The ecology of cancer prevalence across species: Cancer prevalence is highest in desert species and high trophic levels

The ecology in which species live and evolve likely affects their health and vulnerability to diseases including cancer. Using 14,267 necropsy records across 244 vertebrate species, we tested if animals in low productivity habitats, with large habitat range, high body temperature and weight-inferred estimates of metabolic rates, and in high trophic levels (from lowest to highest: herbivores, invertivores, primary carnivores, and secondary carnivores) are linked with having increased prevalence of neoplasia. This study found that: (1) habitat productivity negatively correlated with the prevalence of malignancy and neoplasia across tissues, and malignancy and neoplasia in gastrointestinal tissues; (2) inferred metabolic rates negatively correlated with the prevalence of neoplasia; and (3) trophic levels positively correlated with malignancy and neoplasia prevalence in both mammals and non-mammals. However, only the correlations with trophic levels remained significant after Bonferroni corrections for multiple testing. There are several mechanisms that might explain these findings, including the biomagnification of carcinogens in higher trophic levels, as well as tradeoffs between cancer suppression versus reproduction and survival in low productivity environments.


Introduction
Very little is currently understood about why some organisms evolve to be more susceptible to cancer than others, and what tradeoffs constrain the evolution of cancer suppression. Previous work has looked at the association between cancer prevalence and intrinsic organismal characteristics such as body size, longevity, placental invasiveness, and litter size [1][2][3] . Because the ecology of a species determines the selective pressures on that species, we hypothesised that aspects of a species ecology may force tradeoffs between cancer suppression and adaptations to the environment. This study investigates if there are associations between cancer prevalence and habitat productivity, habitat range, temperatureand weight-inferred estimates of metabolic rates, and trophic levels. For trophic levels, we classified species by their primary diet as herbivores (plant eaters), invertivores (invertebrate eaters), primary carnivores (herbivore and invertivore eaters), and secondary carnivores (eaters of primary carnivores).
The selective pressures in the ecology of a species shape all aspects of that species' life history, including its susceptibility to cancer. Life history theory predicts that larger and longer-lived animals may have invested more energy in cancer defence mechanisms, e.g. more protection from mutations, and less energy in reproduction than smaller and shorter-lived animals 4 , and thus have lower cancer prevalence. Cancer researchers, however, predict that larger, longer-lived organisms should have more cancer, because they have more cells that could generate cancer over longer lifespans [5][6][7][8] . The observation that larger, longer-lived species do not seem to get more cancer is called Peto's Paradox 1,2,9 . However, there is a life history trait, litter size, that is positively correlated with cancer prevalence across 29 mammals, indicating a possible tradeoff between offspring quantity and quality of somatic maintenance 1 . Cancer vulnerability should be associated with ecological variables Low productivity habitats may select for species that are more susceptible to cancer through a variety of mechanisms including tradeoffs between investment in somatic maintenance versus reproduction and survival in resource limited environments, selection for lower metabolisms and meat-eating. Habitat productivity can be quantified by the grams of carbon in the form of plant glucose produced per square metre per year 10,11 (Fig. 1), in other words the amount of plant materials in the environment. Low productivity may limit an organism's ability to withstand physiological stresses. Birds from habitats with fewer plants have lower resistance to oxidative and non-oxidative stress in response to cadmium, hydrogen peroxide, paraquat, tunicamycin, and methane methylsulfonate [12][13][14][15] . Additionally, a comparison of 59 temperate bird species from 17 families versus 69 tropical bird species from 29 families shows that species living in lower productivity environments, e.g. temperate birds, have higher metabolic rates than species in higher productivity environments, e.g. tropical birds [12][13][14][15] . Higher metabolic rates are linked with increased oxidative stress, immunological stress 16 , and increased cell proliferation, which can lead to cancer 17,18 . Still, a recent study suggests that high metabolic rates might be cancer protective 19 . Birds are known to have higher metabolic rates than mammals [20][21][22][23][24] , and lower cancer prevalence than mammals [25][26][27][28] .
Higher metabolic rates and larger home ranges (km 2 ) have been also observed in species, within the order Carnivora, eating a higher percentage of meat 42 . Therefore, it would be expected that higher trophic levels and species with a larger habitat range would have higher cancer prevalence across vertebrates.
The hypotheses were that: (1) lower habitat productivity; (2) higher trophic levels; (3) larger habitat range; and (4) higher metabolic rates, are associated with higher malignancy and neoplasia prevalence across tissues across species. It was also predicted that the effect of higher trophic levels might be particularly evident in neoplasms of the gastrointestinal tract.
These hypotheses were tested on data collected from 244 vertebrate species from zoos, aquariums, veterinary hospitals, and online resources.
All animals in the study had been held in human care, including zoos, research facilities, rehabilitation centers or were pets that had been submitted to veterinary clinics. The neoplasia prevalence (including both benign or malignant tumors) or malignancy prevalence was calculated by dividing the number of necropsies that reported neoplasms, or only malignancies respectively, by the total number of necropsies for that species. We calculated tissue-specific neoplasia prevalence or malignancy prevalence by dividing the number of necropsies showing a neoplasm or only malignancy in that tissue by the total number of necropsies for that species. Species were analysed for which there were ≥20 necropsies, as has been done in previous analyses 3 .

Data Filtering
All infant records were excluded. Infancy was determined if a record's age was smaller than or equal to that species' age of infancy (either the weaning age in the life history table, or the average of male and female maturity if there was no weaning age). When there was no recorded age of infancy, the record was considered an infant if it contained any of the following key words: "infant", "fetus", "juvenile", "immature", "adolescent", "hatchling", "subadult", "neonate", "placenta", "newborn", "offspring", "fledgling", "snakelet", "brood", "fry", "fingerling". Records for which there was no information about the location of the neoplasia and free-ranging animal records were excluded from this study.
For reference, worldwide, adult humans have a cancer rate of approximately 0.18 (https://ourworldindata.org/grapher/share-of-deaths-by-cause), but there is not an estimate of benign neoplasia prevalence in humans, so humans were not included in our database.

Tissue specific diagnoses
Gastrointestinal malignancy or neoplasia prevalence of each species consists of the malignancy or neoplasia prevalence, respectively, of the oral cavity, esophagus, stomach, gallbladder, bile duct, liver, pancreas, duodenum, small intestine, and colon. A case of a midabdominal mass adhered to a ferret's stomach and small bowel, and a mass in the pancreatic region of another ferret were included in the gastrointestinal data. The remaining abdominal neoplasms that did not classify as neoplasms of the esophagus, stomach, gallbladder, bile duct, liver, pancreas, duodenum, small intestine, or colon, were included in the non-gastrointestinal neoplasia data.

Habitat data collection
To determine the habitat of the 244 species for which there were cancer records, a search was performed predominantly using the Animal Diversity Web Values of habitat productivity for freshwater, tropical, marine, temperate, and desert habitats from previous studies were found 43,44 (Table 1A). The productivity of freshwater habitats was classified as the average productivity of swamps, marshes, and river estuaries, whereas the productivity of marine habitats was classified as the average productivity of coral reefs, algal beds, and open oceans (Table 1A). The average habitat productivity of a species was the productivity of its habitat if that species lived in a single habitat, and the average productivity of its habitats if it lived in two or more habitats (Table 1B). Habitats were ordered from lowest to highest productivity (Table 1).
To find how many habitats each species lives in, we used the habitat information (supplementary data; Table 1), and categorised each species by their number of habitats (one, two or three or more habitats). Few species lived in four or five different habitats, so these these species were classified as living in three or more habitats.

Diet data collection
When considering the effect of diet on neoplasia prevalence, there are potentially two different effects: the diet that the species has evolved to eat and the diet those animals are actually fed under human care as pets or in zoological institutions. The dietary data for the animals in human care are not available for this study.
To find the natural diet of each of the 244 species in our database in the wild, a search was performed in the Animal Diversity Web (https://animaldiversity.org/), the IUCN red list of threatened species (https://www.iucnredlist.org/), and other resources 45-50 . Each ecosystem can have many trophic levels. For example, marine ecosystems usually have more trophic levels than terrestrial ecosystems 51-53 . Each organism may eat many types of food creating complex interactions that appear as a food web 54 . In order to test hypotheses in all ecosystems, based on the trophic pyramid 55 , and the food pyramid of low to high cancer incidence in humans 40,41 , each species was classified according to their primary diet into four trophic levels: herbivores, invertivores, primary carnivores, and secondary carnivores.
Decomposers, such as certain bacteria and fungi, were not included in these analyses, because there are not malignancy or neoplasia prevalence data for these species. In these classifications, herbivores include species eating primarily plants (fruit, seeds, bark, leaves, flowers, roots) and/or fungi. Invertivores include species eating primarily invertebrate animals such as arthropods, cnidarians, echinoderms, comb jellies, annelids, and/or molluscs.
Eating primarily vertebrates automatically assigned a species as a primary carnivore. To verify this animals were evaluated if their prey were primary carnivores. If the prey were also primary carnivores, these species (i.e. the predators of primary carnivores) were classified as secondary carnivores (Supplementary Data). The diet and habitat of a species may change over time. In this paper, species were analysed according to their primary diet in the wild as reported in recent years. If data on the diet of juveniles in the wild were available, we classified these species according to the diet of adults in those species.

Inferred metabolic rates
To find the metabolic rate of each species we used the following formula M -1/4 • e  [63][64][65] . The mass-specific metabolic rate was estimated as the unweighted average of body temperature. In order to use these body temperature values in "M -1/4 • e -E/kT ", the body temperatures was converted from Celsius to Kelvin.

Statistical Analyses
All were performed analyses in R version 4.0.5 66 using the R packages CAPER 67 , phytools 68 , geiger 69 , tidyverse 70 , and powerAnalysis (https://github.com/cran/powerAnalysis), and performed Phylogenetic Generalised Least Squares (PGLS) regressions to take into account the phylogenetic non-independence among species. To perform a PGLS regression, a phylogenetic tree (phyl file) was made using the NCBI Tree creator (https://www.ncbi.nlm.nih.gov/Taxonomy/CommonTree/wwwcmt.cgi). PGLS analyses have also been previously used to compare neoplasia prevalence and life history variables across mammals 1 . In the analyses where the dependent variable was malignancy or neoplasia prevalence, we weighted the analyses by 1/(square root of the number of necropsies per species) (from Revell 68 ). A Bonferroni corrections was performed to adjust for multiple testing.
The trophic level, average habitat, minimum habitat productivity, and number of habitats were used as categorical variables. Whereas the malignancy prevalence, neoplasia prevalence, average body temperature, "M -1/4 • e -E/kT ", and average habitat productivity (Table 1)

Results
This study found that species from lower productivity habitats have more malignancies, more tumors, more gastrointestinal cancer and gastrointestinal neoplasms in general ( Table 2). These associations were examined in two ways. First, using the minimum productivity habitat occupied by a species and controlling for the species position in the trophic pyramid (Fig. 2). Second, using the average productivity of the habitat(s) a species occupies and controlling for the higher levels of neoplasms in mammals 25 (Fig. 3). These correlations, however, between habitat productivity and malignancy or neoplasia prevalence were not significant after applying Bonferroni corrections for multiple testing (Table 2). There were no significant associations overall between habitat range and malignancy prevalence across tissues or neoplasia prevalence across tissues without controlling for increased cancer in mammals, and after applying a Bonferroni correction (Supp. Fig. 2).
There is evidence of a trend (P-value = 0.01), however, in species living in three or more habitats having higher neoplasia prevalence than species living in one habitat (Supp. Fig. 2B).
Higher trophic levels tend to have higher malignancy and neoplasia prevalence across tissues ( Fig. 4), both within mammals and within non-mammals ( Fig. 5 & Supp. Fig. 1; Table   2). Also, in mammals, higher trophic levels have higher non-gastrointestinal malignancy and neoplasia prevalence. Gastrointestinal neoplasia prevalence in mammals seems to largely be determined by phylogenetic closeness (Lambda = 0.99), potentially between the species in the order Carnivora relative to the other mammalian orders, and there was little evidence that gastrointestinal neoplasia prevalence was influenced by trophic level ( Table 2). In contrast, in non-mammals, higher trophic levels had higher non-gastrointestinal neoplasia prevalence as well as higher gastrointestinal malignancy and neoplasia prevalence  Table 2). Most of these relationships remained statistically significant after a Bonferonni multiple testing correction ( Table 2).  Table 2, and Supp. Fig. 3), however, this association was not statistically significant after applying Bonferroni corrections for multiple testing ( Table 2).

[Figure 6]
Average body temperature was not significantly correlated with malignancy or neoplasia prevalence across tissues without controlling for increased cancer in mammals (Supp. Fig. 4). Also, inferred metabolic rates (M -1/4 • e -E/kT ) 0.1 were lower in primary carnivores than invertivores (Supp. Fig. 5).
The majority of species in lower trophic levels in our database lived in higher productivity habitats, whereas the majority of species in higher trophic levels lived in lower productivity habitats (Supp. Fig. 7). When controlled for trophic level, there was still evidence that low productivity habitats are still associated with more neoplasia (P=0.02) and malignancy (P=0.04), though neither was significant after Bonferroni correction for multiple testing.

Discussion
Cancer prevalence varies across species, for mostly unknown reasons 3,25,26,30 . This study hypothesised that ecological variables would explain some of this variance in cancer prevalence. Evidence was found that malignancy and neoplasia prevalence are higher in species from lower productivity habitats (Table 2). There was also evidence that neoplasia prevalence is higher in species with lower inferred metabolic rates ( Table 2).
The strongest effect found was that malignancy and neoplasia prevalence are higher in higher trophic levels, particularly in secondary carnivores (Figs. 4 & 5, Table 2). A recent study in mammals on a different dataset showed an association between cancer mortality risk and animal content in a species' diet 3 . Here this association was expanded across vertebrates, and discovered that this was partly driven by higher levels of tumors, both benign and malignant, in vertebrate species that are secondary carnivores. It is possible that some of the association between trophic level and cancer prevalence was due to biomagnification of non-degradable toxic chemicals, such as heavy metals, pollutants, microplastics, and pesticides 71 , causing detrimental health effects as their concentration increases in animal tissue at every step higher in the trophic pyramid [31][32][33][34][35][36][37]39 . This study also discovered strikingly low rates of neoplasia prevalence in species that mainly feed on insects or other invertebrates Fig. 1). Trophic levels explain much of the variation in malignancy and neoplasia prevalence across tissues across species in our database (R² as high as 0.31; Table   2).
Caution should be used when applying observations across species to within species; however, these observations of increased cancer prevalence at higher trophic levels were consistent with many studies in humans showing an association between cancer and consumption of animal products [72][73][74][75][76][77] . In humans, consuming products low in photosynthates, such as red meat, animal fat, and/or processed food, leads to the production of N-nitroso compounds, secondary bile acids, hydrogen sulfide, and reactive oxygen species 41,78 . These chemicals increase DNA damage and substitution rates, reduce mucus production, degrade the extracellular matrix, cause inflammation and tumor proliferation 40,41,78-82 . Products abundant in photosynthates, such as fruit and vegetables [83][84][85][86][87][88][89][90] , produce more vitamins 41 , antioxidants 41 , flavonoids 41,91 , and glucosinolates 41 , which reduce DNA damage and inflammation 41 , degrade toxins, limit pathogen growth 41 , tighten cellular junctions 41 , maintain the barrier between intestinal cells 41 , and inhibit tumor growth 40,41,79,92 . Perhaps it is no coincidence that the majority of already known cancer-resistant mammals, such as naked-mole rats, blind mole rats, and elephants, are herbivores [93][94][95] . Despite these observations, the story is likely to be more complex. It is likely that species with meat-based diets have evolved specialised adaptations to consuming meat that may mitigate the negative effects of carnivory.
Invertivores in most of these analyses had the lowest median cancer prevalence and neoplasia prevalence of all trophic levels ( Species at high trophic levels tend to live in many habitats 42,102,103 (Supp. Fig. 6) and in low productivity habitats in particular (Supp. Fig. 7). This larger habitat range of carnivores versus herbivores, may be explained by unpredictability in the abundance and wide dispersal of their prey. Carnivores have relatively more mobile prey than herbivores, and this possibly led to adaptations related to movement in a broader habitat range in carnivores than herbivores 104 . Species living in larger habitat ranges, however, do not have higher malignancy or neoplasia prevalence across tissues than species living in one habitat ( Table 2).
DNA base pair substitution (mutation where a pair of bases changes in the DNA) rate is sometimes modelled as a function of the inferred metabolic rate 56-59,105 , and so metabolic rate may be associated with cancer due to its relationship to mutation rate. Additionally, higher metabolic rates are characteristic of higher trophic levels 42, 106,107 . Still, birds have higher metabolic rates than mammals [20][21][22][23][24] , and lower cancer prevalence than mammals [25][26][27][28] .
This study found that inferred metabolic rates did not differ significantly between trophic levels (Supp. Fig. 5), and inferred metabolic rates showed a negative correlation with neoplasia prevalence across tissues ( Figure 6; P-value = 0.006) which was not significant after applying Bonferroni corrections (Table 2) Why haven't species at lower productivity habitats, higher trophic levels, or with lower metabolic rates evolved better cancer defences than species in higher productivity habitats, lower trophic levels, or higher metabolic rates? Resource scarcity in lower productivity habitats and bioaccumulated toxins in higher trophic levels may limit adaptive responses to cancer defences. Imagine a damaged piece of DNA in two different environments: tropics and deserts. In the tropics, the resources for fixing/maintaining 108,109 the broken DNA are more likely to be available than in the deserts. Some tropical birds with slower life histories are more resistant to oxidative stress than temperate birds with faster life histories [12][13][14]110 . Tropical predators may also have more resources for detoxifying the bioaccumulated toxins. In other words, tropical birds may invest more resources in cellular maintenance, such as cancer protection, than temperate species 12,111 . Bioaccumulated natural (and artificial) toxins in high trophic levels, and loss of heat, waste, and dead matter in every trophic level, can reduce population size in high trophic levels and thus limit adaptive responses to future environmental stressors 112 . Therefore, even though damages in the DNA can potentially happen in many different environments and trophic levels, their chance of repair may be higher in species of higher productivity environments and lower trophic levels.

Limitations & future directions
There were 244 species in this database, however, some data were missing or were not precise for many of those species. For example, the adult weight and body temperatures of 188 and 81 species, respectively, were available. This limits statistical power for identifying relationships between variables and cancer prevalence. Some taxa, such as birds, have too few species per habitat or trophic level (e.g. only 8 primary carnivores and only 2 secondary carnivores within birds) to identify associations between habitat or trophic level and cancer prevalence just within that taxon. Another caveat in our study is that precise data on the productivity of the specific habitats inhabited by each species does not exist, nor are there data on the amount of a species range that is covered by each habitat. The number of grams of carbon produced per square metre per year is an average over the whole year based on the habitat of each species 43,44 (Table 1A &  We tested the effect of environmental factors such as habitat productivity, habitat range, and diet, on malignancy and neoplasia prevalence across the tree of life. There may be additional factors, however, within these or neighboring habitats that affect malignancy and neoplasia prevalence across species, that were not tested. The use of hormonal contraceptives in human managed populations may contribute to higher cancer prevalence in some of our species, particularly in females. When testing for a sex bias in cancer risk in the order Carnivora, however, Vincze et al. 3 did not find any significant sex bias in cancer risk, indicating that hormonal contraceptives do not explain the increased cancer prevalence in the populations of Carnivora under human care that were examined in that study. A study from 1977 found that the majority of neoplasms in mammals were found in the lung 25 , whereas the majority of neoplasms in birds and reptiles were lymphosarcomas 25 . Dogs exposed to passive smoking and air pollution may develop cancer 114,115 . Numerous other carcinogens in the environment and diet are known to affect malignancy and neoplasia prevalence in several species 38,39,[116][117][118] , so it would be useful to know the impact of these carcinogens on malignancy and neoplasia across species. More environmental toxins may be found in the tissues of animals living in lower productivity habitats and/or in higher levels of the food pyramid 32,39,119 , which could explain the observed higher malignancy and neoplasia prevalence in lower productivity habitats (Fig. 2 & 3) as well as the higher cancer prevalence across tissues in higher trophic levels ( Fig. 4 & 5; Supp. Fig. 1; Table 2).
We expected what type of food different species would predominantly consume rather than what they occasionally would consume to affect their evolution and development the most. Therefore, we classified species in this database according to their primary diet.
Malignancy and neoplasia prevalence, however, may also depend on whether species are monotrophs or polytrophs (e.g., generalists). Therefore, malignancy and neoplasia prevalence could be compared with the diversity of food that an organism eats, although we do not have data on the whole range of foods that every species eats or a clear biological hypothesis about which group, monotrophs or polytrophs, may have higher cancer prevalence.
Future studies can test whether the correlations between higher trophic levels and cancer prevalence (  Table 2) are also present when comparing carnivorous physiological traits and cancer prevalence. In humans, the risk of developing breast and colorectal cancer is higher in people that mainly work at night 49,120 . Nocturnal and crepuscular species may have higher cancer prevalence than diurnal species, and nocturnal and crepuscular living is more typical of carnivores [121][122][123] . It may be that this association between carnivory and nocturnal or crepuscular living is partially responsible for the relationship we found between higher trophic levels and higher cancer prevalence (Fig. 4 &   5; Supp. Fig. 1; Table 2). The association between carnivores and cancer prevalence may also depend on the microbiome 124 , though little is currently known about the microbiome of most of the species in this dataset.

Conclusion
To the extent that cancer has been an important selective factor on the survival and reproduction of a species, it would be expected that natural selection blunts any associations between ecological factors and cancer prevalence. Tradeoffs in responding to different selective pressures in an organism's ecology, however, may constrain the ability of a species to effectively suppress the development of neoplasia. This study found that malignancy and neoplasia prevalence across tissues was correlated with trophic levels across vertebrates, even when applying Bonferroni corrections for multiple testing ( Table 2). Future studies should distinguish whether this pattern exists due to bioaccumulation of toxins in high trophic levels, due to the consumption of meat per se, or both. An additional hypothesis for potential testing would be the effect of insect-derived chemicals, such as chitin, on the low cancer incidence in invertivores. Species with high inferred metabolic rates and in high productivity habitats have lower neoplasia prevalence but these relationships are not significant after Bonferroni corrections, and the reasons for those potential relationships remain unknown. These results suggest that species with higher metabolic rates and in higher productivity habitats might have more somatic resources with which to invest in cancer defences. A complete theory of metabolic ecology would help distinguish between these possible mechanisms and this knowledge could possibly be used to decrease the burden of cancer for animals managed in human care, free ranging animals, and humans.      Table 2 for statistical analyses.    Tables   Table 1A. Net primary production (in grams of carbon produced per square meter per year) from previous measures of biomass productivity 43,44 .

Producers
Net biomass productivity (gC/(m² • year)) = the mass of photosynthate  tests, the most conservative Bonferroni correction requires a P-value < 0.0019 in order to be considered statistically significant, which are highlighted with a * next to the P-value. In the 1st P-value column the P-value are reported for the first variable (which we call variable A in the multivariate analysis), in the 2nd P-value column the P-value is reported for variable B, in the F-statistics column we report the F-statistics of variable A, and in the "Type of Association" column we report whether there is a positive (+) or negative (-) association between the independent variable A and neoplasm or malignancy prevalence. If the independent variable is categorical, the sign (+ or -) of the majority of between-group comparisons is reported. † indicates that the R² value was not available given that the categorical independent variable only included two groups (herbivores and primary carnivores).

Competing interests
We declare we do not have any conflicts of interest.  T e m p e r a te (s u n ra y s h it th e E a rt h a t a n a n g le , a ir c o o ls a n d d e s c e n d s)

Supplementary Figures
T e m p e r a te (s u n ra y s h it th e E a rt h a t a n a n g le , a ir c o o ls a n d d e s c e n d s ) T e P o la r T e m p e r a te (s u n ra y s h it th e E a rt h a t a n a n g le , a ir c o o ls a n d d e s c e n d s) T e m p e r a te (s u n ra y s h it th e E a rt h a t a n a n g le , a ir c o o ls a n d d e s c e n d s ) ht tha n tem per ate env iron m ent s) en vi ro nm en ts ) (le ss su nl ig ht th an te m pe ra te (le ss su nl ig ht th an te m pe ra te en vi ro nm en ts ) (le ss w at er th an th e tr op ic s) (le ss w at er th an th e tr o p ic s) (s un ra ys hi t th e Ea rt h at an an gl e, le ss su nl ig ht th an th e tr op ic s) (les s sun ligh t tha n tem per ate env iron m ent s) (s u n ra ys h it th e Ea rt h at an an gl e, le ss su n lig h t th an th e tr o p ic s) (s un ra ys hi t th e Ea rt h di re ct ly , m oi st ur e ri se s, co ol s, an d is re le as ed as ra in ) Minimum productivity habitat of a species (gC produced per m² per year)