Dried blood spot sampling of testosterone microdosing in healthy females

.


Introduction
Measurement of circulating steroids such as testosterone (T), androstenedione (A 4 ), 17 hydroxyprogesterone (17OHP) and progesterone (P 4 ) are frequently required to evaluate ovulation as well as hyperandrogenism which may be due to polycystic ovary syndrome, congenital adrenal hyperplasia or, rarely, tumors of the adrenal or ovary or Disorder of Sexual Differentiation [1].However large-scale population, field or home-based self-sampling studies are hindered by the need for skilled and laborious sample processing requiring venipuncture, separation of serum/plasma and frozen storage and shipping.Capillary blood sampling using DBS technology coupled with the high sensitivity and specificity of multianalyte LCMS steroid analysis offers advantages of simpler, cheaper, less intrusive, lower blood volume sampling.It is also less reliant on skilled labor making it more convenient for field and/or self-sampling home studies while avoiding the need for costly frozen storage and shipping [2,3].We have applied DBS sampling to a clinical study of testosterone administration to determine the accuracy of capillary steroid measurements made using samples stored for four years at room temperature compared with the original serum measurements using the same methods.

Study objective and design
The study design and primary outcomes were reported previously [4].Briefly, an open label design comprising three periods and 12 study visits over a 5-week period was employed.During the 14-day run-in participants collected sets of paired capillary (DBS) and venous blood samples and again during 14-day post-treatment period following seven days of daily treatment with a transdermal testosterone gel.The study was registered with the ANZ Clinical Trials Register (ACTRN12619000633189) and all participants signing written informed consent as approved by the Sydney Local Health District Human Ethics Committee (Concord zone) consistent with the Declaration of Helsinki.
Healthy women aged 18-60 years were recruited by local advertising.Exclusion criteria comprised pregnancy or breast feeding; seeking fertility within the next 6 months; use of any hormonal drugs other than oral contraception or drugs that could interfere with testosterone metabolism; contraindications to testosterone administration; major or chronic medical disorders requiring regular prescribed medication; severe or extensive skin disease that would interfere with transdermal application of testosterone gel; history of androgen or other drug abuse within last year or a history of a major psychiatric disease or psychological condition that limits understanding and compliance with study requirements.Precaution was advised for women working in any occupation that requires urine drug testing.None was involved in competitive athletics.

Study procedures
Paired capillary and venous blood samples were taken at each study visit with the seated participant resting and at the same time of day, usually afternoons and without regard to menstrual cycle stage.
Capillary blood samples were collected as described previously by finger prick using a single-use lancet (BD Microtainer, Contact-Activated Lancet 2.0×1.5 mm; Becton Dickinson) and applying blood onto filter cards (Whatman 903 protein saver cards, GE Healthcare) [5].Only a single DBS sample was collected at each timepoint.
Daily treatment was a single actuation of a metered dose of the testosterone gel pump pack (Testogel 1%, Besins Healthcare) with the gel was applied to the same area of skin (abdomen, chest, or shoulder) once daily for 7 days.Each pump actuation provides 1.25 g of 1% gel delivering a dose of 12.5 mg testosterone.Each pump pack was weighed before delivery and after return to measure gel usage.

DBS and serum steroid profiling
DBS assay standards and quality control (QC) samples were created in artificial blood made by diluting washed human erythrocytes with charcoal-stripped plasma to a hematocrit of 42% as described previously [5].DBS samples together with standards and QCs were extracted by a single 6 mm punch that was transferred to a single tube with addition of 100 µl water and vortexing for 30 minutes (solution 1).Serum T was measured as described in the original study report [4] with previously unreported serum A 4 , 17OHP 4 and P 4 added by the same method for this study.
From solution 1, aliquots (10 µL) of each sample were placed in a 96 well plate with addition of 80 µl sodium lauryl sulfate (Sulfolyser, Sysmex UK) solution.The plate was gently mixed and left to stand for 5 minutes before absorbance at 550 nm was read on a Enspire multiple plate reader (Perkin-Elmore) as described [6].The remainder of solution 1 had addition of 25 µl internal standard plus 100 µl water before being vortexed for a minute and then left to stand for 30 min at 4ºC.Then 1.5 ml methyl tert-butyl ether (MTBE) was added with vortexing for a minute followed by standing for 1 hour at 4ºC and then centrifugation (300 rpm, 10 minutes) with the solutions then frozen at − 80ºC for 30 minutes, The unfrozen supernatant was then transferred to new glass tube for drying overnight at 40ºC on a heat block.The dried extracts were then reconstituted in 75 µl 50% ammonium fluoride:water with 50 µl injected into the LCMS system.
LCMS analysis was based on reversed-phase chromatographic separation of the injected sample using ultra-pressure liquid chromatography (UPLC) followed by gradient elution of target steroids on a Phenyl  Hexyl column (100 mm×2.1 mm×1.7 µm, part# 00D-4500-AN, Kinetex) fitted with a Security Guard ULTRA Cartridge (UHPLC Phenyl for 2.1 mm ID columns, part AJ0-8788, Phenomenex).Extracts (50 µl) injected into the chromatography column (flow rate 0.35 ml/ min, column temperature 40º C, total run time 13.5 minutes) were eluted using a variable gradient of 0.2 mM ammonium fluoride (solvent A) and methanol/water (solvent B) to separate target steroids before introducing the eluant into the mass spectrometer without splitting.
The limits of detection (LOD)/limits of quantitation (LOQ) were 0.0625 / 0.25 ng/ml for T and A 4 for A 4 and 0.125 / 0.50 ng/ml for 17OHP 4 and P 4 .Based on quality control samples, the within-day and between-day accuracy (average pooled over low, mid-range and high concentrations, as %) averaged 99.1 and 95.6, respecitvely, for T; 94.8 and 89.3 for A 4 ; 95.5 and 94.9 for 17OHP 4 ; and 106.0 and 100 (two levels in working range) for P 4 .Within-day and between-day reproducibility (coefficient of variation %) was 7.0 and 8.4, respectively, for T; 9.1 and 11.7 for A 4 ; 8.4 and 11.2 for 17OHP 4 and 10.5 and 9.7 for P 4 .
DBS steroid measurements of the last four run-in samples and six post-treatment samples for each participant were expressed in terms of unadjusted whole blood concentrations as well as adjusted for hematocrit to simulate serum measurements.The hematocrit adjustments were either the hematocrit of that original blood sample, or the overall group average hematocrit (38%) or else by an individual estimate of hematocrit from the punched-out DBS sample using the Sulfolyser method.

Hematology profile
Each blood sample had a standard hematological profile measured within an hour of collection on a regularly calibrated, Sysmex XN-450 (Roche, Australia) hematology analyzer using methods recommended by the manufacturer with reference ranges for healthy females [7] as described previously [4].The coefficients of variation were <6% for all standard erythrocyte variables including hematocrit and hemoglobin without bias as the mean values of the three levels of QC samples ranged between 93% and 105% of target for erythrocyte variables.The hemoglobin content of the DBS sample after four years storage was measured using the sodium lauryl sulfate (Sulfolyser) method as described [6].

Data analysis
Data were initially analyzed by descriptive measures and  represented as mean and standard error of the mean (SE) together with median and first (Q1) and third (Q3) quartiles.Comparison between serum and DBS measurements of T and A4 were undertaken by Passing-Bablok regression and Bland-Altman deviance analysis.Linear mixed modeling with repeated measures was used to fit the serial measurements for each set of steroid measurements to provide a goodness of fit estimate for that model according to the Akaike Information Criterion (AIC).AIC provides a measure of goodness of fit for the model based on the likelihood function and numbers of parameters as a measure of entropy reduction of the model for the data allowing comparisons between different models for the same data set.All data analysis used NCSS 2023 Statistical Software (NCSS, LLC.Kaysville, Utah, USA).

Participants
Twelve healthy female participants were recruited, each completing all scheduled visits and without any adverse effects reported.Compliance with study treatment was high according to both measured reweighing of returned bottles and 99% of scheduled doses reported as taken.Their age was 49.1 ± 1 years, height 162 ± 0.5 cm, weight 74.2 ± 1.4 kg, body mass index 28.2 ± 0.5 kg/m 2 and body surface area 1.84 ± 0.02 m 2 .All reported average to excellent health with three using regular prescribed non-hormonal medications for minor illnesses but none were taking hormonal contraceptives or iron supplements.Five had regular menses and the remainder were menopausal.Eight had children (median number 2, range 1-3).Alcohol intake was minimal with four non-drinkers and the others consuming a median of 2-3 standard drinks (range 1-7) per week).

DBS and serum steroid profile
DBS and serum T, whether measured as whole blood without hematocrit adjustment or after individual or groupwise adjustment for hematocrit increased from a pre-treatment baseline to display a large peak at the end of treatment with progressive decline over time towards pre-treatment baseline levels by 14 days after end of treatment (Fig. 1).Serum and DBS measurements of T displayed good agreement (Fig. 2).When analyzed by linear mixed model analysis for repeated measured, whole blood T without adjustment has the lowest AIC (best fit) followed by serum T, followed by adjustment with individual hematocrit values (Table 2).Adjustment by groupwise average hematocrit produced an inferior model and all models were an improvement on that of urine T measured by the same methods.
Serum A 4 , DBS A 4 , 17OHP and P 4 did not differ significantly over the study period and their summary statistics are shown (Table 1).DBS and serum A 4 also displayed good agreement without bias (Fig. 3).Goodness # data from original study using the same LCMS measurement methods [4].

4.
Plot of Passing-Bablok regression (left panel) of serum androstenedione (x axis) compared with DBS androstenedione (y axis) and of Bland-Altman plot (right panel) for deviance between the two measurements.
of fit of models of A 4 measurement followed the same pattern as with T with the unadjusted whole blood measurement having lowest AIC, followed by serum A 4 and then followed by whole blood adjusted for individual or groupwise average hematocrit (Table 2).Similarly, for 17OHP measurement, the unadjusted whole blood measurement has lowest AIC following by whole blood measurement adjusted by individual and then groupwise hematocrit adjustment.Fig.4 3.

Hematology profile
The hemoglobin (129.5 ± 0.6 g/L, median 130 (125− 134) g/L) and hematocrit (38.3 ± 2.0%, 39 (37, 40)%) were stable across all 12 visits (p>0.97).The measurement of hemoglobin content in DBS samples using the sodium lauryl sulfate method produced very low readings which did not correspond to the original hematocrit measurements and were not employed.

Discussion
The present study evaluated the use of DBS capillary blood sampling to assess circulating testosterone and related sex steroids in healthy women before and after daily transdermal administration of testosterone for seven days.There was a good correlation between for DBS and serum T and A 4 concentrations where DBS measurements were expressed as whole blood measurements without or with hematocrit adjustment by individual or groupwise average hematocrit, consistent with previous reports [8,9].The present finding verify that DBS samples provide a very comparable patterns of serum T before and after testosterone administration.Further, using the data entropy reduction metric AIC, we show that the DBS measurement of whole blood T or A 4 without adjustment for hematocrit provides the best alternative model over adjustment by individual or groupwise hematocrit or urine T measurement.The same applied also to DBs and serum A 4 where there was high correlation without bias but that the unadjusted whole blood measurement provided the best model fit.
In the present study the DBS samples were stored at room temperature for four years before analysis using the same steroid LCMS methods as the original serum samples processed at the time of the study [4].This study demonstrates the storage stability of DBS serum samples consistent with, and exceeding previous reports of storage stability of DBS steroids when measured for up to a year of storage [10] as well as extending our findings of storage stability of serum steroids in frozen storage for 10 years [11].These findings suggest DBS sample have long-term, possibly indefinite, stability in storage.
There has been debate over how DBS measurements, made in whole blood, should be adjusted to compare better with the more familiar serum steroid concentrations [9].An immediate solution is to measure hematocrit in each blood sample; however, by requiring concomitant venipuncture this sacrifices the convenience and simplicity of finger-prick DBS sampling.Alternatively, novel devices for accurate volumetric sampling or biochemical measurements that allow for adjustment based on estimates of hematocrit in the punched out DBS samples have been developed [9].We were pleasantly surprised to find in this analysis that whole blood DBS measurements without additional adjustment provided the best model, superior to those of hematocrit-adjusted whole blood or serum T and A 4 measurements.This may be understood in that adding another error-prone measurement (of hematocrit) to the whole blood measurements merely to resemble more familiar serum measurements, the error variances are compounded and perhaps best avoided.The similarity of the temporal pattern of DBS and serum T indicates that the simpler approach preserves the biological signal in the data at least as well as measuring serum steroid alone.Nevertheless, we also confirmed that contemporaneous hematocrit adjustment of DBS whole blood measurements produced a good correlation with serum measurements for T and A 4 .The present findings indicate that whole blood measurements without hematocrit adjustment provided the best model fit over any hematocrit adjustment (with individual adjustment performing better than groupwise adjustment) or urinary T measurement by the same LCMS steroid analysis method.
The strengths of this study include its high participation rate allowing virtually complete sampling for an intensive sampling design.We also provided novel finding including not just a good correlation between hematocrit-adjusted DBS and serum measurements, but that this was durable for over four years of DBS sample storage at room temperature which greatly simplifies the applications of DBS sampling and storage for prolonged periods.Crucially, we also demonstrate that using unadjusted whole blood steroid measurements can provide a superior descriptive model for a study compared with any hematocritbased adjustment.Though the numbers for whole blood steroids will appear unfamiliarly lower that conventional serum measurements, the present study indicates that no biologically important information is lost, and the experimental data is better represented without making hematocrit adjustments.Limitations of this study are the relatively small sample size of participants (n=12) but as each participant contributed 10 sets of paired DBS and serum sample, the number of samples was 120.Nevertheless, a larger reference range using these methods would be desirable.We conclude that the present study's findings are very encouraging for the wider application of the simpler, cheaper, and less intrusive DBS sampling for field, self-sampling home studies as well as clinical trials.

Fig. 1 .
Fig. 1.Plot of serum T (green circle and line), DBS T adjusted for hematocrit in same sample (blue square and line) and whole blood T unadjusted for hematocrit (red diamond and line) including 4 days of run-in and then days 0, 1, 2, 3, 4, 5 and 6 after treatment.The dashed horizontal lines indicate the upper 95% confidence limits for each analytical measurement in the corresponding colors.

Fig. 2 .
Fig. 2. Plot of Passing-Bablok regression (left panel) of serum testosterone (x axis) compared with DBS testosterone (y axis) and of Bland-Altman plot (right panel) for deviance between the two measurements.

Fig. 3 .
Fig. 3. Left panel is a plot of serum androstenedione (green circle and line), DBS androstenedione adjusted for hematocrit in same sample (blue square and line) and whole blood androstenedione unadjusted for hematocrit (red diamond and line) including 4 days of run-in and then days 0, 1, 2, 3, 4, 5 and 6 after treatment.The right panel is DBS 17 hydroxyprogesterone adjusted for hematocrit in same sample (blue square and line) and whole blood 17 hydroxyprogesterone unadjusted for hematocrit (red diamond and line) including 4 days of run-in and then days 0, 1, 2, 3, 4, 5 and 6 after treatment.The dashed horizontal lines indicate the upper 95% confidence limits for each analytical measurement in the corresponding colors.

Table 1
Summary statistics on DBS and serum A4, 17OHP and P4 in healthy women.

Table 2
Summary of goodness of fit of models for the serial measurement of DBS steroids.AIC, Akaike Information Criterion is a number indicating goodness of fit of a model for one set of data, based on the maximal likelihood function and number of model parameters and reflecting entropy reduction by the model.It allows comparison between different models of the same data.Lower AIC (including negative numbers, not absolute values) indicates better a fitting model.The diagonal (variance) estimates the pooled variability of each data point and the off-diagonal (correlation) estimates the correlation between data points. *