Inhibition of Topoisomerase IIα Expression by Transforming Growth Factor-β1 Is Abrogated by the Papillomavirus E7 Protein¹ Free

Topo IIα³belongs to a family of enzymes that affect conformational changes in DNA during the cell cycle (1, 2). Peak topo IIαexpression and activity coincide during the G₂phase of the cell cycle, thus topo IIα protein expression is thought to be a sensitive and specific marker for cell proliferation(3, 4, 5, 6, 7). Cancer cells often have elevated topo IIαexpression, and drugs that target topo IIα are used widely for cancer treatment. One such drug, etoposide, stabilizes the “cleavable complex” form of topo IIα, which results in the production of unrepaired double-strand DNA breaks (8, 9, 10). This DNA damage initiates programmed cell death (apoptosis). Etoposide preferentially targets rapidly proliferating cells that have high topo IIα expression. Thus, bone marrow suppression and gastrointestinal toxicity are dose-limiting side effects of etoposide therapy.

TGF-β1 family members have emerged as promising candidates in pharmacological strategies aimed at minimizing the cytotoxic effects of chemotherapy on normal tissues. For example, TGF-β1 protects normal cells from etoposide-induced cell death (11), yet the mechanism has remained speculative. TGF-β1 reversibly arrests cell cycle progression in the G₁ phase of the cell cycle (12, 13, 14), and the ability to induce G₁ arrest is thought to be necessary for the chemoprotective effects of TGF-β1 (11). TGF-β1-induced G₁ arrest is associated with the inhibition of expression of B-myb and c-myb(15, 16, 17), transcription factors required by cells to exit G₁ and begin DNA synthesis (18, 19). Recently it was shown that the myb transcription factors are responsible for the cell cycle-specific induction of topo IIα mRNA in the S phase (20), although diverse mechanisms may contribute to the regulation of topo IIα expression (6, 21).

The Mv1Lu cell line is nontransformed and sensitive to TGF-β1-induced growth arrest, thereby serving as a useful model system for the study of topo IIα gene expression in a setting of nontumorigenic epithelial cell proliferation. On the basis of a putative pathway in Mv1Lu cells that includes B-myb as a key mediator, we predicted that TGF-β1 would inhibit topo IIαexpression. Furthermore, we predicted that the inhibition of topo IIαwould be dependent on the ability of TGF-β1 to induce a G₁ cell cycle arrest. Because the presence of topo IIα is required for the cytotoxic effect of etoposide, the inhibition of topo IIα expression by TGF-β1 would account for the ability of TGF-β1 to protect cells from etoposide-induced cytotoxicity.

Mv1Lu cells (ATCC CCL-64) were grown in a humidified incubator in 5%CO₂ at 37°C. The growth medium consisted of DMEM (Life Technologies, Inc.) and 5% dialyzed fetal bovine serum (Life Technologies, Inc.).

Porcine TGF-β1 (R&D Systems) was added to a final concentration of 10 ng/ml at the times indicated.

Cell cycle distribution was determined using a standard protocol for FACS analysis after treatment with BrdUrd for 30 min(22).

LXSN and LXSN-16E7 retroviruses were a gift from Dr. D. A. Galloway at the Fred Hutchinson Cancer Research Center(23). Mv1Lu cells were incubated in viral supernatant plus 4 μg/ml polybrene (Sigma Chemical Co.) for 24 h. Cells were then cultured in virus-free medium for 2 days and then selected in growth medium containing 400 μg/ml of G418 (Life Technologies, Inc.). Pooled populations of infected cells were generated from samples infected at 20–50% efficiency based on the percentage of cells surviving selection with G418. The concentration of G418 was reduced to 200 μg/ml once 100% of the uninfected control cells had died.

Cells were harvested at the times indicated in each figure and polyadenylated RNA was isolated using a standard protocol (Fig. 1; Ref. 24). Whole cell lysates were pooled from two to three flasks for each time point for polyadenylated RNA isolation. RNA was resolved by electrophoresis on an agarose gel and transferred to Hybond membrane. Complete transfer was confirmed by ethidium bromide staining.

[³²P]dATP and dCTP (Dupont/NEN) -labeled probes were made using a random primer DNA labeling kit (Boehringer Mannheim). Hybridizations were performed at 42°C using 1 × 10⁶ cpm/ml labeled DNA in hybridization buffer containing 50% formamide. Images and quantification of signal intensity were obtained using a phosphorimager (Molecular Dynamics).

The following cDNA templates were used for random primed labeling:(a) the 0.7-kb BamHI/PstI fragment from plasmid SP65 1B15, containing the rat cyclophilin cDNA(25); (b) a 0.9-kb fragment of the mink PAI-1 cDNA generated using oligo d(T)₄₀-primed mink cDNA (Superscript II; Life Technologies, Inc.) as a template for PCR(Taq polymerase; Roche Diagnostics, Inc.) using the forward primer ctggttctgcccaagttctc and the reverse primer cagggaacatgacaagcaaa and then nested PCR using the forward primer tcatggacagaccctttcctc and reverse primer caagcacacaaggacaagga; (c) the 1.8-kb EcoRI fragment from pc15 htopoIIα containing the human topo IIα cDNA (26); and (d) the 2.8 kb BamHI fragment from SP12 htopoIIβ containing the human topo IIβ cDNA (27). The topoisomerase probes did not show any cross hybridization. Using the probe for topo IIα, we detected an ≈6.5-kb band similar to the 6.2-kb mRNA reported for human topo IIα (26). The topo IIα band was distinct from the band detected using the probe for topo IIβ (not shown). The identity of the PAI-1 fragment was confirmed by sequencing.

Western analysis and immunostaining were performed using standard,published protocols (28). Using a topo IIα-specific antibody (Calbiochem Ab-1), we detected a M_r 170,000 doublet corresponding to the expected size of human topo IIα (27, 29).

Immunostaining of suspended cells for the simultaneous measurement of protein expression and cell cycle phase was performed using a standard immunostaining protocol adapted for cells in suspension. Topo IIα was labeled with FITC as follows. Cells were trypsinized to a single-cell suspension, then pelleted and resuspended in calcium and magnesium-free PBS. The sample size was adjusted to give 2 × 10⁶ cells. Each sample was rinsed in PBS and pelleted. The pellet of cells was resuspended in fixative (2%paraformaldehyde, 0.1% Triton X-100, in PBS) for 15 min on ice. Cells were then pelleted and rinsed twice with cold PBS. The pellet was then resuspended in 1 ml of cold PBS, and 2 ml of cold ethanol were added dropwise while vortexing. The cells were pelleted and resuspended in permeabilization buffer {0.2% Triton X-100 in Tris-buffered saline[10 mm Tris (pH 7.5), 100 mm NaCl, 5 mm KCl} for 5 min at room temperature. Cells were pelleted and resuspended in antibody buffer (0.1% Triton X-100, 1%BSA, 3% normal goat serum, in TBS) for 10 min. At this point each sample was split into two aliquots, one to be incubated with the topo IIα-specific antibody (Calbiochem Ab-1) at a 1:200 dilution and one aliquot to be incubated with an equal concentration of the nonspecific isotype control antibody (Sigma M-5284 mouse IgG1κ). The cells were resuspended and incubated for 30 min with their respective primary antibody in 500 μl of antibody buffer. The cells were pelleted and rinsed in TBS twice for 5 min at room temperature. Each sample was then incubated for 20 min in 500 μl of antibody buffer containing a 1:200 dilution of the secondary antibody [goat antimouse IgG1-FITC (Southern Biotechnology Associates)]. The cells were pelleted and rinsed in TBS twice for 5 min at room temperature.

To determine DNA content, the cells were then pelleted and stained with PI using a standard protocol.⁴

Green fluorescence (FITC) and red fluorescence (PI) were determined simultaneously in each cell (n = 20,000–50,000 cells/sample) using standard FACS analysis. Multicell aggregates were excluded from subsequent analysis. Average green fluorescence/cell was ascertained in each of the three cell cycle fractions distinguished by PI staining. Fraction 1 contained cells with a G₁ DNA content; fraction 2 contained mainly cells with an S DNA content; and fraction 3 contained cells with a G₂-M DNA content. The respective values for nonspecific green fluorescence ascertained using the nonspecific isotype control antibody were subtracted from the total green fluorescence ascertained using the topo IIα-specific antibody for each fraction to obtain the value for topo IIα expression in each cell cycle fraction (Fig. 3).

To determine the effect of TGF-β1-induced growth arrest on topo IIα expression, we compared topo IIα expression in control(untreated) and TGF-β1-treated Mv1Lu cells. Samples of cells were harvested simultaneously to determine topo IIα mRNA expression, topo IIα protein expression, and cell cycle distribution at the times indicated after TGF-β1 treatment (Fig. 1,A). We consistently observed a marked drop in topo IIα mRNA expression between 8 and 12 h after TGF-β1 treatment (Fig. 1, A and B). Comparison of mRNA expression with the cell cycle analysis revealed that most of the cells in S + G₂-M at 8 h after TGF-β1 treatment had completed mitosis and arrested in G₁ by 12 h(Fig. 1 A). Thus, mRNA expression decreases subsequent to the onset of G₁ arrest, which is consistent with our prediction that the inhibition of topo IIα is dependent on the ability of TGF-β1 to arrest growth in the G₁phase.

Topo IIα protein expression consistently decreased beginning 2–3 h after the decrease in mRNA expression, suggesting that the inhibition of mRNA expression mediates the inhibition of protein expression (Fig. 1 B). Comparison with the cell cycle distribution indicated that the inhibition of topo IIα protein expression did not occur until after most of the cells had completed S and G₂-M and accumulated in G₁. Between 12 and 24 h after TGF-β1 treatment, the cell cycle distribution of the TGF-β1-treated cells showed minimal change, yet most of the inhibition in topo IIα protein expression occurred during this 12–24-h interval. Therefore, in Mv1Lu cells, the TGF-β1-induced G₁ arrest occurred before any appreciable decrease in topo IIα protein expression, indicating that most of the decrease occurred while the majority of the cells were in the G₁ phase of the cell cycle. This is consistent with the observation that TGF-β1 treatment in S or G₂-M does not affect cell cycle progression until cells reenter the G₁ phase (12, 13, 14). On the basis of reports that there is minimal synthesis of topo IIαprotein during G₁ (3, 4, 5, 6, 7), it appears that much of the topo IIα protein present during G₁ in rapidly proliferating Mv1Lu cells is residual protein which was synthesized during the previous cell cycle.

To corroborate the observations made by Western analysis, we performed immunostaining of control and TGF-β1-treated Mv1Lu cells (Fig. 2). In untreated, rapidly proliferating cells we found that the topo IIα signal varied in intensity from cell to cell and was restricted to the nucleus, as reported previously(7). In cells treated with TGF-β1 for 24 h, we found a marked inhibition in topo IIα-specific nuclear staining. Indeed, the cells treated with TGF-β1 for 24 h (≈95% in G₁) appeared to have much less topo IIα protein than the control cells (≈65% in G₁; see comparison of cell cycle distribution at 24 h in Fig. 1). This suggested that the average topo IIα protein level during G₁ in TGF-β1-treated cells was less than the average level during G₁ in control cells. This observation is consistent with data in Fig. 1, which suggested that much of the topo IIα protein present during G₁is residual protein that was synthesized during the previous cell cycle. Therefore, a prolongation of the G₁interval by TGF-β1-induced arrest would be expected to eventually result in the elimination of topo IIα protein, just as we found in Fig. 1. By extending the interval from the point at which the cells last produced topo IIα protein, TGF-β1 seems to allow sufficient time to achieve nearly complete degradation of topo IIα protein.

A second feature of the immunostaining was surprising. In the TGF-β1-treated samples we did not see any cells with high topo IIαexpression despite the fact that about 5% of the cells continued to progress through the S and G₂-M phases of the cell cycle. This qualitative observation suggested that the subpopulation of cells that continued to cycle in the presence of TGF-β1 had significantly lower topo IIα expression during the later stages of the cell cycle.

To verify these observations, we developed a novel method for quantitative analysis of protein expression as a function of cell cycle position using immunostaining and FACS analysis. We simultaneously determined topo IIα protein expression and cell cycle position for each cell in samples of 2–5 × 10⁵ cells. As with standard immunostaining, we compared topo IIα expression in untreated (control) cells with that in cells treated with TGF-β1 for 24 h (Fig. 3). In untreated, rapidly proliferating Mv1Lu cells, topo IIα expression increased as cells progressed through the cell cycle. Compared with the level of topo IIαexpression during G₁, expression increased 3-fold in the S-phase fraction, and 7-fold in the G₂-M-phase fraction (Fig. 3,A)consistent with previous reports regarding the cell cycle-specific expression of topo IIα (3, 4, 5, 6). TGF-β1 significantly inhibited topo IIα expression in the G₁ and G₂-M fractions (Fig. 3, A and B).

Thus, the quantitative immunostaining confirmed our unexpected observation that the subpopulation of cells that continued to progress through the G₂-M phases of the cell cycle despite the presence of TGF-β1 had markedly decreased topo IIα expression. Close examination of the FACS data suggested two possibilities. The first is that this subpopulation of Mv1Lu cells is unable to respond to TGF-β1-induced G₁ arrest, and inexplicably always has lower-than-normal topo IIα expression during G₂-M, regardless of exposure to TGF-β1. If this were the case, TGF-β1 treatment would simply unmask the presence of this subpopulation by eliminating the TGF-β1-responsive cells through G₁ arrest. The second possibility is that this subpopulation fails to respond to TGF-β1-induced G₁ growth arrest, yet retains the ability to respond to TGF-β1 by decreasing topo IIα expression. In other words, the data suggest that in some Mv1Lu cells, the inhibition of topo IIα expression by TGF-β1 might occur independent of cell cycle arrest.

To pursue whether TGF-β1 might inhibit topo IIα expression independent of cell cycle arrest, we determined the effect of TGF-β1 on topo IIα expression in Mv1Lu cells that fail to undergo G₁ arrest (Fig. 4,A). It has been shown that inactivation of pRb by viral oncoproteins results in cells that are unable to undergo growth arrest in response to TGF-β1, yet in such cells the effects of TGF-β1 on gene expression that do not require pRb are preserved (30). We established a pooled population of Mv1Lu cells that express the papillomavirus type 16 E7 oncoprotein from a retrovirus-introduced transgene(Mv1Lu-LXSN-E7). We found that Mv1Lu-LXSN-E7 cells did not undergo G₁ arrest in response to TGF-β1, whereas control cells (Mv1Lu-LXSN) responded normally (Fig. 4,A). We have found that TGF-β1-treatment continues to alter the expression of a number of TGF-β1-responsive genes in Mv1Lu-LXSN-E7 cells, including PAI-1 (Fig. 4), topo IIβ (see below), and others (data not shown),indicating that many TGF-β1-mediated effects on gene expression are retained in Mv1Lu-LXSN-E7 cells.

Interestingly, Mv1Lu-LXSN-E7 cells had significantly higher topo IIαmRNA levels than Mv1Lu-LXSN [120.2 ± 16.5% (SD) of control (Mv1Lu-LXSN), P = 0.05]. Thus,inactivation of pRb resulted in an increase in topo IIα mRNA expression. The increase in topo IIα mRNA expression in Mv1Lu-LXSN-E7 cells was not attributable to a greater fraction of cells in the S and G₂-M phases of the cell cycle because the cell cycle distribution was the same for Mv1Lu-LXSN and Mv1Lu-LXSN-E7 cells(data not shown). Rather, this result is consistent with the idea that pRb normally functions in a pathway to inhibit topo IIα synthesis primarily during the G₁ phase of the cell cycle. Thus, inactivation of pRb by E7 would increase topo IIα synthesis primarily during G₁, accounting for the relatively modest increase. In contrast to topo IIα, topo IIβ mRNA expression was not significantly affected by E7 (data not shown).

To determine whether the inhibition of topo IIα expression by TGF-β1 is dependent on pRb-mediated cell cycle arrest, we determined the response of topo IIα to TGF-β1 in Mv1Lu-LXSN and Mv1Lu-LXSN-E7 cells. TGF-β1 treatment inhibited topo IIα mRNA expression by 95%in Mv1Lu-LXSN cells, but TGF-β1 did not significantly affect topo IIα mRNA expression in Mv1Lu-LXSN-E7 cells (Fig. 4, B and C). Similarly, topo IIα protein expression was virtually eliminated after TGF-β1 treatment of Mv1Lu-LXSN cells, but TGF-β1 did not significantly affect topo IIα protein expression in Mv1Lu-LXSN-E7 cells (Fig. 4, B and D). These results indicated that pRb is required for the inhibition of topo IIαexpression by TGF-β1, and the inhibition is associated with pRb-mediated cell cycle arrest.

Of interest, the response of topo IIβ again differed from that of topo IIα. We found that TGF-β1 treatment significantly increased topo IIβ mRNA expression in both Mv1Lu-LXSN and Mv1Lu-LXSN-E7 cells,indicating that the induction of topo IIβ mRNA expression by TGF-β1 does not depend on pRb status [139.6 ± 12.5% (SD) of control (P < 0.01) versus134.9 ± 19.8% (SD) of control (P < 0.05)].

The TGF-β family members have emerged as promising candidates in pharmacological strategies aimed at minimizing the cytotoxic effects of chemotherapy on normal tissues while leaving cancer cells unaffected. The rational for the experimental use of TGF-β1 has been the observation that normal cells are responsive to the growth inhibitory effects of TGF-β1, whereas cancer cells typically have lost the ability to respond to the TGF-β1 growth inhibitory signal. A great deal of optimism has been generated for the clinical feasibility of this approach on the basis of cell culture and animal studies showing that TGF-β1 protects rapidly dividing stem cell pools in the bone marrow and gastrointestinal tract from cell cycle-acting chemotherapy in vitro and in vivo (11, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40). If this experimental approach can be successfully translated to bedside practice, TGF-β1 treatment would transiently arrest normal cells in G₁, but the TGF-β1-resistant tumor cells would continue to progress through the cell cycle, and therefore would remain susceptible to cell cycle-acting chemotherapy. Thus, pretreatment of the cancer patient with TGF-β1 might allow the use of higher (more tumoricidal) doses of chemotherapy,because the side effects caused by chemotherapy-induced damage to normal tissues would be reduced or prevented.

Our results support the idea that, in the proper setting, TGF-β1 might be a useful adjuvant to protect normal cells from cancer chemotherapy. Our report focuses on the regulation of expression of topo IIα, the target of the chemotherapeutic agent etoposide. Etoposide preferentially kills rapidly proliferating cells that have high topo IIα expression. Thus, bone marrow suppression and gastrointestinal toxicity are dose-limiting side effects of etoposide therapy. In addition, secondary leukemias with translocations between chromosomes 11 and 9 appear to occur as a result of treatment with etoposide-like drugs (8, 41, 42). TGF-β treatment protects Mv1Lu cells from cell death induced by treatment with drugs that target topo IIα, including etoposide (11, 37). Our preliminary data indicate that as much as a 10-fold-higher dose of etoposide is needed to kill 90% of Mv1Lu cells if they have been pretreated for 24 h with physiological doses of TGF-β1 (1 ng/ml;data not shown). Because topo IIα expression is required for etoposide-induced cell death, and peak expression of topo IIαnormally occurs during the G₂ phase of the cell cycle, we predicted that TGF-β1 would inhibit topo IIα enzyme expression, thereby protecting cells from etoposide.

We found that TGF-β1 inhibits topo IIα protein expression by a mechanism involving inhibition of topo IIα mRNA expression. Previous work suggests that topo IIα mRNA synthesis is inhibited by TGF-β1 through a pathway mediated by the inhibition of B-mybexpression in Mv1Lu cells (15, 20). In addition, the possibility that TGF-β1-induced G₁ arrest might indirectly affect mRNA stability is suggested by the report that in HeLa cells the half-life of the topo IIα mRNA is shortest during G₁ (6).

In general, reports indicate that there is a tight correlation between the level of topo IIα mRNA expression and protein expression; however topo IIα protein stability has been found to vary during the cell cycle as well. The half-life of topo IIα protein in normal chicken embryo fibroblasts was shortest (1.3 h) during early G₁ (21). Our observations regarding the drop in topo IIα protein expression as Mv1Lu cells arrest in G₁ suggest that the half-life is similarly short in Mv1Lu cells.

Because TGF-β1 did not decrease expression of the topo IIβ isoform,our finding that TGF-β1 decreases expression of topo IIα and decreases etoposide-induced cytotoxicity suggests that topo IIα is the primary target of etoposide in proliferating Mv1Lu cells. It has been demonstrated that decreased topo IIα expression commonly allows cells to escape death caused by etoposide (10, 43),and the degree of inhibition seen in Mv1Lu cells after TGF-β1 would seem to fully account for the ability of TGF-β1 to protect Mv1Lu cells from etoposide. However, our results do not exclude the possibility that TGF-β1 might also inhibit cellular uptake of etoposide, or induce a posttranslational modification in topo IIαthat alters the ability of etoposide to cause DNA damage.

By determining the level of topo IIα protein expression and cell cycle position simultaneously, we were able to gain a novel perspective of the heterogeneity in response of individual cells to TGF-β1. Specifically, in a population of Mv1Lu cells, there is always a small fraction that fail to undergo G₁ arrest, yet features that might distinguish this group of cells to allow them to be studied have been inaccessible previously. We have shown that conclusions that could only be inferred from Western blotting and immunostaining can be demonstrated quantitatively using FACS analysis of individual cells. We found unexpectedly that the subpopulation of cells that continued cycling (failed to undergo G₁ arrest) in the presence of TGF-β1 had decreased topo IIα expression. This observation is consistent with two possibilities. The first is that this sub-population of cells has much lower-than-normal levels of topo IIα expression without TGF-β1 treatment, and that the cells neither arrest growth nor decrease topo IIα in response to TGF-β1. The second possibility is that this subpopulation of cells is able to respond to TGF-β1 by decreasing topo IIα expression, but is able to escape TGF-β1-induced growth arrest, indicating that inhibition of topo IIα expression can occur independent of G₁ cell cycle arrest. The latter possibility regarding this subpopulation would suggest that TGF-β1 might mediate resistance to etoposide in tumor cells that have lost the ability to undergo G₁ arrest but otherwise have the TGF-β1 signaling pathway intact.

This subpopulation of Mv1Lu cells that fails to respond to TGF-β1 growth arrest and has decreased topo IIα expression may prove to be a useful model for some types of etoposide-resistant cancer cells that share these two features. If some types of cancer are able to respond to TGF-β1 by decreasing topo IIα expression without undergoing G₁ arrest, then it seems likely that these cancers might respond better to etoposide if TGF-β1 signaling is blocked. Interestingly, a similar conclusion was reached from studies that showed that neutralizing TGF-β could reverse the resistance of mouse mammary carcinoma cells to the alkylating agents cyclophosphamide and cisplatin in vivo (44, 45, 46). Thus, for some types of cancer, rather than increasing TGF-β1 to protect normal tissues, it may be more beneficial to reduce TGF-β1 signaling in the tumor to augment sensitivity of the tumor cells to chemotherapy. Of interest, several strategies have proven to effectively reduce TGF-β1 signaling in vivo (44, 45, 47, 48, 49, 50).

In summary, we found that the ability of TGF-β1 to inhibit topo IIαexpression correlated generally with the ability of TGF-β1 to induce a G₁ cell cycle arrest in Mv1Lu cells. Inactivation of pRb by expression of the papillomavirus type 16 E7 protein effectively abrogated the ability of TGF-β1 to both arrest growth and inhibit topo IIα expression. However, using quantitative immunohistochemistry, we identified a subpopulation of Mv1Lu cells that failed to undergo G₁ arrest in response to TGF-β1 yet unexpectedly had low levels of topo IIα during G₂-M. The latter observation suggested that topo IIα expression might be dissociable from cell cycle position under rare circumstances.

By further elucidating mechanisms that regulate topo IIα expression,this work may aid in the search for drugs that protect normal cells by specifically inhibiting topo IIα expression in vivo during chemotherapy with agents that target topo IIα. Additionally, the available evidence strongly supports the need for additional investigation to determine whether there is a safe and effective means to transiently modulate TGF-β1 activity to achieve the most successful response in cancer patients undergoing cell cycle-acting cancer chemotherapy. The question of whether increasing or decreasing TGF-β1 might be most beneficial in a particular patient needs to be considered when developing techniques like microarray analysis to reveal the unique features of an individual tumor. Our results offer the suggestion that TGF-β1 will not reduce the sensitivity of pRb-deficient tumors to etoposide, therefore the status of pRb in a tumor might help indicate whether a patient would benefit from pretreatment with a TGF-β1-like agent.

We thank Joe Holden, David Jones, and Bill Carroll for their critical review and helpful comments on the manuscript. We thank Joe Holden for providing the topoisomerase IIα and IIβ cDNAs, as well as the topoisomerase IIα-specific antibody. We are grateful to Pauline Cordray for expert technical assistance.