Chorioamnionitis-exposure alters serum cytokine trends in premature neonates

Determine if chronologic age and/or chorioamnionitis exposure alter normal serum cytokine and chemokine levels in uninfected preterm neonates during their initial NICU stay. A 7-plex immunoassay measured levels of serum IL-1β, IL-6, IL-8, IL-10, TNF-α, CCL2, and CCL3 longitudinally from chorioamnionitis-exposed and unexposed preterm neonates under 33 weeks’ gestation. Chorioamnionitis-exposed and unexposed preterm neonates demonstrated differences in the trends of IL-1β, IL-6, IL-8, IL-10, TNF-α, and CCL2 over the first month of life. The unexposed neonates demonstrated elevated levels of these inflammatory markers in the first two weeks of life with a decrease by the third week of life, while the chorioamnionitis-exposed neonates demonstrated differences over time without a predictable pattern. Chorioamnionitis-exposed and unexposed neonates demonstrated altered IL-10 and TNF-α trajectories over the first twelve weeks of life. Chorioamnionitis induces a state of immune dysregulation in preterm neonates that persists beyond the immediate neonatal period.


INTRODUCTION
Preterm birth, defined as delivery that occurs prior to 37 weeks' gestation, complicates approximately 11% of births globally [1]. Preterm neonates are at risk for numerous morbidities during their initial hospitalization, including an increased risk for sepsis [2]. This infection risk is often attributed to immaturity of the preterm immune system, particularly the innate immune system. Several studies have demonstrated differences in innate immune cytokine levels between term and preterm infants [3,4]. These altered cytokine responses are thought to contribute to a preterm neonate's heightened susceptibility to infection as appropriate cytokine responses are necessary to guide the clearance of microorganisms [5]. However, the natural evolution of innate immune responses in premature neonates is incompletely understood.
Preterm delivery is often complicated and may even be stimulated by intrauterine inflammation and/or infection, termed chorioamnionitis [1]. Chorioamnionitis is present in up to 70% of very preterm deliveries and leads to an initial fetal pro-inflammatory response, including increased expression of the pro-inflammatory cytokines IL-1β, IL-6, IL-8, and TNF-α [6,7]. This fetal inflammatory response alters the developing immune system, resulting in decreased pro-inflammatory cytokine expression when umbilical cord blood monocytes from chorioamnionitis-exposed neonates undergo a secondary challenge with either lipopolysaccharide (LPS) or Staphylococcus epidermidis [8,9]. Chorioamnionitis exposure is known to increase the risk of developing both early and late onset neonatal sepsis, which may be at least partially due to these dampened monocyte responses [10,11]. It is currently unclear how long this chorioamnionitis-induced immune hypo-responsiveness persists, which could impact a preterm infant's already heightened susceptibility to infection beyond the immediate neonatal period.
The primary objective of this study was to determine if serum cytokine and chemokine concentrations in uninfected preterm neonates differed based on chronologic age. A secondary objective was to determine if chorioamnionitis-exposure influenced serum cytokine and chemokine concentrations in uninfected preterm neonates during and beyond the immediate neonatal period. To achieve these objectives, we performed longitudinal cytokine and chemokine profiling using a 7-plex cytokine and chemokine assay to measure concentrations of CCL2, CCL3, IL-1β, IL-6, IL-8, IL-10, and TNF-α in neonatal serum samples. Using less than 200 μL of residual serum from clinically indicated routine blood tests, we compared cytokine and chemokine levels throughout a subject's entire NICU course.

METHODS Patient recruitment and blood collection
This study was approved by the University of Michigan IRB. This study was performed in accordance with the Declaration of Helsinki. This was a single-center study. The University of Michigan C.S. Mott Children's Hospital is a 348 bed tertiary care hospital. The Brandon NICU is a 52-bed level IV NICU and the Von Voigtlander Women's Hospital has 3600 births yearly. After informed written parental consent was obtained, residual serum was collected prospectively from clinically indicated laboratory draws of neonates born at less than 33 weeks' gestational age. Serum samples were collected from 61 patients from birth through 42 weeks' postmenstrual age, death or discharge, whichever occurred first. A power calculation was performed using our previously published data demonstrating decreased IL-8 protein expression in neonatal chorioamnionitisexposed monocytes following stimulation with LPS compared to unexposed monocytes (1023 ± 507.5 pg/mL vs 384.6 ± 156.1 pg/mL) [9]. Using an alpha value of 0.05 and a power of 90%, a sample size of 13 subjects per exposure group was estimated to detect a statistically significant difference between groups. Our subject cohort included 27 chorioamnionitis-exposed and 34 unexposed preterm infants, exceeding this target sample size. The average number of blood draws for enrolled subjects was 14 (range 2-37). Sample collection occurred from April, 2019 through April, 2021.
Histopathologic examination of the placenta was performed by qualified pathologists and the Amsterdam Placental Staging Criteria was used to diagnose chorioamnionitis [9,12,13]. Ten of the chorioamnionitis-exposed subjects also had evidence of funisitis on placental histopathology. Subjects with placental pathology significant for acute chorioamnionitis ± funisitis were included in the chorioamnionitis group as we found no differences in serum cytokines or chemokines between chorioamnionitis only and chorioamnionitis with funisitis exposed preterm neonates ( Supplementary Fig. 1). Subjects without inflammation on placental pathology, even if there was clinical suspicion for chorioamnionitis, were included in the unexposed group. The blood volume collected with each sample varied, as the serum available for testing was what remained after all clinically ordered testing was performed. As 200 µL was required for performance of the cytokine assay, samples were pooled if collected within three days of one another and the subject had no significant change in clinical status. A total of 397 residual serum samples were collected. Samples were frozen and stored in a −80°C freezer prior to use. As the main objective of this study was to evaluate if baseline serum cytokine values differed based on chronologic age and/or chorioamnionitis exposure and several previous studies demonstrated elevated levels of IL-6, IL-10, IL-8 and/or TNF-α around the time of sepsis or necrotizing enterocolitis diagnosis, we excluded samples from longitudinal data evaluation if the subject had a suspected or confirmed infection and was being treated with antibiotics at the time of sample collection to eliminate samples that may falsely elevate baseline cytokine and chemokine levels [14][15][16][17]. This included treatment for sepsis, urinary tract infection, pneumonia, necrotizing enterocolitis or spontaneous intestinal perforation, excluding 100 samples from analysis. Samples were excluded from 13 chorioamnionitis-exposed and 16 unexposed preterm neonates. A total of 297 serum samples were included in the final longitudinal analysis. Samples were again included from these subjects once the antibiotic treatment course had ended.

Multiplexed immunoassays
Microring resonator immunoassays were validated and performed on the Maverick M1 and Matchbox systems (San Diego, CA USA), respectively, as previously described [18][19][20]. The Maverick instruments use microfluidic systems for automated reagent handling. The M1 uses reusable cartridge devices and the Matchbox uses disposable, injection-molded, plug-andplay devices [20]. Microring chips were functionalized with capture antibodies using an amine-reactive, homobifunctional crosslinker to create a 7-plex cytokine and chemokine capture array. Each capture antibody spanned two clusters of four microring sensors in each of the two microfluidic channels, giving n = 8 technical replicates of each target cytokine or chemokine per channel. After introducing the sample to the chip surface, a mixture of all tracer antibodies was flowed across the chip, followed by streptavidin-tagged enzymes and a signal amplification reagent. Assays were performed at a 30 μl/min flow rate for all steps. There was an initial rinse of 5 min with the running buffer to ensure equilibration of the chip prior to sample analysis. The assay included steps as follows: (1) running buffer (2 min); (2) sample (7 min); (3) running buffer rinse (2 min); (4) biotinylated tracer antibodies (7 min); (5) running buffer rinse (2 min); (6) SA-HRP (7 min); (7) running buffer rinse (2 min); (8) 4-CN (7 min); (9) running buffer rinse (2 min). The total assay time was 38 min (Supplementary Fig. 2A).

Immunoassay calibrations
The 7-plex immunoassay was simultaneously calibrated for all analytes in a multiplexed format, as described previously [18]. Serial dilutions from a mixed saturating analyte sample of all multiplexed targets were used to construct eight-point calibration curves correlating net sensor shifts to target concentrations. To quantify, the signal before the enhancement step (t = 29 min) was subtracted from the signal after the final assay rinse step (t = 38 min). These net resonance wavelength shifts (Δpm) were plotted as a function of standard concentration and fit to a four-parametric logistic function ( Supplementary Fig. 2B). Limits of detection (LOD) and quantification (LOQ) were defined as the blank signal plus 3 times and 10 times the standard deviation of the blank, respectively (Supplementary Table 2). Each calibration was performed at least in triplicate for each sample dilution as measured with 8 sensors per technical replicate.

Sample evaluation
All samples contained at least 200 μL of residual serum. Neonatal residual serum samples were analyzed at two dilutions (0.5X and 0.1X) in running buffer using the same steps highlighted in Supplementary Fig. 2A. To quantify, the net shift surrounding the amplification step for each target was correlated to concentration using the corresponding standard calibration curve, 50% serum or 10% serum, matching the serum content of the residual serum dilution. The most appropriate dilution to use for statistical analysis was selected by choosing the dilution with the relative shift closest to the inflection point of the respective calibration curve.

Statistical analysis
Basic statistical analysis was performed in GraphPad Prism 8. Data normality was evaluated using the Shapiro-Wilk test. The ROUT test was used to identify and remove statistically significant outliers. Study group characteristics were compared using the student's t test for quantitative parametric data, the Mann-Whitney test for nonparametric data and the Fisher's exact test for categorical variables. p values of <0.05 were considered significant. Cytokine and chemokine levels were compared between the first and second weeks of life in the same subject using the Wilcoxon matched-pairs signed rank test. If there was more than one data point within these time frames, the data points were averaged to create a single mean level for each week. p values < 0.05 was considered significant for this analysis method.
General Estimating Equations were used in SPSS 28.0.1.0 to evaluate for changes in cytokine trends over the first four weeks of life in the chorioamnionitis-exposed and unexposed groups as the data was longitudinal, paired and non-parametric with missing data points for some subjects. The General Estimating Equations used a robust covariance matrix, an unstructured working correlation matrix and a Tweedie with log link model. If there was more than one data point within each time frame, the data points were averaged to create a single mean level for each week. p values < 0.05 was considered significant for the comparison of overall trends within each exposure group. However, when individual timepoints were compared within exposure groups, p values < 0.01 were considered significant to correct for multiple comparisons.
Cytokine and chemokine levels from each subject were then compared over time by week-of-life (chronologic age) through 12 weeks. When there was more than one data point in a week, all points within that week were averaged to create a single mean cytokine level. Univariate statistics showed that the cytokines were not normally distributed and were largely right skewed, with many zeros, representing cytokine levels below the limit of detection. To transform the data to approximate a normal distribution more appropriate for modeling, the natural log of (cytokine level + x, where x is a positive value that varies based on the cytokine in question) was used. SAS Proc Mixed was used to perform repeated measures regression to look at the effect of chorioamnionitis status on the trajectory of cytokines over time while controlling for gestational age, ethnicity, and birth via C-section, all of which were found to be statistically different between exposure groups. Analyses were restricted to the first twelve weeks of life as the chorioamnionitis-exposed group had no data points beyond the first twelve weeks of life. Autoregressive covariance structure was selected based upon (a) a conceptual understanding of the data (measurements close in time would be expected to more strongly correlated than measurements which are farther away from one another) and (b) lower Akaike information criteria (AIC) in comparison to other covariance structures. The interaction term between week of life and chorioamnionitis status indicated whether or not the cytokine trajectories differed. Least square mean values from SAS Proc Mixed were graphed to allow for a clearer understanding of trajectory differences. p values of < 0.05 were considered significant.

Characteristics of study subjects
A total of 61 preterm neonates were enrolled in this study, including 27 exposed to chorioamnionitis and 34 unexposed. Subjects ranged from 22 to 32 weeks' gestational age at birth and were followed to 42 weeks' postmenstrual age, discharge or death, whichever came first. Table 1 describes characteristics of the two study groups. Chorioamnionitis-exposed preterm neonates were younger, more likely to be African American and more likely to be born by vaginal delivery than unexposed preterm neonates.
Cytokine and chemokine measurements during initial two weeks of life Levels of 7 cytokines and chemokines known to be important in innate immunity were measured in residual neonatal serum (Supplementary Table 3). We first sought to investigate the change in cytokine and chemokine levels from the first week of life to the second in all of the preterm neonates and separated subjects based on chorioamnionitis exposure. We directly compared all cytokine and chemokine levels from each infant averaged over the first week of life to its average levels in the second week of life using a matched comparison, with each infant compared to itself at two different points in time. Levels from the first week of life were significantly higher than those in week two for IL-6 and CCL2 in both chorioamnionitis-exposed and unexposed subjects but IL-8 only demonstrated this trend in unexposed subjects (Fig. 1).

Cytokine and chemokine trends over the first month of life
We then compared the levels of serum cytokines and chemokines in chorioamnionitis-exposed and unexposed subjects by time post birth. The following epochs were compared between the same subject: Unexposed and chorioamnionitis-exposed preterm neonates demonstrated changes in IL-1β, IL-6, IL-8, IL-10, TNF-α and CCL2 during the first month of life ( Fig. 2A-F). Chorioamnionitis-exposed preterm neonates demonstrated changes in CCL3 over the first month of life but unexposed preterm neonates did not (Fig. 2G). In general, unexposed preterm neonates demonstrated elevated levels of IL-1β, IL-6, IL-8, IL-10, TNF-α and CCL2 in the first one to two weeks of life with a decrease to what appears to be baseline levels by the third week of life (Fig. 2). This is in contrast to chorioamnionitis-exposed preterm neonates, who demonstrated differences in cytokine levels over the first month of life but without a predictable pattern (Fig. 2). Direct comparisons between the different time points are detailed in Supplementary Table 4.
Cytokine and chemokine trajectories between chorioamnionitis-exposed and unexposed preterm neonates Repeated measures of regression were then performed to look at the effect of chorioamnionitis status on the trajectory of cytokines over the 12 weeks following birth. This analysis controlled for gestational age, race/ethnicity and mode of delivery, as all of these variables were found to differ between exposure groups on univariate analysis. The trajectories of IL-10 and TNF-α differed between chorioamnionitis-exposed and unexposed neonates, while there were no differences in the trajectories of IL-1β, IL-6, IL-8, CCL2 or CCL3 between groups (Fig. 3).

DISCUSSION
Neonatal infections are a cause of significant morbidity and mortality in preterm neonates during their hospitalization in the NICU [2]. It is known that preterm neonates exposed to chorioamnionitis have an increased risk of developing earlyonset sepsis (blood stream infection that occurs within the first 72 hours of life) [8,11,21]. It is unclear if this infection risk is due to a common pathogen causing both conditions or alterations in the neonatal immune response following chorioamnionitis exposure, or both. Multiple studies have shown that exposure to chorioamnionitis impacts the neonatal immune system by altering gene transcription and innate immune responses [7][8][9]. These altered immune responses include dampened pro-inflammatory cytokine expression when a second pathogen is encountered [8,9]. Appropriate pro-inflammatory cytokine expression is necessary for the clearance of microorganisms, so these chorioamnionitis-induced changes to neonatal immune responses are thought to be at least partially responsible for this increased risk of infection. However, it is unclear how long chorioamnionitisinduced dampened cytokine expression persists, as studies are conflicting about whether chorioamnionitis exposure protects against or increases the risk for developing late onset sepsis (blood stream infection that presents after 72 hours of life) [11,[22][23][24].
To determine if baseline serum cytokine and chemokine concentrations in preterm neonates differed based on chronologic age and/or chorioamnionitis-exposure, we performed longitudinal cytokine and chemokine profiling in very preterm neonates from birth to NICU discharge. We chose a panel of cytokines and chemokines known to be significant contributors to neonatal immune responses. Neonates primarily rely upon the innate immune system early in life to protect against infections due to limited antigen exposure in utero and major deficiencies in adaptive immune responses [25,26]. Innate immune cells, including monocytes, macrophages and neutrophils, require signaling from cellular messengers such as cytokines and chemokines in order to mount a coordinated response to an infectious pathogen [27]. CCL2 and CCL3 are chemokines that recruits monocytes, macrophages and neutrophils to local sites of infection and are necessary for prominent signaling pathways in the neonatal immune system [28]. IL-8 shows similar chemotactic affinity for neutrophils and stimulates bacterial phagocytosis [29]. IL-6, IL-1β and TNF-α are pro-inflammatory cytokines important to the acute phase response necessary to assist in the clearance of microorganisms [30,31]. IL-10 is an immunoregulatory cytokine important for immune homeostasis that also suppresses autoinflammation [32]. We believe this panel of cytokines and chemokines provides a broad overview of neonatal innate immune reactivity.
In this study, we used a novel method of cytokine and chemokine evaluation, using each preterm neonate as its own matched control to compare levels at different chronologic ages. While this method has previously been used to demonstrate a significant decline in IL-1β, IL-6 and TNF-α from DOL 1 to DOL 40 in term neonates, we are the first to use it to evaluate changes in cytokine and chemokine levels over time in preterm neonates [33,34]. We found that in our population of preterm neonates, levels of IL-6 and CCL2 decreased between the first and second weeks of life in both chorioamnionitis-exposed and unexposed groups while IL-8 only decreased in the unexposed group. Nonchorioamnionitis exposed preterm neonates had a consistent decrease in levels of IL-1β, IL-6, IL-8, IL-10, TNF-α and CCL2 over the first month of life, reaching what appeared to be baseline levels around three weeks after birth. This is in contrast to chorioamnionitis-exposed preterm neonates, whose cytokine and chemokine levels demonstrated differences over the first month of life without a consistent pattern based on chronologic age. We additionally found that the trajectory of IL-10 and TNF-α serum levels differed between chorioamnionitis-exposed and unexposed preterm neonates. These findings are important as most of these cytokines and chemokines have been proposed as biomarkers to diagnose or predict prematurity-based complications, including sepsis, necrotizing enterocolitis and bronchopulmonary dysplasia [14,[35][36][37][38][39]. Our findings suggest that chronologic age and chorioamnionitis-exposure should be taken into consideration when using cytokines and chemokines as biomarkers in premature neonates.
The altered cytokine and chemokine responses seen in the chorioamnionitis-exposed preterm neonates is in line with previous reports demonstrating altered cytokine responses from chorioamnionitis-exposed umbilical cord blood monocytes following stimulation with either LPS or Staphylococcus epidermidis [8,9]. The cytokines and chemokines measured in this study are primarily expressed by innate immune cells, which are not typically self-renewing immune populations. How then, can chorioamnionitis-exposure around the time of birth influence cytokine and chemokine expression up to 12 weeks of life when the originally exposed innate immune cells are no longer around? The answer likely involves the concept of trained immunity, which describes long-term functional reprogramming of innate immune cells through epigenetics. We previously showed that chorioamnionitis exposure alters the histone tail modification landscape and subsequent gene expression profile of neonatal monocytes, resulting in a trained immunity phenotype [9]. This phenotype resulted in dampened proinflammatory cytokine expression when chorioamnionitisexposed monocytes encountered a secondary pathogenic stimulus [9]. Based on these findings, it is possible that chorioamnionitis-induced histone modification changes in innate immune cells results in long-lasting "epigenetic memory" and altered immune responses well beyond the immediate neonatal period. It is also possible that chorioamnionitis exposure results in a trained immunity phenotype in bone marrow progenitor cells, which would repopulate the circulating innate immune cells with this altered "epigenetic memory". Alterations in innate immune "epigenetic memory" may provide insight into immune-related complications experienced by chorioamnionitis-exposed neonates, including late onset sepsis, persistent wheezing and asthma [24,40].
Our 7-plex cytokine microring resonator assay was robustly validated for all targets simultaneously to ensure reproducible results across all samples analyzed. Each assay was 38 min to result, creating a quick method for analyzing important clinical samples. Using this multiplexed immunoassay, we were able to collect large amounts of immunological data quickly and with little starting sample volume. Early sepsis detection and prompt initiation of antibiotic therapy significantly improves outcomes in neonatal sepsis [41]. In practice, early sepsis detection is difficult as signs of infection in the neonatal population are often nonspecific. The gold standard test to diagnose a blood stream infection is a blood culture that demonstrates growth of a pathogenic organism, however this often takes at least 24 hours to result [42]. Several biomarkers are commonly used to support or refute the presence of infection, including C-reactive protein, procalcitonin and the presence of many immature forms of neutrophils [43]. These biomarkers are non-specific, and are often more useful to rule infection out rather than diagnose it. Several cytokines have been proposed as useful biomarkers to diagnose neonatal sepsis or necrotizing enterocolitis, including IL-1β, IL-6, IL-8, IL-10 and TNF-α [14,16,17,[35][36][37]. While cytokines hold great promise as biomarkers to diagnose neonatal sepsis, they often take days to result, so are unable to be used to promptly diagnose infection. Our microring resonator assay has the potential to provide cytokine values quickly for the most vulnerable patients, which could allow for the use of cytokine values in real time to accurately predict the risk of sepsis and impact bedside patient care. This study has several limitations. All samples were collected from clinically indicated laboratory tests, so the timing of sample Fig. 1 Comparison of cytokines and chemokines obtained in the first and second weeks of life in preterm neonates. Serum protein levels were measured and compared between the same subject during the first and second weeks of life. If more than one serum level was obtained during each week, then the average level was used for comparison. Serum protein levels are demonstrated for (A) IL-1β in unexposed neonates, (B) IL-1β in chorioamnionitis-exposed neonates, (C) IL-6 in unexposed neonates, (D) IL-6 in chorioamnionitis-exposed neonates, (E) IL-8 in unexposed neonates, (F) IL-8 in chorioamnionitis-exposed neonates, (G) IL-10 in unexposed neonates, (H) IL-10 in chorioamnionitisexposed neonates, (I) TNF-α in unexposed neonates, (J) TNF-α in chorioamnionitis-exposed neonates, (K) CCL2 in unexposed neonates, (L) CCL2 in chorioamnionitis-exposed neonates, (M) CCL3 in unexposed neonates and (N) CCL3 in chorioamnionitis-exposed neonates. First week unexposed n = 22, second week unexposed n = 22, first week chorioamnionitis-exposed n = 15, second week chorioamnionitis-exposed n = 15. Wilcoxon test used to determine statistical significance. *p < 0.05, **p < 0.01, ****p < 0.0001. Mean ± standard error of the mean for each protein level shown below each x-axis label.
collection varied between patients and was not standardized. There were differences between the exposure groups, and chorioamnionitis-exposed subjects were more likely to be born earlier, African American and by vaginal delivery than unexposed subjects. It is unclear if these differences impacted cytokine and chemokine expression. Degree of prematurity and mode of delivery have been shown to impact immune responses in prior studies, so these factors were accounted for in out statistical evaluation [44][45][46]. Samples were excluded from subjects who had a suspected or confirmed infection and were receiving antibiotics at the time of sample collection. We excluded these samples as several previous studies have demonstrated elevated levels of IL-6, IL-10, IL-8 and/or TNF-α around the time of sepsis or necrotizing enterocolitis diagnosis, and we were concerned that these and similar conditions could falsely elevate baseline cytokine measurements and confound the study results [14][15][16][17]. However, this should be taken into account when interpreting and attempting to generalize our results. We did include samples from these patients later during their NICU course once the infection was treated, however, as infectious/inflammatory conditions are common in preterm neonates and this population is likely to make up a large proportion of infants upon which normative values are based. Subject attrition was noted at later time points during the study (8 weeks and beyond) primarily due to patient discharge. This should be taken into account when considering our findings. Additionally, it is unclear if suspected or confirmed infections influence future cytokine and chemokine expression, but this would be an interesting and informative comparison to make in future studies. Furthermore, corrections were not made for clinical differences such as mode of respiratory support, presence of BPD, steroid administration, or PDA treatment. The numbers in this study are not large enough to directly address these potential confounding factors, but future studies containing more subjects would be of benefit.

CONCLUSIONS
This study demonstrated that healthy preterm neonates had a consistent decrease in levels of IL-1β, IL-6, IL-8, IL-10, TNF-α and CCL2 over the first month of life, reaching what appeared to be baseline levels around three weeks after birth. This same pattern of changes was not present in chorioamnionitis-exposed preterm Fig. 2 Longitudinal cytokine and chemokine trends over time in chorioamnionitis-exposed and unexposed preterm neonates. Serum protein levels were measured and compared between the same subject during day of life (DOL) 1-7, 8-14, 15-21, 22-28 and 29 and beyond. If more than one serum level was obtained during each timeframe, then the average level was used for comparison. Serum protein levels are demonstrated in chorioamnionitis-exposed (white circles) and unexposed (black circles) preterm neonates for (A) IL-1β, (B) IL-6, (C) IL-8, (D) IL-10, (E) TNF-α, (F) CCL2 and (G) CCL3. Unexposed n = 28, chorioamnionitis-exposed n = 17. General Estimating Equations were used to determine statistical significance. Circles represent mean levels and error bars represent standard error of the mean. p values for differences in trends over time are shown.