Diabetes Increases Formation of Advanced Glycation End Products on Sarco(endo)plasmic Reticulum Ca²⁺-ATPase

Mouse monoclonal SERCA2a and phospholamban antibodies were obtained from Affinity Bioreagents (Golden, CO). Actin antibodies (C-11) were purchased from Santa Cruz Biotechnology (Santa Cruz, CA). Brevital (methohexital sodium) and NPH Ilentin II (intermediate-acting insulin) were obtained from Eli Lilly (Indianapolis, IN). Other reagents and solvents used were of analytical grade.

All animal procedures were carried out in accordance with guidelines established by institutional animal care and use committees and are described in detail elsewhere (40–42). Briefly, male Sprague-Dawley rats weighing between 180–190 g were anesthetized with Brevital (25 mg/kg i.p.) or isofluorane and then injected via tail vein with either streptozotocin (STZ) in 0.1 mol/l citrate buffer, pH 4.5 (50 mg/kg), or citrate buffer. Three days later, blood glucose levels were determined using a Glucometer II and Glucostix (Peridochrom Glucose GOD-PAP Assay Kit; Roche Molecular Biochemicals, Indianapolis, IN). Throughout this study, animals were housed in pairs (similar weights to minimize dominance) at 22°C with fixed 12-h light/dark cycles and given free access to food and water.

Six weeks after the STZ injection, diabetic animals were randomly divided into two subgroups. One subgroup was placed on an insulin regimen (NPH Ilentin II; Lilly, Indianapolis, IN) for 2 weeks. Tail blood glucose levels of these animals were measured daily (between 9:00 and 10:00 a.m.), and insulin doses were individually adjusted so as to maintain euglycemic states. The other subgroup of diabetic animals continued as nontreated diabetic subjects for 2 additional weeks. Any diabetic animal whose body weight fell by >10% of initial starting body weight during in vivo protocol was killed and eliminated from the study.

Animals were given a single lethal dose of Brevital (75 mg/kg i.p.) or Inactin (thiobutabarbital, sodium salt, 60 mg/kg i.p). Abdominal cavities were opened and blood samples were collected from left renal arteries (43). HbA_1c levels were determined from whole blood using Roche Diagnostics Unimate 5 assay kit (Roche Diagnostics, Indianapolis, IN), and insulin was assayed from serum using Mercoda rat insulin ELISA kit (Alpco Diagnostics, Salem, NH). After removal of blood, chest cavities were opened and hearts from 8-week age-matched control (8C), 8-week STZ-induced diabetic (8D), and 6-week STZ-induced diabetic/2-week insulin-treated (6D-2I) animals were excised. Hearts assigned for measurement of cardiac function using modified Langendorff procedure were immediately placed in oxygenated Kreb-Henseleit buffer (118 mmol/l NaCl, 27.2 mmol/l NaHCO₃, 4.8 mmol/l KCl, 1.2 mmol/l MgSO₄, 1.0 mmol/l KH₂PO₄, 1.25 mmol/l CaCl₂, and 11.0 mmol/l glucose at 37°C), and those to be used for measurement of SERCA2a and PLB content were quick-frozen by embedding in crushed dry ice.

Two procedures were used to compare cardiac function among animals from the three experimental groups. In the first procedure, in vivo left ventricular function (±dP/dt, left ventricular end diastolic pressure, and heart rate) was evaluated in anesthetized animals (at least four animals per group) to ascertain changes in basal cardiac function induced by diabetes. For this, animals were lightly anesthetized with Inactin (20 mg/kg i.p), and a catheter containing a pressure transducer was inserted via the carotid artery into the left ventricle to determine basal heart rate, left ventricular pressure, left ventricular end diastolic pressure, and the derivatives ± dP/dT, as previously described (44). Powerlab data acquisition system (ADInstruments, Colorado Springs, CO) was used for acquiring and analyzing measurements. In the second procedure, isoproterenol-induced contraction/relaxation properties of hearts from 8C, 8D, and 6D-2I animals were compared using a modified isobaric Langendorff procedure (45) under similar perfusion conditions. For this, hearts were excised from animals, cannulated via the aorta, and then perfused retrograde at a flow rate of 10 ml/min with oxygenated standard Krebs-Henseleit buffer (37°C). The right atrium was then removed and the ventricle was pierced at the apex and connected to a force displacement transducer (FT03C; Grass Instrument, Quincy, MA) via silk thread for measurement of developed tension. Arteriovenous node was crushed and the ventricle was then stimulated at a constant pacing rate of 190–200 bpm. After establishing basal isometric tension, hearts were then challenged with freshly prepared isoproterenol bitartrate (1, 10, and 100 nmol/l, given as bolus via a side arm in aortic cannula) and changes in isometric developed tension (at peak) were determined. Changes in relaxation rates (−dT/dt_max) were determined and used in this study as indirect indexes of SERCA2a activity. Hearts used for evaluation of cardiac function employing the Langendorff procedure were not used for biochemical analyses.

Total RNA was extracted separately but simultaneously from 8C, 8D, and 6D-2I rat hearts using Quick Prep total RNA extraction kits (Amersham Pharmacia Biotech, Piscataway, NJ) as previously described (40–42). Thereafter, equivalent amounts of RNA from each of 8C, 8D, and 6D-2I (with distinct 18S and 28S bands on formamide agarose gels, three separate preparations) were used to synthesize 1st strand cDNA. PCRs were then carried out using the following program: 5 min denaturation (94°C), 1 min annealing (56.7°C), 2 min extension (72°C), repeated for a total of 37 cycles. Primers for SERCA2a were sense (5′-GAAGATGAATGATTCGACACG-3′) and antisense (5′-TGGAGAAGTTGTCGTCGG-3′), whereas those for β-actin were sense (5′-CGTAAAGACCTCTATGCCA-3′) and antisense (5′-AGCCATGCCAAATGTCTCAT-3′).

Relative levels of SERCA2a and phospholamban proteins were determined in two steps. First, membrane vesicles were prepared simultaneously from hearts of 8C, 8D, and 6D-2I animals (two hearts per preparation × 3 preparations) using procedures previously described (40–41). Second, 60 μg of total protein from each preparation was solubilized in gel dissociation medium containing 10 mg/ml dithiothreitol and then loaded onto denaturing 4–20% linear gradient polyacrylamide gels. Only those samples used for analysis of SERCA2a were boiled for 15 min before loading onto the gels. The gels were then electrophoresed at 150V at 4°C for 3.5 h for SERCA2a or 1.5 h for PLB. Proteins on the gels were then transferred for 1 h at 400 mA onto polyvinylidene difluoride membrane using a semi-dry procedure, as previously described. The membrane was then probed for SERCA2, PLB, and β-actin proteins, as previously described (40–42).

AGEs were identified using the procedures previously described (42,46,47). In this study, we investigated N^ε-(carboxymethyl)-lysine, imidazolone A, imidazone B, pyrraline, and 1-alkyl-2-formyl-3,4-glycosyl pyrrole molecule (AFGP) as well as crosslinking AGEs crossline and pentosidine (Fig. 1). Two restriction criteria were employed when searching for noncrosslinking AGEs: 1) only a single modification was allowed within the peptide fragment and 2) the peptide must contain one missed cleavage at a lysine or arginine residue (or no miscleavage but containing a histidine residue). For crossline, the crosslinkage must occur between lysine residues on the two peptides, and for pentosidine AGEs the crosslinkage must occur between a lysine residue on one peptide and an arginine residue on the other.

Differences between values from each of the 8C, 8D, and 6D-2I rats were evaluated using the nonparametric Kruskal-Wallis (GBStat version 6.5.4; Dymanic Microsystems, Silver Spring, MD). The data shown are means ± SE and results are considered significantly different if P < 0.05.

Mean body weight of animals at the start of study was 187.2 ± 2.0 g. After 56 days, weight of 8C animals increased to 399.0 ± 8.7 g (n = 12), whereas those of 8D animals increased to 225.4 ± 10.4 g (n = 13, P < 0.05). Although diabetic animals did not gain weight as rapidly, they fed normally and moved around in their cages freely. Two weeks of insulin treatment initiated after 6 weeks diabetes duration restored loss in body weight (279.6 ± 13.1 g, n = 12). Initial average glucose level was 4.4 ± 0.1 mmol/l. Mean blood glucose level of 8C animals remained low (3.9 ± 0.3 mmol/l). Three days after injection with STZ, mean blood glucose levels of diabetic animals were 20.7 ± 0.6 mmol/l. This level increased progressively to a high of 28.5 ± 2.5 mmol/l on day 56. Insulin treatment also lowered mean blood glucose level to 5.1 ± 3.8 mmol/l. Serum insulin levels were significantly lower in 8D animals (0.5 ± 0.1 ng/ml compared with 2.6 ± 0.5 ng/ml in 8C animals). Heart-to-body weight ratios were significantly higher in 8D compared with 8C animals (4.13 ± 0.2 mg/g vs. 3.0 ± 0.1 mg/g). This ratio returned to nearly 8C values with insulin treatment (3.76 ± 0.2 mg/g). HbA_1c levels in 8C, 8D, and 6D-2I animals were 4.3 ± 0.1, 9.5 ± 0.25, and 5.8 ± 0.5%, respectively.

Before conducting biochemical analyses, left ventricular function (in vivo and in vitro) of hearts from 8C, 8D, and 6D-2I animals were evaluated to establish cardiomyopathy. Consistent with other studies, data from both techniques used in this study confirm that diabetes decreases rate of cardiac relaxation in rat hearts.

By inserting a pressure transducer into the left ventricle of lightly anesthetized rats, we found that after 8 weeks of diabetes basal heart rate, left ventricular pressure, and its derivative −dP/dt_max decreases whereas left ventricular end diastolic pressure increases. These changes can be seen clearly (Fig. 2A and B) and show representative 3-s tracings of basal hemodynamics from 8C, 8D, and 6D-2I animals. Most obvious (Fig. 2A) are the decreases in heart rate and the slowing in decrease in left ventricular pressure decay. The latter is shown more clearly (Fig. 2B [8D panel]). On closer examination, it is also very clear that left ventricular end diastolic pressure increases in the diabetic heart, possibly reflecting ventricular stiffness. Two weeks of insulin treatment initiated after 6 weeks of diabetes prevented and/or minimized the slowing in heart rate, the decrease in left ventricular pressure (and its derivatives ± dP/dt), and the increase in left ventricular end diastolic pressure. A summary of hemodynamic changes induced by 8 weeks of diabetes and its reversal with insulin treatment is shown in Table 1.

When paced at 190 bpm in a Langendorff apparatus, hearts from 8D animals developed significantly less tension compared with 8C animals (7.0 ± 0.5 g tension compared with 9.5 ± 1.2 g), consistent with cardiomyopathic hearts. Two weeks of insulin treatment initiated after 6 weeks of diabetes minimally improve basal developed tension (7.2 ± 0.6 g). When hearts from 8C animals were challenged with bolus 1 nmol/l isoproterenol, developed tension increased to 18.1 ± 1.2% over basal within 12 s. This increase persisted for ∼3 s and then gradually subsided. Further increases in developed tension were also obtained when 8C hearts were stimulated with 10 nmol/l and 100 nmol/l isoproterenol (amplitude of the increases were 28.6 ± 2.5 and 24.7 ± 1.3% over basal, respectively). On the other hand, when hearts from 8D animals were stimulated with 1 nmol/l isoproterenol, minimal changes in isometric developed tension were observed. On stimulation with 10 nmol/l isoproterenol, the peak response took almost 35 s to be attained, but developed tension increased by 80.3 ± 5.6%. Interestingly, the peak response also persisted for >2 min and subsided to basal levels 5 min after addition of the drug. Increasing the concentration of isoproterenol to 100 nmol/l resulted in a 70 ± 8.7% increase in developed tension. These data are consistent with the lower sensitivity of diabetic hearts to isoproterenol. When hearts from 6D-2I animals were challenged with 1, 10, and 100 nmol/l isoproterenol, developed tension increased by 20.2 ± 6.5, 35.1 ± 5.4, and 33.2 ± 5.3% over basal, and the time to peak response was of the order of 12–15 s (similar to those of control animals). Responses of hearts from 6D-2I animals also persisted for >90 s. Thus, insulin treatment is able to restore diabetes-induced loss in isoproterenol sensitivity.

Figure 2C shows the summary of the changes in −dT/dt_max when hearts from 8C, 8D, and 6D-2I animals were challenged with 1, 10, and 100 nmol/l isoproterenol. These data are in agreement with those of Dincer et al. (48) and Afzal et al. (49) and clearly show that diabetes decreases relaxation rates. Taken together, data from in vivo ventricular function measurements and isobaric Langendorff procedure establish that after 8 weeks of diabetes left ventricular function was compromised and that 2 weeks of insulin treatment initiated after 6 weeks of diabetes prevented and/or reversed this dysfunction.

After normalizing to concomitant β-actin, mRNA encoding SERCA2a decreased by 15.6 ± 8.3% following 8 weeks of STZ-induced diabetes (Fig. 3A). However, this difference was not significant from age-matched controls (P > 0.05). Steady-state levels of mRNA encoding SERCA2a in hearts from 6D-2I animals were similar to that of 8C animals (96.9 ± 7.1% of control). Nucleotide sequence analyses and restriction mapping confirmed that the PCR products generated in the PCRs were derived from specific amplification of mRNA encoding SERCA2a (data not shown).

After normalization to concomitant β-actin protein, immunoreactive SERCA2a protein from 8D rat hearts was 39.9 ± 7.2% lower than age-matched 8C rats (Fig. 3B) (P < 0.05). On some Western blots it is apparent that diabetes may be altering the electrophoretic mobility (SDS-PAGE) of SERCA2a. In one preparation (made from two hearts), the electrophoretic mobility of SERCA2a from 8D was faster than that of 8C and 6D-2I (see Fig. 2B [middle panel]), whereas in another preparation the electrophoretic mobility of SERCA2a from 8D was slower. The shift in electrophoretic mobility is likely due to posttranslation modifications. Two weeks of insulin-treatment, initiated after 6 weeks of diabetes minimized and/or prevented loss in SERCA2 expression (109.2 ± 2.2% of control).

After 8 weeks of STZ-induced diabetes, steady-state levels of immunoreactive monomeric (dephosphorylated) PLB increased significantly in hearts from 8D rats (Fig. 4A). When normalized to concomitant β-actin protein, the magnitude of the increase was 40 ± 4.2% over control. Two weeks of insulin-treatment initiated after 6 weeks of diabetes prevented (and/or reversed) the increase in expression induced by diabetes. In this study, steady-state levels of phosphorylated pentameric PLB also increased with diabetes. Two distinct phosphorylated forms of PLB proteins were detected, likely suggesting mono- (either on serine-16 or threonine-17) and diphosphorylation (both serine-16 and threonine-17). Two weeks of insulin treatment decreased steady-state levels of immunoreactive diphosphorylated PLB (assuming that the higher running of the two bands reflect the diphosphorylated form). Insulin treatment did not change the amount of monophosphorylated PLB induced by diabetes.

Overnight digestion of SERCA2a from 8C with trypsin afforded 75 peptides with monoisotopic between 700 and 2,500 Da. Of these, 26 (35%) correspond to theoretical trypsin fragments (nonmodified) generated in silico from rat SERCA2a (GenBank accession no. B31982), spanning the entire range of the rat SERCA2a protein with minimum variability (≤560 ppm). Digestion of SERCA2a from 8D animals with trypsin afforded 60 peptides, 14 (23%) of which represented nonmiscleaved/nonmodified theoretical SERCA2a peptides. Only seven of these peptides were a subset of 8C peptides, suggesting that diabetes is inducing changes to SERCA2a that compromise trypsin’s ability to digest it. Trypsin digestion of SERCA2a from 6D-2I animals afforded 50 peptides with monoisotopic ranging from 700 to 2,500 Da. Of these, 22 (44%) corresponded to the masses of nonmodified trypsin-derived peptides from rat SERCA2a, and 16 of them were identical to 8C peptides. Thus, insulin treatment only partially attenuated trypsin-sensitive formation of posttranslational modifications on SERCA2a (modifications at lysine and arginine residues).

The next step was to search the matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass data files to determine whether AGEs are formed on SERCA2a, and if so, what type of AGEs are formed and where they are located on protein. In the first step of the protocol, rat SERCA2a (GenBank accession no. B31982) was digested in silico and M+H⁺ of all possible peptides (including cysteine with iodoacetamide and acrylamide adducts, oxidized methionines, and one miscleavage) were uploaded into the PERL algorithm along with the MALDI-TOF mass data files. The data files were then searched for select noncrosslinking N^ε-(carboxymethyl)-lysine, imidazolone A, imidazone B, pyrraline, and AFGP as well as crosslinking AGEs (crossline and pentosidine). During the analyses, M+H⁺s that were found in the mass files of all three 8C, 8D, and 6D-2I samples and not attributed to SERCA2a peptides were considered contaminants and eliminated.

As an example, search of data files revealed a monoisotopic mass of 1426.71 Da that is prominent in 8D but not in 8C and 6D-2I samples (Fig. 5). Since no peptide generated from trypsin-digestion of SERCA2a possesses an M+H⁺ of 1,426.71 ± 0.5 Da, this M+H⁺ likely reflects the mass of a SERCA2a peptide from 8D samples that contains an AGE adduct. The next step was to determine the nature of the modification and which peptide fragment (amino acid) contained this adduct. For this, the algorithm searched the theoretical SERCA2a digestion file for M+H⁺s that were 58.02, 58.03, 108.02, 142.03, 144.03, 252.11, or 270.07 Da < 1,426.71 ± 0.5 Da. No M+H⁺ of 1,368.69 (i.e., 1,426.71–58.02), 1,318.69, 1,284.68, 1,282.68, 1,368.68, or 1,174.60 Da, respectively, were detected in the theoretical mass file, suggesting that these masses do not correspond to masses of trypsin-derived SERCA2a peptides. However, an M+H⁺ of 1,156.60 ± 0.1 Da was detected. This M+H⁺ corresponds to the mass of the miscleaved peptides ₄₈₁KEFTLEFSR₄₈₉. The subscript numbers represent the location of these peptides on rat SERCA2a and the bold letter indicates the miscleaved site. Therefore, an M+H⁺ of 1,426.71 in 8D samples corresponds to the mass of the peptide ₄₈₁KEFTLEFSR₄₈₉ modified at arginine 481 with an AFGP adduct. As a second example, search of MALDI-TOF mass files revealed a monoisotopic mass of 2,485.12 Da is prominent in 8D but not in 8C and 6D-2I samples (Fig. 6). Using the PERL algorithm, this M+H⁺ was identified as the mass of two peptides (₄₅₂MNVFDTELKGSLK₄₆₄ and ₆₂₉GTAVAICRR₆₃₇) crosslinked with pentosidine via lysine 460 and arginine 635.

Modifications were not found only on peptides from 8D samples. Some peptides from insulin-treated SERCA2a also contained modifications that were either less prominent or absent altogether from SERCA2a of 8D samples and absent from SERCA2a from 8C samples. An M+H⁺ of 899.17 Da was detected in mass data file from 6D-2I SERCA2a samples (Fig. 7), but barely seen in diabetic SERCA2a (899.13 Da) and absent from 8C samples. Analysis of this data revealed that this M+H⁺ is attributed to a single AFGP (+270.07 Da) adduct on the histidine-3 residue within the peptide ₁CTHIR₇ (629.32 Da). It should be pointed out that the M+H⁺ of 908.60 ± 0.3 Da in all three samples (Fig. 7) could reflect a contaminant because this M+H⁺ is not the mass of any of the theoretical SERCA2a peptides.

Unlike in previous studies (45–47), we did not detect modifications resulting from N^ε-(carboxymethyl)-lysine, pyralline and crossline, attesting to the uncertainty as to which AGEs are formed on which protein and under what conditions. We also did not detect the AGEs precursors, Schiff bases, and Amadori products on SERCA2a. A complete listing of AGE modifications detected on SERCA2a is given (Table 2). It should also be mentioned that a few noncrosslinking AGEs were also localized to consensus SERCA2a peptides from 8C, 8D, and 6D-2I samples. These modifications were not included in Table 2 as we felt that they were attributed to the natural aging process and not to the diabetes.

The present study shows for the first time that AGEs are formed on intracellular SERCA2a during chronic diabetes. They are also the first to pinpoint the specific amino acids within the SERCA2a sequence where AGEs modifications occur. These posttranslational modifications were detected by digesting SERCA2a with trypsin, determining the masses of peptides using MALDI-TOF mass spectrometry, and then searching the data files for AGEs using an in-house PERL algorithm. Imidazolone A was found on the A domain (on arginine 164), AFGP was found on the P domain (on lysine 481), and imidazolone A, imidazolone B, and AFGP were found on the N domain (on histidine 526 and arginine 635). In addition, a large number of pentosidine crosslinkages were also found intra- and interdomains. One crosslinkage was found within the A domain (between lysine 135 and arginine 204), two between A and N domains (between lysine 142 and arginine 677, and between lysine 135 and arginine 655), one between P and N domains (between lysine 460 and arginine 636), and one within the N domain (between lysine 683 and arginine 619). It should be pointed out that none of these AGEs adducts were found within the transmembrane helixes. The latter is expected because these regions are not in direct contact with the high cytosolic concentrations of reducing aldose and ketose sugars.

These data also suggest a novel mechanism by which rate of cardiac relaxation can be prolonged during diabetes. By disrupting intra- and interdomain tertiary structures, AGEs complexes can compromise structural movements required for translocating calcium ions from the cytosol to the lumen of the sarcoplasmic reticulum, and this could lead to a lowering in the activity of SERCA2a. As an example, since A, P, and N domains were found to be crosslinked with AGEs during diabetes, it is likely that the open structure conformation required for phosphorylation may not be readily achieved, leading to loss in activity.

A second major finding of the present study is that SERCA2a from 6D-2I animals no longer contained crosslinking AGEs and had fewer noncrosslinking AGEs compared with SERCA2a from 8D animals. Why crosslinking but no noncrosslinking AGEs were found remains unclear. Nonetheless, since rates of relaxation were noticeably improved in 6D-2I animals, it is likely that crosslinking AGEs adducts more severely compromise SERCA2a ability to pump calcium ions than do noncrosslinking AGEs.

We also found that although steady-state levels of mRNA encoding SERCA2a were not significantly altered by diabetes, steady-state levels of immunoreactive SERCA2a protein decreased. One simple explanation for unchanged mRNA levels may be that too many cycles were used for PCRs (measurements were made at the plateau phase of the product formed versus cycle number curve). We are in the process of conducting real-time PCR to address this question. Alternatively, since steady-state protein level is the sum product of its rates of transcription, translation, and degradation, it is likely that either translation rates of SERCA2a protein may be slower than transcription rates or degradation rates of SERCA2a protein are faster than transcription and translation rates combined.

We did not detect any immunoreactive phosphorylated PLBs in control samples. This was very surprising because samples for SDS-PAGE were not heated (heating is known to dissociate the pentamer), and this is inconsistent with the results of Kim et al. (30) and Choi et al. (31). The reason for lack of phosphorylated PLB in control samples is unknown at this time. Consistent with previous studies (30,31), we also found that diabetes increases expression of monomeric unphosphorylated PLB. Since expression of SERCA2a decreases with diabetes, the ratio of SERCA2a to PLB decreases further, resulting in loss of SERCA2a activity. In this study, we also found that diabetes increases the amount of steady-state levels of phosphorylated PLB. Two distinct phosphorylated PLB bands were observed in 8D samples, likely reflecting the mono- and diphosphorylated forms of PLB. At this time, we are uncertain as to which amino residue (serine-16 or threonine-17) is phosphorylated in the monophosphorylated form. These data are also inconsistent with those of Choi et al. (31), who found that diabetes decreases the amount of serine-16 and threonine-17 phosphorylation of PLB. The reason(s) for this discrepancy is not known at this time. We also found that insulin treatment significantly reduces steady-state levels of diphosphorylated PLB, but did not decrease the amount of monophosphorylated PLB.

In conclusion, data from the present study suggest that in addition to a decreases in expression SERCA2a, an increase in PLB and concomitant decreases in SERCA2a-to-PLB ratio, prolongation of cardiac relaxation rate may also be attributed to diabetes-induced increase in formation AGEs (crosslinking as well as noncrosslinking) on SERCA2a. Our data also show 2 weeks of insulin treatment initiated after 6 weeks of diabetes significantly improved cardiac function and also prevented formation of crosslinking AGEs on SERCA2a. In future studies, we intend to treat animals with aminoguanidine (blocker of AGEs) to establish a cause-effect relationship.

This work was supported in part by grants from the National Institutes of Health (HL66898) and the Ralph W. and Grace R. Showalter Foundation.

Mass spectrometry was done at Indiana University School of Medicine, Proteomics Core Facility with support from the Indiana Genomic Initiative (INGEN).

The authors thank Sarah Ingersoll for technical assistance.