During gonadal differentiation of zebrafish, males temporarily develop a “juvenile ovary” that is later degenerated and transformed into a testis (Takahashi,1977; Maack and Segner,2003; Wang et al.,2007). Although histological analysis showed the presence of degenerating oocytes in these transforming gonads and apoptosis was implicated to be involved with the process (Uchida et al.,2002), the molecular mechanism of this “juvenile ovary-to-testis” transformation has never been addressed. Interestingly, this process can be mimicked in females by feeding them with Fadrozole, an inhibitor of the estrogen-synthesizing enzyme, P450 aromatase (Fenske and Segner,2004; Uchida et al.,2004).
P450 aromatase (Cyp19a1) is the most important steroidogenic enzyme for ovarian differentiation due to its essential role in the production of 17β-estradiol (E2; reviewed by Devlin and Nagahama,2002). E2 is believed to be the major sex hormone for inducing and maintaining ovarian development in fish (Yamamoto,1969). Exposing zebrafish to E2 at 100 ng/L before and during the time of sex differentiation resulted in a female-biased sex ratio at maturity (Brion et al.,2004).
Two different cyp19a1 loci have been identified in zebrafish: cyp19a1a is predominantly expressed in the ovary, whereas cyp19a1b mainly in the brain (Chiang et al.,2001). The transcript level of cyp19a1a was found to be 100 times higher in adult ovary than in testis by real-time polymerase chain reaction (PCR); however, there was no obvious difference in the level of cyp19a1b between female and male brains (Sawyer et al.,2006). cyp19a1a showed a pattern of increasing expression level in granulosa cells surrounding stage IB to stage III oocytes by in situ hybridization, but it could not be detected from the testis at any stage (Rodriguez-Mari et al.,2005). These data suggest that cyp19a1a plays more important role in gonadal divergence than cyp19a1b does.
P450 11β-hydroxylase (Cyp11b, also known as P45011β) is the key enzyme for the synthesis of 11-ketotestosterone (11-KT) in testis, the most predominant androgen in fish (Baroiller et al.,1999; Young et al.,2005). 11-KT can induce all stages of spermatogenesis in Japanese eel testes in vitro (Miura et al.,1991). Elevated levels of 11-KT generated by hormone implantation resulted in 100% masculinization in prespawning female honeycomb groupers (Bhandari et al.,2006).
Of interest, Cyp11b uses the same substrate—testosterone as Cyp19a1 (Baroiller et al.,1999; Young et al.,2005). The successful inhibition of Cyp19a1 will probably leave more substrates for Cyp11b to produce more 11-KT. This probability could explain why females could be sex reversed so readily into neo-males when treated by aromatase inhibitor (McAllister and Kime,2003; Fenske and Segner,2004; Uchida et al.,2004).
Anti-Müllerian hormone (AMH or MIS) is the target of SOX9 in the mammalian testis differentiation pathway (De Santa Barbara et al.,1998; Arango et al.,1999). Amh is expressed in the Sertoli cells of fetal testes and induces regression of the Müllerian ducts, the anlage of the female internal reproductive organs that would otherwise differentiate into Fallopian tubes, uterus, and the upper part of the vagina (Munsterberg and Lovell-Badge,1991; Josso et al.,1993; Lee and Donahoe,1993). Moreover, the AMH protein has been found to be able to inhibit the transcription of Cyp19a1 in mammals (Vigier et al.,1989; di Clemente et al.,1992; Josso et al.,1998; Rouiller-Fabre et al.,1998). Ovine fetal ovaries treated with AMH protein release high level of testosterone instead of estradiol (Vigier et al.,1989).
The zebrafish ortholog of amh has been isolated and characterized recently (Rodriguez-Mari et al.,2005; von Hofsten et al.,2005). Similarly to the mammalian orthologs, zebrafish amh is highly expressed in the Sertoli cells after testicular differentiation (Rodriguez-Mari et al.,2005). At the same time, amh transcript could also be detected in the ovarian follicle layer where cyp19a1a is expressed (Chiang et al.,2001; Goto-Kazeto et al.,2004; Rodriguez-Mari et al.,2005), suggesting that they are colocalized in the same cells. However, these studies have not investigated whether amh was involved in the “juvenile ovary-to-testis” transformation. Analysis of the potential role of amh in this process would provide further insight into the complex mechanism of gonadal differentiation.
In this study, we report about the cloning of zebrafish cyp11b, and about the comparative analysis of its expression profiles together with those of cyp19a1a and amh during gonadal development by real-time PCR and in situ hybridization. Individuals at a particular stage were selected based on the gonadal expression of enhanced green fluorescent protein (EGFP) in a vas::egfp transgenic line as described earlier (Wang et al.,2007). The data indicate that down-regulation of cyp19a1a, possibly by amh, might be one factor leading to “juvenile ovary-to-testis” transformation. The data also suggest that cyp11b or its product, 11-ketotestosterone, is unlikely to be the inducer of zebrafish gonad transformation, as proposed earlier for some protogynous fish species (Alam et al.,2006).
The Cyp11b Enzyme of Zebrafish Showed Conservation of All Four Motifs Found Earlier in Other Teleost Orthologs
Our rapid amplification of cDNA ends (RACE) experiments yielded a 1,992-bp cDNA fragment containing a 1,557-bp-long open reading frame for zebrafish cyp11b (DQ650710). When BLAST-ed against the zebrafish assembly featured on Ensembl (Zv6), the results revealed that the locus was located on chromosome 16, spanning a region of 7,062 bp and consisting of 10 exons (Fig. 1A).
The predicted protein sequence of zebrafish Cyp11b shared 74.9% similarities with that of the medaka ortholog (Yokota et al.,2005), 76.0% with one rainbow trout ortholog Cyp11b1 (Kusakabe et al.,2002), 76.2% with the other rainbow trout ortholog Cyp11b2 (Liu et al.,2000), and 76.8% with the ortholog from Japanese eel (Jiang et al.,1996; Fig. 1B). All four conserved motifs identified earlier [binding sites for steroids, oxygen, heme/steroid, and heme (Kusakabe et al.,2002)] were clearly preserved in the zebrafish ortholog as well (Fig. 1B).
cyp11b mRNA Was Localized to Presumptive Leydig Cells in the Adult Testis and Its Level Was Four Magnitudes Higher Than That in the Ovary
The mRNA of cyp11b could be detected in all the adult organs tested by real-time PCR (Fig. 2A). However, the transcript level was at least two magnitudes higher in the testis than in any other tissue, when compared with the ovarian level a difference of nearly four magnitudes was recorded. Accordingly, the level of 11-KT was approximately 200 times higher in testes than in ovaries when quantified by enzyme-linked immunosorbent assay (ELISA; Fig. 2B).
When the adult testis was analyzed by in situ hybridization, cyp11b mRNA was clearly detected in presumptive Leydig cells, which are usually organized into small clusters of three to six cells located among the clusters of germ cells (Fig. 3A). Larger clusters of presumptive Leydig cells could also be found usually on the edge of testis sections (Fig. 3B). cyp11b mRNA could not be detected from other types of cells in testis, such as germ cells and presumptive Sertoli cells. In zebrafish testis, germ cells were found in clusters, within which all the cells developed synchronously from spermatogonia to spermatocytes and sperm cells (Fig. 3E). Sertoli and Leydig cells were located among larger clusters of germ cells. Sertoli cells showed irregularly shaped nuclei and usually stayed solitary along the membrane of germ cell clusters (Leo van der Ven and Piet Wester, http://www.rivm.nl/fishtoxpat/). Although trace amount of cyp11b transcript could be detected from ovaries by real-time RT-PCR, it could not be found in any cell type by in situ hybridization (Fig. 3C).
Quantitative Analysis of amh, cyp11b, and cyp19a1a Showed Sexually Dimorphic Expression in the Zebrafish Gonads
The expression levels of amh, cyp19a1a, and cyp11b during gonad differentiation were tested in transgenic offspring expressing the vas::egfp transgene. Quantitative analysis performed by real-time PCR has shown that the transcript levels of all three genes observed in zygotes have decreased drastically by the end of the first day of embryogenesis (Fig. 4). During the 1–3 wpf, cyp19a1a became up-regulated, whereas amh and cyp11b became slightly down-regulated in all individuals tested, independently from their EGFP expression. At 3 wpf, dimorphic expression between EGFP-positive and EGFP-negative individuals could be observed for all these three genes, cyp19a1a being higher in the former, and amh and cyp11b higher in the latter. At 4 wpf, the transcript levels of amh and cyp11b were over 300 and 600 times higher in the testis than in ovary, respectively. In contrast, the level of cyp19a1a transcript was over a magnitude higher in the ovary than in the testes. While both amh and cyp11b showed increasing expression during testicular development from 3 wpf onward, cyp19a1a showed first a decrease, then from 5 wpf onward a largely stable level during ovarian development.
cyp19a1a Was Down-regulated, While amh and cyp11b Were Both Up-regulated During “Juvenile Ovary-to-Testis” Transformation
With the aid of the vas::egfp transgenic line, we selected some individuals with continuously increasing EGFP-derived fluorescence in their gonads (4 wpf; Fig. 5A) and others exhibiting decreasing EGFP levels, the hallmark of transforming gonads (5 wpf; Fig. 5B). The former were shown earlier to develop normal (i.e., nontransforming) ovaries, whereas the latter were undergoing “ovary-to-testis” transformation (Wang et al.,2007). In the “normal” ovaries with increasing EGFP fluorescence cyp19a1a was expressed in the somatic cells (presumptive granulosa cells) surrounding the healthy oocytes (Fig. 5C), but could not be detected anywhere in the transforming gonads with decreasing EGFP (Fig. 5D). In contrast, the amh transcript could not be detected in “normal” ovaries (Fig. 5E), instead, it was highly expressed in cells (presumptive Sertoli cells) surrounding the degenerating oocytes in the transforming gonads (Fig. 5F). The expression of amh persisted in the transforming gonads around the cavities (Fig. 5F star) from which the oocytes appeared to be completely eliminated presumably by apoptosis (Uchida et al.,2002). cyp11b was also only expressed in the transforming gonads; the signal was usually localized to the margins of the transforming gonad, without showing any relationship to the position of degenerating oocytes (Fig. 5G,H). Also, based on the observed signal intensities on the sections, the expression level of cyp11b appeared to be substantially lower than that of amh.
amh Expression Preceded That of cyp11b in the Transforming Gonads
Individuals at the beginning (early stages; Fig. 6A) and in the middle (Fig. 6B) of their gonadal transformation (Wang et al.,2007) were also selected for studying the developmental expression pattern of amh and cyp11b during the testis differentiation process. At the early stages, the expression of amh was not uniform across the whole gonad. It could not be detected in those regions that were still full of oocytes (Fig. 6C), but it was clearly expressed in regions containing few or no oocytes (Fig. 6E). In contrast, cyp11b could not be detected anywhere in such kinds of gonads (Fig. 6G, adjacent section to that shown on Fig. 6E). As the transformation process progressed further, amh could be detected in all sections of the gonads (Fig. 6D,F). It was often found in cells surrounding a cluster of germ cells, and cells beside degenerating oocytes. At the same time, cyp11b mRNA has appeared gradually, first on the edge of the gonad and later also inside the gonad, presumably in the Leydig cells (Fig. 6H, adjacent region to that of 6F).
The down-regulation of cyp19a1a and sequential up-regulation of first amh and then cyp11b during gonadal transformation as observed on adjacent sections of the same transforming testis by in situ hybridization were also confirmed by the real-time PCR data (Fig. 7). In a normal ovary at 28 dpf both genes showed very low expression levels. The transforming gonads of all six specimen (32–35 days postfertilization [dpf]) showed reciprocal expression levels for amh and cyp11b vs. cyp19a1a, the higher the level of former two, the lower that of the latter. Both amh and cyp11b showed a dramatic increase during gonadal transformation, but cyp11b was obviously delayed compared with amh (see the expression levels in OT1 individual) and became highly expressed only when the testes were fully differentiated (see 42d T individual on Fig. 7).
Sexually Dimorphic Expression of amh, cyp11b, and cyp19a1a
We have shown earlier that the fluorescent intensity of the reporter gene in the gonad can be used to select individuals at various stages of their gonad differentiation from 16 dpf onward (Wang et al.,2007). In this study, we have analyzed the expression levels of amh, cyp11b, and cyp19a1a in the gonads of zebrafish individuals transgenic to vas::egfp (Krovel and Olsen,2002).
From 1 to 3 wpf, the transcript level of cyp19a1a has increased in both EGFP-positive transgenic individuals (with ovaries) and EGFP-negative transgenic ones (with undifferentiated gonads), although it was slightly higher in the former. This finding suggest that cyp19a1a plays an important role in the early differentiation of ovary, because initially all zebrafish individuals develop a “juvenile ovarian” structure (Takahashi,1977; Maack and Segner,2003; Wang et al.,2007). From 4 wpf onward, cyp19a1a was clearly expressed at a higher level in the differentiated ovary than in the testis.
Transcription of amh and cyp11b showed a sudden increase after 3 wpf, when some individuals entered the ovary-to-testis transformation process indicated by a short transient EGFP expression in their gonads. This increase has indicated that these two genes are likely to be important for male differentiation. The sexually dimorphic expression of cyp19a1a, amh, and cyp11b was maintained in both sexes until adulthood with certain variations at some stages. Notably, all three genes seemed to show a slightly decreasing trend during ovarian development from 3 wpf onward. This finding was probably due to the fact that the relative amount of β-actin transcript had increased, as the oocytes grew in size (from 10 μ to 800 μ in diameter), but the granulosa and theca cells (in which the three genes were expressed) did not. Therefore, this decreasing trend may not reflect the actual expression levels of genes in the ovarian somatic cells.
The expression levels of these three genes were also studied in detail during gonadal transformation (4 to 6 wpf) of some individuals. The transcript levels of all three genes in these individuals were between those of 4 wpf ovaries and 6 wpf testes (Fig. 7). The transcript level of cyp19a1a in transforming gonads was lower than those measured in the ovary at 4 wpf, but still higher than those in fully differentiated testes (6 wpf). In contrast, amh and cyp11b showed the opposite trend.
These results support earlier observations (McAllister and Kime,2003; Fenske and Segner,2004; Uchida et al.,2004) that cyp19a1a probably plays an important role for the differentiation of ovary and the maintenance of the ovarian fate. The down-regulation of cyp19a1 may be one of the reasons leading to the degeneration of ovary during gonadal transformation in males. At the same time, the data also indicate that amh and cyp11b are probably involved in the differentiation of testis.
amh Is a Good Candidate Gene for Revealing the Mechanism of Gonadal Transformation in Male Zebrafish
In our study, amh mRNA could not be detected in 4 wpf normal ovaries with stage I oocytes, confirming the results of Rodriguez-Mari et al. (2005). After the ovaries started their transformation into testes, amh could be gradually detected. It was not only expressed in areas surrounding germ cells, but also in the near vicinity of degenerating oocytes. When tested by real-time PCR, the increase of amh level was accompanied by the decrease of cyp19a1a levels. Cyp19a1a was expressed in the presumptive granulosa cells around some stage I oocytes in nontransforming ovaries (Rodriguez-Mari et al.,2005), but it could not be detected during the process of juvenile ovary-to-testis transformation. Thus, at the beginning the oocytes were surrounded by ovarian supporting (presumptive granulosa) cells expressing cyp19a1a when they were still alive, and later by presumptive Sertoli cells expressing amh while they were degenerating. There is evidence that granulosa and Sertoli cells have a common precursor in mice (Albrecht and Eicher,2001), but the question whether the zebrafish amh-expressing cells originate from the former cyp19a1a expressing cells or from new precursors could not be answered at this point. Further work may be done to find out whether (1) amh and cyp19a1a are colocalized in the same cells at any stage during gonadal transformation and (2) ovarian supporting cells of zebrafish might have the ability to de-differentiate into precursors and later differentiate into Sertoli cells, or (3) transdifferentiate into Sertoli cells as it was described for rat (Guigon et al.,2005).
Our results support the hypothesis that the peptide hormone gene amh may inhibit the expression of zebrafish cyp19a1a proposed earlier by Rodriguez-Mari et al. (2005), as it was shown in mammals (Vigier et al.,1989; di Clemente et al.,1992; Josso et al.,1998). Because chemical inhibition of the Cyp19a1a enzyme can cause the transformation of zebrafish ovaries to testes in vivo (McAllister and Kime,2003; Fenske and Segner,2004; Uchida et al.,2004), amh might be a rather important factor leading to the natural gonadal transformation in males.
The Most Predominant Male Steroid Hormone 11-KT Is Not the First Signal During Zebrafish Testicular Differentiation
Steroid hormones were proposed to be the main inducers driving gonadal differentiation in fish (Yamamoto,1969). This suggestion was later supported by finding steroid-producing cells before phenotypic gonadal differentiation in salmonids using various molecular and histological methods (for review, see Nakamura et al.,1998; Devlin and Nagahama,2002). The roles of 11-KT were investigated in protogynous hermaphrodite fish species that showed similar gonadal morphology during the natural sex reversal process to that observed in juvenile zebrafish (Wang et al.,2007). Increased 11-KT concentrations were observed during the natural ovary–testis transition in several species, such as Thalassoma dupperrey (Nakamura et al.,1989), Sparisoma viride (Cardwell and Liley,1991), and Epinephelus merra (Bhandari et al.,2003; Alam et al.,2006). Artificial female-to-male sex reversal could also be induced by treating individuals of two of these species with 11-KT (Cardwell and Liley,1991; Bhandari et al.,2006). Thus, Alam et al. (2006) proposed that 11-KT may provide the stimulus for female to degenerate oocytes and initiate sex change in these species.
However, all the above studies seem to place too much emphasis on the function of steroid hormone produced by Leydig cells and neglect the factors originating from Sertoli cells during testicular differentiation. In zebrafish amh has been shown to be expressed in Sertoli cells (Rodriguez-Mari et al.,2005; von Hofsten et al.,2005), the same cellular localization as demonstrated for mammalian AMH (reviewed by Rey et al.,2003). In zebrafish amh-positive cells could even be detected in undifferentiated gonads as revealed by in situ hybridization (Rodriguez-Mari et al.,2005).
In our present study, we compared the developmental expression pattern of the early Sertoli marker amh with Leydig cell-derived cyp11b gene by in situ hybridization and real-time PCR. The results showed that amh was up-regulated earlier and to a higher level than cyp11b, suggesting that 11-KT, the product of Cyp11b, was not the earliest factor during testicular differentiation. Therefore, 11-KT should not be considered to be the primary inducer of the natural gonad transformation in zebrafish, but rather an important downstream player of the process. In agreement with data reported earlier for Japanese medaka (Matsuda et al.,2002; Kobayashi et al.,2004; Matsuda,2005) and for mammals (for review, see Brennan and Capel,2004; Ross and Capel,2005), our data seem to support earlier differentiation of Sertoli cells than Leydig cells in the developing testis.
Fish Keeping and Breeding
All zebrafish were kept in AHAB recirculation systems (Aquatic Habitats, Apopka, FL) at ambient temperature (26–28°C) and at a 14-hr light / 10-hr dark cycle. Hemizygous embryos from the transgenic vas::egfp zebrafish line (AB background) were a kind gift from Dr. Lisbeth Charlotte Olsen (SARS, Bergen, Norway). They were raised to maturity in our lab, and inter-crossed to generate homozygous individuals. All transgenic offspring used for experiments were obtained by crossing homozygous transgenic males with wild-type AB females.
RNA Isolation and Real-Time PCR
Before the age of 1 wpf, whole embryos were used for RNA isolation. During the 1–3 wpf periods, the gonads would be still too small to be isolated, therefore, cropped trunk regions containing the gonad were used for the analysis at those stages. For samples over 4 wpf, gonads were isolated from EGFP-positive and EGFP-negative individuals, respectively. RNAs of 3 wpf EGFP-positive individuals were collected in two ways, from the isolated gonads of some individuals and from the body trunks of other with similar EGFP intensity to create a normalizing factor to link the gene expression levels in 1–3 wpf with those of over 4 wpf. Individuals with transforming and nontransforming gonads were sorted on the basis of the dynamic changes of EGFP expression as described earlier by Wang et al. (2007).
From samples weighing more than 100 mg (e.g., pooled embryos, larvae, and adult organs) RNA was isolated by Trizol-based procedure (Invitrogen, catalog no. 15596-026), whereas from smaller amounts of tissue (e.g., isolated gonads at 4 wpf) the RNeasy kit (Qiagen, catalog no. 74104) was used. All RNA samples were treated by RNase-free DNase to remove possible genomic DNA contamination. First-strand cDNA was synthesized by SuperScript III Reverse Transcriptase (Invitrogen, catalog no. 18080-093) using oligo(dT)20 as primer. Real-time PCR was performed using MyiQ Single-Color Real-Time PCR Detection System from Bio-Rad. The reaction (20-μl final volume) contained the following: 10 μl of iQ SYBR Green Supermix, 0.4 μl of each primer (10 μM), 1 μl of cDNA, and water. Both forward and reverse primers were designed to span two exons to avoid amplification from genomic DNA (see the Supplementary Table for primer sequences, which can be viewed at http://www.interscience.wiley.com/jpages/1058-8388/suppmat). The size of PCR products was approximately 180–200 bp. Reactions were run in duplicate for each sample.
A 10× dilution series of plasmid containing the cDNA insert was analyzed at the same time to produce a standard curve (correlation coefficient > 0.999). From such a standard curve, the absolute copy number of transcript can be determined by obtaining the value of threshold cycles (CT). The relative transcript levels between different samples were normalized to the same amount of β-actin.
Cloning of Zebrafish cyp11b Full-Length cDNA
To obtain the zebrafish ortholog of cyp11b, we first used a rainbow trout cyp11b1 sequence (AF179894) to search through nonmammalian expressed sequence tags (ESTs) by blastn in GenBank. The best hit was a 477-bp-long zebrafish EST (EB933597), which showed 85% identity with the query. We then used this EST to screen through the zebrafish genome assembly of Ensembl (Sanger Institute, Hinxton, Cambridge, UK) and obtained the sequence of the zebrafish cyp11b genomic locus (Zv5_NA4630). The predicted cDNA from this locus had 68% identity with the full length of rainbow trout cyp11b1. RACE was then performed by using FirstChoice RLM-RACE kit (Ambion; catalog no. 1700), with specific primers designed from the predicted cDNA (see the Supplementary Table for primer sequences). All PCR products were cloned into pGEM-T Easy Vector (catalog no. A1360), and at least 12 colonies for each clone type were sequenced. Full-length sequence was finally obtained by assembling these PCR fragments by Sequencher software (Gene Codes Corp, Ann Arbor, MI). Protein alignments were produced with ClustalX (Thompson et al.,1997).
Determining the Concentration of 11-KT by ELISA
ELISA was performed using 11-keto-testosterone EIA Kit (Cayman Chemical, catalog no. 582751). Three testes and three ovaries were isolated from the adult zebrafish, respectively, and then macerated in distilled water on ice. Macerated samples were centrifuged at 14,000 rpm for 2 min, and the supernatant was loaded directly onto the plate provided by the kit together with a series of standards. The results were normalized by the total protein measured by Bradford assay.
In Situ Hybridization on Sections
The digoxigenin-labeled RNA probes, both sense and antisense, were produced from plasmids containing the cDNA of the interest gene using T7 or SP6 transcriptase (Promega). After purification, the probes were further hydrolyzed into short fragments approximately 300–400 nucleotides in length with carbonate buffer (final concentration, 40 mM NaHCO3 and 60 mM Na2CO3). To study the expression patterns of amh, cyp19a1a, and cyp11b during the “juvenile ovary-to-testis transformation” process, individuals were selected based on the decreasing fluorescence in vas::egfp transgenic zebrafish (Wang et al.,2007). Specimens were fixed in 4% paraformaldehyde in PBS (pH 7.4) at 4°C overnight. Later, they were embedded in 2% agar, soaked in 30% sucrose, and then frozen in OCT medium to be sectioned (12 μm) by cryomicrotome (Leica). The protocol for hybridization was a kind gift from the Postlethwait lab: it followed the method published earlier by Strahle et al. (1994) and later modified by Jowett et al. (1995), but used 70°C instead of 55°C for hybridization and 65–70°C for following washes. The final pictures were taken by Zeiss Axioplan 2 microscope with Nikon DXM1200F digital camera and ACT-1 software.
The authors thank Woei Chang Liew and Alex Kuok Weai Chang for their help with the ELISA assay; Anne Vatland Krøvel and Lisbeth Charlotte Olsen for providing the vas::egfp transgenic zebrafish line; John H. Postlethwait for their protocol for in situ hybridization; Richard Bartfai for his help in the assembly of the cyp11b locus and for critical reading of the early versions of the manuscript; the members of the Reproductive Genomics Lab at TLL for discussions and advice; and an anonymous referee for useful suggestions. Xingang Wang acknowledges the useful suggestions from Vladimir Korzh, Philippa Melamed, and Toshie Kai.