High early life mortality in free-ranging dogs is largely influenced by humans

Free-ranging dogs are a ubiquitous part of human habitations in many developing countries, leading a life of scavengers dependent on human wastes for survival. The effective management of free-ranging dogs calls for understanding of their population dynamics. Life expectancy at birth and early life mortality are important factors that shape life-histories of mammals. We carried out a five year-long census based study in seven locations of West Bengal, India, to understand the pattern of population growth and factors affecting early life mortality in free-ranging dogs. We observed high rates of mortality, with only ~19% of the 364 pups from 95 observed litters surviving till the reproductive age; 63% of total mortality being human influenced. While living near people increases resource availability for dogs, it also has deep adverse impacts on their population growth, making the dog-human relationship on streets highly complex.

correlated with life expectancy at birth 30,31 . Since a large number of births are observed in free-ranging dogs every year, but population growth does not appear to be very high 13,32 , we speculate that high early life mortality might be an important factor responsible for controlling population growth in free-ranging dogs. In this study we estimate mortality rates in the early life of free-ranging dogs and assess the factors affecting their survival. Since onset of sexual maturity in dogs begins at 6 months of age, pups were tracked from birth to the end of the seventh month of their age or death, whichever was earlier. This ensures that we considered the entire juvenile period, and did not consider sexually mature individuals in estimating early life mortality.

Results
Demography. Pups started to appear in the population in October, with the number of pups and juveniles reaching a peak during the months of December and January. The average litter size at birth was 3.98 ± 2 (median 4, quartiles 3-5, N = 108) and there was no bias towards any particular sex (Male: Female: 1:1.04; Mann Whitney U test: U = 5621, df = 104, 104, P = 0.625). The death/disappearance rate overtook the birth rate from January, and the net number of pups and juveniles began to decrease significantly in the population (Linear regression: R 2 = 0.848, β = − 0.921, P = 0.003, Supplementary Information 1), with the number of newborn pups reaching zero by the end of February (Fig. 1). Only 18.96% of total observed pups reached their 7 months of age. There was a negative correlation between dog age and survival (Linear regression: R 2 = 0.985, β = − 0.992, P < 0.0001, Fig. 2) and this trend was comparable over the 5 years of sampling (Linear regression comparison: F = 1.412, P = 0.254). The highest rate of mortality (30.47%) was observed at the 4 th month of age.
Survival analysis. Survival analysis yielded a plot of survival probabilities of the pups corresponding to each ordered time at which the event of removal occurred (Fig. 3a). The median of the curve corresponds to 82 days (N = 364, 95%CL: 72-92 days). In the Cox mixed-effect model, the variables 'sex' and 'habitat' showed significant   Figure 3b shows the stratified survival curves for males and females separately. The median value of survival time for females was 112.5 days (N = 164, 95% CL: 93-136 days) while that for males was much lower, at 80.5 days (N = 156, 95% CL: 68-92 days). A log-rank test confirmed a significant difference (Chisq = 8.1, df = 1, p = 0.0045) between the two curves, showing that the survival probabilities of males were significantly lower than that of females. Figure 3c shows the stratified survival curves for urban and suburban populations separately. The median value of survival time for the urban population was 95 days (N = 150, 95%CL: 75-149 days) while that of the suburban population was 71 days (N = 214, 95%CL: 59-86 days). Log-rank test confirmed a significant difference (Chisq = 26.7, df = 1, p = 2.33e-07) between the two curves. However the survival rates of males and females remained comparable over the two types of habitat (as sex and habitat together had no effect on mortality) (Supplementary Information 2, Model 6).

Causes of mortality.
Only 32% of the total mortality was by natural causes and for 5% of total mortality there were no reliable records. The remaining 63% of the total mortality was influenced directly (accidents/ poisoning/ beatings) or indirectly (taken away from the population) by humans. Natural causes were the prevalent reason for mortality (52%) at age class 0-1 month when only 3% of the mortality was human influenced. From 1 month of pup age human influenced death or disappearance became prevalent (50%) and this trend remained unchanged until the 5 th month of age. From the 5 th month onwards 70-80% of the total mortality was caused by juveniles going missing (Table 1) which indicated the onset of dispersal from the population.
The cumulative incidence curve ( Fig. 4) showed the proportion of individuals removed from the population for a specific cause as time passed, in presence of the other competing causes. Males and females showed significantly different incidences for categories 2 (taken by human) and 4 (road accidents) ( Table 2; Fig. 4). This suggested that humans preferentially took away male pups and female pups faced more road accidents. Urban and suburban habitats showed significantly different incidences for categories 1 (natural death) and 3 (murdered by human) (Supplementary Information 3), suggesting higher human intervention in suburban habitats.
Simulation model. The simulation model (Supplementary Information 4) generated a highly skewed population at 7 months, with an expected male:female ratio of 1:3.53, while the real population had a much better sex ratio of 1:1.56. However, the trend for the distribution of sexes at each age class was similar in the model and the data (Fig. 5), suggesting that the observed skew was by chance alone. Thus the higher deaths of females due to road accidents was a consequence of selective removal of male pups from the population, and not an intrinsic tendency of females to be more accident prone.

Discussion
Early-life mortality in free-ranging dogs was observed to be very high, with only about 19% of the pups reaching the age of sexual maturity, implying an even lower lifetime survival rate. The pattern of mortality over pup age in months did not vary across the four years of sampling, and can thus be considered to be the actual trend in the population of free-ranging dogs. Highest rate of mortality were observed in the 4 th month of pup age, which is in agreement with an earlier report of 67% mortality by 4 months 33 . This can be attributed to increased mobility and onset of dispersal in juveniles around the 4 th month of age 25,34 . Our data also showed an increased incidence of missing individuals after the 5 th month of age, suggesting increased dispersal. Studies on other canids that live around human settlements have also suggested higher mortality rates for juveniles, as compared to that of adults [35][36][37] .
Humans have been identified as a major cause of mortality for many species including coyotes and foxes, that share space with humans, in both rural and urban environments [38][39][40][41][42] , and free-ranging dogs are no exception 16 . Most of these studies suggest road kills and/or hunting as human-induced causes of deaths of animals living in human dominated landscapes. Our study too revealed an extremely high incidence (62%) of death or disappearance due to human influenced factors. Interestingly, the human influenced deaths increased and remained high after the first month of age, which is the time at which pups are no longer in the protective environment of dens, and social play increases 43,44 . Hence the vulnerability of pups to accidents, and their accessibility to humans who carry them away either for adoption or for a passing fancy increases around this time, leading to increased human influenced mortality, including brutality towards the pups.
The overall higher rates of mortality, both natural and human influenced, observed in urban areas as compared to semi-urban areas is intriguing. Not only were the pups and juveniles more likely to die of natural causes like disease and starvation in the suburbs, where human densities are lower than in cities, but the interference of humans on their lives was also high. The relatively lower engagement of people in the cities with the dogs, and a higher abundance of shelters and resources in cities 16 probably causes this difference in mortality levels. These results resonate with observations on carnivores such as the red fox (Vulpes vulpes), coyote (Canis latrans), Eurasian badger (Meles meles) and raccoon (Procyon lotor) that have not only adapted to urban habitats, but have exploited anthropogenic shelters and food sources to achieve higher population densities than in natural habitats 42 . Thus the dependence of dogs on humans for their survival 16 , and the anthropogenic factors that may influence their population dynamics at later stages of life, like socio-cultural factors 15 , need to be understood in more detail for better management of free-ranging dog populations in urban habitats. People preferentially remove male pups from the population, and our simulation shows that this leads to higher mortality of females due to accidents, due to the sheer skew in the numbers of male and female pups/ juveniles in the resulting population. Thus, humans not only are responsible for a high proportion of the mortality of free-ranging dogs in early life, but also cause a skew in the sex ratio of the cohort that attains sexual maturity. Random sampling from the population of free-ranging dogs of West Bengal have shown that the sex ratio in the dog population does not deviate significantly from 1:1, when both adults and juveniles are considered 32 . Since we observed a skew in the sex ratio at 7 months of age, this suggests that factors like adult mortality and dispersion can lead to stabilization of the sex ratio. Indeed, our simulation suggests that the skew itself can lead to higher   . The localities included residential as well as commercial areas, and were selected randomly based on convenience of sampling within human habitations. Each of those localities had a study area of 1-2 km 2 .
The observer visited the pre-selected area at least twice or if possible thrice a day at random times and walked on all roads and lanes to locate pups. Each visit took 1-2 hours of time depending on the size of the study area. Whenever pups were sighted, a record was made including details like the location of the litter, litter size, sex of the pups and their date of birth (approximate, when actual date of birth was not known). These details were used later to track individual pups. 95 of these litters (having a total of 364 pups) were followed up to the age of 7 months. For each of the cases of pup death (or disappearance) we recorded the causes of death either from direct observations or relying on reports by the local people. We carried out linear regression analyses in StatixtiXL version 1.11 to understand the pattern of mortality in the population and for comparing the data from the different years.
Demography. 108 mother-litter groups of free-ranging dogs were located between 2010 and 2015, of which 95 litters comprising of 364 pups in total were tracked till the 7 th month of age, while for the remaining 13 litters (66 pups) only information at birth was available. Tracking relied on the coat colour and patch positions on their body. All deaths and disappearances were considered under the category of mortality. The various factors that caused death could be broadly divided into natural causes (disease, climatic factors, predation by other dogs or jackals, and injury from fights), human influenced deaths (poisoning, beating, malnutrition, road accidents, taken away from the population) and disappearances for unknown reasons. Pups are often taken away by humans (mostly children), and our qualitative observations show that while some of these pups are adopted as pets, most are abandoned. The abandoned pups mostly die of starvation, unless they are further adopted by other people. Abandoned juveniles typically face aggression from adult dogs if they are not released back in their own groups, and they too rarely survive. Hence pups and juveniles taken by humans were considered to be dead for this analysis as these were effectively removed from the gene pool, unless we saw them return to their natal groups within the span of our observations.

Survival analysis.
In the survival analysis the survival function S(t) is defined by the probability that an individual survives longer than some specified time t. In our case, 'survival' means survival in the population, so all the individuals that were removed from the population by various causes observed were considered as 'not-survived' . The 'survival time' means the age in days of the individual at which the 'event' of his/her removal occurred. The event of death/removal is also typically referred as a 'failure' . Since the individuals were followed until 30 weeks of their age, and no information about the survival time is available after that period, the survival time of the individuals are considered as 'right-censored' . In order to understand the influence of the various factors in shaping the population structure of free-ranging dogs, we carried out a survival analysis 45,46 by the Kaplan-Meir (KM) method, using the 'Survival' package 47,48 of the statistical computation software R 49 . 'Survfit' function of 'KM method' generates the survival curve considering the survival function S(t) as the dependent variable while survival time was considered as the independent variable (Fig. 3a). KM survival estimates for males and females and urban and suburban populations separately (Fig. 3b,c). In order to test the effect of explanatory variables; litter size, sex and habitat, on the survival time of the individuals, we used the Cox proportional hazard (PH) model 50

Causes of mortality.
We analyzed the effect of all types of causes (natural death, human influenced death, disappearance or gone missing from the population and death or disappearance due to unknown reasons) behind pups' mortality, for each age class (0-1 month, 1-2, 2-3 and so on till 7 month of pup age). Later we stratified the 'human influenced death' category into three subcategories such as the death or disappearance records for 'taken by humans' and 'murdered by humans' and 'death in road accident' . We used competing risk analysis to understand the contribution of various factors to mortality in the population. We divided the event of "death" into various categories. Therefore, considering the fact that more than one event is possible, the individual can experience only one of the several different types of events over the study period. These categorical events are 1) 'natural death' , 2) 'taken by human' , 3) 'murdered by human' , 4) 'death in road accident' and 5) 'no information' (includes 'missing' and 'unknown') ( Table 2). We used a cumulative incidence curve (CIC) to model competing risks survival data using the 'cmprsk' package 52 of R (Fig. 4). CIC provides estimates of the marginal probability of an event in the presence of competing events. A modified chi-square statistic 53 , for differences in incidence among the different sexes was used for comparisons.
Simulation model. We were intrigued by the observation that a higher percentage of females died due to road accidents, and wanted to test if females are intrinsically more prone to road accidents, or this observation was the result of a skew caused by selective removal of males. We carried out a simulation using the observed mortality rates applied to an initial population of 1000 dogs having a sex ratio of 1:1. Two kinds of mortality rates were used in the model-taken by humans and others.
Ethical statement. No dogs were harmed during this work. All work reported here was purely observation based, and did not involve handling of dogs in any manner. The methods reported in this paper were approved by the animal ethics committee of IISER Kolkata (approval number: 1385/ac/10/CPCSEA), and in accordance with approved guidelines of animal rights regulations of the Government of India.