The Distribution of Small Mammals and its Relationship with Ticks in Two Coastal Areas
of Norway
Fredrik Munkelien
Master of Science thesis
Centre for Ecological and Evolutionary Synthesis, Department of Biosciences
University of Oslo 31.05.2018
© Fredrik Munkelien Year: 2018
Title: The Distribution of Small Mammals and its Relationship with Ticks in Two Coastal Areas of Norway
Author: Fredrik Munkelien
Supervisors: Professor Atle Mysterud and associate professor Torbjørn Håkan Ergon http://www.duo.uio.no/
Acknowledgements
This master thesis was conducted at the CEES (Centre for Ecological and Evolutionary Synthesis) under the supervision of my main supervisor professor Atle Mysterud and co- supervisor associate professor Torbjørn Håkan Ergon. Since I was young, I have been interested in how everything works, and in nature especially, and my parents have always been encouraging in this regard. As such, ever since I learned to read, and probably for some time before that, I have wanted to work with science. I therefore want to give a special thanks to Atle for letting me take part in this project, for which I am very grateful, and I am also very thankful for his excellent supervision which has been of tremendous help to me. I am of course also very thankful for the great help and guidance Torbjørn provided me, especially regarding the programming and statistical analysis, which has been invaluable to me. I must also thank Anders Herland, Vetle Malmer Stigum and Harald Linløkken for helping me out in both the field and in the lab, and Jeroen van der Kooij for his expert help in species determination of the small mammals. I also want to thank my fellow students, who have been of great support during this project. Last but not least, I must also thank my friends and especially my family, for encouraging me from an early age. My parents taught me to appreciate nature, and without them, I am certain I would never have studied science, let alone biology. This has truly been an interesting project, and I have learned quite a lot working on it!
Abstract
In northern Europe, the tick Ixodes ricinus is of particular concern, as it is the principal vector of the spirochete Borrelia burgdorferi sensu lato (s.l.) which is known to cause Lyme disease in humans, among other pathogens. Small mammals such as rodents and shrews are found to be the main host of I. ricinus larvae and competent reservoirs for tick-borne pathogens. As such, it is necessary to better understand the ecology of these small mammal tick hosts in order to increase the knowledge on tick-borne pathogens and how they spread. This thesis therefore focused on three of the most important small mammal hosts for I. ricinus larvae in Scandinavia, the bank vole (Myodes glareolus), the wood mouse (Apodemus sylvaticus), and the common shrew (Sorex araneus). The main aim of this thesis was to investigate to what extent habitat affect the occupancy and abundance of different small mammal species and their tick loads.
There were two study areas used in this thesis; Førde, which is located in Sogn of Fjordane, and Son, which is located in Akershus, both in Norway. As expected, there were differences between the rodents and shrews in regards of their population structure, and the common shrew dominated in terms of abundance in Førde and constituted about 54% of the catch in Førde, while the rodents were the most abundant in Son with the bank vole constituting about 42% of the catch. The abundances were found to vary between the years and season, but overall, the abundance of the common shrew was found to be more stable than the rodents’. The spatial distributions of the small mammals increased in years with increasing abundances suggesting density-driven occupancy patterns. In order to gain insight into how the occupancy, abundance and tick load of these small mammals is affected by the habitat, several variables were measured at each trapping station. Vegetation cover by different groups and species of plants was represented by two principal component axes (PCA), in addition to structural variables such as hole in soil, hole under stone and canopy cover. There was in general too low spatial variation to explain their occupancy by the habitat variables, consistent with these small mammals being habitat generalists. While their abundance could be partially explained by these habitat variables, there was however little consistency in which specific variables affected the abundance between species and regions. The load of I. ricinus larvae on the small mammals was found to fluctuate both yearly and seasonally, and it was generally higher in shaded areas.
However, the variables affecting the tick load differed from those affecting the small mammal
generalist nature and high abundance of the small mammal species found in the both regions make them important as host for I. ricinus, and hence they likely play important roles in the transmission cycle of tick-borne diseases.
Contents
1. Introduction ... 1
2. Methods ... 4
2.1 Study areas ... 4
2.2 Capture and study design ... 5
2.3 Laboratory work ... 8
2.4 Measurements of habitat variables ... 8
2.5 Analysis ... 10
2.5.1 Habitat variables ... 10
2.5.2 Multi-season occupancy model ... 10
2.5.3 Zero-inflated Poisson model ... 12
2.5.4 Tick load ... 13
3. Results ... 14
3.1 Description of capture data ... 14
3.2 Relationships among vegetation variables ... 15
3.3 Multi-season occupancy model – Bank vole ... 16
3.4 Abundance and occupancy – Bank vole ... 18
3.5 Abundance and occupancy – Wood mouse ... 21
3.6 Abundance and occupancy – Common shrew ... 24
3.7 Overview of tick load ... 29
3.8 Analysis of tick load – Bank vole ... 31
3.9 Analysis of tick load – Wood mouse ... 34
3.10 Analysis of tick load – Common shrew ... 37
4. Discussion ... 40
4.1 Estimation of occupancy probability and abundance ... 40
4.2 Small mammal communities and temporal variation ... 41
4.3 Small mammal communities and habitat ... 42
4.4 Tick load on small mammals and habitat ... 44
5. Conclusion ... 46
6. References ... 47
A3. Results from multi-season occupancy model – Occupancy and detection ... 53
A4. Results from multi-season occupancy model – Wood mouse ... 56
A5. Results from multi-season occupancy model – Common shrew ... 57
A7. Correlation between the numerical variables ... 58
1. Introduction
In Europe, one of the ectoparasite groups of most concern to the public is the ticks, and in particular the tick Ixodes ricinus (Medlock et al. 2013). I. ricinus is widely distributed and well known for spreading pathogens to both livestock and humans. I. ricinus has four stages of development, of which 3 requires a blood meal in order to moult and reproduce: larva, nymph, and adult. During the subsequent feeding on hosts, the tick may vector pathogens from one host to the next.
I. ricinus is a generalist in terms of choice of host, and they feed on a large variety of vertebrates like birds, reptiles, and mammals (Gern et al. 1998, De Sousa et al. 2012). As a result of the wide host range, humans are exposed to a number of tick-borne pathogens, such as the spirochete Borrelia burgdorferi sensu lato complex, which is of great concern to human health as it causes Lyme disease (Franke et al. 2013). Understanding the mechanisms of vector- borne diseases and how they spread is of great importance in order to mitigate the effects of these diseases. To do so, it is necessary to gain an understanding of the hosts of the ticks, and in particular of those hosts also transmitting the pathogens.
Studies have shown that small mammals, such as rodents and shrews, are the preferred hosts of I. ricinus larvae (Aeschlimann 1991). However, there is little information about spatial variation on the main larval tick hosts (Kilpatrick et al. 2017), but one may assume that small mammal communities differ regionally. Studies have shown that the rodent species wood mouse (Apodemus sylvaticus) and bank vole (Myodes glareolus) are very competent B.
burgdorferi reservoir hosts, as well as the common shrew (Sorex araneus) (Gern et al. 1998, Bown et al. 2011, Mysterud et al. 2015, Hofmeester et al. 2016), while the rodents are known to produce the majority of the infected nymphs (Hofmeester et al. 2016).
Since small mammal tick hosts are important pathogen reservoirs and their distribution and density affect the tick density (Ostfeld et al. 2006), a better understanding of the distribution and ecology of small mammals over a longer time period would thus help understand the epidemiology of tick-borne diseases. The environmental variables of a site reflect what resources are available for food and shelter, and hence also the probability that a given species is present at a given site or not (occupancy), and if present - its abundance. As habitats differ in
are insectivores whose diets consist mostly of invertebrates, supplemented with vegetable matter such as coniferous seeds (Crowcroft 1957, Rudge 1968, Myllymäki and Paasikallio 1972). Due to the insectivorous diet of shrews, any effect of vegetation is thought to be mostly secondary by sustaining invertebrate populations, as vegetation plays a lesser dietary role and their burrows are underground (Crowcroft 1957). Rodent populations at northern latitudes often show abundance cycles (Steen et al. 1996, Strann et al. 2002). The shrew populations are less well studied, but found to be more stable than the rodent population with less pronounced cycles – if any at all (Strann et al. 2002).
Current studies of host use of the I. ricinus do not take into account variation in host species density and distribution due to differing habitats, and they implicitly assume that their sample of hosts is representative of the small mammal community composition in the ecosystems. However, the detection probability is linked to several processes such as abundance and may differ between host species. One could expect differences between rodents and shrews in terms of occupancy and abundance as they are likely to differ in habitat use. How different small mammals utilise different habitats, may to some extent be found through examining how environmental variables affect the probability of occupancy and abundance. By using occupancy modelling, it can be determined whether certain habitats seem more favourable to the different species of small mammals (MacKenzie 2006).
Although there are several qualitative studies on host use of the I. ricinus, there are few quantitative studies on this parasite-host-relationship in Scandinavia. This thesis therefore aims to increase the understanding of the habitat use of small mammal tick hosts in two coastal areas over several years with fluctuations in population abundance, and the thesis is part of work connected to the NFR-funded project #EcoTick. In order to study the ecology of small mammal hosts, I used both available and new data (1871 catches) gathered as part of this thesis from two different ecosystems. The data consists of both occupancy and abundance of certain species, as well as the tick data for each animal caught. Measurements of habitat variables were gathered at the trapping stations. These datasets were combined to perform occupancy modelling. In the following sections, I will further elaborate on the methods used in this thesis.
This thesis aims to answer:
• To what extent do small mammal species and groups (shrews vs. rodents) differ in occupancy and abundance and can this be explained by specific habitat variables?
• Are the distributions of the small mammals density dependent, that is, does occupancy vary over years depending on population abundances? And if so, is this similar across species and groups?
• To what extent can accounting for habitat affect predictions of tick load in small mammals?
2. Methods
2.1 Study areas
Two established study areas, Førde and Son, were used as in previous studies (Mysterud et al.
2015, Linløkken 2017). Both areas are coastal, but they differ in topography and climate.
The Førde area consists of Førde and Askvoll municipality, both situated in Sogn og Fjordane county in western Norway (Figure 1). Much of Førde is steep and hilly, being located by a fjord. The climate in Førde is typical for western Norway, having mild winters and high levels of precipitation. The bedrock is pre-Cambrian and is part of the Sveconorwegian orogeny (Bingen et al. 2008). It is principally composed of igneous rocks such as granite and syenite, and metamorphic rock such as gneisses which are to a large part transformed from granitic rock (Norges Geologiske Undersøkelse 2018). Førde lies mostly in the boreonemoral zone (Abrahamsen 1977). The vegetation in Førde consists mostly of areas with deciduous and coniferous forests, and sheep pastures. The deciduous forests are mostly composed of birch (Betula spp.), alder (Alnus incana) and to some extent hazel (Corylus avellana). The hazel might to some degree have a similar role as the oak have in Son by replenishing the rodents’
pantry, especially during the winter. The coniferous forests consist mostly of planted Norway spruce (Picea abies), however, there are also areas with pine (Pinus sylvestris).
Son is located in Vestby municipality, which resides in Akershus county (Figure 1). Son is relatively flat, typically for eastern Norway. Being located by the Oslo fjord, the climate is relatively stable, with moderate amounts of precipitation. The bedrock in Son consists mostly of granitic gneisses and paragneiss, and it is part of the Sveconorwegian orogeny (Bingen et al.
2008, Norges Geologiske Undersøkelse 2018). Son is to a large degree an agricultural area, although intercepted by patches of forest. Similar to Førde, Son is within the boreonemoral zone (Abrahamsen 1977). Son consists mostly of mixed forests along with planted coniferous forests.
The mixed forests are dominated by spruce, birch and alder, while the coniferous areas are mostly being dominated by spruce. The flora in Son also contain some southern tree species such as oak (Quercus robur and Q. petraea), which are found in mixed forests and in smaller islands of oak forest. The oaks are found to provide an important food source for bank voles, especially during mast years (Selås et al. 2002), and this may contribute to increased winter survival as the bank voles may cache acorns in winter storages.
2.2 Capture and study design
Capture was done in the spring and fall in the years 2014-2017 in Førde and Son. Data from 2016-2017 has been collected as a part of this thesis, however, data from Førde fall 2017 were not processed in time for this thesis. To prevent local depletion of the small mammals, each station was placed at least 500 m apart. The trapping stations were placed along the main road of the respective areas, with a minimum distance at 50 m from the road.
In Førde there were two transects called Førde-Vest and Angedalen (Figure 1). The Førde-Vest transect was located in western Førde municipality. This transect was further extended into Askvoll municipality with 10 stations. The Førde-Vest transect was located along Førdefjorden, with several of the trapping stations being close to the sea. The Angedalen transect was located north-east in Førde municipality and was more inland compared to the Førde-Vest transect. There was in total 50 trapping stations in Førde. In Son, there was a single trapping area, and the trapping stations in Son were more widespread (Figure 1), rather than following linear transects as in Førde. In Son, there were 25 trapping stations.
Figure 1: Map of the three different transects Førde Vest/Askvoll and Angedalen (Sogn og Fjordane county), and Son (Akershus county), Norway. Trapping stations are represented by black dots. From Linløkken (2017) with permission from the author.
Small mammals were trapped by the aid of Ugglan multiple-capture live traps using the small quadrate method (Myllymäki et al. 1971). At each trapping station, 4 traps were placed at approximately each corner of a 15 m x 15 m square (Figure 2). In order to enhance the probability of capture, the traps were placed strategically in the terrain within a 2 m radius of each corner.
The capture was done over the course of 4 days each session. In the first day, the traps were prepared and baited. The traps were baited with oats and carrots, the latter as a source of water. The bait was in sufficient quantity to ensure survival for at least 24 hours. In the following three days, traps were checked for catch and rebaited if necessary. The catch was humanely killed by dislocation of the cervical vertebrae, and then individually stored in zip- lock bags and registered.
Figure 2: Overview of a trapping station were the traps are represented by red dots. The vegetation data was collected in a 30 m x 30 m grid based on the trap layout, with a total of 25 points per station.
At each of the 25 points, I measured the vegetation coverage within a 0.5 m x 0.5 m metal frame, in addition to structural variables in close vicinity. At the centre point (C3) I measured station variables which were general to the station as a whole rather than a single point.
2.3 Laboratory work
After capture, the catch was checked for ticks during laboratory work, and the tick load per individual was measured. This was done by first collecting ticks by using tweezers and a magnifying glass. The time used per individual was standardised to 20 minutes each. Our method is similar to previous work by the group (Mysterud et al. 2015, Linløkken 2017). Even though the species differ in size and hence surface body area to search, this procedure is unlikely to yield bias, as very few ticks were found in the last five minutes of examination except in rare cases. When the 20 minutes had ended, the collected ticks were counted, and their species (I.
ricinus or I. trianguliceps) and instar stages (larvae, nymphs, and adult males or females) were determined. If other ectoparasites were present, such as mites, fleas, lice, etc., those were also collected in the same manner. The categorised ectoparasites were stored in marked containers corresponding to the individual small mammals, and the ticks and other ectoparasites were stored in different containers. This procedure allows for further analyses to be done on the genetic material for later projects. Data about the small mammals were also collected, including species identity. The species of the small mammals were further confirmed by an expert, Jeroen van der Kooij, to ensure correct determination.
2.4 Measurements of habitat variables
The habitat measurements were done as described in Figure 2 and conducted in September 2016 in Son, and in June 2017 in Førde. The vegetation coverage (in %) was estimated by rough visual assessment inside the 0.5 m x 0.5 m metal frame at each point. Some points had less than 100% coverage by the vegetation variables, due to objects or plant species that were not included in the survey covering some proportion of the area estimated. Most often this was due to rocks, tree branches and dead leaves.
The vegetation coverage data included the following 8 species/groups:
• Bilberries (Vaccinium myrtillus)
• Woody shrub (Ericaceae)
• Grass (Poaceae)
• Herbs
• Peat mosses (Sphagnum)
• Other mosses (Bryophyta)
• Lichens
• Ferns (Polypodiopsida)
Each point was also surveyed for structural habitat variables thought to be of importance for presence and abundance of small mammals. These variables were merely measured if present nearby (within 2 m) each point or not. The structural variables included:
• Water/mud (presence of water or mud)
• Hole in soil (in the ground or under trees)
• Hole under stone (below, between rocks, or otherwise directly adjacent to stone) Hole in soil and hole under stone at first seem similar, but they differ as holes in soil are thought to be mostly made by an animal, presumably small mammals, while hole under stone may also occur ‘naturally’. Both provide shelter for small mammals, but hole in soil may as such potentially be a better indicator for small mammal presence.
In order to use vegetation coverage data and the above mentioned structural variables in further analyses, I converted the variables from 25 data points per station to mean per station. I also estimated some additional structural variables that were more general to each station, which were estimated from the centre point of the station. These variables included:
• Vegetation type (deciduous, coniferous or mixed forest, and pasture)
• Canopy cover
Vegetation type was determined for station area itself only, not accounting for possibly different vegetation types in adjacent areas, while canopy cover was measured by a spherical densiometer model C (Lemmon 1956).
2.5 Analysis
All analysis was done in R version 3.4.1 (R Development Core Team 2017). Model selection was based on Akaike Information Criterion (AIC). To avoid complicated models, the model with the fewest parameters within 2 ΔAIC was chosen. For the multi-season occupancy models and the zero-inflated Poisson models, model selection was based on AICc to account for low sample sizes. Otherwise, the model selection procedure was identical. I chose to focus analysis on the three most common small mammal species which are present in both Førde and Son, which also are the most important tick larvae hosts. These species were the bank vole, the wood mouse and the common shrew.
2.5.1 Habitat variables
There were in total eight variables describing vegetation coverage. I performed a principal component analysis (PCA) on these eight variables to determine how the different variables were correlated with each other. The resulting PCA axes were then used as explanatory variables representing vegetation, being more tractable than all 8 variables partly representing the same information when correlated. The two first principal component axes (PCA1 and PCA2) explained most of the variation (56%) and were therefore used.
In addition to PCA1 and PCA2 on vegetation coverage, I also wanted to include the other habitat data. The habitat variables were vegetation type, canopy cover, water/mud, and hole in soil and hole under stone. The numeric variables were standardised for optimal model fitting performance. This makes the mean set to 0 and SD of 1.
2.5.2 Multi-season occupancy model
Occupancy modelling allows estimation of how environmental variables affect the probability of occupancy of a certain species at a given site (MacKenzie 2006). Several different processes may affect whether you catch a given small mammal species at any particular time. Even if a species is present at the site, it is not certain it is caught (detected). The detection process may be further complicated by extinction and recolonization between each trapping session in a given location. Given that certain species of small mammals experience periodical fluctuations, an approach which accounts for this seasonal variation is therefore preferred. By using multi- season occupancy modelling over several years and seasons, one can not only estimate how different habitat variables affect occupancy and detection of small mammals but also how this
may vary over time (MacKenzie 2006). As such, by using a multi-season approach, I avoid any potential biases that may occur by chance from using a single season ‘snapshot’ of the population development.
I used the R package ‘unmarked’, which allows us to do multi-season single species analyses (Fiske and Chandler 2011). The ‘colext’ function in ‘unmarked’ allows for such an analysis with four sub-models: ‘Initial’, which represents the probability for occupancy at the different sites for the first year; ‘colonisation’, which represents the probability that an unoccupied site is colonised from one year to another; ‘extinction’, which represents the probability of an occupied site going extinct from one year to another; and ‘detection’, which represents the probability of detecting a species given that it is present on a site (Fiske and Chandler 2011).
All estimates are given on a logit scale.
Due to the four different model components, there was a limit to how many variables can be entered due to limited sample sizes. To get convergence, I had to limit the number of parameters based in part of an initial screening of data as well as prior biological knowledge.
The ‘initial’ sub-model of the occupancy model included 4 parameters: PCA1, PCA2, region, and a structural variable. This structural variable could be either vegetation type, canopy cover, water/mud, hole in soil, or hole under stone and was selected by AICc.
Due to differences in species composition and landscape, I expected there to be difference between Son and Førde in terms of occupancy. On the other hand, I expected the habitat variables to be of greater importance than region for the colonisation and extinction probabilities, as vegetation and habitat structure often limit the resources available for the individual animals. Therefore, I excluded region from the ‘colonisation’ and ‘extinction’ sub- models, while retaining PCA1, PCA2 and the habitat parameter from the ‘initial’ sub-model.
Given that a site is occupied, I do not expect the habitat variables to significantly influence the detection probability. Rather, I expect temporal and spatial variation in the detection probability across seasons and years due to fluctuations in population abundances, which likely differs between Son and Førde. There are also variations species-wise in the relative population abundances between Førde and Son. Because of this, in the ‘detection’ sub-model I included the interaction region:year:season.
2.5.3 Zero-inflated Poisson model
The multi-season occupancy model was very complex and time-consuming to work with, and models yielded wide confidence intervals, indicating insufficient data to estimate environmental effects precisely. There was a need for an alternative approach on how to model the different habitat variables’ effect on the small mammals’ choice of habitat. A zero-inflated Poisson (ZIP) distribution is more likely to contain zeros than a Poisson distribution (Zuur et al. 2009). A zero in a ZIP distribution may stem from one of two sources; either absence of the species at a site or failure to detect it even when it is present (Zuur et al. 2009). By using a ZIP model on the total number of individuals caught during the study period without taking into account yearly and seasonal variation, the zeros are more likely to be caused by absence from a given site rather than detection failure. As zero-inflation is likely to be due to aversion of certain habitats by this approach, the probability for occupancy is the same as 1 minus the probability of zero-inflation, or in other words 1 minus the probability of absence. By modelling for zero-inflation, it is then possible to see how the different variables affect the occupancy.
While being simpler than the multi-season occupancy model, this approach is, however, less informative as it does not take into account either colonisation or extinction.
The R package ‘pscl’ support ZIP models (Zeileis et al. 2008), and it has been used for this purpose in this thesis. This package allows modelling of both zero-inflation and the expected count given occupancy. In cases when a ZIP model could not be fitted for a given species due to too few zeros in the data, I fitted the data with a Poisson model by using the
‘glm’ function built into R, in order to at least get a count model for the abundance (this model assumes that the probability of occupancy is one). The candidate covariates and factors were PCA1, PCA2, canopy cover, hole in soil, hole under stone, vegetation type, year, season and the interaction year:season.
2.5.4 Tick load
Tick load was analysed using negative binomial models in the package ‘glmmADMB’ (family
= ‘nbinom’) in R (Fournier et al. 2012, Skaug et al. 2013). A negative binomial model is often used when analysing data on load of ectoparasites, which accounts for the fact that ticks are known to aggregate on a relatively few animals (Shaw and Dobson 1995). Such models therefore often give a better fit than for example Poisson models often used for many other types of count data, and this has been confirmed in earlier analyses of a subset of the data used here (Mysterud et al. 2015). These models give an overdispersion parameter (alpha), which describes how well the variance in a model fits a negative binomial distribution. Using the
‘nbinom’ family, the variance of the distribution fitted to the data (Var(Y) is related to the mean (E(Y)) as Var(Y) = E(Y) * (1 + E(Y) / alpha). Hence a low alpha signifies greater variation than expected from a Poisson distribution where Var(Y) = E(Y), while the variation will converge towards a Poisson distribution at higher alphas.
As different small mammal species utilise the habitat differently, it makes sense to analyse the tick load species-wise. I regard it easier to identify what environmental variables that are of interest without using complex models, i.e., with too many interaction terms. I also fitted separate models for each region to avoid too complicated interactions terms in the model, as I expect different results from Førde and Son, as the areas differ in small mammal species composition, in landscape and climate.
The response variable used in the tick load models was the load of I. ricinus larvae. The focus on I. ricinus larvae was chosen due to its importance as a vector for the pathogen causing Lyme borreliosis (Franke et al. 2013). However, the small mammal populations in Norway also feed another tick species, the I. trianguliceps, which is considered a rodent specialist. As I.
trianguliceps is a specialist on rodents and linked to their burrows, it does not act as a direct disease vector to humans or livestock. However, it may aid in maintaining high infection levels of several pathogens in the small mammal populations. The tick load of I. trianguliceps was not considered in this thesis as its importance for the Lyme borreliosis epidemiology is lessened compared to I. ricinus.
The candidate covariates and factors were PCA1, PCA2, canopy cover, hole in soil, hole under stone, vegetation type, year, season and the interaction year:season. I used station as a
3. Results
3.1 Description of capture data
The capture for all years 2014-17, except Førde fall 2017, included 1871 individual small mammals. In Son, capture consisted of five species: Bank vole, field vole (Microtus agrestis), wood mouse, common shrew, pygmy shrew (Sorex minutus) and a single water shrew (Neomys fodiens). In Førde, there were in addition: yellow-necked mouse (Apodemus flavicollis), taiga shrew (Sorex isodon), and a single house mouse (Mus musculus). Despite Son having half as many trapping stations as Førde, the total number of trapped animals were almost equal (Table 1). This indicates a higher population density in Son compared to Førde, although Førde, with 9 species captured, had a higher species richness of small mammals, with only 6 species. As the bank vole, the wood mouse and the common shrew are the most important small mammal hosts for I. ricinus larvae in Scandinavia, and they are abundant in both regions, I chose to focus on them in this thesis.
Table 1: An overview of the trapped mammal species for years 2014-17 in Førde and Son, Norway.
Total catch Percentage
Species Førde Son Førde Son
Bank vole 60 387 6.40 % 41.70 %
Field vole 42 1 4.48 % 0.11 %
Wood mouse 83 124 8.85 % 13.36 %
Yellow-necked mouse 20 0 2.13 % 0.00 %
House mouse 1 0 0.11 % 0.00 %
Common shrew 510 359 54.37 % 38.69 %
Pygmy shrew 49 56 5.22 % 6.03 %
Taiga shrew 158 0 16.84 % 0.00 %
Eurasian water shrew 15 1 1.60 % 0.11 %
Sum 938 928 100.00 % 100.00 %
3.2 Relationships among vegetation variables
The Principal Component Analysis (PCA) showed that coverage of herbs and grasses were correlated and linked to the lower end of PCA1 (Figure 3). The higher end of PCA1 was linked to higher coverage of bilberries and other woody shrubs. Along PCA2, ferns and mosses position themselves at the lower end of the axis, while grasses were found at the higher end.
Figure 3: A Principal Component Analysis (PCA) of vegetation coverage at trapping stations in Førde and Son, Norway. Coverage of herbs, grasses and ferns were on the lower end of PCA1, while lichens, woody shrubs, peat mosses and bilberries were one the higher end. On PCA2, ferns and mosses were
3.3 Multi-season occupancy model – Bank vole
The multi-season occupancy model for the bank vole included the variables PCA1, PCA2, hole under stone, and region in the ‘initial’ sub-model. The ‘colonisation’ and ‘extinction’ sub-model included PCA1, PCA2 and hole under stone. The ‘detection’ sub-model included the interaction between region:year:season, and the factor capture day number.
In the ‘initial’ sub-model, region had a strong positive effect on the probability of initial occupancy (Table 2), and initial occupancy was found to be higher in Son than Førde. PCA1 and PCA2, on the other hand, had no detectable effect on the probability of initial occupancy, while the presence of hole under stone had a strong positive effect.
In the ‘colonisation’ sub-model, neither the PCA axes nor presence of holes under stone had a clear effect on colonising (Table 2). In the ‘extinction’ sub-model, PCA1 had a negative effect, and PCA2 had a significant positive effect. This indicates that habitats with a high coverage of grasses may be unfavourable to the bank vole. Hole under stone had no statistically significant effect on extinction.
In the ‘detection’ sub-model, many season-year interactions had a statistically significant effect, implying that detection probability varied a lot through time (Table 2).
Naturally, there was overall higher detection probability in the fall when the population is correspondingly larger, which was apparent for both regions. Interestingly, in 2016 in Son there was a sharp peak in detection probability, with high detection probability even in the spring.
This corresponds well with the catch data which had a comparable increase in the number of bank voles caught in 2016. There was found no statistically significant difference in detectability over the three days of capture, suggesting no marked population depletion.
Table 2: Parameter estimates from the multi-season occupancy model for the bank vole. Note that in the interaction SF is short for the region Førde (Sogn og Fjordane). SE = standard error. z = standard score. p = probability value.
Initial
Parameter Estimate SE z p
(Intercept) -2.762 0.970 -2.848 0.004
PCA1 0.014 0.935 0.015 0.988
PCA2 -0.416 0.731 -0.569 0.569
Hole under stone 9.814 5.015 1.957 0.050
Region (Son vs Førde) 2.301 1.097 2.098 0.036
Colonisation
(Intercept) -0.954 0.424 -2.2523 0.024
PCA1 2.177 1.343 1.6214 0.105
PCA2 1.035 0.959 1.0789 0.281
Hole under stone -0.481 5.376 -0.0895 0.929
Extinction
(Intercept) -1.060 0.427 -2.483 0.013
PCA1 -1.700 0.897 –1.892 0.058
PCA2 2.190 0.814 2.688 0.007
Hole under stone -2.360 2.544 -0.928 0.353
Detection
Region:Year:Season (SF.y2014.fall) -0.851 0.551 -1.543 0.123 Region:Year:Season (SF.y2014.spring) -1.593 0.621 -2.566 0.010 Region:Year:Season (SF.y2015.fall) -2.288 0.453 -5.050 <0.001 Region:Year:Season (SF.y2015.spring) -3.532 0.754 -4.683 <0.001 Region:Year:Season (SF.y2016.fall) -1.939 0.432 -4.490 <0.001 Region:Year:Season (SF.y2016.spring) -3.580 0.750 -4.775 <0.001 Region:Year:Season (SF.y2017.spring) -3.615 0.750 -4.823 <0.001 Region:Year:Season (Son.y2014.fall) 1.005 0.333 3.014 0.003 Region:Year:Season (Son.y2014.spring) -0.614 0.494 -1.242 0.214 Region:Year:Season (Son.y2015.fall) 1.028 0.422 2.433 0.015 Region:Year:Season (Son.y2015.spring) -1.003 0.380 -2.636 0.008 Region:Year:Season (Son.y2016.fall) 1.782 0.387 4.604 <0.001 Region:Year:Season (Son.y2016.spring) 1.294 0.551 2.349 0.019 Region:Year:Season (Son.y2017.fall) -0.080 0.338 -0.237 0.813 Region:Year:Season (Son.y2017.spring) -1.124 0.363 -3.096 0.002
capture day no 2 0.370 0.231 1.601 0.109
capture day no 3 -0.278 0.236 -1.176 0.240
3.4 Abundance and occupancy – Bank vole
The most parsimonious model (in terms of lowest AICc values) for the bank vole in Førde included the covariates PCA1, PCA2 and holes in soil (Table 3), while the most parsimonious model for the bank vole in Son included the covariates PCA1, PCA2, hole under stone and canopy cover.
In the zero-inflation part (Table 4), none of the covariates had any significant effect on occupancy probability. In the count part for the bank vole in Førde (Table 4), PCA1, PCA2 and hole in soil had a positive effect on abundance (Figure 4). This suggests that bank voles are more often caught in areas that contain bilberries, woody shrubs and holes in soil, while caught less often in areas containing herbs, ferns, mosses with few holes in soil.
In the zero-inflation part for the bank vole in Son (Table 4), no covariates had a significant effect on the occupancy. In the count part for the bank vole in Son (Table 4), PCA1 and hole under stone had a positive effect on the abundance (Figure 5), while PCA2 and canopy cover had a negative effect on the abundance. This indicates that relatively open areas with holes in soil, bilberries and woody shrubs experience a higher abundance of bank voles compared to denser and grassy areas.
Table 3: Results from model selection for the zero-inflated Poisson model the bank vole in Førde and Son. x = term included in the model. AICc = Akaike information criteria. ΔAICc = difference in AICc value between the given model and the most parsimonious model. The most parsimonious model (the simplest model within 2 ΔAICc) is marked in bold type.
PCA1 PCA2 Canopy cover Hole in soil Hole under stone Vegetation type AICc ΔAICc
Førde
x x 165.569 22.340
x x x 170.857 27.628
x x x 160.935 17.705
x x x 144.615 1.385
x x x 168.612 25.382
x x x x 143.230 0.000
x x x x x 145.266 2.037
x x x x 145.448 2.219
x x x x x 150.206 6.976
Son
x x 269.988 18.965
x x x 258.892 7.869
x x x 266.399 15.376
x x x 267.579 16.555
x x x 261.436 10.412
x x x x 262.470 11.447
x x x x x 251.023 0.000
x x x x x x Failed to converge
x x x x x Failed to converge
x x x x 252.807 1.784
Table 4: Parameter estimates by the zero-inflation and count parts for the bank vole in Førde and Son.
Estimates are given on a log scale. SE = standard error. z = standard score. p = probability value.
Førde – Zero- inflation part Parameter Estimate SE z p
(Intercept) 0.221 0.472 0.467 0.640
PCA1 -0.013 0.716 -0.018 0.986
PCA2 1.765 1.150 1.535 0.125
Hole under stone -0.251 0.500 -0.503 0.615
Førde – Count part (Intercept) 0.838 0.190 4.418 <0.001
PCA1 1.085 0.331 3.281 0.001
PCA2 0.741 0.354 2.095 0.036
Hole under stone 0.568 0.117 4.862 <0.001
Son – Zero- inflation part (Intercept) -2.862 1.117 -2.562 0.010
PCA1 -0.236 2.226 -0.106 0.916
PCA2 0.976 1.652 0.590 0.555
Canopy cover -0.949 1.061 -0.894 0.371
Hole under stone 0.007 1.314 0.005 0.996
Son – Count part
(Intercept) 2.821 0.058 48.393 <0.001
PCA1 0.370 0.127 2.907 0.004
PCA2 -0.631 0.124 -5.067 <0.001
Canopy cover -0.331 0.066 -5.038 <0.001
Hole under stone 0.226 0.053 4.292 <0.001
3.5 Abundance and occupancy – Wood mouse
The most parsimonious model (in terms of lowest AICc values) for the wood mouse in Førde included the covariates PCA1 and PCA2 (Table 5). The most parsimonious model for the wood mouse in Son included the covariates PCA1, PCA2, hole in soil, and hole under stone.
There was no statistically significant effect by the covariates on neither the count part nor the zero-inflation part for the wood mouse in Førde (Table 6, Figure 4). In the zero-inflation part for the wood mouse in Son (Table 6), PCA1 had a strong negative effect on occupancy.
The other covariates, PCA2 and hole under stone, had no significant effect on occupancy. In the count part (Table 6), PCA1 had no effect on the abundance, while PCA2 had a positive effect. Hole under stone had a negative effect on the abundance (Figure 5). This suggests that the wood mouse prefers grassy areas with fewer holes under stone, compared to areas rich in mosses, ferns and holes under stone.
Table 5: Results from model selection for the zero-inflated Poisson model the wood mouse in Førde and Son. x = term included in the model. AICc = Akaike information criteria. ΔAICc = difference in AICc value between the given model and the most parsimonious model. The most parsimonious model (the simplest model within 2 ΔAICc) is marked in bold type.
PCA1 PCA2 Canopy cover Hole in soil Hole under stone Vegetation type AICc ΔAICc
Førde
x x 197.464 0.000
x x x 201.196 3.732
x x x 200.526 3.062
x x x 198.277 0.813
x x x 207.949 10.485
Son
x x 161.476 5.566
x x x 169.752 13.842
x x x 159.457 3.547
x x x 155.910 0.000
x x x 165.414 9.504
x x x x 166.496 10.587
x x x x 159.609 3.699
x x x x 164.603 8.693
Table 6: Parameter estimates by the zero-inflation and count parts for the wood mouse in Førde and Son. Estimates are given on a log scale. SE = standard error. z = standard score. p = probability value.
Førde – Zero- inflation part Parameter Estimate SE z p
(Intercept) -0.423 0.343 -1.233 0.217
PCA1 0.392 0.598 0.656 0.512
PCA2 0.015 0.547 0.028 0.977
Førde – Count part
(Intercept) 0.916 0.146 6.283 <0.001
PCA1 -0.386 0.269 -1.436 0.151
PCA2 0.214 0.206 1.038 0.299
Son – Zero- inflation part (Intercept) -6.731 3.552 -1.895 0.058
PCA1 3.130 1.600 1.956 0.051
PCA2 2.074 2.056 1.009 0.313
Hole under stone -5.994 4.374 -1.370 0.171
Son – Count part (Intercept) 1.673 0.112 14.927 <0.001
PCA1 -0.215 0.194 -1.106 0.269
PCA2 0.528 0.203 2.597 0.010
Hole under stone -0.431 0.149 -2.903 0.004
3.6 Abundance and occupancy – Common shrew
The most parsimonious model (in terms of lowest AICc values) for the common shrew in Førde included the covariates PCA1, PCA2 and hole in soil (Table 7). The most parsimonious model for the common shrew in Son included the covariates PCA1, PCA2 and hole under stone.
Due to the abundance of the common shrew and its presence at all the stations in both Førde and Son, it was not possible to use zero-inflation to predict the probability of occupancy. In the count models for the common shrew in Førde and Son (Table 8), neither PCA1 nor PCA2 had any statistically significant effect on the abundance. Hole in soil had a slight negative effect on abundance in Førde (Figure 4), but a slight positive effect in Son (Figure 5).
Table 7: Results from model selection for the Poisson models for the common shrew in Førde and Son.
x = term included in the model. AICc = Akaike information criteria. ΔAICc = difference in AICc value between the given model and the most parsimonious model. The most parsimonious model (the
simplest model within 2 ΔAICc) is marked in bold type.
PCA1 PCA2 Canopy cover Hole in soil Hole under stone Vegetation type AICc ΔAICc
Førde
x x 394.735 1.610
x x x 397.034 3.909
x x x 393.133 0.008
x x x 397.000 3.875
x x x 394.697 1.572
x x x x 395.508 2.383
x x x x 395.578 2.453
x x x x 393.125 0.000
x x x x x 395.917 2.792
x x x x x 395.723 2.598
Son
x x 179.350 2.889
x x x 180.952 4.492
x x x 176.461 0.000
x x x 179.705 3.245
x x x 182.145 5.684
x x x x 177.403 0.942
x x x x 178.538 2.078
x x x x 179.330 2.870
Table 8: Parameter estimates by the Poisson (count) models for the common shrew in Førde and Son.
Estimates are given on a log scale. SE = standard error. z = standard score. p = probability value.
Førde – Count model
Parameter Estimate SE z p
(Intercept) 2.281 0.051 44.685 <0.001
PCA1 0.068 0.077 0.875 0.382
PCA2 0.024 0.075 0.323 0.746
Hole in soil -0.153 0.078 -1.964 0.050
Son – Count model
(Intercept) 2.616 0.061 43.162 <0.001
PCA1 -0.095 0.116 -0.820 0.412
PCA2 -0.044 0.124 -0.355 0.723
Hole in soil 0.103 0.043 2.400 0.016
Figure 5: Predicted effect by the different habitat variables on the probability of occupancy (first row) and the abundance (second row) of small mammals in Son, Norway. Note that some species are missing in some of the plots as the habitat variable was not included in the species’ most parsimonious model.
Occupancy = the probability of presence of a given species at a given site (trapping station). Abundance = the number of small mammals captured per station.
3.7 Overview of tick load
The tick load differed both between regions and between species (Figure 6, Figure 7). The tick load was lower in Førde compared to Son. Most of the hosts had few to none ticks. The bank vole had a tick prevalence of 92.6%, the highest of the investigated hosts. The wood mouse had a lower tick prevalence of 72%, while the common shrew had a lower yet tick prevalence of 54.6%.
Figure 6: The distribution of ticks on the small mammal hosts, based on catch data. Note that the scale differs on both the x and y-axis between the areas and host species.
Figure 7: The tick load (number of individual ticks per host) of I. ricinus larvae on the different small mammals by the regions Førde and Son, predicted by years and season by the negative binomial models. Note that scale on the y-axis differs.
3.8 Analysis of tick load – Bank vole
The most parsimonious model (in terms of lowest AIC values) of load of I. ricinus larvae on bank vole in Førde included the factors year, season, and the covariate PCA1 (Table 9). The most parsimonious model for bank vole in Son included the factors year and season, the covariates canopy cover and hole in soil, and the interaction year:season (Table 9).
For the bank vole in Førde (Table 10), PCA1 had a strong positive effect on the tick load, indicating that the tick load is higher in areas rich in woody shrubs compared to areas rich in herbs. The tick load was highest in 2014 (Figure 7), while lowest in 2015. The tick load was also higher in the fall compared to the spring.
For the bank vole in Son (Table 10), there was a significant negative effect of canopy cover and hole in soil on the tick load, indicating a higher tick load in areas that are more open and with holes in soil. As there was interaction between year and season in this model, there was not detected any consistent effect by year and season as main factors (Figure 7). There was little annual variation in the tick load, and the tick load was found to be lower in the spring than the fall, except for 2014 where the tick load was equal in both seasons.
Table 9: Results from model selection for the load of I. ricinus larvae on the bank vole. x = term included in the model. AIC = Akaike information criteria. ΔAIC = difference in AIC value between the given model and the most parsimonious model. The most parsimonious model (the simplest model within 2 ΔAIC) is marked in bold type.
Year Season Vear:Season PCA1 PCA2 Canopy cover Hole under stone Hole in soil Vegetation type AIC ΔAIC
Førde
x x x 270.394 2.682
x x x x 267.844 0.132
x x x x 272.188 4.476
x x x x 271.150 3.438
x x x x 271.160 3.448
x x x x 271.428 3.716
x x x x 271.428 3.716
x x x 268.218 0.506
x x x x x 269.542 1.830
x x x x x 269.052 1.340
x x x x x 268.684 0.972
x x x x x 267.712 0.000
x x x x x 271.030 3.318
Son
x x x 2323.440 5.860
x x x x 2325.420 7.840
x x x x 2325.420 7.840
x x x x 2319.620 2.040
x x x x 2325.420 7.840
x x x x 2320.800 3.220
x x x x 2322.760 5.180
x x x 2319.940 2.360
x x x x x 2320.860 3.280
x x x x x 2321.600 4.020
x x x x x 2321.220 3.640
x x x x x 2317.580 0.000
x x x x x 2320.360 2.780