Blood-Borne RNA Correlates with Disease Activity and IFN-Stimulated Gene Expression in Systemic Lupus Erythematosus

This article is featured in In This Issue, p.2555

Systemic lupus erythematosus (SLE) is a chronic autoimmune disease that is characterized by diverse disease manifestations impacting multiple organs, including the skin, CNS, joints, vasculature, and kidneys (1). A central component of the disease is the presence of autoantibodies directed against a variety of nuclear Ags (2, 3). The normal process of cell death and turnover results in accumulation of circulating nuclear Ags in SLE as a result of impaired clearance of the cellular debris, and this process is thought to be involved in the pathogenesis of the disease (4–6). Autoantibodies complexed with RNA-containing Ags lead to the formation of immune complexes (ICs) that are known to activate immune cell function, disrupt vascular integrity, and cause organ dysfunction (7–10).

Blood cells isolated from SLE patients overexpress many IFN-inducible genes (11–13). The presence of this gene expression signature has led to the widely accepted hypothesis that IFN is overproduced and contributes to the systemic inflammation that is characteristic of SLE. A recent analysis of the SLE patient IFN gene signature suggested that the expression profile is a composite of genes induced by IFN-α, INF-β, and IFN-γ (14).

Type 1 IFNs function as a key part of host innate immunity. Viruses and other pathogens contain elements that are recognized by host cell pattern recognition receptors that, when ligated, trigger IFN production (15). Among these receptors are nucleic acid–sensing TLRs that recognize distinct nucleic acid motifs, with TLR3- and TLR7- recognizing RNA molecules (16, 17). A large body of data from animal studies, as well as human studies, demonstrates the importance of TLR7 in activating IFN production in SLE. For example, in genetic studies of several mouse strains, TLR7 is the sole requirement for accelerated autoimmunity (18, 19). Additionally, human genetic studies in SLE patients similarly revealed the central role of TLR7 and IFN production in the disease (20, 21). These TLRs are localized within endosomal compartments where they are sequestered from host-derived ligands (16, 22). However, host cell nucleic acids can activate nucleic acid–sensing TLRs following internalization into immune system cells (23–27). One candidate ligand, U1 RNA, is a noncoding ubiquitously expressed transcript that functions as a component of the spliceosome, a ribonucleoprotein assembly involved in the processing of pre-mRNAs, and is known to bind to multiple protein Ags recognized by SLE patient autoantibodies (28, 29). When bound to an Ab, the RNA-containing Ag can be internalized via receptor-mediated endocytosis facilitated by Fc, complement, and BCRs (24, 30, 31), resulting in delivery of the U1-RNA to endosomal TLRs and activation of downstream inflammatory pathways. Numerous studies demonstrated the ability of ICs to effectively activate TLRs. For example, mixing SLE patient sera with necrotic or apoptotic cell extracts results in the production of IFN by healthy donor blood cells in an IgG-dependent and RNase-sensitive manner (32, 33). ICs also activate the NLRP3 inflammasome in human monocytes, leading to production of IL-1β (34) and activation of the oxidative burst response in neutrophils (35, 36).

Assays for detecting circulating ICs were described, but methodologies for quantifying specific RNAs associated with these complexes were not (37–39). In this study, we provide the first direct measurement, to our knowledge, of RNA molecules complexed with SLE patient autoantibodies. The target RNAs, U1 and Y1, were recovered in association with protein A–bound Igs from a broad cross-section of 228 SLE patients. The amount of U1 and Y1 RNA associated with the captured Abs was significantly elevated in SLE compared with healthy volunteers (HVs). Elevated levels of Ab-bound RNA correlated with disease activity and the expression of IFN-inducible genes. Therefore, this study establishes a link between these blood-borne RNA molecules and SLE disease manifestations, providing the rationale for a therapeutic approach aimed at reducing extracellular RNA.

A total of 228 subjects fulfilling 4 of the 11 American College of Rheumatology criteria for SLE was enrolled from six clinical sites in the United States. The only requirement for entry into this study was meeting SLE Classification Criteria; there were no exclusion criteria. A seventh site (PlasmaLab International) recruited 20 HVs. The study protocol was reviewed and approved by the Western Institutional Review Board (WIRB Pro. No. 20121576) for all sites except one, which used its own institutional review board. Prior to sample donation, each donor signed an institutional review board–approved informed consent form allowing sample collection. Principal Investigators also completed a case report form that included an assessment of concomitant medications and disease activity (Safety of Estrogens in Lupus Erythematosus National Assessment [SELENA]–SLE Disease Activity Index [SLEDAI]) (40).

A Tempus blood RNA tube (Life Technologies) and a heparin Vacutainer tube were obtained from each subject. Plasma was prepared at each of the six clinical sites shortly after each blood draw by a single centrifugation in a clinical tabletop centrifuge, and the plasma samples were transferred to tubes containing recombinant RNase inhibitor, resulting in a final concentration of 1000 U/ml (RNasin Plus; Promega). RNase inhibition is crucial to prevent degradation of plasma RNA by blood-borne RNase activity during processing (41). The resulting plasma and Tempus tube samples were stored at −80 and −20°C, respectively.

Protein A Plus Agarose (Thermo Scientific Pierce) was used to capture total IgG from plasma. For each sample, three replicate Protein A and a control capture (performed using unconjugated agarose beads) were conducted in parallel. One-milliliter aliquots of each thawed plasma sample were centrifuged at 16,000 × g for 5 min at 4°C. A total of 200 μl of the resulting supernatants was transferred into each of three 15-ml tubes containing 4.4 ml of PBS, 0.05% Tween-20, and 400 μl of Protein A Plus Agarose (50% suspension) and a single 15-ml tube containing unconjugated agarose beads in the same buffer. Tubes were incubated for 1 h at 4°C with gentle mixing, after which the mixtures were transferred to disposable columns, and resins were collected. Resins were washed (gravity flow) with three 5-ml volumes of PBS, 0.05% Tween, suspended in 1 ml of PBS, 0.05% Tween, transferred to 1.7-ml microcentrifuge tubes, and collected by brief centrifugation; the resulting supernatants were discarded. RNA was recovered from the washed beads using miRNeasy Micro Kits (QIAGEN), per the manufacturer’s instructions. RNA was eluted in 15 μl of RNase-free water and stored at −80°C.

cDNA synthesis was performed using High-Capacity cDNA Reverse Transcription Kits (Life Technologies); 4 μl of each RNA isolate was reverse transcribed in a 20-μl reaction, according to the manufacturer’s instructions. A control reaction lacking reverse-transcriptase was conducted for each plasma sample.

Levels of U1 and Y1 RNA were measured by quantitative real-time PCR (qPCR) using hydrolysis probe-based detection. All primers and probes were synthesized by Integrated DNA Technologies (Coralville, IA). For detection of U1, amplification primers were 5′-CGGAGTGCAATGGATAAG-3′ and 5′-CTGGCAGGGGAGATAC-3′, and the detection probe was 5′-/56-FAM/CCTGGGAAA/ZEN/ACCACCTTCGTG/3-IowaBlackFQ/-3′. Primer sequences for Y1 were 5′-GGGAAAGAGTAGAACAAGGAGT-3′ and 5′-TGGTCCGAAGGTAGTGAGTTA-3′, and 5′-/5HEX/TCGATCTGT/ZEN/AACTGACTGTGAACAATC/3-IowaBlackFQ/-3′ was used as the detection probe.

For qPCR, 2 μl of cDNA was added to 25-μl reactions containing 1× Brilliant Multiplex Master Mix (Agilent Technologies), 30 nM ROX reference dye, 125 nM U1 and Y1 primers, and 62.5 nM U1 and Y1 probes. Incubation conditions were 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min in an Agilent Technologies MX3005P qPCR instrument. A standard curve containing from 10¹ to 10⁷ copies of linearized plasmid DNA containing U1 and Y1 sequences was included in each qPCR run. Ct values were obtained using MxPro software (Agilent Technologies), and standard curves were used to convert Ct values into copy numbers. For each isolate, the background signal detected in cDNA derived from the control bead capture was subtracted from the signal derived from protein A capture. The resulting specific signal was expressed as 1000 copies/ml of input plasma (kcopies/ml). Triplicate U1 and Y1 measurements displayed median coefficients of variation of 18 and 13%, respectively, over the 228 samples analyzed (data not shown).

RNA isolation, quality control, and Fluidigm Dynamic Array (48 × 48) qPCR analysis were performed by EA Genomics Lab of Q² Solutions. The array measured 186 transcripts; 6 of these were included as reference genes for normalization. The remaining 180 genes were selected based on meta-analyses of published and unpublished SLE clinical study gene-expression datasets compiled by the Benaroya Research Institute (13, 42–44). Genes that differentiated healthy individuals from SLE patients or showed a link to disease severity were selected. From this list of modules, representative genes were selected for inclusion in the array. The array was dominated by the presence of known IFN-responsive genes. In addition, genes associated with neutrophil, B cell, and myeloid modules were included. Each RNA isolate was assayed twice, and resulting Ct values were inverted by taking the number of effective cycles in the assay and subtracting the Ct value reported for the gene. As a result, reported Ct values correlate positively with quantity. ΔCt (or normalized) values are corrected relative to the measured set of reference genes. Each target Ct value was normalized to create ΔCt values relative to the average of inverted Ct values of the six reference genes; thus, ΔCt values are also positively correlated with quantity. The IFN signature metric (ISM) score was calculated for each subject as follows: mean Ct value of CMPK2, EPSTI, and HERC5 − mean Ct value of the six reference genes included in the dynamic array (45). This score was multiplied by −1 to give the relative log₂-scaled expression the correct directionality. Subjects with ISM scores > 1.0 were classified as ISM-high.

Autoantibody analysis was conducted by the Department of Laboratory Medicine at the University of Washington. Anti-dsDNA levels were measured using a commercial ELISA (Bio-Rad), with modifications to the manufacturer’s protocol designed to reduce background. A panel of common autoantibodies was measured using Bio-Plex 2200 kits (Bio-Rad). Criteria for scoring a sample positive were based on historical experience testing samples from HVs. For anti-dsDNA, a value > 14 U/ml was considered positive. The following Ab index (AI) values were considered positive: SSA/Ro52 (AI > 1.2), SSA/Ro60 (AI > 1.2), SSB/La (AI > 6.0), Sm (AI > 3.0), SmRNP (AI > 5.0), U1 RNP 68 (AI > 5.0), and U1 RNP A (AI > 5.0). For samples above the normal reference range, absolute AI values were determined following dilution.

Comparisons of median levels of circulating biomarkers and disease status (HV versus SLE) were assessed with a Mann–Whitney U test. The correlations between autoantibodies and autoantibody-bound RNA and SLEDAI score were quantified using Spearman’s rank correlation coefficient (ρ), whereas the correlations among autoantibodies and autoantibody-bound RNA were assessed using the Pearson product-moment correlation coefficient. The Fisher exact test was used to test for association between categorical variables.

The study enrolled a broad cross-section of SLE subjects with various disease activity and SLE manifestations that was characteristic of private practice and academic medical center settings. The SLE patient and HV cohorts consisted of 90% females, with average ages of 45 and 41 y, respectively. The average SELENA-SLEDAI score of the SLE cohort was 3.89; 29% of the subjects had a SELENA-SLEDAI score ≥ 6, and 29% had inactive disease, with a SLEDAI score of 0 (Supplemental Table I).

An assay for measuring the amount of RNA associated with Abs was developed to analyze patients with respect to this potential disease mediator. Total plasma IgG was captured with Protein A, including free Abs and ICs. Specific primer/probe sets were designed to allow reverse-transcription qPCR (RT-qPCR) measurement of U1 and Y1 RNA simultaneously. Each of these small noncoding RNAs is known to associate with protein Ags recognized by SLE patient autoantibodies (29, 46).

The quantities of U1 and Y1 RNA recovered in association with Abs isolated from HVs and SLE patients are shown in Fig. 1. Many HVs displayed detectable Ab-bound U1 RNA, and Ab-bound Y1 RNA was detected from all 20 HV samples. RNA measurements from HVs were specific, because levels of U1 or Y1 transcripts recovered from the same plasma samples using Protein A–free agarose beads were 8- and 200-fold less for U1 and Y1, respectively (data not shown). Omission of the reverse-transcription step resulted in greatly attenuated signals, confirming RNA dependence. SLE patients were scored positive if they had U1 or Y1 RNA 2 SDs above the mean level observed in HVs (2200 kcopies/ml [U1] and 17,000 kcopies/ml [Y1]). U1 and Y1 RNA levels were significantly elevated in SLE patients relative to HVs (Fig. 1A, 1B). Levels of U1 RNA were elevated in a greater percentage of SLE patients (59%) than were Y1 RNA levels (29%), and the median difference between SLE and healthy donors was greater for U1 than for Y1 RNA. To determine whether Abs recovered from rheumatoid arthritis (RA) patients were complexed with U1 or Y1 RNA, plasma samples from five patients were subjected to Ab capture and RT-qPCR analysis. Levels of U1 and Y1 RNA recovered from each of the RA patients were comparable to levels observed in HVs (Fig. 1C).

To analyze the daily variability in circulating RNA, a longitudinal substudy was undertaken by obtaining measurements on five consecutive days. The observed fluctuations in autoantibody-bound U1 and Y1 RNA generally varied in a 2–3-fold range over the 5-d period, with subject 4 showing a 1-d 9-fold change (Fig. 2A). The U1 RNA levels remained above the HV threshold during the 5-d study. To assess the day-to-day reproducibility of Ab capture and quantitation from the same sample, repeat measurements of eight plasma samples from the longitudinal study were performed. For all eight samples, the second independent measurements yielded levels of U1 and Y1 RNA in close agreement with the original determination (Fig. 2B). The variability in the repeat measurements averaged 16% for U1 and 13% for Y1 measurements among the group of eight samples.

Examination of the relationship between autoantibody-bound RNA and disease activity demonstrated a statistically significant correlation between each of the two RNAs measured and disease activity, as measured by SLEDAI score (Fig. 3). In contrast, analysis of the seven autoantibodies tested in the study did not show the same strength of correlation with SLEDAI as did the RNA molecules (Supplemental Fig. 1). Of the 228 total SLE subjects in the study, 97 were seronegative for the presence of the tested RNA autoantibodies. Interestingly, despite being negative for the RNA autoantibodies, >40% of these subjects had elevated levels of autoantibody-bound RNA and showed significant associations with IFN gene-expression status and SLEDAI score (Fig. 4).

The panel of autoantibodies assessed in this study included four that bind Ags known to complex with U1 RNA (Sm, SmRNP, U1 RNP 68, U1 RNP A) (29) and three that are known to bind to Y1 RNA (SSA/Ro52, SSA/Ro60, SSB/La) (46). Within this group of SLE patients, there was a highly significant positive correlation in the quantities of the two RNAs associated with these autoantibodies (Supplemental Fig. 2A, 2B). Anti-SmRNP displayed the strongest associations with Ab-bound U1 RNA (Table I), whereas anti-SSA/Ro60 displayed the strongest associations with Ab-bound Y1 RNA (Table II). However, there was no correlation between autoantibody titer and the corresponding Ab-bound RNA (Supplemental Fig. 2C, 2D).

RNA was prepared from whole blood obtained from each SLE patient and 20 HVs. The expression level of 180 genes thought to be involved in SLE and 6 housekeeping genes were analyzed by qPCR array. The identity and grouping of the genes in the array are shown in Supplemental Table II. A heat map of normalized Ct values for all 228 SLE patients and 20 HVs is shown in Fig. 5. A group of 55 known IFN-regulated genes clearly separates the subjects into two groups. The low IFN group (green) consists, in large part, of the 20 HVs and those SLE patients lacking autoantibody-bound U1 and Y1 RNA, whereas the high IFN group (red) contains most of the patients positive for autoantibody-bound RNA. In addition, within the IFN high group there are interesting changes to a cluster of genes associated with the neutrophil module that further separates these patients into two fairly distinct subgroups. The subjects’ autoantibody-bound RNA status is shown at the bottom of the heat map. The U1⁺ or Y1⁺ subjects tend to cluster in the IFN-high group.

Based on calculated ISM scores (45), SLE patients in this study were grouped into IFN signature positive (ISM ≥ 1; 64%), and IFN signature negative (ISM < 1; 36%). Elevated levels of autoantibody-bound RNA were significantly associated with the IFN signature positive group (Tables I, II), with U1 RNA showing a stronger association. Also, as noted previously, the association between RNA and ISM status persisted in the subset of SLE patients who were negative for all tested RNA autoantibodies (Fig. 4).

To further investigate the relationship between autoantibody-bound RNA and IFN-regulated gene expression, patients were divided into four groups based on RNA status. Group 1 was negative for U1 and Y1 (U1⁻/Y1⁻); group 2 consisted of all patients positive for U1, regardless of Y1 status (U1⁺); group 3 was U1⁺ and negative for Y1 (U1⁺/Y1⁻), and group 4 was positive for Y1 and negative for U1 (U1⁻/Y1⁺). Volcano plots were constructed for each of these four groups using the corresponding ΔΔCt values (relative to HVs). Comparison of the volcano plots reveals significant differences in the magnitude of gene expression and associated p values among the four groups of SLE patients (Fig. 6). Most striking were the gene-expression changes and p values in the RNA⁻ (U1⁻/Y1⁻) group compared with the three other groups that were RNA⁺. All three groups positive for U1 or Y1 RNA displayed similar increases in gene expression. The decreased statistical significance seen in the U1⁻/Y1⁺ plot likely reflects the small number of subjects (n = 15) in this group. The top 10 upregulated transcripts observed in the U1⁺ group were known IFN-regulated genes (12). Within this group, expression change ranged from 46-fold for IFI27 to 7-fold for OAS3. By comparison, the same genes were only modestly upregulated in the U1⁻/Y1⁻ group (Fig. 7). This group of genes was also the top 10 upregulated genes in the Y1⁺/U1⁻ group and, with the exception of CCL2 replacing OAS3, the top upregulated genes in the U1⁺/Y1⁻ group. Relative expression levels of the 180 transcripts measured were highly significantly correlated between the U1⁺/Y1⁻ and U1⁻/Y1⁺ groups (data not shown).

The inflammatory nature of RNA and its role in activation of the innate immune system in autoimmune diseases is widely recognized (7, 23–27). To our knowledge, the present study is the first to systematically quantitate circulating RNA in a large cross-section of SLE patients. Several compelling findings arise from this work. First, plasma from SLE (but not RA) patients demonstrated elevated levels of Ab-bound U1 and/or Y1 RNA relative to plasma from HVs. Thus, increased Ab-bound RNA appears to be a characteristic specific to SLE. Second, the abundance of Ab-bound RNA correlated with disease activity in this cohort of 228 SLE patients. Finally, the presence of autoantibody-bound RNA is highly correlated with the expression of IFN-stimulated genes; conversely patients lacking these RNA molecules demonstrated gene-expression patterns similar to HVs. Taken together with the well-established ability of RNA to stimulate IFN production, the data suggest the involvement of RNA in driving the IFN pathway and disease activity in SLE.

Analysis of the correlation between the autoantibodies tested in this study and SLEDAI scores revealed that there was not a similar degree of correlation between the presence of these autoantibodies and SLEDAI score. The results suggest that RNA complexed with autoantibodies is more closely associated with disease activity and IFN gene stimulation than the autoantibodies themselves. Given the dynamic nature of cellular apoptosis and turnover relative to the largely consistent levels of RNA autoantibodies, Ag production may be the governing moiety modulating inflammatory pathway activation, with autoantibodies serving as delivery vehicles transporting the RNA into proximity with pattern recognition receptors.

Analysis of blood cell RNA expression using a qPCR dynamic array clearly separated SLE patients and HVs based on their IFN gene-expression profile. Among patients demonstrating upregulation of IFN-responsive transcripts, other genes included in the array showed differential expression patterns, including a set of genes associated with neutrophils that appear to subdivide the IFN⁺ patient group. The presence of RNA complexed with autoantibodies was highly enriched in those patients with elevated expression of IFN-responsive transcripts. Further analysis of the four subgroups of patients positive for U1 and/or Y1 and those negative for these RNA molecules demonstrated a highly statistically significant correlation between the presence of self-antigens containing U1 or Y1 and a group of 10 well-known INF-stimulated genes (Figs. 6, 7).

U1 and Y1 RNAs were analyzed in the current study based on their known association with protein Ags recognized by SLE patient autoantibodies, but it is likely that Abs may be found in association with other RNA species as well (47). The absence of robust expression of IFN-responsive genes in the U1⁻/Y⁻ subgroup suggests that U1 and/or Y1 RNA was the primary driver of IFN gene expression in these SLE patients. Some subjects within this subgroup had elevated levels of autoantibodies, so the absence of IFN-inducible gene expression is not simply a reflection of the lack of RNA-binding capacity.

Although highly significant associations were observed between the presence of elevated Ab-bound RNA and the presence of RNA autoantibodies, nearly half of the SLE patients who were positive for Ab-bound U1 RNA were negative for the well-characterized autoantibodies measured in this study. This finding suggests that these subjects may have autoantibodies directed against other Ags that can associate with U1 RNA, such as the U1C protein (48), or autoantibodies that bind directly to U1 RNA (49). Many protein Ags were detected as targets of SLE patient autoantibodies (50).

There was no correlation observed between Ab titers and the level of Ab-bound RNA. This may reflect the vast excess of autoantibody relative to RNA-containing Ag, because autoantibody concentrations in serum are known to be in the μg/ml range (51), and the Ig-bound RNA copy numbers measured in this study are orders of magnitude lower (pg/ml range; data not shown). This finding lends further support to the hypothesis that RNA-containing Ags, rather than the autoantibodies themselves, are the primary determinant in driving inflammation (52).

The presence of autoantibodies directed against RNA-binding proteins was reported to be correlated with increased expression of IFN-inducible genes (9, 41); therefore, it is not surprising that elevated levels of Ab-bound RNA were also associated with an IFN-induced gene-expression signature. However, it was notable that patients negative for all tested RNA autoantibodies still had a strong association between Ab-bound RNA and IFN-stimulated gene expression. Furthermore, there was an association between Ab-bound RNA and elevated SLEDAI scores in these autoantibody-negative patients. Therefore, direct measurement of Ab-bound RNA provides information beyond that available from autoantibody measurements alone.

Precisely measuring the amount of Ab-complexed RNA circulating in vivo is a technically challenging undertaking, given that considerable RNase activity is also circulating in the blood (41). Avoiding RNA degradation during sampling necessitates immediate RNase inhibition, as was done in this study. In addition, avoiding the liberation of RNA-containing Ag during plasma preparation is difficult and may lead to the accumulation of additional autoantibodies complexed with RNA Ags that are formed ex vivo during plasma preparation. Notwithstanding the technical limitations noted, it is clear from the correlations observed in this study among patients positive for U1 or Y1 RNA, IFN-stimulated gene expression, and disease activity that Ag-bound RNA molecules are circulating in the blood of these patients and constitute an upstream stimulus driving the production of IFN-α and immune system activation.

SLE flares are known to be triggered by various external stimuli, such as UV light, viral infections, or certain drugs (53, 54). Presumably, the liberation of additional RNA-bound Ag as a result of any of these factors leads to increased Ab-complexed RNA, further exacerbating inflammation. Our longitudinal substudy quantitated daily U1 and Y1 RNA levels over a 5-d period in a subset of four patients to examine fluctuations in these Ab-complexed RNAs over time. In this analysis, fluctuations of 2–3 fold in Ab-bound RNA measurements were typical. Once an IC forms in the circulation, it is expected to exist transiently (55–57). Steady-state levels of circulating RNA-containing ICs result from the interplay of several biological processes, including release of RNA and associated protein Ags from dying cells, the concentration of autoantibodies in the circulation, the levels of blood-borne RNase, and the rate of clearance of ICs via receptor-mediated processes. Therefore, the steady-state level of RNA-containing ICs is likely to result from the integration of several biological systems and may be quite different among SLE patients.

In this study, low levels of Ab-bound U1 and Y1 RNA were observed in the HVs examined. Given that antinuclear Abs are known to exist in some healthy individuals, this finding is perhaps not unexpected (58, 59). The large amounts of Ab-bound RNA found in many of the SLE patients examined in this study are consistent with the reported defect in the normal clearance mechanism that appears to result in accumulation of cellular debris in the circulation (4–6). Coupled with a loss of self-tolerance and the development of autoantibodies, the presence of circulating RNA provides a proximal trigger of the inflammatory cascade in SLE. The findings presented in this study highlight for the first time, to our knowledge, the connection among circulating RNA, IFN-stimulated gene expression, and SLE disease activity. Therefore, a therapeutic approach aimed at reducing the amount of circulating RNA represents a potentially effective strategy for the treatment of SLE.

We thank all study participants for important contributions to this investigation and extend a special thanks to the clinical site coordinators (Dawn Dubois, Cathy Loeffler, Pam Kivitz, Jasmine Becella, Lidia Espino, Sanita Kandasami, and Jolene Washington) and their teams for efforts in enrolling patients and collecting samples. We also thank Robert Gerwein for expertise with statistical analysis of the data.

J.A.P., D.J.B., J.R.D., and K.B.E. are consultants to Resolve Therapeutics, LLC, and J.A.P., J.R.D., and K.B.E. own equity in the company. The other authors have no financial conflicts of interest.