Introduction

For nearly a century, caloric restriction has been known to extend lifespan and delay age-associated pathology in laboratory animals [1,2,3,4,5]. More recently, in both animals and humans, a variety of alternative “anti-aging” diet modalities have been described which are providing new mechanistic insights and potential clinical applications [6]. These diets include intermittent fasting [7, 8], fasting mimicking diets [9, 10], ketogenic diets [11,12,13,14,15], protein or essential amino acid restriction [16, 17], and time-restricted feeding [18,19,20].

These diets have been most extensively studied in rodents in controlled laboratory settings, due to ease of administering diets on a specific schedule and an enhanced ability to tease apart the mechanisms through which they act. Time-restricted feeding studies in rodents suggest improvements in several metabolic parameters, including glucose and insulin homeostasis, energy expenditure, hepatic pathology, resistance to different obesogenic diets, and improved circadian rhythm maintenance during aging [21,22,23]. In one study, mice who experienced time-restricted feeding demonstrated an 11% extension in lifespan [18]. Additionally, several studies demonstrate that caloric restriction and intermittent fasting play a protective role in maintaining and enhancing cognitive function, including memory and spatial learning [24,25,26,27,28]. It remains unclear, however, whether these benefits in laboratory animals are generally due to reduced caloric intake or meal frequency or both [6].

Despite mainstream popularization in humans of several of these diets, the beneficial health effects of time-restricted feeding outside of a laboratory setting are less clear. In some human studies, only mild improvements in body composition and cardiovascular risk factors were detected [29], even when subjects also reduced their daily caloric intake [30]. In other studies, detrimental effects on glucose homeostasis were observed with time-restricted feeding [31]. Finally, while some studies have found potential cognitive benefits, especially for memory in older adults [32, 33], other studies have shown no effect of fasting on cognition [34, 35].

Companion dogs provide a potentially powerful animal model to elucidate the relationship between diet and age-related health outcomes [36]. Having co-evolved alongside people for thousands of years [37], companion dogs share human environments, experience similar diseases, and receive similar medical care. Once-daily feeding in dogs serves as a natural model for the intermittent fasting/time-restricted feeding protocols currently being studied both in preclinical rodent models and in human trials [38].

The Dog Aging Project is a large-scale research initiative following thousands of companion dogs over their lifetimes to better understand how biology, lifestyle, and environment impact healthy aging [39, 40]. Participating owners report annually on a variety of aspects related to their dog’s diet, primary and secondary activities, social and physical environments, medications, and health conditions. In the current study, we used cross-sectional data collected in the first year of the Dog Aging Project to ask if feeding frequency is associated with cognitive function and health conditions. Specifically, we hypothesized that dogs fed once-a-day would display lower rates of physical health issues and better cognitive scores compared to dogs fed more frequently.

Methods

Subjects

All dogs had been recruited to join the Dog Aging Project (DAP) via mainstream media, social media, or word of mouth. Their owners then completed the relevant online surveys between December 26, 2019 and December 31, 2020 [41]. Study data were collected and managed using REDCap electronic data capture tools hosted through the DAP [42, 43]. REDCap (Research Electronic Data Capture) is a secure, web-based software platform designed to support data capture for research studies, providing (1) an intuitive interface for validated data capture, (2) audit trails for tracking data manipulation and export procedures, (3) automated export procedures for seamless data downloads to common statistical packages, and (4) procedures for data integration and interoperability with external sources.

Instruments

The first survey that participants completed was the Health and Life Experience Survey (HLES), which collects information regarding dog demographics, physical activity, environment, dog behavior, diet, medications and preventives, health status, and owner demographics. In the current investigation, we were principally interested in feeding frequency and health status, and we identified a priori health conditions that could plausibly be affected by feeding frequency.

After completing HLES, all participants were offered the opportunity to complete the Canine Social and Learned Behavior Survey (CSLB), which measures cognitive function. The CSLB, renamed by the DAP, is the same as the Canine Cognitive Dysfunction Rating Scale (CCDR) [44], with only a handful of minor wording changes. The CCDR has been validated as a clinical instrument to detect cognitive dysfunction over a specific threshold. In this work, we use CSLB as a quantitative score for detecting deterioration towards cognitive dysfunction and acknowledge that it is not yet validated for this purpose. The Canine Cognitive Dysfunction Rating Scale was presented to participants as the Canine Social and Learned Behavior Survey to avoid the negative connotations of the phrase “cognitive dysfunction.” This instrument asks owners to indicate the frequency with which their dogs exhibit behaviors indicative of dementia (i.e., disengagement from social activity; difficulty in navigation, searching, and recognition). Based on owner responses, dogs receive a score that ranges from 16 to 80, where higher scores are indicative of worse cognitive function (i.e., more cognitive dysfunction).

During the study time period, 27,541 DAP participants completed HLES, and 20,096 DAP participants completed CSLB.

Ethical note

The University of Washington IRB deemed that recruitment of dog owners for the Dog Aging Project, and the administration and content of the DAP Health and Life Experience Survey (HLES), is human subjects research that qualifies for Category 2 exempt status (IRB ID no. 5988, effective 10/30/2018). No interactions between researchers and privately owned dogs occurred; therefore, IACUC oversight was not required.

Inclusion/exclusion criteria

Given that meal frequency is adjusted as puppies mature, we specified age of inclusion as 1 ≤ age < 18 years for all health outcomes. For the CSLB outcome, we specified age of inclusion as 6 ≤ age < 18 years, as 6 years is the youngest age indicated in the literature where signs of cognitive decline can start to appear in dogs [45,46,47]. Less than 5% of dogs in the DAP are intact, and these dogs were excluded, as well as dogs (< 1%) whose owners reported that their diet was “not at all consistent.” Thus, in our final sample, all dogs were spayed or neutered due to exclusion criteria, and slightly less than half of dogs were male. About one-fifth of dogs received daily or more frequent omega-3 or other fatty acid supplementation in their diet.

We studied health conditions that were reported in the nine broad categories on HLES that could plausibly be affected by feeding frequency: dental or oral disease, skin disorders, orthopedic disorders, gastrointestinal disorders, cancer or tumors, kidney or urinary disorders, cardiac disorders, neurological disorders, and liver or pancreas disorders. The other broad categories of health conditions reported in HLES were not analyzed because they were either based on temporary situational and/or environmental factors and thus unlikely to be associated with feeding frequency (e.g., trauma, ingesting toxic substances, infectious or parasitic disorders), were infrequently reported and thus had a very small sample size (less than 3.5% of the total sample; e.g., respiratory disorders, endocrine disorders, reproductive system disorders, immune-mediated disorders, and hematopoietic disorders), or there was no compelling rationale as to why feeding frequency would affect them (e.g., ear, nose, and throat disorders, eye disorders).

For the health categories examined in this investigation, all participants were assigned a binary score (affected/unaffected). Dogs were considered “affected” if their owner reported them to have at least one relevant condition within a given category. However, we did not consider any congenital health outcomes as affected: since animals were born with these conditions, their feeding regimen was by definition instituted after onset and could therefore not have affected the development of the condition. Similarly, disorders linked to transient situational factors, including infectious diseases and trauma, were not considered as affected, as the circumstantial nature of these instances made them unlikely to be affected by feeding frequency.

See Supplementary Information 1 for details of all inclusion/exclusion criteria, as well as the specific conditions that qualified a dog as affected within the dental or oral disease, skin disorders, orthopedic disorders, gastrointestinal disorders, cancer or tumors, kidney or urinary disorders, cardiac disorders, neurological disorders, and liver or pancreas disorders categories.

After applying exclusion criteria, the final sample consisted of responses from 24,238 HLES surveys and 10,474 CSLB surveys. The CSLB was always completed at least 1 week after completion of HLES. Most participants in the final sample (88%) completed CSLB within 3 months of completing HLES and always within a year (range: 7 to 352 days, average: 47 days).

Explanatory variables

We analyzed feeding frequency as a binary exposure, comparing dogs fed once-daily to dogs fed more frequently. Specifically, owners were asked “How many times per day is your dog fed?” The dogs of owners who answered “Once” were sorted into the once-daily category, whereas the dogs of owners who answered “Twice”, “Three or more,” or “Free-fed (filling up bowl when empty or always having food available)” were sorted into the fed-more-frequently category. In all analyses, 8% of the total sample was fed once daily (Tables 1 and 2).

Table 1 Characteristics of 10,474 dogs in CSLB analysis (76 pure breeds included)
Full size table
Table 2 Characteristics of 24,238 dogs in analysis of health outcomes (100 pure breeds included)
Full size table

In our analyses, we adjusted for sex (spayed female or neutered male), age, breed for purebred dogs, and body size (as captured by weight) for mixed breed dogs. We also adjusted for whether the owner reported daily omega-3 (or other fatty acid) diet supplementation for all analyses except for dental/oral disorders and liver/pancreas disorders, as there is evidence in the literature that fatty acids can have beneficial effects on cognitive function, skin, cardiac, gastrointestinal, renal, orthopedic, and neoplastic outcomes [48,49,50,51,52,53,54,55,56]. For analysis of CSLB, we additionally adjusted for two factors that are thought to affect cognitive function: physical activity level [57] and whether the dog has a history of training (according to the dog’s primary or secondary activity indicated by the owner; e.g., show dogs, service dogs, and dogs trained for field trials vs. pets/companion dogs; see Supplementary Information 2 for full details) [58].

We adjusted for the breed of purebred dogs as a categorical variable. After inspecting the distribution of weight by breed, we subdivided standard poodles into two breeds for the analysis, large poodles (weight ≥ 13.6 kg (30 lb)) and small poodles (weight < 13.6 kg (30 lb)). Although there are over 200 breeds represented in DAP data, for each analysis, we only included breeds that had at least one exposed and one unexposed dog because breeds without variance in the exposure cannot inform the exposure-outcome association. We also restricted our analyses to breeds with at least 10 dogs meeting inclusion criteria. These restrictions reduced the number of breeds to 76 breeds for the CSLB analysis and 100 breeds for the analyses of health outcomes.

Statistical methods

All statistical analyses were carried out in R v.4.0.3 [59]. Age was flexibly modeled using natural splines with interior knots at 7, 10, and 14 years for CSLB analysis and interior knots at 2, 7, and 13 years for health outcomes [60]. Weight was similarly modeled using natural splines with interior knots at 14, 48, and 79 lbs. In each instance, interior knots are at approximately the 10th, 50th, and 90th percentile of each variable. We explored more elaborate adjustment models (e.g., 4 or 5 interior knots), but these were not supported by metrics such as AIC, and examination of some results suggested overfitting.

To adjust for physical activity, we performed principal component analysis on three HLES-reported activity variables: lifestyle activity level (reported as not active, moderately active, or very active over the past year), average activity intensity level (reported as low: walking, medium: jogging, or vigorous: sprinting, such as playing fetch or Frisbee), and average daily time spent physically active (reported in hours and minutes). Parallel analysis recommended retaining one principal component. This principal component captured 52% of the variance, and we used the loadings onto the first principal component as a physical activity score (PA-score). We adjusted for PA-score using natural splines with interior knots at approximately the 10th, 50th, and 90th percentiles.

Mixed breed dogs were included as a separate category of breed, and we adjusted for body size, as measured by weight, for mixed breed dogs only by constructing a variable weight*MB where weight is the dog’s weight and MB = 1 for mixed breed dogs and MB = 0 for purebred dogs. This is analogous to grouping mixed breed dogs by weight and including each group as a breed, except that our approach uses continuous weight information.

We used linear regression for analysis of CSLB and logistic regression for analysis of all health outcomes. For linear regression, the large number of parameters in the model due to 76 breeds does not cause statistical issues. However, large models are problematic for logistic regression when using conventional maximum likelihood model-fitting [61]. Therefore, we fit the logistic models using a conditional likelihood, where the conditioning was on the breed categories. This approach allowed breed to be in the model without necessitating the estimation of 100 breed parameters. Due to the large size of the dataset, maximizing the exact conditional likelihood was not computationally feasible, and we used the Efron approximation (reported in Supplementary Information 3). We used robust standard error estimates for all regression analyses. All hypothesis tests were two-sided and we did not adjust for multiple comparisons.

Secondary analyses

For analysis of CSLB, we performed a sensitivity analysis excluding 2% of dogs with scores < 20 (n = 10,288), as these very low scores are implausible and highly suggestive that the owner did not recognize that some survey questions reflected a bidirectional rather than a unidirectional scale.

For analysis of the nine categories of health conditions, we repeated the analysis using only mixed-breed dogs and dogs from the 10 most common breeds, fitting the model with ordinary logistic regression. The 10 most common breeds were Australian shepherd, beagle, Border collie, Chihuahua, dachshund, German shepherd, golden retriever, Labrador retriever, poodles (large), and pugs (Table 2). We also treated these analyses as secondary analyses to assess the robustness of our findings, including robustness given use of the Efron approximation for the primary analyses.

For all ten outcomes, we repeated our primary analyses but analyzed feeding frequency as a four-level unordered categorical factor.

Results

In the CSLB analysis (n = 10,474 dogs), 56% of dogs were mixed breed, and the remaining dogs belonged to 76 breeds (Table 1; Supplementary Information 4). Dogs fed once per day had, on average, a 0.62 point lower CSLB score than dogs fed more than once per day (95%: 0.27, 0.97; p < 0.001), adjusting for age, sex, weight (for mixed breed dogs), breed (for purebred dogs), cognitive activity, physical activity level, and fatty acid supplementation (Fig. 1; see Supplementary Information 3 for the full model). This effect size of 0.62 points is roughly the same difference in mean CSLB score between 11- and 7-year-old dogs. Results were very similar in the sensitivity analysis excluding 2% of dogs with very low CSLB scores (n = 10,288, 0.50 point lower CSLB score (95% CI: 0.20, 0.80), p = 0.0012).

Fig. 1
figure 1

Scatterplot of CSLB Scores vs. age with superimposed trend lines. Darker points represent dogs fed once daily and other points represent dogs fed more frequently. Trend lines were constructed separately for the two groups using natural splines. Dogs fed once daily have slightly lower mean CSLB score at all ages (6 ≤ age < 18 years)

Full size image

In the analyses of health conditions (n = 24,238 dogs), 55% of dogs were mixed breed, and the remaining dogs belonged to 100 breeds (Table 2; Supplementary Information 5). For five of the nine categories of health conditions analyzed, we found evidence that being fed once per day vs. more often is associated with lower odds of having the health condition (Fig. 2; Table 3). Adjusted odds ratios were less than one and statistically significant for gastrointestinal, dental/oral, orthopedic, kidney/urinary, and liver/pancreas health conditions. Adjusted odds ratios were also less than one for the remaining four categories of health conditions (cardiac, skin, neurological, cancer), but not statistically significant (see Supplementary Information 3 for detailed model results). Results were similar in secondary analyses including only mixed breed dogs and dogs from the ten most common breeds (see Supplementary Information 3 for full report).

Fig. 2
figure 2

Summary of results for analysis of health conditions. Circles represent point estimates of adjusted odds ratios, with darker circles indicating statistically significant results. Bars represent 95% confidence intervals. Odds ratios less than 1 indicate lower odds of the outcome among dogs fed once daily; odds ratios greater than 1 indicate higher odds of the outcome among dogs fed once daily

Full size image
Table 3 Estimated odds ratios of specific health condition for dogs fed once per day compared to more frequently, adjusted for sex, age, breed for purebred dogs, and body size (as captured by weight) for mixed breed dogs. All analyses except liver/pancreas and dental/oral are also adjusted for omega-3 (or other fatty acid) supplementation
Full size table

In the secondary analyses treating feeding frequency as a four-level unordered categorical variable, results were very consistent with the primary analysis for dogs fed twice-daily and dogs fed three or more times daily (see Supplementary Information 6). These dogs had higher CSLB scores and higher odds of a health condition compared to dogs fed once daily. Results were somewhat more ambiguous for free-fed dogs. In particular, for dental/oral and kidney/urinary conditions, free-fed dogs in our data were less likely to have a health condition compared to dogs fed once daily, although results were statistically insignificant.

Discussion

Using observational data from the Dog Aging Project, this is the largest study to date of feeding frequency conducted in companion dogs. We found that adult dogs fed once daily have better average cognitive scores and are less likely to have gastrointestinal, dental/oral, orthopedic, kidney/urinary, and liver/pancreas health conditions than dogs fed more frequently. While it is important to note that this study does not demonstrate causality, our observations are consistent with prior work in laboratory mice and observational studies in humans [62] suggesting that diets that restrict the timing of feeding are associated with better cognitive function and physical health.

In addition to being able to observe the animals in a naturalistic versus laboratory setting, one of the major strengths of our investigation is the large sample size of dogs included (CSLB assessment: 10,474 dogs; all other health conditions: 24,238 dogs). Furthermore, our statistical methods used flexible adjustment of continuous covariates (age, weight, and physical activity), reducing the possibility of residual confounding by these factors.

The key limitation of this work is that it is a cross-sectional analysis. The analysis implicitly assumes that each dog’s feeding frequency has been the same over some meaningful period of the dog’s life. Importantly, we cannot rule out the possibility that dog owners shifted to more frequent feeding in response to health conditions, and observed associations are due in whole or part to reverse causality. This is a particular concern for gastrointestinal conditions and liver conditions, which are the two categories of health conditions with the strongest observed associations. Such a shift would not be captured in the current dataset as owners reported on their dog’s current feeding frequency but did not provide information on feeding frequency history. In the future, we will gather this information through annual “snapshots” since participants complete HLES each year. As the Dog Aging Project accrues longitudinal data over the next several years, investigations can compare dogs with different feeding frequencies who do not have a health condition and prospectively examine incidence of the condition. Such analyses can provide stronger evidence for a causal effect of feeding frequency on health.

It is plausible that once-daily feeding tends to result in lower overall caloric intake compared to more frequent feeding. Since we do not have data on caloric intake, we cannot analyze whether observed associations are mediated by caloric intake or reflect a possible effect of feeding frequency through other pathways. Caloric restriction has been previously reported to extend lifespan and improve health in Labrador retrievers maintained in a laboratory setting [5], although it has not yet been studied in other breeds of dogs or in companion dogs.

This study has other limitations. All data are owner-reported and thus subject to error in recall and interpretation. However, while a given owner’s responses on their dog’s cognitive function, physical activity, and other health conditions might reflect individual differences in interpretation and reporting errors, it is unlikely that these would generate the specific associations we observed. We were also unable to account for dogs reported as fed once-daily but who received snacks and treats throughout the day. Although HLES gathers data on frequency of treats, we did not use these data because the caloric content of treats was unknown. Furthermore, we did not have adequately detailed data to account for the composition of the diets fed, although we do know that the majority of dogs (over 80%) ate a diet consisting primarily of dry, commercial dog food. Finally, due to the rarity of intact dogs in our sample, analyses included only spayed and neutered dogs. While age at spay or neuter might be an important factor for some health outcomes [63, 64], this information was not incorporated into our analyses because data on the timing of gonadectomy were not available with sufficient detail or completeness.

Studies of obesity, including possible associations with feeding frequency [23], will be an important area of future research. This investigation did not consider obesity because information on dogs’ body condition scores was not available. We anticipate that these data will be available in the future when owners share their dogs’ veterinary electronic medical records (VEMR) with the Dog Aging Project.

We view once-daily feeding to be a model of time-restricted feeding and did not expect to see associations between health and feeding frequency among dogs fed more frequently. However, our secondary analyses revealed an intriguing trend for all nine categories of health conditions (not including CSLB). Specifically, we observe higher odds of a health condition from once-daily to twice-daily to three-or-more-times daily feeding. On the other hand, free-fed dogs often did not follow this trend. We speculate that free-fed dogs represent a heterogeneous group that includes dogs with high caloric intake as well as dogs whose owners provide constant access to food because the dog eats very little. We also recognize that free-fed dogs may naturally restrict their period of food consumption each day. While these results are intriguing, we see these analyses as exploratory and hypothesis-generating.

Given the limitations of this cross-sectional, observational study, the results of this investigation should not be used to make decisions about the feeding or clinical care of companion dogs. However, if supported by future studies, it may be prudent to revisit the currently predominant recommendation that adult dogs be fed twice daily. The rationale for twice-daily feeding in dogs is obscure (although see [65]), and our study suggests that more frequent feeding may, in fact, be suboptimal for several age-related health outcomes.

We view these results as an exciting first step of an ongoing exploration of the impact of diet on companion dogs living in human environments. Given the intense interest in, and popularization of, “longevity diets” such as intermittent fasting and time-restricted feeding, these types of studies in dogs are both timely and important. We believe these studies will ultimately offer insights into factors that promote health and longevity for both dogs and humans.