An independent comparison of the Novolytix salivary melatonin radioimmunoassay with the new Novolytix salivary melatonin enzyme‐linked immunosorbent assay

The dim light melatonin onset (DLMO) is the current gold standard biomarker of the timing of the central circadian clock in humans and is often assessed from saliva samples. To date, only one commercially available salivary melatonin assay is considered accurate at the low daytime levels required to accurately detect the DLMO (Novolytix RIA RK‐DSM2). The aim of this study was to conduct the first independent evaluation of a newly improved enzyme‐linked immunosorbent assay (ELISA; Novolytix MLTN‐96) and compare it with the recommended radioimmunoassay (RIA)—both in terms of melatonin concentrations and derived DLMOs. Twenty participants (15 females, 18–59 years old) provided saliva samples every 30 min in dim light starting 6 h before their habitual bedtime, yielding a total of 260 saliva samples. Both the RIA and ELISA yielded daytime melatonin concentrations <2 pg/mL, indicating adequate accuracy to detect the DLMO. The melatonin concentrations from the two assays were highly correlated (r = .94, p < .001), although the RIA yielded lower levels of melatonin concentration than the ELISA, on average by 0.70 pg/mL (p = .006). Seventeen DLMOs were calculated from the melatonin profiles and the DLMOs from both assays were not statistically different (p = .36) and were highly correlated (r = .97, p < .001). Two DLMOs derived from the RIA occurred more than 30 min earlier than the DLMO derived from the ELISA. These results indicate that the new Novolytix ELISA is an appropriate assay to use if the Novolytix RIA is not feasible or available.

The current gold standard biomarker of the timing of the central circadian clock in humans is the dim light melatonin onset (DLMO). 1,2Melatonin levels are typically very low during the day, but begin to rise 1-3 h before habitual sleep onset time, remain elevated during the night, and return to low daytime levels usually within 1 h of habitual wake time. 3Melatonin is suppressed by light and thus the onset of its secretion must be measured in dim light. 4Although melatonin concentrations in plasma are approximately threefold higher than in saliva, the timing of the DLMO in these two fluids is similar. 5hus, for clinical and research purposes, melatonin is often assessed non-invasively from saliva samples, albeit at lower concentrations.Indeed, the salivary DLMO is a key tool for understanding the contribution of the endogenous circadian system to human health. 6t is recommended that the salivary DLMO be calculated with a low threshold, such as 3 pg/mL. 6,7lthough there are multiple commercially available salivary melatonin assays, only some are accurate at the low levels of melatonin that is required for the accurate determination of the salivary DLMO. 8In a recent review of melatonin assays, any salivary melatonin assay that registered daytime levels greater than 2 pg/mL was considered inaccurate. 8The only commercially available salivary melatonin assay that met this criteria was the Buhlmann Laboratories radioimmunoassay (RIA, RK-DSM2, average daytime levels of approximately 1 pg/mL).The next best performing salivary melatonin assay was the Buhlmann Laboratories enzyme-linked immunosorbent assay (ELISA, EK-DSM), although with average daytime levels of approximately 3 pg/mL, this was still considered inaccurately high. 8Thus, it is clear that the choice of melatonin assay will have critical consequences for the accurate assessment of the salivary DLMO and only one assay at the time of the 2019 review met the criteria of low daytime values of <2 pg/mL.
RIAs, although potentially more sensitive and accurate than ELISAs, involve the use of a radioactive tracer, which many university campuses will not permit in their facilities.For this reason, the use of RIAs by researchers often requires the additional expense of shipping samples to an external laboratory to conduct the RIA.By contrast, it is not uncommon to find university-based laboratories capable of running ELISAs, which removes the need and cost of shipping samples.Thus, a better performing salivary melatonin ELISA, registering average daytime levels at less than 2 pg/mL, will widen the available choice of accurate salivary melatonin assays, and enable more circadian researchers to conduct their own accurate melatonin assays inhouse.
In January 2021, Buhlmann Laboratories entered into a royalty agreement, relinquishing their responsibilities and rights for the melatonin assays to Novolytix.The RIA salivary melatonin assay (RK-DSM2) remains unchanged, however Novolytix stated that they have improved their ELISA salivary melatonin assay (MLTN-96), and report a high concordance between their RIA and new ELISA when tested on 84 samples, as part of an internal study (r2 = .931). 9 No melatonin profiles were displayed, excluding the examination of daytime melatonin levels.Furthermore, the difference in DLMO detection between these RIA and ELISA salivary melatonin assays has not been examined.Thus, the primary aim of this study was to conduct the first independent comparison of the Novolytix RIA for salivary melatonin versus their improved ELISA for salivary melatonin, to help inform researchers on the potential use of these assays.

| Participants
There were 20 participants (five males, 15 females, 18-59 years, mean age ± SD: 33.1 ± 11.3 years).The majority were non-Hispanic white (70%), with two Hispanic more than one race participants (10%), three non-Hispanic Black participants (15%), and one non-Hispanic Asian participant (5%).Eight of the participants were recruited specifically for this study (May-August 2022), with the remaining 12 having generated saliva samples (with excess saliva available for this study) when they participated in earlier studies (between September 2019 and September 2021).Of these 12, there were four healthy light drinkers, 10 five people with delayed sleep-wake phase disorder 11 and three women with postpartum depression (unpublished, NIH grant R01 MH121531).Otherwise, all participants were free of any other chronic mental or physical disease and were not taking any prescription medications (apart from two participants taking hormonal birth control and one participant taking an antidepressant medication).None had traveled across time zones nor engaged in shift work in the past month.Twelve participants underwent a urine toxicology screen (for cocaine, amphetamine, methamphetamine, marijuana, opiate, phencyclidine, barbiturates, and benzodiazepines) and tested negative for all drugs.All participants were instructed to be drug free during their study participation.All participants gave written informed consent before participating.All studies were conducted at the University of Michigan according to the principles outlined by the Helsinki Declaration and the Institutional Review Board approved all of the study protocols.

| Saliva collection
Each participant provided saliva samples every 30 min using Salivettes starting 6 h before their individual habitual bedtime.This resulted in the collection of 13 samples from each participant.Participants were instructed not to consume any alcohol or caffeine in the prior 24 h and no nonsteroidal anti-inflammatory drugs in the prior 72 h, to avoid confounding melatonin levels. 12,13Twelve participants collected saliva samples while seated in the laboratory in dim light (<5 lux as measured with an Extech EA33 light meter, at the angle of gaze).The dim red light was generated from a Sunlite 25A/TR/CD2 incandescent 25 watt A19 red colored lamp placed behind an opaque white light shade, and positioned at least 4.6 m to the right of each participant.Eight participants collected saliva samples at home using standardized procedures that included objective verification of dim light levels (<50 lux, with addition of blue wavelength blocking glasses 14 ) and correct sample timing. 15,16Specifically, during the home saliva collections, light was continuously monitored with an Actiwatch Spectrum (Respironics, 30 s epochs, with wrist band removed, worn around the neck on a cord).The blue wavelength blocking glasses consisted of a SCT-Orange Lens in a Skyper frame (Uvex), with 0% light transmission for <540 nm.All samples were immediately frozen after collection, and transferred to a −20°C freezer for storage.
In preparing the saliva samples for this study, each participant's 13 frozen saliva samples were thawed, shaken for 5 s, and then centrifuged for 5 min before 0.5 mL was pipetted into a separate empty salivette for the ELISA, and 0.9 mL was pipetted into a separate empty salivette for the RIA (which requires more saliva volume).In this way, each saliva sample was split into two duplicate samples before being refrozen at −20°C.
The samples for the RIA were coded with a Y followed by a randomly generated two-digit code (using Microsoft Excel).Similarly, the samples for the ELISA were coded with an X, followed by a different randomly generated two-digit code.These procedures prevented any determination of which samples for the RIA were the same as the samples for the ELISA.
The saliva samples were shipped in dry ice to Novolytix for RIA and ELISA analysis.After an extensive search within the United States, this laboratory was identified as the only laboratory that could conduct both the RIA and ELISAs on the duplicate samples, thereby avoiding potential differences due to variations in individual laboratory practices.As described above, due to the random labeling system, the laboratory was not able to identify nor match the duplicate saliva samples to each other.Furthermore, samples were shipped and assayed in two separate batches, with the duplicate samples not always shipped and assayed together.The samples were assayed in September and November 2022.
The assay details are reported by Novolytix as follows. 9The RIA is reported to have a lower limit of detection of 0.2 pg/mL and the intra-assay and interassay coefficients of variation for low, medium and high concentrations of melatonin are reported in Table 1.The newly improved ELISA is reported to have a lower limit of detection of 0.5 pg/mL, and the intra-assay and interassay coefficients of variation for low, medium and high concentrations of melatonin are reported in Table 2.A DLMO was calculated for each melatonin profile as the clock time (with linear interpolation) when the melatonin concentration exceeded 3 pg/mL, as per a consensus report. 7The average daytime value was determined from the first three samples in each melatonin profile.

| Statistical methods
We visualized the correlation and calculated a Pearson's correlation coefficient between melatonin concentrations assessed using the two assays.Considering 20 The intra-and inter-assay variabilities of the Buhlmann/Novolytix RIA for salivary melatonin (RK-DSM2).

Intra-assay precision
Inter-assay precision participants, the study had 89% power to detect a Pearson's correlation coefficient (r) of .6 or larger assuming an r of .9.We also calculated this correlation separately for the samples collected in the laboratory, versus samples collected at home.We compared the daytime levels of melatonin concentration (average of first three points in each melatonin profile) between assays using a paired t test.
The distribution of the overall melatonin concentrations showed skewness to the right due to the low baseline values before the rise in melatonin.The concentrations were recorded as zero when below 0.1 pg/mL and we therefore log-transformed the concentration after adding 0.09 to the observed concentrations.We then tested for the difference in log-transformed melatonin concentrations between the two assays using a mixed model with participant as random intercepts and with autoregressive structure of order 1 to account for correlation from repeated assessments within individuals.We also extended the model to see if the difference between assays in the melatonin concentrations increased with the number of days between sample collection and the day they were assayed (even though all samples were frozen immediately after collection at −20°C) and separately examined if the differences increased with higher melatonin concentrations, likely from greater assay variability at higher melatonin concentrations.Finally, we assessed if the assay difference varied depending on whether the saliva samples were collected in the laboratory or at home.For ease of interpretation, we did similar analyses using the raw (not log-transformed) melatonin concentration data and sought for consistency in substantive findings.
The DLMOs derived from the two assays were compared with a paired t test to examine the statistical significance of the difference.Pearson's correlations were calculated to examine the linear association between the two DLMOs in the entire sample, and in the samples collected in the laboratory versus at home.Bland-Altman plots were also created (the mean difference between pairs of data plotted against the mean value of the pairs 17 ).The plots indicate the limits of agreement as the interval of two standard deviations of the measurement differences either side of the mean difference.We also determined a priori the limits of maximum acceptable differences in DLMO values between any two methods as an absolute value of 30 min, as this has been defined as a clinically significant difference in the DLMO. 18All analyses were conducted using Stata 15.1 (StataCorp).Statistical significance was determined with a two-sided test p < .05.

| Melatonin concentrations
Twenty participants (12 in laboratory, eight at home) with 13 samples each were included in the analyses.The average daytime melatonin concentrations were 1.0 pg/mL (SD 1.1) in the RIA melatonin profiles and 1.1 pg/mL (SD 1.2) in the ELISA melatonin profiles, and this difference was not statistically significant (p = .67),nor was it associated with the location of sample collection (laboratory or home, p = .58).Combined across the 13 samples each from the 20 participants, the logtransformed melatonin concentration data between the two assays were similar, with a Pearson's correlation coefficient of .82(p < .001).The mixed model analysis of the log-transformed melatonin concentration difference, with the difference calculated as the ELISA minus the RIA, showed the difference in overall concentration to be significantly positive (β = 0.38, p = .007),indicating higher overall mean melatonin concentrations by ELISA than RIA across the melatonin profile.When location of sample collection was examined, the difference between the ELISA and RIA was not significant for samples collected in the laboratory (β = 0.14, p = .10)but the ELISA melatonin concentrations were significantly higher than the RIA melatonin concentrations in the samples collected at home (β = 0.72, p = .01).The number of days between sample collection and day of assay was not The intra-and inter-assay variabilities of the Novolytix ELISA for salivary melatonin (MLTN-96).

Intra-assay precision
Inter predictive of the magnitude of the difference in melatonin concentration by the two assays (all samples p = .27,laboratory samples p = .24,home samples p = .87).In examining if differences were related to greater assay variability at higher melatonin concentrations, we found melatonin concentration level itself by RIA to be a significant predictor of the magnitude of the difference (p = .005);however, the expected difference between the two assays at mean melatonin concentration remained significant (p = .003).
Analyses of the non-log-transformed (raw) mean difference in melatonin concentration by the two assays gave substantively similar results.The Pearson's correlation coefficient between the two assays for the entire sample was 0.94 (p < .001; Figure 1) and with similar correlations observed in the laboratory and home collected samples (Figure 1).The same correlations are shown in Supporting Information S1: Figure 1S for samples with melatonin concentration <10 pg/mL.The overall unadjusted difference between the assays was 0.70 pg/mL (p = .006),with a similar difference observed in the laboratory collected samples (0.66 pg/mL, p = .09)and home collected samples (0.72 pg/mL, p = .001).There was no significant effect of the number of days between sample collection and day of assay (p = .57).The two melatonin profiles showing the smallest and largest average difference in magnitude in melatonin concentration between the two assays are also shown, and both were collected in the laboratory (Figure 2).

| DLMO
Three participants had melatonin profiles in which the DLMO could not be calculated, because the melatonin concentrations did not rise above the threshold of 3 pg/mL.In the remaining 17 participants, DLMOs derived from both assays showed no notable skewness or need for transformation.The DLMOs derived from the RIA were comparable to the DLMOs derived from the ELISA (21.6 ± 1.9 vs. 21.7 ± 1.7 h, respectively, p = .36).The correlation between the DLMOs derived from the two assays was high (r = .97,p < .001; Figure 3).This correlation remained high whether DLMOs were derived from samples collected in the laboratory (r = .92,p < .001)versus at home (r = .99,p < .001).
The Bland-Altman plot revealed that the mean difference between the paired values was 0.12 h and only one participant's DLMO fell outside the 95% limits of agreement ([−0.89,1.13], Figure 4).This individual had a large difference between the DLMOs-the RIA derived DLMO occurred 1.9 h earlier than the ELISA derived DLMO (Figures 4 and 5).This same participant, and an additional participant, both had DLMO differences of greater than 30 min, which is considered clinically meaningful.The second participant also had

| DISCUSSION
To our knowledge, this is the first independent comparison of what others have identified as the most accurate commercially available salivary melatonin assay (Novolytix RIA RK-DSM2) and a newly improved ELISA (Novolytix MLTN-96), for the measurement of salivary melatonin.In this test of 260 duplicate samples and 17 individual duplicate DLMOs, we found that both assays yielded daytime melatonin concentrations of <2 pg/mL, and thus both assays met the criteria of being accurate at low levels of melatonin, as required for the accurate detection of the DLMO. 8Although the ELISA yielded on average higher melatonin concentrations than the RIA,  and more so in the samples collected at home, the mean difference was relatively small at <1 pg/mL and the melatonin concentrations were highly correlated.Overall, the DLMOs derived from the two assays were not significantly different, and the correlation between them was very high.However, two participants had RIAderived DLMOs that were more than 30 min from their ELISA-derived DLMOs, and in both cases the RIAderived DLMOs occurred earlier in time.Thus, for 12% of cases (2/17), the DLMOs derived from the two assays were meaningfully different.
No apparent differences existed between the two participants with the largest and smallest differences in melatonin concentrations derived from the two assays.Similarly, no apparent differences existed between the two participants with the largest and smallest differences in the DLMOs derived from the two assays.All four participants had their saliva samples collected in the laboratory under staff supervision, all were not taking any medications, and all passed a urine drug screen.We do note that the two participants with the largest difference in melatonin concentration and DLMO, secreted overall higher levels of melatonin than the two participants with the smallest difference in melatonin concentration and DLMO.This is consistent with our finding that the assay difference increased as melatonin concentration increased, likely due to the greater assay variability at higher melatonin concentrations.Lastly, we note that the ELISA requires less saliva than the RIA-Novolytix reports for duplicate processing, only 0.2 mL is needed for the ELISA, whereas at least 0.8 mL is needed for the RIA.Thus, in some cases where there may be substantial risk for low volume saliva samples (such as home saliva collection), and/or the RIA is not readily available the use of the ELISA may be preferred.The RIA could be preferred for better detection of low levels of melatonin, although in our study this did not systematically affect the resulting DLMOs.
Limitations of this study include that the sample consisted of more women than men, although this is not expected to have influenced the differences between the two assays.Notably, there was a range in how long the saliva samples had been frozen for before being assayedranging from 2 months to almost 3 years.However, the storage time before assay was not related to the differences between the two assays.We also note that the difference between the two assays did not appear to be affected by the use of hormonal birth control in two participants and an antidepressant in one participant.Importantly, this study was designed to examine melatonin levels before habitual bedtime and not across a 24 h period, and therefore future work should examine the assay difference at higher peak levels of melatonin.Lastly, although our random labeling procedures prevented determination of which samples for the RIA were the same as the samples for the ELISA, we note that Novolytix may have conducted the assays with more than usual care, given their understanding that the data were to be submitted for publication.
In conclusion, the results of this study provide the first independent confirmation that the new ELISA from Novolytix meets the accuracy criteria set by others to have average daytime levels below 2 pg/mL. 8Overall, the new ELISA produced higher levels of melatonin than the RIA, but the difference was small, and the resulting DLMOs did not significantly differ and were highly correlated.Two of 17 cases had RIA-derived DLMOs occur more than 30 min earlier than the ELISA-derived DLMOs, which is clinically meaningful.Overall, the results suggest that if the RIA is not available or feasible for researchers, the next best assay to use is the new Novolytix ELISA (MLTN-96).This is important, because not all melatonin assays perform equally well, and as others have highlighted, if the melatonin assay used is not accurate, this can result in the publication of studies that are seriously flawed. 8

F I G U R E 1
The salivary melatonin concentrations from the enzyme-linked immunosorbent assay (ELISA) versus the salivary melatonin concentrations from the radioimmunoassay (RIA) (n = 260).The two measures were highly correlated.The solid line is the correlation line, the dashed line is the line of perfect concordance for the full sample.The samples collected in the laboratory are shown in closed circles, the samples collected at home are shown in open circles.

F
I G U R E 2 The two individual salivary melatonin profiles showing the largest (A) and smallest (B) average difference in melatonin concentration between the two assays.The radioimmunoassay is shown in closed circles, the enzyme-linked immunosorbent assay in open circles.The dashed horizontal line represents the dim light melatonin onset threshold of 3 pg/mL.Both melatonin profiles were collected in the laboratory.their RIA derived DLMO occur earlier than the ELISA derived DLMO, by 0.64 h.Both participants collected their saliva samples in the laboratory.The remaining 16 pairs of RIA and ELISA derived melatonin profiles with DLMO threshold are shown in Supporting Information S2: Figure 2S.

F I G U R E 3
The dim light melatonin onsets (DLMOs) derived from the enzyme-linked immunosorbent assay (ELISA) versus the DLMOs derived from the radioimmunoassay (RIA) (n = 17).The two DLMOs were highly correlated.The solid line is the correlation line, the dashed line is the line of perfect concordance in the full sample.The DLMOs derived from samples collected in the laboratory are shown in closed circles, the DLMOs derived from samples collected at home are shown in open circles.F I G U R E 4 A Bland-Altman plot of the dim light melatonin onsets (DLMOs) derived from the enzyme-linked immunosorbent assay (ELISA) and radioimmunoassay (RIA) (n = 17).The dashed lines represent the limits of agreement as defined as 2 SDs (95%) of the measurement differences on either side of the mean difference.The DLMOs derived from the samples collected in the laboratory are shown in closed circles, the DLMOs derived from samples collected at home are shown in open circles.

F I G U R E 5
The two individual salivary melatonin profiles showing the largest (A) and smallest (B) difference in dim light melatonin onsets (DLMOs) between the two assays.The radioimmunoassay (RIA) is shown in closed circles, the enzymelinked immunosorbent assay (ELISA) in open circles.The dashed horizontal line represents the DLMO threshold of 3 pg/mL.Both melatonin profiles were collected in the laboratory.