Food for Thought: Lactating Coquerel’s Sifaka (Propithecus Coquereli) Eat Foods High in Protein and Fiber During the Lean Season

Infant-bearing, Coquerel’s sifaka (Propithecus coquereli) undergo gestation during a lean seasonal climate with weaning occurring during the abundant season. During this time, nutrient demand increases due to placental transport to the fetus and to the infant postpartum by milk. Females respond to this increased demand by ingesting larger food quantities, reducing expenditure, and/or using their nutrient stores. We collected foods (N=75) exploited by lactating females (N=10) in Ankarafantsika National Park, Madagascar to examine the nutritional landscape within which sifakas forage. We measured food nitrogen, neutral detergent �ber (NDF), acid detergent �ber (ADF), gross energy (GE) and ash to estimate crude protein (CP), available protein (AP), �ber, mineral content and metabolizable energy (ME). Two signi�cant PCA (principal component analysis) axes corresponded to high protein and high �ber-low ME explaining 91.6% of the variance. Cluster 1 is categorized by foods that contained higher AP and cluster 2 is categorized by higher �ber foods. P. coquereli rely on a diverse range of foods inclusive of those with high AP and ME, but also high �ber foods with low ME. We hypothesize that the high �ber, low ME foods may be important for maintaining the gut microbiome.


Introduction
The nutrient content of foods eaten by wild primates is highly variable and resources are not interchangeable [1][2][3] . The nutrient quantities required for proper primate nutrition are contingent on body size, metabolism, digestive anatomy and physiology, sex, life history, and habitat quality [4][5][6][7][8][9] . Food selection indicates varying nutritional needs 10 by prioritizing nutrients to meet distinct nutritional goals within environmental constraints 11 . Assessing the nutrients and energy available from foods helps gauge these speci c parameters within this contextual framework. One effective way to examine these constraints is to measure the nutrient content of foods consumed by individual animals to explore the nutritional options in their habitat.
The taxonomic Family Indriidae is composed of mostly folivorous-frugivorous primates endemic to Madagascar that have evolved an extensive small intestine and enlarged hindgut to assist with nutrient extraction 5 . The enlarged lower gut characteristic of hindgut fermenters consists of the caecum, a portion of the large intestine, and colon 12 that stretches to 13-15 times the animal's body length, thereby requiring a 24-48 hour gut-passage time in Propithecus spp. 13,14 . The lower gut serves as a fermentation chamber to aid in ber digestion (Lambert 1998) with large populations of microbes housed in the caecum (Campbell et al. 1999). Microbes found in the caecum and colon are capable of fermenting ber, in turn producing energy for indriids in the form of short-chain volatile fatty acids (primarily acetate, butyrate and propionate), as well as amino acids, vitamins and a host of other bioactive molecules that may bene t the host 15 .
Indriids are challenged with the unpredictability in abundance and distribution of food resources due to the extreme seasonality within the region 16 . Additionally, the majority of lemur species including indriids give birth during the dry, lean season when resources are of lower quality (i.e., reduced protein and energy availability) and wean infants during the wet, abundant season when resources are higher quality (i.e., greater protein and energy availability) 17,18 . P. coquereli infants are born predominantly during the lean season from June-August and weaned during the abundant season from January-February 19,20 . This reproductive strategy intensi es the already high energetic demands on lactating females since infants are behaviorally and nutritionally dependent when resources are most seasonally depletive. As an example of a related species, female Verreaux's sifaka (Propithecus verreauxi) increase their overall food intake during late lactation; including increased intakes of crude protein, fat, non-structural carbohydrates and energy relative to males 6 . During gestation, sex differences in macronutrient intakes and energy were not present (Koch et al. 2017). Even with a greater nutrient intake during late lactation, lactating P. verreauxi lose 18% of their body weight throughout the dry season 21 .
In the present study, we investigate the nutrient content of foods selected by lactating P. coquereli during the lean season. We assessed protein, ber, energy, and minerals to explore the nutrients available to lactating females from which we characterize the nutritional landscape in which sifakas forage and feed.

Study Site
This study was conducted in Ankarafantsika National Park (ANP), Madagascar. ANP is a dry deciduous forest with a pronounced lean (dry) season from May to September 22 with the greater number of P. coquereli infants being born during this time; i.e., late May to August 19,23 . Forested areas are experiencing anthropogenic disturbance from slash-and-burn agriculture, re, human tra c, unregulated presence and herding of domestic cattle, bushmeat hunting and hole digging for Dioscorea maciba tuber extraction [24][25][26] , which increases food scarcity during the lean season. Soils are either red, speckled, or white, with red soil containing the highest water content and white sand the lowest 25 . Many tree species grow in nutrient poor, acidic white sands and a thick layer of loose sand is present on the soil surface because of sandstone erosion 27,28 . Flora are speciose and the forest understory is moderately thick with sparse leaf litter (Lourenço & Goodman, 2006).

Plant Collection
The collection of plants that were consumed by ten habituated P. coquereli lactating females occurred from June to December of 2010 and 2011 for 93 hours over 52 weeks (26 consecutive weeks/season). Plant parts identi ed included: leaves, fruits, owers, buds, and bark. Samples were stored in manila envelopes until they were transported to a propane drying oven at the end of each focal follow.

Plant Processing and Preservation
Samples were dried on-site in a propane oven at a maximum of 50°C using a max/min digital thermometer (HBE International Inc.) until a constant weight was reached for at least 48 hours 29 . Samples were weighed daily to determine dry weights and not exposed to direct sunlight to limit post-collection changes in nutrient composition. Samples were placed in 3M SCC Dri-Shield 2000 moisture barrier bags with silica gel and stored in plastic containers in a concrete storage area.

Chemical Analyses and Calculations
Laboratory assays were conducted at the Nutrition Laboratory, Smithsonian National Zoo and Conservation Biology Institute. Dry food samples were re-dried at 55° C for a minimum of 48 hours and ground to achieve a homogeneous subsample. Plant material was ground using a Wiley mill or with a ceramic mortar and pestle depending on consistency and sample size and passed either through a 0.38 mm sieve (CHN procedure) or 0.86 mm sieve. Assays included: nitrogen (N) as an index for protein, neutral detergent ber (NDF), acid detergent ber (ADF), gross energy (GE) (kcal/g), and ash as an index for total mineral content. N content were measured using a combustion method (Dumas method) in a PerkinElmer 2400 Series II Analyzer (PerkinElmer, Waltham, MA). The ANKOM ber procedure using an ANKOM Fiber 200 Analyzer or the Van Soest ber procedure 30 were used for neutral detergent (NDF) and acid detergent ber (ADF determination). We did not assay ADL (acid detergent lignin) which would have represented the indigestible ber fraction and acknowledge this may have affected our results and interpretation. GE of samples (kcal/g) was measured using adiabatic bomb calorimetry to measure the heat from sample combustion. Pellets were formed from 0.25-0.75 g of sample and re-dried for one hour at 60°C. A Parr 1241 Adiabatic Calorimeter (Parr Instrument Company, Moline, IL) was used to measure GE. Samples were considered for re-assay if duplicates varied by >0.2 kcal/g. Total mineral content was determined by ashing the samples in a mu e furnace. Crucibles were lled with 0.25-0.50 g of sample and heated for six hours at 450° C.
We estimated crude protein (CP) following Maynard and Loosli 31 ; available protein (AP) following 32 ; and metabolizable energy (ME) using values for energy not available from NDF from Campbell, et al. 33 and Conklin-Brittain, et al. 34 .
The value of 2.17 kcal/g for energy not available from NDF was estimated using the value 61% NDF digestion factor 33 and accounting for the energy lost to microbial metabolism estimated by as 1kcal/g of NDF 34 . Thus, energy lost from NDF is estimated to be: The mean value for NDF digestion was for captive foods 33 , and thus likely represents a maximum for wild foods, so our estimated nonprotein ME is likely an overestimate.

Statistical Analysis
A total of 139 plant samples were assayed, however, there were some duplicate samples of the same food type (e.g., fruit, leaf) and plant species collected from different locations or times. Duplicate samples were averaged to produce macronutrient values for a unique species-plant part except in the case of four species that displayed an apparent seasonal difference in nutrient composition (Table 1). These eight samples were treated as different foods, based on the macronutrient composition. This resulted in 75 unique sifaka foods for which we report data ( Table 2). All nutrient results are reported on a dry matter basis to control for the effect of variable water content. Values are reported as mean ± SEM and range. Pearson's correlation was used to assess associations among nutrients assayed. Data were analyzed using SPSS 20.0, IBM Corp, Armonk NY.   Exploratory statistics were used to describe the variation in sifaka foods. Principal component analysis (PCA) was conducted on the nutrient values to reduce the number of parameters (CP, AP, NDF, ADF, GE, ME, and ash). Only axes with an eigen-value greater than one were considered signi cant. The PCA was considered signi cant if Bartlett's Test for Sphericity was signi cant and the Kaiser-Meyer-Olkin measure of sampling adequacy was greater or equal to 0.5 35 . The number of signi cant axes from the PCA was used to set the k value for the k-means cluster analysis on the same parameter set.

Results
Nutrient values for the 75 unique plant foods are given in Table 2. The sifakas selected foods representing 48 unique plant taxa with a wide range of nutrient content. AP, digestible protein not bound in ber, ranged from 0.0-24.4%, with a mean of 10.3 ± 0.7% and median of 10.2%. NDF ranged from 8.3-81.8% with a mean of 38.2 ± 2.1% and median of 34.8%. ADF ranged from 6.2-75.2% with a mean of 27.8 ± 1.8% and median of 23.7%. ME ranged from 1.92 kcal/g to 4.96 kcal/g, with a mean of 3.49 ± 0.07 kcal/g and median of 3.56 kcal/g. Ash (total minerals) ranged from 1.85-19.37%, with a mean of 4.95 ± 0.32% and median of 4.22%. Four foods showed seasonal differences in nutrient composition, with the highest percentage of change in the amount of protein in manary (Dalbergia trichophylla) fruit from the end of the lean to the beginning of the wet season (Table 1).
Except for bark, plant part does not categorize sifaka foods by nutrient composition, as all plant part categories had examples of high and low values for all nutrients. For example, the mean and range of NDF content of leaves (35.8%, 17.1 -81.8%) was virtually the same as the mean and range of NDF for fruit (37.7%, 8.6 -78.5%). Although the mean value for available protein for leaves (12.7±0.9%) was numerically higher than that for fruit (7.6±1.3%), the range again was essentially identical for the two plant parts (0 -24.4% and 0 -22.0%). Bark contained mostly ber, with essentially no available protein ( Table 2).
The best t PCA model contained only ve of the seven parameters (CP, AP, NDF, ADF, and ME). The best model found two signi cant axes (eigen-values greater than one) that can be categorized as high protein and high ber. These two axes (protein factor and ber factor) explained 91.6% of the variation in nutrient content between the foods. Bartlett's Test of Sphericity was signi cant (Chi-square = 435. 6, df = 10, p<0.001) and the Kaiser-Meyer-Olkin measure of sampling adequacy was 0.659, suggesting that sampling is adequate.
Estimated ME was signi cantly negatively correlated with the ber factor score from the PCA (r = -0.867, p<0.001; Figure 1) but was not associated with the protein factor score. Ash was positively correlated with the protein factor score (r = 0.314, p = 0.012) but was not correlated with the ber factor score.
The cluster analysis had k set to 2 based on the number of signi cant axes from the PCA. Cluster 1 foods (N=52) were higher in AP and lower in ber ( Table 3). The foods in cluster 2 (N=14) were higher in ber and lower in estimated ME (Table 3). Nine foods could not be ascribed to a cluster because they were missing GE data, and thus an estimated ME could not be calculated. Figures 2 through 4 display how the foods in the two clusters differ. Cluster 1 foods displayed a positive correlation between the protein and ber factor scores (r = 0.580, p<0.001, Figure 2) while cluster 2 foods showed no association (r = -0.136, p=0.642, Figure 2). Both cluster 1 and cluster 2 foods had negative correlations between estimated ME and the ber factor score (r = -0. 728, p<0.001 and r = -0.855, p<0.001). Cluster 1 foods had a tendency for estimated ME to be negatively associated with the protein factor score (r = -0.270, p=0.053), but there was no association between estimated ME and the protein factor score for cluster 2 foods.   Figure 4 displays the lower and less variable AP for cluster 2 foods. In addition, there is no relationship between AP and ADF for cluster 1 foods (Figure 4), but a signi cant decline in AP with ADF for cluster 2 foods (r = -0.717, p=0.004). The ratio of AP to CP differed between clusters 1 (0.88±.01) and 2 (0.53±0.1; p<0.001), indicating that a greater percentage of protein was bound to the ADF fraction for cluster 2 foods. Cluster 2 foods had a lower protein-to-ber ratio whether expressed as CP-to-NDF (0.48±0.04 versus 0.09±0.01, p<0.001) or CPto-ADF (0.70±0.05 versus 0.12±0.02, p<0.001).
Cluster 2 foods were comprised of all 4 bark samples, 6 of 18 fruit samples, 4 of 29 leaf samples, but no buds or owers. There were 6 samples, 3 from cluster 1 and 3 from cluster 2, that overlap in the ber and protein factor space ( Figure 2). The cluster 1 foods were lower in NDF (44.3±0.7% versus 50.4±0.7%, p=0.004) with no overlap, but otherwise did not differ from the cluster 2 foods (Table 4). Table 4 Foods from clusters 1 and 2 overlapping in the protein factor-ber factor space +

Discussion
We found that lactating P. coquereli exploited a nutritionally diverse set of foods that varied widely for all measured nutrients and included many high ber foods. The PCA indicated that available protein, ber and metabolizable energy accounted for over 91% of the variation among these foods. Our analysis revealed two potential categories of foods in our dataset, visually represented in Figure 2. The relationship between ME and ADF ( Figure 3) and AP and ADF ( Figure 4) visually demonstrates the separation between the clusters for ber. However, estimated ME and AP shows considerable overlap between the two clusters, suggesting that sifaka foods could be described by a nutritional gradient. This approach is supported since some foods were moderate to higher in protein and metabolizable energy while lower in ber, and other foods were lower in protein and metabolizable energy while higher in ber. The gradient approach may better re ect the continuous nature of nutrient values, particularly for foods on the cluster boundaries (Figures 1, 2, and 4). However, the six foods overlapping in protein and ber factor space do differ in NDF (Table 4) and the two clusters vary in the proportion of ber bound to ADF. Both these factors support the hypothesis that these foods cluster into at least two nutritionally distinct groups. We propose that these two food types will have different physiological and metabolic effects, with cluster 1 foods contributing more to the ingesting sifaka's nutritional status directly while cluster 2 foods will affect nutritional status through effects on the sifaka gut microbiome.
Protein and ber were the most consistently variable nutrients in the sifaka foods, which also varied considerably in the protein-to-ber ratio. Primates are estimated to require a minimum of 14% protein per dry matter basis for reproduction, 7-11% for growth and development 36 , and 6.4-8% crude protein in their diet to satisfy maintenance nutritional requirements 37 . The cluster 1 foods consumed by lactating P. coquereli had a mean of 12.0% available protein, which exceeds minimum protein requirements for primate maintenance, and growth and development, while nearly meeting the estimated reproductive nutritional requirements. Cluster 1 foods had a high ratio of AP-to-CP, supporting the hypothesis that they are good protein sources.
Lactating P. coquereli appear to have a diet quite high in ber (means of 38.2% NDF, 27.8% ADF) with a relatively low protein-to-ber ratio without experiencing adverse effects and routinely consumed high ber foods during the lean season (Table 3). Frequently consumed foods of gestating ring-tailed lemurs (Lemur catta) during the dry season contained less than 21% ADF 3 . During lactation, eight of ten of the most frequently consumed foods contained less than 30% ADF, while none of the foods contained over 50% ADF 3 . The black-andwhite ruffed lemur (Varecia variegata), consumed fruits, leaves and owers with ADF content of approximately 30% 38 . The average ADF content of leaves eaten by the larger-bodied Indri (Indri indri) was 53%, and the fruit, leaves and owers consumed by diademed sifakas (Propithecus diadema) averaged between 30 and 50% ADF 39 . The ber levels for these larger lemur species are comparable to our results for P. coquereli. Although, the sifakas did include many high protein/low ber foods in their diet, suggesting that exploiting different foods has functionally distinctive physiological and metabolic consequences.
High ber food consumption may be a residual effect of lactating P. coquereli unselectively exploiting the foods available in the forest during the lean season. We emphasize that this also has biological relevance, since it provides an assessment of seasonally available nutrients consumed during the critical period of infant development. During the lean season in a dry deciduous forest the availability of foods high in available protein and metabolizable energy may be insu cient, thereby constraining females to select di cult to digest resources to meet energy requirements. Perhaps the increased demand placed by lactation in conjunction with the food constraints of the lean season force sifakas living in dry deciduous forest to ingest the high ber foods.
However, Propithecus spp. are hindgut fermenters 5 with highly specialized gut microbiomes that vary depending on seasonal fruit availability 40 . Dietary plant ber only become nutritious after its microbial conversion into vital nutrients like short-chain fatty acids 41 , facilitated by speci c cellulose-degrading microbes present in the sifaka gut and an increased functional capacity for ber metabolism 42 . The specialized morphology of hindgut fermenters (enlarged caecum and elongated colon) could enable the e cient digestion of brous materials, increasing nutrient extraction from di cult to digest resources. Our ndings are consistent with previous studies that have shown sifakas to be seasonally exible folivores, a novel dietary strategy that may mitigate potential energetic de cits [43][44][45] . Recent evidence demonstrates sifakas possess molecular adaptions to folivory including rapidly evolving gene pathways that aid in xenobiotic metabolism and nutrient absorption, which may assist in the detoxi cation of plant compounds while maximizing nutritional gain from leaves 46 . This capacity for augmented nutrient uptake 46 would be advantageous to foraging throughout periods of pronounced seasonality in Madagascar 18,47 .
Variation in dietary ber is a critical component to understanding gut microbiomes in folivores and has been shown to affect microbial diversity in P. coquereli 42 . Sifaka gut microbiomes have been found to be signi cantly richer and more diverse in comparison to generalist and frugivorous lemurs 42 . Less inter-individual variation in sifaka gut microbiomes is exhibited relative to frugivorous V.
variegata and generalist L. catta, suggesting that sifakas may be less exible in terms of their diet 42 and more susceptible to habitat disturbance 48 .
Captive P. coquereli provisioned with a more diverse diet that included local wild plant species had signi cantly richer, more diverse gut microbiomes in comparison to when their standard diet was supplemented with winged-sumac only 48 . Signi cantly higher concentrations of short-chain fatty acids, including acetate and propionate, and moderately greater concentrations of butyrate were present in P. coquereli colonic metabolomes when provisioned a more diverse diet 48 . Additionally, the same study found that individuals given the opportunity to forage more naturally in forested enclosures, even for limited durations, maintained greater gut microbiome diversity relative to conspeci cs without forest access (Green et al., 2018). This supports that ber consumption can have a profound in uence on gut microbiome structure and function. It is possible that a high-ber diet is a requirement for sifakas to maintain their coevolved microbiota. We posit that many if not all the cluster 2 foods in our study may have a greater effect on the sifaka gut microbiome than a direct nutritional effect on the host animal. In other words, cluster 2 foods may be important for maintaining gut health by feeding the microbiome, while cluster 1 foods more directly affect the nutritional plane of the sifakas.
Sifakas are exceptionally di cult to maintain in captivity due to their specialized digestive anatomy and highly folivorous diet 49,50 . Our results suggest that incorporating high-ber foods (ADF greater than 30% or even 40%) into captive diets would better replicate foods consumed in the wild. Table 3 highlights two leading commercial products for leaf-eating primates in various life cycle stages, health, and seasonality versus our eld data collected on lactating sifakas during the lean season. The commercial supplements contain higher concentrations of protein (CP) and lower concentrations of ber. Both Marion and Mazuri provide their products as supplements to foraging and non-foraging fruit and vegetable produce diets. Because of this, percent nutritional values of the various nutrients do not represent the overall lemurs' diet, but only that of the commercial product itself. Similarly, food selection in the wild depends on environmental factors and does not necessarily re ect the ideal composition for the health of sifakas without food supply constraints as in captivity. While we acknowledge the limitations of juxtaposing a partial wild diet to a partial captive diet, it is presented here to highlight the importance of incorporating nutritional diversity in captive diet design based on wild plant foods acquired by lemurs. We suggest that incorporating foods like the cluster 2 foods in this study may be helpful for dietary management of captive sifakas, possibly by improving gut health through effects on the microbiome.
Consistent with previous studies 11,29,51−53 , our results con rm that botanical category (e.g., fruit versus leaf) is a poor means by which to assess the nutritional contribution a food will make to animals that consume it. Fruit is often equated with high water and high nonstructural carbohydrate (sugar) content; however, wild fruits can be substantially different in nutrient pro le from domesticated fruits, and often are similar to leaves, buds, and owers, as seen in our study. The fruits in this study were not different from leaves in ber content. The NDF content of fruit in our study ranged from 8.6-78.5% and the mean NDF for fruit (37.7%) was numerically higher than the mean NDF for leaves (35.8%). Sifakas ingest high ber foods, whether those foods are classi ed as leaves, fruit, owers, or buds. Our results also con rm that wild plant foods can vary seasonally in nutrient content, cautioning that the nutritional consequences of consuming some foods can differ by time of year.
In summary, infant-bearing P. coquereli's employ a mixed-diet strategy consuming foods with wide ranges in percent nutrient content to compensate for nutrient de ciencies in multiple plant parts and food availability. Food sources clustered into two categories: high in protein and low-to-moderate in ber; or high in ber and low in metabolizable energy. Relationship between metabolizable energy (ME) and ber by cluster in foods consumed by lactating P. coquereli Relationship between protein and ber by cluster in foods consumed by lactating P. coquereli Page 19/20

Figure 3
Relationships between metabolizable energy (ME) and acid detergent ber (ADF) of foods consumed by lactating P. coquereli determined from PCA followed by cluster analysis