Hunger Pandemic: Urban Rodents' Boom and Bust During COVID-19

Shortly after the enactment of restrictions aimed at limiting the spread of COVID-19, local governments and public health authorities around the world reported an increased sighting of rats. We combined multi-catch rodent station data, rodent bait stations data, and rodent-related residents’ complaints data to explore the effects that social distancing and lockdown measures might have had on the rodent population within the City of Sydney, Australia. We found that rodent captures, activity, and rodent related residents’ complaints increased during the COVID-19 related lockdown period, followed by a steep decline post-lockdown. We found no changes in the geographical distribution of any of our indices of rodent abundance. We hypothesize that lockdown measures resulted in an increase in rodent activity driven by a reduction in human-derived food resources. This might have increased the mortality rate, triggering a population crash. There is a high chance that the surviving individuals might be rodenticide resistant. It is possible that the onset of COVID-19 might have disrupted commensal rodent populations, with profound implications for the future management of these ubiquitous urban indicator species.


Introduction
Commensal rodents are abundant and pervasive pest species that cause vast damage to infrastructure, destroy crops, and spread diseases throughout the world [1][2][3][4] . In urban areas throughout the world, when a rodent pest becomes overabundant they contaminate food, damage infrastructure, increase risk of re by gnawing on electrical wiring and pose a risk to public health as diseases carriers 2,[5][6][7][8][9][10][11][12][13][14] . Annually, pest rodents cost billions of dollars in losses of food 15,16 . In many cases, pest rodents have become dependent on humans for food and harborage. Changes in human behaviors are known to affect commensal rodent populations 17 . Rodent abundance in cities has been found to be closely linked to socioeconomic conditions, accessibility to structures that offer nesting places, and human-derived food resources 18,19 . For these reasons, rodent control methods commonly include strategies to limit access to public garbage containers and potential nesting places to reduce rodent abundance 20 .
Previous studies have suggested that changes in human behavior following natural catastrophes typically result in changes in rat populations [21][22][23] . For example, large storms such as hurricanes, can often cause large spikes in rodent abundance 22,23 . It is thought that in urban environments, where this pattern has been recorded, the increased rodent abundance might be due to a process of counter-urbanization 24-26 . Shortly after a natural disaster, the urban human population decreases via emigration, which increases the abandonment of idle or degraded infrastructure and thus increases the availability of habitat for pest species 27 . For example, greater levels of abandonment in New Orleans following Hurricane Katrina appeared to cause an increase in the commensal rodent populations 26 . This increase in commensal rodent abundance can potentially increase the risk of zoonotic diseases transmission in the area [10][11][12] .
The appearance and rapid spread of SARS-CoV-2 in the human population during 2020 can arguably be classi ed as a natural disaster and, as such, it is expected to have a similar effect on rodent pest populations around the world. Following the rapid increase in COVID-19 cases at the beginning of 2020, governments around the world started to enact preventative measures aimed at limiting the spread of SARS-CoV-2. Shortly after, reports from local governments and public health authorities around the world implicated the closures of restaurants and food-related venues with increased sightings of rats 28-34 . In some cases, these reports have included rodents engaging in aberrant behaviors, such as rats being active during the day and in close proximity to humans 31,35,36 , as well as rats consuming conspeci cs (e.g. muricide or cannibalism) 31, 37 . A perceived increase in the abundance of rats is su cient to cause intense fear and have severe negative effects on mental health [38][39][40] . These effects on mental health can be potentiated by an increased use of social media during the pandemic 41,42 and the spread of 'panic' across the population 43 . Thus, given the very signi cant implications for public health risk pest rodents pose [7][8][9][38][39][40]44 , it is critical to better understand how pest rodents respond to global events such as COVID-19.
Recently, Parsons, et al 2020 released a study preprint in which they investigated how stay-at-home measures affected rat sightings. They analyzed rat-related public complaints in New York City and Tokyo, Japan and surveyed pest control companies in the United States, Canada, Japan and Poland 45 . They found that rat sightings were geographically speci c, with each city showing different patterns of rat-related public requests either increasing (i.e. Tokyo) or decreasing (i.e. New York City) 45 . Further, they reported a positive association between rat sightings and food service establishments in both cities, with the formation of new rat sighting hotspots during the lockdown period 45 . Parsons, et al 2020 argued that the strong association between rat sightings and cafes or restaurants, as well as the development of new rat sightings hotspots suggests mass movements of rats triggered by lockdown. Moreover, they suggest that this pattern is not observed in Warsaw, Poland due to the lack of clustering of restaurants 45 . An important caveat of the Parsons, et al 2020 study is that it used public perception as a proxy for rodent abundance and movements. Although previous studies have found a reliable relationship between public complaints and rodent abundance 46 , these measurements have not been validated during disruptive periods such as COVID-19. It has been suggested that public perceptions can be affected by cognitive biases potentiated by COVID-19 restrictions and the increased use of social media [41][42][43] Building on anecdotal reports, we investigated how the COVID-19 pandemic restrictions affected pest rodent trapping success, activity, and residents' perceptions within the City of Sydney Local Government Area, New South Wales, Australia. From January 2020, the Australian Federal Government enacted a series of preventative measures to limit the spread of COVID-19. These preventative measures included limits on the number of attendees at social gatherings, mandatory 14 days self-quarantine of all travelers entering Australia, mandatory closure of all non-essential businesses, as well as border closures. Similar to other parts of the world, shortly after these measurements were put in place, different media outlets started to report a seemingly increase in rodent sightings 31-33 . We used data on rodent trapping success, rodent activity and rodent related resident's complaints received by the City of Sydney Council as part of their pest monitoring and control program to determine whether the enactment of COVID-19 preventative measures affected the rodent pest population, as well as the residents' perception of the abundance of rodent pests.

Spatial analysis of Multi-Catch stations, bait stations and residents' complaints
Multi-Catch rodent stations were operated within seven of the eleven Statistical Areas Level 2 (SA2) within the Council, whereas rodent bait stations were operated in all eleven SA2. Rodent related residents' complaints were made in all SA2 (Fig. 3). For all three measures, their mean centers and directional ellipses were approximately equivalent in the pre-lockdown, lockdown and post-lockdown periods (Fig. 4). Signi cant correlations were found between residents' complaints pre-lockdown versus both lockdown (r SP 0.791, P = 0.004) and post-lockdown (r SP 0.709, P = 0.015) periods. Multi-catch rodent stations pre-versus post-lockdown periods (r SP 0.829, P = 0.021), and multi-catch rodent stations versus bait stations pre-lockdown (r SP -0.775, P = 0.041) were also correlated.

Discussion
Overall, we found the general patterns of rodent catches, rodent activity and rodent related residents' complaints to be consistent across measurements (Tables 1, 2 and 3; Fig 1). All three measurements were highest during the lockdown period and lowest post-lockdown (Fig 1). Rodent catches were slowly declining pre-lockdown, while an abrupt spike in catches during lockdown was seen, with an almost immediate crash and a slow recovery during post-lockdown (Table 1, Fig 2A). In contrast, rodent activity seemed to have been on the rise pre-lockdown, with lockdown triggering a steep decline in activity that continued post-lockdown (Table 2, Fig 2B). We found no temporal changes in the number of rodent related complaints received by the council (Table 3, Fig. 4). We found no spatial changes for any of the measurements between pre-lockdown, lockdown and post-lockdown periods (Fig. 4). The spatial distribution of multi-catch rodent stations seemed to have changed during lockdown and returned to the pre-lockdown distribution in the post-lockdown. Multi-catch and bait stations data pre-lockdown appear related, but this relationship seemed to have been disrupted during lockdown and continued to be disrupted post-lockdown.
Despite the higher levels of rodent catches, activity and sighting during lockdown is consistent with rodent population observations after hurricanes 22,23 , whereas the rapid decline at the end of the lockdown period is not. We hypothesize that this might be due to the undeniable temporal and physical differences between a climatic event and a pandemic. Hurricanes are short-lived, with vast physical effects on the landscape, while the COVID-19 pandemic has had a long-term effect on human behavior and no tangible physical consequences in terms of infrastructure. Hurricanes might cause a shift in habitat characteristics, potentially increasing landscape heterogeneity thus encouraging rodent abundance 2,47 . In contrast, social restriction, which involves closure of restaurants, cafes and other food venues 48 , might have only reduced or eliminated human-derived food resources where they have been plentiful before. Commensal rodents show high levels of neophobia 49,50 and taste aversion 51,52 resulting in high levels of 'trap-shyness' 53,54 and low bait acceptance 55,56 . However, a reduction in food resources might have driven animals to engage in "bold" behaviors during foraging whilst in a lower physiological state [57][58][59][60] . These hunger-driven behaviors might explain reports of rats feeding in close proximity to humans 31,35,36 as well cannibalism 31,37 . Ultimately, hunger might have caused these animals to overcome their neophobia and taste aversion, resulting in a decrease in trap shyness and higher bait acceptance, driving an increase in mortality by electrocution or poisoning.
A higher mortality by lethal traps and rodenticide, as well as the decrease in carrying capacity by the reduction of food resources, can explain the steep decline in catches, activity and sightings that we found after the lockdown period 61,62 . Moreover, it is highly unlikely that the peaks in rodent activity, catches and complaints would be due to an increase in the rodent population, given that the lockdown period lasted only 45 days. Even rats, which are known for their proli c reproduction 63,64 , would not be able to reproduce and mature in such a limited time frame. The recovery in the population is therefore expected to be more gradual, like the steady but slow increase in activity and rodent catches we found in the post-lockdown period. Remarkably, rodent related residents' complaints seem to mirror rodent activity and tradability, similarly to what has been reported during periods of no disturbance 46 . This is regardless of the potential for cognitive biases in residents perceptions that have been reported during COVID-19 [41][42][43] .
Given that our data did not cover several years, we were unable to account for natural seasonal cycles in the rodent population. Several studies have shown that urban rodent populations follow a seasonal gradient that re ect both human changes in behaviors and temperature 19,46,65,66 . Colder months seem to trigger lower rodent activity, that then increase towards spring and peaks in summer 46,65,66 . Our pre-lockdown multi-catch station data seems to support this, but not so our bait station data. It is possible that in a subtropical City such as Sydney (average minimum temperature 15.7•C 67 ) the effect of seasonal changes in temperature might not be as strong as that detected in laboratory studies 65,66 and more temperate cities 46 . Additionally, it has been well documented that cities are "heat islands" that experience signi cantly milder winters than surrounding areas 68 . This might be more pronounced in coastal cities like Sydney. Moreover, the expected seasonal changes in rodent activity cannot explain the abrupt increase and decline in catches, nor the abrupt decline in rodent activity during lockdown. Therefore, we argue that the effects we report are solely due to the changes in human behaviors, and unintended effects on the rodent population, elicited by the COVID-19 restrictions.
Interestingly, we found no evidence of spatial changes driven by the lockdown. This supports the ndings Parsons et. al 2020 reported from Warsaw, Poland but contrast with their results from New York City and Tokyo 45 . They suggested that COVID-19 lockdown measures trigger an increase in rodent movement and potential massive migrations, based on the increased association of rats and food service establishments and the formation of new hotspots of rat sightings in New York City 45 . Our data suggest the contrary and based on the well-known site delity pest rodents species show 69 , it is di cult to reconcile that the effects measured are not localized. In the case of Tokyo and New York City, the result may re ect cognitive bias where residents are spending more time at home during social distancing measure are more likely to see rodents in a different area, and thus report and call pest controllers. Parsons et. al 2020, does suggest that this pattern of movement was not recorded in Warsaw, potentially because of the lack of restaurant clusters in that city, a situation that maybe be similar to the one in Sydney.
If the peak in rodent activity was indeed due to abnormal foraging behavior caused by starvation followed by a population crash, it is highly possible that the lockdown acted as a genetic bottleneck. The City of Sydney Council currently has 942 rodent bait stations deployed, versus a maximum of 60 multi-catch rodent stations deployed on any one day. It follows that rodents would be signi cantly more likely to encounter a bait station than a multi-catch trap. Thus, an increase in acceptance of rodenticide baits could be the main cause of a population crash, driven by the reduction of human-derived food resources during lockdown. Similarly to other reported cases, after such a mortality event, the genetic variation within the remaining population could be up to 90% lower than the original population 70 . Australia is currently one of the only countries in the world where rodenticide resistance has not yet been detected. Thus far a limited number of studies have looked into the subject, either through feeding trials 71,72 or genetic detection of reported mutations in the VKORC1 gene responsible for conferring such resistance 73 . Thus, there is a high possibility some individuals inhabiting Australian urban areas may carry mutations, either novel or previously reported 74 , conferring on them rodenticide resistance. It follows that a higher proportion of the lockdown surviving individuals could be genetically resistant to rodenticides 75 . These remaining individuals would then become the "founding gene pool" that would give rise to a genetically distinct and potentially rodenticide resistant population. This "new" population would not take long to recover and repopulate the area 64 , evident by the rapid increase in rodent catches and activity we report post-lockdown. This is a very different interpretation to that offered by Parsons et. al 2020, where the assumption of the mass movement of rats might drive an increase in genetic variation due to interbreeding between not previously connected populations 76,77 .
Although the risks of commensal rodents to be infected or transmit SARS-CoV-2 are low 78 , we know that these animals pose other health risks 2,7-14 . Thus, an increase in rodent-human interactions has the potential to place further strains on health systems around the world. This could become even more pronounced if rodenticide resistance in these pests become widespread 79 . It is possible that the onset of COVID-19 might have disrupted not only human behavior, but also commensal rodent populations, with profound implications for the future management of these species.

Study Area
The City of Sydney (hereafter Council) is the largest city, by population, in Australia, with 240,229 residents 80 . It

Data sourcing
All data used in this research was obtained from the Council. As part of Council's ongoing rodent control operations, pest management contractors have deployed multi-catch rodent stations as well as rodent bait stations across the Council. Rodent captures and activity are recorded regularly, and the data is stored by the Council. The Council also receives residents' complaints through phone calls, emails or through electronic complaint forms found on the Council's website. These complaints are compiled and stored by the Council.

Multi-Catch rodent stations dataset
Flick SMART Multi-Catch rodent station is an internationally patented rodent trap design 81 . This trap consists of a trigger mechanism that kills the animal by an electric current. The trap has a built-in programmable computer with a SIM card, enabling it to communicate via the mobile network when it has been activated. This trap can catch multiple animals (eight maximum), before it needs to be serviced. Block; Flocoumafen: Storm Secure Block and Storm Soft Bait). Baits were randomly rotated at each station, to prevent rodents developing aversion to any bait. Each station was checked regularly (mean: 10.56 Days ± SE: 0.06) and scored according to the rodent activity i.e. low: no bait consumed and no visual signs; or high: bait consumed and visual signs present. Following strategic pest management, bait station with consistent low activity score were relocated. The location for all bait stations was recorded in UTM coordinates to facilitate spatial analyses.

Rodent related residents' complaints dataset
We accessed all rodent related complaints made to the Council from January 2019 to August 2020. Complaints were received through phone calls, emails or through electronic complaint forms found on the Council's website. All complaints contained the date and street address. Identifying information was removed from the complaint dataset, with street address transformed to UTM coordinates to facilitate spatial analyses.

COVID-19 pandemic restrictions
Following the rapid increase in COVID-19 cases at the beginning of 2020, the Australian Federal Government enacted a series of preventative measures to limit the spread of the disease. These preventative measures included limits in the number of attendees at social gatherings, mandatory 14 days self-quarantine of all travelers entering Australia, mandatory closure of all non-essential businesses, as well as border closures. We used the publicly available timeline of these measures 48 , to classify the datasets into three "Periods". Pre-lockdown was de ned as the period prior to March 31 st , 2020; the Lockdown was de ned as the period from the April 1 st to May 15 th , 2020; and Post-lockdown was de ned as the period from May 16 th onwards.

Statistical Analyses
We rst wanted to explore if there was any effect of Covid-19 restrictions overall on pest rodent population and residents' perception within the Council. For this purpose, we performed statistical analyses in R 4.0.2 (R Development Core Team, 2019). The Multi-Catch rodent station dataset was analyzed by Generalized Linear Mixed Models (GLMM) using the functions lmer, glmer and glmer.nb from the package "lme4" version 1.1.23 82 and the function glmmTMB from the package "glmmTMB" version 1.0.2.1 83 for model construction. The rodent bait station dataset was also analyzed by GLMMs with a binomial distribution, using the function glmer from the package "lme4" version 1.1.23 (Bates et al., 2015). The rodent related residents' complaints dataset was analyzed by General Linear Models (GLM) using the functions lm, glm and glm.nb from the package MASS version 7.3.51.6 84 for model construction. With the exception of the models constructed to test the rodent bait station dataset, residual plots and the Pearson's dispersion test were used to identify the best distribution and link for each model 85 . For model selection and re nement we used the function AICc from the package "MuMIn" version 1.43.17 86 . We calculated Δm between models and excluded models with Δm > 2 as having substantially less support 87 .
To generate P-values, Wald Chi-square tests were applied to both GLMMs and a Chi-square analysis of deviance was applied to the GLM using the function Anova from the package "car" version 3.0.9 88 . Post-hoc pairwise comparisons with Tukey adjustments were carried out by the functions emmeans, and pairs from the package "emmeans" version 1.5.0 89 , and the function cld from the package "multcomp" version 1.4.13 90 . Graphs were constructed using package "ggplot2" version 3.3.2 91 .
Each model used aimed to test the effects of COVID-19 restrictions on the number of rodent catches per day (Multi-Catch rodent stations), the level of rodent activity (Rodent bait stations) and residents' perceptions of rodent activity (number of complaints). Period (i.e. Pre-lockdown, Lockdown and Post-lockdown), date, and the interaction between date and period were included as xed factors. To account for intrinsic differences in rodent catches and/or activity based on location, the models aimed to test rodent catches and rodent activity included location as the only random factor. Additionally, in the case of the Multi-Catch rodent station data set, to account for differences in trapping effort due to variable number of traps deployed across the sampling period, the number of active traps was included in the models as an offset.