Introduction

The independent secondary transition to water by terrestrial ancestors of the marine mammal groups Cetacea (whales, dolphins, and porpoises), Sirenia (manatees and the dugong), and Pinnipedia (seals, sea lions, and walruses) is a classic example of macroevolution1. Cetaceans emerged around 55 million years ago (Mya), approximately five million years after sirenians and 20 Mya before pinnipeds2,3,4. Among marine mammals, cetaceans are today the most numerous, with ~100 species (15 baleen whale species and 85 toothed whale species) resulting from a rapid proliferation of crown cetacean taxa over ~15 Mya beginning at the early Oligocene (~34 Mya) argued to be driven by paleogeographic transitions and major shifts in ocean climate and productivity4,5,6.

Across the animal kingdom, the forms seen today result from millions of years of evolution and are tightly associated with ecological shifts that manifested as evolutionary steps with intermediate forms7. For example, morphological studies have shed light on auditory system innovations since the first cetacean (stem lineage) emerged ~55 Mya, revealing the origin of underwater hearing during the early Eocene (~51–48 Mya), high-frequency hearing in the first toothed whales during the late Eocene (~36 Mya), and ultrasonic hearing and echolocation in toothed whales in the mid-Oligocene (27–24 Mya)5,8,9.

The integumentary system (skin and its appendages) of cetaceans is no exception of a stepwise evolutionary process. Maintaining a stable body temperature (homeothermy) is essential for metabolism and to inhabit diverse environments. Water conducts heat from the body much faster than air10 and temperature decreases with water depth11,12. It follows that marine mammals must have evolved adaptations to minimize heat loss and provide mechanical protection. To this end, the integument of modern cetaceans has thick outer (epidermis) and inner (hypodermis) layers11,13. Recent molecular genetics literature shows that the integumentary system largely evolved through gene loss events as the earliest cetaceans became obligate marine14,15. The studies support a restructuring of the epidermis and dermis and loss of skin appendages (e.g., sweat glands and hair follicles)15,16,17,18,19,20,21,22,23. In contrast, and somewhat surprising given its importance, little is known about the gene repertoire that underlies the evolution of the cetacean hypodermis, blubber. Blubber is a hydrostatic pressure-resistant and insulating vascularized layer of subcutaneous adipose tissue also found in sirenians and pinnipeds. The outer blubber is thermoregulatory and is not appreciably compressed at depth (i.e., it is mainly structural), while the inner blubber serves as an energy supply (it is more metabolically active, like terrestrial mammal adipose tissue)24,25,26.

Studies in the last decade suggest that cetaceans have an altered energy metabolism that presumably underlies their thickened blubber (e.g., see refs. 27,28,29,30), but the key underlying genes have remained largely unknown. Recent studies provide some insights into the molecular basis of blubber in marine mammals26,29,30,31,32,33,34,35. Most reports are on UCP1 (uncoupling protein 1), expressed by brown (and brown-like) adipocytes, the primary effector of heat generation in response to cold by adipose tissue (termed non-shivering thermogenesis)36. Inactivating UCP1 mutations in mammals have been associated with the evolution of a large body size, a morphological trait that decreases heat loss and the reliance on energetically costly cold-induced non-shivering thermogenesis35. UCP1 is intact in all pinnipeds except for the large-bodied elephant seals that emerged in the late Pliocene ~25 million years after the first pinniped37. The gene became inactivated in paenungulates ~58 Mya, in a common ancestor of terrestrial hyraxes (whose early Eocene ancestor was substantially larger) and elephants, and fully aquatic sirenians23,35,38. Some paenungulate species evolved additional cold adaptations during later climatic cooling, as illustrated by the fur of woolly mammoths in the early Pliocene (~5 Mya)39,40 and the thick, hyperkeratotic epidermis and gigantism of the Steller’s sea cow41.

Cetacean UCP1 mutations first arose ~53 Mya32,35 in a semi-aquatic stem lineage with a small body size compared to most extant cetaceans (e.g., the wolf-sized Pakicetus5). The early Eocene was a time of global warming, with sea surface temperatures (SSTs) ~10 °C higher than today42. By the late Eocene (~36 Mya), fully aquatic cetaceans had emerged, including the crown cetacean lineage4,5. Paraphyletic stem cetaceans (archaeocetes) and crown cetaceans had also dispersed towards colder higher latitudes (e.g., North America and Antarctica)43,44,45. This time was characterized by decreasing global temperatures that culminated in the Eocene/Oligocene Transition Antarctic ice sheet expansion 34–33 Mya46,47,48. While later crown cetacean lineages increased in size, a shift towards gigantism does not appear to have been the norm in the late Eocene/early Oliogene49. For example, whereas the crown cetacean Llanocetus spp. of Antarctica evolved a large body size, its smaller conspecifics in South America and other cetacean species in Antarctica and elsewhere did not44,50. Until recently, it was assumed that UCP1 function was lost in all cetaceans because it evolved under relaxed selection32 and harbors inactivating (amorphic) mutations33,34,35. However, mouse UCP1 with cetacean mutations (a premature stop codon and a 4-bp indel) appears to encode a partially functional thermogenic protein (i.e., the gene harbors hypomorphic mutations), and immunohistochemical and western blot analysis on Yangtze River dolphin (Lipotes vexillifer) blubber also revealed UCP1 expression32. The gene likely became completely inactivated in some lineages, including the superfamilies Inioidea (Pontoporia + Inia) and the species-rich Delphinidae, long after the origin of the crown cetacean clade32. Because cetacean UCP1 appears to have retained some thermogenic function in stem cetaceans and most genes associated with their thick epidermis were lost by the early–mid Eocene15, it begs the question of how cetaceans adapted to later colder waters if gigantism was not the norm.

Adipose tissue, the insulating component of blubber, is a heterogeneous and dynamic endocrine organ that responds to stimuli such as cold-induced β-adrenergic receptor activation to meet metabolic and energetic demands by adipocyte differentiation and remodeling with a minimal change in cell number51. Mammals possess functionally and morphologically distinct adipose tissue cell types—broadly white, brown, and brown-like (also known as beige or brite) adipocytes36. White adipocytes are specialized for the release and storage of lipids (primarily triacylglycerol, TAG) within lipid droplets, while brown and brown-like adipocytes release energy in the form of heat via catabolism of stored lipids52. In this work, we hypothesized that adipose tissue-associated genes other than UCP1 lost partial or complete function and allowed cetaceans to maintain body temperature in colder waters by increasing blubber thickness (i.e., regulating adipose tissue homeostasis) instead of body size. Our comparative genomics analysis identified inactivating mutations in the β3-adrenergic receptor (ADRB3), a master adipose tissue regulator induced by cold exposure, in the crown cetacean lineage. Cetacean ADRB3 is functionally inactive and likely restrain adipose tissue UCP1-mediated thermogenesis and lipolysis to promote lipid storage (thermal insulation). We propose that its loss was adaptive and necessary for cetacean survival, radiations, and diversifications during a time of unprecedented cooling of the world’s oceans.

Results

Cetaceans have candidate inactivating mutations in the β3-adrenergic receptor gene

We employed a recently established computational pipeline53 to identify 85 candidate gene loss events in cetacean genomes (Supplementary Data 1). The literature mining tool Rummagene54 revealed the cetacean loss candidates showed enrichment for genes previously reported to be inactivated in a study of toothed and baleen whales17 (11 genes; Fisher’s exact test Benjamini-Hochberg P = 2.79 × 10−8; Supplementary Data 2). Shared losses with our data included genes associated with salivation (SLC4A9) and thrombosis (F12; coagulation factor XII), as well as long-chain fatty acid elongase 3 (ELOVL3). ELOVL3 was first described as a gene restricted to sebaceous glands and the epithelial cells of hair follicles55, and its loss by cetaceans has been attributed to their lack of these skin appendages15,19, but the gene is also a marker of cold-induced adipose tissue thermogenesis56,57,58,59. Because the gene is highly conserved, other than a shared premature stop codon, and showed no evidence of relaxed selection, the inactivation of ELOVL3 could not be reliably dated beyond that it must have occurred before the divergence of toothed and baleen whales in the late Eocene5. Manual inspection of our 85-gene set further revealed ADRB3 (β3-adrenergic receptor), a gene essential for brown adipocyte lipolysis and thermogenesis that acts upstream of ELOVL3 and UCP156.

At first glance, ADRB3 appeared intact in cetaceans since NCBI RefSeq gene annotations are available but a closer look at the records revealed otherwise (Supplementary Note 1 and Fig. S1). We compared cetaceans with 98 other mammals, revealing extensive ADRB3 loss in the cetacean clade (Fig. S2). An analysis of genome assemblies and sequencing reads showed that, in addition to various lineage-specific mutations, an inactivating mutation (a 1-bp deletion at the 118th base position of the first exon) is present in all cetaceans except for beaked whales (Ziphiidae; exon region missing due to a large indel) and river dolphins in the superfamilies Inioidea (Pontoporia + Inia) and Lipotoidea (Lipotes) (Fig. 1; Supplementary Figs. S3 and S4). The deletion at nucleotide 118 in most cetaceans was validated by interrogating genome sequencing reads (Table S2) and by Sanger sequencing polymerase chain reaction (PCR) amplicons (Fig. 1). Lipotes vexillifer, Pontoporia blainvillei, and Inia geoffrensis ADRB3 show evidence of reversion to the ancestral guanine at nucleotide 118 (i.e., a back-mutation), a rare event in species with a long generation time and low population size60,61,62. While L. vexillifer and P. blainvillei acquired additional ADRB3 inactivating mutations, I. geoffrensis appears to have an intact gene (Supplementary Note 2). At first, this observation seems paradoxical. However, the downstream ADRB3 effector gene UCP1 has several large mutations in the Inioidea superfamily, likely inactivating it, in contrast to its partially functional L. vexillifer counterpart (ref. 32 and Fig. S5). Although the blubber of I. geoffrensis is thin, it is as thick as that of Lipotes vexillifer (~20 mm dorsal fin blubber63,64). Together with a complete loss of UCP1, we speculate that an intact ADRB3 reflects that the Inea lineage has been restricted to warmer waters than Lipotes and Pontoporia.

Fig. 1: Loss of β3-adrenergic receptor (ADRB3) in the crown cetacean lineage.
Fig. 1: Loss of β3-adrenergic receptor (ADRB3) in the crown cetacean lineage.
Full size image

Overview of gene-inactivating mutations in the cetacean lineage. The two coding exons of ADRB3 are represented using open rectangles. TAG and TAA indicate a premature stop codon; numbers indicate the size of small indels; and a dotted line indicates large exon deletions. The Sanger sequencing chromatograph shows ADRB3 frameshift mutations in common bottlenose dolphin (Tursiops truncatus), Indo-Pacific humpback dolphin (Sousa chinensis), and humpback whale (Megaptera novaeangliae). The gray callout box shows validation by reads from the NCBI Sequence Read Archive (SRA), with a 1-bp deletion at nucleotide 118 shown in red. Silhouette images are from PhyloPic (CC BY 4.0).

Five models implemented in PAML codeml65 were used to examine the selection acting on cetacean ADRB3 (Table S3). In model A, the overall selection pressure for all species was estimated to be ~0.16 (P < 0.05), which is significantly smaller than 1, indicating strong purifying selection in mammals. Moreover, cetacean ADRB3 had a ω2 of ~0.66 (model C; P < 0.05), which is significantly higher than the ω value (~0.13) observed from lineages with intact ADRB3 coding sequences, indicating that the cetacean gene has evolved under relaxed selection. We compared models C and D, which assigns a fixed ω2 = 1.00 in pseudogenized branches, to further evaluate selective pressures. This analysis showed that functional constraints on ADRB3 are not completely removed in cetaceans (model D vs model C; P < 0.05). RELAX (general descriptive model)66 showed relaxed selection in cetaceans (K = 0.14 and P = 0.0001; Fig. S6). These results suggest that cetacean ADRB3 has evolved under some functional constraint, as previously reported for cetacean taste receptor genes67. ADRB3 is a G protein-coupled receptor (GPCR), a protein composed of seven transmembrane domains (Fig. S3). It has been argued that the selection pressures on GPCRs, including the related ADRB2, reflect their structure and depend on residues buried and exposed to the plasma membrane and ligands68,69.

To investigate when ADRB3 was lost by cetaceans, we used a method described by Sharma and colleagues based on the rate of nonsynonymous to synonymous substitutions and species divergence times70. The gene was estimated lost approximately ~32.0–34.4 million years ago (Table S4), within the age of crown cetaceans, the most recent common ancestor of toothed and baleen whales5.

Cetacean ADRB3 is functionally inactive

Given that cetacean ADRB3 would encode a protein lacking all seven β3-adrenergic receptor transmembrane domains—the common bottlenose dolphin open reading frame is disrupted after 57 amino acids, while the mouse encodes a 377-AA protein (Fig. S1)—we hypothesized it is a gene devoid of function.

Environmental (cold-increased ADRB3 gene expression) or chemical β3-adrenergic stimuli promote the differentiation of brown-like adipocytes from mature white adipocytes and preadipocytes within white adipose tissue (WAT)—termed “browning”. A notable feature of brown-like adipocytes is that they can revert back to white adipocyte-like cells (termed “whitening”) and, like brown adipocytes, harbor numerous small lipid droplets compared to the large single lipid droplet of white adipocytes52. To evaluate cetacean ADRB3 function, we injected adeno-associated virus (AAV) vectors containing common bottlenose dolphin (AAV-Ttru-ADRB3; harbors a premature stop codon due to a one-base deletion) and mouse (AAV-Mmus-ADRB3) ADRB3 sequences into the inguinal WAT (iWAT) of adult male C57BL/6J mice. After AAV transduction, we challenged the animals with a high-fat diet (60% fat) for six weeks to generate diet-induced obese (DIO) mice. We reasoned that iWAT cell populations and functional responses to a high-fat diet by our DIO mice expressing mouse and dolphin ADRB3 would be markedly different (Fig. 2a). A high-fat diet stimulates the differentiation of mouse brown-like adipocytes to white-like adipocytes51,71. In contrast, chemical ADRB3 agonists or forced ADRB3 overexpression stimulates the dedifferentiation of white adipocytes to brown-like adipocytes in addition to promoting lipolysis and cold-induced mitochondrial thermogenesis of resident brown-like adipocytes56,71,72.

Fig. 2: Comparative analysis demonstrates lack of in vivo function by cetacean ADRB3.
Fig. 2: Comparative analysis demonstrates lack of in vivo function by cetacean ADRB3.
Full size image

a Overview of the expected outcome on resident adipocyte pools in iWAT of mice overexpressing cetacean and mouse ADRB3 (β3-adrenergic receptor) after a six-week high-fat diet (HFD). An HFD induces the differentiation of brown-like adipocytes to lipid-storing white adipocytes, while ADRB3 promotes the differentiation of white adipocytes into thermogenic brown-like adipocytes and promotes lipid utilization. If cetacean ADRB3 is not functional, their iWAT should store more lipids. Illustrations of dolphin and mouse courtesy of 1983NatalyaTalala/Shutterstock.com and ArtoPhotoDesigno Studio/Shutterstock.com, respectively. b, c Mouse body weight and food intake. d Weight of iWAT, eWAT, BAT, and liver after six-weeks. e Representative HE staining images of mice iWAT after six weeks (scale bar, 50 μm). Lipid droplet size was quantified after staining with oil red O. f Expression of the adipose lipid storage gene Pparg (PPARγ) and the thermogenic brown/brown-like adipose marker genes Ppargc1a and Prdm16 after six weeks. g Serum levels of triglyceride (TAG), total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL) after six weeks. Data in all subfigures presented as mean expression ± S.D. relative to vector control, with circles, rectangles, and triangles representing individual data points (i.e., each represents one mouse) per experiment (ns denotes not significant, a P  >  0.05; *a P < 0.05). Source data are provided as a Source Data file. The P values are from the two-way ANOVA. Full statistical details and P values are provided in Table S6.

Quantitative RT-PCR confirmed successful AAV transduction (Fig. S7). AAV-Ttru-ADRB3 and control (empty AAV vector) mice gained significantly more weight than AAV-Mmus-ADRB3 mice (Fig. 2b) but there was no difference in food intake (Fig. 2c). No changes were observed in glucose tolerance tests (GTT) and insulin tolerance tests (ITT) at four and five weeks after injection, respectively (Fig. S8). To confirm that the body weight difference was due to iWAT lipid accumulation, we measured the weight of different tissues. Only iWAT showed a significant difference, with the AAV-Mmus-ADRB3 mice tissue weighing less than that of AAV-Ttru-ADRB3 and control mice (Fig. 2d). The mouse iWAT adipocytes of AAV-Mmus-ADRB3 mice had smaller lipid droplets (Fig. 2e). In agreement, the iWAT expression of the lipid storage gene Pparg (peroxisome proliferator-activated receptor gamma; PPARγ) suggested a shift towards lipid catabolism by AAV-Mmus-ADRB3 iWAT (Fig. 2f), while the expression of the thermogenic brown/brown-like marker genes Ppargc1a and Prdm1673 was higher (Fig. 2f). The mice injected with AAV-Mmus-ADRB3 also exhibited reduced serum TAG, but no difference in the levels of total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL) (Fig. 2g). Because we observed no change in the weight of other fat deposits, including the liver (Fig. 2d), the lower serum TAG level may reflect more efficient continuous uptake and lipolysis of dietary TAGs as fuel by Mmus-ADRB3 iWAT adipocytes.

We also examined cetacean ADRB3 in the 3T3-L1 cell line, a complementary model to study β3-adrenergic receptor function (see “Methods”). Cell assays and expression profiling revealed that only overexpression of mouse ADRB3 reflected the expected response to chronic β3-adrenergic receptor activation, while the dolphin form and mouse ADRB3 mutated to match the dolphin 1-base pair deletion at base 118 mirrored the vector control (Fig. 3 and Supplementary Note 3). Mmus-ADRB3 adipocytes had lower TAG levels (Fig. 3b) and immunoblotting showed a higher expression of adipose triglyceride lipase (ATGL; encoded by Atgl, also known as Pnpla2), an enzyme that initiates TAG breakdown from lipid droplets and whose stability is predominantly regulated at the protein level74,75 (Fig. 3c and Supplementary Fig. S9). A seminal difference between brown or brown-like adipocytes and white adipocytes is the relative abundance of mitochondria expressing UCP152. The 3T3-L1 cell line expresses this gene upon β3-adrenergic stimulation, albeit at low levels76. While our RNA-seq did not detect mouse UCP1 transcripts, immunoblotting confirmed higher UCP1 expression by Mmus-ADRB3 adipocytes (Fig. 3d and Supplementary Fig. S10). The ratio of mitochondrial to nuclear DNA (nDNA) (Fig. 3e) and mitochondrial reactive oxygen species (ROS) levels was also higher in Mmus-ADRB3 cells (Fig. 3f). Elevated mitochondrial ROS supports an adipocyte thermogenic identity77,78. In agreement with the cell assay data, the expression of Trib1 (Tribbles homolog 1) and Cmpk2 (cytidine monophosphate kinase 2) that regulate mitochondrial DNA (mtDNA) synthesis79,80 was higher in Mmus-ADRB3 adipocytes (Fig. 3a and Supplementary Fig. S10). Trib1 is essential for adipose tissue thermogenesis in mice and 3T3-L1 adipocytes79. These results and the in vivo data support the hypothesis that cetacean ADRB3 has been rendered non-functional.

Fig. 3: ADRB3 overexpression in the murine adipocyte cell line 3T3-L1 supports that the cetacean form is non-functional.
Fig. 3: ADRB3 overexpression in the murine adipocyte cell line 3T3-L1 supports that the cetacean form is non-functional.
Full size image

a Heat map using pairwise Euclidean distance of the expression profiles of 376 genes differentially expressed in Mmus-ADRB3 cells. The right panel highlights selected genes associated with adipogenesis and browning, and brown-like and brown adipocyte function. b Intracellular levels of triacylglycerol (TAG). c Top, a representative immunoblot of ATGL and GAPDH. Numbers on the right side of the gels represent molecular weight marker locations (values in kDa). Full images of blots in Supplementary Fig. S9. Bottom, bar graph showing ImageJ densitometry of three independent experiments, presented as mean arbitrary density units ± S.D. relative to empty vector (NC), while dots represent individual data points per experiment. d Top, a representative immunoblot of UCP1 and β-actin (ACTB). Bottom, densitometry. Annotated as in (e). Full images of blots in Fig. S9. e Number of mitochondria estimated using the ratio of mitochondrial DNA (mtDNA) to nuclear DNA (nDNA). f Mitochondrial ROS estimated using the fluorescent probe MitoSOX. NC denotes negative control, cells transfected with empty vector. Data in all subfigures presented as mean expression ± S.D. relative to vector control, with circles, rectangles, and triangles representing individual data points per experiment (ns denotes not significant, a P  >  0.05). Each dot in subfigures b–f represents an independent experiment. Source data are provided as a Source Data file. The P values are from the one-way ANOVA. Full statistical details and P values are provided in Table S6.

Discussion

Our in silico analyses, coupled with experimental validation, reveal that cetaceans (with Inia geoffrensis being an exception) have lost the gene encoding β3-adrenergic receptor (ADRB3), a master regulator of adipose tissue lipolysis and cold-induced non-shivering thermogenesis56,58. We also report and confirm cetacean loss of the gene encoding ELOVL3, a downstream effector of ADRB3 that supplies fatty acids necessary to activate UCP181. Such a shift in adipose tissue function by gene changes is not unprecedented. A recent study suggests that UCP1, although originating ~420 Mya, only became thermogenic in the ancestor of placental mammals (i.e., after ~180 Mya)82.

Studies, on microbes to mammals, show that gene loss can be adaptive, particularly when species face environmental change83,84. In line with this concept, we speculate that the cetacean inactivation of adipose tissue (i.e., blubber-associated genes) may best be understood through a geological epoch lens. That is to say, blubber evolved along the ~55 Mya cetacean transition to a fully aquatic environment. We estimated that ADRB3 was inactivated in the crown cetacean lineage ~32.0–34.4 Mya. While there is variability and uncertainty in time estimates from molecular and fossil data, ADRB3 function must have been lost prior to the divergence of toothed whales and baleen whales and also before the early Oligocene radiation of toothed whales estimated at ~34.1 Mya4. This positions the cetacean ADRB3 loss within the Eocene/Oligocene Transition, a period marked by dramatic cooling of the world’s oceans that was, until then unprecedented in the evolution of mammals5,46,47,48 (Fig. 4). We speculate that the hypomorphic UCP1 mutations were a stem cetacean innovation that balanced cold-induced non-shivering thermogenesis and thermal insulation during the warm early to late Eocene. In contrast, amorphic ADRB3 mutations in crown cetaceans abolished blubber non-shivering thermogenesis and promoted thermal insulation (i.e., facilitated blubber thickness), providing an adaptation necessary to conquer the cooling environment during cetacean radiation. Interestingly, UCP1 is completely lost in many extant lineages, including the South American dolphin lineage Inia that appears to have restored ADRB3 via a rare back-mutation—illustrating the diverse environmental conditions the ancestors of the ~100 species of cetaceans observed today later adapted to.

Fig. 4: Inactivation of blubber-associated genes during cetacean evolution and thermogenic ability.
Fig. 4: Inactivation of blubber-associated genes during cetacean evolution and thermogenic ability.
Full size image

The top panel shows a LOESS (locally estimated scatterplot smoothing) plot of the sea surface temperature (SST) during cetacean evolution. The line represents the LOESS best fit, while the lighter shading corresponds to the 90% confidence window of the fit. The data includes 2143 TEX86 Bayspar-derived SST records from 0–40°N/S latitudes generated by ref. 42. The lower panels depict Cenozoic epochs (a circled X indicates a gene inactivation event), major cetacean lineages, and hypothesized thermogenic ability (NST denotes UCP1-mediated non-shivering thermogenesis). Pakicetus, Basilosaurid, and extant whale illustrations courtesy of Encyclopædia Britannica, Inc.

A thickened, structurally remodeled integument is a critical homeothermic innovation, and cetacean blubber innovation did not cease with the loss of cold-induced non-shivering thermogenesis. Among toothed whales, the superfamily Physeteroidea—sperm whales (Kogiidae and Physeteridae) and beaked whales (Ziphiidae)—emerged during the early to mid-Miocene (~23–16 Mya)85. These species are capable of extreme dives, at depths greater than 1000 m for about an hour or longer, and appear to have evolved a more rigid, pressure-resistant outer blubber by storing fatty acids as wax esters as an alternative to TAGs within their lipid droplets24. Whether the shift towards wax esters primarily resulted from gene inactivation or other genomic changes remains to be determined. A recent study identified a sperm whale and beaked whale-specific amino acid substitution in the fatty acid enzyme ELOVL1 associated with abnormal wax ester composition of mouse tear films30, but a change in its function has yet to be experimentally demonstrated. Blubber is also not the only thermally insulating and rigid layer of the cetacean integument: they have a thick lipid-rich epidermis13. Because ADRB3 is mainly expressed by adipocytes, its loss would not influence epidermal lipogenesis, however, a late Eocene/early Oligocene loss of an epidermal lipid homeostasis master regulator cannot be ruled out.

In summary, our work suggests that the evolution of cetacean blubber is distinct from other marine mammals. We provide evidence that the loss of ADRB3 function was sufficient for cetacean blubber to cease cold-induced non-shivering thermogenesis in favor of thermal insulation. Somewhat surprisingly, this loss was not an ancient event in the cetacean lineage but occurred in the crown ancestor of toothed and baleen whales during a time of global ocean cooling. Instead, early cetacean taxa likely had a much thinner blubber and were capable of limited cold-induced non-shivering thermogenesis due to partial loss-of-function mutations of the downstream ADRB3 effector UCP1. The insights gained from our research should advance the study of marine mammal thermoregulatory strategies.

Methods

Gene loss pipeline

To screen for gene loss in cetaceans, defined as genes with premature stop codons and/or frameshifts in a species, we employed a recently published pipeline53. Briefly, the longest human protein sequence for each gene was mapped to the genomes of the common bottlenose dolphin Tursiops truncatus (assembly mTurTru1.mat.Y) and sperm whale Physeter macrocephalus (assembly ASM283717v5) using BLAT v3686 and GenBlastA v1.0.187. Next, the mapped genomic regions and 1000 base pairs of flanking sequence were examined for disruptions (nonsense mutations and frameshifts) to the gene coding sequences using GeneWise v2.4.188. False positives with disruptive mutations introduced by GeneWise, sequencing errors, or annotation errors were removed following the steps in ref. 53. Candidate pseudogenes with multiple disruptions were manually inspected to remove short or low-quality alignments. Rummagene54, a text mining tool that as of November 2024 scans gene sets derived from the supporting material of 150,600 PubMed Central articles was used to screen for genes identified in previous cetacean comparative genomics studies. To manually investigate the function of genes with candidate inactivating mutations in cetaceans, we perused Alliance of Genome Resources v7.5.0 Gene Descriptions (https://www.alliancegenome.org)89.

Validation of ADRB3 inactivating mutations

Genomes of 27 species from across the cetacean phylogeny were used to explore the adaptive evolution of ADRB3 (Table S1). ADRB3-coding DNA sequences (CDS) of common bottlenose dolphin (Tursiops truncatus) and mammals other than cetaceans were downloaded from the National Center for Biotechnology Information (NCBI). The remaining cetacean ADRB3 sequences were obtained through BLAST searches using the sequence of human (Homo sapiens) and common bottlenose dolphin as queries. The sequences were aligned in MEGA790 with the MUSCLE91 plug-in. To verify putative inactivating mutations in cetaceans, we interrogated genomic reads in the NCBI Sequence Read Archive (SRA) database92 (Table S2). Inactivating mutations were further confirmed by PCR amplification of genomic DNA (primers in Table S5) from the common bottlenose dolphin (Tursiops truncatus), Indo-Pacific humpback dolphin (Sousa chinensis), and humpback whale (Megaptera novaeangliae). ADRB3 sequences from 98 non-cetacean mammalian species were retrieved from OrthoMaM v1093.

Selection pressure analysis

To assess the evolutionary rate of genes lost in the cetacean lineage, we employed five models to determine the ω value (dN/dS) in different branches, as described in a recent study32. We also applied the RELAX model implemented in Datamonkey (http://www.datamonkey.org)66 to determine whether the lost genes evolved under relaxed selection in cetaceans. Gene loss events were dated using the same procedure employed by ref. 70 (based on refs. 94,95). Briefly, a gene is assumed to have evolved under selective pressure before it was inactivated and neutrally after being inactivated (i.e., it should accumulate synonymous and non-synonymous mutations at an equal rate). The Ka/Ks value (K) estimated for this entire branch is considered the average of the Ka/Ks value for the part of the branch where the gene was under selection (Ks) and the Ka/Ks value for the part of the branch where the gene evolved neutrally (Kn = 1), weighted by the proportion of time for which the gene evolved under selection (Ts/T) and neutrally (Tn/T): K = Ks×Ts/T + Kn × Tn/T, where T represents the time since the split from the last common ancestor. Using the lower and upper bound of the confidence interval for the species divergence time T obtained from TimeTree96 and Ka/Ks value for mammals with a functional gene, one can estimate a lower and upper bound for Tn as Tn = T × (K − Ks)/(1 − Ks) to estimate how long a gene has evolved neutrally. A limitation of this method is that a cetacean gene loss cannot be reliably dated if a gene of interest is missing or heavily fragmented in any non-cetacean species within our dataset, or if the gene is highly conserved in cetaceans beyond a single stop codon or other inactivating mutations.

Overexpression of ADRB3 in inguinal white adipose tissue of diet-induced obese mice

Because background strain and sex contribute to the ease of generating DIO (diet-induced obese) mice97, we procured male C57BL/6J mice from the Shanghai Model Organisms Center. All animal experiments were performed in accordance with policies of the National Ministry of Health and approved by the by Nanjing Normal University Institutional Animal Care and Use Committee (IACUC-2025044).

All mice were housed at 21 ± 1 °C with a humidity of 50 ± 5%, with a 12-h light-dark cycle with ad libitum access to food and water. AAV vectors (pAAV-CMV-MCS-EF1-GdGreen-WPRE) harboring common bottlenose dolphin and mouse ADRB3 were purchased from Obio Technology (Shanghai). After the mice were anesthetized, AAV was injected into their iWAT at multiple spots (4–5 spots per fat pad). The total volume was 50 μl, and the virus titer was 8 × 1010 viral genomes (vg) per fat pad. After AAV transduction, the mice were fed a high-fat diet (HFD) containing 60% energy from fat (XTHF60, Jiangsu Xietong Pharmaceutical Bio-engineering Co.) for six weeks.

For the GTT, mice were injected intraperitoneally with 20% glucose (2 mg/g body weigh) after fasting for 16 h (from 5 pm to 9 am), and blood glucose obtained from tail bleeds was monitored at 0, 15, 30, 60, and 120 min after administration using a glucometer monitor (Roche). For the ITT, mice were injected intraperitoneally with recombinant human insulin (Beyotime) (0.5 U/kg body weight) after fasting for 4 h (from 10 am to 2 pm), and blood glucose levels were then monitored. After six weeks on a high-fat diet, serum lipid levels—TC, triacylglycerol (TAG; also known as triglyceride, TG), HDL-C, and LDL-C—were evaluated with assay kits from Nanjing Jian Cheng Biotech.

Oil red O (Sigma) staining was performed as suggested by the manufacturer (Beyotime). Eight days after induction of differentiation, cells were fixed with 4% PFA Fix Solution for 10 min at room temperature and covered with staining wash for 20 s. Cells were then stained oil red O solution for 40 min and washed with staining wash solution and PBS for 30 s. After extracting oil red O with 100% isopropanol, the extracted dye was quantified by reading the absorbance at 510 nm.

Overexpression of ADRB3 in the murine adipocyte cell line 3T3-L1

WAT is a highly dynamic, plastic tissue with multiple cell types98,99,100,101,102, complicating the interpretation of cell function from omics data. Hence, we employed the murine adipocyte cell line 3T3-L1 to assess gene expression changes following ADRB3 overexpression. This ostensibly white adipocyte cell line also exhibits characteristics of other adipocyte lineages, including brown and brown-like adipocytes (i.e., it is polyclonal)76,103. Like most current two-dimensional cell culture models, it does not fully capture all characteristics of its primary counterparts76 but it is a useful complementary model to assess cetacean ADRB3 function.

Murine 3T3-L1 fibroblast (donated by Zhu Li of the Farm Animal Genetic Resources Exploration and Innovation Key Laboratory of Sichuan Province, Sichuan Agricultural University, Chengdu, Sichuan, China) were grown in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin—streptomycin (P/S) at 37 °C until confluence. To induce adipocyte differentiation, the cells were cultured for three days in an induction medium containing DMEM with 5 μg/ml insulin (Beyotime), 1 μM dexamethasone (Beyotime), 0.5 mM 3-isobutyl-1-methylxanthine (IBMX; Aladdin), 1 μM triiodothyronine (T3, Beyotime), and 1 μM rosiglitazone (Aladdin). The cells were next maintained in DMEM supplemented with 5 μg/ml insulin, 1 μM T3, and 1 μM rosiglitazone for another four days before the media was changed into DMEM containing only 10% FBS and 1% P/S at day seven.

Lentiviral overexpression constructs (pLV4ltr-PGK-ZsGreen(2 A)Puro-CMV) with common bottlenose dolphin ADRB3 (Ttru-ADRB3), mouse ADRB3 (Mmus-ADRB3), or mouse ADRB3 with a dolphin mutation at nucleotide 118 (Mmus-ADRB3-1del) were generated by Corues Biotechnology Company (Nanjing, China). When the confluency rate reached 70–80%, cells were passaged, cultured to 50–60% confluence, and infected with lentivirus at 40 multiplicities of infection. Two days after transfection 2 μg/ml puromycin was applied to select stable transfectants and cultured for an additional three days, followed by RNA and protein extractions and cell assays. The infection efficiency was assessed by EGFP expression three days after infection.

Intracellular triacylglycerol (TAG) levels in differentiated 3T3-L1 adipocytes were measured using a TG assay kit (Nanjing Jian Cheng) following the manufacturer’s instructions. The number of mitochondria was estimated using the ratio of mtDNA to nDNA. Total DNA was isolated using a DNA isolation kit (QIAGEN). The DNA levels of ND1 and HK2 were determined by real-time PCR quantification. The relative mtDNA content was evaluated by the ratio of DNA levels between ND1 and HK2, as previously described104. The fluorescent probe MitoSOX105 was used to measure mitochondrial matrix superoxide. Cells were incubated with 5 μM MitoSOX Red (Yeasen, Shanghai) for 10 min at 37 °C and then washed with PBS three times. A microplate reader (BioTek, USA) measured the level of Mito-SOX fluorescence at an excitation wavelength of 396 nm and an emission wavelength of 580 nm. The whole process was performed in the dark.

Bulk 3T3-L1 RNA sequencing and processing

Total RNA was extracted from 3T3-L1 cells using TRIzol (Vazyme). Library preparation of 1 μg of total RNA, using a TruSeq RNA Sample Preparation Kit v2 (Illumina, San Diego, CA), and 150-bp paired sequencing on the Illumina HiSeq X Ten platform was performed by Novogene (Beijing, China). RNA-seq reads were filtered using Flexbar v3.4.0106,107 with default settings (also removes reads with any uncalled bases). Transcripts were quantified using Salmon v1.10.1108 with the mouse GENCODE M36 (GRCm39) reference transcriptome and protein-coding annotations (https://www.gencodegenes.org/mouse/release_M36.html)109. The R package DESeq2 v1.46.0110 was employed to identify differentially expressed genes between sample groups. Briefly, the gene-level count data were normalized using the median of ratios method. For differentially expressed analysis, paired analysis was performed using the generalized linear model in DESeq2, and statistical significance was obtained with default DESeq2 parameters (Wald test and Benjamini-Hochberg P value correction). The statistical significance of intersections of genes with differential expression in Mmus-ADRB3 3T3-L3 adipocytes (i.e., compared to Ttru-ADRB3, Mmus-ADRB3-1del, and vector control cells) was determined using the R package SuperExactTest v1.1.0111. Heat maps were created using the R packages heatmap.3 (https://www.rdocumentation.org/packages/GMD/versions/0.3.3/topics/heatmap.3) and the palette pals::kovesi.linear_bmy_10_95_c78 from paletteer v1.6.0112. Normalized DESeq2 counts are available at https://zenodo.org/records/16990893.

General molecular biology methods

Genomic DNA was extracted using a TIANamp Genomic DNA Kit (TIANGEN, DP304) and amplified using 2×Es Taq MasterMix (CWBIO, CW2313S). For quantitative RT-PCR, total RNA was isolated using TRIzol Extraction Reagent (Vazyme, Nanjing, China). Template complementary DNA was generated using a Hifair III cDNA Synthesis SuperMix (Vazyme,Q321-02), and quantitative real-time RT-PCR was performed on a Roche LightCycler 480 II instrument (Roche Applied Science, Mannheim, Germany) using ChamQ SYBR qPCR Master Mix (Without ROX) (Vazyme) with GADPH as an internal normalization control. Primer sequences are listed in Table S5.

For western blot analysis, adipocyte-differentiated 3T3-L1 cells were washed twice with ice-cold PBS and lysed in RIPA buffer containing 1% protease inhibitor and phosphatase inhibitor (Beyotime, P1050). Cell lysates were sonicated and clarified by centrifugation at 13,680 g for 10 min at 4 °C. Protein concentrations of cell supernatants were determined using a BCA Protein Quantification Kit (Vazyme, E112-01/02) using BSA as the standard. Protein samples were incubated with primary antibodies against UCP1 (Proteintech, 23673-1-AP, 1:1000), ATGL (Proteintech, 55190-1-AP, 1:1000), β-actin (Proteintech, 66009-1-lg,1:40000), and GADPH (Proteintech, 60004-1-Ig, 1:100000). Horseradish peroxidase (HRP)-conjugated antirabbit IgG and anti-mouse IgG (Proteintech) were used as secondary antibodies. The blots were visualized using a chemiluminescence kit (Vazyme, E422-01).

Estimates of the thermal evolution of the world’s oceans

SST data from the upper ocean across the Cenozoic (from 66 Mya) was obtained from ref. 42. The SST data comprised TEX86 values, a palaeothermometer based upon membrane lipids of the mesophilic marine archaeal phylum Nitrososphaerota113. Because low-latitude records are scarce across the Cenozoic middle- and low-latitude SST records were combined, as performed by ref. 42. The data set contained 675 high-latitude (above 50°N/S) and 2143 low/middle-latitude (0–40°N/S) records. We omitted high-latitude SST records in our plot for two reasons: There was limited data from the high-latitude locations from the Eocene-Oligocene transition (~34–33 Mya) and early Miocene (~18–23 Mya), regions with anomalous SST distribution also excluded from plots generated by ref. 42, and cetacean species were not resident at high latitudes until the late Eocene44,114. A scatter plot of SST was generated using the R package “ggplot2” after local regression (locally estimated scatterplot smoothing (LOESS)) with a smoothing factor of 0.02.