The use of animals in this study was in accordance with the guidelines of the regulations of experiments on animals and was approved by the animal Ethics and Welfare committee of the Northwest Institute of Plateau Biology, Chinese Academy of Science.

The laboratory experiments were conducted at the Northwest Institute of Plateau Biology, Chinese Academy of Sciences, Xining, China. Root voles were housed singly in clear polyethylene cages (36 × 20 × 17 cm³) with wood shavings and maintained at 20 ± 2 °C under a 12:12 h light: dark cycle. Food and water were availed ad libitum. Twenty voles, six months and older, of each sex from a laboratory colony were divided into two groups: coccidia-infected (hereafter PA+) and parasite-free groups (hereafter PA-). Half of the PA+ and PA- groups were exposed to predator odor (hereafter PR+PA+ or PR+PA-), and the other half to a control odor (hereafter PR-PA+ or PR-PA-). Each of the four treatments involved five voles per sex, and the initial vole body mass of the four treatments did not differ (F_3,36 = 0.187, p = 0.904).

Voles in the PA+ group were once orally administered with 2000 coccidia oocysts suspended in 0.1 mL distilled water on June 3^rd, 2019. Their oocyte levels were comparable to the oocysts per gram in the feces of coccidia-infected root voles studied by Shang et al. (2019). Meanwhile, each vole in the PA- group was treated with a single orogastric gavage dose of 0.1 mL combinatorial anthelmintic comprising 6.25 × 10^-4 mL diclazuril solution (Weierkong, Sichuan) and a 2 mg ivermectin tablet (Weierkong). Combinatorial anthelmintics can effectively expel nematodes, cestodes, and coccidia (Yang et al. 2018).

Our pilot study found that the latency period for coccidia infection in root voles was 6–7 days, and the maximum oocyst output occurred 9–10 days post-infection. Accordingly, we measured nociceptive responses on June 13^th, 2019.

Voles were exposed to predator or control odors on June 4–13^th, 2019. Silver fox, Vulpes vulpes (Linnaeus, 1758), odor was used to stimulate predation risk, while the rabbit odor, Oryctolagus cuniculus f. domesticus (Linnaeus, 1758) was used as control (Wang and Liu 2002a, Bian et al. 2005b). Since the major predators in the study area – Buteo hemilasius Temminck & Schlegel, 1844 or Mustela altaica Pallas, 1811 – are protected animals in China, capturing and collecting fresh feces and urine for 15 consecutive days was challenging. Wang and Liu (2002a, 2002b) found that the silver fox odor could change the behavioral responses of root voles. Thus, our study used silver foxes for predator stimulation instead of natural predators.

Fresh silver fox and rabbit feces and urine were collected in trays under the animal cages daily. Each tray was washed with 500 mL distilled water, and the washing water strained through a filter with a 100 mesh screen (Bian et al. 2005a). Filtered solutions from each animal species collected at different times were thoroughly mixed. At the onset of the laboratory experiment, filter papers infused with predator or control odors were randomly placed in the vole cages three to four times a day between 8:00 am and 11:00 pm. This period was chosen because root voles are primarily diurnal (Sun et al. 1982). Each exposure to predator or control odor lasted 30–60 s.

The nociceptive responses of voles were measured on June 3^rd, 2019, prior to parasitic infection. The initial nociceptive response latency did not differ among voles in the four treatments (F_3,36 = 0.165, p = 0.919). Nociception was measured based on the latency of foot-lifting or licking responses to an aversive thermal stimulus (“hot plate,” CAT.NO.T-91-S, CT, USA). Each measurement was replicated thrice in each individual. The individual was immediately removed from the heated surface following the response display and returned to its cage. If no response was observed within 60 s, the test was terminated, and the vole returned to its cage (Kavaliers and Colwell 1994). In the present study, all voles displayed nociceptive responses within 60 s.

Field experiments were conducted at the Haibei Alpine Meadow Ecosystem Research Station, Menyuan County, approximately 155 km north of Xining, Qinghai Province, China (37°37’N, 101°12’E). The station has an elevation of 3200 m, is surrounded by mountains, and has an average annual temperature and precipitation of -1.6 °C and 560 mm, respectively (Li et al. 2004).

Root vole populations in this area fluctuate annually, usually with relatively low numbers in late winter and spring, increasing throughout the breeding season, and declining after the breeding season; multi-year cycles are weak or absent (Jiang et al. 1991). In this study area, root voles prefer dense vegetation, mainly Elymus nutans (Poaceae), in habitat selection. The average population in the habitat ranged from 217 to 280 voles ha^-1 in autumn, even up to 400 ha^-1 where grazing activities were limited (Bian et al. 1994, Sun et al. 2002). The breeding season typically lasts from April to October. Juveniles reach puberty and breeding age at approximately 50 and 70 days, respectively (Liang et al. 1982). The primary predators in the study area are falcons, Falco tinnunculus Linnaeus, 1758, buzzards, Buteo hemilasius Temminck & Schlegel, 1844, and weasels, Mustela altaica Pallas, 1811.

The field experiments were carried out in four 0.15 ha (50 × 30 m) outdoor enclosures located in an old E. nutans meadow. Major plants included E. nutans, Poa spp., Thalictrum alpinum, and Kobresia humilis. The vegetative cover provided a dense leaf layer, forming a natural refuge for root voles. The enclosures were constructed using galvanized steel panels (1.5 and 0.5 m above and below ground, respectively) but without wire mesh roofs. Further, the enclosures had a series of low panels (~0.3 m high) along the exterior walls every 10 m, allowing terrestrial and avian predators to enter but prevented voles from exiting the enclosures. The vegetation conditions are similar in each enclosure. Each enclosure was equipped with 60 laboratory-made wooden traps (Bian et al. 2011), spaced in 5 × 5 m grids.

Forty-eight voles of each sex, six months or older, from a laboratory colony were used to establish founder populations on October 16, 2017. The voles were divided into two nociception levels according to thermal response latency (“hot plate,” CAT.NO.T-91-S, CT, USA); high response latency group (hereafter group H; 53.92 ± 0.11) and low response latency group (hereafter group L; 49.3 ± 0.09). The nociceptive response latency of group H was significantly higher than group L (F_1,94 = 1011.93, p < 0.001). Earmarked voles from group H were released into two enclosures, while earmarked voles from group L were released into the other two enclosures. Each enclosure contained 12 voles per sex, and each treatment was conducted in duplicate. The density of the founder population (160 voles ha^-1) was in line with natural densities in autumn (Bian et al. 1994, Sun et al. 2002).

Prior to the experiment, all voles were treated with a combinatorial anthelmintic to eliminate parasites and ensure homogeneity. Besides, all enclosures were trapped for two weeks to remove small resident mammals. We also ensured the initial vole body mass did not differ between the enclosures (F_1,94 = 0.004, p = 0.95).

Live trapping began on October 28^th, 2017, after the voles had acclimated to the enclosures for two weeks, and lasted for 141 days (at the end of March 17^th, 2018). Standard capture-record-recapture methods were used throughout the present study. Six trapping sessions were conducted, each consisting of three trapping days. The time interval between two trapping sessions was approximately one month. Each trap was baited with carrots, set between 8:00 am and 5:30 pm, checked every two hours and closed when trapping did not occur. Following capture, the individual was identified and their sex recorded.

We estimated the apparent survival (hereafter “survival”) and recapture probability (hereafter “recapture”) using the standard open population Cormack-Jolly-Seber model (Lebreton et al. 1992) in the MARK program (White and Burnham 1999). The recapture probability was evaluated under the assumption that the individual was alive and in the sample (Cooch and White 2006). The data comprised a capture history of 96 voles in six trapping sessions from October 28^th, 2017, to March 17^th, 2018. First, RELEASE in the MARK program was used to conduct a goodness-of-fit test for the global models, namely ΦTR × T, with both vole survival and recapture dependent on treatment, “TR,” and time, “T.” The goodness-of-fit of the global model was assessed by testing the assumptions of independence and homogeneity of individuals in the enclosures. The goodness-of-fit tests were not significant for voles (tests 2 and 3, RELEASE: χ² = 12.64, df = 18, p = 0.81), suggesting that voles in the enclosures were independent and that the model fits were acceptable. We then used a bootstrap-based goodness-of-fit test to estimate the c-hat value (a variance inflation factor; 1.92), which was adjusted to 1.92 in the global models.

Second, we selected the models as described in our previous study (Yang et al. 2018). Briefly, parsimonious models were selected based on the QAICc values, which allows a compromise between bias and precision when the global model does not fit the data (Anderson et al. 1994) and incorporates a variance inflation factor. Third, we tested the hypothesis that nociception influences vole overwinter survival. For the test, a parsimonious model containing the treatment factor was compared with neighboring populations without the factor, using QAICc values. Subsequently, the model average was estimated from the mean monthly apparent survival probability.

We used the minimum number known alive method to estimate population sizes across trapping sessions in each enclosure. Mark-recapture sampling trials of known populations in the enclosures showed that the minimum number known alive was the best estimate of the actual population size relative to other estimators (Chambers et al. 1999, Bian et al. 2011). The rate of population change for each enclosure was calculated using the following equation: r_t = (1/T) ln (N_t+1/N_t), where N_t is the population density at time t, N_t+1 is the population density during the subsequent trapping session, i.e., at time t +1, while T is the time interval between trapping sessions (Klemola et al. 2002).

The vole population size (Poisson distribution) was analyzed using generalized linear mixed models, with log link functions in the SPSS v. 20 program (IBM, Armonk, NY, USA). Continuous variables were analyzed using a linear model. Data sampled repeatedly were analyzed using the repeated measures method, and all models were simplified by eliminating non-significant (p > 0.05) interactions. Post hoc comparisons of significant effects were computed using the sequential Bonferroni post hoc procedure.

In the analyses of nociceptive responses, treatments were input into the models as fixed factors, and individual IDs were put as the random factors. In the analyses of population change rate and density, treatments and trapping sessions were input as fixed factors to test the primary and interaction effects separately. Meanwhile, enclosures were input as random factors. Since no sex differences were found for any parameter, the data for males and females were pooled during analyses.

We found an effect of treatment on vole nociceptive responses (F_3,34 = 8.89, p < 0.001). Compared with uninfected voles exposed to the control odor (50.70 ± 0.41; range from 49.07 to 53.7), uninfected voles exposed to the predator odor had increased nociceptive latencies (53.65 ± 0.41; range from 51.73 to 55.77), indicating the induction of analgesia (PR-PA- vs. PR+PA-, p < 0.001). However, the response latencies of infected voles after exposure to the predator odor (52.11 ± 0.41; range from 51.40 to 53.23) were lower than those of uninfected voles (PR+PA+ vs. PR+PA-, p < 0.05; Fig. 1). Thus, coccidia infection altered the vole response to predator odor.

Among the various models describing survival, model 1, 2 and 3 were parsimonious models (Table 1). The differences of QAICc value between models 1, 2 and 3 are less than 2 (the differences between model 1 and 2, 3 is 0.11 and 1.71, respectively), thus these models are considered equally valid models. The model 1, 2 and 3 included the effect of time, treatment and interaction between time and treatment, indicating that time and treatment affected vole overwinter survival (Table 1). The average overwinter survival rates in group H and group L were 0.772 ± 0.01 and 0.755 ± 0.01, respectively (Fig. 2).

The population change rate was affected by time (F_5,12 = 10.277, p < 0.05) and the interaction between time and treatment (F_5,12 = 0.785, p < 0.05). However, no effect of treatment alone was found (F_1,12 = 0.06, p = 0.81), indicating that only time and its interaction with the treatment affected population change rate. The average population change rates in the group H and group L were -0.008 ± 0.01 and -0.012 ± 0.01, respectively (Fig. 3). Meanwhile, the population density was affected by time (F_5,12 = 53.203, p < 0.001) and treatment (F_1,12 = 14.735, p < 0.05), but not the interaction between time and treatment (F_5,12 = 0.736, p = 0.611). The group L vole populations demonstrated a higher density than group H (p < 0.05). Although the former had a 23.7 % higher density than the latter during the adaptive phase (the first 13 days), its population declined sharply to a size similar to group H at the end of the experiments (March 2018, Fig. 4).