Land use gradients drive spatial variation in Lassa fever host communities in Eastern Sierra Leone.

Rodent
Ecology
Zoonosis
Lassa Fever
In Preparation
Landuse Change
Author
Affiliation

David Simons

The Royal Veterinary College

Published

November 2, 2022

Abstract

The natal multimammate mouse (Mastomys natalensis) is the primary reservoir species of the zoonotic infectious disease, Lassa fever (Lassa mammarenavirus). This disease is endemic to Sierra Leone with the highest incidence of human infection reported from the Eastern Province. The spatial occurrence and abundance of this rodent species is regulated by the human environment and biotic interactions within small mammal communities, little is known about these effects even in highly endemic areas of Lassa fever. We conducted a rodent trapping study at four village study sites between 2020-2023, comprising 40,152 trap nights to understand how M. natalensis is distributed across a gradient of landuse types and how this may be influenced by the broad small mammal community structure. We conducted a Bayesian multiple species occupancy model, accounting for imperfect detection, to test the hypothesis that M. natalensis were more likely to occur within human dominated landuse types. We found that M. natalensis occurrence increased from less to more human dominated landscapes. We further found that this effect was not replicated across scales as the probability of occurrence in peri-urban settings was lower than in rural settings. Interactions within the small mammal community appeared to moderate the occurrence of M. natalensis, with the presence of Mus musculus, but not Rattus rattus reducing the probability of occurrence of M. natalensis. This finding may explain prior observations of lower-than-expected human cases of Lassa Fever fever from urban settings in endemic regions. Our findings highlight the spatially heterogeneous distribution of rodent species across landuse gradients with implications for the hazard of Lassa fever outbreaks.

This work is currently in preparation with additional data expected

Motivation

This study was designed to investigate the effect of changes in landuse on the occurrence of different rodent species in a Lassa fever endemic region of Eastern Sierra Leone. While habitat preferences of Mastomys natalensis have been well studied in West Africa to better understand the risk to human populations from Lassa mammarenavirus spillover (Fichet‐Calvet et al. 2010), there is limited understanding of the wider rodent species communities. Biotic interactions between different species will drive occurrence and abundance in different habitats and this study was designed to understand how these species co-exist in these habitats.

We conducted repeated, systematic, rodent trapping in the Eastern province of Sierra Leone, along a landuse gradient to model the association of landuse and occurrence of M. natalensis and more generally small mammal communities. We aimed to investigate the following questions. First, what is the diversity of rodent communities in varied landuse types in Eastern Sierra Leone? Second, how do patterns of landuse affect the occupancy of M. natalensis and other sympatric rodents? Finally, is there evidence that the local spatial distribution of M. natalensis is regulated by biotic interactions with co-occurring species? We expect these analyses to further our understanding of rodent community structures that may explain observed patterns of Lassa fever spillover.

Method

A protocol was developed prior to a pilot trapping session in November 2020. This protocol is archived on the Open Science Framework. All field data is collected using the Open Data Kit (ODK).

Rodent sampling

We conducted rodent trapping surveys between October 2020-February 2023 within and around four village study sites (Baiama; latitude = 7.8375, longitude = -11.2683, Lalehun; latitude = 8.1973, longitude = -11.0803, Lambayama; latitude = 7.8505, longitude = -11.1969, and Seilama; latitude = 8.1224, longitude = -11.1936) in the Lassa fever endemic zone of the Eastern Province of Sierra Leone (Figure 1A.). Surveys were conducted within trapping grids along a landuse gradient of anthropogenic disturbance comprising, forest, agriculture (including fallow and currently in-use areas), and villages (within and outside of permanent structures) (See Supplementary Material 1 for images representative of trapping grid locations). Trapping grids were designated during the initial trapping survey session, one grid was deployed in forest land use, three to four grids were deployed in agricultural land with two grids deployed in village land use. For one village study site, Lambayama, there were no local forest areas, so this landuse type was omitted (Figure 1B-E). Trapping survey sessions within each village occurred four times annually with two trapping surveys in each of the rainy and dry seasons (May to November and December to April, respectively), giving a total of 9 trapping sessions over the study period.

Figure 1. A) Location of village study sites (coloured labels), in Eastern Sierra Leone, major cities are shown with white labels. The inset map shows the location of Sierra Leone in West Africa.

Figure 1. B) The location of standardised trapping grid cells in Baiama. Darker colours refer to more trapnights obtained from a 49m2 grid cell. The minimum number of trap nights within a grid cell was 4 trap nights and maximum was 68.

Figure 1. C) The location of standardised trapping grid cells in Lalehun. Darker colours refer to more trapnights obtained from a 49m2 grid cell. The minimum number of trap nights within a grid cell was 4 trap nights and maximum was 68.

Figure 1. D) The location of standardised trapping grid cells in Lambayama. Darker colours refer to more trapnights obtained from a 49m2 grid cell. The minimum number of trap nights within a grid cell was 4 trap nights and maximum was 60.

Figure 1. E) Location of standardised trapping grid cells in Seilama. Darker colours refer to more trapnights obtained from a 49m2 grid cell. The minimum number of trap nights within a grid cell was 4 trap nights and maximum was 96.

Species classification

Taxonomic identification was performed in the field based on external characteristics using a taxonomic key, including external morphological measurements and characteristics, developed from Kingdon and Monadjem (Kingdon and Happold 2013; Monadjem et al. 2015). Morphological identification alone is unable to distinguish some small-mammal species within the study area at species level. Therefore, molecular identification was performed on whole blood, tissue or dried blood spots. Samples were stored at -20°C until processing, genomic DNA was extracted using QIAGEN DNAeasy kits as per the manufacturers instructions (Supplementary Material 2) (QIAGEN 2023). DNA extracts were amplified using platinum Taq polymerase (Invitrogen) and cytochrome B primers (Bangura et al. 2021). DNA amplification was assessed through gel electrophoreisis with successful amplification products undergoing Sanger sequencing. Attribution of obtained sequences to rodent species was through the BLAST programme comparing NCBI species records for rodent cytochrome B to our sample sequences (Altschul et al. 1990).

Description of rodent detection and species community structure

Adequacy of sampling effort was assessed using species accumulation curves produced for each village study site and each land use type within a village study site, suggesting sufficient effort to detect the expected rodent species within these categories. We constructed detection/non-detection histories for each grid cell and rodent species, assigning “1” when the species was detected and “0” otherwise. We describe species communities at multiple spatial scales. First, all species identified across all village sites and land use types. Second, all species identified within a village study site. Third, all species identified within a single land use type within a single village study site. We report species richness and Shannon diversity at these different spatial scales.

Estimating the effect of land use on species occurrence and richness

To adjust for differential probabilities of detection that may be driven by environmental conditions and trapping effort during the trapping study and between species, we use a Bayesian spatial latent factor multi-species occupancy model that incorporates residual species correlations, imperfect detection and spatial autocorrelation. Variable selection was informed by a pre-specified conceptual model. Models were defined using the sfMsPGOcc function in the spOccupancy package in the R statistical computing language (Doser et al. 2022).

Results

Rodent detection and species community structure

During the study period 530 individuals were detected from 30,364 trap-nights across the four village study sites (1.7% trap-success (TS)). The greatest number of individuals, highest species richness and Shannon diversity values were obtained in the agricultural landuse type, meanwhile, TS was greatest within village landuse settings (i.e., within and outside of permanent structures) (Table 1). The village study site of Seilama had the highest overall TS, species’ richness and Shannon diversity and unlike the three other village study sites had the greatest TS in agricultural landuse. Species richness in Seilama was twice that of the peri-urban village study site (Lambayama) with relatively high Shannon diversity across all landuse types. The sole peri-urban village study site (Lambayama) located within the expanding boundaries of Kenema city, had the lowest species’ richness and Shannon diversity with the majority of rodents detected within the village landuse type.

The most commonly detected rodent species across all village study sites and land use types was M. natalensis (N = 99, 18.7%), followed by Praomys spp. (N = 81, 15.2%), R. rattus (N = 71, 13.3%), M. musculus (N = 57, 10.7%) and Lophuromys sikapusi (N = 47, 8.8%). Mastomys natalensis and R. rattus were detected at all village study sites, although M. natalensis was not detected in forest landuse types (Figure 2.). Conversely, Hybomys planifrons and Gerbilliscus kempii were only detected in a single village study site, with H. planifrons detected in forest landuse and G. kempii in agricultural landuse types. The invasive rodent species M. musculus was only detected in the Lambayama and Seilama village study sites within village landuse types. The detection rate (the number of individuals detected per 1000 TN) varied by species, landuse type and village study site. The greatest rate of detection was for M. musculus in the Lambayama village study site, with the other commensal species M. natalensis and R. rattus having high detection rates across multiple village study sites within village landuse types. Praomys spp. had the highest detection rates in forest and agricultural landuse types.

Figure 2. Detection rate of rodent species in land use type. The detection rate per 1000 TN and the number of detections of each rodent species in the three land use types across all four village study sites are shown. Landuse category is shown on the x-axis, with species name on the y-axis. The plots are panelled by village study site. The three rural village study sites are Baiama, Lalehun and Seilama, the single peri-urban village study site is Lambayama. The absolute number of detections of each species in each landuse type in each village is shown in the label. The colour of the tile corresponds to the detection rate per 1,000 trap nights. Mastomys natalensis was detected at relatively high rates in the village landuse type from all study villages, although at a lower rate in Lambayama where M. musculus was detected at the highest rate. Rattus rattus was detected in the village and agricultural landuse type at all study villages, although at a greater rate in the village landuse type. Rodent species that were not detected in village land use types were generally less frequently detected throughout the study (i.e., Lemniscomys spp., Malacomys spp. and M. minutoides)

Estimating the effect of land use on species occurrence and richness

We drew posterior samples from the most parsimonious Bayesian occupancy model incorporating spatial autocorrelation to estimate the probability of occurrence of a species within a trapping grid cell (see Supplementary Material 8. for model selection). Occurrence terms included landuse type, village study site and scaled terms for distance to the nearest permanent structure and elevation. Detection terms included scaled precipitation and trapping effort (TN) and the fraction of a full moon. We found three patterns of probability of occurrence (\(\psi\)) within a trapping grid cell for the seven included species (Figure 3.), marginal effects of the remaining parameters are shown in Supplementary Material 9. First, M. natalensis, R. rattus and M. musculus had greatest probabilities of occurrence in village landuse types with lower occurrence probabilities in agricultural and forest landuse types. Mastomys natalensis differed from the two commensal, invasive species (R rattus and M musculus) as their probability of occurrence in agricultural settings was generally high. Second, Praomys spp. had high probability of occurrence in forest landuse types with lower probabilities in agricultural and village landuse types. Finally, Crocidura spp, Lophuromys spp and Mus minutoides had their highest probabilities of occurrence in agricultural land use with lower probabilities of occurrence in forest and village landuse. No species showed high probability of occurrence across all land use types, consistent with species being adapted to distinct ecological niches.

Figure 3. Probability of species occurrence across a landuse gradient. The probability of occurrence (\(\\psi\)), within different landuse types, for the seven small mammal species with more than 10 detections is shown. Each point is the median of the predicted probability of occurrence for a species obtained from the posterior distribution at a trapping grid cell. Predictions were obtained for each of the 1,939 trapping grid cells. The y-axis corresponds to the probability of occurrence (\(\\psi\)) at that trapping grid cell stratified by landuse type (x-axis and point colour) for each species. The range of points indicates confidence of the modelled estimate for that landuse type. For example the narrow range of probabilities for Mastomys natalensis in forest landuse types (0-13%) is suggestive that the probability of this species occurring within forest settings to be very low. The wide range of probabilities for some species, such as, Praomys spp. in agricultural landuse types, between 0-90% is suggestive that in some agricultural grid cells the probability of occurrence was very low while in others it was very likely to occur.

The probability of occurrence within a trapping grid cell of some species within the same landuse types showed wide variability for some species. To further explore this we stratified village study sites by human population density into rural and peri-urban (rural <= 500 individuals per 1km2). The probability of occurrence of M natalensis was importantly different, with high probability of occurrence in both agricultural and village landuse settings in rural areas but substantially lower probability in peri-urban village study sites. The same pattern was observed for R. rattus. For the rodent species predicted to have lower probability of occurrence in village landuse settings, namely, Praomys spp, Lophuromys sikapusi and M. minutoides probabilities of occurrence were greater in all landuse types in rural areas compared to peri-urban areas. Shrew species were predicted to have similar probabilities of occurrence in rural and peri-urban areas. In contrast, M. musculus was predicted to have a low probability of occurrence in all landuse types in rural areas, with high values only for village landuse settings in peri-urban areas. The occurrence probabilities for the three commensal species (M. natalensis, R. rattus and M. musculus) suggest that competition may be reducing the occurrence of M. natalensis and R. rattus in the presence of M. musculus as in it’s absence these two species have high occurrence probabilities in village landuse types.

Figure 4. Probability of species occurrence across a land use gradient stratified by rural and peri-urban village study sites. The probability of occurrence (\(\\psi\)), within different land use types, for the seven small mammal species with more than 10 detections is shown. Each point is the median of the predicted probability of occurrence for a species obtained from the posterior distribution at a trapping grid cell. Predictions were obtained for each of the 1,939 trapping grid cells. The y-axis corresponds to the probability of occurrence (\(\\psi\)) at that trapping grid cell stratified by both landuse type and whether the trapping grid cell was located in a rural or peri-urban setting (x-axis and point colour) for each species. Mastomys natalensis shows an important difference in the predicted probability of occurrence in village and agricultural landuse types between rural and peri-urban settings, with a greater than 50% decrease between rural and peri-urban settings. Mus musculus shows an inverse pattern where the predicted probability of occupancy is importantly increased in peri-urban settings and remains very low in rural settings.

Co-occurrence of species

We hypothesised that the local spatial distribution of M. natalensis is regulated by biotic interactions with co-occurring species. Our tests for species correlations supported this for M. natalensis and other species’ of the rodent communities (Figure 5.). We observed that in landuse types where both M. natalensis and M. musculus co-occurred the presence of one species led to a reduction in the probability of occurrence at a grid cell level of the other with a statistically significant very weak negative correlation observed (Spearman’s \(\rho\) = -0.15, p < 0.001). This negative relationship was not observed between M. natalensis and the other commensal, invasive rodent R. rattus, where a strong positive correlation between probabilities of occurrences in both agricultural (\(\rho\) = 0.86, p < 0.001) and village (\(\rho\) = 0.84, p < 0.001) landuse settings was observed. Generally, within village landuse types, high probabilities for the presence of M. musculus was associated with lower probabilities for all other rodent species. This was not replicated for M. natalensis and R. rattus, which did not have a similar effect on the presence of the native rodent species Praomys spp and L. sikapusi. Within agricultural landuse types the probability for co-occurrence between rodent species were high. Generally, across all landuse types, the presence of shrew species’ had a negative correlation with the presence of rodent species’.

Figure 5. Spearman’s rank correlations for the modelled probability of occurrence of species pairs in different landuse types. Positive values (blue shades) represent positive correlation coefficients where an increase in the probability of Species 1 or 2 is associated with an increased probability of Species 1 or 2. Negative values (red shades) represent negative correlation coefficients where an increase in the probability of Species 1 or 2 is associated with a decrease in the probability of Species 1 or 2. Numbers in bold typeface and indicated with an asterisk () are statistically significant at a level of p* <= 0.005. Grey tiles are used where no detections of the species pair in the landuse type were observed in the underlying data informing the model and so were not considered for this analysis to limit inference from limited data.

Data availability and draft manuscript

Data are available in the project’s GitHub repository. The most recent version of the draft currently being shared with co-authors is V3.

Additional discussion

An interactive report has been produced for collaborators, it may take some time to load.

I presented data from the first year of trapping at the 2022 Ecology and Evolution of Infectious Diseases conference, the slides are available and the talk is embedded below.

References

Altschul, S. F., W. Gish, W. Miller, E. W. Myers, and D. J. Lipman. 1990. “Basic Local Alignment Search Tool.” Journal of Molecular Biology 215 (3): 403–10. https://doi.org/10.1016/S0022-2836(05)80360-2.
Bangura, Umaru, Jacob Buanie, Joyce Lamin, Christopher Davis, Gedeon Ngiala Bongo, Michael Dawson, Rashid Ansumana, et al. 2021. “Lassa Virus Circulation in Small Mammal Populations in Bo District, Sierra Leone.” BIOLOGY-BASEL 10 (1). https://doi.org/10.3390/biology10010028.
Doser, Jeffrey W., Andrew O. Finley, Marc Kéry, and Elise F. Zipkin. 2022. spOccupancy: An r Package for Single-Species, Multi-Species, and Integrated Spatial Occupancy Models.” Methods in Ecology and Evolution. https://doi.org/10.1111/2041-210X.13897.
Fichet‐Calvet, Elisabeth, Leen Audenaert, Patrick Barrière, and Erik Verheyen. 2010. “Diversity, Dynamics and Reproduction in a Community of Small Mammals in Upper Guinea, with Emphasis on Pygmy Mice Ecology.” African Journal of Ecology 48 (3): 600–614. https://doi.org/10.1111/j.1365-2028.2009.01144.x.
Kingdon, Jonathan, and David Happold. 2013. “Mammals of Africa” 4.
Monadjem, Ara, Peter J. Taylor, Christiane Denys, and Fenton P. D. Cotterill. 2015. Rodents of Sub-Saharan Africa: A Biogeographic and Taxonomic Synthesis. Berlin, München, Boston: DE GRUYTER. https://doi.org/10.1515/9783110301915.
QIAGEN. 2023. DNeasy Blood & Tissue Kits.” January 20, 2023. https://www.qiagen.com/us/products/discovery-and-translational-research/dna-rna-purification/dna-purification/genomic-dna/dneasy-blood-and-tissue-kit.

Citation

BibTeX citation:
@online{simons2022,
  author = {Simons, David},
  title = {Land Use Gradients Drive Spatial Variation in {Lassa} Fever
    Host Communities in {Eastern} {Sierra} {Leone.}},
  date = {2022-11-02},
  url = {https://www.dsimons.org/lassa/rodent_trapping.html},
  langid = {en},
  abstract = {The natal multimammate mouse (*Mastomys natalensis*) is
    the primary reservoir species of the zoonotic infectious disease,
    Lassa fever (*Lassa mammarenavirus*). This disease is endemic to
    Sierra Leone with the highest incidence of human infection reported
    from the Eastern Province. The spatial occurrence and abundance of
    this rodent species is regulated by the human environment and biotic
    interactions within small mammal communities, little is known about
    these effects even in highly endemic areas of Lassa fever. We
    conducted a rodent trapping study at four village study sites
    between 2020-2023, comprising 40,152 trap nights to understand how
    *M. natalensis* is distributed across a gradient of landuse types
    and how this may be influenced by the broad small mammal community
    structure. We conducted a Bayesian multiple species occupancy model,
    accounting for imperfect detection, to test the hypothesis that *M.
    natalensis* were more likely to occur within human dominated landuse
    types. We found that *M. natalensis* occurrence increased from less
    to more human dominated landscapes. We further found that this
    effect was not replicated across scales as the probability of
    occurrence in peri-urban settings was lower than in rural settings.
    Interactions within the small mammal community appeared to moderate
    the occurrence of *M. natalensis*, with the presence of *Mus
    musculus*, but not *Rattus rattus* reducing the probability of
    occurrence of *M. natalensis*. This finding may explain prior
    observations of lower-than-expected human cases of Lassa Fever fever
    from urban settings in endemic regions. Our findings highlight the
    spatially heterogeneous distribution of rodent species across
    landuse gradients with implications for the hazard of Lassa fever
    outbreaks.}
}
For attribution, please cite this work as:
Simons, David. 2022. “Land Use Gradients Drive Spatial Variation in Lassa Fever Host Communities in Eastern Sierra Leone.” November 2, 2022. https://www.dsimons.org/lassa/rodent_trapping.html.