Purification of bovine lens cell-to-cell channels composed of connexin44 and connexin50

Lens fiber cells are connected by extensive networks of junctions that cover >60% of their surface. The junctions belong to two different types. The overwhelming majority are 11-13 nm thick and exhibit an asymmetric structure formed by tetragonal arrays adhering to a protein-free apposing membrane (Lo and Harding; Costello et al., 1985, 1989; Zampighi et al., 1989). The 11-13 nm thick junctions are composed of a 28 kDa protein called the ‘major intrinsic protein’ (MIP) (Bok et al., 1982; Zampighi et al., 1989). Proteins of the MIP family are expressed in plants, yeasts, mammalian erythrocytes and bacteria. These proteins function in the transport of water and neutral molecules through membranes. Expression in Xenopus oocytes has shown that MIP functions as a water channel (Zampighi et al., 1995). Therefore, the structure and function of MIP eliminate the 1113 nm thick junctions in mediating lens cell communication.

Lens fiber cells are also connected through a thicker (16-18 nm) junction composed of a protein called MP70 (Kistler et al., 1985; Gruijters et al., 1987; Zampighi et al., 1989). This protein was identified as a connexin because its N terminus amino acid sequence was 40-50% identical with those of liver connexin32 and connexin26 and heart connexin43 (Kistler et al., 1988). Also, the N terminus sequence (including three degenerate positions) matched the deduced sequences of mouse lens connexin50 (White et al., 1992), rat lens connexin46 (Paul et al., 1991), bovine lens connexin44 (Gupta et al., 1994) and chick lens connexin56 (Rup et al., 1993). The identification of junctions composed of connexins indicated that lens fiber cells communicate via gap junctions channels.

However, the 16-18 nm thick lens junctions do not exhibit channels arranged in bidimensional plaques (crystalline or random), an important characteristic of gap junctions. This has complicated the demonstration that MP70 forms channels spanning the two plasma membranes and the intervening extracellular space, the hallmark of communication via gap junctions. An approach used to answer this question has been to treat isolated 16-18 nm thick junctions with nonionic detergent and to study the particles released into the solution by negative staining (Kistler and Bullivant, 1988). They showed that some particles released by the treatment exhibited sizes and shapes of ‘complete’ cell-to-cell channels in negatively stained preparations (Kistler et al., 1993). In this paper, we used this approach and developed a procedure that purifies milligram quantities of stable, monodisperse, and delipidated channels that, in negative staining appear as particles ∼11 nm diameter and ∼16 nm in length. We show that the ‘complete’ channels are composed of equal amounts of phosphorylated connexin44 and connexin50, probably assembled in the same channel.

Calf and adult lenses and fetal bovine eyes were obtained from Pel-Freez (Rogers AK). Octyl-β-D-glucopyranoside (OG) was purchased from Bachem (Bubendorf, Switzerland), decyl-β-D-maltopyranoside (DeM) from Sigma Chemical Co. (St Louis, MO); calf intestine alkaline phosphatase (20 U/μl), endoglycosidase H (200 U/200 μl) and neuraminidase (1 U/100 μl) from Boehringer (Mannheim, Germany). HPTLC silica 60 gel plates (without fluorescent indicator) from Merck (Darmstadt, Germany) and phosphomolybdic acid from Sigma Chemical Co.

Buffer A: 2 mM NHCO₃, 3 mM EDTA, 100 μM PMSF, 0.08 μM aprotinin, 1 μM pepstatin, 1 μM leupeptin, pH 8.0.

Buffer B: 4 mM Tris-HCl, 5 mM EDTA, 1.5 mM NaN₃, pH 8.0.

Buffer 1: 10 mM HEPES, 50 mM NaCl, 5 mM EDTA, 3 mM NaN_s, 1% OG, 10% glycerol, 5 mM DTT, pH 7.0.

Buffer 2: Buffer 1 plus 1 M NaCl.

Buffer 3: Buffer 1 plus 0.2% DeM instead of OG.

Buffer 4: Buffer 3 plus 1 M NaCl.

Buffer 5: 10 mM HEPES, 50 mM NaCl, 3 mM NaN_s, 0.1% DeM, 10% glycerol, pH 7.5.

Plasma membranes were isolated from fetal, calf and adult bovine lenses as described previously (Ehring et al., 1990; Zampighi et al., 1982; Zampighi et al., 1989). The procedure used 100-200 lenses from fetal, 3-month-old calves, or the cortical region of adult lenses. Decapsulated lenses were homogenized in 300 ml of Buffer A. The homogenate was diluted, filtered and centrifuged at 2,000 g for 10 minutes at 4°C. The pellets were resuspended in Buffer B and centrifuged at 2,000 g for 10 minutes. The pellets were resuspended in Buffer B with 4 M urea, centrifuged at 27,000 g for 20 minutes and then resuspended in Buffer B with 7 M urea an d centrifuged at 50,000 g for 90 minutes. The urea-extracted pellets were washed twice in Buffer B and stored at − 70°C in 200 μl aliquots (5-10 mg/ml). The procedure yielded 0.56 mg, 0.72 mg and 0.17 mg of plasma membrane protein per gram of fetal, calf and adult lens, respectively.

Plasma membranes were resuspended in Buffer 1 without OG at a protein concentration of 2 mg/ml. Solubilization was performed by adding solid OG to the suspension up to a final concentration of 25 mg OG per mg of membrane protein (∼3.5%). The solution was bath-sonicated 3 times for 10 seconds each, left on ice for 30 minutes, and centrifuged in a Beckman TLA-100 Ultracentrifuge (rotor TLA 100.2) at 120,000 g, at 4°C for 60 minutes. The supernatant was diluted (1+1 v/v) with Buffer 1 and called ‘OG solubilized proteins’.

The OG-solubilized proteins were applied to a MonoS (HR5/5) FPLC column (Pharmacia Fine Chemicals) equilibrated with Buffer 1. The ‘flow-through peak’ (area shaded in Fig. 1A, first panel) was collected and concentrated 10 times in a Centriprep30 microconcentrator (Amicon, Danvers, MA). The concentrate was diluted (1+9) with Buffer 3. This step replaced most of the OG with DeM, a detergent that stabilized the channels and permitted further purification. The fraction was applied to a MonoQ (HR 5/%) FPLC column (Pharmacia Fine Chemicals) equilibrated in Buffer 3. The column was eluted at a flow rate of 1 ml/minute with a continuous gradient formed by Buffers 3 and 4. The protein peak eluting at 350-450 mM NaCl (area shaded in Fig. 1A, second panel) was collected and concentrated to a final volume of 500 μl in a Centriprep100 microconcentrator, and then injected into a Superose 6 (HR 10/30) FPLC column (Pharmacia) equilibrated in Buffer 3. Fractions eluting from 6.5 ml to 10.5 ml contained ‘purified cell-to-cell channels’ (shaded area in Fig. 1A, third panel).

Starting with plasma membranes, the procedure took less than a day and yielded 0.5-1 mg of purified cell-to-cell channels per 10 mg of membrane protein. Protein concentration was determined by either the BCA (Pierce Chemicals, Rockford IL) or the Lowry (1951) assays.

SDS-PAGE was performed according to the procedure of Laemmli (1970) and immunoblotting according to the procedure of Towbin et al. (1979). The proteins were transferred onto nitrocellulose membranes in a semi-dry Milliblot-Electroblotter (millipore, Bedford, MA) at 250 mA for 30 minutes. Blocking was done with 4% bovine serum albumin in phosphate buffered saline (PBS) for 30 minutes at 45°C. The blots were probed with a monoclonal anti-MP70 (connexin50) or a polyclonal rat anti-connexin46 diluted in phosphate buffered saline. Primary antibodies were visualized with biotinylated horse anti-rat IgG or goat anti-rabbit IgG using a Vectastain ABC kit (Vector Laboratories, Burlingame, CA).

Coomassie Blue stained SDS-PAGE gels of purified cell-to-cell channel fractions were quantified using a Molecular Dynamic Computing Densitometer Model 3000. The purified cell-to-cell channel fractions (area shaded in Fig. 1A, third panel) were treated with alkaline phosphatase (see below). A ratio of 1 to 1.2 of connexin50 to connexin44 was estimated in three different fractions.

Connexin44 was probed with a polyclonal antibody against residues 411-416 of rat connexin46 (Paul et al., 1991), an antibody that also recognizes bovine connexin44 (Gupta et al., 1994). Connexin50 was probed with the monoclonal 6-4-B2-C6 against MP70 (Kistler et al., 1985), an antibody that also recognizes mouse connexin50 (White et al., 1992).

Dephosphorylation was performed by adding 25 U alkaline phosphatase to 2.5 nmoles of purified cell-to-cell channels. The samples were incubated for 30 minutes at room temperature and chromatographed in a Superose 6 column to remove the enzyme from the solution.

Deglycosylation was performed by adding 90 mU endoglycosidase H and 100 mU neuraminidase in 50 mM sodium acetate buffer pH 5.2 to 0.25 nmoles of purified cell-to-cell channels. The samples were incubated for 12 hours at 4°C and analyzed by SDS-PAGE.

Lipids were extracted from plasma membranes isolated from fetal, calf and adult lenses (0.2 mg membrane protein) and from 0.2 mg of purified cell-to-cell channels (0.3 nmoles) according to the method of Bligh and Dyer (1959). The chloroform phases were applied on HPLTC silica 60 gel plates. The plates were developed with CHCl₃/MeOH/H₂O (20:6.7:1 parts) and the lipids were visualized with phosphomolybdic acid.

Aliquots (∼1 μl) of purified proteins at a concentration of ∼0.5 mg/ml were spread on glow-discharged carbon-coated grids. Excess solution was blotted by touching the side of the grid on a filter paper. The grids were washed by touching the surface of 5-10 drops of distilled water and stained with a 2% uranyl acetate. The grids were examined in a Philips EM301 or CM12 electron microscope. Measurement of particle diameter and length were performed directly on the negatives in a Nikon comparator.

Cell-to-cell channels were purified from plasma membranes isolated from fetal and calf lenses. SDS-PAGE shows that the plasma membranes were composed of numerous protein bands, the principal of which migrated at ∼26 kDa and is called MIP (Fig. 1B, lane 1 and Fig. 4A). MIP was present in similar amounts in membranes isolated from fetal, calf and adult bovine lenses (Fig. 4A). On the other hand, the cell-to-cell channels are composed of MP70, a protein that appears as a diffuse 55-75 kDa band in SDS-PAGE (Fig. 1B, lane 1 and Fig. 4A).

The first step in the purification procedure transferred the plasma membrane proteins into nonionic detergents micelles (‘solubilization’). The aim was to produce stable, monodisperse channel/detergent complexes suitable for subsequent purification by column chromatography. Treatment with octyl-β-D-glucopyranoside (OG) in Buffer 1 solubilized >60% of the total plasma membrane proteins (Ehring et al., 1991). Comparison of SDS-PAGE gels of the plasma membranes and the OG-solubilized proteins showed that these conditions effectively solubilized most types of intrinsic proteins, including MIP and the cell-to-cell channels (Fig. 1B, lanes 1 and 2).

Purification of the channel/detergent complexes required chromatography of the OG-solubilized proteins in three different columns. Chromatography in a cation exchange column (MonoS) eluted cell-to-cell channels in the ‘flowthrough’ fractions (area shaded in the first panel of Fig. 1A). SDS-PAGE showed a dramatic decrease in the amount of MIP and a substantial increase in the amount of the diffuse 55-75 kDa band in the fractions (MP70) (Fig. 1B, lane 3).

Further purification required exchanging OG for decyl-β - maltopyranoside (DeM). OG was effective in transferring the intrinsic proteins from lipid bilayers into micelles. Probably because of its short acyl chain, however, OG produced unstable complexes which remained in solution for only short periods of time (∼10 hours). Reduced stability induced unspecific protein-protein aggregation which handicapped further purification. Replacement of OG with DeM formed complexes that remained stable in solution for over extended periods of time (months) and allowed for the purification of the channels.

The DeM/channel complex binds to the anion exchange column MonoQ. This permitted removal of most of the intrinsic lipids from the channel/DeM complex by extensive washes with Buffer 3. The channels were eluted from the MonoQ in a peak at 0.35-0.45 M NaCl with a continuous NaCl gradient formed by mixing Buffer 3 and 4 (area shaded in Fig. 1A, second panel). SDS-PAGE showed that the principal protein of the MonoQ peak was the diffuse 55-75 kDa band (MP70). Contaminants appear as weak bands at ∼35 kDa and 26 kDa (Fig. 1B, lane 4).

Chromatography of the channel fractions in a Superose 6 gel filtration column removed the contaminants seen after the cation exchange purification step. Proteins eluted in the void volume (6.5 ml) and in a shoulder that extended up to 12 ml into the internal volume of the Superose 6 column (Fig. 1A, third panel). SDS-PAGE showed that the diffuse 55-75 kDa band eluted from 6.5 ml to 10.5 ml which indicated heterogeneity in the size of the channel/DeM complex. Therefore, all fractions containing the diffuse 55-75 kDa band were studied in the electron microscope.

Negative staining showed that fractions eluting from 7.5 ml to 9.5 ml (shaded in Fig. 1A, third panel) contained annuli of about 11±0.6 nm (s.d., n=105) in diameter and particles 16±0.8 nm in length (s.d., n=96) and ∼11 nm in width (Fig. 2). The ∼16 nm long particles were composed of two domains joined through a 3 nm wide, lightly stained region (Fig. 2 inset). SDS-PAGE showed that these fractions contained principally the 55-75 kDa band and only a small amount of weaker bands at ∼35 kDa (Fig. 1B, lane 5). These fractions are referred to as ‘purified cell-to-cell channels’.

Fractions eluting from 6.5 ml to 7.5 ml also contained principally the diffuse 55-75 kDa protein band, however, negative staining showed aggregates composed of 3-6 channels (not shown, but see Kistler et al., 1993). Such an aggregate can explain why those channels eluted in the column void volume. They were discarded. Negative staining showed that fractions eluting from 9.5 ml to 10.5 ml contained single channels of similar shape and dimensions to those in the ‘purified cell-to-cell channel’ fractions. However, SDS-PAGE showed a substantial increase in the amount of the protein bands migrating at ∼35 kDa that most likely represent endogenous proteolytic fragments of the channels (Kistler et al., 1990). They were also discarded.

Therefore, SDS-PAGE and negative staining showed that fractions eluting from 7.5 ml to 9.5 ml from the Superose 6 column contained ‘complete’, monodisperse cell-to-cell channels composed principally of a diffuse 55-75 kDa protein band.

Immunoblots showed that the upper part of the diffuse band was recognized by anti-MP70, a monoclonal that crossreacts with sheep and mouse connexin50 (Fig. 1C, second lane) (Kistler et al., 1985; White et al., 1992). The lower part was recognized by an anti-rat connexin46, a polyclonal that also crossreacts with bovine connexin44 (Fig. 1C, third lane) (Gupta et al., 1994). Both antibodies also recognized lower molecular mass bands migrating at about 50 kDa and 40 kDa, respectively (Fig. 1C, lanes 2 and 3). Whereas the weaker bands correlate well with the molecular masses of connexins 44 and 50, the stronger diffuse bands were 15-20 kDa larger than the molecular mass predicted from the amino acid sequence. We refer to the strong bands as the ‘modified’ and to the weaker bands as the ‘unmodified’ forms of connexins 44 and 50.

To account for the decreased electrophoretic mobility of the modified connexins 44 and 50, we tested for glycosylation and phosphorylation Deglycosylation with endoglycosidase H and neuroaminidase did not alter either the appearance or the position of the 55-75 kDa band (not shown). Dephosphorylation with alkaline phosphatase on the other hand changed the diffuse band into two distinct bands (Fig. 3A). The sharper bands were the ‘modified’ forms of connexins 44 and 50 because their molecular masses were larger than those predicted from the sequence (Fig. 3B and C). Densitometry determined a ratio of 1.2 connexin44 to 1 connexin50.

Immunoblotting showed that alkaline phosphatase affected both connexins differently. Dephosphorylation of connexin50 showed double bands at the position of both the ‘modified’ and ‘unmodified’ forms (Fig. 3B) in channels purified from either fetal or calf lenses. In contrast, dephosphorylation shifted the mobility of the modified form of connexin44 from calf lenses (Fig. 3D) but not from fetal lenses (Fig. 3C). The slightly higher mobility of fetal connexin44 was also seen in plasma membranes (Fig. 4B) suggesting that it is not an artifact of the purification procedure.

We have also determined the relative amounts of ‘modified’ and ‘unmodified’ forms of connexins 44 and 50 in fetal, calf and the cortex of adult lenses. ‘Modified’ connexin44 was present in similar quantities in membranes isolated from fetal and calf lenses but it was greatly decreased in membranes from adult lenses (Fig. 4B). ‘Unmodified’ connexin44, on the other hand, was present in membranes from fetal lenses only (Figs 3C and D, and 4B, first lane). Previous studies did not detect ‘unmodified’ connexin44 in fetal lenses probably because of low protein loading (Jarvis et al., 1993). In contrast, ‘modified’ and ‘unmodified’ forms of connexin50 were present in equal amounts in plasma membranes from fetal, calf and the cortex of adult lenses. Thus, it appears that the two connexins are processed independently.

Lens plasma membranes have an unusual lipid composition that includes the highest concentration of cholesterol and sphingomyelin in the body (Fig. 5, lanes 1-3). Thin layer chromatography indicates that the purification procedure used in this study removed cholesterol, sphingomyelin and other phospholipids from the purified preparations. This indicates that intrinsic lipids do not participate in the stability of the channels.

Purified cell-to-cell channels appeared as annuli ∼11 nm diameter and as particles ∼16 nm in length and ∼11 nm wide. The ∼16 nm particles must correspond to side views of ‘complete’ cell-to-cell channels. They are long enough to span the junctional membranes and the intervening extracellular gap and also exhibit two distinct domains joined by a constriction that collects stain molecules. This view is consistent with a channel composed of two hemichannels joined through the extracellular domains (Unwin and Zampighi, 1980, Unwin and Ennis, 1984). The annulus, on the other hand, exhibits a larger diameter than that of the channels in the junction (∼8 nm) (Revel and Karnovsky, 1967). Such a discrepancy is not unexpected since the channels were purified as a complex with DeM molecules, a fact that should increase their diameters, but not their lengths.

While the ∼16 nm long particles can only be best explained as the side view of a ‘complete’ channel (dodecamer), the ∼11 nm annulus may correspond to the top view of either a ‘complete’ channel or a hemichannel (hexamer). Gel filtration chromatography showed that top and side views eluted in the same fractions, indicating that the annulus and the ∼16 nm long particle must have very similar, if not identical, molecular masses. Taken together, the data indicate that the purified fractions must contain ‘complete’ cell-to-cell channels, not a mixture of channels and hemichannels.

The protein composition of lens cell-to-cell channels has been a topic of controversy (for review see Zampighi et al., 1992). Due to its abundance and location in junctions, early work advanced the hypothesis that MIP formed the lens cell-to-cell channels (Peracchia and Peracchia, 1980). This view was challenged by the identification of the 16-18 nm thick junctions (Zampighi et al., 1982) and the demonstration that they were composed of MP70 (Kistler et al., 1985; Kistler and Bullivant, 1987; Gruijters et al., 1987; Zampighi et al., 1989). The observation that purified ‘complete’ channels are composed of equal amounts of connexins 44 and 50 provides strong support to the view that channels located in the 16-18 nm thick junction mediate lens cell communication.

Gap junctions are extremely resistant to solubilization with mild detergents, a property that greatly facilitated their isolation but has hampered the purification of individual channels and hemichannels. Presently, connexin32 from liver or overproduced in a baculovirus/insect cell expression system (Stauffer et al., 1991), and MP70 from lens fiber cells (Kistler and Bullivant, 1988; Kistler et al., 1993) have been solubilized under mild conditions. Connexin32 was purified by column chromatography and shown to be assembled in hexamers (hemichannels) by single particle rotational averaging (Stauffer et al., 1991). Lens MP70 was only partially purified and shown by negative staining to be assembled in ‘complete’ channels (Kistler et al., 1993). Our studies extend the work of Kistler and co-workers by showing that ‘complete’, stable, monodisperse and delipidated channels can be purified in large quantities from bovine lenses. Chemical and structural methods may now be used to study the channels to learn about their high resolution structure and function.

The purified channels were composed principally of phosphorylated connexin44 and connexin50 and of only small quantities of the dephosphorylated forms. This result is in agreement with previous studies showing that mouse connexin50, rat connexin46 and bovine connexin44 are phosphorylated in plasma membrane fractions (White et al., 1992; Paul et al., 1991; Gupta et al., 1994) and connexin46 in cultured fiber cells (Jiang et al., 1993). In crude lens homogenates, dephosphorylation with alkaline phosphatase shifted the mobility of connexin44 to the molecular mass of the unphosphorylated form (Gupta et al., 1994). In the purified channel fraction, however, alkaline phosphatase transformed the diffuse 55-75-kDa band into two sharper bands, but it did not decrease the mobility to that of the dephosphorylated forms. The discrepancy may be due to differences in the preparations (purified channels versus crude lens homogenates), to the experimental conditions used in the enzymatic reaction, or to the possibility that the enzyme dephosphorylated only partially the channel/detergent complexes.

The identification of channels composed of phosphorylated connexins may be important to the lens. Previous studies have shown that phosphorylation may control gap junction assembly (Musil et al., 1990) or can modify channel properties such as gating (Burt and Spray, 1988) and unitary conductance (Saez et al., 1986). The importance of phosphorylation in channel function has been supported by expression in heterologous systems. Injection of cDNA coding for bovine connexin44 and rat connexin46 in Xenopus oocytes expressed the unphosphorylated form of both connexins. This resulted in single oocyte depolarization and lysis (Gupta et al., 1994; Paul et al., 1991). Although these results can be rationalized by assuming that hemichannels formed of unphosphorylated connexin44 are open, it remains to be explained why the expression of unphosphorylated connexin50 (or of any other connexin) does not lyse single oocytes.

Assembly of cell-to-cell channels is a complex process that involves the interaction of connexin within the plane of the membrane to form hemichannels, and across the intercellular space to form the ‘complete’ channel. Cells expressing a single type of connexin assemble homotypic cell-to-cell channels. Cells expressing two (Kuraoka et al., 1993; Nicholson et al., 1987; Traub et al., 1989; Zhang and Nicholson, 1989) or more types of connexins can assemble either homotypical or ‘mixed’ cell-to-cell channels. Homotypic channels can be localized in the same gap junction or be segregated into different junctions. Depending on the subunit composition of the hemichannel, ‘mixed’ cell-to-cell channels can be assembled ‘heterotypically’ (hemichannels composed of a single connexin) or heteromerically (hemichannels composed of two or more types of connexins).

Heterotypic channels have been assembled by pairing Xenopus oocytes expressing different types of connexins (Swenson et al., 1989; Werner et al., 1989). Conclusive evidence for ‘mixed’ cell-to-cell channels in tissues, however, does not exist. In the lens, connexin46 and connexin50 have been localized in the same fiber cells, and more importantly, in the same junction (Paul et al., 1991; Louis et al., 1993; TenBroek et al., 1992). This paper shows that the purified channels contain equal amounts of connexins 44 and 50. It seems unlikely that the fractions are composed of mixtures of homotypic channels assembled of either connexin44 or connexin50. The channels were purified based on differences in their molecular mass, hydrodynamic radii and charge. This strategy should have separated homotypic channels because the calculated pI of connexins 44 and 50 differs by more than 2 pH units. Therefore, the simplest explanation that can rationalize our results in purified cell-to-cell channels, and those in plasma membranes (Paul et al., 1991; TenBroek et al., 1992) and in fiber cells (TenBroek et al., 1994) is that connexins 44 and 50 are assembled in the same channel (‘mixed’ or ‘hybrid’).

Expression of multiple connexins may represent a ‘safety’ mechanism that ensures formation of cell-to-cell channels when a particular connexin is not expressed, is expressed at low levels, or is vulnerable to endogenous proteolysis. It may also be used to alter channel properties such as conductance, gating, or voltage-dependence. For example, heterotypic channels assembled from liver connexin32 and connexin26 exhibited voltage dependence and asymmetry not present in homotypic channels formed by either connexin (Barrio et al., 1991; Rubin et al., 1992; Versalis et al., 1993). It is not possible to determine which, if any, of these possible functions is occurring in the lens. Perhaps the complexity detected in the architecture of lens cell-to-cell channels is related to the vectorial transport of ions, metabolites, and water between posterior and anterior surfaces, which is crucial in maintaining lens transparency and homeostasis.

We thank Dr J. Kistler and Dr D. Paul for providing the monoclonal antibody against MP70 (connexin50) and the polyclonal against connexin46. The technical assistance of Mr M. Kreman is also acknowledged. The work was supported by a Royal Society Fellowship, the Medical Research Council, the Department of Trade and Industry, Zeneca and Glaxo under the protein engineering LINK program (N.K.) and National Institute of Health grant EY-04110 (G.A.Z.).