Glycoprotein Hormone Assembly in the Endoplasmic Reticulum

All three human glycoprotein hormone heterodimers are assembled in the endoplasmic reticulum by threading the glycosylated end of α-subunit loop two (α2) beneath a disulfide “latched” strand of the β-subunit known as the “seatbelt.” This remarkable event occurs efficiently even though the seatbelt effectively blocks the reverse process, thereby stabilizing each heterodimer. Studies described here show that assembly is facilitated by the formation, disruption, and reformation of a loop within the seatbelt that is stabilized by the most easily reduced disulfide in the free β-subunit. We refer to this disulfide as the “tensor” because it shortens the seatbelt, thereby securing the heterodimer. Formation of the tensor disulfide appears to precede and facilitate seatbelt latching in most human choriogonadotropin β-subunit molecules. Subsequent disruption of the tensor disulfide elongates the seatbelt, thereby increasing the space beneath the seatbelt and the β-subunit core. This permits the formation of hydrogen bonds between backbone atoms of the β-subunit cystine knot and the tensor loop with backbone atoms in loop α2, a process that causes the glycosylated end of loop α2tobe threaded between the seatbelt and the β-subunit core. Contacts between the tensor loop and loop α2 promote reformation of the tensor disulfide, which explains why it is more stable in the heterodimer than in the uncombined β-subunit. These findings unravel the puzzling nature of how a threading mechanism can be used in the endoplasmic reticulum to assemble glycoprotein hormones that have essential roles in vertebrate reproduction and thyroid function.

The glycoprotein hormones are heterodimers of two cystine knot proteins (1)(2)(3) in which a glycosylated loop of one subunit (loop ␣2) 1 is surrounded by a strand of the other "like a seatbelt" (1). This topology raises questions as to how these heterodimers might be assembled. We have found that the human glycoprotein hormone subunits combine by a process in which the glycosylated end of loop ␣2 is threaded beneath the seatbelt while it is latched (22). Although the hCG heterodimer can be assembled by a mechanism in which the seatbelt is wrapped around loop ␣2 after the subunits dock (4,5), this appears to be a minor pathway that can be used to form some hormone analogs that are unable to latch their seatbelts to ␤-subunit loop 1. This "salvage" pathway may have had a role in the evolution of glycoprotein hormones in some teleost fish (23).
Purified glycoprotein hormone subunits have long been known to recombine slowly in vitro in oxidizing conditions (6), a phenomenon that occurs while all the disulfides in both subunits remain intact (7). This showed that assembly can occur by a mechanism in which the glycosylated end of loop ␣2 is threaded beneath the seatbelt. hCG assembly is accelerated substantially by protein-disulfide isomerase (8) and low concentrations of reducing agents, however (7). Furthermore, ␤-mercaptoethanol-catalyzed assembly is blocked by agents that react with thiols, e.g. iodoacetate (7), an indication that threading is limited by one or more disulfide bonds that must reform for the heterodimer to be stable after assembly is completed. Using a highly sensitive procedure capable of detecting and identifying trace amounts of free thiols, we found that only one of the 11 hCG disulfides was disrupted significantly during ␤-mercaptoethanol-catalyzed assembly (7). Because this disulfide stabilizes a small loop in the ␤-subunit seatbelt, this finding implies that reduction of this disulfide enhanced subunit combination in vitro by elongating the seatbelt. This would increase the size of the hole in the ␤-subunit, thereby facilitating the passage of the glycosylated end of loop ␣2. Whereas these studies showed how ␤-mercaptoethanol facilitated threading, because the reverse process must be inhibited to stabilize the heterodimer, it remained unclear as to why ␤-mercaptoethanol did not facilitate heterodimer dissociation.
Studies of hCG assembly in cells using pulse-chase methods (4) indicated that the disulfide that stabilizes the small seatbelt loop forms before the disulfide that latches the seatbelt to loop ␤1 (4). Based on our observations that most hCG is assembled while the seatbelt remains latched (22) and that this is impeded in vitro by the small loop in the seatbelt (7), we reasoned that the small seatbelt loop might break and reform during heterodimer assembly in the ER. Because of the transient nature of this process, we expected that formation, disruption, and reformation of this disulfide would not be detected using pulse-chase methods. To study this process in cells, we took advantage of an hCG antibody that can distinguish ␤-subunit isoforms in which the seatbelt is latched normally from those in which it is latched to alternate sites. By observing how free cysteines in the ␣and ␤-subunits influence seatbelt latching and hormone assembly, we found that formation, disruption, and reformation of the disulfide that stabilizes the small seatbelt loop is critical for efficient heterodimer assembly in cells. Because this disulfide stabilizes the small loop that shortens the seatbelt, we refer to it as the "tensor." As shown here, the tensor disulfide appears to function as a redox-regulated switch. Formation of the tensor disulfide facilitates latching of the end of the seatbelt to the ␤-subunit core. Disruption of the tensor disulfide after the seatbelt has been latched facilitates threading. Reformation of the tensor disulfide after the glycosylated end of loop ␣2 has been threaded beneath the seatbelt completes assembly and stabilizes the heterodimer.

EXPERIMENTAL PROCEDURES
The sources of reagents and methods of analyses used in these studies are described in the preceding manuscript (22). Constructs used in this study (Fig. 1), which were produced by standard polymerase chain reaction and cassette mutagenesis methods (9) were sequenced prior to use. The name of each construct used in these studies reflects its amino acid sequence. For example, hCG␤-C93A refers to an hCG ␤-subunit analog in which ␤Cys 93 is converted to alanine. Because ␤Cys 93 forms a disulfide in the heterodimer with ␤Cys 100 , converting ␤Cys 93 to alanine leaves ␤Cys 100 unpaired, making it available to form a disulfide with nearby cysteines. During these studies we prepared several hCG analogs and monitored them using well characterized monoclonal antibodies. The relative locations of the mutations and the antibody binding sites in the free ␤-subunit and in the heterodimer are summarized in Figs. 2 and 3. The rationale for each of the studies is described in the text and figure legends.

The Tensor Loop Is Usually Formed Before the Seatbelt Is
Latched-Most hCG is assembled in the endoplasmic reticulum by a threading mechanism in which ␣2 passes beneath the seatbelt through a hole in the ␤-subunit (22). We had found that subunit combination was facilitated in vitro when the tensor disulfide (i.e. ␤93-␤100) was disrupted (7) and considered the possibility that this disulfide forms in the endoplasmic reticulum only after the heterodimer is assembled. Pulse-chase analyses (4) suggested that the tensor disulfide forms before the seatbelt latch disulfide (i.e. ␤110-␤26). Because the seatbelt is latched prior to hCG assembly (22), this implies that the tensor loop forms before the subunits dock, a phenomenon that would retard threading (7). Using the rationale outlined in Fig.  2, we re-examined the order in which the tensor and seatbelt latch disulfides are formed by comparing the abilities of ␤-subunit analogs lacking one or both tensor cysteines to latch their seatbelts to ␤Cys 26 . Initial studies were designed to learn if the seatbelt would become latched in hCG ␤-subunit analogs that lacked the ability to form the tensor loop (Fig. 2, upper panel). To be certain that passage of the ␤-subunit through the secretory pathway did not influence our results, we conducted these studies using hCG ␤-subunit analogs that are secreted as well as those that are retained within the cell because of the presence of an ER retention signal at their carboxyl terminus (i.e. KDEL) (10). We have found that this signal delays heterodimer secretion (22), presumably by keeping the ␤-subunit in the ER.
To deduce the relative timing of seatbelt latching and tensor loop formation, we studied how adding or removing ␤-subunit cysteines influenced seatbelt latch formation in the free ␤-subunit. In these studies we quantified the total amount of ␤-subunit with a sandwich immunoassay using monoclonal antibodies B101 for capture and 125 I-B110 for detection (Fig. 3). This assay detects those molecules that have formed the ␤-subunit core, i.e. the portion of the molecule created by the formation of the cystine knot. We monitored the fraction of the total heterodimer in which the seatbelt was latched to ␤Cys 26 using a sandwich assay employing monoclonal antibodies B101 for capture and 125 I-B111 for detection (Fig. 3). B111 does not recognize the hCG ␤-subunit when the seatbelt is latched to cysteines that are added to other sites in the ␤-subunit. As described next, using this property of B111, we determined that formation of the tensor disulfide usually occurs prior to formation of the seatbelt latch disulfide.
hCG ␤-subunit analogs lacking both tensor cysteines latched their seatbelts to ␤Cys 26 , a finding that showed that the tensor disulfide does not need to form for the seatbelt to be latched properly (Table I, rows 1 and 2). For example, B111 recognized ␤-subunit analogs hCG␤-C93A,C100A and hCG-C93A,C100A-KDEL, which lack cysteines ␤Cys 93 and ␤Cys 100 (Table I, rows 1 and 2, column C93A and C100A), as well as or better than hCG␤ and hCG␤-KDEL, which contain both tensor cysteines (Table I, rows 1 and 2, column C93 and C100). This observation does not mean that the seatbelt is latched before the tensor disulfide in most ␤-subunit molecules, however. As described next, other data suggested that the tensor disulfide usually forms first. The amino acid sequences of the hCG ␣and ␤-subunits are illustrated in single letter format. The substitutions made are indicated above the amino acid sequence and each mutant can be identified by its name. Thus, ␣-Q5C represents an analog of the ␣-subunit in which ␣Gln 5 was converted to cysteine. Analogs that contained the additional sequence "KDEL" were secreted slowly (22) and are presumed to be retained in the endoplasmic reticulum as has been observed for other proteins containing this signal (10). Thus, hCG␤-C26A-KDEL refers to an analog of the hCG ␤-subunit in which the natural seatbelt latch site at ␤Cys 26 was converted to alanine and four additional residues (Lys-Asp-Glu-Leu) were fused to its COOH terminus. hCG␤-␦93:100DA refers to an analog of the hCG ␤-subunit in which residues Cys 93 -Arg 94 -Arg 95 -Ser 96 -Thr 97 -Thr 98 -Asp 99 -Cys 100 , the entire tensor loop, were replaced by aspartic acid and alanine. Modeling suggested that this analog would have a seatbelt comparable in length to that of the native ␤-subunit when the tensor disulfide is present. The hFSH and hTSH ␤-subunit constructs encode the natural amino acid sequences (not shown). The hFSH-KDEL construct encodes hFSH ␤-subunit residues 1-108 and the sequences of hCG ␤-subunit residues 115-145 and KDEL fused to its carboxyl terminus to give the sequence shown.
To determine whether a tensor cysteine might serve as a potential seatbelt latch site, we replaced one tensor cysteine with alanine and compared the abilities of B110 and B111 to detect the resulting ␤-subunits. Analogs containing only a single tensor cysteine recognized B110 much better than B111, showing that the seatbelt was not latched to ␤Cys 26 in a majority of these ␤-subunits. For example, only 25, 42, 14, and 24% of hCG-␤C93A, hCG␤-C93A-KDEL, hCG␤-C100A, and hCG␤-C100A-KDEL subunits appeared to have their seatbelts latched to ␤Cys 26 as reflected by differences in their abilities to be recognized by B110 and B111 (Table I, row 3 -6). The loss in B111 binding relative to that of B110 caused by elimination of one tensor cysteine appeared because of the competition of the remaining tensor cysteine with ␤Cys 26 for formation of the seatbelt latch disulfide, not to a change in folding of the remainder of the ␤-subunit. This is because changes in other parts of the ␤-subunit would have disrupted the binding of antibodies B101 and B110 to the ␤-subunit core. The finding that the seatbelt was not latched properly in most ␤-subunit molecules that contain only a single tensor cysteine (Table I, rows 3-6) is in marked contrast to the finding that the seatbelt was latched properly when both tensor cysteines were present or when both tensor cysteines were missing (Table I, rows 1 and 2). This suggests that the tensor cysteines have the potential to become latched transiently to the seatbelt before the seatbelt is latched and, as discussed next, supports the notion that the tensor disulfide is usually formed before the seatbelt is latched.
Residues ␤Cys 93 , ␤Cys 100 , and ␤Cys 110 are located near one another in the seatbelt. Therefore, before the seatbelt is latched, ␤Cys 110 is likely to be nearer tensor cysteines ␤Cys 93 and ␤Cys 100 than it is to ␤Cys 26 , the cysteine in loop ␤1 with which it will ultimately form the seatbelt latch disulfide. Consequently, ␤Cys 110 may compete with ␤Cys 93 and ␤Cys 100 for formation of the tensor disulfide. In analogs lacking both tensor cysteines, the seatbelt has no latch site other than ␤Cys 26 , 2 which would account for the finding that it appeared to be latched exclusively to this cysteine (Table I, data rows 1 and 2). In contrast, when one of the tensor disulfides is eliminated, ␤Cys 110 has two potential latch sites, i.e. ␤Cys 26 and either ␤Cys 93 or ␤Cys 100 . The finding that the seatbelt was latched to ␤Cys 26 in only a fraction of the ␤-subunit molecules that contained a single tensor cysteine suggests that ␤Cys 110 has little intrinsic tendency to be latched to ␤Cys 26 . Its proximity to ␤Cys 93 and ␤Cys 100 makes it more likely to form a stable disulfide with either of these tensor cysteines unless it is prevented from doing so.
What prevents the seatbelt from becoming latched stably to a tensor cysteine when both cysteines are present? The most likely explanation is that the tensor disulfide forms first or is more stable than either the ␤Cys 110 -␤Cys 93 and ␤Cys 110 -␤Cys 100 disulfides. Physical constraints on the positions of the latter disulfides cause them to remain near ␤Cys 100 or ␤Cys 93 , which would enable them to be disrupted by a disulfide exchange involving ␤Cys 100 or ␤Cys 93 , respectively. This would form the tensor disulfide and enable ␤Cys 110 to form the seatbelt latch disulfide with ␤Cys 26 . An exchange of this type cannot occur in ␤-subunit analogs that have only one tensor cysteine, which would account for the reduction in abilities of hCG␤-C100A and hCG␤-C93A to latch their seatbelts to ␤Cys 26 (Table I, data rows 3-6). The probability that ␤Cys 110 forms a disulfide with ␤Cys 26 before it forms a disulfide with ␤Cys 93 or ␤Cys 100 can be estimated from the fraction of hCG␤-C100A and hCG␤-C93A that is recognized by B111, respectively (Table I).
Whereas ␤Cys 110 would also be capable of disrupting a disulfide between ␤Cys 93 and ␤Cys 100 , it would be less likely to do so because its location at the end of the seatbelt does not constrain it to an area near the tensor disulfide. Thus, the findings that seatbelt latch disulfide formation is impaired by eliminating either ␤Cys 93 or ␤Cys 100 , but not by eliminating both cysteines, suggest that the tensor disulfide forms before the seatbelt is latched in most ␤-subunit molecules.
The finding that removal of one tensor cysteine prevented the hCG seatbelt from being latched properly in a majority of ␤-subunit molecules suggested that ␤Cys 110 at the end of the seatbelt is not constrained to a region near ␤Cys 26 . We tested the possibility that ␤Cys 110 scans the ␤-subunit to find its normal latch site by monitoring the abilities of cysteines added to the ␤-subunit to compete with ␤Cys 26 for formation of the seatbelt latch disulfide. The rationale for this is described in FIG. 2. Rationale for determining the influence of the tensor disulfide on formation of the seatbelt latch disulfide. A cysteine was introduced into the hCG ␤-subunit coding sequence either by changing the codons for either ␤Cys 93 or ␤Cys 100 to that for alanine (top panel) or by changing a codon for a residue in another part of the ␤-subunit to that for cysteine (bottom panel). Constructs were expressed individually in COS-7 cells and ␤-subunit analogs that were secreted or retained in the cells (KDEL constructs only) were measured in a sandwich immunoassay using B101 for capture and 125 I-B110 for detection relative to a purified hCG ␤-subunit standard. B110 binding provided an estimate of the total amount of ␤-subunit that had well folded ␤-core. It did not provide an indication of how the seatbelt was latched, however. Measurement of seatbelt latch formation was made using B101 for capture and 125 I-B111 for detection relative to the same hCG ␤-subunit standard. The ability of the ␤-subunit analogs to bind B111 relative to B110 provided an estimate of the fraction of the material in which the seatbelt is latched to ␤Cys 26 , its normal site.
the lower panel of Fig. 2. Several cysteines were found to compete with ␤Cys 26 for formation of the seatbelt latch site. These included those in place of ␤Leu 16 , ␤Ile 33 , ␤Ala 35 , ␤Phe 64 , ␤Ala 83 , ␤Ala 91 , and ␤Ser 96 (Table I). For example, the presence of a cysteine in place of ␤Leu 16 caused a 6-fold loss in the ability of the ␤-subunit to be recognized by B111 relative to B110, indicating that the seatbelt was latched to ␤Cys 26 in only one of six ␤-subunit molecules (i.e. 16% of the ␤-subunit). Similar results were found when the seatbelt had a choice between ␤Cys 26 and a cysteine in place of ␤Ile 33 . Some cysteine substitutions had relatively little influence on seatbelt latching. Thus, replacing ␤Arg 8 or ␤Asn 77 with cysteine was accompanied by a smaller reduction in the abilities of the ␤-subunit to be recognized by B111 (Table I). With the exception of the cysteine in place of ␤Arg 8 , most cysteines competed with ␤Cys 26 for latching the seatbelt. Consequently, addition of cysteine to most sites in loops ␤1 and ␤3 reduced the ability of the ␤-subunit to be recognized by B111. Cysteines that were closer to the origin of the seatbelt (i.e. ␤Ala 91 ) were usually more efficient competitors. 3 These observations support the notion that the hCG seatbelt does not have a strict propensity to be latched to ␤Cys 26 .
These studies supported the notion that ␤Cys 110 "scans" the surface of the ␤-subunit by a constrained random walk to find ␤Cys 26 , its natural cysteine "partner." By shortening the length of the seatbelt, formation of the tensor disulfide would restrict the surface of the ␤-subunit that can be scanned, making it easier for ␤Cys 110 to find ␤Cys 26 . To test this notion, we repeated many of these studies with an hCG ␤-subunit analog that was unable to form the tensor disulfide. As can be seen (Table I), the ability to form the tensor disulfide influenced the abilities of cysteines to compete with ␤Cys 26 . As a result, when the tensor disulfide was removed, there was a significant reduction in B111 binding. This indicated that the seatbelt had a greater tendency to be latched to the cysteine added to the ␤-subunit and a lower tendency to be latched to ␤Cys 26 . In only one case (i.e. hCG␤-A83C) was the abilities of the seatbelt to be latched to ␤Cys 26 increased when the tensor disulfide was removed. These observations are also consistent with the notion that the tensor disulfide is formed before the seatbelt is latched in most hCG ␤-subunit molecules.
The Tensor Loop Disulfide Appears to Be Disrupted Transiently during Assembly of Heterodimers by the Threading Pathway-Pulse-chase analyses (4) and the data just presented indicate that the tensor loop is formed before the seatbelt is latched in most hCG ␤-subunit molecules. Because the hCG seatbelt is usually latched prior to heterodimer assembly (22), we expect that the tensor disulfide would also be present prior to heterodimer assembly, at least transiently. Studies using purified hCG ␣and ␤-subunits showed that disruption of the tensor disulfide facilitated assembly in vitro (7) and we considered the possibility that the tensor disulfide might also be disrupted during assembly in the endoplasmic reticulum. A clue to the manner in which we might identify free tensor cysteines during threading came from the observation that a fraction of some heterodimers having an additional cysteine in their ␣-subunits were acid stable (22). For this to occur, the cysteine in the ␣-subunit would need to be located near a free cysteine in the ␤-subunit.
As discussed elsewhere (22), two of the 12 ␤-subunit cysteines are more likely than the other 10 to participate in the TABLE I Competition of ␤-subunit cysteines with ␤Cys 26 for seatbelt residue ␤Cys 110 and influence of the tensor disulfide The total ␤-subunit was measured in sandwich assays employing antibodies B101 for capture (loop 2) and 125 I-B110 (loops 1 and 3) or 125 I-B112 (loop 3) for detection. Latching the seatbelt to ␤Cys 26 was determined using B101 for capture and 125 I-B111 for detection. Purified hCG␤ that had been isolated from the heterodimer was used as the standard in all assays. Values shown represent the ratio of material determined in assays employing B111 relative to that measured in assays employing B110 or B112. The distance values were measured between the C␣ carbons from the start of the seatbelt ␤Cys 90 to the free cysteine. The endoplasmic reticulum-trapped analogs contained the KDEL sequence at their carboxyl termini. Values in the column denoted "Cys 93 and Cys 100 " were for analogs that contain both tensor cysteines and that are expected to contain the tenor loop except as noted. Those in the column denoted "C93A and C100A" were for analogs in which both tensor cysteines had been replaced by alanine and that are unable to form the disulfide that stabilizes the tensor loop. All experimental values are means of triplicates Ϯ S.E. Data  intersubunit cross-link with cysteines added to the ␣-subunit. The finding that cross-linked heterodimers containing ␣-subunit analogs having an unpaired cysteine were recognized by monoclonal antibodies B110 and B111 showed that the cystine knot is intact, the seatbelt is latched properly, and the disulfide at the tips of loops ␤1 and ␤3 had not been disrupted (22). Thus, we considered it more likely that the intersubunit disulfide cross-link involved cysteines that form the tensor disulfide, the most readily reduced disulfide in the ␤-subunit (7), than any other ␤-subunit cysteines. Furthermore, modeling supported the notion that the locations of the tensor cysteines would be more likely to enable them to form disulfides with cysteines that had been added to parts of the ␣-subunit that passed beneath the seatbelt during threading (not shown). Our earlier studies suggested that one of the tensor cysteines can be disulfide bridged to the ␣-subunit (22), a finding that supports the notion that the tensor disulfide was disrupted during threading and we investigated this possibility further. Analogs of the ␤-subunit that lacked one or both tensor cysteines, i.e. hCG␤-C93A, hCG␤-C100A, and hCG␤-C93A,C100A, were not incorporated into heterodimers containing the native ␣-subunit (Table II, data row 1). This confirmed earlier observations (11) and showed that the presence of a functional tensor loop was essential for heterodimer stability. In contrast, hCG␤-C93A and hCG␤-C100A, but not hCG␤-C93A,C100A, formed heterodimers with several ␣-subunit analogs that contained an additional cysteine (Table II, data columns 1, 2, and 4). Heterodimers that contained hCG␤-C93A are likely to contain an intersubunit disulfide between the ␣-subunit and ␤Cys 100 (Table  II, data column 1). The cysteine substitution in ␣2 that gave rise to the most cross-linked heterodimer, i.e. ␣-S43C, is nearest ␤Cys 100 (1,2). Several other cysteines that had been substituted for residues in ␣2 were able to participate in intersubunit crosslinks with hCG␤-C93A, indicating that they can also be bridged to ␤Cys 100 . Cysteines that had been added to the carboxyl-terminal portions of the ␣-subunit also became cross-linked to this ␤-subunit analog, including those in place of ␣Tyr 88 and ␣Ser 92 .
Cysteines that had been added to the ␣-subunit were also capable of being cross-linked to ␤Cys 93 , but this process appeared to be less efficient than formation of a cross-link with ␤Cys 100 . As a result, significantly less heterodimer formed when hCG␤-C100A was co-expressed with the ␣-subunit analogs (Table II, data column 2). This is consistent with the crystal structure, which shows that most of the cysteines added to the ␣-subunit are closer to ␤Cys 100 than they are to ␤Cys 93 . It may also suggest that ␣2 passes nearer ␤Cys 100 during threading. A notable exception to this generalization was the cysteine in ␣-N52C. ␣Asn 52 appears to be nearer ␤Cys 93 , a factor that may contribute to the ability of ␣-N52C to be crosslinked to this tensor cysteine.
We have found that the seatbelt can be latched to a cysteine added to the ␣-subunit (5). Whereas it might be anticipated that heterodimers containing an additional ␣-subunit cysteine are also stabilized by latching their seatbelts to the ␣-subunit, this appears unlikely. Formation of a disulfide between ␤Cys 110 and the cysteine added to the ␣-subunit was observed only when ␤Cys 26 , the normal seatbelt latch site, had been replaced (5). Furthermore, we found that the seatbelts of analogs containing hCG␤-C93A were latched to ␤Cys 26 because these acid-stable heterodimers were recognized by B111 (Table  II). Presumably these analogs contain a cross-link between ␤Cys 100 and the ␣-subunit. Some of the heterodimers containing hCG␤-C100A that we presumed to have a cross-link between ␤Cys 93 and the ␣-subunit were not recognized by B111, however. Modeling suggested that it would be more difficult to form this cross-link without affecting the conformation of the heterodimer, a phenomenon that may have disrupted B111 binding. The ability of B111 to bind analogs containing hCG␤-C93A was about half of that expected relative to its ability to bind hCG. This suggested that the conformation of the seatbelt had also been altered somewhat by this cross-link.
To learn if a tensor cysteine is required for formation of a  Table I. Coordinates for the C␣ atoms of the ␤-subunit derived from the crystal structure of hCG (1) were used to make the wire diagrams shown here in black. The C␣ atoms of the residues discussed in the text are shown space-filled and numbered. The relative binding sites of the antibodies used for measurement are also indicated. Note that the tensor disulfide is formed by residues Cys 93 -Cys 100 and the seatbelt latch disulfide is normally formed by residues Cys 26 -Cys 110 . The seatbelt extends from residues Ala 91 to Cys 110 . It is unlikely that the positions of ␤-subunit loop 2 (top of the figure) or the seatbelt occupy the same positions in the free ␤-subunit as they do in the heterodimer. Because the structure of the uncombined ␤-subunit is not known, this figure has been derived from the heterodimer.
cross-linked heterodimer, we co-expressed several ␣-subunit analogs with hCG␤-C93A,C100A, an hCG ␤-subunit analog that lacks both tensor cysteines. The trace amounts of heterodimers produced in these experiments were too small to detect.
We anticipated that formation of an intersubunit disulfide between the ␣-subunit and a tensor cysteine would require that cysteines added to the ␣-subunit be near at least one tensor cysteine. This would permit formation of a disulfide without distorting the conformation of the heterodimer so severely that it would prevent its recognition by A113 and B110. As can be seen by reference to Fig. 4, only those cysteines added to the ␣-subunit that are in the vicinity of the tensor cysteines were capable of forming a cross-linked heterodimer. None of the heterodimers that contained ␣-subunit analogs ␣-Q5C and ␣-M71C) were acid stable (Table III, data rows 2 and 14), most likely because the cysteines in these ␣-subunits are distant from the tensor cysteines (Fig. 4). Furthermore, being near the tensor cysteines was not sufficient to form a cross-link. Thus, whereas a cysteine at the COOH terminus of the ␣-subunit, ␣S92C, can be cross-linked efficiently to hCG␤-C93A and hCG␤-C100A (Table II, data row 20), it became cross-linked poorly to hCG␤ (Table III, data row 16). This showed that formation of the cross-link occurred only while the cysteine in the ␣-subunit was most likely to be constrained near a free tensor cysteine or while loop ␣2 being threaded beneath the seatbelt.
We presumed that the disulfide formed between a tensor cysteine and a cysteine added to the ␣-subunit could have formed only during assembly that occurs by a threading mechanism, not during assembly that occurs by a wraparound mechanism. To test this prediction, we co-expressed ␣-L41C or ␣-S43C with the native hCG ␤-subunit and with hCG␤-C26A,C110A, an analog that lacks the seatbelt latch site and the cysteine at the end of the seatbelt. This analog has both tensor cysteines and would be expected to be cross-linked to either ␣-L41C or ␣-S43C if a disulfide can form between a tensor cysteine and the ␣-subunit while the seatbelt is unlatched. Consistent with earlier observations, co-expression of ␣-L41C and ␣-S43C with hCG␤ led to a small amount of acidstable heterodimer that was readily recognized by B111 (Table   IV, rows 1 and 3). We did not detect formation of any heterodimer when ␣-L41C or ␣-S43C were co-expressed with hCG␤-C26A,C110A, however (Table IV, rows 2 and 4). Thus, formation of a cross-link between a tensor cysteine and a cysteine in loop ␣2 occurs only when the seatbelt is latched, most likely while the tensor disulfide is disrupted as loop ␣2 is being threaded beneath the seatbelt.
Only a small fraction of most heterodimers that contained an additional ␣-subunit cysteine was stable at acid pH (Table III). Because the remaining fraction lacked an intersubunit disulfide, it might have been formed by a threading process that occurred while the tensor disulfide remained intact. Indeed, high concentrations of the glycoprotein hormone subunits have long been known to recombine in vitro in oxidizing media (6), conditions in which the tensor disulfide and the seatbelt latch disulfide remain intact (7). Efforts to test the possibility that threading can occur while the tensor disulfide is intact led us to study the formation of heterodimers containing ␤-subunit constructs that lacked the tensor loop. These were prepared by replacing the tensor loop (i.e. Cys-Arg-Arg-Ser-Thr-Thr-Asp-Cys) with either Asp or with Asp and Ala. The latter substitution approximates the length of the tensor disulfide. This is seen by comparing the distance between the C␣ carbons in the tensor disulfide, i.e. 6.2 Å, with that between the C␣ atoms of a typical dipeptide, which can range from 5.3 to 6.9 Å depending on whether it is in a helix or a sheet. Molecular modeling indicated that neither of these changes would disrupt the heterodimer and that the seatbelt of the latter analog would be essentially the same length as that of the seatbelt after the tensor disulfide had formed. COS-7 cells that were co-transfected with the native ␣-subunit and hCG␤-␦93:100D, hCG␤-␦93:100DA, or the KDEL derivatives of these analogs failed to produce heterodimers (Table V, data for KDEL derivatives not shown). Both analogs were capable of docking with the ␣-subunit as shown by their ability to disrupt hCG assembly (Table V). These findings suggested that the tensor loop is essential for assembly, most likely because it enables the seatbelt to be elongated, thereby increasing the size of the hole beneath the seatbelt.
We continued to be puzzled by the observation that only a TABLE II Influence of tensor cysteines on formation of an intersubunit cross-link to a cysteine added to the ␣-subunit COS-7 cells were transfected with the indicated analog and ␤-subunit constructs. These ␤-subunits are unable to form the tensor disulfide and did not combine stably with the ␣-subunit unless it contained a cysteine at an appropriate location. The free tensor cysteine in hCG␤-C93A is ␤Cys 100 . That in hCG␤-C100A is ␤93. Some acid stable heterodimers containing hCG␤-C93A were tested for their abilities to be recognized by B111. Their recognition by B111 showed their seatbelts were latched to ␤Cys 26 , not the ␣-subunit. These data suggest either tensor cysteine can be cross-linked to the ␣-subunit, depending on the location of the added cysteine. fraction of the heterodimers that contain an additional ␣-subunit cysteine were cross-linked and considered three mechanisms that would explain the formation of the intersubunit disulfide cross-link. One suggested that the disulfide formed during threading but that this was a relatively inefficient process because of the transient nature of threading. Alternatively, the intersubunit disulfide might form readily but then be disrupted as the result of a disulfide exchange caused by an attack of the other tensor cysteine that resulted in formation of the tensor disulfide. In this model, the small amount of crosslinked heterodimer that remains is a kinetically trapped reaction intermediate. The final explanation we considered was that the intersubunit disulfide formed after the heterodimer had been assembled by a disulfide exchange and was created by an attack of the ␣-subunit cysteine on the tensor disulfide. The first two explanations, which we could not distinguish experi-mentally, imply that the tensor disulfide is more stable than the intersubunit disulfide. The third explanation suggests that the intersubunit disulfide is more stable than the tensor disulfide. We reasoned that if the tensor disulfide were more stable than the intersubunit disulfide, treatment of the cross-linked heterodimers with low concentrations of reducing agent would promote a disulfide exchange leading to disruption of the intersubunit disulfide and formation of the tensor disulfide. If this occurred, the heterodimer would remain intact but no longer be acid stable. Treatment of the acid-stable fraction of several hCG analogs with 1 mM ␤-mercaptoethanol rendered them acid unstable without causing them to dissociate (Table  VI). This showed that the intersubunit disulfide is the least stable disulfide in the heterodimer. Thus, it must have formed during assembly while the tensor disulfide had been disrupted, not by an attack of the cysteine in the ␣-subunit on the tensor FIG. 4. Models depicting the positions of residues that appear to form an intersubunit disulfide with an hCG ␤-subunit tensor cysteine relative to those that do not. Co-expression of hCG ␤-subunit with some ␣-subunit analogs that contain an additional cysteine leads to the production of acid-stable heterodimers that are capable of binding B111 (Table III). This shows that the seatbelts of these analogs are latched to ␤Cys 26 in loop ␤1, similar to that in hCG. The acid stabilities and B111 recognition properties of these heterodimers are similar to those of heterodimers containing ␤-subunit analogs missing one tensor cysteine and ␣-subunit analogs having an additional cysteine. Residues of the ␣-subunit that appear to form an intersubunit disulfide cross-link when expressed with the native hCG ␤-subunit are shown in white with black lettering. Not all these are labeled because of their proximity to one another. The position of the cysteine in ␣-S92C, which does not form an intersubunit disulfide cross-link with hCG␤ but can form an intersubunit cross-link when expressed with an hCG ␤-subunit missing one tensor cysteine is shown in dark gray with white lettering. Some ␣-subunit analogs, such as ␣-Q5C and ␣-M71C, do not form an intersubunit disulfide with the hCG ␤-subunit during heterodimer assembly in the ER. These are shown in dark gray with white lettering and encircled with a broken white line. Note that all of these three residues do not pass under the seatbelt during threading. Residue 88 does not pass under the seatbelt either, but is constrained to be near the tensor cysteines after threading. disulfide after assembly had been completed.
A fraction of the heterodimer produced when hFSH and hTSH ␤-subunits were co-expressed with some ␣-subunit analogs containing an additional cysteine was also stable at acid pH (Table VII). This suggested that similar to hCG, analogs of hFSH and hTSH that have an additional cysteine in their ␣-subunits can become cross-linked by an intersubunit disulfide. As in the case of the hCG analogs just discussed, we observed good cross-linking with ␣-S43C. This is most likely because of its proximity to hFSH␤-Cys 94 or hTSH␤-Cys 95 , which correspond to hCG␤-Cys 100 . Substitution of cysteines for residues ␣Thr 46 , ␣Met 47 , ␣Leu 48 , and ␣Val 50 also resulted in the formation of intersubunit disulfides with all three glycoprotein hormone ␤-subunits. These residues are also located in loop ␣2 and would pass beneath the seatbelt during threading. We noted some differences in the residues that appeared to be cross-linked in hCG, hFSH, and hTSH, however. For example, we did not detect the formation of any cross-link containing ␣-K45C in hFSH and hTSH (Table VII, legend), an analog of the ␣-subunit that led to the formation of small amount of cross-linked heterodimers containing hCG␤-KDEL (Tables III  and VII). We also failed to detect the formation of hFSH or hTSH analogs that contained a cross-link with ␣-Y88C. This may reflect subtle differences in the pathways taken by ␣2 during threading of each type of heterodimer. It may also reflect differences in the compositions of the ␤-subunits, notably their seatbelts, which is the portion of the ␤-subunit primarily responsible for its influence on biological activity (9,(12)(13)(14). We were unable to characterize the hFSH and hTSH analogs beyond their acid stability because of the lack of an antibody that recognized the seatbelt latch disulfide in these hormone analogs and did not study them further.
Heterodimer Assembly Appears to Be Driven by Differences in the Redox Potential of the Tensor Disulfide in the ␤-Subunit and the Heterodimer-Based on the known mobility of loop ␣2 (15), we envision that threading involves motions of this loop and TABLE III Cross-linked hCG analogs detected in the ER hCG␤-KDEL and the indicated ␣-subunit analogs were co-transfected into COS-7 cells and quantified in A113/ 125 I-B110 assays. Note, the analog containing M71C was quantified in A113/ 125 I-B112 assays. The fraction of cross-linked heterodimer isolated from the cells was identified by its acid stability. Heterodimer remaining after acid treatment was also detected in A113/ 125 I-B111 assays (column 3). The ability of B111 to bind this material was not quite as good as its ability to bind hCG, a phenomenon that may have reflected the influence of the cross-link on the conformation of the seatbelt. No heterodimer formed when these ␣-subunits were co-transfected with hCG␤-C93A,C100A-KDEL (Table II), showing that the cross-link appears to involve a tensor cysteine. The ability of B111 to bind these acid-stable analogs contrasts its inability to bind acid stable analogs in which the seatbelt is latched to the ␣-subunit. This showed that the acid stability of these analogs is not due to the cross-linking of ␤Cys 110 to the ␣ -subunit.  IV Acid stable heterodimers containing the hCG ␤-subunit and ␣-subunit analogs having unpaired cysteines did not form unless the seatbelt was latched to ␤Cys 26 Constructs encoding the indicated ␣and ␤-subunits were transfected into COS-7 cells and heterodimer secreted into the medium was monitored by sandwich immunoassays employing ␣-subunit antibody A113 for detection and radioiodinated ␤-subunit monoclonal antibodies B110 or B111 for detection as described (22). The latter detects formation of a seatbelt latch between ␤Cys 26 and ␤Cys 110 .  V Influence of the tensor loop on heterodimer formation COS-7 cells were transfected with constructs encoding the native human ␣-subunit and hCG␤ or the indicated hCG ␤-subunit analogs. Heterodimer formation was monitored in A113/ 125 I-B110 sandwich immunoassays. Analogs hCG␤-␦(93:100)D and hCG␤-␦(93:100)DA, which lack the tensor loop were detected in B111 assays indicating that their seatbelts are latched to ␤Cys 26 (not shown). The seatbelt of hCG␤-C93A,C100A is also known to be latched to ␤Cys 26 based on its ability to be recognized by B111 (Table I).
To learn if the tensor disulfide is more stable in the heterodimer than in the free ␤-subunit, we treated equimolar amounts of hCG and hCG␤ with 0 -2 mM BME for 15 min at 37°C and then blocked the resulting free thiols with an excess of iodoacetate (IA). Consistent with the finding that treatment of hCG␤ with low concentrations of BME disrupted only its tensor disulfide (7), BME/IA treatment blocked the ability of hCG␤ to combine with the ␣-subunit (Fig. 5a), but did not alter its recognition by conformation-dependent antibodies B101, B111, or B112 (Fig. 5b). The abilities of these antibodies to recognize the ␤-subunit showed that mild reduction and alkylation did not affect the subunit core or the seatbelt latch disulfide. The finding that BME/IA treatment altered the ability of the ␤-subunit to be incorporated into heterodimers is consistent with the known sensitivity of the tensor disulfide to reduction (7). In contrast, similar BME/IA treatment did not influence the stability of hCG (Fig. 5a) or its ability to be recognized by these antibodies (Fig. 5b). If these concentrations of BME had disrupted the tensor disulfide in the heterodimer, they would have rendered ␤Cys 93 and/or ␤Cys 100 capable of being alkylated and thereby promoted heterodimer dissociation. The finding that the free ␤-subunit was rendered incapable of being incorporated into the heterodimer by concentrations of BME/IA that had no influence on the stability of the heterodimer shows that the tensor disulfide is more stable in the heterodimer than in the free ␤-subunit. The increased stability of the tensor disulfide to reducing agents shows how assembly of the heterodimer can be driven by the redox potential of the ER. Disruption of the tensor loop would permit threading during assembly. The increased stability of the tensor disulfide following the completion of threading and heterodimer formation (Fig. 5) would prevent loop ␣2 and its attached oligosaccharide from passing beneath the seatbelt, a process needed to reverse assembly.

The Tensor Disulfide Has Multiple Roles in Heterodimer
Assembly, the First of Which Is to Facilitate Formation of the Seatbelt Latch Disulfide-Before it is latched, the end of the seatbelt appears to be a highly mobile portion of the hCG ␤-subunit. This explains the ability of hCG␤-C26A, an analog lacking the normal ␤1 seatbelt latch site, to latch its seatbelt to cysteines introduced into the ␣-subunit (5) or into other parts of the ␤-subunit (22). It also accounts for the abilities of cysteines that are introduced into the ␤-subunit to compete with ␤Cys 26 as a seatbelt latch site before the subunits dock (Table I). Except for cysteines added to some portions of the aminoterminal end of the ␤-subunit, those that are within the distance capable of being reached by ␤Cys 110 can serve as seatbelt latch sites. Their abilities to compete with ␤Cys 26 as a latch site appears to vary inversely with their distance from ␤Cys 90 , the residue that tethers the amino-terminal end of the seatbelt. This suggests that the seatbelt latch disulfide forms after a constrained random walk over much of the ␤-subunit surface, not a predetermined fold that puts it adjacent to ␤Cys 26 .
The findings that the seatbelt is mobile and that it can become latched to the tensor cysteines creates potential problems for latching the seatbelt to ␤Cys 26 . As shown (Table I), either tensor cysteine (i.e. ␤Cys 93 or ␤Cys 100 ) can compete with ␤Cys 26 for latching the seatbelt. Indeed, the proximity of the tensor cysteines to ␤Cys 110 makes it likely that the seatbelt would become latched to either ␤Cys 93 or ␤Cys 100 before it becomes latched to ␤Cys 26 . Indeed, the apparent ability of hCG␤-C93A,C100A to be recognized better than hCG␤ in B111 assays than in B110 assays (Table I) might reflect the possibility that the seatbelt becomes latched to either ␤Cys 93 or ␤Cys 100 in some hCG␤ molecules expressed in cultured cells. Nonetheless, we cannot exclude the possibility that differences in B111 recognition occur because of differences in the conformations of the seatbelt caused by the presence and absence of the tensor loop. The problem caused by improper latching of the ␤-subunit would be largely avoided if the tensor disulfide forms rapidly, thereby preventing ␤Cys 93 and ␤Cys 100 from serving as potential seatbelt latch sites. Furthermore, the proximity of the tensor cysteines would be expected to facilitate a disulfide exchange that disrupts an inappropriate disulfide between the other tensor cysteine and seatbelt residue ␤Cys 110 . This would result in formation of the tensor disulfide before the seatbelt latch disulfide.
In addition to eliminating the potential competition between the tensor cysteines and the seatbelt, formation of the tensor disulfide would shorten the seatbelt. This would reduce the area of the ␤-subunit that can be scanned by the end of the seatbelt before it becomes located in the vicinity of ␤Cys 26 and, The acid-stable fraction of heterodimers, which were presumed to contain a disulfide cross-link between a tensor cysteine and a cysteine added to ␣2, were treated with 1 mM BME. The resulting material was tested for its acid stability and its ability to be recognized by B111 in A113/B111 sandwich immunoassays. These data show that mild reduction disrupted the intersubunit cross-link and rendered the heterodimer acid unstable. We suggest this is due to the disruption of the intersubunit disulfide caused by formation of the tensor disulfide. In addition to the analogs described here, we tested several other analogs with these ␤-subunit analogs, none of which contained a significant amount of cross-link. These included: ␣-R35C, ␣-R42C, ␣-K45C, ␣-Q50C, ␣-Y88C, ␣-H90C, and ␣-S92C. as a consequence, it would facilitate latch formation, as was found (Table I). Considered together these findings suggest that the tensor disulfide forms before the seatbelt is latched, at least transiently. Disruption and Reformation of the Tensor Disulfide Have Key Roles in Heterodimer Assembly-Most human glycoprotein hormone heterodimers are assembled by a process in which a part of the ␣-subunit and its attached oligosaccharide are threaded beneath the seatbelt through a hole in the ␤-subunit (22). As shown here, assembly of hCG in the endoplasmic reticulum is assisted by disruption and reformation of the tensor disulfide. Disruption of the tensor disulfide elongates the seatbelt, which is expected to facilitate passage of loop ␣2 and its associated oligosaccharide beneath the seatbelt during threading. Reformation of the tensor disulfide following threading shortens the seatbelt and would be expected to retard heterodimer dissociation by hindering the glycosylated end of loop ␣2 from passing through the hole in the ␤-subunit.
The role of the tensor disulfide in assembly was first detected during studies to uncover the mechanism by which reducing agents potentiate hCG assembly in vitro, a process found to occur exclusively by threading (7). The tensor disulfide was reduced more readily than any other disulfide in the hCG ␣or ␤-subunits and was essentially the only disulfide disrupted by concentrations of reducing agents that facilitated assembly optimally (7). The finding that the tensor disulfide is more stable in the heterodimer than in the free ␤-subunit (Fig. 5) showed that concentrations of reducing agents sufficient to disrupt the tensor disulfide before assembly is initiated would not prevent reformation of the tensor disulfide once assembly is completed. This explains the ability of a mild reducing environment to potentiate hCG assembly in vitro and in the endoplasmic reticulum. Thus, the tensor disulfide can be viewed as a redox-sensitive switch that opens before the subunits have combined and closes afterward to secure the heterodimer. By itself, disruption and reformation of the tensor disulfide would not be sufficient to drive assembly of the heterodimer, however. As described elsewhere (24), contacts between the amino-terminal portions of the hCG subunits and between ␣-subunit loops 1 and 3 with a portion of ␤-subunit loop 2 appear to have key roles in subunit docking. Based on the crystal structures of hCG and hFSH, which reveal several hydrogen bond contacts between the backbones of loop ␣2 and portions of the ␤-subunit cystine knot, we suggest that formation of these hydrogen bonds drives the migration of loop ␣2 under the seatbelt to create an unstable heterodimer (Fig. 6). Contacts between the tensor loop and loop ␣2 appear to stabilize the tensor disulfide, which then stabilizes the heterodimer.
We tested the notion that hCG assembly in the endoplasmic reticulum requires transient disruption of the tensor disulfide using ␤-subunit analogs in which the small loop in the seatbelt was replaced with either aspartic acid or aspartic acid and alanine. Molecular modeling showed that the length of the seatbelt in these analogs would be similar to that in hCG when the tensor disulfide is formed. In principle, because the hCG subunits can combine while the tensor disulfide remains intact (7), ␤-subunits having a dipeptide in place of the small seatbelt loop would be expected to combine with the ␣-subunit during assembly in cells unless this process requires disruption of the tensor disulfide. Neither of the ␤-subunits that contained an Asp or Asp-Ala in place of the tensor loop were incorporated into heterodimers even though each appeared to latch their seatbelts 4 and to dock with the ␣-subunit (Table V). This suggested that increasing the length of the seatbelt by disrupting the tensor disulfide, a phenomenon that enhanced assembly in vitro (7), may be crucial for heterodimer assembly in the endoplasmic reticulum. 4 hCG␤-␦93:100DA, an hCG ␤-subunit analog in which the tensor loop is replaced by aspartic acid and alanine, was readily secreted from cells and detected in assays employing B101 for capture and either 125 I-B110 or 125 I-B111 for detection. This indicated that except for the absence of the tensor loop, its structure was similar to that of the hCG ␤-subunit and its seatbelt had been latched normally. We were unable to detect any heterodimer when either hCG␤-␦93:100DA or an analog having the KDEL endoplasmic reticulum retention signal were coexpressed with the ␣-subunit.
FIG. 5. Influence of BME and IA treatment on hCG and the free ␤-subunit. Equimolar amounts of hCG and free ␤-subunit (10 Ϫ10 mol/5 l, 2 ϫ 10 Ϫ7 M) were treated with 0, 0.17, 0.5, and 2.0 mM ␤-mercaptoethanol (15 min, 37°C). The reaction was terminated by the addition of iodoacetate (final concentration, 10 mM) and aliquots were taken to determine the amount of hCG that had dissociated and the ability of the ␤-subunit to combine with the ␣-subunit. These low concentrations of BME did not promote subunit dissociation, measured in A113/B111, A113/B112, B101/B111, and/or B101/B112 sandwich assays (not shown). The higher concentrations of BME followed by IA treatment blocked the ability of the free ␤-subunit to combine with the ␣-subunit, p Ͻ 0.003 (A), but had no influence on the seatbelt latch disulfide of hCG or the free ␤-subunit detected as the ratio of B111/B112 binding (B). The amount of hCG recovered during the recombination study was 78% of the theoretical limit.
The transient nature of the threading process precluded its direct measurement. The notion that the tensor disulfide is disrupted during threading of hCG, hFSH, and hTSH is supported by the observation that one of the tensor cysteines becomes disulfide bridged to the ␣-subunit during the assembly of heterodimers that contain an unpaired ␣-subunit cysteine. Several observations support the notion that the tensor disulfide is involved in this cross-link. First, either tensor cysteine can become cross-linked to cysteines that have been substituted for several different ␣-subunit residues (Table II). This shows that the cross-link is feasible. Second, at least one tensor cysteine was required to form the cross-link. Replacing both tensor cysteines with alanine prevented the cross-link from forming. Third, only those cysteines in the ␣-subunit that are near a tensor cysteine were found to participate in the intersubunit cross-link. None of the analogs that contained a cysteine at a more remote site became cross-linked. Fourth, formation of this cross-link was detected only when the ␣-subunit was co-transfected with ␤-subunits that were capable of latching their seatbelts. And finally, all the cross-linked analogs tested were detected by B111, a phenomenon that would not have been detected if seatbelt residue ␤Cys 110 were latched to any other cysteine in the molecule. This showed that the end of the seatbelt was latched to ␤Cys 26 , not to the cysteine in the ␣-subunit.
We considered the possibility that the intersubunit crosslink was formed after the heterodimer had been assembled rather than during the process of threading. The finding that the intersubunit disulfide could be readily disrupted by mild reduction excluded this possibility (Table VI). These observations revealed that the cross-link is less stable than the tensor disulfide and that the cross-linked analogs are likely to be unstable intermediates that became trapped kinetically during threading. As a result, there would be no tendency to form a stable intersubunit disulfide after the tensor disulfide had been formed even in the reducing environment in the endoplasmic reticulum. The notion that cross-linked heterodimers are unstable intermediates is also consistent with the finding that only a fraction of the total heterodimers contained a cross-link.
Differences in the Redox Potential of the Tensor Disulfide in the Free ␤-Subunit and the Heterodimer Appear to Drive Gly-coprotein Hormone Assembly by the Threading Pathway-The tensor disulfide is less stable in the free subunit than in the heterodimer (Fig. 5) and its disruption would facilitate threading of ␣2 by increasing the space that is available for passage of the glycosylated end of ␣2. Indeed, the latter may have the greatest requirement for space because the rate of assembly in the absence of this oligosaccharide exceeds that in its presence, a phenomenon that can be used to prepare hormone analogs lacking this oligosaccharide (16). Completion of assembly, a phenomenon that stabilizes the tensor disulfide, would impede passage of ␣2 beneath the seatbelt and contribute to heterodimer stability. This would explain the discrepancies noted in the kinetics of subunit association and heterodimer dissociation (17), the acceleration of hCG assembly by reducing agents (7), and the influence of protein-disulfide isomerase on assembly (8). Changes in the size and composition of the tensor loop, which would be expected to affect its formation, have been found to adversely affect heterodimer assembly by mammalian cells (18).
Why is the stability of the tensor disulfide in the heterodimer greater than that in the free ␤-subunit? The structure of the seatbelt in the free ␤-subunit has not been determined. Because this region of the seatbelt is not recognized by heterodimer-specific antibodies to epitopes that include portions of the NH 2 -terminal end of the seatbelt (19), it is likely to have a different structure in the free ␤-subunit than in the heterodimer. In hCG, the backbone atoms of ␣-subunit residues ␣Val 53 ,Ser 55 ,Ser 57 form hydrogen bonds with hCG␤-Asp 99 ,Gly 101 and possibly hCG␤-Thr 97 (1, 2), a phenomenon that would constrain tensor cysteine hCG␤-Cys 100 to a region nearby hCG␤-Cys 93 (Fig. 6). In hFSH, these ␣-subunit residues form hydrogen bonds with hFSH␤-Ser 91 ,Asp 93 ,Thr 95 (3), thereby constraining tensor cysteine hFSH␤-Cys 94 nearby hFSH␤-Cys 87 . Consequently, the reducing environment of the endoplasmic reticulum is less likely to disrupt a disulfide between the tensor cysteines in the heterodimer than in the free ␤-subunit. The ␣-subunit residues that participate in this network are held in a ␤-sheet with residues in ␣2 that are in contact with the ␤-subunit cystine knot (Fig. 6). Thus, assembly of the heterodimer stabilizes the position of the tensor loop relative to the ␤-subunit cystine knot, something that is un-FIG. 6. Hydrogen bonds in hCG that are expected to drive threading and to stabilize the ␤93-␤100 tensor disulfide. The orientation of the heterodimer (left panel) is positioned to illustrate hydrogen bonds that stabilize a ␤-sheet structure formed by residues in both subunits (1). Hydrogen bonds that stabilize other portions of the heterodimer have been omitted for clarity. Color code: white, ␣-subunit; dark gray, ␤-subunit; thin white lines (left panel) or thick broken lines (right panel), hydrogen bonds. A similar arrangement is seen in hFSH (3). The corresponding hFSH ␤-subunit residues are: ␤Trp 27 ,␤Cys 28 ,␤Ala 29 ,␤Gly 30 ,␤Tyr 31 and ␤Thr 92 ,␤Asp 93 ,␤Cys 94 ,␤Thr 95 . likely to occur in the free ␤-subunit.
Antibody Tools Are Useful for Structural Analyses-The studies described here and in the companion manuscripts (22)(23)(24) depended on the use of monoclonal antibodies to conformation-sensitive epitopes to evaluate the structures of various folding intermediates (20,21). The most important of these was B111, the antibody that can detect an epitope that is formed when the hCG seatbelt is latched normally. This permitted studies in which various cysteines were allowed to compete with ␤Cys 26 and to determine when this disulfide was latched. Whereas it would have been preferable to use high resolution techniques such as crystallography or nuclear magnetic resonance spectroscopy to identify these intermediates, these techniques do not have the sensitivity required for the analysis of nanogram quantities of materials that can be produced readily. Even discounting the challenges of determining the structures of these intermediates by NMR and crystallography, it would have been cost prohibitive to make the larger amounts of materials required.
hCG contains a total of 11 disulfide bonds, making it possible that the introduction of cysteines into either subunit might disrupt one or more of these disulfides and alter its structure. With the exception of the seatbelt latch site and the apparent formation of disulfides between a tensor cysteine and the ␣-subunit observed during threading, we did not detect any signs that the cysteines we introduced or removed altered the structures of either subunit despite the fact that we analyzed a large number of cysteine containing constructs. Indeed, we sought to test the robustness of our approach by creating and testing a large panel of cysteine containing analogs. All of the observations that we made are internally consistent, a phenomenon that would be unlikely if some cysteine mutations had disrupted the structure of the hormone.
Our dependence on antibodies for these studies raises the possibility that we missed important assembly intermediates that are not recognized by any antibodies in our panel. For example, because all the antibodies used in these studies are conformation dependent, we would not observe heterodimer assembly that occurs before formation of the subunit cores, a phenomenon that depends on formation of their cystine knots. Whereas we cannot exclude the possibility that some assembly occurs by this route, the fact that we can account for most, if not all the heterodimer that is formed, makes it unlikely. As noted earlier (22), the difficulty of distinguishing dead end folding intermediates is one reason that we chose not to use pulsechase methods for these studies. Finally, we attempted to exclude the possibility that our observations would be affected by changes to the protein that occur during secretion using analogs that are preferentially retained within the cell. We observed similar phenomena using analogs that lacked or contained the KDEL retention signal, indicating that our conclusions cannot be because of changes to the hormone that occur during its migration through the secretory pathway.