The Phosphoinositide 3-Kinase Regulatory Subunit p85α Can Exert Tumor Suppressor Properties through Negative Regulation of Growth Factor Signaling

The class I_A phosphoinositide 3-kinase (PI3K) pathway plays a central role in growth factor signaling and oncogenesis. On activation of a receptor tyrosine kinase (RTK), PI3K is activated and generates the second messenger phosphatidylinositol (3,4,5)-trisphosphate (PIP₃), which activates critical downstream targets such as Akt and mammalian target of rapamycin (mTOR). When dysregulated, the PI3K pathway has a causal role in many forms of cancer, including those of the breast (1), colon (2), and liver (3).

The PI3K enzyme is an obligate heterodimer with an SH2-containing regulatory subunit (p85) and a p110 catalytic subunit (4). The primary function of the p85 subunit is to bind, stabilize, and inhibit the p110 catalytic subunit until RTK activation (5). Oncogenic mutations of PI3K enzyme can affect the kinase domain but more commonly disable the ability of p85 to inhibit p110, thus leading to unchecked constitutive activity (6). Such cancer-causing mutations in p85 have been documented in a murine lymphoma model (7), as well as in human ovarian cancer, colon cancer (8), and glioblastoma (9).

Although p85 can function as an oncogene, knockouts (KO) of various isoforms of p85 subunits have revealed that p85 may endogenously function as a tumor suppressor (10). For instance, mice with heterozygous deletion of Pik3rl (11), the gene encoding p85α and its shorter isoforms p55α and p50α, display increased hepatic Akt activation and improved insulin sensitivity despite an overall decrease in PI3K activity (12). In a similar vein, mice with a liver-specific KO of Pik3r1 (L-Pik3r1KO) have augmented hepatic PI3K activity and improved metabolism. In addition, mice that are heterozygous for both Pik3r1 and Pten display increased Akt and S6 activation and a 2-fold increase of intestinal neoplasia compared with Pten heterozygotes alone (13). The mechanisms of this negative regulation by the p85 monomer in these situations are multifaceted. We and others have previously shown that in most cells there is a stoichiometric imbalance between p85 and p110, with the former being in excess of the latter (14, 15). This imbalance results in a net inhibition of binding of the p85-p110 heterodimer to receptor phosphotyrosines (14). The p85 monomer also has significant effects independent of its regulation of the p110 catalytic subunit, including the sequestration of insulin receptor substrate proteins and positive regulation of PTEN function (reviewed in ref. 16). Despite these significant data on the negative regulation of growth factor signaling, no studies have directly addressed the extent to which Pik3r1 alone can function to modify tumor growth and development in vivo.

Hepatocellular carcinoma (HCC) is one of the most lethal of human cancers, with a 5-year survival of <5% (17). Worldwide, HCC is the fifth most common cancer and the third most common cause of cancer death (18). At the molecular level, there is ample literature suggesting that hepatic tumorigenesis is promoted by dysregulated growth factor signaling through RTKs, such as the insulin and insulin-like growth factor-I receptors (IGF-IR), c-MET, or epidermal growth factor receptor (EGFR), which strongly promote tumor growth, antiapoptosis, and chemotherapy resistance, primarily via the PI3K pathway (19).

In the present study, we have sought to define the role of the p85 regulatory subunit of PI3K in carcinogenesis by examining PIK3R1 expression in human cancers and determining effects of liver-specific deletion of Pik3r1 (L-Pik3r1KO) on formation of liver tumors. Using the Oncomine database, we find that p85α expression is significantly reduced in many human cancers, including HCC. More importantly, we find that mice lacking Pik3rl in the liver develop HCC by ∼14 months of age, often with aggressive characteristics including lung metastases. The formation of these liver tumors correlated with increased levels of PIP₃, increased Akt activation, increased PTEN phosphorylation, and decreased PTEN protein and mRNA expression. Thus, PIK3R1 may function as an important tumor suppressor gene in the liver and may play a similar role in other tissues.

All animals were housed on a 12-hour light-dark cycle and fed a standard rodent chow. All protocols for animal use and euthanasia were approved by the Animal Care Use Committee of the Joslin Diabetes Center and Harvard Medical School in accordance with NIH guidelines. All mice in this study were on a mixed genetic background of 129Sv-C57BL/6-FVB.

The Oncomine 3.0 database was interrogated for expression levels of PIK3R1 using the online interface http://www.oncomine.org/ (discussed and reviewed in ref. 20). The data from Oncomine were converted to raw expression levels by taking the inverse log₂ of the raw values. Expression levels of PIK3R1 were then normalized to the corresponding control tissue in each experiment. Gene expression from liver tissue in lox/lox and L-Pik3r1KO mice was assessed by quantitative reverse transcription-PCR (RT-PCR), as described previously (21). Primers can be found in Supplementary Table S1.

Tissues were harvested immediately and stained in H&E using standard techniques. Phospho-Akt and PIP₃ immunofluorescence were performed as described previously (22).

Tissue homogenates were prepared in a tissue homogenization buffer from livers in the random-fed state, except when indicated (21). Rabbit polyclonal pan-p85α antibody was generated as described previously (23); otherwise, all other antibodies were purchased from Cell Signaling Technology. Protein expression was quantified by densitometry using NIH ImageJ and presented as ±SEM.

Hepatocytes were isolated by a collagenase digestion technique as described previously (24). For experiments involving growth factor stimulation, hepatocytes were serum starved for 10 hours in DMEM-H plus glutamine and antibiotics and then stimulated for 15 minutes with saline, human insulin (100 nmol/L; Novo Nordisk), EGF (20 ng/mL; R&D Systems), or platelet-derived growth factor (PDGF)-AB (1 ng/mL; R&D Systems).

Data are presented as ±SEM. Student's t test was used for statistical analysis between two groups. Statistical significance between multiple treatment groups was determined by ANOVA and Tukey's t test.

Changes in expression of the PI3K p85 regulatory subunit modify insulin/IGF-I action and PI3K activity (12) and enhance tumorigenic effects of Pten deficiency (13). To determine the potential role of the p85 regulatory subunit of PI3K in human cancer, we analyzed expression levels of PIK3R1, the gene that encodes p85, in Oncomine, an online database of >18,000 gene array studies (available online, http://www.oncomine.org/, or see ref. 20). The data collection was interrogated for differences in PIK3R1 mRNA expression between cancerous tissue and corresponding normal tissue, primarily from human tissue samples. PIK3R1 was significantly decreased in samples from human cancer tissues compared with control tissues (P < 0.05) in ∼30 different cancer studies (Supplementary Table S2). These data included >1,600 tissue specimens representing cancers of the prostate, liver, lung, breast, kidney, and others (Supplementary Table S2). For instance, PIK3R1 expression was significantly decreased by 17% to 75% in prostate cancer and by 19% to 46% in lung cancer (25–28) and reduced by 18% (P < 0.001) in bladder cancer (29), 22% (P < 0.01) in ovarian cancer (30), and 18% (P < 0.05) in breast cancer (Fig. 1A and B; ref. 31). These data show a pattern of reduced expression in human cancer and suggest a possible role of p85 as a tumor suppressor in a number of situations. In the case of HCC, the expression patterns of p85 showed a correlation with grade of malignancy (Fig. 1C). Thus, p85 mRNA levels were inversely correlated with stage of HCC, both at stage IIIA (32) and metastatic disease within the liver (Fig. 1D; ref. 33). These data suggest a potential negative regulatory effect of p85 on tumorigenesis. To directly test this hypothesis in vivo, we created a mouse model with a tissue-specific KO of Pik3r1 in the liver.

Mice lacking Pik3r1 in the liver (L-Pik3r1KO) were generated by crossing mice homozygous for a floxed allele of Pik3r1 with those heterozygous for the albumin-Cre transgene, as described previously (22). Although p85α plays a critical role in hepatic insulin signaling (16), we wondered if this molecule also negatively regulated other signaling pathways, such as the EGFR and PDGF receptor, because they have both been shown to play critical roles in HCC progression (34, 35). We isolated primary hepatocytes from L-Pik3r1KO mice and stimulated them with insulin, PDGF, or EGF and assessed the response of downstream mediators Akt and mitogen-activated protein kinase (MAPK; Fig. 2). As observed in other cell types that lack p85α (11), L-Pik3r1KO primary hepatocytes showed a 1.5-fold increase in insulin-stimulated Akt phosphorylation but no differences in insulin-stimulated MAPK phosphorylation (Fig. 2; Supplementary Fig. S1). Primary hepatocytes from L-Pik3r1KO mice also displayed augmented Akt phosphorylation in response to EGF or PDGF stimulation by 1.6- and 1.8-fold, respectively (Fig. 2), whereas there were no differences between control and KO hepatocytes in EGF- or PDGF-stimulated MAPK phosphorylation (Supplementary Fig. S1). Thus, deletion of p85α selectively enhances insulin and growth factor action through the Akt pathway, with no change in their effects on the MAPK pathway.

Because HCC rarely occurs without preceding liver damage, we analyzed livers from younger L-Pik3r1KO and lox/lox controls to determine the presence of premalignant pathology. A tabular summary of these findings is presented as Supplementary Table S3. At 6 months of age, there were no detectable abnormalities in lox/lox control livers, whereas four of ten 6-month-old KO mice examined showed evidence of hepatitis and necrosis (Fig. 3). On gross inspection, small necrotic lesions were readily visible on the surface of the mouse livers (Fig. 3A, left), and histology of these same livers showed significant neutrophilic infiltrate throughout the liver and periportal foci of necrotic hepatocytes with monocytic and neutrophilic infiltration (representative section in Fig. 3A, right). Seven of 12 young L-Pik3r1KO mice and 10 of 14 older KO mice showed evidence of this chronic hepatitis by liver histology. Gram stain and culture of these lesions revealed no evidence of bacterial infection.⁸

These L-Pik3r1KO mice also exhibited significant hepatic glycogen accumulation beginning at 2 months of age as determined by periodic acid-Schiff (PAS) staining, which selectively binds to glycogen within cells (Fig. 3B). This abnormal glycogen deposition is likely a consequence of increased insulin sensitivity. Indeed, 6-month-old L-Pik3r1KO mice have lower levels of fasting blood glucose and insulin (Supplementary Fig. S2A and B) as well as improved intraperitoneal glucose tolerance tests (Supplementary Fig. S2C). In addition, expression of the essential gluconeogenic enzymes phosphoenolpyruvate carboxykinase (PEPCK) and glucose-6 phosphatase (G6Pase) was attenuated by 50% and 70%, respectively, in L-Pik3r1KO livers (Supplementary Fig. S2D). Both PEPCK and G6Pase are key hepatic enzymes that break down glycogen during fasting to maintain blood glucose levels. Thus, livers of L-Pik3r1KO mice accumulate abnormally high amounts of glycogen, most likely due to enhanced insulin sensitivity and decreased gluconeogenesis. These glycogen deposits then create a source of inflammation and associated hepatosteatosis, which is similar to the pathobiology of human glycogen storage diseases (36).

At 10 to 12 months of age, this inflammation and glycogen deposition progressed to the formation of precancerous dysplastic nodules. In humans, HCC is often preceded by the development of small dysplastic nodules seen on biopsy or magnetic resonance imaging (MRI; refs. 37, 38). In 5 of 12 L-Pik3r1KO mice ages 10 to 12 months that exhibited diffuse low- and high-grade dysplastic nodules within the liver parenchyma, the nodules were readily apparent on gross dissection (Fig. 3C, left, black arrows) and at the microscopic level (Fig. 3C, right). In addition, we performed micro-MRI on several L-Pik3r1KO mice before sacrifice and found that these lesions seemed radiologically similar to dysplastic nodules in humans, as exhibited by contrast enhancement characteristics on T1-weighted imaging (Supplementary Fig. S3).

The heightened sensitivity to multiple growth factors in L-Pik3r1KO hepatocytes led us to posit that the chronic loss of Pik3r1 could be hepatocarcinogenic in older mice. Indeed, 86% (12 of 14) of the mice ages 14 to 20 months spontaneously formed HCC, whereas none of the control lox/lox mice had any evidence of cancer (see Supplementary Table S3). One lox/lox mouse developed a benign hepatic adenoma at age 23 months, which is an entirely normal finding in male BL/6 or 129 mice of this advanced age (39). On gross dissection, these older L-Pik3r1KO mice displayed large irregular livers that were infiltrated with tumors (Fig. 4A). Several of the livers exhibited cystic lesions that were filled with bile and necrotic debris (Fig. 4A, green arrow). Other mice exhibited massive livers that filled the entire peritoneal cavity, accompanied with intracapsular hemorrhage (Fig. 4A, red arrow). The livers of L-Pik3r1KO mice over 14 months of age were also larger than age-matched controls, with an average liver weight of L-Pik3r1KO mice of 2.0 ± 0.3 g (n = 11) compared with age-matched wild-type (WT) controls with average liver weights of 1.3 ± 0.1 g (n = 6; P < 0.05, compared with KO), primarily due to tumor weight. When adjusted for body weight, livers lacking Pik3r1 were nearly double the relative size of control livers (6.9% versus 3.4%, KO versus WT; P < 0.01).

Histologic analysis of these mice from 14 to 24 months of age often revealed significant infiltration of the normal liver parenchyma with HCC and dysplastic nodules. The histologic features of the HCC in L-Pik3r1KO mice were consistent with a high-grade cancer, with a high nuclear/cytoplasm ratio, prominent nucleoli, coarsened trabeculae with layers three to six cells wide, and multiple mitoses in nearly every high-powered field (Fig. 4B, left). In addition, many livers displayed prominent bile duct hyperplasia and periportal inflammation (Fig. 4B, right). These high-grade lesions were strongly correlated with the presence of pulmonary metastases, which occurred in ∼60% (7 of 12) of the KO mice with HCC (Fig. 4C). Of note, metastases were detected exclusively in the lung and were only observed in mice older than 16 months of age.

To confirm that inflammation may have played a role in the pathophysiology of these tumors, we performed quantitative RT-PCR analysis on HCC samples from L-Pik3r1KO mice (14–18 mo old) and compared them with age-matched controls (Fig. 4D). HCC samples showed a 3.0 ± 0.4–fold increase in the expression of CD68, a specific macrophage marker that is elevated in human HCC (40). In addition, this increase was associated with a 3.1 ± 0.98–fold increase in the expression of tumor necrosis factor-α (TNF-α); however, there was no statistically significant difference in the expression of interleukin-6 (IL-6).

To determine the molecular mechanisms by which Pik3r1 deletion promotes hepatocarcinogenesis, we performed Western blots on liver lysates from KO and control mice at ages between 6 and 18 months. These liver extracts were prepared from mice in a random-fed state without exogenous hormone stimulation. Compared with controls, whole liver extracts of L-Pik3r1KO mice showed an 80% to 90% decrease in p85α expression and a complete abrogation of p55α and p50α expression (Fig. 5A). Because the p85 subunit of PI3K is known to stabilize the p110α subunit (5), the loss of the p85 expression also resulted in a 75% to 85% decrease in levels of the catalytic p110α subunit of PI3K.

As expected, there was little activation of the PI3K or MAPK pathways in lox/lox mice at 6 months or 16 to 18 months, and little activation in 6-month-old L-Pik3r1KO livers, because these mice were not stimulated with exogenous growth factors before harvesting their livers (Fig. 5A). On the other hand, older L-Pik3r1KO livers that contained HCC showed a striking upregulation of Akt phosphorylation, by 3.84 ± 0.35–fold (Supplementary Fig. S4A; P < 0.01), compared with 6-month-old lox/lox controls. Likewise, MAPK phosphorylation/activity was significantly enhanced in KO mice with HCC by 2.37 ± 0.22–fold compared with young lox/lox mice (Supplementary Fig. S4B; P < 0.001). The significantly increased Akt phosphorylation in tumor samples was confirmed by direct visualization with immunofluorescent staining. Phospho-Akt detected in situ was increased by 4-fold in tumor samples compared with lox/lox and KO controls (Fig. 5B, quantitated in C). These signaling changes are likely cell autonomous, and not a result of systemic endocrine signals, because blood glucose and serum levels of insulin did not differ between 18-month-old control and L-Pik3r1KO mice (Supplementary Fig. S5A and B).

The mTOR pathway is critical for the activating protein synthesis and cell growth and is regulated by PI3K/Akt via the phosphorylation and inhibition of TSC1 and TSC2. The increased activity of Akt in the L-Pik3r1KO mice with HCC resulted in a significant increase in the phosphorylation of TSC2, which lies upstream of mTOR (Fig. 5D). In addition, the phosphorylation of ribosomal S6 protein, which is the ultimate downstream target of mTOR, was increased 8-fold (Fig. 5D).

The activation of Akt requires the binding of the enzyme to the second messenger PIP₃ via its pleckstrin homology (PH) domain. To determine levels of PIP₃ in the tumors, we used an in situ immunofluorescence technique that allowed us to semiquantitatively measure PIP₃ levels in hepatocytes (Fig. 6A; ref. 22). This technique showed that very little PIP₃ accumulated in the control lox/lox and young L-Pik3r1KO mice, which was expected because these livers were not exposed to stimulation before their collection. Tumor samples from L-Pik3r1KO mice, however, displayed significant PIP₃ accumulation. Quantitation of immunofluorescence levels showed that the HCC from L-Pik3r1KO mice had 4.92 ± 0.3–fold more PIP₃ than age-matched control lox/lox mice (1.0 ± 0.06) or younger L-Pik3r1KO mice (0.96 ± 0.05).

This enhancement of PIP₃ levels suggested that degradation by PTEN or some other lipid phosphatase might be dysregulated. We have previously shown that the loss of p85 partially abrogates PTEN activity and decreases PIP₃ turnover (12). Western blots on liver lysates from lox/lox and L-Pik3r1KO mice found a 53 ± 5.8% decrease in PTEN protein expression in 18-month-old KO mice compared with age-matched lox/lox controls and younger KO animals (Fig. 6B). This correlated with a ∼60% decrease in Pten mRNA as determined by quantitative PCR analysis of RNA transcripts from tumor samples compared with age-matched control livers (Fig. 6C).

Phosphorylation of PTEN on its COOH-terminal tail has also been shown to negatively regulate PTEN activity at the protein levels (41). Western blotting with an antibody that recognizes multiple phosphorylation sites in the COOH terminus of PTEN revealed a modest decrease in overall levels of phospho-PTEN, but after adjusting for overall decreased PTEN expression, this represented a 2.1 ± 0.4–fold increase in level of PTEN phosphorylation at the protein level (Fig. 6B) Taken together, these data show that PTEN protein and mRNA expression may be altered in the setting of chronic p85α deletion and that PTEN activity may also be negatively regulated by a state of relative hyperphosphorylation.

Over the past 2 decades, a number of studies have shown an integral role of PI3K in tumorigenesis (reviewed in ref. 4). This can occur through direct activation of PI3K by oncoproteins, such as polyoma middle T and RTKs, deletion of the PIP₃ phosphatase PTEN, and genetic mutation of proteins in the PI3K signaling pathway, including p85, p110, and Akt. In the present study, we provide evidence that decreased expression of the PI3K p85 regulatory subunit can also directly lead to tumor formation in a novel murine model of HCC.

An initial clue of the role of expression of the regulatory subunit of PI3K in cancer comes from the finding that PIK3R1 expression is decreased in many types of human cancers, including prostate, lung, breast, and HCC (Fig. 1A–D). Other human studies have also suggested the possible importance of p85 in human carcinogenesis. For instance, a functional missense mutation in PIK3R1 resulting in reduced p85 expression has been strongly linked with colon cancer (42). Likewise, the PIK3R1 gene is located on human chromosome 5q13, a region that is commonly deleted in cancers that use the PI3K pathways for growth, including HCC (43). Because other putative tumor suppressors involved in DNA damage repair are contained within the 5q13 chromosomal region, including XRCC4 and hRad17 (44), further studies must be performed to determine the extent to which PIK3R1 haploinsufficiency might account for the association of the 5q13 chromosomal deletion and cancer. However, in the case of HCC, the data from this study indicate that PIK3R1 may indeed be a meaningful tumor suppressor gene.

In our mouse model of a selective deletion of Pik3r1 in liver, a complex combination of cellular and molecular events contributes to progressive pathologic events that ultimately cause HCC, as illustrated schematically in Fig. 6D. Many features of this murine HCC model resemble human HCC. Human HCC is preceded by chronic inflammation from infectious or chemical etiologies (18), whereas this mouse model develops chronic hepatitis from abnormal glycogen deposition that arises from an intrinsic enhancement of hepatic insulin sensitivity (Fig. 3; ref. 12). Inflammation is readily observed at the cellular level via histology and at the molecular levels with increased levels of a macrophage marker and TNF-α (Fig. 4D). These chronic hepatitis and HCC are also described in mouse models of glycogen storage disease, as well as in humans with this disorder (36).

By 14 months of age, almost all L-Pik3r1KO mice develop an aggressive HCC as shown by the presence of pulmonary metastases in 60% of the mice with HCC, which is an uncommon feature in murine cancer models that exhibit spontaneous tumor growth. These results agree with our analysis of clinical human data that show a correlation of decreased Pik3r1 expression with stage IIIA or metastatic HCC, supporting the notion that p85 can serve as an endogenous tumor suppressor and that reduced levels of Pik3r1 may enable the formation of HCC.

The molecular mechanisms of HCC formation in L-Pik3r1KO mice are similar to those of human HCC in several respects. Notably, human HCC is correlated with dysregulated growth factor signaling, particularly through RTKs such as IGF-IR, PDGFR, and EGFR. L-Pik3r1KO mice have exquisite growth factor sensitivity, particularly through the PI3K/Akt axis, as shown by enhanced Akt signaling through insulin/IGF-I and other RTKs, including PDGFR and EGFR/ErbB3, which have all been implicated in human cancers (Fig. 2).

This enhancement of Akt signaling is the net effect of several contributing factors. We have previously described that p85 has a positive regulatory effect on PTEN, where the loss of p85 leads to decreased PTEN function and decreased breakdown of PIP₃, leading to accumulation of this second messenger and resulting in enhanced downstream signaling, most notably by PH domain–containing proteins such as Akt. In this study, we also find that despite a significant defect in the enzyme capacity of PI3K, tumor specimens from L-Pik3r1KO mice displayed five times more hepatic PIP₃ than control mice and a resultant 4-fold increase in Akt activation.

The mechanism of PTEN regulation by p85 in the context of HCC seems to be due to a diminished expression of PTEN protein levels that is a result of both decreased PTEN transcript production and increased phosphorylation of the PTEN protein. Phosphorylation of the COOH-terminal tail of PTEN is known to play a crucial role in regulating both its expression and its activity in vitro, but it is unclear how much this contributes to this human cancer or this murine model (41, 45). Nonetheless, the loss of PTEN expression that occurs in the older L-Pik3r1KO mice likely accelerates the formation of HCC, as it has been suggested to do in humans (46). The mechanism of how p85 might modulate PTEN mRNA expression or PTEN phosphorylation is unclear, and further studies must be conducted to determine the regulation of PTEN phosphorylation in terms of membrane recruitment of the enzyme, epigenetic events regulating PTEN (47), as well as the potential effects of p85 deletion on regulation of other PIP₃ phosphatases.

Despite many similarities with human HCC, this murine model differs in several key respects. First, the gross morphology of this murine cancer is different from the human form, marked by the presence of bilious cysts and bile duct proliferation. These histologic findings may reflect the role of p85 in bile canalicular transport (48). Second, despite the correlation between abrogated hepatic p85 and reduced PTEN expression in the murine HCC, many of the human cancer studies with decreased PIK3R1 expression did not also show a concomitant decrease in PTEN expression (Supplementary Fig. S6). The reason for this discrepancy is unknown but likely stems from the fact that the gene expression data may not adequately capture the multiple different ways that PTEN function can be compromised, including functional mutations that do not alter RNA transcript production (49), promoter methylation (47), or posttranslational regulation (50). Moreover, it may be possible that the HCC in different individuals or different populations results from different inciting events, such as infection (such as hepatitis) versus toxin (ethanol or aflatoxin B), and these could have completely different mechanisms that are masked in the gene array data.

In summary, we have shown that the p85 regulatory subunit of PI3K functions as an inhibitor of growth factor signaling and a novel tumor suppressor in the liver and most likely in other cancers as well. Increasing or restoring expression of p85 may thus represent an exciting potential therapeutic possibility against cancer.

C.R. Kahn: consultant/advisory board, Plexicon, Inc. The other authors disclosed no potential conflicts of interest.

We thank Michael Rourk for his technical assistance and Jonathan Glickman for his discussion on the HCC pathology.

Grant Support: NIH grants DK33201 and DK55545, Joslin Diabetes and Endocrinology Research Center grants DK34834 (C.R. Kahn) and GM41890 and CA089021 (L.C. Cantley), Southern Association for Institutional Research grant 2U24CA092782-07 (R. Weissleder), American Diabetes Association Medical Scholars Award (C.M. Taniguchi), and Medical Scientist Training Program Scholarship (Harvard Medical School).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.