To assess the Y RNAs, a family of homologous RNAs that bind to the Ro autoantigen, for the ability to contribute to autoimmune disease by activating RNA-responsive Toll-like receptors (TLRs).
To assess the Y RNAs, a family of homologous RNAs that bind to the Ro autoantigen, for the ability to contribute to autoimmune disease by activating RNA-responsive Toll-like receptors (TLRs).
Using cell lines expressing or stably transfected with TLR-3, TLR-7, or TLR-8, we determined the patterns of RNA-specific TLR activation by in vitro transcripts of all of the known murine and human Y RNAs. Next, 8–10-week-old female mice were exposed to a single 50-μg subcutaneous injection of mouse Y1 or mouse Y3 RNA, and the effects were observed.
Y RNA family members differed in their TLR reactivities. Both human and mouse Y3 RNAs, but not other human or mouse Y RNAs, prominently induced TLR-3 activation. Although most human and mouse Y RNAs activated TLR-7 efficiently, mouse Y3 RNA and human Y5 RNA did not. Single subcutaneous injections of mice with either mouse Y1 RNA or mouse Y3 RNA induced or inhibited lymphoid infiltrates in different target organs based on the Y RNA and TLR status of the mouse used. Mouse Y1 RNA induced kidney lesions in TLR-3–intact mice but not in TLR-3–knockout mice. In contrast, mouse Y3 RNA treatment was associated with nephritis in TLR-3–knockout mice but not in TLR-3–intact mice. Sialoadenitis developed in untreated TLR-3−/− mice and in TLR-3−/− mice treated with mouse Y3 RNA, but sialoadenitis was not present in TLR-3−/− mice treated with mouse Y1 RNA.
Y RNAs can induce innate immune responses and influence clinical manifestations of autoimmunity, suggesting that they are relevant to syndromes of anti-Ro autoimmunity. Distinct patterns of tissue targeting can be seen after exposure to different Y RNAs, in a pattern that correlates with the innate immune signals they induce. Thus, the balance of innate immune signals induced by exposure to endogenous Y RNAs may help determine the nature of the clinical syndrome in anti-Ro autoimmunity.
The ability of RNP autoantigens to bind to RNA molecules with innate immune stimulatory properties has been advanced as a hypothesis for the high prevalence of anti-RNP responses in the setting of systemic autoimmunity (1). The Ro/SSA RNP autoantigen, one of the most prevalent and best characterized target antigens in systemic autoimmunity, represents an important avenue for exploring this hypothesis. Ro-associated autoimmunity also typifies the degree of variability in clinical syndromes and end organ targeting that can exist with antibodies to a single well-defined autoantigen. Ro is a prominent autoantigen targeted in cases of Sjögren's syndrome (SS), systemic lupus, subacute lupus, scleroderma, mixed connective tissue disease (MCTD), and primary biliary cirrhosis (2–5). No satisfactory explanation exists for the wide clinical variability associated with anti-Ro antibodies.
The Ro 60 protein directly binds RNA molecules through a typical RNA recognition motif (2). The substantial majority of the protein exists in vivo bound to a series of small cytoplasmic RNAs called the Y RNAs (6). In humans and nonrodent mammals, Y1, Y3, Y4, and Y5 RNAs exist, while in rodents, only Y1 and Y3 RNAs have been identified (7). Y RNA molecules are strongly conserved phylogenetically with regard to sequence, doubled-stranded RNA motifs, and overall structure (8).
Y RNAs have been implicated in anti-Ro autoimmunity, by studies showing autoantibodies directed against Ro–Y RNA complexes (9) and by epitope mapping showing that immunodominant anti-Ro responses are specific for the Y RNA–binding region in a conformationally sensitive manner (10). Moreover, other Y RNA–binding partners are autoantigens. La/SSB has been reported to directly bind Y RNAs as part of a macromolecule with Ro 60 (9). Furthermore, when a series of novel proteins were identified based on Y1 RNA binding, all of them were found to be lupus autoantigens (11). Antibodies directly targeting Y5 RNA in patients with Ro-positive autoimmune disease have also been described (12). Recently, 2 groups of investigators observed Toll-like receptor (TLR)–inducing activity in Y RNA–containing Ro particles derived from living cells (13, 14), although the specific effects of individual Y RNAs on either TLR activation or clinical expression of disease were not described.
In this study, we examined the in vitro effects of human and mouse Y RNA molecules on TLR-3, TLR-7, and TLR-8. We observed that Y RNAs are similar to U1 RNA, another autoantigen-associated RNA, in terms of their efficiency of activation of TLRs, but that the particular TLRs activated by the Y RNAs are neither the same as those activated by U1 RNA nor are they uniform across all of the Y RNAs. The clinical effects of a single subcutaneous dose of mouse Y RNA in experimental mice were reminiscent of the effects of exposure to U1 RNA that we previously reported (15), in terms of the impact on end organ manifestations of autoimmunity. However, the patterns of end organ effects differed between U1 RNA and the Y RNAs and also differed between individual Y RNAs with different TLR stimulation profiles. These results introduce the possibility that variations in the clinical manifestations of anti-Ro–associated syndromes could be attributable to differential innate immune activation by Y RNAs.
Plasmids encoding the Y RNA sequences for human Y1, Y3, Y4, and Y5 as well as mouse Y1 and Y3 RNA were a gift from Dr. Darise Farris and Dr. John Harley (Oklahoma Medical Research Foundation). Each plasmid was resequenced to confirm 100% fidelity to the published sequences of the corresponding Y RNAs. A similar plasmid containing human U1 RNA as a control stimulus was used as previously described (16). RNAs were produced by in vitro transcription of SP64 plasmids containing inserts corresponding to the sequences of the human and mouse RNAs.
Transcripts of each of the RNAs were purified using a standard protocol. Plasmid DNA was eliminated with TURBO DNase treatment, according to the manufacturer's instructions (Ambion, Austin, TX). The RNAs were extracted with phenol–chloroform followed by chloroform, then ethanol precipitated. The resulting products were resuspended in sterile RNase-free and DNase-free diethyl pyrocarbonate–treated water and stored at −80°C until validated and used. Each RNA was validated by nonreducing agarose gels to contain only RNA of the predicted molecular weight, without multimerization. Samples were subjected to RNase and DNase treatments to confirm that the bands visualized were RNase-sensitive and DNase-resistant, in addition to undergoing tests for endotoxin contamination.
HEK 293 cell lines stably transfected to express either TLR-3, TLR-7, or TLR-8 were used to measure the TLR-inducing activities of the test RNAs, as previously described (15). Assays were performed in which cells were transiently transfected with a luciferase reporter gene linked to either an NF-κB or an interferon-β (IFNβ) promoter (17). Twenty-four hours after transient transfection, 106 cells were exposed to U1 RNA or control stimuli for 4 hours at 37°C, allowed to react with Steady-Glo stable luciferase substrate (Promega, Madison, WI), and assayed within 30 minutes in duplicate for luminescence. Additionally, the same TLR–stably transfected cell lines in the absence of luciferase reporter gene transient transfection were incubated for 24 hours under standard culture conditions with the test RNAs or control stimuli and assayed for interleukin-8 (IL-8) secretion into supernatants by enzyme-linked immunosorbent assay (ELISA; R&D Systems, Minneapolis, MN) (18).
The human TLR-3– and TLR-5–expressing endometrial cell line RL-95-2 (henceforth referred to as RL-95) (American Type Culture Collection, Manassas, VA) was also used to test for TLR-3 reactivity, as previously described (16). Briefly, confluent cell cultures were exposed to test and control RNAs for 24 hours under standard culture conditions, then supernatants were removed and assayed for IL-6 using an ELISA (R&D Systems).
B6 × 129 mice transgenic for the extracellular domain of HLA–DR4 (DR4 mice) and B6 × 129 mice deficient in TLR-3 (TLR-3−/− mice) were used as previously described (15). We chose to use DR4 mice for this study despite the fact that primary SS and anti-Ro responses have been linked to alternate major histocompatibility complex (MHC) haplotypes (19–21), so that the responses of the Y RNAs would be most comparable with our previous work using U1 RNA (15). In making this choice, we were also aware that anti-Ro antibodies develop in fully 25% of HLA–DR4–positive patients with MCTD (4).
Female 8–10-week-old mice were immunized once subcutaneously with 50 μg of the indicated stimulus in sterile phosphate buffered saline, and were then followed up with monthly urinalyses and blood withdrawals for 2 months, as previously described (15). In mice receiving Freund's complete adjuvant (CFA), 50 μl of a standardized preparation (Difco, Detroit, MI) was used as previously described (15). Control mice receiving no stimulus underwent an identical sham procedure and identical clinical followup, but they received no subcutaneous injection. Two months after immunization, the mice were killed, and tissue was processed for histologic evaluation as previously described (15). In addition to renal and lung assessments, as previously reported, salivary glands were also obtained at the time that the mice were killed and were graded for the number of foci of inflammation among at least 50 cells/4 mm2 of specimen. Mice with a focus score >1 (indicating >1 focus of inflammation among at least 50 cells/4 mm2) were prospectively designated as having sialoadenitis.
Mice receiving no stimulus underwent an identical sham immunization procedure (without a subcutaneous injection), were housed together with the RNA-injected mice, and received identical clinical followup care. All mouse studies were approved by the Institutional Animal Care and Use Committee at our center, and all animals were housed in American Association for Accreditation of Laboratory Animal Care–approved facilities. None of the animals described here have been previously reported.
For all mice, serum samples were obtained at the time of study entry and when the mice were killed. Immunoblots were performed on the latter serum samples from all mice at 1:1,000 dilutions, using intact and apoptotic Jurkat cells as the substrates, as previously described (22).
All numeric results are reported as the mean ± SD. Statistical analysis was performed using Prism 3.0 software (GraphPad Software, San Diego, CA). Categorical variables were compared using Fisher's exact test, and mean values were compared using Student's unpaired 2-tailed t-test.
We produced transcripts of the human Y1, Y3, Y4, and Y5 RNAs and the mouse Y1 and Y3 RNAs by in vitro transcription and observed uniform bands of the predicted molecular weights for noncomplexed Y RNA species on agarose gels (results not shown). Each RNA was sensitive to RNase but not DNase, and significant endotoxin contamination was not observed (results not shown).
Using TLR-3–expressing RL-95 cells or, similarly, TLR-3–transgenic 293 cells (results not shown), we observed that U1 RNA and poly(I-C) were potent TLR activators, as previously described (15, 16). Both human and mouse Y3 RNA also clearly activated TLR-3–expressing cells above baseline levels, although somewhat less efficiently than equal doses of U1 RNA (Figure 1). TLR-3 activation by the other human and mouse Y RNAs was minimal.
The human Y RNAs induced activation of TLR-7–expressing 293 cells similar to that seen with milligram-per-milligram doses of U1 RNA, as reflected by expression levels of an IFNβ promoter–linked luciferase reporter construct, with human Y5 RNA showing the lowest mean level of activation (Figure 2A). Although murine Y1 RNA was similar to human Y1 RNA in terms of TLR-7 induction, mouse Y3 RNA was a much less efficient inducer of TLR-7 (Figure 2B). As reflected by the differences in the magnitude of the human Y1 RNA signal (shown in Figures 2A and B), we observed some variation, from test to test, in the magnitude of the signals in our 293 cell experiments; however, the relative magnitude of responses to each test stimulus remained consistent. Similar results were seen with an NF-κB promoter–linked reporter construct and by measuring IL-8 production (results not shown).
TLR-8–expressing 293 cells were not activated by U1 RNA (Figure 3A). Increased activation of an IFNβ promoter–linked reporter construct in TLR-8–expressing 293 cells was seen with human Y3 RNA, mouse Y1 RNA, and mouse Y3 RNA (P = 0.01, P = 0.02, and P = 0.01, respectively, versus U1 RNA, by Student's t-test), while human Y1 RNA failed to show a significant effect (Figure 3A). Human Y4 RNA and human Y5 RNA also did not appear to have significant TLR-8 stimulatory effects, using the IFNβ promoter reporter (results not shown). However, using an NF-κB–linked reporter construct in the same TLR-8–expressing 293 cells, exposure to human Y3 RNA, human Y4 RNA, and particularly human Y5 RNA showed trends toward increased activation compared with U1 RNA (Figure 3B).
In assays for TLR-3, TLR-7, and TLR-8 activity, results were consistent regardless of whether the Y RNAs were pretreated with lipofectin to inhibit RNase-induced degradation and to facilitate transport of RNAs across the plasma membrane (data not shown). Substantial amounts of undigested RNA could be recovered from culture supernatants after 4 hours of incubation of unprotected RNA with either RL-95 cells or TLR-expressing 293 cells, for each of the Y RNAs tested (data not shown). When the SP64 plasmid from which the RNAs were transcribed was subjected to transcription and purification in the absence of an RNA-coding insert, no observable product remained after standard purification, and the use of this “mock transcribed” product failed to induce TLR-3, TLR-7, or TLR-8 activity in any of our assays (data not shown).
To confirm our in vitro observations that Y RNA molecules are capable of stimulating TLRs, we subcutaneously injected DR4 mice and TLR-3−/− mice with mouse Y RNAs or U1 RNA. Experiments were not performed using human Y RNAs, because these are not native mouse structures. Two months after being injected, the mice were killed and analyzed by blinded readers for the presence of significant salivary lymphoid foci (at least 1 foci of >50 inflammatory cells/4 mm2 of tissue) and for the presence of pulmonary interstitial infiltrates or glomerulonephritis, as previously described (15). Among the DR4 mice, a limited degree of salivary gland inflammation was seen in mice that were not immunized (Table 1). Exposure to either mouse Y3 RNA or U1 RNA was associated with the absence of significant salivary gland lesions and was not associated with nephritis (Table 1). These results differed from those obtained with mouse Y1 RNA, which induced a moderate rate of renal disease (as documented by the presence of glomerular hypercellularity, red cell casts, and/or proteinuria [results not shown]), although also in the absence of significant salivary gland infiltrates. Pulmonary infiltrates were not seen in the untreated mice or in either the mouse Y1– or mouse Y3–treated animals but were present in 2 of 5 U1 RNA–treated mice, consistent with our previous findings (15). In a DR4 mouse treated with 50 μl of CFA, no lung, renal, or salivary lesions developed.
|Mouse strain/stimulus||Salivary lesions||Salivary gland focus score, mean ± SD||Nephritis|
|Untreated||1/3||1.3 ± 0.9||0/3|
|Mouse Y1||0/5||0.4 ± 0.2||2/5|
|Mouse Y3||0/4||0 ± 0||0/4|
|U1 RNA||0/5||0.4 ± 0.2||0/5|
|Untreated||2/2†||4.5 ± 0.5||0/2|
|Mouse Y1||0/5||0.2 ± 0.2‡||0/5|
|Mouse Y3||2/5||0.8 ± 0.5§||2/5|
|U1 RNA||0/2||0 ± 0||2/2†|
Among the TLR-3−/− mice, marked salivary gland infiltrates were seen in untreated mice (Figure 4 and Table 1), without lung or kidney lesions. Because the number of untreated TLR-3−/− mice in the experimental cohort was small, we subsequently confirmed the presence of at least 2 foci of inflammation, in the absence of lung or kidney disease, in 5 of 5 additional untreated female TLR-3−/− mice. Because these 5 animals were not part of a group contemporaneous with the other experimental mice, they were excluded from statistical analyses. Nonetheless, exposure to mouse Y1 RNA in TLR-3−/− mice clearly was associated with a reduction in the rate of sialoadenitis compared with that in nonimmunized mice (P = 0.048 by Fisher's exact test) as well as a reduction in the severity of salivary gland lesions (P = 0.0002 by Student's t-test). Sialoadenitis was not seen after U1 RNA treatment of TLR-3−/− mice. In the mouse Y3 RNA–treated TLR-3−/− mice, sialoadenitis persisted, although the level of severity (as reflected by the mean ± SD salivary gland focus score) decreased from 4.5 ± 0.5 to 0.8 ± 0.5 (P = 0.008 by Student's t-test). Consistent with our previous results (15), U1 RNA treatment was associated with moderate levels of nephritis (Table 1). To our surprise, mouse Y1–treated TLR-3−/− mice did not develop renal lesions (P = 0.048 versus U1 RNA–treated mice, by Fisher's exact test), but a modest rate of nephritis was observed among mouse Y3 RNA–treated mice (Table 1). Neither mouse Y1 RNA nor mouse Y3 RNA induced lung disease in TLR-3−/− mice. In 2 TLR-3−/− mice treated with 50 μl of CFA, no lung, renal, or salivary lesions developed.
When assayed by immunoblot versus Jurkat cell lysates, sera from DR4 mice treated with U1 RNA, Y RNA, or CFA showed no evidence of anti-Ro or anti-La antibodies and minimal if any other detectable autoantibodies—even those animals with clinical manifestations of autoimmunity (Figure 5A). Immunoblots (not shown) using control versus apoptotic cell lysates confirmed that antifodrin antibodies were not observed (23). In the TLR-3−/− mice, induction of autoantibodies was also minimal. No mice developed anti-Ro or anti-La responses detectable by immunoblot. However, one of the TLR-3−/− mice treated with mouse Y3 RNA that developed renal disease and showed the absence of salivary gland disease did develop anti–U1 RNP antibodies (Figure 5B). Several additional antibody specificities were also present in this mouse.
Recent reports by other investigators have shown that Y RNA particles extracted from cells are capable of activating dendritic cells through TLRs, including TLR-7 (13, 14). This study, however, is the first to show that different Y RNAs vary in their TLR stimulatory capacity, that exposure to Y RNAs alone is capable of influencing autoimmunity in an animal model, and that different Y RNAs induce different patterns of end organ involvement.
In terms of differential TLR activation by Y RNAs, the ability of only human and murine Y3 RNAs to prominently induce TLR-3 was most striking. In mice (but not in humans), Y3 RNA also failed to prominently activate TLR-7. Thus, in contrast to U1 RNA, which activates both TLR-3 and TLR-7, the murine Y1 and Y3 RNAs were relatively selective inducers of TLR-7 and TLR-3, respectively.
Although TLR recognition of self determinants might be predicted to be a universally adverse event in terms of the risk of development of autoimmunity, our results, in accordance with those reported by Christensen and colleagues (24), suggest a more nuanced role of TLRs in self recognition and autoimmunity. When a self molecule is recognized by a TLR, the effects depend on which TLRs are activated. In mice expressing both TLR-3 and TLR-7, the TLR-7–inducing mouse Y1 RNA induced nephritis, but the TLR-3–inducing mouse Y3 RNA did not. Furthermore, the effects of TLR activation appear to have tissue specificity: RNA-induced activation of some TLRs induced increased inflammation in some tissues but not others. TLR-3– and TLR-7– inducing U1 RNA induced lung disease, but TLR-7–specific mouse Y1 RNA did not. Moreover, under at least some circumstances in some tissues, TLR activation can lead to a reduction rather than an augmentation of inflammation. Thus, spontaneous sialadenitis could be diminished by either TLR-3 or TLR-7 agonists.
We previously observed that activation of U1 RNA in the context of autoantigen immunization can induce autoimmune disease with clinical features that differ depending on the balance of TLR-3 versus TLR-7 signaling (15). We and other investigators have linked TLR-7 signaling to lupus manifestations including nephritis (15, 25), while TLR-3 signaling was linked to MCTD-like lung disease (15). Among patients with anti-Ro responses, multiple distinct clinical syndromes have also been described, some of which encompass features reminiscent of those attributed to TLR-7 and TLR-3 signaling in response to U1 RNA. These include associations of anti-Ro responses with interstitial lung disease similar to those seen in human MCTD and associations with typical manifestations of systemic lupus erythematosus inclusive of nephritis (26). Consistent with our previous results, only mouse Y1 RNA, a stimulator of TLR-7 but not TLR-3, induced nephritis in TLR-3–intact DR4 mice. However, the absence of nephritis in the mouse Y1 RNA–treated TLR-3−/− mice suggests that other mechanisms in addition to or permissive of TLR-7 activation must be needed to induce lupus-like autoimmunity. For example, the inability of mouse Y1 RNA to induce an apparent autoantibody response in TLR-3−/− mice may have prevented the formation of mouse Y1 RNA–containing immune complexes needed to facilitate proimmune TLR-7 signaling (14).
The fact that mouse Y3 RNA did not induce lung disease in TLR-3–intact mice, as we have seen for U1 RNA, likewise suggests that the TLR-3 signal alone does not appear to be sufficient for the development of pulmonary disease. The fact that human Y3 RNA shares both TLR-3 and TLR-7 stimulation profiles with U1 RNA suggests that inflammation mediated by endogenous Y3 RNA in humans may have more similarities to that mediated by U1 RNA than that mediated by endogenous Y3 RNA in mice. Alternatively, the ability of mouse Y3 RNA to induce nephritis in TLR-3−/− mice implies that mouse Y3 RNA may induce additional proinflammatory systems as well as TLR-3. Other innate immune sensing systems that, like TLR-3, are specific for double-stranded RNA have been described, such as the protein kinase R and retinoic acid inducible gene I RNA helicase systems (17). Studies of these pathways may also be relevant to an understanding of the effects of Y RNAs on clinical autoimmunity.
The ability of Y RNAs to differentially activate multiple TLRs and to differentially modulate the development of salivary gland lymphocytic foci raises the question of whether innate immune activation plays a role in the pathogenesis of SS, the clinical syndrome most closely associated with autoimmune responses to both the Ro and La Y RNA binding proteins. We observed that mouse Y1 RNA and U1 RNA were able to inhibit spontaneous sialoadenitis in our TLR-3−/− test animals, suggesting that TLR-7 signals may be able to functionally replace tonic TLR-3 signals to inhibit lymphocytic infiltration of the salivary glands.
A model system expressing additional MHC haplotypes known to favor the development of anti-Ro antibodies would be needed to explore the reasons why the autoantibody responses we observed did not include anti-Ro or anti-La. Of note, anti-Ro and anti-La antibodies generally have not been observed in murine models of spontaneous SS (27–30) and have been induced along with sialoadenitis in mice only after repeated hyperimmunization with a non–TLR-3 or non–TLR-7 stimulatory adjuvant (31). It will be of interest in future studies to determine whether any of the antigens targeted in the U1 RNP autoantibody–reactive TLR-3−/− mouse we observed are also Y RNA–binding lupus autoantigens as identified by Fabini et al (11). The possibility that Y RNA stimulation, even in the absence of apparent Ro or La immune responses, can give rise to U1 RNP antibodies raises additional questions about patterns of intermolecular spreading observed between the mouse Y RNA–associated Ro and U1 RNA–associated U1 RNP macromolecules (32).
Immune responses to Y5 RNA–conjugated Ro 60–associated complexes are reported to be frequent in humans with anti-Ro responses (33). The fact that Y5 RNA is a relatively weak activator of TLR-7 and TLR-3 might account for the tendency for anti-Ro immune responses to persist for extended periods of time in many patients before flares of clinically significant autoimmune disease emerge, in contrast to U1 RNP responses that are closely temporally tied to clinical disease expression (34). Studies of TLR expression in salivary glands have shown up-regulated expression of TLR-8 and TLR-9 (not TLR-3 or TLR-7) in patients compared with controls (35). If Y5 RNA induces a TLR-8 signal in vivo in the absence of TLR-3 and TLR-7 signals that appear to down-regulate sialoadenitis, this could also be linked to the induction of human SS. A model system that (unlike the mice described here) expresses TLR-8 with human-like activation properties will be required to explore whether Y RNAs acting on TLR-8 induce a Ro-associated clinical syndrome (36).
The Y1 RNA has been reported to be a major Y RNA detected in intact cells (37). Thus, our results suggest a mechanism by which Ro 60–knockout animals might be at increased risk of lupus-like disease (38). If, under normal circumstances, Ro 60 acts as a sink for intracellular Y1 RNA, then the absence of Ro 60 could lead to excessive exposure of Y1 RNA. This could lead to excess TLR-7 stimulation and hence lupus-like autoimmunity such as the nephritis seen in our mouse Y1 RNA–treated DR4 mice.
In summary, we observed that individual Y RNAs have distinct patterns of innate immunity induction, and that exposure to specific Y RNAs has the potential to induce different patterns of autoimmune tissue targeting. The results of this study suggest that, in addition to such factors as the differential tissue expression of autoantigens and the fine specificity of autoimmune responses at the B cell and T cell levels, innate immunity signals may have substantial effects on the patterns of clinical expression of autoimmune syndromes.
Dr. Greidinger had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study design. Greidinger, Nassiri, Barber, Hoffman.
Acquisition of data. Greidinger, Zang, Martinez, Jaimes, Nassiri, Bejarano, Barber.
Analysis and interpretation of data. Greidinger, Zang, Martinez, Jaimes, Nassiri, Bejarano, Hoffman.
Manuscript preparation. Greidinger, Martinez, Hoffman.
Statistical analysis. Greidinger.
We thank Drs. Darise Farris and John Harley for their generous contributions to this work.