Gourds and squashes (Cucurbita spp.) adapted to megafaunal extinction and ecological anachronism through domestication

Significance

Squashes, pumpkins, and gourds belonging to the genus Cucurbita were domesticated on several occasions throughout the Americas, beginning around 10,000 years ago. The wild forms of these species are unpalatably bitter to humans and other extant mammals, but their seeds are present in mastodon dung deposits, demonstrating that they may have been dispersed by large-bodied herbivores undeterred by their bitterness. However, Cucurbita may have been poorly adapted to a landscape lacking these large dispersal partners. Our study proposes a link between the disappearance of megafaunal mammals from the landscape, the decline of wild Cucurbita populations, and, ultimately, the evolution of domesticated Cucurbita alongside human cultivators.

Abstract

The genus Cucurbita (squashes, pumpkins, gourds) contains numerous domesticated lineages with ancient New World origins. It was broadly distributed in the past but has declined to the point that several of the crops’ progenitor species are scarce or unknown in the wild. We hypothesize that Holocene ecological shifts and megafaunal extinctions severely impacted wild Cucurbita, whereas their domestic counterparts adapted to changing conditions via symbiosis with human cultivators. First, we used high-throughput sequencing to analyze complete plastid genomes of 91 total Cucurbita samples, comprising ancient (n = 19), modern wild (n = 30), and modern domestic (n = 42) taxa. This analysis demonstrates independent domestication in eastern North America, evidence of a previously unknown pathway to domestication in northeastern Mexico, and broad archaeological distributions of taxa currently unknown in the wild. Further, sequence similarity between distant wild populations suggests recent fragmentation. Collectively, these results point to wild-type declines coinciding with widespread domestication. Second, we hypothesize that the disappearance of large herbivores struck a critical ecological blow against wild Cucurbita, and we take initial steps to consider this hypothesis through cross-mammal analyses of bitter taste receptor gene repertoires. Directly, megafauna consumed Cucurbita fruits and dispersed their seeds; wild Cucurbita were likely left without mutualistic dispersal partners in the Holocene because they are unpalatable to smaller surviving mammals with more bitter taste receptor genes. Indirectly, megafauna maintained mosaic-like landscapes ideal for Cucurbita, and vegetative changes following the megafaunal extinctions likely crowded out their disturbed-ground niche. Thus, anthropogenic landscapes provided favorable growth habitats and willing dispersal partners in the wake of ecological upheaval.

The wild precursors of domestic squashes (Cucurbita spp.) are adapted for a landscape inhabited by large herbivores. Their robust pepo fruits (1) were dispersed by large mammals, as revealed by intact Cucurbita seeds in mastodon dung deposits (2). Furthermore, Cucurbita is a weedy genus (3) well suited to the mosaic-like landscapes maintained by megafauna, which offer an abundance of disturbed habitat in a niche-diverse ecosystem (4). However, all megafauna >1,000 kg disappeared from the Americas by the early Holocene—through ecological shifts, human predation, or some combination of both (5)—leaving Cucurbita in a turbulent ecosystem lacking its mutualistic partners. Cucurbita is among many such anachronistic New World taxa (6), including avocado (Persea americana), chocolate (Theobroma cacao), Osage orange (Maclura pomifera), honey locust (Gleditsia triacanthos), and tree calabash (Crescentia cujete). Distributions of ancient samples show that some species of Cucurbita were distributed much more broadly in the past (e.g., refs. 7 and 8), suggesting that the Holocene has witnessed extirpation and refugiation of certain wild types. The decline of wild Cucurbita may have stemmed from both its loss of dispersal mutualists and the changing habitat. Other anachronistic plants have partnered with substitute dispersers (9), but most wild Cucurbita have not adapted in this way (10).

Although Cucurbita species declined in the wild, they thrived in domestication. The genus contains at least five domesticated species and has been used extensively throughout the Holocene (11), beginning around 10,000 B.P. in Mexico (12). Cucurbita domestication was a widespread phenomenon involving numerous wild precursor lineages throughout the Americas (11, 13–15), and these taxa now contain hundreds of cultivars and landraces grown worldwide (11, 15). In this study, we use a two-stage approach to study the simultaneous decline of wild Cucurbita and rise of its domestic counterparts. First, to assess aspects of Cucurbita domestication history and phylogeographic patterning, we recover and analyze complete plastid genomes from 91 total archaeological (n = 19) and modern wild (n = 30) and domestic (n = 42) Cucurbita specimens. Second, wild Cucurbita may have adapted poorly to dispersal anachronism partly for physiological reasons. Specifically, it produces a cytotoxic suite of cucurbitacins—harshly bitter triterpenoid compounds (16)—that deter small mammals (17) but appear to be harmless to megafauna. As such, Cucurbita was palatable to some megafauna (2) but possibly not to the majority of smaller New World mammals that survived into the Holocene; thus, Cucurbita survival was threatened by the disruption of its propagation strategy. By screening 46 mammal genomes for bitter taste receptor-encoding taste receptor type 2 (TAS2R) genes, we evaluate the hypothesis that mammal body size may be related directly to the palatability of Cucurbita and other bitter species. If smaller mammals have adapted mechanisms for better detecting and avoiding bitter, potentially toxic compounds in lower doses, Cucurbita may have coadapted specifically for a landscape containing large mammals; such adaptation would increase the potential impacts of a major shift toward an ecosystem lacking them. In sum, we combine archaeogenetic and comparative genomic techniques to probe the natural history of a genus that is ecologically anachronistic in the wild but is prolific in cultivation.

Results

Plastid Genome Analysis.

Using Illumina shotgun sequencing, we recovered and assembled complete plastid genomes from 12 of the 14 accepted extant Cucurbita species, representing 18 total species, subspecies, and varietal taxa. Using our completed assemblies for C. pepo ssp. pepo and C. moschata, we then designed and synthesized biotinylated RNA probes (18) to enrich additional modern and ancient DNA isolates for targeted sequencing of complete plastid genomes. Using this enrichment strategy, we tested 32 archaeological Cucurbita samples, yielding 19 ancient plastid genomes sufficiently complete for analysis. We also sequenced an additional 53 modern plastid genomes using a combination of shotgun sequencing and targeted enrichment, bringing the total sample size to 91 (n = 19 ancient; n = 72 modern) (Tables S1–S3), and reconstructed a Cucurbita plastid genome phylogeny from this dataset (Fig. 1).

Table S1.

Sample details for Cucurbita accessions analyzed: archaeological samples

Laboratory ID	Taxon	Site	Site no.	Accession no.	Tissue
101	C. pepo ssp. ovifera	Putnam Shelter, AR	3WA4	32-44-291	Rind
102	C. pepo ssp. ovifera	Salts Bluff, AR	3BE18	33-12-125	Rind
103	C. pepo ssp. ovifera	23BYM4, MO	23BYM4	32-34-107	Rind
104	C. pepo ssp. ovifera	Craddock Shelter, AR	3CW2	34-7-130	Rind
105	C. pepo ssp. ovifera	Cob Cave, AR	3NW6	31-72-25	Rind
106	C. argyrosperma	Eden's Bluff, AR	3BE6	31-72-25	Rind
107	Cucurbita sp.	Salts Cave, KY	15HT4	CRF-81	Rind
108	Cucurbita sp.	Salts Cave, KY	15HT4	CRF-85	Rind
109	Cucurbita sp.	Salts Cave, KY	15HT4	CRF-95	Rind
110	C. pepo ssp. ovifera	Phillips Spring, MO	23HI216	1502 seed 8	Seed
111	Cucurbita sp.	Phillips Spring, MO	23HI216	1502 seed 13	Seed
112	C. argyrosperma	Coxcatlan Cave, Puebla, Mexico	TC50	XI 39/5	Rind
113	Cucurbita sp.	Coxcatlan Cave, Puebla, Mexico	TC50	XI 79/6	Rind
114	C. argyrosperma	Coxcatlan Cave, Puebla, Mexico	TC50	VII 70/8	Rind
115	C. argyrosperma	Coxcatlan Cave, Puebla, Mexico	TC50	V 37-4	Rind
116	Cucurbita sp.	Romero's Cave, Tamaulipas, Mexico	TMC247	occ 2 247/14	Rind
118	Cucurbita sp.	Romero's Cave, Tamaulipas, Mexico	TMC247	occ 5 247/72	Rind
119	Cucurbita sp.	Romero's Cave, Tamaulipas, Mexico	TMC247	occ 10 247 32	Rind
120	Cucurbita sp.	Romero's Cave, Tamaulipas, Mexico	TMC247	occ 11 247/93	Rind
121	C. pepo ssp. pepo	Romero's Cave, Tamaulipas, Mexico	TMC247	occ 14 247/243	Rind
122	C. moschata	Romero's Cave, Tamaulipas, Mexico	TMC247	occ 16 51/16	Rind
123	C. pepo ssp. fraterna	Romero's Cave, Tamaulipas, Mexico	TMC247	occ 8 247/89	Rind
139	C. pepo ssp. pepo	Guila Naquitz, Oaxaca, Mexico	OC43	INAH42(6N18) Zona B1 Cuadro C11	Rind
140	C. pepo ssp. pepo	Guila Naquitz, Oaxaca, Mexico	OC43	INAH58(6N20) Zona B(B1) Cuadro B9	Rind
141	Cucurbita sp.	Jemez Cave, NM		759 Lab 16801	Rind
142	C. pepo ssp. pepo	El Riego Cave, Puebla, Mexico	TC50	Square16(S3E3) Level 3 Peduncle 21	Peduncle
143	Cucurbita sp.	Newt Kash Hollow, KY	15MF1	10842 Lab 16390	Rind
144	C. pepo ssp. pepo	Bandelier Cave, NM		2143 B	Rind
145	Cucurbita sp.	Cloudsplitter Rockshelter, KY	15MF36	FS1580 ES8145	Rind
146	C. pepo ssp. pepo	Jemez Cave, NM		969 Lab 16291	Seed
147	Cucurbita sp.	Mammoth Cave, KY	15ED1	87276 B16 Wartly	Rind
148	C. pepo ssp. pepo	Bat Cave, NM		LA44182 Bag 2144 SV45A	Rind

Table S3.

Sample details for Cucurbita accessions analyzed: modern samples

Laboratory ID	Taxon	Status	Location (wild samples)	US Department of Agriculture accession no.	Sequencing strategy	Percent sites represented
010	C. pepo ssp. pepo	Cultivar		PI 508469	Shotgun/LR PCR	99.85
011	C. pepo ssp. ovifera var. ovifera	Cultivar		PI 518687	Shotgun/LR PCR	99.85
012	C. pepo ssp. ovifera var. texana	Wild	Texas	Ames 26891	Shotgun/LR PCR	99.80
013	C. pepo ssp. ovifera var. ozarkana	Wild	Arkansas	Ames 26607	Shotgun/LR PCR	99.80
014	C. pepo ssp. fraterna	Wild	Tamaulipas, Mexico	PI 532354	Shotgun/LR PCR	99.80
015	C. moschata	Domestic		PI 209116	Shotgun/LR PCR	99.86
016	C. argyrosperma	Landrace		PI 511929	Shotgun	99.86
017	C. argyrosperma var. palmeri	Wild	Sinaloa, Mexico	PI 512211	Shotgun/LR PCR	99.11
018	C. argyrosperma ssp. sororia	Wild	Chiapas, Mexico	PI 532404	Shotgun/LR PCR	99.86
020	C. maxima ssp. andreana	Domestic		G 29254	Shotgun/LR PCR	99.86
021	C. ecuadorensis	Wild	Casco, Ecuador	PI 432445	Shotgun/LR PCR	99.85
022	C. okeechobeensis	Wild	Florida	N/A (LAN, sample no. S-600)	Shotgun/LR PCR	99.86
023	C. okeechobeensis ssp. martinezii	Wild	Veracruz, Mexico	PI 540900	Shotgun/LR PCR	99.83
024	C. lundelliana	Wild	Guatemala	Grif 9452	Shotgun/LR PCR	99.85
025	C. ficifolia	Domestic		PI 512680	Shotgun	99.86
026	C. digitata	Wild	Sonora, Mexico	PI 240879	Shotgun/LR PCR	99.84
027	C. pedatifolia	Wild	San Luis Potosí, Mexico	PI 442290	Shotgun/LR PCR	99.86
028	C. cordata	Wild	Mexico	PI 653839	Shotgun/LR PCR	99.86
029	C. foetidissima	Wild	Oklahoma	PARL 264	Shotgun	99.82
036	C. pepo ssp. fraterna	Wild	Tamaulipas, Mexico	PI 532356	Shotgun	99.77
037	C. pepo ssp. ovifera var. ozarkana	Wild	Illinois	Ames 26875	Shotgun	99.76
038	C. pepo ssp. ovifera var. texana	Wild	Texas	PI 614686	Shotgun	99.73
039	C. pepo ssp. ovifera var. ovifera	Cultivar		PI 267755	Shotgun	99.77
040	C. pepo ssp. ovifera var. ovifera	Cultivar		NSL 180768	Shotgun	99.79
041	C. moschata	Domestic		PI 244707	Shotgun	99.85
042	C. moschata	Domestic		PI 442263	Shotgun	99.85
043	C. moschata	Domestic		PI 634693	Shotgun	99.81
044	C. pepo ssp. pepo	Cultivar		PI 615105	Plastid capture	99.01
045	C. pepo ssp. pepo	Cultivar		PI 615088	Plastid capture	98.34
046	C. pepo ssp. ovifera var. ovifera	Cultivar		PI 595838	Plastid capture	98.56
047	C. pepo ssp. ovifera var. ovifera	Cultivar		PI 615111	Plastid capture	98.01
048	C. pepo ssp. ovifera var. ovifera	Cultivar		PI 615114	Plastid capture	97.31
049	C. pepo ssp. pepo	Cultivar		PI 615108	Plastid capture	96.14
050	C. pepo ssp. ovifera var. ovifera	Cultivar		PI 615152	Plastid capture	96.87
051	C. pepo ssp. ovifera var. ovifera	Cultivar		PI 615115	Plastid capture	97.97
052	C. pepo ssp. ovifera var. texana	Wild	Mississippi	PI 614693	Plastid capture	98.01
053	C. pepo ssp. ovifera var. texana	Wild	Mississippi	PI 614692	Plastid capture	98.44
054	C. pepo ssp. ovifera var. texana	Wild	Texas	PI 614689	Plastid capture	97.72
055	C. pepo ssp. ovifera var. texana	Wild	Texas	PI 614687	Plastid capture	92.44
056	C. pepo ssp. ovifera var. texana	Wild	Mississippi	PI 614701	Plastid capture	93.81
057	C. pepo ssp. ovifera var. ozarkana	Wild	Oklahoma	Ames 26610	Plastid capture	95.74
058	C. pepo ssp. ovifera var. ozarkana	Wild	Illinois	Ames 26616	Plastid capture	96.96
059	C. pepo ssp. ovifera var. ozarkana	Wild	Louisiana	Ames 26884	Plastid capture	93.45
060	C. pepo ssp. ovifera var. ozarkana	Wild	Mississippi	Ames 26882	Plastid capture	96.12
061	C. pepo ssp. ovifera var. ozarkana	Wild	Kentucky	Ames 26617	Plastid capture	86.99
062	C. pepo ssp. ovifera var. ozarkana	Wild	Missouri	Ames 26612	Plastid capture	92.02
063	C. pepo ssp. ovifera var. ozarkana	Wild	Arkansas	Ames 26890	Plastid capture	91.06
064	C. pepo ssp. fraterna	Wild	Tamaulipas, Mexico	PI 532355	Plastid capture	92.29
065	C. pepo ssp. fraterna	Wild	Tamaulipas, Mexico	PI 614683	Plastid capture	88.18
071	C. moschata	Domestic		N/A (LAN, “Seminole Pumpkin”)	Plastid capture	99.83
072	C. moschata	Domestic		PI 211998	Plastid capture	99.82
073	C. moschata	Domestic		PI 169409	Plastid capture	99.84
074	C. moschata	Domestic		PI 183258	Plastid capture	99.80
075	C. moschata	Domestic		PI 483347	Plastid capture	99.85
076	C. moschata	Domestic		PI 419083	Plastid capture	99.85
077	C. moschata	Domestic		PI 200822	Plastid capture	99.78
078	C. moschata	Domestic		PI 249565	Plastid capture	99.85
079	C. moschata	Domestic		PI 287532	Plastid capture	99.85
080	C. moschata	Domestic		PI 357918	Plastid capture	99.85
081	C. moschata	Domestic		PI 357918	Plastid capture	99.86
082	C. moschata	Domestic		PI 524427	Plastid capture	99.84
084	C. moschata	Domestic		PI 490351	Plastid capture	99.85
085	C. moschata	Domestic		PI 200736	Plastid capture	99.80
086	C. moschata	Domestic		PI 438548	Plastid capture	99.86
087	C. moschata	Domestic		PI 406848	Plastid capture	99.85
088	C. moschata	Domestic		PI 369346	Plastid capture	99.85
089	C. moschata	Domestic		PI 194570	Plastid capture	99.85
090	C. moschata	Domestic		PI 458728	Plastid capture	99.86
091	C. moschata	Domestic		PI 441725	Plastid capture	99.85
092	C. moschata	Domestic		PI 498429	Plastid capture	99.85
093	C. moschata	Domestic		PI 475750	Plastid capture	99.84
094	C. moschata	Domestic		PI 458650	Plastid capture	99.84

Fig. 1.

Sample map and plastid genome phylogeny of Cucurbita. All nodes shown are supported with Bayesian posterior probability of 1 and maximum likelihood support of 100/100 bootstrap replicates, with two exceptions: (i) C. okeechobeensis/C. okeechobeensis ssp. martinezii node: 93 bootstrap replicates; (ii) C. argyrosperma/C. okeechobeensis/C. lundelliana node: 95 bootstrap replicates. Tree tip labels in color represent taxa recovered among our archaeological samples, with color-matching lines indicating sampling sites and diamonds showing sampling locations of modern wild counterparts, if applicable (present for C. pepo ssp. ovifera, C. pepo ssp. fraterna, and C. argyrosperma). Labeled black diamonds show the sampling locations of other wild accessions analyzed as single taxonomic representatives. Among the four wild perennial species analyzed (Tables S1–S3), C. cordata is not shown, because sampling records for this accession are not specific regarding collection locality. C. maxima, a South American species with wild and domestic members, was sampled from a wild accession of C. maxima ssp. andreana from central Argentina and so is not shown.

Table S2.

Read alignment details for Cucurbita accessions analyzed: archaeological samples

Laboratory ID	Merged read pairs obtained	Percent reads on-target	Percent sites represented	Mean depth of coverage	Nucleotides masked for final consensus
101	4,461,973	3.49	98.30	85.21	8
102	5,888,667	2.50	98.04	89.39	8
103	5,679,328	3.80	99.15	173.55	8
104	9,329,829	2.16	99.13	125.02	8
105	6,926,007	1.27	97.01	43.79	8
106	5,180,946	2.15	97.69	56.13	8
107	4,970,331	0.03	21.40	0.75	N/A
108	3,850,699	0.06	34.08	1.10	N/A
109	4,422,071	0.04	20.50	0.89	N/A
110	5,557,862	0.76	98.48	23.28	8
111	5,024,510	0.08	40.41	1.72	N/A
112	5,620,282	2.96	99.23	106.34	9
113	4,267,521	0.02	10.88	0.46	N/A
114	6,444,043	3.01	99.06	108.83	4
115	5,140,918	4.01	98.78	136.50	8
116	4,615,057	0.05	23.18	1.12	N/A
118	19,996,947	0.02	42.40	2.04	N/A
119	4,927,607	0.03	13.75	0.69	N/A
120	4,203,398	0.03	12.00	0.63	N/A
121	5,877,218	3.57	99.50	129.44	8
122	6,177,243	3.78	99.77	169.71	5
123	18,192,689	0.14	82.22	11.18	5
139	23,463,065	0.04	68.10	3.96	8
140	20,493,606	0.09	89.37	7.43	7
141	16,149,579	0.02	32.47	1.39	N/A
142	8,495,477	2.73	99.76	162.22	10
143	8,572,258	0.03	33.88	1.51	N/A
144	10,665,311	2.13	99.79	163.77	10
145	6,091,069	0.05	27.64	1.50	N/A
146	33,314,376	0.02	66.54	3.95	8
147	5,208,273	0.05	22.79	1.25	N/A
148	4,175,335	1.33	97.53	28.47	10

C. pepo comprises three subspecies: ssp. pepo, ssp. ovifera, and ssp. fraterna. Of these, C. pepo ssp. pepo (zucchini, pumpkins, and summer and winter squashes) appears to have been domesticated in the Oaxaca Valley region around 10 kya (12) and has no known extant wild type. C. pepo ssp. ovifera contains wild varieties found in patches throughout the southeastern United States, including C. pepo ssp. ovifera var. ozarkana, and C. pepo ssp. ovifera var. texana, as well as a wide variety of domestic types (scallop and acorn squashes, ornamental gourds). Archaeologically, it appears that ssp. ovifera domestication took place in eastern North America (ENA) alongside a suite of other crop species (e.g., 19). C. pepo ssp. fraterna is a wild form restricted to a narrow range in northeastern Mexico and unknown as a domesticate (20). However, it is genetically similar to ssp. ovifera and therefore has been hypothesized as an alternate source population of domesticated C. pepo in ENA (14).

We found that members of ssp. ovifera form a well-supported clade that includes modern cultivars and wild gourds from ENA as well as all archaeological ENA C. pepo samples. There was no signal of ssp. ovifera among ancient samples from Mexico, the alternative possible source region (14), nor in modern free-living ssp. fraterna in that region. This absence strongly supports local evolution of domesticated C. pepo ssp. ovifera in ENA, in agreement with archaeological evidence of an increase in seed size over time and molecular data clustering modern ssp. ovifera cultivars with ENA wild populations (21–23). However, although ssp. fraterna is known only in the wild at present and is absent from the ENA archaeological record, we observed a ssp. fraterna plastid sequence in an archaeological rind fragment from Romero’s Cave in northeastern Mexico. The specimen shows a thick (>2 mm) rind and a furrow typical only of domesticated Cucurbita. We obtained a direct date from the rind of 4,614–4,836 calibrated calendar years BP (2σ range). This result might represent cross-pollination of native wild ssp. fraterna by domesticated ssp. pepo leading to the expression of some domestic-type characters or an independent domestication trajectory that did not contribute to modern plastid diversity among cultivars. In either circumstance, this finding demonstrates a previously unknown role of ssp. fraterna in the crop fields of ancient Mexico. Additionally, we observed archaeological ssp. pepo in cave sites ranging from Oaxaca to the southwestern United States, suggesting either prolific human-mediated dispersal following domestication or a geographically widespread evolution of domestic forms.

Finally, we found that three endemic populations spanning the Gulf of Mexico, currently defined as distinct taxa (24), are essentially indistinguishable based on the plastid genome. C. okeechobeensis is restricted to the Florida peninsula, on the shores of Lake Okeechobee and the St. John’s River basin, C. okeechobeensis ssp. martinezii is found on the north coastal plain of Veracruz, and C. lundelliana grows on the limestone plains of the Yucatan Peninsula (24), but the three taxa differ at only six single-nucleotide variants across the 156.6-kb plastid genome (average pairwise nucleotide diversity: π = 0.0043%) despite their geographic separation. Along with the undifferentiated plastid genome structure among patchy C. pepo ssp. ovifera in ENA (Fig. 1), this result may reflect recent fragmentation of a previously contiguous population.

In summary, we observed a pattern that supports hypotheses for widespread Holocene domestication of Cucurbita and suggests that this process coincided with dramatic habitat and range fractionation and wild plant extinction. C. pepo ssp. pepo is thought to be extinct in the wild, but we find seven archaeological examples from this lineage spanning more than 2,000 km and dating back nearly 10,000 y. We observe at least one other independent domestication episode of C. pepo, that of ssp. ovifera in ENA alongside a suite of other native crops, and a possible third, arrested domestication trajectory involving ssp. fraterna in northeastern Mexico. We also observe genetic similarity between disparate endemic taxa (C. okeechobeensis, C. okeechobeensis ssp. martinezii, and C. lundelliana), suggesting recent separation of these populations.

Analysis of Ability to Detect Bitter-Tasting Compounds.

We screened 46 therian mammal genomes to assess general ability to detect bitter-tasting compounds by analyzing functional TAS2R gene number variation. We excluded the poorest quality assemblies [length-weighted midpoint scaffold size of the genomic assembly (scaffold N50) <250,000 nt; n = 8] because of unreliable gene detectability (Materials and Methods). Among the 38 remaining genomes, we detected a combined total of 851 intact TAS2R genes, ranging from a low of 8 in the genome of the West Indian manatee (Trichechus manatus latirostris) to a high of 46 in the genome of the common shrew (Sorex araneus). We tested for relationships among TAS2R count, body mass, and dietary breadth (Materials and Methods) while controlling for phylogenetic nonindependence.

When all species are treated as independent, TAS2R count is significantly negatively correlated with both body mass (Pearson’s product–moment correlation; r² = 0.236; P = 0.002) (Fig. 2A) and greater dietary specialization (r² = 0.194; P = 0.0056) (Fig. 2B). When all variables are analyzed simultaneously with a generalized linear model, body mass and dietary breadth are each significant predictors of TAS2R count when controlling for the other variable (P = 0.0043 and P = 0.0122, respectively), but the significance of this relationship is strongest when both variables are considered together, and explains a substantial proportion of the variance in TAS2R count (P = 0.0004; multiple r² = 0.363). When taxonomic nonindependence among our sample is controlled for by using a phylogenetic generalized least squares test (Materials and Methods), the associations of body mass and dietary breadth with TAS2R count are still significantly stronger than expected from chance (TAS2R–size: P = 0.044; TAS2R–diet: P = 0.015; overall model: P = 0.007, multiple r² = 0.246). The TAS2R–body size association is most pronounced among smaller-bodied taxa; when only the very largest mammal species are excluded from the analysis, the inverse relationships between the number of intact TAS2R genes and both body size and diet are highly significant even when controlling for phylogeny (among n = 35 mammals <1,000 kg, TAS2R–size: P = 0.0013; TAS2R–diet: P = 0.001; overall model: P = 4.08 × 10⁻⁵; multiple r² = 0.468). To summarize, smaller species, and especially those with more diverse diets, tend to have a larger repertoire of functional bitter taste receptor genes, and thus they likely are better able to detect and avoid potentially toxic bitter compounds while foraging.

Fig. 2.

TAS2R gene count and phenotypic traits in 38 high-quality mammal genomes. (A) Log₁₀-scaled body mass vs. TAS2R count. The regression line assumes taxonomic independence, and point shading corresponds to diet specialization as in B (blue: least specialized; tan: most specialized). Representative species shown from left to right are common shrew, mouse, tarsier, human, giant panda, West Indian manatee, and African elephant. (B) Boxplots of TAS2R count by diet specialization. Category 1 consists of broad dietary generalists (e.g., human, mouse), and category 4 consists of highly specialized feeders (e.g., giant panda).

Discussion

Holocene Anachronism.

The Holocene decline of wild Cucurbita was likely driven, at least in part, by the nearly complete disappearance of herbivores ≥1,000 kg, with whom they appear to have had an important symbiotic relationship. Wild Cucurbita are “too bitter for humans and well-fed livestock” (10), but larger mammals are much more tolerant of bitter plant compounds than smaller ones. Larger species are likely able to metabolize or pass moderate toxins harmlessly as part of a high daily biomass intake. For example, even if we ignore any additional physiological or behavioral detoxification strategies and the large digestive throughput of hindgut fermenters, a fully-grown African elephant [averaging 4,540 kg (22)] would need to ingest an estimated 7.5–23 kg of wild Cucurbita gourds—approximately 75–230 whole fruits—over a short timespan (25, 26) to approach a lethal dose of cucurbitacins (27). Indeed, several bitter species of Cucurbitaceae are eaten and dispersed by African elephants in the present (10). As another illustration of these concepts, the one-horned rhinoceros in lowland Nepal [1,500–2,000 kg (22)] preferentially feeds on the bitter fruits of trewia (Trewia nudiflora, Euphorbiaceae), daily consuming up to hundreds of fruits that are unpalatable to smaller mammals who are concomitantly incapable of dispersing the ∼1-cm-diameter seeds (10, 28). Humans have reported severe symptoms of cucurbitacin toxicity after consuming only “one or two bites” of cucurbitacin-tainted squash (29). The presence of Cucurbita seeds in mastodon dung is clear evidence that, like their extant counterparts, North American megafauna were effective Cucurbitaceae dispersers, physiologically undeterred by cucurbitacin bitterness.

Larger numbers of intact TAS2R genes are associated with greater ability to detect and avoid toxins via their bitterness, and interspecific variation in copy number is hypothesized to reflect ecological and evolutionary diversity (30–32). Previous studies have reported that herbivores have slightly higher TAS2R copies than carnivores, presumably to detect plant-based toxins more acutely (30), and that the common ancestor of modern whales completely lost functional TAS2R copies, probably as a result of feeding behaviors and the switch to a marine environment (33). Similarly, in our analysis, dietary breadth is significantly correlated with TAS2R count, presumably because a wider range of possible food species necessitates sensitivity to a more diverse range of toxic compounds, and because narrow dietary specialists have little need for toxin detection. Because very large-bodied species are more physiologically resilient to moderately toxic compounds, they too require fewer TAS2R gene copies. That is, bitter compounds signal toxicity to foraging animals, but in the absence of serious toxic threat, large mammals do not benefit by acutely recognizing this danger signal. Therefore, selection to maintain a broad TAS2R repertoire in larger species may be relaxed, leading to gene loss. Furthermore, although we cannot test this hypothesis in the present study, TAS2R pseudogenization could even be adaptive for megafauna if the consumption of bitter fruit was a benefit to survival, i.e., if acuity to bitter tastes would be an unnecessary barrier to herbivory.

The now-extinct megafaunal herbivores of the Americas were sufficiently large that they were likely to have been protected against moderate toxins, whereas the smaller surviving mammals are not. Accordingly, smaller mammals, and especially those with broad dietary habits, tend to have larger suites of TAS2R genes to detect and avoid toxic compounds. These smaller species generally would be unable to consume Cucurbita fruits whole and pass the intact seeds in the manner of mastodons; therefore, as seed predators instead of dispersal mutualists, the smaller species pose a threat to Cucurbita. Thus the toxicity of wild Cucurbita and related species may be, in part, an adaptive strategy to facilitate beneficial seed dispersal by very large mammals while warding off smaller seed predators. The resulting unpalatability of Cucurbita to small-bodied mammals, combined with the megafaunal extinctions, would have left the plants lacking plausible dispersal partners and the associated nutrient-rich dung deposits important in seed germination and seedling development (28); thus intergenerational seed propagation would be curtailed and offspring would be more vulnerable to localized stressors.

Moreover, Cucurbita prefers field edges, floodplains, and other disturbed habitats and would have thrived in the niche-diverse, mosaic-like landscapes maintained by large herbivores (4). In the wake of these herbivores’ extinction, large-scale zonal vegetative patterns replaced the fine-grained variation maintained by the megafauna (34), and weedy taxa would have been crowded out as their disturbed-ground niche disappeared. Taken together, these transitions could have been instrumental in driving wild Cucurbita into relict, refugial patches, causing fragmentation of previously contiguous distributions, and ultimately, in some cases, leading to the extinction of wild forms.

Domestication as an Adaptive Strategy.

Against the backdrop of these ecological changes, the anthropogenic landscape provided a disturbed habitat readily colonized by weedy species, and humans fulfilled the dispersal role. Our phylogenetic results support previous hypotheses that the domestication process was widespread, occurring independently many times throughout the Americas, including in ENA (C. pepo ssp. ovifera), Mesoamerica (C. pepo ssp. pepo and C. argyrosperma, possibly C. pepo ssp. fraterna, C. moschata, and C. ficifolia), and South America (C. maxima and possibly C. ecuadorensis and C. moschata) (35). ENA is a long-debated putative source region for Cucurbita pepo ssp. ovifera var. ovifera, a hypothesis that is strongly supported by the clustering of ovifera cultivars with modern and ancient ENA gourds in our analysis. C. pepo ssp. pepo—unknown in the wild—originated in the Valley of Oaxaca around 10,000 y B.P (12) but also was prominent among our samples from northern Mexico and the southwestern United States, several hundred miles to the north. This result suggests either (i) a widespread ancient distribution of this subspecies and the emergence of domestic forms across a broad geographic range, illustrating a range-wide shift to human landscapes as an adaptive strategy; or (ii) prolific human-mediated dispersal of this taxon throughout Mesoamerica, illustrating the massive reproductive and dispersal potential for plants under cultivation. Either possibility underscores the adaptive potential of domestication (36). One ancient C. pepo ssp. fraterna sample from northeastern Mexico also indicates the possible presence of this taxon in ancient crop fields. Presently, C. pepo ssp. fraterna exists only in the wild, suggesting that our ancient fraterna sample may represent a lost domestic lineage or an independent, subsequently arrested, pathway to domestication.

Reproductive Isolation.

The fixation of domesticated phenotypes requires a level of reproductive isolation between cultivated and wild forms to prevent the constant back-crossing of unfavorable traits into populations under human selection (37, 38). For example, emmer wheat required removal from its source region to express the phenotypic effects of domestication robustly (39), and it has been suggested that maize flourished into its dramatic, familiar, hyper-domesticated forms only after being transported out of southern Mesoamerica (40). This requirement may vary by species, with the level of isolation being linked to the genetic configuration and cultural importance of domestication traits. In Cucurbita, the requirement for reproductive isolation is especially pronounced, with cross-fertilization from wild forms producing unpalatable fruits (10), likely through the up-regulation of cucurbitacin production. That is, without isolation from wild stands, it is unlikely that Cucurbita could have evolved into a form suitable for human consumption. During the Holocene, the collapse of wild populations following the onset of a relationship with humans ultimately may have prevented gene flow from wild types constantly recharging crop fields with the allelic predisposition for a tough rind and bitter cucurbitacins. Cucurbita species have thrived in the Holocene largely by partnering with humans through domestication as wild populations dwindled and have been bred to lower their natural defenses for human palatability. At the same time, the fixation of important domestication alleles conferring palatability might have been impossible if wild plants had not largely receded and disappeared.

Materials and Methods

Plastid Genome Analysis.

Sample materials.

Details of all modern and ancient Cucurbita samples analyzed are reported in Tables S1–S3. Modern samples were provided by L.A.N. and the US Department of Agriculture Agricultural Research Service National Plant Germplasm System. Access to ancient samples was granted by B.D.S., the Illinois State Museum, the National Park Service, and the University Museum at the University of Arkansas.

DNA isolation, sequencing, and assembly.

In modern samples, we used sterile razor blades to excise embryonic tissue from seeds and extracted whole genomic DNA from the embryos using Qiagen DNeasy Plant Mini Kits, following the manufacturer’s protocol after grinding tissue in the kit lysis buffer with sterile pellet pestles. In modern-sample representatives from each separate taxon, we first PCR-amplified long, overlapping fragments of the plastid genome with primers in conserved regions (see Table S4 for primer sequences) designed using an alignment of cucumber (Cucumis sativus, Cucurbitaceae) and karaka (Corynocarpus laevigatus, Corynocarpaceae) plastid genomes, both order Cucurbitales. We were unable to amplify complete plastid genomes in any taxon using long-range PCR (LR-PCR), but we recovered several long fragments to assist with plastid genome assembly. We separately sheared LR-PCR product pools and whole genomic DNA to ∼350 bp using a Covaris model S2, size-selected the sheared DNA at ∼350 bp on a 1.6% low-melt agarose gel, and prepared barcoded Illumina libraries (41). We pooled all barcoded libraries equally among samples, with 20% of estimated pool molarity comprising LR-PCR products and the remainder whole genomic DNA. We sequenced the pool on one lane of an Illumina HiSeq 2000 flow cell with a 101-nt paired-end configuration. To acquire additional C. pepo and C. moschata plastid genome sequences to compare with the initial assemblies, we shotgun-sequenced eight additional plastid genomes on one lane of an Illumina HiSeq 2500 flow cell with a 151-nt paired-end configuration in Rapid Run mode.

Table S4.

LR-PCR primer sequences

Primer ID	Primer sequence (5′ – 3′)
CuLR1F_14	ACGGGAATTGAACCCGCGCATGGTG
CuLR1R_10234	GCGGGAATCGAACCCGCATCGTTAGC
CuLR2F_7978	GCCAAATTGCCCGAGGCCTACGCTTT
CuLR2R_19445	GCCGAGTAGGCGGATTGGTCCGAGT
CuLR3F_17745	AGATCGCTTCTTCCAAAGCACGCCC
CuLR3R_28990	ATTTGCAGTCCCCCGCCTTACCGCT
CuLR4F_27271	CCTACTCACACGAGCCCATATCCTTGCT
CuLR4R_36971	GGAGCACGCAGATCCCAAAAACGCA
CuLR5F_35924	TGGGCCGGGAATGCCCGACTTATCA
CuLR5R_44861	TTCCGACCGGGGAGAACAGGCCATT
CuLR6F_43879	CTCGAGAAATGACCCGGTCTGGCCC
CuLR6R_53080	CGGGAATAGAACCAATGGGCGATGCT
CuLR7F_51582	GGGCCATCCTACCCAACTTTCAGGCA
CuLR7R_61940	GCCGGAAATACTAGGCCCACTAAAGGCA
CuLR8F_60937	GGAGGAGCACGCATGCAAGAAGGAAG
CuLR8R_70507	TCGGGGGCAAACGCCTACGAAAAGA
CuLR9F_70068	TGGCCAAGGGTAAAGATGCCCGAGT
CuLR9R_79605	CATTGGGCCATGCGGGTTCTCCGTA
CuLR10F_79451	TCATGTCCGGTTCCTTCGGGGGATGG
CuLR10R_88670	ACTGGATGCACGCCAATGGGACCCT
CuLR11F_IR_88137	CTACGGCTCCATTGCGTGTGCTCGG
CuLR11R_IR_97912	TGGGTTCAAGCTTTCCCCAGCCCCT
CuLR12F_IR_97208	CGTGCATGAGACTTTCATCTCGCACGGC
CuLR12R_IR_107874	TTCCTTGACCTTCCGGCACTGGGCA
CuLR13F_107734	AAGGGGTGCCTCCTCACAAAGGGGG
CuLR13R_117642	TCATGGATTTATTGGCGCTGCGCTT
CuLR14F_115832	ATCTCGGTTCGAGTCCGAGTGGCGG
CuLR14R_124372	GCTTCTCATCTGTTATGGCTTGGCCCT
CuLR15F_124203	AGCCGCGACTCCCCCGATACGAAAA
CuLR15R_134752	CGCCGAAGATGAACGGGGCTAAGCG
CuLR16F_end_146561	TGAATAGGACGAACCGCCCCGTGGT
CuLR16R_end_668	TAGCTGGTGTATTCGGCGGCTCCCT
CuLR_alt_8F_61002	TGGAAAAGCAAGTTGATCGGTTAAT
CuLR_alt_8R_70245	CTTCTAGAATCCTTTGTTCCCTTCA
CuLR_alt_7F_52080	CTAAGCGGGCTCACATAACAGAAAT
CuLR_alt_7R_61259	TAGAGTTCGGGTTCGAATTCCATAG
CuLR_alt_5F_36058	GTGGTAGAGTAACGCCATGGTAAGG
CuLR_alt_5R_44044	AATCTTTATGCTCAAAACCCCGATT
CuLR_alt_2F_10163	GTCAGTACCTAGCCGGGCTTTTT
CuLR_alt_2R_18853	GAGATGATGGAAGCAGGAGTTCATT
CuLR_alt_13BF_111751	CTCCCAATTGGTTGGACCGTAG
CuLR_alt_13BR_116450	AGAAAAGTGGGAGAGAAGGGGTTT

We divided LR-PCR–derived sequence data into groups of 100,000 read pairs and shotgun sequence data into 1,000,000 pair subsets to normalize coverage for de novo assembly. We used Velvet 1.2.08 to iterate over LR-PCR subsets to produce long, high-quality contigs from the high-coverage enriched plastid regions. We combined all LR-PCR contigs per sample and then used Velvet (42) to iterate over shotgun data subsets using stringent parameters (long k-mer value and high coverage cutoff) and included LR-PCR contigs as long reads to anchor the assembly. We combined all resulting contigs and reiterated over shotgun subsets using new contigs as long reads until no substantial increase in contig length and quality was observed (two to five iterations). We used Burrows–Wheeler Aligner (BWA) (43) to align resulting contigs from one taxon, Cucurbita pepo ssp. fraterna, to the plastid genome of cantaloupe (Cucumis melo, Cucurbitaceae), a relative with an estimated 30-Mya divergence date from Cucurbita (44). We then used the resulting alignment to identify contigs covering the majority of the plastid genome and used pairwise alignment to assemble them to the cantaloupe reference sequence. We used bash utilities to close the remaining short gaps by reiteratively querying contig terminal sequences from the short-read data and realigning matching short reads to the working assembly until reaching the next adjacent contig. We used YASRA (45) with LASTZ (46) to map shotgun reads from other taxa to the circularized C. pepo ssp. fraterna plastid reference assembly reiteratively and closed gaps using de novo contigs and manual gap-filling as described above. Two short stretches (estimated <200 nt based on comparison with the cantaloupe reference sequence) in the long single-copy region containing tandem repeats were unresolvable using short-read data, one in the rps4-ndhJ intergenic spacer region and the other apparently within the coding region of accD. These regions were hard-masked for analysis. After the initial assembly of plastid genomes across all taxa, we used BWA (43) and SAMtools (47) to align short reads from subsequent samples to the appropriate reference and call consensus sequences.

Following plastid genome assembly for all available taxa, we used the C. pepo ssp. pepo and C. moschata plastid consensus sequences to design target capture probes for plastid DNA enrichment by RNA hybridization (18). We extracted DNA and prepared sequencing libraries from 22 additional modern specimens of C. pepo and 23 additional modern C. moschata. In addition, we extracted DNA from 33 archaeological Cucurbita rind, seed, and peduncle specimens using an N-phenacylthilazolium bromide (PTB)-based extraction protocol (48) and prepared sequencing libraries (41). Because of the fragmentary, low-copy, and contamination-prone nature of ancient DNA, stringent protocols were observed during ancient DNA extraction and library preparation. All pre-PCR handling of ancient tissue and DNA was carried out in a dedicated sterile laboratory physically isolated from any molecular biology building, with HEPA filtration, positive pressure, frequent decontamination using a strong bleach solution, and workflow protocols to prevent modern contamination of the clean laboratory. We enriched modern and ancient DNA sequencing libraries for plastid DNA using the MYbaits kit (Mycroarray), according to the manufacturer’s protocol but with two extra wash repetitions and an 18-cycle PCR reamplification using primers IS5 and IS6 (41) after hybridization. For the modern samples, we combined the barcoded libraries into pools of 10 samples before the capture step and carried each pool through a single capture reaction. The enriched libraries were sequenced in parallel on an Illumina HiSeq 2500.

For modern libraries, we used BWA (43) to map reads to the appropriate reference plastid assembly and then used the SAMtools (47) mpileup function and a custom Perl script to call majority-rule haplotype consensus sequences. For ancient samples, we first merged overlapping paired-end reads after Kircher (49) and then BWA-mapped (43) merged reads to the C. pepo ssp. fraterna plastid reference sequence. We estimated the level of cytosine deamination at fragment ends—characteristic ancient DNA damage most prevalent in single-stranded overhangs (50)—by analyzing base mismatches to the reference sequence relative to base position on the fragment, revealing an average 8-nt interval of elevated 5′ C->T bias and 3′ G->A bias (Fig. S1). This observation helped authenticate our ancient DNA data and indicated that the region within 8 nt of read ends, on average, might be prone to damage-derived sequence error. To mitigate any effects of these biases in our analysis, we hard-masked all potentially damaged positions (5′ thymines and 3′ adenines) within appropriate distance of fragment ends (Fig. S1 and Tables S1–S3). We then called haplotype consensus sequences as above, enforcing a minimum 2× nonredundant coverage and 80% site identity for a consensus base call. Under these requirements, we recovered plastid assemblies between 10.9% and 99.8% complete for all ancient samples (Tables S1–S3). We discarded samples with <65% coverage of the single-copy regions and one of the two inverted repeat regions, leaving a total of 19 ancient samples for analysis, 15 of which had ≥97% coverage.

Fig. S1.

DNA misincorporation patterns resulting from ancient DNA deamination. X-axis: distance on mapped read from 5′ (Left) and 3′ (Right) termini; Y-axis: proportion of reference C matched with read T (Left, red), and reference G match with read A (Right, blue).

We aligned all plastid genome sequences (n = 91) using MAFFT (51), manually checked the alignment, and manually masked any short regions of poor mapping quality caused by long homopolymer repeats prone to polymerase and sequencing error (aligned and curated sequences provided as Dataset S1). We used a log-likelihood ratio test to select the Hasegawa, Kishino, and Yano (HKY) substitution model with four gamma-distributed rate categories, and used PhyML (52) with 100 bootstrap replicates to reconstruct a phylogenetic tree (Fig. 1). We also confirmed the tree topology by running a Bayesian phylogenetic analysis to convergence using BEAST 2 (53), enforcing a strict molecular clock and implementing the same evolutionary model described for the maximum likelihood analysis.

Mammalian TAS2R Analysis.

TAS2R gene identification.

We retrieved 46 mammal genome assemblies from public repositories (Tables S5 and S6), representing all scaffold- or chromosome-level therian mammal genome assemblies available at the time of data acquisition (April, 2014), excluding the bottlenose dolphin. Extant whales possess only pseudogenized TAS2R homologs because of the complete loss of functional TAS2R genes in a common ancestor living 36–53 Mya (33). Therefore, dolphins have never possessed a suite of functional bitter taste receptors (33), rendering analysis of subsequent evolution of functional TAS2R suites alongside phenotypic and behavioral variables moot. To identify intact, putatively functional TAS2R gene copies in the mammalian genomes, we first adapted the strategy of Li and Zhang (30) by generating BLAST databases (54) from each genome assembly and queried them for the annotated human (n = 25) and mouse (n = 35) TAS2R protein sequences (30) using TBlastN, enforcing an e-value threshold of ≤10⁻¹⁰. This approach previously has proven robust for localizing TAS2R variants across diverse species, with conserved transmembrane domains in TAS2R genes providing reliable amino acid targets for the TBlastN queries (30). From the TBlastN output, we generated a bed file for each species containing genomic coordinates of candidate TAS2R regions, merged overlapping regions, added 100 nucleotides to each flank, and used BEDtools (55) to generate nucleotide FASTA files for the corresponding regions from the genome assemblies. We then used a Perl script to translate these nucleotide data into amino acid sequences in all six reading frames and return the longest putative intact gene from each region, defined previously (30) as having at least 270 amino acid residues, starting and ending with start and stop codons, and being uninterrupted by premature stop codons. We used MAFFT (51) to align all resulting protein sequences and visually inspected the result to ensure plausible multiple alignment among all copies, validating these genes as TAS2R homologs. This strategy resulted in the recovery of 933 intact TAS2R copies among the 46 species analyzed. To minimize any remaining taxonomic or dietary acquisition biases associated with using human and mouse query sequences, we repeated the entire screening process a second time, using all TAS2R genes recovered in the first pass as input queries. In the second pass, we recovered 20 additional genes across 14 different species, bringing the total TAS2R count to 953 to carry into our analyses (provided as Datasets S2 and S3).

Table S5.

Mammal genome assemblies analyzed, TAS2R count, and assembly quality information

Species	Common name	Assembly	TAS2R copies, first pass	TAS2R copies, second pass	Body mass estimate, kg	Diet specialization	Scaffold N50	Included
Ailuropoda melanoleuca	Giant panda	ailMel1	16	16	102.5	4	1281781	Yes
Bos taurus	Cow	bosTau7	21	21	755	3	2599288	Yes
Callithrix jacchus	Common marmoset	calJac1	21	23	0.33	3	5167444	Yes
Canis familiaris	Dog	canFam3	15	16	51.5	3	45876610	Yes
Cavia porcellus	Guinea pig	cavPor3	31	31	0.9	3	27942054	Yes
Ceratotherium simum	White rhinoceros	cerSim1	27	27	2,520	3	26277727	Yes
Choloepus hoffmanni	Two-toed sloth	choHof1	4	4	6	4	9667	No
Dasypus novemcinctus	Nine-banded armadillo	dasNov3	10	10	5.65	2	1687935	Yes
Daubentonia madagascariensis	Aye-aye	dauMad	10	11	2.5925	4	13597	No
Dipodomys ordii	Ord's kangaroo rat	dipOrd1	9	10	0.0755	2	36427	No
Echinops telfairi	Lesser hedgehog tenrec	echTel2	14	17	0.2025	2	45764842	Yes
Equus caballus	Horse	equCab2	21	22	1,150	3	46749900	Yes
Erinaceus europaeus	Western European hedgehog	eriEur2	20	20	1	2	3264618	Yes
Felis catus	Cat	felCat5	14	14	4.75	3	4658941	Yes
Gorilla gorilla gorilla	Western lowland gorilla	gorGor3	24	24	180	3	913458	Yes
Heterocephalus glaber	Naked mole rat	hetGla2	21	21	0.055	4	20532749	Yes
Homo sapiens	Human	hg19	24	24	70	1	46395641	Yes
Loxodonta africana	African bush elephant	loxAfr3	20	20	4,540	2	46401353	Yes
Macaca mulatta	Rhesus macaque	rheMac3	25	25	8	2	1660975	Yes
Macropus eugenii	Tammar wallaby	macEug2	19	20	6.55	3	36602	No
Microcebus murinus	Gray mouse lemur	micMur1	10	10	0.06	2	140884	No
Monodelphis domestica	Gray short-tailed possum	monDom5	27	27	0.1225	2	59809810	Yes
Mus musculus	House mouse	mm10	36	36	0.021	1	52589046	Yes
Mustela putorius furo	Ferret	musFur1	15	15	1.5	3	9335154	Yes
Myotis lucifugus	Little brown bat	myoLuc2	28	29	0.0095	3	4293315	Yes
Nomascus leucogenys	Northern white-cheeked gibbon	nomLeu3	17	17	5.7	3	52956880	Yes
Ochotona princeps	American pika	ochPri2	15	15	0.1485	2	88760	No
Oryctolagus cuniculus	European rabbit	oryCun2	28	29	2	2	35972871	Yes
Otolemur garnettii	Northern greater galago	otoGar3	22	22	0.7715	3	13852661	Yes
Ovis aries	Sheep	oviAri3	15	15	110	3	100079507	Yes
Pan troglodytes	Chimpanzee	panTro4	26	26	48	1	8925874	Yes
Papio anubis	Olive baboon	papAnu2	29	29	19.5	1	528927	Yes
Pongo abelii	Sumatran orangutan	ponAbe2	23	24	60	3	747460	Yes
Procavia capensis	Rock hyrax	proCap1	15	15	4	2	24297	No
Pteropus vampyrus	Large flying fox	pteVam1	16	16	0.85	4	124060	No
Rattus norvegicus	Brown rat	rn5	36	36	0.32	1	2178346	Yes
Saimiri boliviensis	Black-capped squirrel monkey	saiBol1	22	22	0.75	2	18744880	Yes
Sarcophilus harrisii	Tasmanian devil	sarHar1	19	19	8	3	1847106	Yes
Sorex araneus	Common shrew	sorAra2	44	46	0.0095	2	22794405	Yes
Spermophilus tridecemlineatus	Thirteen-lined ground squirrel	speTri2	19	19	0.125	2	8192786	Yes
Sus scrofa	Pig	susScr3	15	16	169	1	576008	Yes
Tarsius syrichta	Philippine tarsier	tarSyr2	23	25	0.125	3	401181	Yes
Trichechus manatus latirostris	West Indian manatee	triMan1	6	8	400	3	14442683	Yes
Tupaia chinensis	Chinese tree shrew	tupChi_1.0	35	35	0.385	2	3670124	Yes
Ursus maritimus	Polar bear	ursMar	14	14	475	4	16000000	Yes
Vicugna pacos	Alpaca	vicPac2	12	12	60	4	7263804	Yes

Table S6.

Species and dietary notes

Species	Dietary notes
Ailuropoda melanoleuca	>99% bamboo, rarely some fruits and small animals
Bos taurus	Grass specialist, some tougher vegetation
Callithrix jacchus	Mostly plant exudates, various other miscellany
Canis familiaris	Terrestrial vertebrates
Cavia porcellus	Grass specialist, but some other plants
Ceratotherium simum	Grazer
Choloepus hoffmanni	Strict folivore
Dasypus novemcinctus	Generalist carnivore
Daubentonia madagascariensis	Canarium and larvae only
Dipodomys ordii	Primarily seeds, grasshoppers and moths seasonally
Echinops telfairi	Invertebrates, molluscs, fruit
Equus caballus	Grazer
Erinaceus europaeus	Omnivorous, primarily insects
Felis catus	Terrestrial vertebrates
Gorilla gorilla gorilla	Primarily folivore, some other plants and occasional insects
Heterocephalus glaber	Tuberous structures
Homo sapiens	Broad-spectrum omnivore
Loxodonta africana	Generalist herbivore
Macaca mulatta	Generalist herbivore
Macropus eugenii	Grass specialist
Microcebus murinus	Primarily insects, small reptiles and amphibians, some plants
Monodelphis domestica	Generalist omnivore
Mus musculus	Broad-spectrum omnivore
Mustela putorius furo	Terrestrial vertebrates
Myotis lucifugus	Insects only
Nomascus leucogenys	Fruit, some other plant matter, and insects
Ochotona princeps	Generalist herbivore
Oryctolagus cuniculus	Generalist herbivore
Otolemur garnettii	Fruit and insects
Ovis aries	Grass specialist
Pan troglodytes	Broad-spectrum omnivore
Papio anubis	Broad-spectrum omnivore
Pongo abelii	Primarily fruit, other plant matter, and insects/eggs to 5%
Procavia capensis	Generalist herbivore
Pteropus vampyrus	Flowers, nectar, fruit
Rattus norvegicus	Broad-spectrum omnivore
Saimiri boliviensis	Primarily fruit and insects, but with other contributers
Sarcophilus harrisii	Scavenger carnivore
Sorex araneus	Invertebrates only
Spermophilus tridecemlineatus	Generalist omnivore
Sus scrofa	Broad-spectrum omnivore
Tarsius syrichta	Obligate carnivore, birds, reptiles and amphibians, insects, arthropods
Trichechus manatus latirostris	Sea grasses, occasional fish and invertebrates
Tupaia chinensis	General omnivore, but some uncertainty and interspecies variation
Ursus maritimus	Terrestrial vertebrates, prefers sea mammals, lipivore
Vicugna pacos	Grazer with strict physiological requirements

Phenotypic character assignments.

Body mass was assigned based on typical values for adult males of each species from the University of Michigan Animal Diversity Web (ADW) (56) or from adults in general when sexual dimorphism data were not available. For dietary breadth, we assigned each mammal species a value from 1 to 4, where 1 signifies a broad-spectrum dietary generalist (omnivorous species with a wide range of foraging behaviors, e.g., mouse, pig, human) and 4 indicates a narrow dietary specialist (species with extremely specialized dietary behaviors, especially species for which the diet may consist almost entirely of a single species or a few similar species, e.g., bamboo for the giant panda or highland grasses for the alpaca), using ADW descriptions of diet as guidelines (Tables S5 and S6). Diet breadth assignations were made without prior knowledge of TAS2R count to avoid undue influence on category choice.

Analysis of TAS2R copy number variation.

To test for the effects of genome assembly quality on the sensitivity of our technique to detect TAS2R copies in poorer assemblies, we first tested for a relationship between log₁₀ scaffold N50 and TAS2R count using a linear model alongside log₁₀ body mass and diet breadth, using the R statistical package (57). We identified a highly significant correlation between scaffold N50 and TAS2R count (P = 0.003). This effect apparently is caused by difficulty in identifying intact TAS2R variants in more fragmentary, lower-quality genome assemblies. When assemblies with N50 <50 kb (n = 5) were excluded, significance decreased dramatically (P = 0.103). Conservatively, we chose to exclude assemblies with N50 <250 kb (n = 8) from analysis, eliminating the significant relationship between assembly quality and intact TAS2R count (P = 0.863). To analyze associations between phenotypic characters and TAS2R gene count while correcting for phylogenetic influence, we implemented a phylogenetic generalized least squares (pGLS) approach within the “caper” R package (58). We constructed an ultrametric phylogenetic mammal supertree manually from the literature (Fig. S2 and Dataset S4) (59–61). We used maximum likelihood optimizations of the delta, kappa, and lambda branch length transformations implemented in the pGLS (detailed in ref. 58).

Fig. S2.

Ultrametric supertree used for pGLS analysis of 38 mammal genomes.