia. (2006) Is olfactory detection of hum an cancer by dogs based on major histocompatiblitiy complexdependant odor components? A possible cure and a precocious diagnosis of cancer. Medical Hypothesis 66, 270-272. 3. Beets, M. (1970) The molecular parameters of olfactory response. Pharmacol Rev 22, 1-34. 4. Bouton, M.E. (2007) Learning and be havior; A contemporary synthesis. Sinauer Associates, Inc. Publishers Sunderland, Massachusetts. 5. Bottino, R. Castriconi, L. Moretta, A. Moretta. (2005) Cellular ligands of activating NK receptors. Trends Immunol 26, 221. 6. Brechbuhl, J., Klaey, M., Broillet, M.C. (2008) Grueneberg ganglion cells m ediate alarm pheromone detection in mice. Science 321, 1092. 7. Brennan P.A., K. M. Kendrick. (2006) Mammalian social odors: attraction and individual recognition. Philos. Trans. R. Soc. London Ser. B 361, 2061. 8. Buck, LB. (2000) The molecular archite cture of odor and pheromone sensing in mammals. Cell 100, 611-618. 9. Buck L, R. Axel. (1991) A novel mu ltigene family may encode odorant receptors: A molecular basis for odor recognition. Cell 65, 175-187. 10. Burnet, F.M. (1970) The concept of immunol ogical surveillance. Prog. Exp. Tumor Res. 13, 1 27. 11. Campoli, M., S. Ferrone. (2008) Tumor escape mechanisms: potential role of soluble HLA antigens and NK cells activating li gands. Tissue Antigens 72, 321-334. 12. Chang, C.C., M. Campoli S. Ferrone. (2004) HL A class I antigen expression in malignant cells: why does it not always correlate with CTL-mediated lysis? Curr Opin Immunol 16, 644. 13. Com isarow, M.B., A.G. Marshall. (1974a) F ourier transform ion cyclotron resonance spectroscopy. Chem. Phys. Lett. 25, 282-283. 14. Comisarow, M.B., A.G. Marshall. (1974b) Frequency-sweep fourier transform ion cyclotron resonance spectros copy. Chem. Phy. Lett. 26, 489-490.

PAGE 82

Canine 76 15. Cumm ings, A.J., R.C. Knibb, M. Erlewyn-Lajeunesse, R.M. King, G. Roberts, J.S.A. Lucas. (2010) Management of nut allergy influe nces quality of life and anxiety in children and their mothers. Pediatr Aller gy Immuno. (Epub ahead of print) 16. Del Punta K., A. Puche, N.C. Adams, I. R odriguez, P. Mombaerts. (2002) A divergent pattern of sensory axonal proj ections is rendered convergent by second-order neurons in the accessory olfactory bulb. Neuron 35, 1057. 17. Dulac, C., R. Axel. (1995) A novel family of genes encoding putative pheromone receptors in mammals. Cell 83, 195. 18. Eggert, F., W.M. Muller-Ruchholtz, R. Ferstl. (1999) Olfactory cues associated with the major histocompatiblity complex. Genetica 104, 191-197. 19. Firestein, S. (1992) Elec trical signals in olfactory tran sduction. Curr. Opin. Neurobiol. 2, 444. 20. Fleischer, J., K. Schwarzenbacher, H. Breer. (2007) Expression of trace amine-associated receptors in the Grueneberg ganglion. Chem. Senses 32, 623631. 21. Foster, F.S., A. Falconer, A.R. Dodson, A. F. Norman, N. Dennis, A. Fletcher, C. Southgate, A. Dowe, D. Dearna ley, S, Jhavar, R. Eeles, A. Feber, C.S. Cooper. (2004) Transcription factor E2F3 overexpressed in pros tate cancer independently predicts clinical outcome. Oncogene 32, 5871-5879. 22. Giglione C, M. Pierre, T. Meinnel. (2000) Peptide deformylase as a target for new generation, broad spectrum antimicrobial agents. Mol Microbiol. 36(6), 1197 1205. 23. Guillamon, A., S. Segovia. (1997) Sex differenc es in the vomeronasal system. Sensory Input and Behavior. Brain Research Bulletin 44, 377-382. 24. Gordon, R.T., C.B. Schatz, L.J. Myers, M. Ko sty, C. Gonczy, J. Kroener, M. Tran, P. Kurtzhals, S. Heath, J.A. Koziol, N. Arthur, M. Gabriel, J. Hemping, G. Hemping, S. Nesbitt, L. Tucker-Clark, J. Zaayer. (2008) The use of canines in the detection of human cancers. J Altern Complement Med.14, 61-67. 25. Gotzsche, P.C., M. Nielsen. (2006) Screening for breast cancer with mammography. Cochrane Database Syst Rev. 4, CD001877. 26. Haitina, T., R. Fredriksson, S. Foord, H. Schith, D. Gloriam. (2009) The G proteincoupled receptor subset of the dog genome is mo re similar to that in human than rodents. BMC genomics 10, 24. 27. Halasz, N. (1990) The vertebrate olfactory system; Chemical neur oanatomy, function and development. Akademiai Kiado, Budapest.

PAGE 83

Canine 77 28. Hallem E.A., M.G. Ho, J.R. Carlson. (2004) The molecular basis of odor coding in the Drosophila antenna. Cell 117, 965. 29. Hanna, N., R.C. Burton (1981) Definitive evidence that natural killer (NK) cells inhibit experimental tumor metastases in vi vo. The Journal of Immunology 127, 5 1754-1758. 30. Harris, R., K.N. Lohr. (2002) Screening for pr ostate cancer: an update of the evidence for the U.S. Preventive Services Task Force. Ann Intern Med. 137, 917-929. 31. Horvath, G., G.A. Jarverud, S. Jarverud, I. Horvath. (2008) Human ovarian carcinomas detected by specific odor. Integr Cancer Ther. 7, 76-80. 32. Howard, J.C. (1997) H-2 and mating preferences Nature 266, 406. 33. Huebner, R.J., G.J. Todaro (1969) O ncogenes of RNA Tumor viruses as determinants of cancer. Viral Carcinogenesis branch, Na tional Caner Insititute communication. 34. Jacobson L, D. Trotier, K.B. Dving. (1998) Anatomical description of a new organ in the nose of domesticated anim al s by Ludvig Jacobson (1813). Chem Senses 23(6), 743-54. 35. Jager E, M. J. Ringhoffe r, M. Karbach, F. Oesch, Knuth (1996) A I nverse relationship of melanocyte differentiation antigen expression in melanoma tissues and CD8+ cytotoxic-Tcell responses: evidence for imm unoselection of antig en-loss variants in vivo. Int J Cancer 66, 470. 36. Jett, J.R. (2005) Limitations of screening fo r lung cancer with low-dose spiral computed tomography. Clin Cancer Res. 11, 4988s-4992s. 37. Jin, X., D.E. Bauer, S.E. Tuttleton, S. Lewin, A. Gettie, J. Blanchard, C.E. Irwin, J.T. Safrit, J. Mittler, L. Weinberger, L.G. Kost rikis, L. Zhang, A.S. Perelson, D.D. Ho (1999 ) Dramatic Rise in Plasma Viremia after CD8+ T Cell Depletion in Simian Immunodeficiency Virus-infected Macaques. J. Exp. Med. 189, 6. 38. Johnston, J..M. (1998) Enhanced canine explosives detection, unpublished report, Institute for Biological Detection Systems, Auburn University, Alabama. 39. Kalat, J.W. (2007) Biological Psychology 9th edition. Thomson Wadsworth Canada. 40. Kaluza J.F., F. Gussing, S. Bohm, H. Breer, J. Strotmann. (2004) Olfactory receptors in the mouse septal organ. J Neurosci Res 76:442. 41. Kandil, R.M., C..M. Davis. (2009) Shellfish allergy in children. Pediatr Allergy Immunol. 20(5), 408-14. 42. Karp, G. (2008) Cell and m olecular bi ology; Concepts and experiments 5th edition. John and Wiley Sons, Inc. Danvers, MA 662-726.

PAGE 84

Canine 78 43. Katada, S., T. Hirokawa, Y. Oka, M. Suwa, K. Touhara. (2005) Structural basis for a broad but selective ligand spectrum of a mouse olfactory recept or: mapping the odorant-binding site. J. Neurosci. 25, 1806. 44. Katada S, M. Tanaka, K. Touhara. (2004) Struct ural determinants for membrane trafficking and G protein selectivity of a mouse olfactory receptor. J. Neurochem. 90, 1453. 45. Kato, A, K. Touhara. (2009) Mammalian olf actory receptors: pharmacology, G protein coupling and desensitization. Cell Mol. Life Sci. 66, 3743-3753. 46. Kavaliers, M., E. Choleris, A. Agmo, D.W. Pfaff. (2004) Olfactory mediated parasite recognition and avoidance linking gene s to behavior. Horm. Behav. 46, 272. 47. Kavaliers, M., E. Choleris A. gmo W. John Braun, D. D. Colwell L. J. Muglia S, Ogawa D. W. Pfaff (2006) Inadvertent social informati on and the avoidance of parasitized m ale mice: a role for oxytocin. Pr oc. Natl Acad .Sci. USA 103, 4293. 48. Kavaliers, M., E. Choleris, D.W. Pfaff. (2005) Genes, odors and the recognition of parasitized individuals by rodents. Trends Parasitol. 21, 423. 49. Kimoto H, K. Sato, F. Nodari, S. Haga, T. E. Holy, K. Touhara. (2007). Sexand strainspecific expression and vomeronasal activity of mouse ESP family peptides. Curr. Biol. 17, 1879. 50. Leinders-Zufall T, P. Brennan, P. Widmayer, A. Maul-Pavicic, M. Jage r, X.H. Li, H. Breer, F. Zufall, T. Boehm. (2004) MHC class I peptides as chemosensory signals in the vomeronasal organ. Science 306, 1033. 51. Leinders-Zufall T, R.E. Cockerhan, S. Michalak is, M. Biel, D.L. Garbers, R.R. Reed, F. Zufall, S.D. Munger. (2007) Contribution of the receptor guanyly ll cyclase GC-D to chemosensory function in the olfactory ep ithelium. Proc Natl Acad Sci USA 104, 14507 14512. 52. Levai O, T. Feistel, H. Breer, J. Strotmann J (2006) Cells in the vom eronasal organ express odorant receptors but project to the accesso ry olfactory bulb. J Comp Neurol 498, 476. 53. Liberles SD, L.B. Buck. (2006) A second class of chemosensory receptors in the olfactory epithelium. Nature 442, 645. 54. Liberles, S.D., L.F. Horowitz, D. Kuang, J.J. Contos, K.L. Wilson, J. SiltbergLiberles, D.A. Liberles, L.B. Buck. (2009) Formyl peptide receptors are candidate chemosensory receptors in the vomeronasal organ. Proc. Natl. Acad. Sci. USA 106, 9842. 55. Liman E.R., D.P. Corey, C. Dulac. (1999) TRP2: a candidate tran sduction channel for mammalian pheromone sensory signaling. Proc. Natl. Acad. Sci. USA 96, 5791.

PAGE 85

Canine 79 56. Lin, W ., J. Arellano, B. Slotni ck, D. Restrepo. (2004). Odors detected by mice deficient in cyclic nucleotide-gated channel subunit A2 stimulate the main olfactory system. J. Neurosci. 24, 3703. 57. Luskin, M.B., J.L. Price. (1983) The topographi ca organization of asso ciational fivers of the olfactory system in the rat, including centr igucal givers to the olfactory bulb. J. Comp. Neurol. 216, 264-291. 58. Machado R., D. Laskowski, O. Deffenderfer T. Burch, S. Zheng, P. Mazzone, T. Mekhail C. Jennings J.K. Stoller J. Pyle J. Duncan R.A. Dweik, S.C. Erzurum (2005) Detection of lung cancer by sensor array analyses of exhaled breath. A mJ Respir Crit Care Med 171, 1286. 59. Macrides, F., H.B. Eichenbaum, W.B. Forb es. (1982) Temoporal relationship between sniffing and the limbic O rhythm during odor di scrimination reversal learning. J. Neurosci. 2, 1705-1717. 60. Martin, D.R., R.C. Semelka. (2006) Health e ffects of ionising radiation from diagnostic CT. Lancet. 367, 1712-1714. 61. Mazzone, P. (2009) Sniffing out lu ng cancer. Nature Nanotechnology 4, 621. 62. Mazzone P., J. Hammel, R. Dweik, J. Na, C. Czich, D. Laskowski, T. Mekhail. (2007) Diagnosis of lung cancer by the analysis of exhaled breath with a colorimetric sensor array. Thorax. 62, 565. 63. McCulloch M., T. Jezierski, M. Broffman, A. Hubbard, K. Turner, and T, Janecki. (2006) Diagnostic accuracy of canine scent detection in earlyand late-stage lung and breast cancers. Integr Cancer Ther 5, 30-39. 64. Migeotte, I., D. Communi, M. Parmentier. ( 2006) Formyl peptide receptors: a promiscuous subfamily of G protein-coupled receptors cont rolling immune responses. Cytokine Growth Factor Rev. 17, 501. 65. Montag, S., M. Frank, H. Ulmer, D. Wernet, W. Gopel and H.G. Rammensee. (2001) Electronic nose detects major histocompa tibility complex-dependent prerenal and postrenal odor components, Proc Natl Acad Sci USA 98, 9249. 66. Moser, E., M. McCulloch. (2010) "Canine Scent Detection of Hum an Cancers: a Review of Methods and Accuracy." (in press, Journal of Veterinary Behavior) 67. Munger, SD. (2009) Noses within noses. Nature 459. 68. Niewenhuys, R. (1963) The comparative anat omy of the actinopterygian forebrain. J. Hirnforsh. 6, 171-192.

PAGE 86

Canine 80 69. Novotny M.V. (2003) Pherom ones, binding protei ns and receptor responses in rodents. Biochem Soc Trans 31, 117. 70. OConnor, S., T. Jacob (2008) Neuropharm ocology of the olfactory bulb. Current Molecular Pharmacology. 1, 181-190. 71. Olender, T., T. Fuchs, C. Linhart, R. Sham ir, M. Adams, F. Kalush, M. Khen, and D. Lancet. (2004) The canine olf actory subgenome. Genomics 83, 361. 72. Onoda, N., K. Imamura, E. Obata, M. Ilino. (1984) Response selec tivity of neocortical neurons to specific odors in the rabbit. J. Neurophysio. 52, 638-651. 73. Pai, M., M. McCulloch, J.D. Gorman, N. Pai, W. Enanoria, G. Kennedy, P. Tharyan, J.M. Colford. (2004) Systematic reviews and meta-analyse s: an illustrated, st ep-by-step guide. Natl Med J India. 17, 86-95. 74. Pearse-Pratt R, H. Schellinck, R. Brown, B. Roser. (1992) Evolutionary and immunological implications of the role of the MHC in olfactory signaling. In: Doty RL, Mller-Schwarze D (eds). Chem ical Signals in Vertebra tes, VI. Plenum Press: New York, 167. 75. Pearse-Pratt R, H. Schellinck, R. Brown, B. Singh, B. Roser (1999) Soluble MHC antigens and olfactory recognition of genetic individuality: the mechanism. Genetica 104, 223. 76. Peng, G., U. Tisch, O. Adams, M. Hakim, N. Shehada, Y. Broza, S. Billan, R. AbdahBortnyak, H. Haick. (2009) Diagnosing l ung cancer in exhaled breath using gold nanoparticles. Nature Nanotechnology 4, 669-673. 77. Penn D., W. Potts (1998) How do ma jor histocompatibility complex genes influence odor and mating preferences ? Adv Immunol. 69, 411. 78. Pickel, D.P., G.P. Manucy, D.B. W alker, S.B. Hall, J.C. Walker. (2004) Evidence for canine olfactory detectio n of melanoma. Appl. Anim. Behav. Sci. 89, 107-116. 79. Poo, C., J.S. Isaacson. (2009). Odor representati ons in olfactory cortex: sparse coding, global inhibition, and oscillations. Neuron 62, 850. 80. Potts, W.K., E.K. Wakeland. (1993) Evolution of MHC genetic diversity; A tale of incest, pestilence, and sexual preference. Trends in Genet. 9, 181-187. 81. Prices J.L. (1990). Olfactory system. In The Human Nervous System, ed. G Paxinos, 979 1001. San Diego: Academic 82. Pryor, K. (1999) Dont Shoot shoot the dog!: Th e new art of teaching and training. Bantam Books New York.

PAGE 87

Canine 81 83. Quignon, P., E. Kirknes s, E. Cadieu, N. T ouleimat, R. Guyon, C. Renier, C. Hitte, C. Andr,C. Fraser, F. Galibert. (2003) Comp arison of the canine and human olfactory receptor gene repertoires. Genome Biology, 4. 84. Quignon, P., M. Giraud, M. Rimbault, P. Lavi gne, S. Tacher, E. Morin, E. Retout, A. Valin, K. Lindblad-Toh, J. Nicolas, F. Galiber t. (2005) The dog and ra t olfactory receptor repertoires. Genome Biology 6. 85. Ressler, K.J., S.L. Sullivan, L.B. Buck. (1993) A zonal organization of odorant receptor gene expression in the olfact ory epithelium. Cell, 73, 597-609. 86. Riviere, S., L. Challet, D. Fluegge, M. Sp ehr, I. Rodriguez. (2009). Formyl peptide receptor-like proteins are a novel family of vomeronasal chemosensors. Nature 459, 574 577. 87. Robin, S., S. Tacher, M. Rimbault, A. Vaysse1, S. Drano1, C. Andr, C. Hitte, F. Galibert. Genetic diversity of canine ol factory receptors. BMC Genomics 10, 21. 88. Rodrigues I, U. Boehm. (2008) Pheremone sensing in mice. Probl Cell Differ, 47, 1-20. 89. Rouqier, S., D. Giorgi. (2007) Olfactory recept or gene repertoires in mammals. Mutation Research 616, 95-102. 90. Sam, M., S. Vora, B. Malnic, W. Ma, M.V. Novotny, L.B. Buck. (2001). Neuropharmacology. Odorants may arouse in stinctive behaviours. Nature 412, 142. 91. Seliger, B., M.J. Maeurer, S. Ferrone. (2000) Antigen-processing machinery breakdown and tumor growth Immunology Today 21, 455-464. 92. Shimura T, M. Hagihara, K. Yamamoto, K. Takebe, B. Munkhbat K. Ogoshi T. Mitomi, Y. Nagamachi, K. Tsuji (1994) Quantification of serum-so luble HLA class I antigens in patients with gastric cancer. Hum Immunol. 40, 183. 93. Singh, P.B., R.E. Brown, B. Roser. (1987) MHC antigens in urine as olfactory recognition cues. Nature 327, 161. 94. Singer, A.G., G.K. Beaucham p, K. Yamazak i. (1997) Volatile signals of the major histocompatibility complex in male mouse urine. Proc Natl Acad Sci USA 94, 2210. 95. Singer, A.G., H. Tsuchiya, J. Wellington, G. Beauchamp, K. Yamazaki, (1993) Chemistry of odortypes in mice: Fractionation and bioa ssay. Journal of Chemical Ecology 19, 3: 569579. 96. Sinha, R., D.E. Anderson, S.S McDonald, P. (2003) Greenwald. Cancer Risk and Diet in India. J Postgrad Med 49, 222-8.

PAGE 88

Canine 82 97. Slotnick, B.M. (1985) O lfactor y discrimination in rats with anterior anygdala lesions. Behavior and Neuroscience 99, 956-963. 98. Solouki, T., J. Szulejko, J. Bennet, L. Grah am. (2004) A preconcentr ator coupled to a GC/FTMS: advantages of sef-contained i onization, mass measurement accuracy, and high mass resolving power for GC applications Journal for American Society of Mass Spectrometry 15, 1191-1200. 99. Staubli, U., D. Fraser, M. Kessler, G. Lync h. (1986) Studeis on retrograde and enterfograde amnesia of olfactory memory after denerva tion of the hippocampus by entorhinal cortex lesions. Behav. Neur al. Biol. 46, 432-444. 100. Stephan AB, E.Y. Shum S. Hirsh, K.D. Cygnar, J. Reisert, H. Zhao. (2009) ANO2 is the cilial calcium-activated chloride channel th at may mediate olfactor y amplification. Proc Natl Acad Sci USA 106, 11776. 101. Stowers L, T.E. Holy, M. Meister, C. Dulac, G. Koentges. (2002) Loss of sex discrimination and male-male aggression in mice deficient for TRP2. Science 295, 1493 500. 102. Su, C, K. Menuz, J.R. Carlson. (2009) Olfactory perception: receptors, cells, cells, and circuits. Cell 139. 103. Takami, S., P.C. Graziadei. (1991) Light microscopic golgi st udy of mitral/tufted cells in the accessory olfactory bulb of th e adult rat. J. Comp. Neurol. 311, 65. 104. Thorndike, E.L. (1911) Animal intelligence : Experimental studies. New York: Macmillan. 105. Touhara, K., L.B. Vosshall. (2009) Sensing odorants and pheromones with chemosensory receptors. Annu. Rev. Physiol. 71, 307-32. 106. Von Dannecker L.E., A.F. Mercadante, B. Malnic. (2006) Ric-8B promotes functional expression of odorant receptors. Proc. Natl. Acad. Sci. USA 103, 9310. 107. Walker, D.B., P. Cavnar, J. Taylor, J. L. Taylor, D. H. Pickel, S. B. Hall, J. C. Suarez. (2003) Canine olfactory thresho lds to N-am yl acetate measured naturalistically. Paper presented at: Meeting of the Association for Chemoreception Sciences; Sarasota, Fla. 108. Wedikind C., S. Furi. (1997) Body odor pref erences in men and women; dothey aim for specific MHC combinations or simply he terosygosity? Proc R Soc Lond B 264, 1471-1479. 109. Wedekind, C., D. Penn. (2000) MHC genes, body odors, and odor preferences. Nephrol Dial Transplant. 15, 1269-1271. 110. Wedikind C., T. Seeback, F. Bettens, A. J. Peapke. (1995) MHC dependant mate preferences in humans, Proc R Soc. London Ser B 260, 245-249.

PAGE 89

Canine 83 111. Williams, D.A. (1994) Conditioned reinforcement; neglected or outmoded explanatory construct? Psychonomic Bulletin. 111, 275-290. 112. Williams, H., A. Pembroke. (1989) Sniffer dogs in the melanoma clinic? Lancet 1, 734. 113. Willis, C.M., S.M. Church, C.M. Guest, W.A. Cook, M. McCarthy, A.J. Bransbury, M.R.T. Church, J.C.T Church. (2004) Olfactory detec tion of human bladder cancer by dogs: proof of principle study. Bmj 329, 712-718. 114. Wilson, E.O. (1963) Pheromones. Sci. Am. 208, 100. 115. Zufall, F., T. Leinders-Zufall. (2000) The cellular and molecular basis of odor adaptation. Chem Senses 25, 473. 116. http://www.personal.kent.edu/~cearley/ChemWrld/smell/olf_bulb.gif 117. http://abclocal.go.com/wls/sto ry? section=resources&id=6766540 118. http://www.peanutdog.com/index.html 119. http://www.cdc.gov/cancer/ski n/statistics/state.htm 120. http://www.magnet.fsu.edu/education/tu torials/m agnetacademy/fticr/index.html