Measuring maternal serum screening markers for Down’s syndrome in plasma collected for cell-free DNA testing

Abstract

Objectives

To determine whether maternal plasma collected in cell-free DNA stabilizing tubes is suitable for measuring prenatal screening ‘serum’ markers.

Methods

Matched plasma and serum samples were collected from 41 second trimester and 42 first trimester non-Down’s syndrome pregnancies. Second trimester samples were tested for alpha-fetoprotein, unconjugated estriol, human chorionic gonadotropin, and inhibin-A (Beckman Coulter DxI immunoassay). First trimester samples were tested for human chorionic gonadotropin and pregnancy-associated plasma protein A. Method comparisons performed for each marker compared plasma and serum results. Down’s syndrome likelihood ratios in serum and plasma were compared.

Results

Plasma and serum results for all markers were highly correlated (r > 0.983) but for all, plasma results differed, usually by proportional amounts. After conversion to multiples of the median using sample type-specific medians, the logarithmic standard deviations in serum and plasma did not differ (all p > 0.37). Likelihood ratios for the first and second trimester marker combinations were highly correlated and closely agreed (log likelihood ratios range 1.005 to 1.032; 1.000 indicates complete agreement).

Conclusions

These results using specialized plasma collection tubes are similar to those of our earlier study showing that plasma collected in EDTA tubes is suitable for ‘serum’ Down’s syndrome screening. Laboratories must account for proportional changes by computing new plasma medians or modifying existing serum medians. Using a portion of the plasma from cell-free DNA collection tubes for ‘serum screening’ may have an advantage in programmes that are reflexively testing cell-free DNA, as only one sample type need be collected.

Introduction

Prenatal screening for Down’s syndrome and other chromosomal disorders has historically been performed using a combination of serum and ultrasound markers ¹ but has evolved to include sequencing of circulating cell-free (cf)DNA in maternal plasma.^2–4 Screening singleton pregnancies via cfDNA sequencing allows detection of 99% of Down’s syndrome, 96% of trisomy 18, and 91% of trisomy 13 pregnancies, with less than a 0.5% overall false-positive rate, ⁵ but failure rates must be considered. In 13 studies utilizing the same cfDNA testing methodology, the initial failure rates ranged between 0.1% and 6.3%. ⁵ However, the average failure rate in the four studies that also included routine testing of a duplicate/subsequent sample was 67% lower than in the 10 studies not testing a second sample (online supplementary Figure 1). Sex chromosome aneuploidy and foetal gender can also be reported using the same sample. This improved method of screening is recommended by the American College of Obstetricians and Gynecologists for women with specific risk factors ⁶ (e.g., serum screen positive result, age 35 or older, ultrasound findings, previous history). Choosing to have cfDNA sequencing as the primary prenatal screening test is common for women at higher risk in the United States, for whom testing is frequently now covered by health insurance. Some women with a ‘low-risk’ pregnancy are also choosing such testing, but often must pay several hundred dollars out-of-pocket.

Table 1.

Summary of six maternal screening marker results with serum compared with those same markers in plasma.

					Median plasma/serum ratio (p) a		Standard deviation (MoM) b
Analyte	Source	Trimester	N	R	Mass units	MoM	Serum	Plasma	F (p) c
AFP	BC DxI	Second	41	0.9927	0.734 (0.58) [0.715–0.753]	1.028 (0.73) [1.000–1.055]	0.1961	0.1922	1.04 (0.90)
uE3	BC DxI	Second	41	0.9887	1.185 (0.42) [1.150–1.221]	0.966 (0.78) [0.937–0.995]	0.1768	0.1682	1.10 (0.75)
hCG	BC DxI	Second	41	0.9917	0.873 (0.18) [0.853–0.893]	0.986 (0.012) [0.963–1.010]	0.1773	0.1667	1.13 (0.70)
Inhibin-A	BC DxI	Second	41	0.9850	0.671 (0.66) [0.653–0.689]	0.984 (0.53) [0.958–1.002]	0.2213	0.2115	1.09 (0.78)
hCG	BC DxI	First	42	0.9852	0.883 (0.30) [0.864–0.902]	0.998 (0.25) [0.976–1.021]	0.1204	0.1149	1.10 (0.77)
PAPP-A	BC DxI	First	42	0.9937	1.163 (0.037) [1.123–1.202]	1.001 (0.028) [0.963–1.038]	0.2659	0.2536	1.10 (0.76)
Inhibin-A	Ansh	Second	41	0.9860	0.828 (0.014) [0.797–0.859]	1.039 (0.036) [0.998–1.080]	0.2548	0.2563	1.01 (0.97)
PAPP-A	Ansh	First	42	0.9953	0.713 (0.57) [0.690–0.736]	1.018 (0.93) [0.982–1.054]	0.2888	0.2738	1.11 (0.73)

AFP: alpha-fetoprotein; BC: Beckman Coulter DxI; uE3: unconjugated estriol; hCG: human chorionic gonadotropin; inhibin-A: dimeric inhibin-A; PAPP-A: pregnancy-associated plasma protein-A; Ansh: Ansh Laboratory reagents; MoM: multiples of the median.

a

The p-value is from a test of the slope from the Bland–Altman plot. A significant p-value indicates the ratio is dependent on the values (e.g., the two sets of results are non-proportional) with a non-significant finding suggesting a proportional relationship. The 95% confidence intervals are provided in brackets.

b

The standard deviations are computed from probability plots (online supplementary Figures 3 and 4) after a logarithmic transformation using a linear regression of the data between the 10th and 90th centiles.

c

The F-value is the ratio of the log variances such that the ratio is > 1.0, and the associated p-value is a test of equal variances.

Implementation strategies for cfDNA include contingent screening, ⁷ in which cfDNA testing is based on initial screening by serum markers utilizing a lower risk cut-off than usual (e.g., screen positive if risk is 1:500 or 1:1000 or greater), and screen positive women (5% to 10% of all tested) are offered follow-up cfDNA or diagnostic testing. An alternative strategy, reflexive testing, ⁸ also relies first on serum screening, but collects a sample suitable for cfDNA testing in all women. The cfDNA testing is then performed on women with screen positive results, usually in the 5% to 10% range, without the need to alert them to return for another blood collection. Both protocols have the potential to increase the Down’s syndrome detection rate while simultaneously reducing the need for amniocentesis owing to a much reduced false-positive rate. As cfDNA testing requires a plasma sample that is most often collected in a container specifically designed to stabilize the cfDNA matrix (e.g., Cell-Free DNA BCT®, Streck, Inc., Omaha, NE), it might be advantageous to be able to test for ‘serum’ markers from that same sample, especially in a reflexive protocol. We have previously reported that plasma collected with an anticoagulant (EDTA, a lavender or purple top tube) and processed within 6 h is suitable for testing ‘serum’ markers using new appropriate medians. ⁹ The current study seeks to extend these findings for plasma samples collected in cfDNA stabilizing plasma ‘Streck’ tubes. This could provide added convenience (and some cost savings) for samples drawn for reflexive cfDNA testing protocols.

Methods

Sample collection was performed under the direction of StemExpress (Placerville, CA), a contracted research organization. The study was approved by Biomedical Research Institute of America IRB (San Diego, CA). Women aged 18 or older, with a singleton pregnancy, who attended one of six prenatal care sites in northern California and Washington State were eligible. Women were informed about the study objectives and had their questions answered. They provided written consent for the collection of two 10 mL blood samples and recording of limited pregnancy-related information (maternal age, gestational age, and number of foetuses). The aim was to collect at least 40 matched plasma and serum samples from pregnant women in the late first trimester (targeting 10–13 weeks’), and another 40 matched samples in the early second trimester (targeting 14–18 weeks’). The first blood sample was collected in a standard red top serum collection tube, and the second in a cfDNA Streck blood collection tube. No information was to be returned to the women or their health care providers. Nucleic acids in the plasma (Streck) sample would not be sequenced; only biochemical markers would be tested. All samples were labelled with a unique study ID that could not identify any individual participant. The plasma (Streck) and serum samples were processed according to standardized protocols, with the plasma sample prepared according to protocols that would be used were the sample to be subject to cfDNA testing. All samples were stored in California at −80℃ until shipment, on dry ice, to Women & Infants Hospital in Rhode Island for testing.

Samples were stored in Rhode Island at −80℃, then thawed and tested within two days of receipt. First trimester samples were assayed for total human chorionic gonadotrophin (hCG) on the DxI autoanalyzer (Beckman Coulter, La Brea, CA) and pregnancy-associated plasma protein-A (PAPP-A) on two platforms: the DxI and a manual assay using reagents obtained from Ansh Laboratories (Webster, TX). Second trimester samples were assayed for total hCG, unconjugated estriol (uE3), alpha-fetoprotein (AFP), and dimeric inhibin-A on the DxI platform. Inhibin-A was also measured using a manual assay based on reagents obtained from Ansh Laboratories. Two analytes (PAPP-A and inhibin-A) were measured on a second platform to confirm that multiple methods could be used.

Paired sample results for each analyte on each testing platform were subject to a formal method comparison, including a Bland–Altman analysis, ¹⁰ to determine whether the results in the two sample types were identical, proportionally different or non-proportionally different. Differences that were proportional could be accounted for by computing new gestational age-specific median values. The data from the paired samples were then used to compute gestational age-specific regressed medians for serum and plasma (Streck) samples using log-linear regression analyses for AFP, uE3, and PAPP-A, negative exponential regression for hCG, and second-order polynomial regression for inhibin-A. Results were then converted to multiples of the median, and a formal method comparison was performed. The population standard deviation was derived from probability plots, after logarithmic transformations, using a linear regression of the data between the 10th and 90th centiles. These population log standard deviations were then compared between the plasma (Streck) and serum samples, as well as with the expected values from the literature. Likelihood ratios for combinations of serum (and plasma) markers expressed as multiples of the median (MoM) were computed using published parameters. ¹¹ Likelihood ratios (and Down’s syndrome risks) are highly right-skewed, and comparisons were performed after a logarithmic transformation. If the Bland–Altman analysis showed the differences to be non-proportional, the subsequent logarithmic standard deviations and Down’s syndrome likelihood ratios were examined to determine whether there was an important clinical impact. Standard deviations were compared using the F-test with two-tailed levels of significance at 0.05.

Results

Paired plasma (Streck) and serum samples were collected between January and March 2015, with 42 first trimester and 41 second trimester sets available for testing and analysis. AFP, uE3, hCG, inhibin-A, and PAPP-A were measured in both plasma (Streck) and serum samples on the Beckman Coulter DxI platform over two days in October 2015. During that same time, inhibin-A and PAPP-A were measured in all samples, using a manual assay with reagents from Ansh Laboratories. Assays were performed according to manufacturer’s specifications, and assay coefficients of variation (CV) were less than 10%.

Figure 1 shows a comparison of the results from four second trimester markers (AFP, uE3, hCG, and inhibin-A in rows 1 through 4) in paired second trimester maternal serum and plasma (Streck) samples, as tested on the Beckman Coulter DxI platform. The scatterplot for AFP (row 1, column A) shows a high correlation between the values (0.993), but the plasma values tend to be lower (i.e., fall below the line of equality; Y = X). The corresponding Bland–Altman plot for AFP (row 1, column B) shows the average of the serum and plasma AFP values (logarithmic horizontal axis) versus the ratio of the plasma to serum value (vertical axis). The horizontal line at 1.0 indicates where the two values are identical. The median ratio is 0.734, indicating that plasma levels of AFP are about 27% lower than corresponding serum levels in the same woman. A regression analysis indicates that the ratio does not change significantly over the range of average values (p = 0.58). Results in mass units were then converted to MoMs based on serum/plasma-specific medians for each marker (online supplementary Figure 2). A second Bland–Altman plot for AFP MoM levels (row 1, column C) shows the ratio of 1.029 is close to 1.00, indicating that conversion to sample type-specific MoM levels has accounted for the proportional difference. Table 1 contains results of the statistical analysis relating to AFP measurements (row 1). In addition to the correlation (0.9927), ratios in mass units and MoM (0.734 and 1.028, respectively), and an analysis of the slope of the Bland–Altman plot (neither have a significant slope, indicating only a proportional difference). The estimated logarithmic standard deviations in both the serum (0.1961) and plasma (0.1922) samples were computed based on probability plots (online supplementary Figure 3), and tested for equality (p = 0.90). Figure 1 and Table 1 contain equivalent analyses for uE3, hCG, and inhibin-A. Nearly all differences between serum and plasma (Streck) were proportional (hCG MoM being the exception). None of the logarithmic standard deviations computed based on probability plots (online supplementary Figures 3 and 4) for the MoM levels were significantly different. The differences between assays were easily accounted for by conversion of results to sample-specific MoM levels.

Figure 1.

Analysis of four second trimester measurements in 41 maternal serum and plasma (Streck) on the Beckman Coulter (BC) DxI platform. Data for AFP, uE3, hCG, and inhibin-A are shown in rows 1 through 4. The related statistical results for these figures are provided in Table 1. The first figure in each row (column a) is a scatterplot of the paired serum/plasma results. The dashed line is the line of identity. For AFP, hCG, and inhibin-A, the observations fall below that line, indicating that the results for plasma are lower than those for serum. The opposite was found for uE3. In the second column (b), the associated Bland–Altman plots are provided for the raw analyte measurements. The x-axis shows the geometric average of the two values while the ratio of plasma to serum is provided on the y-axis. The horizontal line at 1.00 indicates where the two results would be equal. The solid horizontal line indicates the average ratio, while the thicker, slanted line shows the results of a regression analysis. The third column of figures (c) also contains Bland–Altman plots, but after the data have been converted to multiples of the median (MoM) by dividing results by the gestational age-specific median levels specific to each sample type. For all of these second trimester analytes, conversion to sample type-specific MoM results in a ratio of plasma to serum of approximately 1.00.

Figure 2 and Table 1 show the same analyses for first trimester markers hCG and PAPP-A measured on the Beckman Coulter DxI platform. The first trimester hCG results are proportionally different (lower in plasma), but there is a significant relationship between the PAPP-A plasma/serum MoM ratio across the range of PAPP-A values. However, the logarithmic standard deviations for the MoM levels were quite similar for both hCG and PAPP-A for serum and plasma.

Figure 2.

Analysis of two first trimester measurements in 42 maternal serum and plasma (Streck) on the Beckman Coulter (BC) DxI platform. Data for hCG and PAPP-A are shown in rows 1 and 2. The related statistical results for these figures are provided in Table 1. The first figure in each row (column a) is a scatterplot of the paired serum/plasma results. The dashed line is the line of identity. For hCG, the observations fall below that line, indicating that the results for plasma are lower than those for serum. For PAPP-A, results fall above the line. In the second column (b), the associated Bland–Altman plots are provided for the raw analyte measurements. The x-axis shows the geometric average of the two values while the ratio of plasma to serum is provided on the y-axis. The horizontal line at 1.00 indicates where the two results would be equal. The solid horizontal line indicates the average ratio, while the thicker, slanted line shows the results of a regression analysis. The third column of figures (c) also contains Bland–Altman plots, but after the data have been converted to multiples of the median (MoM) by dividing results by the gestational age-specific median levels specific to each sample type. For all of these first trimester analytes, conversion to sample type-specific MoM results in a ratio of plasma to serum of approximately 1.00.

Figure 3 and Table 1 show similar analyses for second trimester inhibin-A and first trimester PAPP-A using a manual assay and Ansh Laboratory reagents. The inhibin-A results show a significant relationship between the plasma/serum ratio across the range of values for both mass units and MoM levels. PAPP-A results were found to be proportional. Again, the logarithmic standard deviations were quite similar.

Figure 3.

Analysis of one first and one second trimester measurements in maternal serum and plasma (Streck) using Ansh Laboratory reagents. Inhibin data from 41 second trimester samples and PAPP-A from 42 first trimester samples are shown in rows 1 and 2. The related statistical results for these figures are provided in Table 1. The first figure in each row (column a) is a scatterplot of the paired serum/plasma results. The dashed line is the line of identity. For both inhibin-A and PAPP-A, the results fall below the line indicating the plasma measurements are lower than serum measurements. In the second column (b), the associated Bland–Altman plots are provided for the raw analyte measurements. The x-axis shows the geometric average of the two values while the ratio of plasma to serum is provided on the y-axis. The horizontal line at 1.00 indicates where the two results would be equal. The solid horizontal line indicates the average ratio, while the thicker, slanted line shows the results of a regression analysis. The third column of figures (c) also contains Bland–Altman plots, but after the data have been converted to multiples of the median (MoM) by dividing results by the gestational age-specific median levels specific to each sample type. For both of these analytes, conversion to sample type-specific MoMs results in a ratio of plasma to serum of approximately 1.00.

The Down’s syndrome likelihood ratios based on the second trimester quadruple markers were compared in the matched serum and plasma samples (online supplementary Figure 5). Both correlations in serum and plasma are high (0.979 and 0.985), and the observations fall on the line of identity (median ratios of the plasma to serum log likelihood ratios were 1.011 and 1.004, respectively). A value of 1.000 would indicate agreement.

The Down’s syndrome likelihood ratios based on the first trimester PAPP-A and hCG markers were also compared (online supplementary Figure 6). Again, the correlations are high (0.980 and 0.985 for PAPP-A on the DxI platform and using Ansh reagents, respectively) and observations fall on the line of identity (median ratios of the plasma to serum log likelihood ratio were 1.032 and 1.016, respectively).

Discussion

Our study shows high correlation and consistency of computed Down’s syndrome likelihood ratios in plasma (Streck) and serum using both first and second trimester marker combinations. These findings further confirm our earlier results ⁹ that plasma (EDTA) is a suitable sample type for multiple marker ‘serum’ screening for Down’s syndrome and other chromosomal abnormalities for at least two assay methodologies. Together, these findings show that if plasma were to be used, regardless of whether from an EDTA or Streck tube, some method to account for proportional differences in assay values from serum must be utilized. This most likely would take the form of new plasma-specific medians that would be developed and monitored according to existing professional guidelines.

For laboratories where the main sample type is serum, with a limited number of plasma samples being tested, existing serum medians could be adjusted by a set of constant factors to generate suitable plasma medians. In our study, for example, the AFP median for serum could be converted to a median appropriate for plasma by multiplying by 0.734 (Table 1). This adjustment could be viewed as a ‘plasma’ factor, and used in the same way as is the current practice to adjust for maternal race or insulin-dependent diabetes status. Another adjustment option for markers where the plasma results are proportionally higher than serum is to dilute the plasma accordingly before measurement, so that the serum median can be applied. Laboratories should verify these correction factors or dilution protocols before reporting clinical results in plasma. Such adjustments should be subject to long-term monitoring of the median plasma results with a target of 1.00 MoM. Once assay results are expressed as MoMs, the computation of Down’s syndrome risk would remain the same.

Our current findings are similar to those we previously reported for EDTA plasma samples. In plasma, regardless of tube type, levels of AFP were reduced relative to serum (EDTA ratio 0.87; Streck ratio 0.73), but levels of uE3 were increased (EDTA 1.17; Streck 1.18). PAPP-A measurements were also higher in plasma than in serum (EDTA ratio 1.32; Streck 1.10). However, second trimester hCG and inhibin-A levels were more markedly altered by use of Streck rather than EDTA plasma tubes. For hCG, levels were similar to serum in EDTA plasma (ratio 0.99), but reduced in Streck tubes (ratio 0.87). While we previously reported that inhibin-A levels were not different between EDTA plasma and serum using the DxI assay method (ratio 1.01), the Streck plasma showed much lower inhibin-A result (ratio 0.67). This is at least partly dependent on the assay method, as the Ansh reagents showed a smaller reduction in inhibin-A levels (0.83) from Streck tubes.

The method of assay also had differential impact on PAPP-A results. With the DxI assay, PAPP-A levels were higher in EDTA or Streck plasma than in serum, but either plasma samples gave lower results when the Ansh assay was used. We sought to explore the cause of this difference. Postulating interference of EDTA in their assay, investigators at Ansh Labs changed the PAPP-A assay dilution buffer to exclude EDTA, and also tried an assay using an alternative detection antibody, more distant from the metal chelating region of the molecule. As a result, there was no change in the PAPP-A results with and without EDTA in the dilution buffers. However, the new detection antibody resulted in PAPP-A levels in EDTA plasma that were higher than those in serum, as had been observed for the DxI assay method. These data suggest that EDTA interferes with the PAPP-A assay when the metal chelating region of the molecule is in close proximity to the antibody binding site.

Our study has limitations. The number of subjects is relatively small, with just over 40 in each trimester, and no cases of aneuploidy included. However, a larger number of samples from 57 Down’s syndrome and 342 euploid pregnancies were included in our previous study, ⁹ which used plasma from EDTA tubes and found similar results. Together, these two studies confirm our finding that plasma is a suitable sample for ‘serum’ screening for at least the two testing methodologies employed. Data on maternal weight were not collected and, therefore, it was not possible to adjust MoM levels for maternal weight. Thus, the logarithmic standard deviations reported here are expected to be slightly higher than published estimates.

Using maternal plasma rather than serum samples to perform multiple marker prenatal screening is possible. Either EDTA (purple top) or Streck (mottled black and tan top) plasma collection tubes appear suitable. The values in plasma and serum are highly correlated but are proportionally different (or only modestly non-proportional), thus requiring separate medians or the use of an adjustment factor. The possibility of using plasma samples in this way might be of most benefit in contingent or reflexive screening models, where collection of only one tube type could add convenience and costs savings to the programme.