Deep phenotyping in zebrafish reveals genetic and diet-induced adiposity changes that may inform disease risk

      The regional distribution of adipose tissues is implicated in a wide range of diseases. For example, proportional increases in visceral adipose tissue increase the risk for insulin resistance, diabetes, and CVD. Zebrafish offer a tractable model system by which to obtain unbiased and quantitative phenotypic information on regional adiposity, and deep phenotyping can explore complex disease-related adiposity traits. To facilitate deep phenotyping of zebrafish adiposity traits, we used pairwise correlations between 67 adiposity traits to generate stage-specific adiposity profiles that describe changing adiposity patterns and relationships during growth. Linear discriminant analysis classified individual fish according to an adiposity profile with 87.5% accuracy. Deep phenotyping of eight previously uncharacterized zebrafish mutants identified neuropilin 2b as a novel gene that alters adipose distribution. When we applied deep phenotyping to identify changes in adiposity during diet manipulations, zebrafish that underwent food restriction and refeeding had widespread adiposity changes when compared with continuously fed, equivalently sized control animals. In particular, internal adipose tissues (e.g., visceral adipose) exhibited a reduced capacity to replenish lipid following food restriction. Together, these results in zebrafish establish a new deep phenotyping technique as an unbiased and quantitative method to help uncover new relationships between genotype, diet, and adiposity.
      Adipose tissues (ATs) are lipid-rich organs that supply and sequester circulating lipid in response to systemic energy demands. ATs thus provide “energetic insurance” to individuals and confer selective advantages during periods of adverse physiological stresses. In modern societies, where food availability is high and energy expenditure is low, ATs accumulate large quantities of lipid, which can initiate a range of secondary metabolic abnormalities that result in increased susceptibility to diabetes, CVD, and cancer. A wide range of adiposity traits can influence disease risk. For example, general adiposity levels, as measured by BMI, are associated with increased risk for disease (
      • Gadde K.M.
      • Martin C.K.
      • Berthoud H.R.
      • Heymsfield S.B.
      Obesity: pathophysiology and management.
      ). In turn, the regional distribution of AT can also influence disease risk. For example, accumulation of visceral AT (VAT) in the abdominal cavity in close proximity to visceral organs is associated with an increased risk for insulin resistance and a sequelae of accompanying diseases such as diabetes and CVD (
      • Fox C.S.
      • Massaro J.M.
      • Hoffmann U.
      • Pou K.M.
      • Maurovich-Horvat P.
      • Liu C.Y.
      • Vasan R.S.
      • Murabito J.M.
      • Meigs J.B.
      • Cupples L.A.
      • et al.
      Abdominal visceral and subcutaneous adipose tissue compartments: association with metabolic risk factors in the Framingham Heart Study.
      ,
      • Karpe F.
      • Pinnick K.E.
      Biology of upper-body and lower-body adipose tissue–link to whole-body phenotypes.
      ). Conversely, accumulation of subcutaneous AT (SAT) in a peripheral location at the hips and upper thighs is associated with reduced disease risk (
      • McLaughlin T.
      • Lamendola C.
      • Liu A.
      • Abbasi F.
      Preferential fat deposition in subcutaneous versus visceral depots is associated with insulin sensitivity.
      ). Understanding how genetics and the environment influence these diverse adiposity traits will be key to treating and predicting the metabolic consequences of obesity.
      Zebrafish are a tropical freshwater fish that offer a tractable model system to study AT biology. Zebrafish AT is morphologically and molecularly homologous to mammalian white AT (WAT) (
      • Flynn 3rd, E.J.
      • Trent C.M.
      • Rawls J.F.
      Ontogeny and nutritional control of adipogenesis in zebrafish (Danio rerio).
      ,
      • Imrie D.
      • Sadler K.C.
      White adipose tissue development in zebrafish is regulated by both developmental time and fish size.
      ). Zebrafish AT also appears to be functionally conserved to mammalian WAT and accumulates lipid during periods of chronic energy excess and mobilizes lipid during energy deficiency (
      • Flynn 3rd, E.J.
      • Trent C.M.
      • Rawls J.F.
      Ontogeny and nutritional control of adipogenesis in zebrafish (Danio rerio).
      ,
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ,
      • Oka T.
      • Nishimura Y.
      • Zang L.
      • Hirano M.
      • Shimada Y.
      • Wang Z.
      • Umemoto N.
      • Kuroyanagi J.
      • Nishimura N.
      • Tanaka T.
      Diet-induced obesity in zebrafish shares common pathophysiological pathways with mammalian obesity.
      ). The molecular pathways that regulate adiposity also appear to be conserved between zebrafish and mammals, as typified by zebrafish growth hormone (gh) mutants (
      • McMenamin S.K.
      • Minchin J.E.
      • Gordon T.N.
      • Rawls J.F.
      • Parichy D.M.
      Dwarfism and increased adiposity in the gh1 mutant zebrafish vizzini.
      ). Importantly, zebrafish AT can be visualized and quantified in vivo at both cell and whole-animal resolutions (
      • Minchin J.E.
      • Rawls J.F.
      In vivo analysis of white adipose tissue in zebrafish.
      ,
      • Minchin J.E.
      • Rawls J.F.
      In vivo imaging and quantification of regional adiposity in zebrafish.
      ). We recently utilized these imaging properties to comprehensively identify, characterize, and quantify the full complement of zebrafish ATs at distinct developmental stages (
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ). This methodology generates unbiased and quantitative phenotypic information on a comprehensive range of adiposity traits. We reasoned that such multivariate data could be utilized for deep phenotyping of adiposity traits. Briefly, deep phenotyping is defined as the precise and comprehensive analysis of phenotypic abnormalities (
      • Robinson P.N.
      Deep phenotyping for precision medicine.
      ) and is useful for identifying traits and complex phenotypic signatures that define health or disease (
      • Hur M.
      • Gistelinck C.A.
      • Huber P.
      • Lee J.
      • Thompson M.H.
      • Monstad-Rios A.T.
      • Watson C.J.
      • McMenamin S.K.
      • Willaert A.
      • Parichy D.M.
      • et al.
      MicroCT-based phenomics in the zebrafish skeleton reveals virtues of deep phenotyping in a distributed organ system.
      ,
      • Hur M.
      • Gistelinck C.A.
      • Huber P.
      • Lee J.
      • Thompson M.H.
      • Monstad-Rios A.T.
      • Watson C.J.
      • McMenamin S.K.
      • Willaert A.
      • Parichy D.M.
      • et al.
      MicroCT-based phenomics in the zebrafish skeleton reveals virtues of deep phenotyping in a distributed organ system.
      ,
      • San-Miguel A.
      • Kurshan P.T.
      • Crane M.M.
      • Zhao Y.
      • McGrath P.T.
      • Shen K.
      • Lu H.
      Deep phenotyping unveils hidden traits and genetic relations in subtle mutants.
      ). Deep phenotyping is particularly powerful when such traits may be quantitative and subtle, as found with adiposity.
      In this study, we generated stage-specific adiposity profiles that comprehensively capture patterns and relationships in adiposity dynamics and adipose distribution. To classify individuals according to expected patterns of adiposity, we applied linear discriminant analysis (LDA) to the adiposity profiles. We utilized this methodology to screen eight zebrafish mutants and identified neuropilin 2b (nrp2b) as a novel gene that promotes adiposity in zebrafish. Finally, we applied our deep phenotyping strategy to identify adiposity changes that occur following diet manipulation. We identified that food restriction induced profound changes in fat distribution, even after animals had fully regained weight. Closer analysis revealed that internal ATs (IATs), including VAT, did not fully regain lipid following refeeding, resulting in altered fat distribution. Altogether, we develop methodology to quantitatively extract phenotypic information for detecting gene and diet-induced adiposity phenotypes.

      MATERIALS AND METHODS

       Zebrafish husbandry, strains, and data availability

      Zebrafish experiments conformed to the US Public Health Service Policy on the Humane Care and Use of Laboratory Animals, using protocols approved by the Institutional Animal Care and Use Committees of the University of North Carolina at Chapel Hill (UNC) and Duke University. Zebrafish husbandry for experiments conducted at either UNC or Duke were performed as previously described (
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ). The Ekkwill (EKW) and WIK WT data used as the LDA training set (N = 456) was previously published (
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ) and is available for download at DataDryad (https://doi.org/10.5061/dryad.98470). The gh mutant data used as a positive control for LDA were previously published (
      • McMenamin S.K.
      • Minchin J.E.
      • Gordon T.N.
      • Rawls J.F.
      • Parichy D.M.
      Dwarfism and increased adiposity in the gh1 mutant zebrafish vizzini.
      ), and raw images are available for download at DataDryad (https://doi.org/10.5061/dryad.vv34p8h). The eight zebrafish mutants used for LDA were obtained from the Wellcome Trust Sanger Institute's Zebrafish Mutation Project (ZMP) or from Zebrafish International Resource Center (
      • Kettleborough R.N.
      • Busch-Nentwich E.M.
      • Harvey S.A.
      • Dooley C.M.
      • de Bruijn E.
      • van Eeden F.
      • Sealy I.
      • White R.J.
      • Herd C.
      • Nijman I.J.
      • et al.
      A systematic genome-wide analysis of zebrafish protein-coding gene function.
      ). Embryos were received at Duke on a Hubrecht long-fin background. In-crosses were performed between heterozygous individuals to generate experimental clutches. Genotypes were determined by Sanger sequencing. Images used for quantification were deposited at DataDryad (https://doi.org/10.5061/dryad.vv34p8h). All alleles used in this study are included in Table 1. The food restriction and refeeding was performed in EKW WT fish at Duke, and the raw images are available for download at DataDryad (https://doi.org/10.5061/dryad.vv34p8h).
      TABLE 1Zebrafish mutants assessed for adiposity changes in this study
      GeneGene SymbolAlleleAllele ConsequenceNo. of Fish ScreenedPercent Misclassified (Phenotypic)
      nrp2bnrp2bsa18942Essential splice site3435.3%
      nuclear receptor subfamily group a member 3nr4a3sa2842Nonsense3414.7%
      proprotein convertase subtilisin/kexin type 1pcsk1sa1558Essential splice site339%
      semaphorin 3aasema3aasa10241Nonsense3514%
      semaphorin 3fbsema3fbsa14466Nonsense4714.8%
      semaphorin 3gbsema3gbsa21283Nonsense2114.2%
      sarcospansspnsa2992Nonsense3616.7%
      transmembrane protein 160tmem160sa1347Nonsense2913.7%
      ghghwp22e1Premature stop codon13036%

       Nile Red staining, image analysis, and adipose and stage classification

      Nile Red (Sigma-Aldrich, catalog no. N1142) was dissolved in acetone at a concentration of 1.25 mg/ml and diluted to 0.5 µg/ml in system water for lipid staining (
      • Minchin J.E.
      • Rawls J.F.
      In vivo analysis of white adipose tissue in zebrafish.
      ,
      • Minchin J.E.
      • Rawls J.F.
      In vivo imaging and quantification of regional adiposity in zebrafish.
      ). Zebrafish were incubated in the diluted Nile Red for 30 min as previously described (
      • Minchin J.E.
      • Rawls J.F.
      In vivo analysis of white adipose tissue in zebrafish.
      ,
      • Minchin J.E.
      • Rawls J.F.
      In vivo imaging and quantification of regional adiposity in zebrafish.
      ). Following the staining, melanosomes were contracted by incubation in 10 mg/ml epinephrine (Sigma-Aldrich, catalog no. E4375) for 5 min, and zebrafish were anesthetized in 1.34 g/l MS222 (Sigma-Aldrich, catalog no. A5040) for 3 min and mounted on 3% methylcellulose (Sigma-Aldrich, catalog no. MO387). The right-hand side of each fish was imaged on a Leica MZ205FA fluorescence stereomicroscope equipped with a Leica DFC365 FX fluorescence camera and using a GFP2 bandpass filter (Leica Microsystems, catalog no. 10447407). All analyses were conducted in FIJI/ImageJ (version 1.51i) (
      • Schindelin J.
      • Arganda-Carreras I.
      • Frise E.
      • Kaynig V.
      • Longair M.
      • Pietzsch T.
      • Preibisch S.
      • Rueden C.
      • Saalfeld S.
      • Schmid B.
      • et al.
      Fiji: an open-source platform for biological-image analysis.
      ). For AT area measurements, two copies of each image were opened; one for taking measurements and one for comparisons to maximize accuracy of adipose segmentation. Standard length (SL), height at the anterior margin of the anal fin (HAA), and body area were used to determine zebrafish size and were measured using the line and polygon tools as previously described (
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ,
      • Parichy D.M.
      • Elizondo M.R.
      • Mills M.G.
      • Gordon T.N.
      • Engeszer R.E.
      Normal table of postembryonic zebrafish development: staging by externally visible anatomy of the living fish.
      ). AT areas were defined by manually set thresholds based on pixel intensities (
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ). For ATs that did not touch, the magic wand tool was used to select AT area. For touching ATs, the polygon tool was used to trace the AT outline, and where a dividing line between the ATs was not visible, a straight line was drawn between the farthest distinguishing AT points. Lateral view images were used for all measurements. ATs were classified and measured as described in ref.
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      . A schematic illustrating the location of each AT in zebrafish is included in supplemental Fig. S1. The stages used for generating adiposity profiles incorporated both postembryonic and adiposity milestones as previously documented (
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ,
      • Parichy D.M.
      • Elizondo M.R.
      • Mills M.G.
      • Gordon T.N.
      • Engeszer R.E.
      Normal table of postembryonic zebrafish development: staging by externally visible anatomy of the living fish.
      ). As previously described, Nile Red also labels neutral lipid within the liver, intestinal epithelium, and the blood (
      • Flynn 3rd, E.J.
      • Trent C.M.
      • Rawls J.F.
      Ontogeny and nutritional control of adipogenesis in zebrafish (Danio rerio).
      ,
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ); however, lipid at these sites can be readily distinguished from AT. Adiposity measurements used in this study are included in supplemental Tables S1 and S2.

       Food restriction and refeeding experiments

      Food restriction and refeeding experiments were conducted largely as described (
      • Flynn 3rd, E.J.
      • Trent C.M.
      • Rawls J.F.
      Ontogeny and nutritional control of adipogenesis in zebrafish (Danio rerio).
      ), with the following differences; zebrafish at 38 days postfertilization [dpf; 11.4 ± 0.5 mm SL (mean ± SD); dorsal fin ray SAT (DFRSAT) stage] were housed individually in six-well plates containing ∼3 ml of system water. No food was administered during food restriction (days 1–11). Upon refeeding (days 12–22), fish were fed Artemia franciscana and powdered food as per normal husbandry procedures at Duke University as previously described (
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ). Every 24 h, 0.5 µg/ml Nile Red was added to the wells, and the fluorescent lipid signal was imaged on a Leica MZ205FA fluorescence stereomicroscope equipped with a Leica DFC365 FX fluorescence camera as described above. Lipid deposits were thus evaluated daily, and food restriction was stopped after 11 days once lipid was mobilized from all AT sites in all animals. System water was replaced daily during both starvation and refeeding periods.

       Statistical analyses

      Stage-specific adiposity profiles were generated using pairwise correlations between 63 adiposity traits (listed in Table 2) using the multivariate platform in JMP Pro 13 (SAS, NC). Adiposity profiles were constructed for 14 stages (supplemental Fig. S2 and supplemental Tables S3–S16). Heatmaps to visualize the adiposity profiles were generated in JMP and show pairwise intertrait correlations expressed as the Pearson's correlation coefficient (1 is a total positive correlation between traits, 0 is no correlation, and −1 is a total inverse correlation). LDA was performed to classify individual fish according to their adiposity profiles. LDA was performed in JMP using the discriminant analysis platform and a linear common covariance method. LDA is a general linear model that derives discriminant functions—linear combinations of variables—to maximize the probability of assigning observations to a predefined group. The discriminant functions for each group (adiposity profile) (Cprofile) followed refs.
      • Tabachnick B.
      • Fidell L.
      and
      • Quinn G.P.
      • Keough M.J.
      Experimental Design and Analysis for Biologists.
      and are expressed as: Cprofile = i + c1(log10 a1) + c2(log10 a2) + ··· + cn(log10 an), where c is the coefficient of the classification equation (
      • Tabachnick B.
      • Fidell L.
      ), i is the constant for each group (adiposity profile) as determined by multiplying the matrix of classification coefficients for that group by the matrix of means for each adiposity trait variable (a) of that group. All 63 adiposity traits were used as variables for LDA. To define the adiposity characteristics, adiposity profiles from 456 WT fish were set as a validation group (training set). Analysis of interclutch variability in LDA within the training set revealed misclassification rates of 15.2% (mean) with a SE of 2.4% (25 independent clutches). This baseline rate was used to identify phenotypic fish during additional comparisons. We defined “nonnormal” phenotypes if clutches had a percent misclassification rate >3 ± SE from the training set rate. These clutches were evaluated further as phenotypic. Principal components analysis (PCA) of adiposity profiles during food restriction (days 1–11) and refeeding (days 12–22) was conducted in JMP using the default estimation method. The mean recovery, or regain, of AT-lipid at day 22 (following food restriction and refeeding) was calculated as a percent of AT-lipid at day 1. Hierarchical clustering of AT-lipid regain at day 22 was performed in JMP using the Ward method (
      • Murtagh F.
      • Legendre P.
      Ward's hierarchical agglomerative clustering method: which algorithms implement Ward's criterion?.
      ). Student's t-tests were used for pairwise comparisons, and one-way ANOVA followed by Tukey's posthoc test was used for multiple groups. ANCOVA was used to test for differences between groups following linear regression. Statistical significance was set to α = 0.05. Graphs were plotted in R using ggplot2 (
      • Wickham H.
      ,
      R Core Team. R: A language and environment for statistical computing..
      ). The stage-specific adiposity profiles can be found in supplemental Fig. 2, and the accompanying correlation matrices are in supplemental Tables 3–18.
      TABLE 2Morphological traits used to construct adiposity profiles
      TraitTrait AcronymTrait Category
      SL (µm)SLBody size
      HAA (µm)HAABody size
      Body area (µm2)BABody size
      PVAT (µm2)PVATAT area
      Abdominal VAT (µm2)AVATAT area
      Renal VAT (µm2)RVATAT area
      CVAT (µm2)CVATAT grouping
      Anterior CVAT (µm2)aCVATAT area
      Posterior CVAT (µm2)pCVATAT area
      VAT (µm2)VATAT area
      Abdominal SAT (µm2)ASATAT area
      Lateral SAT (µm2)LSATAT area
      Dorsal SAT (µm2)DSATAT grouping
      Anterior DSAT (µm2)aDSATAT area
      pDSAT (µm2)pDSATAT area
      Ventral SAT (µm2)VSATAT area
      Truncal SAT (µm2)TSATAT grouping
      DFRSAT (µm2)DFRSATAT area
      Anal fin ray cluster SAT (µm2)AFCSATAT area
      AFRSAT (µm2)AFRSATAT area
      Caudal fin ray SAT (µm2)CFRSATAT area
      Pectoral fin SAT (µm2)PECSATAT grouping
      Loose PECSAT (µm2)lPECSATAT area
      Anterior PECSAT (µm2)aPECSATAT area
      Posterior PECSAT (µm2)pPECSATAT area
      Appendicular SAT (µm2)APPSATAT grouping
      Central IM (µm2)cIMAT area
      Dorsal IM (µm2)dIMAT area
      Ventral IM (µm2)vIMAT area
      Intermuscular (µm2)IMAT grouping
      Dorsal POS (µm2)dPOSAT area
      Central POS (µm2)cPOSAT area
      Ventral POS (µm2)vPOSAT area
      Paraosseal (µm2)POSAT grouping
      Esophageal (µm2)OESAT area
      Nonvisceral AT (µm2)NVATAT grouping
      Dorsal OPC (µm2)dOPCAT area
      Ventral OPC (µm2)vOPCAT area
      Opercular (µm2)OPCAT grouping
      Ocular (µm2)OCUAT area
      Branchihyal (µm2)BHDAT area
      Ceratohyal (µm2)CHDAT area
      Urihyal (µm2)UHDAT area
      Hyal (µm2)HYDAT grouping
      Cranial SAT (µm2)CSATAT grouping
      Total AT (µm2)TOTALAT grouping
      SAT (µm2)SATAT grouping
      IAT (µm2)IATAT grouping
      VAT:IATAT ratio
      NVAT:IATAT ratio
      IAT:TOTALAT ratio
      SAT:TOTALAT ratio
      CSAT:SATAT ratio
      APPSAT:SATAT ratio
      TSAT:SATAT ratio
      VAT:SATAT ratio
      NVAT:SATAT ratio
      CSAT:IATAT ratio
      TSAT:IATAT ratio
      APPSAT:IATAT ratio
      VAT:TSATAT ratio
      VAT:CSATAT ratio
      VAT:APPSATAT ratio
      NVAT:TSATAT ratio
      NVAT:CSATAT ratio
      NVAT:APPSATAT ratio
      NVAT:VATAT ratio
      TSAT:CSATAT ratio
      APPSAT:CSATAT ratio
      APPSAT:TSATAT ratio

      RESULTS

       Stage-specific phenotypic profiles capture adiposity patterns and relationships

      Whole-animal in vivo imaging in zebrafish enables the quantification of all ATs in a single animal and can thus be used to reveal a wide range of adiposity traits, including changes in fat levels, ectopic localization to specific organs, and changes in regional distribution (
      • Flynn 3rd, E.J.
      • Trent C.M.
      • Rawls J.F.
      Ontogeny and nutritional control of adipogenesis in zebrafish (Danio rerio).
      ,
      • Minchin J.E.
      • Rawls J.F.
      In vivo analysis of white adipose tissue in zebrafish.
      ,
      • Minchin J.E.
      • Rawls J.F.
      In vivo imaging and quantification of regional adiposity in zebrafish.
      ). We reasoned that collecting large amounts of quantitative adiposity data could be used to classify individuals based on adiposity traits and identify subtle, quantitative adiposity phenotypes. We previously showed that stage-matched zebrafish exhibit stereotypical adiposity patterns (Fig. 1A) (
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ). Therefore, we utilized existing data to generate stage-specific phenotypic profiles that capture adiposity information (hereafter called adiposity profiles). In total, 456 WT zebrafish across a range of sizes, stages, and ages were used to construct adiposity profiles (Fig. 1B) (
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ). From these 456 fish, 67 traits were quantified in each fish (Table 1). The 67 traits included i) AT area measurements, ii) measures of body size (including SL, HAA, and body area), iii) composite AT groupings (e.g., total AT, SAT, and VAT), and iv) AT proportionality assessments (e.g., VAT:SAT). Correlations were computed between each trait and assessed to determine how adiposity relationships and patterns change in fish of distinct stages (Fig. 1B). As expected, considerable differences were observed in adiposity profiles at distinct stages (Fig. 1B, C). For example, pancreatic VAT (PVAT) was initially positively correlated with SL [pectoral fin bud (PB) stage], before becoming progressively more inversely correlated with SL in larger fish [stage squamation through anterior (SA)] (Fig. 1C). In conclusion, adiposity profiles capture dynamic changes in adiposity patterns at distinct developmental stages and will be useful indicators of “normal” adiposity levels and variation.
      Figure thumbnail gr1
      Fig. 1Generation of stage-specific adiposity profiles in zebrafish. A: Fluorescence stereomicroscope images of Nile Red-stained zebrafish at two distinct postembryonic stages [pelvic fin ray appearance (PR) and SA]. Note the increasingly complex and diversifying distribution patterns of AT in SA fish relative to PR. White arrows correspond to posterior dorsal SAT (pDSAT), black arrowheads correspond to central paraosseal (cPOS), and white arrowheads correspond to anal fin ray SAT (AFRSAT) Scale bars are 1 mm. B: To determine WT adiposity dynamics, 67 adiposity traits were measured in 456 WT zebrafish. PB is from Parichy et al. (
      • Parichy D.M.
      • Elizondo M.R.
      • Mills M.G.
      • Gordon T.N.
      • Engeszer R.E.
      Normal table of postembryonic zebrafish development: staging by externally visible anatomy of the living fish.
      ). Each stage consisted of >10 fish. For each stage, intertrait correlations were determined and used to create an adiposity profile that encapsulates relationship dynamics of individual ATs. C: Example intertrait correlations are given for PVAT and demonstrate the changing relationship between PVAT and SL across distinct stages.

       LDA accurately classifies zebrafish according to the adiposity profile

      To effectively use stage-specific adiposity profiles as a base for in-depth phenotypic profiling we utilized LDA. LDA is a data dimensionality reduction technique that can assign membership of a group based on accompanying covariate values. LDA has been used previously to i) classify morphological phenotypes into distinct groups (
      • Bertsatos A.
      • Papageorgopoulou C.
      • Valakos E.
      • Chovalopoulou M.E.
      Investigating the sex-related geometric variation of the human cranium.
      ,
      • Suhail Z.
      • Denton E.R.E.
      • Zwiggelaar R.
      Classification of micro-calcification in mammograms using scalable linear Fisher discriminant analysis.
      ), ii) predict disease outcomes based on current symptoms (
      • Oh J.
      • Cho D.
      • Park J.
      • Na S.H.
      • Kim J.
      • Heo J.
      • Shin C.S.
      • Kim J.J.
      • Park J.Y.
      • Lee B.
      Prediction and early detection of delirium in the intensive care unit by using heart rate variability and machine learning.
      ), and iii) predict future business success based on current financial parameters (
      • Kočišová K.
      • Mišanková M.
      Discriminant analysis as a tool for forecasting companys financial health.
      ). We reasoned that LDA could also be applied to adiposity profiles and used to classify individual fish according to expected adiposity traits. As a training set, we applied the adiposity profiles from the 456 WT fish described above and assessed whether LDA was able to accurately assign fish to correct developmental stages (Fig. 2A). Based on adiposity profiles, LDA assigned 87.5% of the WT fish into correct stages (Fig. 2A). We next assessed how robust the LDA classification method was across multiple independent clutches. We divided the training set data into its 25 constituent clutches and applied LDA to each clutch, resulting in an average classification rate of 84.8 ± 2.4% (mean ± SE) (Fig. 2E). Next, as a proof-of-concept, we used zebrafish gh mutants to determine whether LDA can be used to detect adiposity phenotypes. We previously showed that gh mutants have increased adiposity and retarded somatic growth relative to size-matched WT siblings (Fig. 2B, C) (
      • McMenamin S.K.
      • Minchin J.E.
      • Gordon T.N.
      • Rawls J.F.
      • Parichy D.M.
      Dwarfism and increased adiposity in the gh1 mutant zebrafish vizzini.
      ). From McMenamin et al. (
      • McMenamin S.K.
      • Minchin J.E.
      • Gordon T.N.
      • Rawls J.F.
      • Parichy D.M.
      Dwarfism and increased adiposity in the gh1 mutant zebrafish vizzini.
      ), we applied LDA to the adiposity profiles of 62 homozygous gh mutants and 60 WT siblings from three clutches (Fig. 2D). Adiposity profiles from the 456 WT zebrafish were used as a training set (Fig. 2D). Within the training set, LDA identified 15.2% of animals as phenotypic, setting a low background misclassification rate (Fig. 2E). From the gh mutant-only data, LDA correctly identified phenotypes based on adiposity profiles in 60% of cases (Fig. 2E). However, when the clutch was considered as a whole (containing both WT siblings and mutants), LDA classified 36% of individuals as phenotypic (P = 0.003) (Fig. 2E). Based on the genotype data, the true misclassification rate was expected to be 50.8%; therefore, although LDA was underperforming, it could be effectively used to flag the clutch as phenotypic. Taken together, LDA can be used to identify phenotypic adiposity traits based on adiposity profiles.
      Figure thumbnail gr2
      Fig. 2LDA can be used to identify and classify phenotypic fish based on adiposity profiles. A: LDA was used as a training set to classify 456 WT fish according to stage-specific adiposity profiles. From this training set, LDA was able to accurately classify fish based on adiposity profiles in 87.5% of instances. B: Zebrafish homozygous for the ghwp22e1 mutation have increased total AT (dotted outlines delineate AT). C: Quantification of total AT area in WT sibling and gh mutants. *** P < 0.0001 (Student's t-test). D: Schematic detailing how the LDA was used to analyze gh mutant and WT sibling fish relative to the training set. E: Of the 122 gh and WT sibling fish, LDA detected 36% as phenotypic (all). The WT sibling fish were not classified as phenotypic relative to baseline classification from the training set (WT). The gh homozygous mutant fish were classified as phenotypic in 60% of instances.

       Identification of nrp2b as a novel gene implicated in zebrafish adiposity

      We next sought to apply LDA to help identify novel zebrafish adipose mutants. We obtained eight mutant lines generated by the ZMP (Table 1) (
      • Kettleborough R.N.
      • Busch-Nentwich E.M.
      • Harvey S.A.
      • Dooley C.M.
      • de Bruijn E.
      • van Eeden F.
      • Sealy I.
      • White R.J.
      • Herd C.
      • Nijman I.J.
      • et al.
      A systematic genome-wide analysis of zebrafish protein-coding gene function.
      ). The eight ZMP mutants were selected as candidate regulators of adiposity based on published experimental evidence or by genome-wide association studies (
      • Shungin D.
      • Winkler T.W.
      • Croteau-Chonka D.C.
      • Ferreira T.
      • Locke A.E.
      • Magi R.
      • Strawbridge R.J.
      • Pers T.H.
      • Fischer K.
      • Justice A.E.
      • et al.
      New genetic loci link adipose and insulin biology to body fat distribution.
      ,
      • Heid I.M.
      • Jackson A.U.
      • Randall J.C.
      • Winkler T.W.
      • Qi L.
      • Steinthorsdottir V.
      • Thorleifsson G.
      • Zillikens M.C.
      • Speliotes E.K.
      • Magi R.
      • et al.
      Meta-analysis identifies 13 new loci associated with waist-hip ratio and reveals sexual dimorphism in the genetic basis of fat distribution.
      ,
      • Mejhert N.
      • Wilfling F.
      • Esteve D.
      • Galitzky J.
      • Pellegrinelli V.
      • Kolditz C.I.
      • Viguerie N.
      • Tordjman J.
      • Naslund E.
      • Trayhurn P.
      • et al.
      Semaphorin 3C is a novel adipokine linked to extracellular matrix composition.
      ,
      • Shimizu I.
      • Yoshida Y.
      • Moriya J.
      • Nojima A.
      • Uemura A.
      • Kobayashi Y.
      • Minamino T.
      Semaphorin3E-induced inflammation contributes to insulin resistance in dietary obesity.
      ,
      • Walton R.G.
      • Zhu X.
      • Tian L.
      • Heywood E.B.
      • Liu J.
      • Hill H.S.
      • Liu J.
      • Bruemmer D.
      • Yang Q.
      • Fu Y.
      • et al.
      AP2–NR4A3 transgenic mice display reduced serum epinephrine because of increased catecholamine catabolism in adipose tissue.
      ,
      • Wilson A.M.
      • Shao Z.
      • Grenier V.
      • Mawambo G.
      • Daudelin J.F.
      • Dejda A.
      • Pilon F.
      • Popovic N.
      • Boulet S.
      • Parinot C.
      • et al.
      Neuropilin-1 expression in adipose tissue macrophages protects against obesity and metabolic syndrome.
      ,
      • Löffler D.
      • Behrendt S.
      • Creemers J.W.
      • Klammt J.
      • Aust G.
      • Stanik J.
      • Kiess W.
      • Kovacs P.
      • Korner A.
      Functional and clinical relevance of novel and known PCSK1 variants for childhood obesity and glucose metabolism.
      ,
      • Heinonen S.
      • Buzkova J.
      • Muniandy M.
      • Kaksonen R.
      • Ollikainen M.
      • Ismail K.
      • Hakkarainen A.
      • Lundbom J.
      • Lundbom N.
      • Vuolteenaho K.
      • et al.
      Impaired mitochondrial biogenesis in adipose tissue in acquired obesity.
      ). Heterozygous carriers from the ZMP mutants were intercrossed, the offspring was raised to ∼30 dpf, and then Nile Red was used to visualize AT levels (Fig. 3). Following imaging, the fish were genotyped to identify WT and heterozygous and homozygous mutants. Adiposity profiles were constructed for each fish, and LDA was used to identify differences in adiposity relative to the training set of 465 WT fish. Of the eight ZMP mutant lines, only nrp2b robustly presented mutant identification rates higher than expected from WT fish (Fig. 3A), suggesting that mutation of nrp2b may influence adiposity in zebrafish. The other seven lines all exhibited misclassification rates within the expected WT range and were therefore classed as phenotypically normal (Fig. 3A). Closer analysis of the nrp2b fish stained by Nile Red revealed a robust reduction in total adiposity, which was validated by analyzing two additional and independent clutches (Fig. 3B, C). In conclusion, we utilized LDA to screen eight new zebrafish mutants for adiposity phenotypes and identified nrp2b as a potential novel modifier of adiposity.
      Figure thumbnail gr3
      Fig. 3LDA can be used as a screening tool to identify mutants with adiposity phenotypes in zebrafish. A: LDA misclassification rates (percent phenotypic fish) for the eight ZMP mutants, gh mutants (positive control), and the WT training set. Error bars denote the SEM across multiple clutches. Bars without error bars are from single clutches. Details of the alleles for each mutant line can be found in . B: Fluorescence stereomicroscope images of Nile Red-stained WT sibling and two nrp2bsa18942 homozygous mutant zebrafish. Note the reduced adiposity levels in nrp2bsa18942 mutants. C: Two additional independent clutches of WT siblings and nrp2bsa18942 mutants were assessed to validate the nrp2b phenotype. Linear regression (straight lines with 95% CIs noted) was used to evaluate adiposity relative to SL, and ANCOVA was used to test for differences between the groups. F1,83 = 52.4, P < 0.0001.

       LDA identifies differences in adiposity traits following food restriction and refeeding

      We reasoned that adiposity profiles and analysis by LDA could also be applied to identify diet-induced changes in adiposity. To test this, we subjected zebrafish to prolonged food restriction followed by refeeding until total adiposity levels were completely replenished. Previous studies have shown that zebrafish robustly mobilize and regain AT-localized lipid in response to food levels (
      • Flynn 3rd, E.J.
      • Trent C.M.
      • Rawls J.F.
      Ontogeny and nutritional control of adipogenesis in zebrafish (Danio rerio).
      ). Furthermore, zebrafish recover from acute food restriction with no visible health detriments (
      • Flynn 3rd, E.J.
      • Trent C.M.
      • Rawls J.F.
      Ontogeny and nutritional control of adipogenesis in zebrafish (Danio rerio).
      ). We subjected 10 WT (EKW) zebrafish to complete food restriction for 11 days, followed by 11 days of refeeding (Fig. 4). To build an accurate depiction of adiposity dynamics, the fish were stained with Nile Red, and their AT-localized lipid was imaged daily during the course of the diet manipulation (Fig. 4A). AT-localized lipid was fully mobilized by 11 days of food restriction, as judged by the lack of Nile Red+ lipid within AT (Fig. 4A). At this point, zebrafish were refed a normal diet (see Materials and Methods) until total adiposity had reached levels equivalent to size-matched normally fed individuals (Figs. 4A and 5A). Adiposity profiles were generated for each fish at each day of diet manipulation (days 1–22), and PCA was used to identify whether adiposity trait dynamics were different during food restriction and refeeding (Fig. 4B, C). Strikingly, PCA traced distinct trajectories for food-restricted and refed zebrafish, which did not overlap (Fig. 4B, C), suggesting that food restriction and refeeding elicits changes in AT distribution even after complete recovery of total AT levels. We next applied LDA to determine whether adiposity profiles differed after food restriction and refeeding. We generated adiposity profiles of fully refed animals (day 22) and “continuously fed” animals that were matched for both size (SL) and developmental stage (Fig. 5B and supplemental Tables S17 and S18). LDA misclassified ∼90% of food-restricted and refed zebrafish (day 22) when compared with stage-matched control animals (Fig. 5C). Thus, food restriction and refeeding leads to wide-ranging adiposity differences relative to normally fed animals.
      Figure thumbnail gr4
      Fig. 4Food restriction and subsequent refeeding leads to changes in adipose distribution. A: Fluorescence stereomicroscope images of Nile Red-stained zebrafish reveal the mobilization and reduction of lipid within AT during 11 days of food restriction (magenta; days 1–11) and the redeposition and increase in lipid within adipose during 11 days of refeeding (blue; days 12–22). Scale bars are 1 mm. B: PCA of adiposity traits during food restriction and refeeding reveals that food restriction leads to altered adipose distribution. Note that fed fish (light magenta; day 1) do not colocalize with refed animals (dark blue; day 22). C: PCA of food restriction and refeeding reveals the daily changes in adipose distribution.
      Figure thumbnail gr5
      Fig. 5LDA can be used to classify food restriction-induced changes in adiposity. A: Chart showing the reduction and subsequent regain of total AT area during food restriction (days 1–11) and refeeding (days 12–22). Black line connects the mean total AT area at each day. Black error bars at each day denote the SD in total AT area. Gray lines indicate the total AT area of individual fish. Red bars indicate the SD of total AT area of size-matched fed animals. Note that at day 22, refed animals have regained total AT to an equivalent level as size-matched continuously fed animals. B: Adiposity profiles showing intertrait correlations at stage DFRSAT in food-restricted/refed animals and size-matched continuously fed animals. C: LDA classifies food-restricted/refed animals as phenotypic in 90% of instances.

       Differential capacities to replenish lipid postfood restriction lead to fat distribution differences in zebrafish

      As food restriction and refeeding alters fat distribution even after total adiposity levels have been restored, we next closely assessed how individual ATs responded to diet manipulation. We previously identified 34 regionally distinct zebrafish ATs that could be classified into two main divisions: IAT, which contains the VAT, and SAT (
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ). Calculation of the percent recovery of each AT (day 22 levels relative to day 1 levels) revealed that IATs had a more limited capacity to replenish lipid when compared with SAT (Fig. 6A). The exception to this was cardiac VAT (CVAT), which exhibited very high levels of lipid replenishment following refeeding (Fig. 6A). When food-restricted and refed animals (day 22) were compared with size-matched continuously fed animals, it become clear that IAT was unable to recover lipid levels to WT levels (Fig. 6B), whereas SAT exhibited robust levels of replenishment to WT levels (Fig. 6C). Indeed, the average IAT recovery was ∼80%, whereas SAT recovery was ∼160% (Fig. 6D). These distinct capacities to replenish lipid following food restriction and refeeding resulted in a reduced IAT:SAT ratio when compared with stage-matched continuously fed control fish (Fig. 6E). Taken together, food restriction and refeeding led to altered fat distribution, even after full replenishment of AT levels. These changes were caused by a general failure of IATs to replenish lipid levels.
      Figure thumbnail gr6
      Fig. 6IATs have a decreased capacity to replenish lipid stores following food restriction. A: Hierarchical clustering of individual ATs based on percent replenishment of lipid (day 22 lipid as a percent of day 1 lipid). Four clusters were identified (nos. 1–4) and ranked according to replenishment rates. Color-code indicates percent lipid replenishment (purple/blue, low; green/red, high). IATs are colored magenta. SAT are colored green, and non-AT measures (e.g., SL) are colored black. Pie charts depict the proportion of AT types in each cluster. Note that clusters that replenish lipid to a high degree (i.e., clusters 3 and 4) are increasingly composed of SATs. B, C: AT-lipid mobilization and replenishment dynamics over the course of food-restriction (days 1–11) and refeeding (days 12–22). The black line links mean AT area values at each timepoint. The black error bars denote SD around the mean at each timepoint. The gray lines denote each individual fish (N = 10). The red (IAT; B) or green (SAT; C) bars denote the SD around the mean of size and stage-matched continuously fed WT fish. These bars represent the expected IAT and SAT quantities in normal fish. Note that IAT after food restriction and refeeding does not recover to expected WT levels. D: The percent recovery of IAT and SAT reveals significant differences between the distinct AT divisions. E: The IAT:SAT ratio is significantly reduced in food-restricted and refed animals (day 22) when compared with size and stage-matched continuously fed fish. aCVAT, anterior CVAT; aDSAT, anterior dorsal SAT; AFCSAT, anal fin ray cluster SAT; APPSAT, appendicular SAT; ASAT, abdominal SAT; AVAT, abdominal VAT; BA, body area; BHD, branchihyal; CFRSAT, caudal fin ray SAT; CHD, ceratohyal; cPOS, central paraosseal; CSAT, cranial SAT; dOPC, dorsal oligodendrocyte progenitor cell; dPOS, dorsal paraosseal; DSAT, dorsal SAT; Food res., food-restricted and refed animals; HYD, hyal; IM, intermuscular; LPECSAT, ‘loose’ PECSAT; LSAT, lateral SAT; NVAT, nonvisceral AT; OCU, ocular; OPC, opercular; pCVAT, posterior CVAT; PECSAT, pectoral SAT; PELSAT, pelvic fin SAT; POS, paraosseal; pPECSAT, posterior pectoral SAT; RVAT, renal VAT; TSAT, truncal SAT; UHD, urihyal; vOPC, ventral opercular; vPOS, ventral paraosseal; VSAT, ventral SAT.

      DISCUSSION

      Deep phenotyping offers considerable potential for the identification of parameters that signify health and disease. Comprehensive and quantitative phenotyping methods are particularly important for adiposity traits that often exhibit complex and continuous phenotypes that are often not adequately assessed by qualitative methods. In this study, we utilize in vivo imaging in zebrafish to develop adiposity profiles that capture unique patterns and relationships that define an individual's adiposity. We further use LDA to assign individuals based on adiposity phenotype and develop this methodology to screen for both genetic and diet-induced changes in adiposity.
      Developing quantitative methodology that can robustly evaluate complex morphological and functional phenotypic changes is essential for understanding and interpreting health and disease states. Most studies have so far focused on “large-effect” phenotypes. However, for complex quantitative traits, phenotypes are often subtler (
      • San-Miguel A.
      • Kurshan P.T.
      • Crane M.M.
      • Zhao Y.
      • McGrath P.T.
      • Shen K.
      • Lu H.
      Deep phenotyping unveils hidden traits and genetic relations in subtle mutants.
      ). Adiposity is a continuous, complex, quantitative trait; therefore, suitable phenotyping methods are needed. The methodology reported here leverages our recent identification and developmental analysis of zebrafish ATs to generate integrated adiposity profiles for individuals (
      • Minchin J.E.
      • Rawls J.F.
      A classification system for zebrafish adipose tissues.
      ). The adiposity profiles generated in this study capture the patterns and relationships that define an individual's adiposity and can thus be used to define expected adiposity. LDA represents a tractable strategy to assign individuals to groups according to adiposity profile. Indeed, we found that LDA robustly and accurately classified individuals with a low basal misclassification rate of 12.5%. gh mutants were used as a positive control, and LDA was able to classify. Therefore, LDA appears to be a robust method to identify multiple aspects of a pleiotropic phenotype. To maximize the deep phenotyping methods described here, it will be helpful to develop automated image analysis methods. Automating accurate segmentation of zebrafish ATs is challenging, owing to their irregular shape; however, the large signal-to-noise ratio of fluorescent lipophilic dyes, such as Nile Red, makes them ideal reagents to facilitate automated image segmentation.
      As a proof-of-principle, we evaluated whether adiposity profiles and LDA were useful as phenotyping methods in a genetic screen. In total, we tested eight ZMP mutants that have not been previously implicated in adiposity changes. LDA was able to identify a single mutant in nrp2b as having an adiposity phenotype. Closer analysis revealed that nrp2b has fairly subtle defects in total AT and VAT, which were difficult to detect by eye. Neuropilins (Nrps) are coreceptors for secreted Semaphorin ligands and cobind with Plexin receptors. Nrps have been implicated in neurogenesis and angiogenesis (
      • Mirakaj V.
      • Rosenberger P.
      Immunomodulatory functions of neuronal guidance proteins.
      ,
      • Álvarez-Aznar A.
      • Muhl L.
      • Gaengel K.
      VEGF receptor tyrosine kinases: key regulators of vascular function.
      ); however, no role has so far been found in adiposity. We recently identified that Plexin D1 (plxnd1) mutant zebrafish have altered regional adiposity, characterized by reduced levels of VAT (
      • Minchin J.E.
      • Dahlman I.
      • Harvey C.J.
      • Mejhert N.
      • Singh M.K.
      • Epstein J.A.
      • Arner P.
      • Torres-Vazquez J.
      • Rawls J.F.
      Plexin D1 determines body fat distribution by regulating the type V collagen microenvironment in visceral adipose tissue.
      ,
      • Minchin J.E.N.
      • Rawls J.F.
      Elucidating the role of plexin D1 in body fat distribution and susceptibility to metabolic disease using a zebrafish model system.
      ). We speculate that Nrp2b in zebrafish may be involved in Plxnd1-mediated VAT growth, possibly by facilitating the binding of Sema3 ligands to the Plxnd1 receptor. Importantly, we did not verify the effect of mutation on the affected gene. This was due to keeping the screen higher-throughput, but leads to the caveat that the specific alleles tested may not lead to large effects (i.e., loss of transcript or protein) (
      • Anderson J.L.
      • Mulligan T.S.
      • Shen M.C.
      • Wang H.
      • Scahill C.M.
      • Tan F.J.
      • Du S.J.
      • Busch-Nentwich E.M.
      • Farber S.A.
      mRNA processing in mutant zebrafish lines generated by chemical and CRISPR-mediated mutagenesis produces unexpected transcripts that escape nonsense-mediated decay.
      ). Therefore, our results should be interpreted such that the tested alleles have no phenotypes, but do not necessarily represent the function of the gene. Evaluating adiposity phenotypes with additional alleles will be essential for properly determining gene function.
      As an additional proof-of-principle, we used LDA to assess the effects of food restriction and refeeding to induce widespread adiposity changes. Food restriction and refeeding in mammalian models can lead to AT distribution differences (
      • Colman R.J.
      • Ramsey J.J.
      • Roecker E.B.
      • Havighurst T.
      • Hudson J.C.
      • Kemnitz J.W.
      Body fat distribution with long-term dietary restriction in adult male rhesus macaques.
      ,
      • Goss A.M.
      • Goree L.L.
      • Ellis A.C.
      • Chandler-Laney P.C.
      • Casazza K.
      • Lockhart M.E.
      • Gower B.A.
      Effects of diet macronutrient composition on body composition and fat distribution during weight maintenance and weight loss.
      ). For example, calorie restriction or alternate-day fasting in mouse results in preferential visceral fat loss (
      • Varady K.A.
      • Allister C.A.
      • Roohk D.J.
      • Hellerstein M.K.
      Improvements in body fat distribution and circulating adiponectin by alternate-day fasting versus calorie restriction.
      ). In zebrafish, we identified a striking phenotype whereby IATs, which are mainly composed of VAT, do not fully replenish lipid following refeeding. These changes were evident even in animals with equivalent total levels of AT as normally fed individuals. SATs have a greater capacity to redeposit lipid when compared with VATs, and refed fish display decreased IAT:SAT ratios. As accumulation of SAT is associated with reduced risk of metabolic disease in humans (
      • McLaughlin T.
      • Lamendola C.
      • Liu A.
      • Abbasi F.
      Preferential fat deposition in subcutaneous versus visceral depots is associated with insulin sensitivity.
      ) and zebrafish (
      • Minchin J.E.
      • Dahlman I.
      • Harvey C.J.
      • Mejhert N.
      • Singh M.K.
      • Epstein J.A.
      • Arner P.
      • Torres-Vazquez J.
      • Rawls J.F.
      Plexin D1 determines body fat distribution by regulating the type V collagen microenvironment in visceral adipose tissue.
      ), our results raise the possibility that the increased SAT accumulation induced by long-term food restriction and refeeding may have beneficial health effects.
      In conclusion, we present here a new method to comprehensively assess the adiposity phenotype in zebrafish. This methodology takes advantage of the fact that a wide range of adiposity traits can be quickly and accurately quantified in live zebrafish, thus providing a complete readout of an individual's adiposity profile. We anticipate that this method will be useful to identify additional new genetic and environmental factors governing AT development and physiology.

      Acknowledgments

      The authors thank the Zebrafish International Resource Center for providing mutant alleles and Steven Farber at the Carnegie Institution for helpful discussions.

      Supplementary Material

      REFERENCES

        • Gadde K.M.
        • Martin C.K.
        • Berthoud H.R.
        • Heymsfield S.B.
        Obesity: pathophysiology and management.
        J. Am. Coll. Cardiol. 2018; 71: 69-84
        • Fox C.S.
        • Massaro J.M.
        • Hoffmann U.
        • Pou K.M.
        • Maurovich-Horvat P.
        • Liu C.Y.
        • Vasan R.S.
        • Murabito J.M.
        • Meigs J.B.
        • Cupples L.A.
        • et al.
        Abdominal visceral and subcutaneous adipose tissue compartments: association with metabolic risk factors in the Framingham Heart Study.
        Circulation. 2007; 116: 39-48
        • Karpe F.
        • Pinnick K.E.
        Biology of upper-body and lower-body adipose tissue–link to whole-body phenotypes.
        Nat. Rev. Endocrinol. 2015; 11: 90-100
        • McLaughlin T.
        • Lamendola C.
        • Liu A.
        • Abbasi F.
        Preferential fat deposition in subcutaneous versus visceral depots is associated with insulin sensitivity.
        J. Clin. Endocrinol. Metab. 2011; 96: E1756-E1760
        • Flynn 3rd, E.J.
        • Trent C.M.
        • Rawls J.F.
        Ontogeny and nutritional control of adipogenesis in zebrafish (Danio rerio).
        J. Lipid Res. 2009; 50: 1641-1652
        • Imrie D.
        • Sadler K.C.
        White adipose tissue development in zebrafish is regulated by both developmental time and fish size.
        Dev. Dyn. 2010; 239: 3013-3023
        • Minchin J.E.
        • Rawls J.F.
        A classification system for zebrafish adipose tissues.
        Dis. Model. Mech. 2017; 10: 797-809
        • Oka T.
        • Nishimura Y.
        • Zang L.
        • Hirano M.
        • Shimada Y.
        • Wang Z.
        • Umemoto N.
        • Kuroyanagi J.
        • Nishimura N.
        • Tanaka T.
        Diet-induced obesity in zebrafish shares common pathophysiological pathways with mammalian obesity.
        BMC Physiol. 2010; 10: 21
        • McMenamin S.K.
        • Minchin J.E.
        • Gordon T.N.
        • Rawls J.F.
        • Parichy D.M.
        Dwarfism and increased adiposity in the gh1 mutant zebrafish vizzini.
        Endocrinology. 2013; 154: 1476-1487
        • Minchin J.E.
        • Rawls J.F.
        In vivo analysis of white adipose tissue in zebrafish.
        Methods Cell Biol. 2011; 105: 63-86
        • Minchin J.E.
        • Rawls J.F.
        In vivo imaging and quantification of regional adiposity in zebrafish.
        Methods Cell Biol. 2017; 138: 3-27
        • Robinson P.N.
        Deep phenotyping for precision medicine.
        Hum. Mutat. 2012; 33: 777-780
        • Hur M.
        • Gistelinck C.A.
        • Huber P.
        • Lee J.
        • Thompson M.H.
        • Monstad-Rios A.T.
        • Watson C.J.
        • McMenamin S.K.
        • Willaert A.
        • Parichy D.M.
        • et al.
        MicroCT-based phenomics in the zebrafish skeleton reveals virtues of deep phenotyping in a distributed organ system.
        Zebrafish. 2018; 15: 77-78
        • Hur M.
        • Gistelinck C.A.
        • Huber P.
        • Lee J.
        • Thompson M.H.
        • Monstad-Rios A.T.
        • Watson C.J.
        • McMenamin S.K.
        • Willaert A.
        • Parichy D.M.
        • et al.
        MicroCT-based phenomics in the zebrafish skeleton reveals virtues of deep phenotyping in a distributed organ system.
        eLife. 2017; 6: e26014
        • San-Miguel A.
        • Kurshan P.T.
        • Crane M.M.
        • Zhao Y.
        • McGrath P.T.
        • Shen K.
        • Lu H.
        Deep phenotyping unveils hidden traits and genetic relations in subtle mutants.
        Nat. Commun. 2016; 7: 12990
        • Kettleborough R.N.
        • Busch-Nentwich E.M.
        • Harvey S.A.
        • Dooley C.M.
        • de Bruijn E.
        • van Eeden F.
        • Sealy I.
        • White R.J.
        • Herd C.
        • Nijman I.J.
        • et al.
        A systematic genome-wide analysis of zebrafish protein-coding gene function.
        Nature. 2013; 496: 494-497
        • Schindelin J.
        • Arganda-Carreras I.
        • Frise E.
        • Kaynig V.
        • Longair M.
        • Pietzsch T.
        • Preibisch S.
        • Rueden C.
        • Saalfeld S.
        • Schmid B.
        • et al.
        Fiji: an open-source platform for biological-image analysis.
        Nat. Methods. 2012; 9: 676-682
        • Parichy D.M.
        • Elizondo M.R.
        • Mills M.G.
        • Gordon T.N.
        • Engeszer R.E.
        Normal table of postembryonic zebrafish development: staging by externally visible anatomy of the living fish.
        Dev. Dyn. 2009; 238: 2975-3015
        • Tabachnick B.
        • Fidell L.
        Using Multivariate Statistics. 1996;
        • Quinn G.P.
        • Keough M.J.
        Experimental Design and Analysis for Biologists.
        Cambridge University Press. 2002;
        • Murtagh F.
        • Legendre P.
        Ward's hierarchical agglomerative clustering method: which algorithms implement Ward's criterion?.
        J. Classif. 2014; 31: 274-295
        • Wickham H.
        ggplot2: Elegant Graphics for Data Analysis. 2009;
      1. R Core Team. R: A language and environment for statistical computing..
        R Foundation for Statistical Computing. 2013;
        • Bertsatos A.
        • Papageorgopoulou C.
        • Valakos E.
        • Chovalopoulou M.E.
        Investigating the sex-related geometric variation of the human cranium.
        Int. J. Legal Med. 2018;
        • Suhail Z.
        • Denton E.R.E.
        • Zwiggelaar R.
        Classification of micro-calcification in mammograms using scalable linear Fisher discriminant analysis.
        Med. Biol. Eng. Comput. 2018;
        • Oh J.
        • Cho D.
        • Park J.
        • Na S.H.
        • Kim J.
        • Heo J.
        • Shin C.S.
        • Kim J.J.
        • Park J.Y.
        • Lee B.
        Prediction and early detection of delirium in the intensive care unit by using heart rate variability and machine learning.
        Physiol. Meas. 2018; 39: 035004
        • Kočišová K.
        • Mišanková M.
        Discriminant analysis as a tool for forecasting companys financial health.
        Procedia Soc. Behav. Sci. 2014; 110: 1148-1157
        • Shungin D.
        • Winkler T.W.
        • Croteau-Chonka D.C.
        • Ferreira T.
        • Locke A.E.
        • Magi R.
        • Strawbridge R.J.
        • Pers T.H.
        • Fischer K.
        • Justice A.E.
        • et al.
        New genetic loci link adipose and insulin biology to body fat distribution.
        Nature. 2015; 518: 187-196
        • Heid I.M.
        • Jackson A.U.
        • Randall J.C.
        • Winkler T.W.
        • Qi L.
        • Steinthorsdottir V.
        • Thorleifsson G.
        • Zillikens M.C.
        • Speliotes E.K.
        • Magi R.
        • et al.
        Meta-analysis identifies 13 new loci associated with waist-hip ratio and reveals sexual dimorphism in the genetic basis of fat distribution.
        Nat. Genet. 2010; 42: 949-960
        • Mejhert N.
        • Wilfling F.
        • Esteve D.
        • Galitzky J.
        • Pellegrinelli V.
        • Kolditz C.I.
        • Viguerie N.
        • Tordjman J.
        • Naslund E.
        • Trayhurn P.
        • et al.
        Semaphorin 3C is a novel adipokine linked to extracellular matrix composition.
        Diabetologia. 2013; 56: 1792-1801
        • Shimizu I.
        • Yoshida Y.
        • Moriya J.
        • Nojima A.
        • Uemura A.
        • Kobayashi Y.
        • Minamino T.
        Semaphorin3E-induced inflammation contributes to insulin resistance in dietary obesity.
        Cell Metab. 2013; 18: 491-504
        • Walton R.G.
        • Zhu X.
        • Tian L.
        • Heywood E.B.
        • Liu J.
        • Hill H.S.
        • Liu J.
        • Bruemmer D.
        • Yang Q.
        • Fu Y.
        • et al.
        AP2–NR4A3 transgenic mice display reduced serum epinephrine because of increased catecholamine catabolism in adipose tissue.
        Am. J. Physiol. Endocrinol. Metab. 2016; 311: E69-E81
        • Wilson A.M.
        • Shao Z.
        • Grenier V.
        • Mawambo G.
        • Daudelin J.F.
        • Dejda A.
        • Pilon F.
        • Popovic N.
        • Boulet S.
        • Parinot C.
        • et al.
        Neuropilin-1 expression in adipose tissue macrophages protects against obesity and metabolic syndrome.
        Sci. Immunol. 2018; 3: eaan4626
        • Löffler D.
        • Behrendt S.
        • Creemers J.W.
        • Klammt J.
        • Aust G.
        • Stanik J.
        • Kiess W.
        • Kovacs P.
        • Korner A.
        Functional and clinical relevance of novel and known PCSK1 variants for childhood obesity and glucose metabolism.
        Mol. Metab. 2016; 6: 295-305
        • Heinonen S.
        • Buzkova J.
        • Muniandy M.
        • Kaksonen R.
        • Ollikainen M.
        • Ismail K.
        • Hakkarainen A.
        • Lundbom J.
        • Lundbom N.
        • Vuolteenaho K.
        • et al.
        Impaired mitochondrial biogenesis in adipose tissue in acquired obesity.
        Diabetes. 2015; 64: 3135-3145
        • Mirakaj V.
        • Rosenberger P.
        Immunomodulatory functions of neuronal guidance proteins.
        Trends Immunol. 2017; 38: 444-456
        • Álvarez-Aznar A.
        • Muhl L.
        • Gaengel K.
        VEGF receptor tyrosine kinases: key regulators of vascular function.
        Curr. Top. Dev. Biol. 2017; 123: 433-482
        • Minchin J.E.
        • Dahlman I.
        • Harvey C.J.
        • Mejhert N.
        • Singh M.K.
        • Epstein J.A.
        • Arner P.
        • Torres-Vazquez J.
        • Rawls J.F.
        Plexin D1 determines body fat distribution by regulating the type V collagen microenvironment in visceral adipose tissue.
        Proc. Natl. Acad. Sci. USA. 2015; 112: 4363-4368
        • Minchin J.E.N.
        • Rawls J.F.
        Elucidating the role of plexin D1 in body fat distribution and susceptibility to metabolic disease using a zebrafish model system.
        Adipocyte. 2017; 6: 277-283
        • Anderson J.L.
        • Mulligan T.S.
        • Shen M.C.
        • Wang H.
        • Scahill C.M.
        • Tan F.J.
        • Du S.J.
        • Busch-Nentwich E.M.
        • Farber S.A.
        mRNA processing in mutant zebrafish lines generated by chemical and CRISPR-mediated mutagenesis produces unexpected transcripts that escape nonsense-mediated decay.
        PLoS Genet. 2017; 13: e1007105
        • Colman R.J.
        • Ramsey J.J.
        • Roecker E.B.
        • Havighurst T.
        • Hudson J.C.
        • Kemnitz J.W.
        Body fat distribution with long-term dietary restriction in adult male rhesus macaques.
        J. Gerontol. A Biol. Sci. Med. Sci. 1999; 54: B283-B290
        • Goss A.M.
        • Goree L.L.
        • Ellis A.C.
        • Chandler-Laney P.C.
        • Casazza K.
        • Lockhart M.E.
        • Gower B.A.
        Effects of diet macronutrient composition on body composition and fat distribution during weight maintenance and weight loss.
        Obesity (Silver Spring). 2013; 21: 1139-1142
        • Varady K.A.
        • Allister C.A.
        • Roohk D.J.
        • Hellerstein M.K.
        Improvements in body fat distribution and circulating adiponectin by alternate-day fasting versus calorie restriction.
        J. Nutr. Biochem. 2010; 21: 188-195